switch to xmake for package management (#611)

This change removes our dependency on vcpkg for package management, in favour of bringing some code in-tree in the `thirdparty` folder as well as using the xmake build-in package management feature. For the latter, all the package definitions are maintained in the zen repo itself, in the `repo` folder.

It should now also be easier to build the project as it will no longer depend on having the right version of vcpkg installed, which has been a common problem for new people coming in to the codebase. Now you should only need xmake to build.

* Bumps xmake requirement on github runners to 2.9.9 to resolve an issue where xmake on Windows invokes cmake with `v144` toolchain which does not exist
* BLAKE3 is now in-tree at `thirdparty/blake3`
* cpr is now in-tree at `thirdparty/cpr`
* cxxopts is now in-tree at `thirdparty/cxxopts`
* fmt is now in-tree at `thirdparty/fmt`
* robin-map is now in-tree at `thirdparty/robin-map`
* ryml is now in-tree at `thirdparty/ryml`
* sol2 is now in-tree at `thirdparty/sol2`
* spdlog is now in-tree at `thirdparty/spdlog`
* utfcpp is now in-tree at `thirdparty/utfcpp`
* xmake package repo definitions is in `repo`
* implemented support for sanitizers. ASAN is supported on windows, TSAN, UBSAN, MSAN etc are supported on Linux/MacOS though I have not yet tested it extensively on MacOS
* the zencore encryption implementation also now supports using mbedTLS which is used on MacOS, though for now we still use openssl on Linux
* crashpad
* bumps libcurl to 8.11.0 (from 8.8.0) which should address a rare build upload bug

102 files changed, 52218 insertions, 0 deletions

diff --git a/thirdparty/blake3/.cargo/config.toml b/thirdparty/blake3/.cargo/config.toml
new file mode 100644
index 000000000..8e7c3afd3
--- /dev/null
+++ b/thirdparty/blake3/.cargo/config.toml
@@ -0,0 +1,2 @@
+[target.wasm32-wasip1]
+runner = "wasmtime"
diff --git a/thirdparty/blake3/.git-blame-ignore-revs b/thirdparty/blake3/.git-blame-ignore-revs
new file mode 100644
index 000000000..6e814e69e
--- /dev/null
+++ b/thirdparty/blake3/.git-blame-ignore-revs
@@ -0,0 +1,2 @@
+# CMakeLists.txt whitespace fixups
+3e14f865d30271c74fc68d417af488ea91b66d48
diff --git a/thirdparty/blake3/.github/workflows/build_b3sum.py b/thirdparty/blake3/.github/workflows/build_b3sum.py
new file mode 100644
index 000000000..f0e1787c7
--- /dev/null
+++ b/thirdparty/blake3/.github/workflows/build_b3sum.py
@@ -0,0 +1,38 @@
+#! /usr/bin/env python3
+
+from pathlib import Path
+import platform
+import shutil
+import subprocess
+import sys
+
+ROOT = Path(__file__).parent.parent.parent
+RUST_TARGET = sys.argv[1]
+
+subprocess.run(
+    ["cargo", "build", "--target", sys.argv[1], "--release"], cwd=ROOT / "b3sum"
+)
+
+if platform.system() == "Windows":
+    original_exe_name = "b3sum.exe"
+else:
+    original_exe_name = "b3sum"
+
+if platform.system() == "Windows":
+    new_exe_name = "b3sum_windows_x64_bin.exe"
+elif platform.system() == "Darwin":
+    new_exe_name = "b3sum_macos_x64_bin"
+elif platform.system() == "Linux":
+    new_exe_name = "b3sum_linux_x64_bin"
+else:
+    raise RuntimeError("Unexpected platform: " + platform.system())
+
+# Copy the built binary so that it has the upload name we want.
+out_dir = ROOT / "b3sum/target" / RUST_TARGET / "release"
+original_exe_path = str(out_dir / original_exe_name)
+new_exe_path = str(out_dir / new_exe_name)
+print("copying", repr(original_exe_path), "to", repr(new_exe_path))
+shutil.copyfile(original_exe_path, new_exe_path)
+
+# This lets the subsequent upload step get the filepath.
+print("::set-output name=bin_path::" + new_exe_path)
diff --git a/thirdparty/blake3/.github/workflows/ci.yml b/thirdparty/blake3/.github/workflows/ci.yml
new file mode 100644
index 000000000..296716f52
--- /dev/null
+++ b/thirdparty/blake3/.github/workflows/ci.yml
@@ -0,0 +1,491 @@
+name: tests
+
+on:
+  push:
+    branches:
+      - "*"
+    # not on tags
+  pull_request:
+
+env:
+  BLAKE3_CI: "1"
+  RUSTFLAGS: "-D warnings"
+  RUST_BACKTRACE: "1"
+
+jobs:
+  library_tests:
+    name: ${{ matrix.target.name }} ${{ matrix.channel }}
+    runs-on: ${{ matrix.target.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        target: [
+          { "os": "ubuntu-latest",  "toolchain": "x86_64-unknown-linux-gnu", "name": "Linux GNU" },
+          { "os": "macOS-latest",   "toolchain": "aarch64-apple-darwin",     "name": "macOS" },
+          { "os": "windows-latest", "toolchain": "x86_64-pc-windows-msvc",   "name": "Windows MSVC" },
+          { "os": "windows-latest", "toolchain": "x86_64-pc-windows-gnu",    "name": "Windows GNU" }
+        ]
+        channel: ["stable", "beta", "nightly"]
+
+    steps:
+    - uses: actions/checkout@v4
+    - uses: dtolnay/rust-toolchain@master
+      with:
+        toolchain: ${{ format('{0}-{1}', matrix.channel, matrix.target.toolchain) }}
+    # Print the compiler version, for debugging.
+    - name: print compiler version
+      run: cargo run --quiet
+      working-directory: ./tools/compiler_version
+    # Print out instruction set support, for debugging.
+    - name: print instruction set support
+      run: cargo run --quiet
+      working-directory: ./tools/instruction_set_support
+    # Default tests plus Rayon and trait implementations.
+    - run: cargo test --features=rayon,traits-preview,serde,zeroize
+    # Same but with only one thread in the Rayon pool. This can find deadlocks.
+    - name: "again with RAYON_NUM_THREADS=1"
+      run: cargo test --features=rayon,traits-preview,serde,zeroize
+      env:
+        RAYON_NUM_THREADS: 1
+    # The mmap feature by itself (update_mmap_rayon is omitted).
+    - run: cargo test --features=mmap
+    # All public features put together.
+    - run: cargo test --features=mmap,rayon,traits-preview,serde,zeroize
+    # no_std tests.
+    - run: cargo test --no-default-features
+
+    # A matrix of different test settings:
+    # - debug vs release
+    # - assembly vs Rust+C intrinsics vs pure Rust intrinsics
+    # - different levels of SIMD support
+    #
+    # Full SIMD support.
+    - run: cargo test --features=
+    - run: cargo test --features=prefer_intrinsics
+    - run: cargo test --features=pure
+    - run: cargo test --features= --release
+    - run: cargo test --features=prefer_intrinsics --release
+    - run: cargo test --features=pure --release
+    # No AVX-512.
+    - run: cargo test --features=no_avx512
+    - run: cargo test --features=no_avx512,prefer_intrinsics
+    - run: cargo test --features=no_avx512,pure
+    - run: cargo test --features=no_avx512 --release
+    - run: cargo test --features=no_avx512,prefer_intrinsics --release
+    - run: cargo test --features=no_avx512,pure --release
+    # No AVX2.
+    - run: cargo test --features=no_avx512,no_avx2
+    - run: cargo test --features=no_avx512,no_avx2,prefer_intrinsics
+    - run: cargo test --features=no_avx512,no_avx2,pure
+    - run: cargo test --features=no_avx512,no_avx2 --release
+    - run: cargo test --features=no_avx512,no_avx2,prefer_intrinsics --release
+    - run: cargo test --features=no_avx512,no_avx2,pure --release
+    # No SSE4.1
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,prefer_intrinsics
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,pure
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41 --release
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,prefer_intrinsics --release
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,pure --release
+    # No SSE2
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,no_sse2
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,no_sse2,prefer_intrinsics
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,no_sse2,pure
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,no_sse2 --release
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,no_sse2,prefer_intrinsics --release
+    - run: cargo test --features=no_avx512,no_avx2,no_sse41,no_sse2,pure --release
+
+    # Test benchmarks. RUSTC_BOOTSTRAP=1 lets this run on non-nightly toolchains.
+    - run: cargo test --benches --features=rayon
+      env:
+        RUSTC_BOOTSTRAP: 1
+    # Test vectors.
+    - name: test vectors
+      run: cargo test
+      working-directory: ./test_vectors
+    - name: test vectors intrinsics
+      run: cargo test --features=prefer_intrinsics
+      working-directory: ./test_vectors
+    - name: test vectors pure
+      run: cargo test --features=pure
+      working-directory: ./test_vectors
+    # Test C code.
+    - name: cargo test C bindings assembly
+      run: cargo test
+      working-directory: ./c/blake3_c_rust_bindings
+    - name: cargo test C bindings intrinsics
+      run: cargo test --features=prefer_intrinsics
+      working-directory: ./c/blake3_c_rust_bindings
+    - name: cargo test C bindings no AVX-512
+      run: cargo test
+      working-directory: ./c/blake3_c_rust_bindings
+      env:
+        CFLAGS: -DBLAKE3_NO_AVX512
+    - name: cargo test C bindings no AVX2
+      run: cargo test
+      working-directory: ./c/blake3_c_rust_bindings
+      env:
+        CFLAGS: -DBLAKE3_NO_AVX512 -DBLAKE3_NO_AVX2
+    - name: cargo test C bindings no SSE41
+      run: cargo test
+      working-directory: ./c/blake3_c_rust_bindings
+      env:
+        CFLAGS: -DBLAKE3_NO_AVX512 -DBLAKE3_NO_AVX2 -DBLAKE3_NO_SSE41
+    - name: cargo test C bindings no SSE2
+      run: cargo test
+      working-directory: ./c/blake3_c_rust_bindings
+      env:
+        CFLAGS: -DBLAKE3_NO_AVX512 -DBLAKE3_NO_AVX2 -DBLAKE3_NO_SSE41 -DBLAKE3_NO_SSE2
+    # Reference impl doc test.
+    - name: reference impl doc test
+      run: cargo test
+      working-directory: ./reference_impl
+
+  msrv_build:
+    name: MSRV build ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: ["ubuntu-latest", "macOS-latest", "windows-latest"]
+    steps:
+    - uses: actions/checkout@v4
+    # The current MSRV. This crate doesn't have an official MSRV policy,
+    # but in practice we'll probably do what libc does:
+    # https://github.com/rust-lang/libs-team/issues/72.
+    # This test target is here so that we notice if we accidentally bump
+    # the MSRV, but it's not a promise that we won't bump it.
+    - uses: dtolnay/[email protected]
+    - run: cargo build --features=mmap,rayon,traits-preview,serde,zeroize
+
+  b3sum_tests:
+    name: b3sum ${{ matrix.target.name }} ${{ matrix.channel }}
+    runs-on: ${{ matrix.target.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        target: [
+          { "os": "ubuntu-latest",  "toolchain": "x86_64-unknown-linux-gnu", "name": "Linux GNU" },
+          { "os": "macOS-latest",   "toolchain": "aarch64-apple-darwin",      "name": "macOS" },
+          { "os": "windows-latest", "toolchain": "x86_64-pc-windows-msvc",   "name": "Windows MSVC" },
+          { "os": "windows-latest", "toolchain": "x86_64-pc-windows-gnu",    "name": "Windows GNU" }
+        ]
+        channel: ["stable", "beta", "nightly"]
+
+    steps:
+    - uses: actions/checkout@v4
+    - uses: dtolnay/rust-toolchain@master
+      with:
+        toolchain: ${{ format('{0}-{1}', matrix.channel, matrix.target.toolchain) }}
+    # Test b3sum.
+    - name: test b3sum
+      run: cargo test
+      working-directory: ./b3sum
+    - name: test b3sum --no-default-features
+      run: cargo test --no-default-features
+      working-directory: ./b3sum
+
+  cross_tests:
+    name: cross ${{ matrix.arch }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        arch:
+          - i586-unknown-linux-musl
+          - i686-unknown-linux-musl
+          - armv7-unknown-linux-gnueabihf
+          - aarch64-unknown-linux-gnu
+          # Big-endian targets. See https://twitter.com/burntsushi5/status/1695483429997945092.
+          - powerpc64-unknown-linux-gnu
+          - s390x-unknown-linux-gnu
+
+    steps:
+    - uses: actions/checkout@v4
+    - uses: dtolnay/rust-toolchain@stable
+    - run: cargo install cross
+    # Test the portable implementation on everything.
+    - run: cross test --target ${{ matrix.arch }}
+    # Test building for ancient i386 processors without guaranteed SSE2 support.
+    - run: cross rustc --target ${{ matrix.arch }} -- -C target-cpu=i386
+      if: startsWith(matrix.arch, 'i586-') || startsWith(matrix.arch, 'i686-')
+    # Test the NEON implementation on ARM targets.
+    - run: cross test --target ${{ matrix.arch }} --features=neon
+      if: startsWith(matrix.arch, 'armv7-') || startsWith(matrix.arch, 'aarch64-')
+    # NEON is enabled by default on aarch64, disabling it through the no_neon feature.
+    - run: cross test --target ${{ matrix.arch }} --features=no_neon
+      if: startsWith(matrix.arch, 'aarch64-')
+    # Test vectors. Note that this uses a hacky script due to path dependency limitations.
+    - run: ./test_vectors/cross_test.sh --target ${{ matrix.arch }}
+    # C code. Same issue with the hacky script.
+    - run: ./c/blake3_c_rust_bindings/cross_test.sh --target ${{ matrix.arch }}
+    - run: ./c/blake3_c_rust_bindings/cross_test.sh --target ${{ matrix.arch }} --features=neon
+      if: startsWith(matrix.arch, 'armv7-') || startsWith(matrix.arch, 'aarch64-')
+
+  wasm_tests:
+    name: WASM tests
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+    - uses: dtolnay/rust-toolchain@stable
+      with:
+        targets: wasm32-wasip1
+    - name: install Wasmtime
+      run: |
+        curl https://wasmtime.dev/install.sh -sSf | bash
+        echo PATH: $PATH
+        mkdir -p ~/.local/bin
+        ln -s ~/.wasmtime/bin/wasmtime ~/.local/bin/wasmtime
+    - run: cargo test --target wasm32-wasip1
+    - run: cargo test --target wasm32-wasip1 --no-default-features
+    - run: cargo test --target wasm32-wasip1 --features wasm32_simd
+    - run: cargo test --target wasm32-wasip1 --no-default-features --features wasm32_simd
+    - run: cargo test --target wasm32-wasip1 --benches --features=wasm32_simd
+      env:
+        RUSTC_BOOTSTRAP: 1
+    - name: test vectors w/o SIMD
+      run: cargo test --target wasm32-wasip1
+      working-directory: ./test_vectors
+    - name: test vectors w/ SIMD
+      run: cargo test --target wasm32-wasip1 --features wasm32_simd
+      working-directory: ./test_vectors
+
+  cargo_xwin_test:
+    name: cargo xwin test
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+    - run: docker run -v $(pwd):/io -w /io messense/cargo-xwin cargo xwin test --target x86_64-pc-windows-msvc --features=mmap,rayon,traits-preview,serde,zeroize
+
+  # Currently only on x86.
+  cmake_c_tests:
+    name: CMake C tests SIMD=${{ matrix.simd }} TBB=${{ matrix.use_tbb }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        use_tbb: ["OFF", "ON"]
+        simd: ["x86-intrinsics", "amd64-asm"]
+    steps:
+    - uses: actions/checkout@v4
+    - run: |
+        sudo apt-get update
+        sudo apt-get install ninja-build libtbb-dev libtbb12
+    # Test the intrinsics-based and assembly-based implementations.
+    - run: |
+        cmake --fresh -S c -B c/build -G Ninja -DBLAKE3_TESTING=ON -DBLAKE3_TESTING_CI=ON "-DBLAKE3_SIMD_TYPE=${{ matrix.simd }}"
+        cmake --build c/build --target test
+        cat c/build/Testing/Temporary/LastTest.log
+    - run: |
+        cmake --fresh -S c -B c/build -G Ninja -DBLAKE3_TESTING=ON -DBLAKE3_TESTING_CI=ON "-DBLAKE3_SIMD_TYPE=${{ matrix.simd }}" -DBLAKE3_NO_SSE2=1
+        cmake --build c/build --target test
+        cat c/build/Testing/Temporary/LastTest.log
+    - run: |
+        cmake --fresh -S c -B c/build -G Ninja -DBLAKE3_TESTING=ON -DBLAKE3_TESTING_CI=ON "-DBLAKE3_SIMD_TYPE=${{ matrix.simd }}" -DBLAKE3_NO_SSE2=1 -DBLAKE3_NO_SSE41=1
+        cmake --build c/build --target test
+        cat c/build/Testing/Temporary/LastTest.log
+    - run: |
+        cmake --fresh -S c -B c/build -G Ninja -DBLAKE3_TESTING=ON -DBLAKE3_TESTING_CI=ON "-DBLAKE3_SIMD_TYPE=${{ matrix.simd }}" "-DBLAKE3_NO_SSE2=1" "-DBLAKE3_NO_SSE41=1" "-DBLAKE3_NO_AVX2=1"
+        cmake --build c/build --target test
+        cat c/build/Testing/Temporary/LastTest.log
+    - run: |
+        cmake --fresh -S c -B c/build -G Ninja -DBLAKE3_TESTING=ON -DBLAKE3_TESTING_CI=ON "-DBLAKE3_SIMD_TYPE=${{ matrix.simd }}" "-DBLAKE3_NO_SSE2=1" "-DBLAKE3_NO_SSE41=1" "-DBLAKE3_NO_AVX2=1" "-DBLAKE3_NO_AVX512=1"
+        cmake --build c/build --target test
+        cat c/build/Testing/Temporary/LastTest.log
+    # Test with TBB disabled/enabled.
+    - run: |
+        cmake --fresh -S c -B c/build -G Ninja -DBLAKE3_TESTING=ON -DBLAKE3_TESTING_CI=ON "-DBLAKE3_USE_TBB=${{ matrix.use_tbb }}"
+        cmake --build c/build --target test
+        cat c/build/Testing/Temporary/LastTest.log
+    # Build the example with TBB disabled/enabled.
+    - run: |
+        cmake --fresh -S c -B c/build -G Ninja -DBLAKE3_TESTING=ON -DBLAKE3_TESTING_CI=ON -DBLAKE3_EXAMPLES=ON "-DBLAKE3_USE_TBB=${{ matrix.use_tbb }}"
+        cmake --build c/build --target blake3-example
+
+  # Currently only on x86.
+  pkg_config_c_tests:
+    name: pkg-config C tests TBB=${{ matrix.use_tbb }} BUILD_SHARED_LIBS=${{ matrix.shared_libs }} STDLIB=${{ matrix.stdlib }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        use_tbb: ["OFF", "ON"]
+        shared_libs: ["OFF", "ON"]
+        stdlib: ["libc++", "libstdc++"]
+    steps:
+    - uses: actions/checkout@v4
+    - name: update packages
+      run: |
+        sudo apt-get update
+        sudo apt-get install ninja-build libtbb-dev libtbb12
+        ${{ matrix.stdlib != 'libc++' || 'sudo apt-get install libc++-dev libc++abi-dev' }}
+    - name: configure cmake
+      run: |
+        export CXXFLAGS=${{ matrix.stdlib == 'libc++' && '-stdlib=libc++' || '' }}
+        export CC=${{ matrix.stdlib == 'libc++' && 'clang' || 'gcc' }}
+        export CXX=${{ matrix.stdlib == 'libc++' && 'clang++' || 'g++' }}
+        cmake --fresh -S c -B c/build -G Ninja -DCMAKE_INSTALL_PREFIX=${{ github.workspace }}/target "-DBLAKE3_USE_TBB=${{ matrix.use_tbb }}" "-DBUILD_SHARED_LIBS=${{ matrix.shared_libs }}" -DCMAKE_VERBOSE_MAKEFILE=1
+    - run: cmake --build c/build --target install
+    - run: mkdir -p ${{ github.workspace }}/target/bin
+    - run: echo "PKG_CONFIG_PATH=${{ github.workspace }}/target/lib/pkgconfig" >> $GITHUB_ENV
+    - run: gcc -O3 -o ${{ github.workspace }}/target/bin/blake3-example c/example.c $(pkg-config --cflags --libs libblake3)
+    - if: matrix.use_tbb == 'ON'
+      run: gcc -O3 -o ${{ github.workspace }}/target/bin/blake3-example-tbb c/example_tbb.c $(pkg-config --cflags --libs libblake3)
+
+  # Note that this jobs builds AArch64 binaries from an x86_64 host.
+  build_apple_silicon:
+    name: build for Apple Silicon
+    runs-on: macOS-latest
+    strategy:
+      fail-fast: false
+    steps:
+    - uses: actions/checkout@v4
+    - uses: dtolnay/rust-toolchain@stable
+      with:
+        targets: aarch64-apple-darwin
+    - name: build blake3
+      run: cargo build --target aarch64-apple-darwin
+    - name: build b3sum
+      run: cargo build --target aarch64-apple-darwin
+      working-directory: ./b3sum
+
+  build_tinycc:
+    name: build with the Tiny C Compiler
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+    - name: install TCC
+      run: sudo apt-get install -y tcc
+    - name: compile
+      run: >
+        tcc -shared -O3 -o libblake3.so \
+          -DBLAKE3_NO_SSE2 -DBLAKE3_NO_SSE41 -DBLAKE3_NO_AVX2 -DBLAKE3_NO_AVX512 \
+          blake3.c blake3_dispatch.c blake3_portable.c
+      working-directory: ./c
+
+  # See https://github.com/BLAKE3-team/BLAKE3/issues/271 for why we test this.
+  # Note that this isn't guaranteed to execute on an AVX-512-supporting server,
+  # but hopefully at least some of the time it will.
+  gcc54:
+    name: "compile and test with GCC 5.4"
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+    - uses: addnab/docker-run-action@v3
+      with:
+        image: gcc:5.4
+        options: -v ${{ github.workspace }}:/work
+        run: |
+          cat /proc/cpuinfo
+          curl https://sh.rustup.rs -sSf | sh -s -- -y --profile minimal
+          cd /work
+          ~/.cargo/bin/cargo test --features prefer_intrinsics
+
+  # CMake build test (Library only).
+  cmake_current_build:
+    name: CMake ${{ matrix.os }} CC=${{ matrix.toolchain.cc }} CXX=${{ matrix.toolchain.cxx }} TBB=${{ matrix.use_tbb }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        cmakeVersion: [latest]
+        ninjaVersion: [latest]
+        os: [ubuntu-latest, macOS-latest, windows-latest]
+        toolchain: [
+          { cc: cl, cxx: cl },
+          { cc: clang, cxx: clang++ },
+          { cc: clang-cl, cxx: clang-cl },
+          { cc: gcc, cxx: g++ },
+        ]
+        use_tbb: [OFF, ON]
+        exclude:
+          - os: macOS-latest
+            toolchain: { cc: cl, cxx: cl }
+          - os: macOS-latest
+            toolchain: { cc: clang-cl, cxx: clang-cl }
+          - os: ubuntu-latest
+            toolchain: { cc: cl, cxx: cl }
+          - os: ubuntu-latest
+            toolchain: { cc: clang-cl, cxx: clang-cl }
+          - os: windows-latest
+            toolchain: { cc: clang, cxx: clang++ }
+            use_tbb: ON
+          - os: windows-latest
+            toolchain: { cc: gcc, cxx: g++ }
+            use_tbb: ON
+    steps:
+      - uses: actions/checkout@v4
+      - uses: lukka/get-cmake@5f6e04f5267c8133f1273bf2103583fc72c46b17
+        with:
+          cmakeVersion: ${{ matrix.cmakeVersion }}
+          ninjaVersion: ${{ matrix.ninjaVersion }}
+      - if: matrix.os == 'macOS-latest'
+        name: Install dependencies on macOS
+        run: |
+          brew update
+          brew install tbb
+      - if: matrix.os == 'ubuntu-latest'
+        name: Install dependencies on Linux
+        run: |
+          sudo apt-get update
+          sudo apt-get install libtbb-dev libtbb12
+      - name: CMake generation, build, install
+        run: |
+          ${{ matrix.os != 'windows-latest' || '& "C:/Program Files/Microsoft Visual Studio/2022/Enterprise/Common7/Tools/Launch-VsDevShell.ps1" -Arch amd64 -SkipAutomaticLocation' }}
+          cmake -S c -B c/build -G Ninja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=${{ github.workspace }}/target -DCMAKE_C_COMPILER=${{ matrix.toolchain.cc }} -DCMAKE_CXX_COMPILER=${{ matrix.toolchain.cxx }} -DBLAKE3_USE_TBB=${{ matrix.use_tbb }} -DBLAKE3_FETCH_TBB=${{ matrix.os == 'windows-latest' && 'YES' || 'NO' }} -DBLAKE3_EXAMPLES=ON
+          cmake --build c/build --target install
+
+  cmake_3-9_build:
+    name: CMake 3.9.6 ubuntu-latest
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: lukka/get-cmake@5f6e04f5267c8133f1273bf2103583fc72c46b17
+        with:
+          cmakeVersion: 3.9.6
+      - name: Create build directory
+        run: mkdir c/build
+      - name: CMake generation
+        run: cmake .. -DCMAKE_INSTALL_PREFIX=${{github.workspace}}/target
+        working-directory: c/build
+      - name: CMake build / install
+        run: make install
+        working-directory: c/build
+
+  miri_smoketest:
+    name: Miri smoketest
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+    - uses: dtolnay/rust-toolchain@nightly
+      with:
+        components: miri
+    # Currently the test search "miri" only matches "test_miri_smoketest", but
+    # we might add more. If this accidentally picks up anything incompatible or
+    # slow, we can narrow it.
+    - run: cargo miri test miri
+
+  tbb_rust_bindings_tests:
+    name: TBB test bindings ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: ["ubuntu-latest", "macOS-latest"]
+    steps:
+    - uses: actions/checkout@v4
+    - uses: dtolnay/rust-toolchain@stable
+    - name: install TBB
+      if: matrix.os == 'ubuntu-latest'
+      run: |
+        sudo apt-get update
+        sudo apt-get install libtbb-dev libtbb12
+    - name: install TBB
+      if: matrix.os == 'macOS-latest'
+      run: |
+        brew install tbb
+        echo "CXXFLAGS=-I$(brew --prefix)/include $CPPFLAGS" >> $GITHUB_ENV
+        echo "RUSTFLAGS=-L$(brew --prefix)/lib $RUSTFLAGS" >> $GITHUB_ENV
+    - name: cargo test C bindings with TBB
+      run: cargo test --features=tbb
+      working-directory: ./c/blake3_c_rust_bindings
diff --git a/thirdparty/blake3/.github/workflows/tag.yml b/thirdparty/blake3/.github/workflows/tag.yml
new file mode 100644
index 000000000..61be4ff99
--- /dev/null
+++ b/thirdparty/blake3/.github/workflows/tag.yml
@@ -0,0 +1,43 @@
+name: publish_b3sum_binaries
+
+on:
+  push:
+    tags:
+      - "*"
+
+env:
+  BLAKE3_CI: "1"
+  RUSTFLAGS: "-D warnings"
+
+jobs:
+  cargo_tests:
+    name: ${{ matrix.target.name }}
+    runs-on: ${{ matrix.target.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        target: [
+          { "os": "ubuntu-latest",  "rust-target": "x86_64-unknown-linux-musl", "name": "Linux" },
+          { "os": "macOS-latest",   "rust-target": "x86_64-apple-darwin",       "name": "macOS" },
+          { "os": "windows-latest", "rust-target": "x86_64-pc-windows-msvc",    "name": "Windows" },
+        ]
+
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v4
+        with:
+          python-version: "3.x"
+      - run: pip install PyGithub
+      - run: sudo apt-get install musl-tools
+        if: matrix.target.os == 'ubuntu-latest'
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          targets: ${{ matrix.target.rust-target }}
+      - name: build b3sum
+        id: build_b3sum
+        run: python -u .github/workflows/build_b3sum.py ${{ matrix.target.rust-target }}
+      - name: upload release asset
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          GITHUB_TAG: ${{ github.ref }}
+        run: python -u .github/workflows/upload_github_release_asset.py ${{ steps.build_b3sum.outputs.bin_path }}
diff --git a/thirdparty/blake3/.github/workflows/upload_github_release_asset.py b/thirdparty/blake3/.github/workflows/upload_github_release_asset.py
new file mode 100755
index 000000000..76340bee3
--- /dev/null
+++ b/thirdparty/blake3/.github/workflows/upload_github_release_asset.py
@@ -0,0 +1,73 @@
+#! /usr/bin/env python3
+
+import github
+import os
+import sys
+import time
+
+RETRIES = 10
+
+g = github.Github(os.environ["GITHUB_TOKEN"])
+tag_name = os.environ["GITHUB_TAG"]
+tag_prefix = "refs/tags/"
+if tag_name.startswith(tag_prefix):
+    tag_name = tag_name[len(tag_prefix) :]
+assert len(sys.argv) == 2
+asset_path = sys.argv[1]
+asset_name = os.path.basename(asset_path)
+
+repo = g.get_repo(os.environ["GITHUB_REPOSITORY"])
+
+tags = list(repo.get_tags())
+
+for tag in tags:
+    if tag.name == tag_name:
+        break
+else:
+    raise RuntimeError("no tag named " + repr(tag_name))
+
+try:
+    print("Creating GitHub release for tag " + repr(tag_name) + "...")
+    repo.create_git_release(tag_name, tag_name, tag.commit.commit.message)
+except github.GithubException as github_error:
+    if github_error.data["errors"][0]["code"] == "already_exists":
+        print("Release for tag " + repr(tag_name) + " already exists.")
+    else:
+        raise
+
+
+def get_release():
+    for i in range(RETRIES):
+        releases = list(repo.get_releases())
+        for release in releases:
+            if release.tag_name == tag_name:
+                return release
+        print(f"Release for tag {repr(tag_name)} not found. Retrying...")
+        time.sleep(1)
+    raise RuntimeError("no release for tag " + repr(tag_name))
+
+
+release = get_release()
+
+print("Uploading " + repr(asset_path) + "...")
+for i in range(RETRIES):
+    try:
+        print("Upload attempt #{} of {}...".format(i + 1, RETRIES))
+        release.upload_asset(asset_path)
+        break
+    except github.GithubException as github_error:
+        # Unfortunately the asset upload API is flaky. Even worse, it often
+        # partially succeeds, returning an error to the caller but leaving the
+        # release in a state where subsequent uploads of the same asset will
+        # fail with an "already_exists" error. (Though the asset is not visible
+        # on github.com, so we can't just declare victory and move on.) If we
+        # detect this case, explicitly delete the asset and continue retrying.
+        print(github_error)
+        for asset in release.get_assets():
+            if asset.name == asset_name:
+                print("Found uploaded asset after failure. Deleting...")
+                asset.delete_asset()
+else:
+    raise RuntimeError("All upload attempts failed.")
+
+print("Success!")
diff --git a/thirdparty/blake3/.gitignore b/thirdparty/blake3/.gitignore
new file mode 100644
index 000000000..fa8d85ac5
--- /dev/null
+++ b/thirdparty/blake3/.gitignore
@@ -0,0 +1,2 @@
+Cargo.lock
+target
diff --git a/thirdparty/blake3/CONTRIBUTING.md b/thirdparty/blake3/CONTRIBUTING.md
new file mode 100644
index 000000000..3a605f255
--- /dev/null
+++ b/thirdparty/blake3/CONTRIBUTING.md
@@ -0,0 +1,31 @@
+# Contributing
+
+We welcome and encourage third-party contributions to BLAKE3, be it reports of issues encountered while using the software or proposals of patches.
+
+## Bug reports
+
+Bugs and other problems should be reported on [GitHub Issues](https://github.com/BLAKE3/BLAKE3/issues).
+
+If you report a bug, please:
+
+* Check that it's not already reported in the [GitHub Issues](https://github.com/BLAKE3/BLAKE3/issues).
+* Provide information to help us diagnose and ideally reproduce the bug.
+
+## Patches
+
+We encourage you to fix a bug via a [GitHub Pull request](https://github.com/BLAKE3/BLAKE3/pulls), preferably after creating a related issue and referring it in the PR.
+
+If you contribute code and submit a patch, please note the following:
+
+* We use Rust's stable branch for developing BLAKE3.
+* Pull requests should target the `master` branch.
+* Try to follow the established Rust [style guidelines](https://doc.rust-lang.org/1.0.0/style/).
+
+Also please make sure to create new unit tests covering your code additions. You can execute the tests by running:
+
+```bash
+cargo test
+```
+
+All third-party contributions will be recognized in the list of contributors.
+
diff --git a/thirdparty/blake3/Cargo.toml b/thirdparty/blake3/Cargo.toml
new file mode 100644
index 000000000..25ab3451d
--- /dev/null
+++ b/thirdparty/blake3/Cargo.toml
@@ -0,0 +1,135 @@
+[package]
+name = "blake3"
+version = "1.8.2"
+authors = ["Jack O'Connor <[email protected]>", "Samuel Neves"]
+description = "the BLAKE3 hash function"
+repository = "https://github.com/BLAKE3-team/BLAKE3"
+license = "CC0-1.0 OR Apache-2.0 OR Apache-2.0 WITH LLVM-exception"
+documentation = "https://docs.rs/blake3"
+readme = "README.md"
+edition = "2021"
+
+[features]
+default = ["std"]
+
+# The NEON implementation does not participate in dynamic feature detection,
+# which is currently x86-only. If "neon" is on, NEON support is assumed. Note
+# that AArch64 always supports NEON, but support on ARMv7 varies. The NEON
+# implementation uses C intrinsics and requires a C compiler.
+neon = []
+
+# The Wasm SIMD implementation does not participate in dynamic feature detection,
+# which is currently x86-only. If "wasm_simd" is on, Wasm SIMD support is assumed.
+# Note that not all Wasm implementations support the Wasm SIMD specification.
+# This may become the default in the future.
+wasm32_simd = []
+
+# This crate uses libstd for std::io trait implementations, and also for
+# runtime CPU feature detection. This feature is enabled by default. If you use
+# --no-default-features, the only way to use the SIMD implementations in this
+# crate is to enable the corresponding instruction sets statically for the
+# entire build, with e.g. RUSTFLAGS="-C target-cpu=native".
+std = []
+
+# The `rayon` feature (disabled by default, but enabled for docs.rs) adds the
+# `update_rayon` and (in combination with `mmap` below) `update_mmap_rayon`
+# methods, for multithreaded hashing. However, even if this feature is enabled,
+# all other APIs remain single-threaded.
+#
+# Implementation detail: We take a dependency on rayon-core instead of rayon,
+# because it builds faster and still includes all the APIs we need.
+rayon = ["dep:rayon-core"]
+
+# The `mmap` feature (disabled by default, but enabled for docs.rs) adds the
+# `update_mmap` and (in combination with `rayon` above) `update_mmap_rayon`
+# helper methods for memory-mapped IO.
+mmap = ["std", "dep:memmap2"]
+
+# Implement the zeroize::Zeroize trait for types in this crate.
+zeroize = ["dep:zeroize", "arrayvec/zeroize"]
+
+# This crate implements traits from the RustCrypto project, exposed here as the
+# "traits-preview" feature. However, these traits aren't stable, and they're
+# expected to change in incompatible ways before they reach 1.0. For that
+# reason, this crate makes no SemVer guarantees for this feature, and callers
+# who use it should expect breaking changes between patch versions of this
+# crate. (The "*-preview" feature name follows the conventions of the RustCrypto
+# "signature" crate.)
+traits-preview = ["dep:digest"]
+
+# ---------- Features below this line are undocumented and unstable. ----------
+# The following features are mainly intended for testing and benchmarking, and
+# they might change or disappear at any time without a major version bump.
+
+# It wasn't originally intended to expose "digest" as its own feature, but the
+# traits-preview feature above predated the "dep:" syntax in Cargo. Version
+# 1.5.2 of this crate started using "dep:" syntax, but that broke some callers
+# in the wild (https://solana.stackexchange.com/q/17787/29050). This feature
+# unbreaks those callers. When Cargo gains the ability to deprecate features,
+# this feature will be deprecated. Note that the relevant trait implementations
+# are still gated by "traits-preview".
+digest = ["dep:digest"]
+
+# By default on x86_64, this crate uses Samuel Neves' hand-written assembly
+# implementations for SSE4.1, AVX2, and AVX512. (These provide both the best
+# runtime performance, and the fastest build times.) And by default on 32-bit
+# x86, this crate uses Rust intrinsics implementations for SSE4.1 and AVX2, and
+# a C intrinsics implementation for AVX-512. In both cases, if a C compiler is
+# not detected, or if AVX-512 support is missing from the detected compiler,
+# build.rs automatically falls back to a pure Rust build. This feature forces
+# that fallback, for testing purposes. (Note that in CI testing, we set the
+# BLAKE3_CI environment variable, which instructs build.rs to error out rather
+# than doing an automatic fallback.)
+pure = []
+
+# As described above, on x86_64 this crate use assembly implementations by
+# default. Enabling the "prefer_intrinsics" feature makes this crate use
+# intrinsics implementations on both 32-bit and 64-bit x86, again for testing
+# purposes.
+prefer_intrinsics = []
+
+# Disable individual instruction sets. CI testing uses these flags to simulate
+# different levels of hardware SIMD support. Note that code for the
+# corresponding instruction set is still compiled; only detection is disabled.
+#
+# As noted above, these flags are *for testing only* and are not stable. It's
+# possible that some users might find that their particular use case performs
+# better if e.g. AVX-512 is disabled, because of issues like CPU downclocking.
+# If that comes up, and if disabling the instruction set here at the feature
+# level turns out to be the right approach, then we can design a stable
+# feature. Until then, we reserve the right to break these features in a patch
+# release.
+no_sse2 = []
+no_sse41 = []
+no_avx2 = []
+no_avx512 = []
+no_neon = []
+
+[package.metadata.docs.rs]
+# Document the rayon/mmap methods and the Serialize/Deserialize/Zeroize impls on docs.rs.
+features = ["mmap", "rayon", "serde", "zeroize"]
+
+[dependencies]
+arrayref = "0.3.5"
+arrayvec = { version = "0.7.4", default-features = false }
+constant_time_eq = { version = "0.3.1", default-features = false }
+cfg-if = "1.0.0"
+digest = { version = "0.10.1", features = ["mac"], optional = true }
+memmap2 = { version = "0.9", optional = true }
+rayon-core = { version = "1.12.1", optional = true }
+serde = { version = "1.0", default-features = false, features = ["derive"], optional = true }
+zeroize = { version = "1", default-features = false, optional = true }
+
+[dev-dependencies]
+hmac = "0.12.0"
+hex = "0.4.2"
+page_size = "0.6.0"
+rand = "0.9.0"
+rand_chacha = "0.9.0"
+reference_impl = { path = "./reference_impl" }
+tempfile = "3.8.0"
+serde_json = "1.0.107"
+ciborium = "0.2.2"
+
+[build-dependencies]
+cc = "1.1.12"
diff --git a/thirdparty/blake3/LICENSE_A2 b/thirdparty/blake3/LICENSE_A2
new file mode 100644
index 000000000..d512ca94d
--- /dev/null
+++ b/thirdparty/blake3/LICENSE_A2
@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2019 Jack O'Connor and Samuel Neves
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
diff --git a/thirdparty/blake3/LICENSE_A2LLVM b/thirdparty/blake3/LICENSE_A2LLVM
new file mode 100644
index 000000000..8d11bec0e
--- /dev/null
+++ b/thirdparty/blake3/LICENSE_A2LLVM
@@ -0,0 +1,219 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+    TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+    1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+    2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+    3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+    4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+    5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+    6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+    7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+    8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+    9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+    END OF TERMS AND CONDITIONS
+
+    APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+    Copyright 2019 Jack O'Connor and Samuel Neves
+
+    Licensed under the Apache License, Version 2.0 (the "License");
+    you may not use this file except in compliance with the License.
+    You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+    Unless required by applicable law or agreed to in writing, software
+    distributed under the License is distributed on an "AS IS" BASIS,
+    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+    See the License for the specific language governing permissions and
+    limitations under the License.
+
+
+---- LLVM Exceptions to the Apache 2.0 License ----
+
+As an exception, if, as a result of your compiling your source code, portions
+of this Software are embedded into an Object form of such source code, you
+may redistribute such embedded portions in such Object form without complying
+with the conditions of Sections 4(a), 4(b) and 4(d) of the License.
+
+In addition, if you combine or link compiled forms of this Software with
+software that is licensed under the GPLv2 ("Combined Software") and if a
+court of competent jurisdiction determines that the patent provision (Section
+3), the indemnity provision (Section 9) or other Section of the License
+conflicts with the conditions of the GPLv2, you may retroactively and
+prospectively choose to deem waived or otherwise exclude such Section(s) of
+the License, but only in their entirety and only with respect to the Combined
+Software.
+
diff --git a/thirdparty/blake3/LICENSE_CC0 b/thirdparty/blake3/LICENSE_CC0
new file mode 100644
index 000000000..0e259d42c
--- /dev/null
+++ b/thirdparty/blake3/LICENSE_CC0
@@ -0,0 +1,121 @@
+Creative Commons Legal Code
+
+CC0 1.0 Universal
+
+    CREATIVE COMMONS CORPORATION IS NOT A LAW FIRM AND DOES NOT PROVIDE
+    LEGAL SERVICES. DISTRIBUTION OF THIS DOCUMENT DOES NOT CREATE AN
+    ATTORNEY-CLIENT RELATIONSHIP. CREATIVE COMMONS PROVIDES THIS
+    INFORMATION ON AN "AS-IS" BASIS. CREATIVE COMMONS MAKES NO WARRANTIES
+    REGARDING THE USE OF THIS DOCUMENT OR THE INFORMATION OR WORKS
+    PROVIDED HEREUNDER, AND DISCLAIMS LIABILITY FOR DAMAGES RESULTING FROM
+    THE USE OF THIS DOCUMENT OR THE INFORMATION OR WORKS PROVIDED
+    HEREUNDER.
+
+Statement of Purpose
+
+The laws of most jurisdictions throughout the world automatically confer
+exclusive Copyright and Related Rights (defined below) upon the creator
+and subsequent owner(s) (each and all, an "owner") of an original work of
+authorship and/or a database (each, a "Work").
+
+Certain owners wish to permanently relinquish those rights to a Work for
+the purpose of contributing to a commons of creative, cultural and
+scientific works ("Commons") that the public can reliably and without fear
+of later claims of infringement build upon, modify, incorporate in other
+works, reuse and redistribute as freely as possible in any form whatsoever
+and for any purposes, including without limitation commercial purposes.
+These owners may contribute to the Commons to promote the ideal of a free
+culture and the further production of creative, cultural and scientific
+works, or to gain reputation or greater distribution for their Work in
+part through the use and efforts of others.
+
+For these and/or other purposes and motivations, and without any
+expectation of additional consideration or compensation, the person
+associating CC0 with a Work (the "Affirmer"), to the extent that he or she
+is an owner of Copyright and Related Rights in the Work, voluntarily
+elects to apply CC0 to the Work and publicly distribute the Work under its
+terms, with knowledge of his or her Copyright and Related Rights in the
+Work and the meaning and intended legal effect of CC0 on those rights.
+
+1. Copyright and Related Rights. A Work made available under CC0 may be
+protected by copyright and related or neighboring rights ("Copyright and
+Related Rights"). Copyright and Related Rights include, but are not
+limited to, the following:
+
+  i. the right to reproduce, adapt, distribute, perform, display,
+     communicate, and translate a Work;
+ ii. moral rights retained by the original author(s) and/or performer(s);
+iii. publicity and privacy rights pertaining to a person's image or
+     likeness depicted in a Work;
+ iv. rights protecting against unfair competition in regards to a Work,
+     subject to the limitations in paragraph 4(a), below;
+  v. rights protecting the extraction, dissemination, use and reuse of data
+     in a Work;
+ vi. database rights (such as those arising under Directive 96/9/EC of the
+     European Parliament and of the Council of 11 March 1996 on the legal
+     protection of databases, and under any national implementation
+     thereof, including any amended or successor version of such
+     directive); and
+vii. other similar, equivalent or corresponding rights throughout the
+     world based on applicable law or treaty, and any national
+     implementations thereof.
+
+2. Waiver. To the greatest extent permitted by, but not in contravention
+of, applicable law, Affirmer hereby overtly, fully, permanently,
+irrevocably and unconditionally waives, abandons, and surrenders all of
+Affirmer's Copyright and Related Rights and associated claims and causes
+of action, whether now known or unknown (including existing as well as
+future claims and causes of action), in the Work (i) in all territories
+worldwide, (ii) for the maximum duration provided by applicable law or
+treaty (including future time extensions), (iii) in any current or future
+medium and for any number of copies, and (iv) for any purpose whatsoever,
+including without limitation commercial, advertising or promotional
+purposes (the "Waiver"). Affirmer makes the Waiver for the benefit of each
+member of the public at large and to the detriment of Affirmer's heirs and
+successors, fully intending that such Waiver shall not be subject to
+revocation, rescission, cancellation, termination, or any other legal or
+equitable action to disrupt the quiet enjoyment of the Work by the public
+as contemplated by Affirmer's express Statement of Purpose.
+
+3. Public License Fallback. Should any part of the Waiver for any reason
+be judged legally invalid or ineffective under applicable law, then the
+Waiver shall be preserved to the maximum extent permitted taking into
+account Affirmer's express Statement of Purpose. In addition, to the
+extent the Waiver is so judged Affirmer hereby grants to each affected
+person a royalty-free, non transferable, non sublicensable, non exclusive,
+irrevocable and unconditional license to exercise Affirmer's Copyright and
+Related Rights in the Work (i) in all territories worldwide, (ii) for the
+maximum duration provided by applicable law or treaty (including future
+time extensions), (iii) in any current or future medium and for any number
+of copies, and (iv) for any purpose whatsoever, including without
+limitation commercial, advertising or promotional purposes (the
+"License"). The License shall be deemed effective as of the date CC0 was
+applied by Affirmer to the Work. Should any part of the License for any
+reason be judged legally invalid or ineffective under applicable law, such
+partial invalidity or ineffectiveness shall not invalidate the remainder
+of the License, and in such case Affirmer hereby affirms that he or she
+will not (i) exercise any of his or her remaining Copyright and Related
+Rights in the Work or (ii) assert any associated claims and causes of
+action with respect to the Work, in either case contrary to Affirmer's
+express Statement of Purpose.
+
+4. Limitations and Disclaimers.
+
+ a. No trademark or patent rights held by Affirmer are waived, abandoned,
+    surrendered, licensed or otherwise affected by this document.
+ b. Affirmer offers the Work as-is and makes no representations or
+    warranties of any kind concerning the Work, express, implied,
+    statutory or otherwise, including without limitation warranties of
+    title, merchantability, fitness for a particular purpose, non
+    infringement, or the absence of latent or other defects, accuracy, or
+    the present or absence of errors, whether or not discoverable, all to
+    the greatest extent permissible under applicable law.
+ c. Affirmer disclaims responsibility for clearing rights of other persons
+    that may apply to the Work or any use thereof, including without
+    limitation any person's Copyright and Related Rights in the Work.
+    Further, Affirmer disclaims responsibility for obtaining any necessary
+    consents, permissions or other rights required for any use of the
+    Work.
+ d. Affirmer understands and acknowledges that Creative Commons is not a
+    party to this document and has no duty or obligation with respect to
+    this CC0 or use of the Work.
diff --git a/thirdparty/blake3/README.md b/thirdparty/blake3/README.md
new file mode 100644
index 000000000..feca7e1bb
--- /dev/null
+++ b/thirdparty/blake3/README.md
@@ -0,0 +1,229 @@
+# BLAKE3
+
+BLAKE3 is a cryptographic hash function that is:
+
+- **Much faster** than MD5, SHA-1, SHA-2, SHA-3, and BLAKE2.
+- **Secure**, unlike MD5 and SHA-1. And secure against length extension,
+  unlike SHA-2.
+- **Highly parallelizable** across any number of threads and SIMD lanes,
+  because it's a Merkle tree on the inside.
+- Capable of **verified streaming** and **incremental updates**, again
+  because it's a Merkle tree.
+- A **PRF**, **MAC**, **KDF**, and **XOF**, as well as a regular hash.
+- **One algorithm with no variants**, which is fast on x86-64 and also
+  on smaller architectures.
+
+The [chart below](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/benchmarks/bar_chart.py)
+is an example benchmark of 16 KiB inputs on a Cascade Lake-SP 8275CL server CPU
+from 2019. For more detailed benchmarks, see the
+[BLAKE3 paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf).
+
+<p align="center">
+<img src="media/speed.svg" alt="performance graph">
+</p>
+
+BLAKE3 is based on an optimized instance of the established hash
+function [BLAKE2](https://blake2.net) and on the [original Bao tree
+mode](https://github.com/oconnor663/bao/blob/master/docs/spec_0.9.1.md).
+The specifications and design rationale are available in the [BLAKE3
+paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf).
+The default output size is 256 bits. The current version of
+[Bao](https://github.com/oconnor663/bao) implements verified streaming
+with BLAKE3.
+
+This repository is the official implementation of BLAKE3. It includes:
+
+* The [`blake3`](https://crates.io/crates/blake3) Rust crate, which
+  includes optimized implementations for SSE2, SSE4.1, AVX2, AVX-512,
+  NEON, and WASM, with automatic runtime CPU feature detection on x86.
+  The `rayon` feature provides multithreading.
+
+* The [`b3sum`](https://crates.io/crates/b3sum) Rust crate, which
+  provides a command line interface. It uses multithreading by default,
+  making it an order of magnitude faster than e.g. `sha256sum` on
+  typical desktop hardware.
+
+* The [C implementation](c), which like the Rust implementation includes SIMD
+  optimizations (all except WASM), CPU feature detection on x86, and optional
+  multithreading. See [`c/README.md`](c/README.md).
+
+* The [Rust reference implementation](reference_impl/reference_impl.rs),
+  which is discussed in Section 5.1 of the [BLAKE3
+  paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf).
+  This implementation is much smaller and simpler than the optimized
+  ones above. If you want to see how BLAKE3 works, or you're writing a
+  port that doesn't need multithreading or SIMD optimizations, start
+  here. Ports of the reference implementation to other languages are
+  hosted in separate repositories
+  ([C](https://github.com/oconnor663/blake3_reference_impl_c),
+  [Python](https://github.com/oconnor663/pure_python_blake3)).
+
+* A [set of test
+  vectors](https://github.com/BLAKE3-team/BLAKE3/blob/master/test_vectors/test_vectors.json)
+  that covers extended outputs, all three modes, and a variety of input
+  lengths.
+
+* [![Actions Status](https://github.com/BLAKE3-team/BLAKE3/workflows/tests/badge.svg)](https://github.com/BLAKE3-team/BLAKE3/actions)
+
+BLAKE3 was designed by:
+
+* [@oconnor663] (Jack O'Connor)
+* [@sneves] (Samuel Neves)
+* [@veorq] (Jean-Philippe Aumasson)
+* [@zookozcash] (Zooko)
+
+The development of BLAKE3 was sponsored by [Electric Coin Company](https://electriccoin.co).
+
+BLAKE3 is also [specified](https://c2sp.org/BLAKE3) in the [Community
+Cryptography Specification Project (C2SP)](https://c2sp.org).
+
+*NOTE: BLAKE3 is not a password hashing algorithm, because it's
+designed to be fast, whereas password hashing should not be fast. If you
+hash passwords to store the hashes or if you derive keys from passwords,
+we recommend [Argon2](https://github.com/P-H-C/phc-winner-argon2).*
+
+## Usage
+
+### The `b3sum` utility
+
+The `b3sum` command line utility prints the BLAKE3 hashes of files or of
+standard input. Prebuilt binaries are available for Linux, Windows, and
+macOS (requiring the [unidentified developer
+workaround](https://support.apple.com/guide/mac-help/open-a-mac-app-from-an-unidentified-developer-mh40616/mac))
+on the [releases page](https://github.com/BLAKE3-team/BLAKE3/releases).
+If you've [installed Rust and
+Cargo](https://doc.rust-lang.org/cargo/getting-started/installation.html),
+you can also build `b3sum` yourself with:
+
+```bash
+cargo install b3sum
+```
+
+If `rustup` didn't configure your `PATH` for you, you might need to go
+looking for the installed binary in e.g. `~/.cargo/bin`. You can test
+out how fast BLAKE3 is on your machine by creating a big file and
+hashing it, for example:
+
+```bash
+# Create a 1 GB file.
+head -c 1000000000 /dev/zero > /tmp/bigfile
+# Hash it with SHA-256.
+time openssl sha256 /tmp/bigfile
+# Hash it with BLAKE3.
+time b3sum /tmp/bigfile
+```
+
+### The `blake3` crate [![docs.rs](https://docs.rs/blake3/badge.svg)](https://docs.rs/blake3)
+
+To use BLAKE3 from Rust code, add a dependency on the `blake3` crate to
+your `Cargo.toml`. Here's an example of hashing some input bytes:
+
+```rust
+// Hash an input all at once.
+let hash1 = blake3::hash(b"foobarbaz");
+
+// Hash an input incrementally.
+let mut hasher = blake3::Hasher::new();
+hasher.update(b"foo");
+hasher.update(b"bar");
+hasher.update(b"baz");
+let hash2 = hasher.finalize();
+assert_eq!(hash1, hash2);
+
+// Extended output. OutputReader also implements Read and Seek.
+let mut output = [0; 1000];
+let mut output_reader = hasher.finalize_xof();
+output_reader.fill(&mut output);
+assert_eq!(hash1, output[..32]);
+
+// Print a hash as hex.
+println!("{}", hash1);
+```
+
+Besides `hash`, BLAKE3 provides two other modes, `keyed_hash` and
+`derive_key`. The `keyed_hash` mode takes a 256-bit key:
+
+```rust
+// MAC an input all at once.
+let example_key = [42u8; 32];
+let mac1 = blake3::keyed_hash(&example_key, b"example input");
+
+// MAC incrementally.
+let mut hasher = blake3::Hasher::new_keyed(&example_key);
+hasher.update(b"example input");
+let mac2 = hasher.finalize();
+assert_eq!(mac1, mac2);
+```
+
+The `derive_key` mode takes a context string and some key material (not a
+password). The context string should be hardcoded, globally unique, and
+application-specific. A good default format for the context string is
+`"[application] [commit timestamp] [purpose]"`:
+
+```rust
+// Derive a couple of subkeys for different purposes.
+const EMAIL_CONTEXT: &str = "BLAKE3 example 2020-01-07 17:10:44 email key";
+const API_CONTEXT: &str = "BLAKE3 example 2020-01-07 17:11:21 API key";
+let input_key_material = b"usually at least 32 random bytes, not a password";
+let email_key = blake3::derive_key(EMAIL_CONTEXT, input_key_material);
+let api_key = blake3::derive_key(API_CONTEXT, input_key_material);
+assert_ne!(email_key, api_key);
+```
+
+### The C implementation
+
+See [`c/README.md`](c/README.md).
+
+### Other implementations
+
+We post links to third-party bindings and implementations on the
+[@BLAKE3team Twitter account](https://twitter.com/BLAKE3team) whenever
+we hear about them. Some highlights include [an optimized Go
+implementation](https://github.com/zeebo/blake3), [Wasm bindings for
+Node.js and browsers](https://github.com/connor4312/blake3), [binary
+wheels for Python](https://github.com/oconnor663/blake3-py), [.NET
+bindings](https://github.com/xoofx/Blake3.NET), and [JNI
+bindings](https://github.com/sken77/BLAKE3jni).
+
+## Contributing
+
+Please see [CONTRIBUTING.md](CONTRIBUTING.md).
+
+## Licenses
+
+This work is released into the public domain with [CC0 1.0](./LICENSE_CC0).
+Alternatively, it is licensed under any of the following:
+
+* [Apache 2.0](./LICENSE_A2)
+* [Apache 2.0 with LLVM exceptions](./LICENSE_A2LLVM)
+
+
+## Adoption & deployment
+
+* [Bazel](https://github.com/bazelbuild/bazel/releases/tag/6.4.0)
+* [Cargo](https://github.com/rust-lang/cargo/pull/14137)
+* [Ccache](https://github.com/ccache/ccache/pull/519)
+* [Chia](https://github.com/Chia-Network/chia-blockchain/blob/main/CHANGELOG.md#10beta8-aka-beta-18---2020-07-16)
+* [Clickhouse](https://github.com/ClickHouse/ClickHouse/blob/master/rust/chcache/Cargo.toml#L7)
+* [Farcaster](https://www.farcaster.xyz/)
+* [IPFS](https://github.com/ipfs/go-verifcid/issues/13)
+* [Iroh](https://www.iroh.computer/blog/blake3-hazmat-api)
+* [LLVM](https://reviews.llvm.org/D121510)
+* [Nix](https://github.com/NixOS/nix/pull/12379)
+* [Nym](https://github.com/nymtech/nym/blob/59056a22c5e6b01a38da2124662bd1fa3c8abef2/common/nymsphinx/params/src/lib.rs#L5)
+* [OpenZFS](https://github.com/openzfs/zfs/)
+* [Redox](https://www.redox-os.org/news/pkgar-introduction/)
+* [Solana](https://docs.rs/solana-program/1.9.5/solana_program/blake3/index.html)
+* [Tekken 8](https://x.com/rodarmor/status/1751567502050771189)
+* [Wasmer](https://github.com/wasmerio/wasmer/blob/4f935a8c162bf604df223003e434e4f7ca253688/lib/cache/src/hash.rs#L21)
+
+
+## Miscellany
+
+- [@veorq] and [@oconnor663] did [an interview with Cryptography FM](https://www.cryptography.fm/3).
+- [@oconnor663] did [an interview with Saito](https://www.youtube.com/watch?v=cJkmIt7yN_E).
+
+[@oconnor663]: https://github.com/oconnor663
+[@sneves]: https://github.com/sneves
+[@veorq]: https://github.com/veorq
+[@zookozcash]: https://github.com/zookozcash
diff --git a/thirdparty/blake3/b3sum/.gitignore b/thirdparty/blake3/b3sum/.gitignore
new file mode 100644
index 000000000..9da4a887b
--- /dev/null
+++ b/thirdparty/blake3/b3sum/.gitignore
@@ -0,0 +1 @@
+!Cargo.lock
diff --git a/thirdparty/blake3/b3sum/Cargo.lock b/thirdparty/blake3/b3sum/Cargo.lock
new file mode 100644
index 000000000..e09a52daf
--- /dev/null
+++ b/thirdparty/blake3/b3sum/Cargo.lock
@@ -0,0 +1,513 @@
+# This file is automatically @generated by Cargo.
+# It is not intended for manual editing.
+version = 4
+
+[[package]]
+name = "anstream"
+version = "0.6.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8acc5369981196006228e28809f761875c0327210a891e941f4c683b3a99529b"
+dependencies = [
+ "anstyle",
+ "anstyle-parse",
+ "anstyle-query",
+ "anstyle-wincon",
+ "colorchoice",
+ "is_terminal_polyfill",
+ "utf8parse",
+]
+
+[[package]]
+name = "anstyle"
+version = "1.0.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "55cc3b69f167a1ef2e161439aa98aed94e6028e5f9a59be9a6ffb47aef1651f9"
+
+[[package]]
+name = "anstyle-parse"
+version = "0.2.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3b2d16507662817a6a20a9ea92df6652ee4f94f914589377d69f3b21bc5798a9"
+dependencies = [
+ "utf8parse",
+]
+
+[[package]]
+name = "anstyle-query"
+version = "1.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "79947af37f4177cfead1110013d678905c37501914fba0efea834c3fe9a8d60c"
+dependencies = [
+ "windows-sys",
+]
+
+[[package]]
+name = "anstyle-wincon"
+version = "3.0.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ca3534e77181a9cc07539ad51f2141fe32f6c3ffd4df76db8ad92346b003ae4e"
+dependencies = [
+ "anstyle",
+ "once_cell",
+ "windows-sys",
+]
+
+[[package]]
+name = "anyhow"
+version = "1.0.98"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e16d2d3311acee920a9eb8d33b8cbc1787ce4a264e85f964c2404b969bdcd487"
+
+[[package]]
+name = "arrayref"
+version = "0.3.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "76a2e8124351fda1ef8aaaa3bbd7ebbcb486bbcd4225aca0aa0d84bb2db8fecb"
+
+[[package]]
+name = "arrayvec"
+version = "0.7.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7c02d123df017efcdfbd739ef81735b36c5ba83ec3c59c80a9d7ecc718f92e50"
+
+[[package]]
+name = "b3sum"
+version = "1.8.2"
+dependencies = [
+ "anyhow",
+ "blake3",
+ "clap",
+ "duct",
+ "hex",
+ "rayon-core",
+ "tempfile",
+ "wild",
+]
+
+[[package]]
+name = "bitflags"
+version = "2.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c8214115b7bf84099f1309324e63141d4c5d7cc26862f97a0a857dbefe165bd"
+
+[[package]]
+name = "blake3"
+version = "1.8.2"
+dependencies = [
+ "arrayref",
+ "arrayvec",
+ "cc",
+ "cfg-if",
+ "constant_time_eq",
+ "memmap2",
+ "rayon-core",
+]
+
+[[package]]
+name = "cc"
+version = "1.2.19"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8e3a13707ac958681c13b39b458c073d0d9bc8a22cb1b2f4c8e55eb72c13f362"
+dependencies = [
+ "shlex",
+]
+
+[[package]]
+name = "cfg-if"
+version = "1.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "baf1de4339761588bc0619e3cbc0120ee582ebb74b53b4efbf79117bd2da40fd"
+
+[[package]]
+name = "clap"
+version = "4.5.37"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eccb054f56cbd38340b380d4a8e69ef1f02f1af43db2f0cc817a4774d80ae071"
+dependencies = [
+ "clap_builder",
+ "clap_derive",
+]
+
+[[package]]
+name = "clap_builder"
+version = "4.5.37"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "efd9466fac8543255d3b1fcad4762c5e116ffe808c8a3043d4263cd4fd4862a2"
+dependencies = [
+ "anstream",
+ "anstyle",
+ "clap_lex",
+ "strsim",
+ "terminal_size",
+]
+
+[[package]]
+name = "clap_derive"
+version = "4.5.32"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09176aae279615badda0765c0c0b3f6ed53f4709118af73cf4655d85d1530cd7"
+dependencies = [
+ "heck",
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
+[[package]]
+name = "clap_lex"
+version = "0.7.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f46ad14479a25103f283c0f10005961cf086d8dc42205bb44c46ac563475dca6"
+
+[[package]]
+name = "colorchoice"
+version = "1.0.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5b63caa9aa9397e2d9480a9b13673856c78d8ac123288526c37d7839f2a86990"
+
+[[package]]
+name = "constant_time_eq"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7c74b8349d32d297c9134b8c88677813a227df8f779daa29bfc29c183fe3dca6"
+
+[[package]]
+name = "crossbeam-deque"
+version = "0.8.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9dd111b7b7f7d55b72c0a6ae361660ee5853c9af73f70c3c2ef6858b950e2e51"
+dependencies = [
+ "crossbeam-epoch",
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-epoch"
+version = "0.9.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5b82ac4a3c2ca9c3460964f020e1402edd5753411d7737aa39c3714ad1b5420e"
+dependencies = [
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-utils"
+version = "0.8.21"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d0a5c400df2834b80a4c3327b3aad3a4c4cd4de0629063962b03235697506a28"
+
+[[package]]
+name = "duct"
+version = "0.13.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e4ab5718d1224b63252cd0c6f74f6480f9ffeb117438a2e0f5cf6d9a4798929c"
+dependencies = [
+ "libc",
+ "once_cell",
+ "os_pipe",
+ "shared_child",
+]
+
+[[package]]
+name = "errno"
+version = "0.3.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "976dd42dc7e85965fe702eb8164f21f450704bdde31faefd6471dba214cb594e"
+dependencies = [
+ "libc",
+ "windows-sys",
+]
+
+[[package]]
+name = "fastrand"
+version = "2.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "37909eebbb50d72f9059c3b6d82c0463f2ff062c9e95845c43a6c9c0355411be"
+
+[[package]]
+name = "getrandom"
+version = "0.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "73fea8450eea4bac3940448fb7ae50d91f034f941199fcd9d909a5a07aa455f0"
+dependencies = [
+ "cfg-if",
+ "libc",
+ "r-efi",
+ "wasi",
+]
+
+[[package]]
+name = "glob"
+version = "0.3.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a8d1add55171497b4705a648c6b583acafb01d58050a51727785f0b2c8e0a2b2"
+
+[[package]]
+name = "heck"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"
+
+[[package]]
+name = "hex"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7f24254aa9a54b5c858eaee2f5bccdb46aaf0e486a595ed5fd8f86ba55232a70"
+
+[[package]]
+name = "is_terminal_polyfill"
+version = "1.70.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7943c866cc5cd64cbc25b2e01621d07fa8eb2a1a23160ee81ce38704e97b8ecf"
+
+[[package]]
+name = "libc"
+version = "0.2.172"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d750af042f7ef4f724306de029d18836c26c1765a54a6a3f094cbd23a7267ffa"
+
+[[package]]
+name = "linux-raw-sys"
+version = "0.9.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cd945864f07fe9f5371a27ad7b52a172b4b499999f1d97574c9fa68373937e12"
+
+[[package]]
+name = "memmap2"
+version = "0.9.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fd3f7eed9d3848f8b98834af67102b720745c4ec028fcd0aa0239277e7de374f"
+dependencies = [
+ "libc",
+]
+
+[[package]]
+name = "once_cell"
+version = "1.21.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "42f5e15c9953c5e4ccceeb2e7382a716482c34515315f7b03532b8b4e8393d2d"
+
+[[package]]
+name = "os_pipe"
+version = "1.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5ffd2b0a5634335b135d5728d84c5e0fd726954b87111f7506a61c502280d982"
+dependencies = [
+ "libc",
+ "windows-sys",
+]
+
+[[package]]
+name = "proc-macro2"
+version = "1.0.95"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "02b3e5e68a3a1a02aad3ec490a98007cbc13c37cbe84a3cd7b8e406d76e7f778"
+dependencies = [
+ "unicode-ident",
+]
+
+[[package]]
+name = "quote"
+version = "1.0.40"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1885c039570dc00dcb4ff087a89e185fd56bae234ddc7f056a945bf36467248d"
+dependencies = [
+ "proc-macro2",
+]
+
+[[package]]
+name = "r-efi"
+version = "5.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "74765f6d916ee2faa39bc8e68e4f3ed8949b48cccdac59983d287a7cb71ce9c5"
+
+[[package]]
+name = "rayon-core"
+version = "1.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1465873a3dfdaa8ae7cb14b4383657caab0b3e8a0aa9ae8e04b044854c8dfce2"
+dependencies = [
+ "crossbeam-deque",
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "rustix"
+version = "1.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d97817398dd4bb2e6da002002db259209759911da105da92bec29ccb12cf58bf"
+dependencies = [
+ "bitflags",
+ "errno",
+ "libc",
+ "linux-raw-sys",
+ "windows-sys",
+]
+
+[[package]]
+name = "shared_child"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09fa9338aed9a1df411814a5b2252f7cd206c55ae9bf2fa763f8de84603aa60c"
+dependencies = [
+ "libc",
+ "windows-sys",
+]
+
+[[package]]
+name = "shlex"
+version = "1.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0fda2ff0d084019ba4d7c6f371c95d8fd75ce3524c3cb8fb653a3023f6323e64"
+
+[[package]]
+name = "strsim"
+version = "0.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7da8b5736845d9f2fcb837ea5d9e2628564b3b043a70948a3f0b778838c5fb4f"
+
+[[package]]
+name = "syn"
+version = "2.0.100"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b09a44accad81e1ba1cd74a32461ba89dee89095ba17b32f5d03683b1b1fc2a0"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
+[[package]]
+name = "tempfile"
+version = "3.19.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7437ac7763b9b123ccf33c338a5cc1bac6f69b45a136c19bdd8a65e3916435bf"
+dependencies = [
+ "fastrand",
+ "getrandom",
+ "once_cell",
+ "rustix",
+ "windows-sys",
+]
+
+[[package]]
+name = "terminal_size"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "45c6481c4829e4cc63825e62c49186a34538b7b2750b73b266581ffb612fb5ed"
+dependencies = [
+ "rustix",
+ "windows-sys",
+]
+
+[[package]]
+name = "unicode-ident"
+version = "1.0.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5a5f39404a5da50712a4c1eecf25e90dd62b613502b7e925fd4e4d19b5c96512"
+
+[[package]]
+name = "utf8parse"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06abde3611657adf66d383f00b093d7faecc7fa57071cce2578660c9f1010821"
+
+[[package]]
+name = "wasi"
+version = "0.14.2+wasi-0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9683f9a5a998d873c0d21fcbe3c083009670149a8fab228644b8bd36b2c48cb3"
+dependencies = [
+ "wit-bindgen-rt",
+]
+
+[[package]]
+name = "wild"
+version = "2.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a3131afc8c575281e1e80f36ed6a092aa502c08b18ed7524e86fbbb12bb410e1"
+dependencies = [
+ "glob",
+]
+
+[[package]]
+name = "windows-sys"
+version = "0.59.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1e38bc4d79ed67fd075bcc251a1c39b32a1776bbe92e5bef1f0bf1f8c531853b"
+dependencies = [
+ "windows-targets",
+]
+
+[[package]]
+name = "windows-targets"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9b724f72796e036ab90c1021d4780d4d3d648aca59e491e6b98e725b84e99973"
+dependencies = [
+ "windows_aarch64_gnullvm",
+ "windows_aarch64_msvc",
+ "windows_i686_gnu",
+ "windows_i686_gnullvm",
+ "windows_i686_msvc",
+ "windows_x86_64_gnu",
+ "windows_x86_64_gnullvm",
+ "windows_x86_64_msvc",
+]
+
+[[package]]
+name = "windows_aarch64_gnullvm"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "32a4622180e7a0ec044bb555404c800bc9fd9ec262ec147edd5989ccd0c02cd3"
+
+[[package]]
+name = "windows_aarch64_msvc"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09ec2a7bb152e2252b53fa7803150007879548bc709c039df7627cabbd05d469"
+
+[[package]]
+name = "windows_i686_gnu"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8e9b5ad5ab802e97eb8e295ac6720e509ee4c243f69d781394014ebfe8bbfa0b"
+
+[[package]]
+name = "windows_i686_gnullvm"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0eee52d38c090b3caa76c563b86c3a4bd71ef1a819287c19d586d7334ae8ed66"
+
+[[package]]
+name = "windows_i686_msvc"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "240948bc05c5e7c6dabba28bf89d89ffce3e303022809e73deaefe4f6ec56c66"
+
+[[package]]
+name = "windows_x86_64_gnu"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "147a5c80aabfbf0c7d901cb5895d1de30ef2907eb21fbbab29ca94c5b08b1a78"
+
+[[package]]
+name = "windows_x86_64_gnullvm"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "24d5b23dc417412679681396f2b49f3de8c1473deb516bd34410872eff51ed0d"
+
+[[package]]
+name = "windows_x86_64_msvc"
+version = "0.52.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec"
+
+[[package]]
+name = "wit-bindgen-rt"
+version = "0.39.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6f42320e61fe2cfd34354ecb597f86f413484a798ba44a8ca1165c58d42da6c1"
+dependencies = [
+ "bitflags",
+]
diff --git a/thirdparty/blake3/b3sum/Cargo.toml b/thirdparty/blake3/b3sum/Cargo.toml
new file mode 100644
index 000000000..8ea930291
--- /dev/null
+++ b/thirdparty/blake3/b3sum/Cargo.toml
@@ -0,0 +1,26 @@
+[package]
+name = "b3sum"
+version = "1.8.2"
+authors = ["Jack O'Connor <[email protected]>"]
+description = "a command line implementation of the BLAKE3 hash function"
+repository = "https://github.com/BLAKE3-team/BLAKE3"
+license = "CC0-1.0 OR Apache-2.0 OR Apache-2.0 WITH LLVM-exception"
+readme = "README.md"
+edition = "2021"
+
+[features]
+neon = ["blake3/neon"]
+prefer_intrinsics = ["blake3/prefer_intrinsics"]
+pure = ["blake3/pure"]
+
+[dependencies]
+anyhow = "1.0.25"
+blake3 = { version = "1.8", path = "..", features = ["mmap", "rayon"] }
+clap = { version = "4.0.8", features = ["derive", "wrap_help"] }
+hex = "0.4.0"
+rayon-core = "1.12.1"
+wild = "2.0.3"
+
+[dev-dependencies]
+duct = "0.13.3"
+tempfile = "3.1.0"
diff --git a/thirdparty/blake3/b3sum/LICENSE_A2 b/thirdparty/blake3/b3sum/LICENSE_A2
new file mode 120000
index 000000000..c7b0be82d
--- /dev/null
+++ b/thirdparty/blake3/b3sum/LICENSE_A2
@@ -0,0 +1 @@
+../LICENSE_A2
\ No newline at end of file
diff --git a/thirdparty/blake3/b3sum/LICENSE_A2LLVM b/thirdparty/blake3/b3sum/LICENSE_A2LLVM
new file mode 120000
index 000000000..c9f3d16a6
--- /dev/null
+++ b/thirdparty/blake3/b3sum/LICENSE_A2LLVM
@@ -0,0 +1 @@
+../LICENSE_A2LLVM
\ No newline at end of file
diff --git a/thirdparty/blake3/b3sum/LICENSE_CC0 b/thirdparty/blake3/b3sum/LICENSE_CC0
new file mode 120000
index 000000000..856562a79
--- /dev/null
+++ b/thirdparty/blake3/b3sum/LICENSE_CC0
@@ -0,0 +1 @@
+../LICENSE_CC0
\ No newline at end of file
diff --git a/thirdparty/blake3/b3sum/README.md b/thirdparty/blake3/b3sum/README.md
new file mode 100644
index 000000000..595ce9e95
--- /dev/null
+++ b/thirdparty/blake3/b3sum/README.md
@@ -0,0 +1,72 @@
+# b3sum
+
+A command line utility for calculating
+[BLAKE3](https://github.com/BLAKE3-team/BLAKE3) hashes, similar to
+Coreutils tools like `b2sum` or `md5sum`.
+
+```
+Usage: b3sum [OPTIONS] [FILE]...
+
+Arguments:
+  [FILE]...  Files to hash, or checkfiles to check
+
+Options:
+      --keyed                 Use the keyed mode, reading the 32-byte key from stdin
+      --derive-key <CONTEXT>  Use the key derivation mode, with the given context string
+  -l, --length <LEN>          The number of output bytes, before hex encoding [default: 32]
+      --seek <SEEK>           The starting output byte offset, before hex encoding [default: 0]
+      --num-threads <NUM>     The maximum number of threads to use
+      --no-mmap               Disable memory mapping
+      --no-names              Omit filenames in the output
+      --raw                   Write raw output bytes to stdout, rather than hex
+      --tag                   Output BSD-style checksums: BLAKE3 ([FILE]) = [HASH]
+  -c, --check                 Read BLAKE3 sums from the [FILE]s and check them
+      --quiet                 Skip printing OK for each checked file
+  -h, --help                  Print help (see more with '--help')
+  -V, --version               Print version
+```
+
+See also [this document about how the `--check` flag
+works](https://github.com/BLAKE3-team/BLAKE3/blob/master/b3sum/what_does_check_do.md).
+
+# Example
+
+Hash the file `foo.txt`:
+
+```bash
+b3sum foo.txt
+```
+
+Time hashing a gigabyte of data, to see how fast it is:
+
+```bash
+# Create a 1 GB file.
+head -c 1000000000 /dev/zero > /tmp/bigfile
+# Hash it with SHA-256.
+time openssl sha256 /tmp/bigfile
+# Hash it with BLAKE3.
+time b3sum /tmp/bigfile
+```
+
+
+# Installation
+
+Prebuilt binaries are available for Linux, Windows, and macOS (requiring
+the [unidentified developer
+workaround](https://support.apple.com/guide/mac-help/open-a-mac-app-from-an-unidentified-developer-mh40616/mac))
+on the [releases page](https://github.com/BLAKE3-team/BLAKE3/releases).
+If you've [installed Rust and
+Cargo](https://doc.rust-lang.org/cargo/getting-started/installation.html),
+you can also build `b3sum` yourself with:
+
+```
+cargo install b3sum
+```
+
+On Linux for example, Cargo will put the compiled binary in
+`~/.cargo/bin`. You might want to add that directory to your `$PATH`, or
+`rustup` might have done it for you when you installed Cargo.
+
+If you want to install directly from this directory, you can run `cargo
+install --path .`. Or you can just build with `cargo build --release`,
+which puts the binary at `./target/release/b3sum`.
diff --git a/thirdparty/blake3/b3sum/src/main.rs b/thirdparty/blake3/b3sum/src/main.rs
new file mode 100644
index 000000000..69a10c837
--- /dev/null
+++ b/thirdparty/blake3/b3sum/src/main.rs
@@ -0,0 +1,564 @@
+use anyhow::{bail, ensure};
+use clap::Parser;
+use std::cmp;
+use std::fs::File;
+use std::io;
+use std::io::prelude::*;
+use std::path::{Path, PathBuf};
+
+#[cfg(test)]
+mod unit_tests;
+
+const NAME: &str = "b3sum";
+
+const DERIVE_KEY_ARG: &str = "derive_key";
+const KEYED_ARG: &str = "keyed";
+const LENGTH_ARG: &str = "length";
+const NO_NAMES_ARG: &str = "no_names";
+const RAW_ARG: &str = "raw";
+const TAG_ARG: &str = "tag";
+const CHECK_ARG: &str = "check";
+
+#[derive(Parser)]
+#[command(version, max_term_width(100))]
+struct Inner {
+    /// Files to hash, or checkfiles to check
+    ///
+    /// When no file is given, or when - is given, read standard input.
+    file: Vec<PathBuf>,
+
+    /// Use the keyed mode, reading the 32-byte key from stdin
+    #[arg(long, requires("file"))]
+    keyed: bool,
+
+    /// Use the key derivation mode, with the given context string
+    ///
+    /// Cannot be used with --keyed.
+    #[arg(long, value_name("CONTEXT"), conflicts_with(KEYED_ARG))]
+    derive_key: Option<String>,
+
+    /// The number of output bytes, before hex encoding
+    #[arg(
+        short,
+        long,
+        default_value_t = blake3::OUT_LEN as u64,
+        value_name("LEN")
+    )]
+    length: u64,
+
+    /// The starting output byte offset, before hex encoding
+    #[arg(long, default_value_t = 0, value_name("SEEK"))]
+    seek: u64,
+
+    /// The maximum number of threads to use
+    ///
+    /// By default, this is the number of logical cores. If this flag is
+    /// omitted, or if its value is 0, RAYON_NUM_THREADS is also respected.
+    #[arg(long, value_name("NUM"))]
+    num_threads: Option<usize>,
+
+    /// Disable memory mapping
+    ///
+    /// Currently this also disables multithreading.
+    #[arg(long)]
+    no_mmap: bool,
+
+    /// Omit filenames in the output
+    #[arg(long)]
+    no_names: bool,
+
+    /// Write raw output bytes to stdout, rather than hex
+    ///
+    /// --no-names is implied. In this case, only a single input is allowed.
+    #[arg(long)]
+    raw: bool,
+
+    /// Output BSD-style checksums: BLAKE3 ([FILE]) = [HASH]
+    #[arg(long)]
+    tag: bool,
+
+    /// Read BLAKE3 sums from the [FILE]s and check them
+    #[arg(
+        short,
+        long,
+        conflicts_with(DERIVE_KEY_ARG),
+        conflicts_with(KEYED_ARG),
+        conflicts_with(LENGTH_ARG),
+        conflicts_with(RAW_ARG),
+        conflicts_with(TAG_ARG),
+        conflicts_with(NO_NAMES_ARG)
+    )]
+    check: bool,
+
+    /// Skip printing OK for each checked file
+    ///
+    /// Must be used with --check.
+    #[arg(long, requires(CHECK_ARG))]
+    quiet: bool,
+}
+
+struct Args {
+    inner: Inner,
+    file_args: Vec<PathBuf>,
+    base_hasher: blake3::Hasher,
+}
+
+impl Args {
+    fn parse() -> anyhow::Result<Self> {
+        // wild::args_os() is equivalent to std::env::args_os() on Unix,
+        // but on Windows it adds support for globbing.
+        let inner = Inner::parse_from(wild::args_os());
+        let file_args = if !inner.file.is_empty() {
+            inner.file.clone()
+        } else {
+            vec!["-".into()]
+        };
+        if inner.raw && file_args.len() > 1 {
+            bail!("Only one filename can be provided when using --raw");
+        }
+        let base_hasher = if inner.keyed {
+            // In keyed mode, since stdin is used for the key, we can't handle
+            // `-` arguments. Input::open handles that case below.
+            blake3::Hasher::new_keyed(&read_key_from_stdin()?)
+        } else if let Some(ref context) = inner.derive_key {
+            blake3::Hasher::new_derive_key(context)
+        } else {
+            blake3::Hasher::new()
+        };
+        Ok(Self {
+            inner,
+            file_args,
+            base_hasher,
+        })
+    }
+
+    fn num_threads(&self) -> Option<usize> {
+        self.inner.num_threads
+    }
+
+    fn check(&self) -> bool {
+        self.inner.check
+    }
+
+    fn raw(&self) -> bool {
+        self.inner.raw
+    }
+
+    fn tag(&self) -> bool {
+        self.inner.tag
+    }
+
+    fn no_mmap(&self) -> bool {
+        self.inner.no_mmap
+    }
+
+    fn no_names(&self) -> bool {
+        self.inner.no_names
+    }
+
+    fn len(&self) -> u64 {
+        self.inner.length
+    }
+
+    fn seek(&self) -> u64 {
+        self.inner.seek
+    }
+
+    fn keyed(&self) -> bool {
+        self.inner.keyed
+    }
+
+    fn quiet(&self) -> bool {
+        self.inner.quiet
+    }
+}
+
+fn hash_path(args: &Args, path: &Path) -> anyhow::Result<blake3::OutputReader> {
+    let mut hasher = args.base_hasher.clone();
+    if path == Path::new("-") {
+        if args.keyed() {
+            bail!("Cannot open `-` in keyed mode");
+        }
+        hasher.update_reader(io::stdin().lock())?;
+    } else if args.no_mmap() {
+        hasher.update_reader(File::open(path)?)?;
+    } else {
+        // The fast path: Try to mmap the file and hash it with multiple threads.
+        hasher.update_mmap_rayon(path)?;
+    }
+    let mut output_reader = hasher.finalize_xof();
+    output_reader.set_position(args.seek());
+    Ok(output_reader)
+}
+
+fn write_hex_output(mut output: blake3::OutputReader, args: &Args) -> anyhow::Result<()> {
+    // Encoding multiples of the 64 bytes is most efficient.
+    // TODO: This computes each output block twice when the --seek argument isn't a multiple of 64.
+    // We'll refactor all of this soon anyway, once SIMD optimizations are available for the XOF.
+    let mut len = args.len();
+    let mut block = [0; blake3::BLOCK_LEN];
+    while len > 0 {
+        output.fill(&mut block);
+        let hex_str = hex::encode(&block[..]);
+        let take_bytes = cmp::min(len, block.len() as u64);
+        print!("{}", &hex_str[..2 * take_bytes as usize]);
+        len -= take_bytes;
+    }
+    Ok(())
+}
+
+fn write_raw_output(output: blake3::OutputReader, args: &Args) -> anyhow::Result<()> {
+    let mut output = output.take(args.len());
+    let stdout = std::io::stdout();
+    let mut handler = stdout.lock();
+    std::io::copy(&mut output, &mut handler)?;
+
+    Ok(())
+}
+
+fn read_key_from_stdin() -> anyhow::Result<[u8; blake3::KEY_LEN]> {
+    let mut bytes = Vec::with_capacity(blake3::KEY_LEN + 1);
+    let n = std::io::stdin()
+        .lock()
+        .take(blake3::KEY_LEN as u64 + 1)
+        .read_to_end(&mut bytes)?;
+    if n < blake3::KEY_LEN {
+        bail!(
+            "expected {} key bytes from stdin, found {}",
+            blake3::KEY_LEN,
+            n,
+        )
+    } else if n > blake3::KEY_LEN {
+        bail!("read more than {} key bytes from stdin", blake3::KEY_LEN)
+    } else {
+        Ok(bytes[..blake3::KEY_LEN].try_into().unwrap())
+    }
+}
+
+struct FilepathString {
+    filepath_string: String,
+    is_escaped: bool,
+}
+
+// returns (string, did_escape)
+fn filepath_to_string(filepath: &Path) -> FilepathString {
+    let unicode_cow = filepath.to_string_lossy();
+    let mut filepath_string = unicode_cow.to_string();
+    // If we're on Windows, normalize backslashes to forward slashes. This
+    // avoids a lot of ugly escaping in the common case, and it makes
+    // checkfiles created on Windows more likely to be portable to Unix. It
+    // also allows us to set a blanket "no backslashes allowed in checkfiles on
+    // Windows" rule, rather than allowing a Unix backslash to potentially get
+    // interpreted as a directory separator on Windows.
+    if cfg!(windows) {
+        filepath_string = filepath_string.replace('\\', "/");
+    }
+    let mut is_escaped = false;
+    if filepath_string.contains(['\\', '\n', '\r']) {
+        filepath_string = filepath_string
+            .replace('\\', "\\\\")
+            .replace('\n', "\\n")
+            .replace('\r', "\\r");
+        is_escaped = true;
+    }
+    FilepathString {
+        filepath_string,
+        is_escaped,
+    }
+}
+
+fn hex_half_byte(c: char) -> anyhow::Result<u8> {
+    // The hex characters in the hash must be lowercase for now, though we
+    // could support uppercase too if we wanted to.
+    if '0' <= c && c <= '9' {
+        return Ok(c as u8 - '0' as u8);
+    }
+    if 'a' <= c && c <= 'f' {
+        return Ok(c as u8 - 'a' as u8 + 10);
+    }
+    bail!("Invalid hex");
+}
+
+// The `check` command is a security tool. That means it's much better for a
+// check to fail more often than it should (a false negative), than for a check
+// to ever succeed when it shouldn't (a false positive). By forbidding certain
+// characters in checked filepaths, we avoid a class of false positives where
+// two different filepaths can get confused with each other.
+fn check_for_invalid_characters(utf8_path: &str) -> anyhow::Result<()> {
+    // Null characters in paths should never happen, but they can result in a
+    // path getting silently truncated on Unix.
+    if utf8_path.contains('\0') {
+        bail!("Null character in path");
+    }
+    // Because we convert invalid UTF-8 sequences in paths to the Unicode
+    // replacement character, multiple different invalid paths can map to the
+    // same UTF-8 string.
+    if utf8_path.contains('�') {
+        bail!("Unicode replacement character in path");
+    }
+    // We normalize all Windows backslashes to forward slashes in our output,
+    // so the only natural way to get a backslash in a checkfile on Windows is
+    // to construct it on Unix and copy it over. (Or of course you could just
+    // doctor it by hand.) To avoid confusing this with a directory separator,
+    // we forbid backslashes entirely on Windows. Note that this check comes
+    // after unescaping has been done.
+    if cfg!(windows) && utf8_path.contains('\\') {
+        bail!("Backslash in path");
+    }
+    Ok(())
+}
+
+fn unescape(mut path: &str) -> anyhow::Result<String> {
+    let mut unescaped = String::with_capacity(2 * path.len());
+    while let Some(i) = path.find('\\') {
+        ensure!(i < path.len() - 1, "Invalid backslash escape");
+        unescaped.push_str(&path[..i]);
+        match path[i + 1..].chars().next().unwrap() {
+            // Anything other than a recognized escape sequence is an error.
+            'n' => unescaped.push_str("\n"),
+            'r' => unescaped.push_str("\r"),
+            '\\' => unescaped.push_str("\\"),
+            _ => bail!("Invalid backslash escape"),
+        }
+        path = &path[i + 2..];
+    }
+    unescaped.push_str(path);
+    Ok(unescaped)
+}
+
+#[derive(Debug)]
+struct ParsedCheckLine {
+    file_string: String,
+    is_escaped: bool,
+    file_path: PathBuf,
+    expected_hash: blake3::Hash,
+}
+
+fn split_untagged_check_line(line_after_slash: &str) -> Option<(&str, &str)> {
+    // Of the form "<hash>  <file>". The file might contain "  ", so we need to split from the
+    // left.
+    line_after_slash.split_once("  ")
+}
+
+fn split_tagged_check_line(line_after_slash: &str) -> Option<(&str, &str)> {
+    // Of the form "BLAKE3 (<file>) = <hash>". The file might contain ") = ", so we need to split
+    // from the *right*.
+    let prefix = "BLAKE3 (";
+    if !line_after_slash.starts_with(prefix) {
+        return None;
+    }
+    line_after_slash[prefix.len()..].rsplit_once(") = ")
+}
+
+fn parse_check_line(mut line: &str) -> anyhow::Result<ParsedCheckLine> {
+    // Trim off the trailing newlines, if any.
+    line = line.trim_end_matches(['\r', '\n']);
+    // If there's a backslash at the front of the line, that means we need to
+    // unescape the path below. This matches the behavior of e.g. md5sum.
+    let Some(first) = line.chars().next() else {
+        bail!("Empty line");
+    };
+    let line_after_slash;
+    let is_escaped;
+    if first == '\\' {
+        is_escaped = true;
+        line_after_slash = &line[1..];
+    } else {
+        is_escaped = false;
+        line_after_slash = line;
+    }
+
+    // Split the line. It might be "<hash>  <file>" or "BLAKE3 (<file>) = <hash>". The latter comes
+    // from the --tag flag.
+    let hash_hex;
+    let file_str;
+    if let Some((left, right)) = split_untagged_check_line(line_after_slash) {
+        hash_hex = left;
+        file_str = right;
+    } else if let Some((left, right)) = split_tagged_check_line(line_after_slash) {
+        file_str = left;
+        hash_hex = right;
+    } else {
+        bail!("Invalid check line format");
+    }
+
+    // Decode the hex hash.
+    ensure!(hash_hex.len() == 2 * blake3::OUT_LEN, "Invalid hash length");
+    let mut hex_chars = hash_hex.chars();
+    let mut hash_bytes = [0; blake3::OUT_LEN];
+    for byte in &mut hash_bytes {
+        let high_char = hex_chars.next().unwrap();
+        let low_char = hex_chars.next().unwrap();
+        *byte = 16 * hex_half_byte(high_char)? + hex_half_byte(low_char)?;
+    }
+    let expected_hash: blake3::Hash = hash_bytes.into();
+
+    // Unescape and validate the filepath.
+    let file_path_string = if is_escaped {
+        unescape(file_str)?
+    } else {
+        file_str.to_string()
+    };
+    ensure!(!file_path_string.is_empty(), "empty file path");
+    check_for_invalid_characters(&file_path_string)?;
+
+    Ok(ParsedCheckLine {
+        file_string: file_str.to_string(),
+        is_escaped,
+        file_path: file_path_string.into(),
+        expected_hash,
+    })
+}
+
+fn hash_one_input(path: &Path, args: &Args) -> anyhow::Result<()> {
+    let output = hash_path(args, path)?;
+    if args.raw() {
+        write_raw_output(output, args)?;
+        return Ok(());
+    }
+    if args.no_names() {
+        write_hex_output(output, args)?;
+        println!();
+        return Ok(());
+    }
+    let FilepathString {
+        filepath_string,
+        is_escaped,
+    } = filepath_to_string(path);
+    if is_escaped {
+        print!("\\");
+    }
+    if args.tag() {
+        print!("BLAKE3 ({}) = ", filepath_string);
+        write_hex_output(output, args)?;
+        println!();
+        return Ok(());
+    }
+    write_hex_output(output, args)?;
+    println!("  {}", filepath_string);
+    Ok(())
+}
+
+// Returns true for success. Having a boolean return value here, instead of
+// passing down the files_failed reference, makes it less likely that we might
+// forget to set it in some error condition.
+fn check_one_line(line: &str, args: &Args) -> bool {
+    let parse_result = parse_check_line(&line);
+    let ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = match parse_result {
+        Ok(parsed) => parsed,
+        Err(e) => {
+            eprintln!("{}: {}", NAME, e);
+            return false;
+        }
+    };
+    let file_string = if is_escaped {
+        "\\".to_string() + &file_string
+    } else {
+        file_string
+    };
+    let found_hash: blake3::Hash;
+    match hash_path(args, &file_path) {
+        Ok(mut output) => {
+            let mut found_hash_bytes = [0; blake3::OUT_LEN];
+            output.fill(&mut found_hash_bytes);
+            found_hash = found_hash_bytes.into();
+        }
+        Err(e) => {
+            println!("{}: FAILED ({})", file_string, e);
+            return false;
+        }
+    };
+    // This is a constant-time comparison.
+    if expected_hash == found_hash {
+        if !args.quiet() {
+            println!("{}: OK", file_string);
+        }
+        true
+    } else {
+        println!("{}: FAILED", file_string);
+        false
+    }
+}
+
+fn check_one_checkfile(path: &Path, args: &Args, files_failed: &mut u64) -> anyhow::Result<()> {
+    let mut file;
+    let stdin;
+    let mut stdin_lock;
+    let mut bufreader: io::BufReader<&mut dyn Read>;
+    if path == Path::new("-") {
+        stdin = io::stdin();
+        stdin_lock = stdin.lock();
+        bufreader = io::BufReader::new(&mut stdin_lock);
+    } else {
+        file = File::open(path)?;
+        bufreader = io::BufReader::new(&mut file);
+    }
+    let mut line = String::new();
+    loop {
+        line.clear();
+        let n = bufreader.read_line(&mut line)?;
+        if n == 0 {
+            return Ok(());
+        }
+        // check_one_line() prints errors and turns them into a success=false
+        // return, so it doesn't return a Result.
+        let success = check_one_line(&line, args);
+        if !success {
+            // We use `files_failed > 0` to indicate a mismatch, so it's important for correctness
+            // that it's impossible for this counter to overflow.
+            *files_failed = files_failed.saturating_add(1);
+        }
+    }
+}
+
+fn main() -> anyhow::Result<()> {
+    let args = Args::parse()?;
+    let mut thread_pool_builder = rayon_core::ThreadPoolBuilder::new();
+    if let Some(num_threads) = args.num_threads() {
+        thread_pool_builder = thread_pool_builder.num_threads(num_threads);
+    }
+    let thread_pool = thread_pool_builder.build()?;
+    thread_pool.install(|| {
+        let mut files_failed = 0u64;
+        // Note that file_args automatically includes `-` if nothing is given.
+        for path in &args.file_args {
+            if args.check() {
+                check_one_checkfile(path, &args, &mut files_failed)?;
+            } else {
+                // Errors encountered in hashing are tolerated and printed to
+                // stderr. This allows e.g. `b3sum *` to print errors for
+                // non-files and keep going. However, if we encounter any
+                // errors we'll still return non-zero at the end.
+                let result = hash_one_input(path, &args);
+                if let Err(e) = result {
+                    files_failed = files_failed.saturating_add(1);
+                    eprintln!("{}: {}: {}", NAME, path.to_string_lossy(), e);
+                }
+            }
+        }
+        if args.check() && files_failed > 0 {
+            eprintln!(
+                "{}: WARNING: {} computed checksum{} did NOT match",
+                NAME,
+                files_failed,
+                if files_failed == 1 { "" } else { "s" },
+            );
+        }
+        std::process::exit(if files_failed > 0 { 1 } else { 0 });
+    })
+}
+
+#[cfg(test)]
+mod test {
+    use clap::CommandFactory;
+
+    #[test]
+    fn test_args() {
+        crate::Inner::command().debug_assert();
+    }
+}
diff --git a/thirdparty/blake3/b3sum/src/unit_tests.rs b/thirdparty/blake3/b3sum/src/unit_tests.rs
new file mode 100644
index 000000000..75f672b4c
--- /dev/null
+++ b/thirdparty/blake3/b3sum/src/unit_tests.rs
@@ -0,0 +1,235 @@
+use std::path::Path;
+
+#[test]
+fn test_parse_check_line() {
+    // =========================
+    // ===== Success Cases =====
+    // =========================
+
+    // the basic case
+    let crate::ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = crate::parse_check_line(
+        "0909090909090909090909090909090909090909090909090909090909090909  foo",
+    )
+    .unwrap();
+    assert_eq!(expected_hash, blake3::Hash::from([0x09; 32]));
+    assert!(!is_escaped);
+    assert_eq!(file_string, "foo");
+    assert_eq!(file_path, Path::new("foo"));
+
+    // regular whitespace
+    let crate::ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = crate::parse_check_line(
+        "fafafafafafafafafafafafafafafafafafafafafafafafafafafafafafafafa   \t\r\n\n\r \t\r\n\n\r",
+    )
+    .unwrap();
+    assert_eq!(expected_hash, blake3::Hash::from([0xfa; 32]));
+    assert!(!is_escaped);
+    assert_eq!(file_string, " \t\r\n\n\r \t");
+    assert_eq!(file_path, Path::new(" \t\r\n\n\r \t"));
+
+    // path is one space
+    let crate::ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = crate::parse_check_line(
+        "4242424242424242424242424242424242424242424242424242424242424242   ",
+    )
+    .unwrap();
+    assert_eq!(expected_hash, blake3::Hash::from([0x42; 32]));
+    assert!(!is_escaped);
+    assert_eq!(file_string, " ");
+    assert_eq!(file_path, Path::new(" "));
+
+    // *Unescaped* backslashes. Note that this line does *not* start with a
+    // backslash, so something like "\" + "n" is interpreted as *two*
+    // characters. We forbid all backslashes on Windows, so this test is
+    // Unix-only.
+    if cfg!(not(windows)) {
+        let crate::ParsedCheckLine {
+            file_string,
+            is_escaped,
+            file_path,
+            expected_hash,
+        } = crate::parse_check_line(
+            "4343434343434343434343434343434343434343434343434343434343434343  fo\\a\\no",
+        )
+        .unwrap();
+        assert_eq!(expected_hash, blake3::Hash::from([0x43; 32]));
+        assert!(!is_escaped);
+        assert_eq!(file_string, "fo\\a\\no");
+        assert_eq!(file_path, Path::new("fo\\a\\no"));
+    }
+
+    // escaped newlines
+    let crate::ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = crate::parse_check_line(
+        "\\4444444444444444444444444444444444444444444444444444444444444444  fo\\r\\n\\n\\ro",
+    )
+    .unwrap();
+    assert_eq!(expected_hash, blake3::Hash::from([0x44; 32]));
+    assert!(is_escaped);
+    assert_eq!(file_string, "fo\\r\\n\\n\\ro");
+    assert_eq!(file_path, Path::new("fo\r\n\n\ro"));
+
+    // Escaped newline and backslash. Again because backslash is not allowed on
+    // Windows, this test is Unix-only.
+    if cfg!(not(windows)) {
+        let crate::ParsedCheckLine {
+            file_string,
+            is_escaped,
+            file_path,
+            expected_hash,
+        } = crate::parse_check_line(
+            "\\4545454545454545454545454545454545454545454545454545454545454545  fo\\n\\\\o",
+        )
+        .unwrap();
+        assert_eq!(expected_hash, blake3::Hash::from([0x45; 32]));
+        assert!(is_escaped);
+        assert_eq!(file_string, "fo\\n\\\\o");
+        assert_eq!(file_path, Path::new("fo\n\\o"));
+    }
+
+    // non-ASCII path
+    let crate::ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = crate::parse_check_line(
+        "4646464646464646464646464646464646464646464646464646464646464646  否认",
+    )
+    .unwrap();
+    assert_eq!(expected_hash, blake3::Hash::from([0x46; 32]));
+    assert!(!is_escaped);
+    assert_eq!(file_string, "否认");
+    assert_eq!(file_path, Path::new("否认"));
+
+    // untagged separator "  " in the file name
+    let crate::ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = crate::parse_check_line(
+        "4747474747474747474747474747474747474747474747474747474747474747  foo  bar",
+    )
+    .unwrap();
+    assert_eq!(expected_hash, blake3::Hash::from([0x47; 32]));
+    assert!(!is_escaped);
+    assert_eq!(file_string, "foo  bar");
+    assert_eq!(file_path, Path::new("foo  bar"));
+
+    // tagged separator ") = " in the file name
+    let crate::ParsedCheckLine {
+        file_string,
+        is_escaped,
+        file_path,
+        expected_hash,
+    } = crate::parse_check_line(
+        "BLAKE3 (foo) = bar) = 4848484848484848484848484848484848484848484848484848484848484848",
+    )
+    .unwrap();
+    assert_eq!(expected_hash, blake3::Hash::from([0x48; 32]));
+    assert!(!is_escaped);
+    assert_eq!(file_string, "foo) = bar");
+    assert_eq!(file_path, Path::new("foo) = bar"));
+
+    // =========================
+    // ===== Failure Cases =====
+    // =========================
+
+    // too short
+    crate::parse_check_line("").unwrap_err();
+    crate::parse_check_line("0").unwrap_err();
+    crate::parse_check_line("00").unwrap_err();
+    crate::parse_check_line("0000000000000000000000000000000000000000000000000000000000000000")
+        .unwrap_err();
+    crate::parse_check_line("0000000000000000000000000000000000000000000000000000000000000000  ")
+        .unwrap_err();
+
+    // not enough spaces
+    crate::parse_check_line("0000000000000000000000000000000000000000000000000000000000000000 foo")
+        .unwrap_err();
+
+    // capital letter hex
+    crate::parse_check_line(
+        "AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA  foo",
+    )
+    .unwrap_err();
+
+    // non-hex hex
+    crate::parse_check_line(
+        "xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx  foo",
+    )
+    .unwrap_err();
+
+    // non-ASCII hex
+    crate::parse_check_line("你好, 我叫杰克. 认识你很高兴. 要不要吃个香蕉?  foo").unwrap_err();
+
+    // invalid escape sequence
+    crate::parse_check_line(
+        "\\0000000000000000000000000000000000000000000000000000000000000000  fo\\o",
+    )
+    .unwrap_err();
+
+    // truncated escape sequence
+    crate::parse_check_line(
+        "\\0000000000000000000000000000000000000000000000000000000000000000  foo\\",
+    )
+    .unwrap_err();
+
+    // null char
+    crate::parse_check_line(
+        "0000000000000000000000000000000000000000000000000000000000000000  fo\0o",
+    )
+    .unwrap_err();
+
+    // Unicode replacement char
+    crate::parse_check_line(
+        "0000000000000000000000000000000000000000000000000000000000000000  fo�o",
+    )
+    .unwrap_err();
+
+    // On Windows only, backslashes are not allowed, escaped or otherwise.
+    if cfg!(windows) {
+        crate::parse_check_line(
+            "0000000000000000000000000000000000000000000000000000000000000000  fo\\o",
+        )
+        .unwrap_err();
+        crate::parse_check_line(
+            "\\0000000000000000000000000000000000000000000000000000000000000000  fo\\\\o",
+        )
+        .unwrap_err();
+    }
+}
+
+#[test]
+fn test_filepath_to_string() {
+    let output = crate::filepath_to_string(Path::new("foo"));
+    assert_eq!(output.filepath_string, "foo");
+    assert!(!output.is_escaped);
+
+    let output = crate::filepath_to_string(Path::new("f\\ \t\r\noo"));
+    if cfg!(windows) {
+        // We normalize backslashes to forward slashes on Windows.
+        assert_eq!(output.filepath_string, "f/ \t\\r\\noo");
+    } else {
+        assert_eq!(output.filepath_string, "f\\\\ \t\\r\\noo");
+    }
+    assert!(output.is_escaped);
+}
diff --git a/thirdparty/blake3/b3sum/tests/cli_tests.rs b/thirdparty/blake3/b3sum/tests/cli_tests.rs
new file mode 100644
index 000000000..3fe173d6b
--- /dev/null
+++ b/thirdparty/blake3/b3sum/tests/cli_tests.rs
@@ -0,0 +1,680 @@
+use duct::cmd;
+use std::ffi::OsString;
+use std::fs;
+use std::io::prelude::*;
+use std::path::PathBuf;
+
+pub fn b3sum_exe() -> PathBuf {
+    env!("CARGO_BIN_EXE_b3sum").into()
+}
+
+#[test]
+fn test_hash_one() {
+    let expected = format!("{}  -", blake3::hash(b"foo").to_hex());
+    let output = cmd!(b3sum_exe()).stdin_bytes("foo").read().unwrap();
+    assert_eq!(&*expected, output);
+}
+
+#[test]
+fn test_hash_one_tag() {
+    let expected = format!("BLAKE3 (-) = {}", blake3::hash(b"foo").to_hex());
+    let output = cmd!(b3sum_exe(), "--tag")
+        .stdin_bytes("foo")
+        .read()
+        .unwrap();
+    assert_eq!(&*expected, output);
+}
+
+#[test]
+fn test_hash_one_raw() {
+    let expected = blake3::hash(b"foo").as_bytes().to_owned();
+    let output = cmd!(b3sum_exe(), "--raw")
+        .stdin_bytes("foo")
+        .stdout_capture()
+        .run()
+        .unwrap()
+        .stdout;
+    assert_eq!(expected, output.as_slice());
+}
+
+#[test]
+fn test_hash_many() {
+    let dir = tempfile::tempdir().unwrap();
+    let file1 = dir.path().join("file1");
+    fs::write(&file1, b"foo").unwrap();
+    let file2 = dir.path().join("file2");
+    fs::write(&file2, b"bar").unwrap();
+
+    let output = cmd!(b3sum_exe(), &file1, &file2).read().unwrap();
+    let foo_hash = blake3::hash(b"foo");
+    let bar_hash = blake3::hash(b"bar");
+    let expected = format!(
+        "{}  {}\n{}  {}",
+        foo_hash.to_hex(),
+        // account for slash normalization on Windows
+        file1.to_string_lossy().replace("\\", "/"),
+        bar_hash.to_hex(),
+        file2.to_string_lossy().replace("\\", "/"),
+    );
+    assert_eq!(expected, output);
+
+    let output_no_names = cmd!(b3sum_exe(), "--no-names", &file1, &file2)
+        .read()
+        .unwrap();
+    let expected_no_names = format!("{}\n{}", foo_hash.to_hex(), bar_hash.to_hex(),);
+    assert_eq!(expected_no_names, output_no_names);
+}
+
+#[test]
+fn test_hash_many_tag() {
+    let dir = tempfile::tempdir().unwrap();
+    let file1 = dir.path().join("file1");
+    fs::write(&file1, b"foo").unwrap();
+    let file2 = dir.path().join("file2");
+    fs::write(&file2, b"bar").unwrap();
+
+    let output = cmd!(b3sum_exe(), "--tag", &file1, &file2).read().unwrap();
+    let foo_hash = blake3::hash(b"foo");
+    let bar_hash = blake3::hash(b"bar");
+    let expected = format!(
+        "BLAKE3 ({}) = {}\nBLAKE3 ({}) = {}",
+        // account for slash normalization on Windows
+        file1.to_string_lossy().replace("\\", "/"),
+        foo_hash.to_hex(),
+        file2.to_string_lossy().replace("\\", "/"),
+        bar_hash.to_hex(),
+    );
+    assert_eq!(expected, output);
+}
+
+#[test]
+fn test_missing_files() {
+    let dir = tempfile::tempdir().unwrap();
+    let file1 = dir.path().join("file1");
+    fs::write(&file1, b"foo").unwrap();
+    let file2 = dir.path().join("file2");
+    fs::write(&file2, b"bar").unwrap();
+
+    let output = cmd!(b3sum_exe(), "file1", "missing_file", "file2")
+        .dir(dir.path())
+        .stdout_capture()
+        .stderr_capture()
+        .unchecked()
+        .run()
+        .unwrap();
+    assert!(!output.status.success());
+
+    let foo_hash = blake3::hash(b"foo");
+    let bar_hash = blake3::hash(b"bar");
+    let expected_stdout = format!(
+        "{}  file1\n{}  file2\n",
+        foo_hash.to_hex(),
+        bar_hash.to_hex(),
+    );
+    assert_eq!(expected_stdout.as_bytes(), &output.stdout[..]);
+
+    let bing_error = fs::File::open(dir.path().join("missing_file")).unwrap_err();
+    let expected_stderr = format!("b3sum: missing_file: {}\n", bing_error.to_string());
+    assert_eq!(expected_stderr.as_bytes(), &output.stderr[..]);
+}
+
+#[test]
+fn test_hash_length_and_seek() {
+    let mut expected = [0; 100];
+    blake3::Hasher::new()
+        .update(b"foo")
+        .finalize_xof()
+        .fill(&mut expected);
+    let output = cmd!(b3sum_exe(), "--raw", "--length=100")
+        .stdin_bytes("foo")
+        .stdout_capture()
+        .run()
+        .unwrap()
+        .stdout;
+    assert_eq!(expected[..], output);
+
+    let short_output = cmd!(b3sum_exe(), "--raw", "--length=99")
+        .stdin_bytes("foo")
+        .stdout_capture()
+        .run()
+        .unwrap()
+        .stdout;
+    assert_eq!(expected[..99], short_output);
+
+    let seek1_output = cmd!(b3sum_exe(), "--raw", "--length=99", "--seek=1")
+        .stdin_bytes("foo")
+        .stdout_capture()
+        .run()
+        .unwrap()
+        .stdout;
+    assert_eq!(expected[1..], seek1_output);
+
+    let seek99_output = cmd!(b3sum_exe(), "--raw", "--length=1", "--seek=99")
+        .stdin_bytes("foo")
+        .stdout_capture()
+        .run()
+        .unwrap()
+        .stdout;
+    assert_eq!(expected[99..], seek99_output);
+}
+
+#[test]
+fn test_keyed() {
+    let key = [42; blake3::KEY_LEN];
+    let f = tempfile::NamedTempFile::new().unwrap();
+    f.as_file().write_all(b"foo").unwrap();
+    f.as_file().flush().unwrap();
+    let expected = blake3::keyed_hash(&key, b"foo").to_hex();
+    let output = cmd!(b3sum_exe(), "--keyed", "--no-names", f.path())
+        .stdin_bytes(&key[..])
+        .read()
+        .unwrap();
+    assert_eq!(&*expected, &*output);
+
+    // Make sure that keys of the wrong length lead to errors.
+    for bad_length in [0, 1, blake3::KEY_LEN - 1, blake3::KEY_LEN + 1] {
+        dbg!(bad_length);
+        let output = cmd!(b3sum_exe(), "--keyed", f.path())
+            .stdin_bytes(vec![0; bad_length])
+            .stdout_capture()
+            .stderr_capture()
+            .unchecked()
+            .run()
+            .unwrap();
+        assert!(!output.status.success());
+        assert!(output.stdout.is_empty());
+        // Make sure the error message is relevant.
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        assert!(stderr.contains("key bytes"));
+    }
+}
+
+#[test]
+fn test_derive_key() {
+    let context = "BLAKE3 2019-12-28 10:28:41 example context";
+    let f = tempfile::NamedTempFile::new().unwrap();
+    f.as_file().write_all(b"key material").unwrap();
+    f.as_file().flush().unwrap();
+    let expected = hex::encode(blake3::derive_key(context, b"key material"));
+    let output = cmd!(b3sum_exe(), "--derive-key", context, "--no-names", f.path())
+        .read()
+        .unwrap();
+    assert_eq!(&*expected, &*output);
+}
+
+#[test]
+fn test_no_mmap() {
+    let f = tempfile::NamedTempFile::new().unwrap();
+    f.as_file().write_all(b"foo").unwrap();
+    f.as_file().flush().unwrap();
+
+    let expected = blake3::hash(b"foo").to_hex();
+    let output = cmd!(b3sum_exe(), "--no-mmap", "--no-names", f.path())
+        .read()
+        .unwrap();
+    assert_eq!(&*expected, &*output);
+}
+
+#[test]
+fn test_length_without_value_is_an_error() {
+    let result = cmd!(b3sum_exe(), "--length")
+        .stdin_bytes("foo")
+        .stderr_capture()
+        .run();
+    assert!(result.is_err());
+}
+
+#[test]
+fn test_raw_with_multi_files_is_an_error() {
+    let f1 = tempfile::NamedTempFile::new().unwrap();
+    let f2 = tempfile::NamedTempFile::new().unwrap();
+
+    // Make sure it doesn't error with just one file
+    let result = cmd!(b3sum_exe(), "--raw", f1.path()).stdout_capture().run();
+    assert!(result.is_ok());
+
+    // Make sure it errors when both file are passed
+    let result = cmd!(b3sum_exe(), "--raw", f1.path(), f2.path())
+        .stderr_capture()
+        .run();
+    assert!(result.is_err());
+}
+
+#[test]
+#[cfg(unix)]
+fn test_newline_and_backslash_escaping_on_unix() {
+    let empty_hash = blake3::hash(b"").to_hex();
+    let dir = tempfile::tempdir().unwrap();
+    fs::create_dir(dir.path().join("subdir")).unwrap();
+    let names = [
+        "abcdef",
+        "abc\ndef",
+        "abc\\def",
+        "abc\rdef",
+        "abc\r\ndef",
+        "subdir/foo",
+    ];
+    let mut paths = Vec::new();
+    for name in &names {
+        let path = dir.path().join(name);
+        println!("creating file at {:?}", path);
+        fs::write(&path, b"").unwrap();
+        paths.push(path);
+    }
+    let output = cmd(b3sum_exe(), &names).dir(dir.path()).read().unwrap();
+    let expected = format!(
+        "\
+{0}  abcdef
+\\{0}  abc\\ndef
+\\{0}  abc\\\\def
+\\{0}  abc\\rdef
+\\{0}  abc\\r\\ndef
+{0}  subdir/foo",
+        empty_hash,
+    );
+    println!("output");
+    println!("======");
+    println!("{}", output);
+    println!();
+    println!("expected");
+    println!("========");
+    println!("{}", expected);
+    println!();
+    assert_eq!(expected, output);
+}
+
+#[test]
+#[cfg(windows)]
+fn test_slash_normalization_on_windows() {
+    let empty_hash = blake3::hash(b"").to_hex();
+    let dir = tempfile::tempdir().unwrap();
+    fs::create_dir(dir.path().join("subdir")).unwrap();
+    // Note that filenames can't contain newlines or backslashes on Windows, so
+    // we don't test escaping here. We only test forward slash and backslash as
+    // directory separators.
+    let names = ["abcdef", "subdir/foo", "subdir\\bar"];
+    let mut paths = Vec::new();
+    for name in &names {
+        let path = dir.path().join(name);
+        println!("creating file at {:?}", path);
+        fs::write(&path, b"").unwrap();
+        paths.push(path);
+    }
+    let output = cmd(b3sum_exe(), &names).dir(dir.path()).read().unwrap();
+    let expected = format!(
+        "\
+{0}  abcdef
+{0}  subdir/foo
+{0}  subdir/bar",
+        empty_hash,
+    );
+    println!("output");
+    println!("======");
+    println!("{}", output);
+    println!();
+    println!("expected");
+    println!("========");
+    println!("{}", expected);
+    println!();
+    assert_eq!(expected, output);
+}
+
+#[test]
+#[cfg(unix)]
+fn test_invalid_unicode_on_unix() {
+    use std::os::unix::ffi::OsStringExt;
+
+    let empty_hash = blake3::hash(b"").to_hex();
+    let dir = tempfile::tempdir().unwrap();
+    let names = ["abcdef".into(), OsString::from_vec(b"abc\xffdef".to_vec())];
+    let mut paths = Vec::new();
+    for name in &names {
+        let path = dir.path().join(name);
+        println!("creating file at {:?}", path);
+        // Note: Some operating systems, macOS in particular, simply don't
+        // allow invalid Unicode in filenames. On those systems, this write
+        // will fail. That's fine, we'll just short-circuit this test in that
+        // case. But assert that at least Linux allows this.
+        let write_result = fs::write(&path, b"");
+        if cfg!(target_os = "linux") {
+            write_result.expect("Linux should allow invalid Unicode");
+        } else if write_result.is_err() {
+            return;
+        }
+        paths.push(path);
+    }
+    let output = cmd(b3sum_exe(), &names).dir(dir.path()).read().unwrap();
+    let expected = format!(
+        "\
+{0}  abcdef
+{0}  abc�def",
+        empty_hash,
+    );
+    println!("output");
+    println!("======");
+    println!("{}", output);
+    println!();
+    println!("expected");
+    println!("========");
+    println!("{}", expected);
+    println!();
+    assert_eq!(expected, output);
+}
+
+#[test]
+#[cfg(windows)]
+fn test_invalid_unicode_on_windows() {
+    use std::os::windows::ffi::OsStringExt;
+
+    let empty_hash = blake3::hash(b"").to_hex();
+    let dir = tempfile::tempdir().unwrap();
+    let surrogate_char = 0xDC00;
+    let bad_unicode_wchars = [
+        'a' as u16,
+        'b' as u16,
+        'c' as u16,
+        surrogate_char,
+        'd' as u16,
+        'e' as u16,
+        'f' as u16,
+    ];
+    let bad_osstring = OsString::from_wide(&bad_unicode_wchars);
+    let names = ["abcdef".into(), bad_osstring];
+    let mut paths = Vec::new();
+    for name in &names {
+        let path = dir.path().join(name);
+        println!("creating file at {:?}", path);
+        fs::write(&path, b"").unwrap();
+        paths.push(path);
+    }
+    let output = cmd(b3sum_exe(), &names).dir(dir.path()).read().unwrap();
+    let expected = format!(
+        "\
+{0}  abcdef
+{0}  abc�def",
+        empty_hash,
+    );
+    println!("output");
+    println!("======");
+    println!("{}", output);
+    println!();
+    println!("expected");
+    println!("========");
+    println!("{}", expected);
+    println!();
+    assert_eq!(expected, output);
+}
+
+#[test]
+fn test_check() {
+    // Make a directory full of files, and make sure the b3sum output in that
+    // directory is what we expect.
+    let a_hash = blake3::hash(b"a").to_hex();
+    let b_hash = blake3::hash(b"b").to_hex();
+    let cd_hash = blake3::hash(b"cd").to_hex();
+    for tagged in [false, true] {
+        let dir = tempfile::tempdir().unwrap();
+        fs::write(dir.path().join("a"), b"a").unwrap();
+        fs::write(dir.path().join("b"), b"b").unwrap();
+        fs::create_dir(dir.path().join("c")).unwrap();
+        fs::write(dir.path().join("c/d"), b"cd").unwrap();
+        dbg!(tagged);
+        let mut args = vec!["a", "b", "c/d"];
+        if tagged {
+            args.push("--tag");
+        }
+        let output = cmd(b3sum_exe(), args)
+            .dir(dir.path())
+            .stdout_capture()
+            .stderr_capture()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let expected_checkfile = if tagged {
+            format!(
+                "BLAKE3 (a) = {}\n\
+                 BLAKE3 (b) = {}\n\
+                 BLAKE3 (c/d) = {}\n",
+                a_hash, b_hash, cd_hash,
+            )
+        } else {
+            format!(
+                "{}  a\n\
+                 {}  b\n\
+                 {}  c/d\n",
+                a_hash, b_hash, cd_hash,
+            )
+        };
+        dbg!(&expected_checkfile);
+        assert_eq!(expected_checkfile, stdout);
+        assert_eq!("", stderr);
+
+        // Now use the output we just validated as a checkfile, passed to stdin.
+        let output = cmd!(b3sum_exe(), "--check")
+            .stdin_bytes(expected_checkfile.as_bytes())
+            .dir(dir.path())
+            .stdout_capture()
+            .stderr_capture()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let expected_check_output = "\
+         a: OK\n\
+         b: OK\n\
+         c/d: OK\n";
+        assert_eq!(expected_check_output, stdout);
+        assert_eq!("", stderr);
+
+        // Check the same file, but with Windows-style newlines.
+        let windows_style = expected_checkfile.replace("\n", "\r\n");
+        let output = cmd!(b3sum_exe(), "--check")
+            .stdin_bytes(windows_style.as_bytes())
+            .dir(dir.path())
+            .stdout_capture()
+            .stderr_capture()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let expected_check_output = "\
+         a: OK\n\
+         b: OK\n\
+         c/d: OK\n";
+        assert_eq!(expected_check_output, stdout);
+        assert_eq!("", stderr);
+
+        // Now pass the same checkfile twice on the command line just for fun.
+        let checkfile_path = dir.path().join("checkfile");
+        fs::write(&checkfile_path, &expected_checkfile).unwrap();
+        let output = cmd!(b3sum_exe(), "--check", &checkfile_path, &checkfile_path)
+            .dir(dir.path())
+            .stdout_capture()
+            .stderr_capture()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let mut double_check_output = String::new();
+        double_check_output.push_str(&expected_check_output);
+        double_check_output.push_str(&expected_check_output);
+        assert_eq!(double_check_output, stdout);
+        assert_eq!("", stderr);
+
+        // Corrupt one of the files and check again.
+        fs::write(dir.path().join("b"), b"CORRUPTION").unwrap();
+        let output = cmd!(b3sum_exe(), "--check", &checkfile_path)
+            .dir(dir.path())
+            .stdout_capture()
+            .stderr_capture()
+            .unchecked()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let expected_check_failure = "\
+        a: OK\n\
+        b: FAILED\n\
+        c/d: OK\n";
+        assert!(!output.status.success());
+        assert_eq!(expected_check_failure, stdout);
+        assert_eq!(
+            "b3sum: WARNING: 1 computed checksum did NOT match\n",
+            stderr,
+        );
+
+        // Delete one of the files and check again.
+        fs::remove_file(dir.path().join("b")).unwrap();
+        let open_file_error = fs::File::open(dir.path().join("b")).unwrap_err();
+        let output = cmd!(b3sum_exe(), "--check", &checkfile_path)
+            .dir(dir.path())
+            .stdout_capture()
+            .stderr_capture()
+            .unchecked()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let expected_check_failure = format!(
+            "a: OK\n\
+         b: FAILED ({})\n\
+         c/d: OK\n",
+            open_file_error,
+        );
+        assert!(!output.status.success());
+        assert_eq!(expected_check_failure, stdout);
+        assert_eq!(
+            "b3sum: WARNING: 1 computed checksum did NOT match\n",
+            stderr,
+        );
+
+        // Confirm that --quiet suppresses the OKs but not the FAILEDs.
+        let output = cmd!(b3sum_exe(), "--check", "--quiet", &checkfile_path)
+            .dir(dir.path())
+            .stdout_capture()
+            .stderr_capture()
+            .unchecked()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let expected_check_failure = format!("b: FAILED ({})\n", open_file_error);
+        assert!(!output.status.success());
+        assert_eq!(expected_check_failure, stdout);
+        assert_eq!(
+            "b3sum: WARNING: 1 computed checksum did NOT match\n",
+            stderr,
+        );
+    }
+}
+
+#[test]
+fn test_check_invalid_characters() {
+    // Check that a null character in the path fails.
+    let output = cmd!(b3sum_exe(), "--check")
+        .stdin_bytes("0000000000000000000000000000000000000000000000000000000000000000  \0")
+        .stdout_capture()
+        .stderr_capture()
+        .unchecked()
+        .run()
+        .unwrap();
+    let stdout = std::str::from_utf8(&output.stdout).unwrap();
+    let stderr = std::str::from_utf8(&output.stderr).unwrap();
+    let expected_stderr = "\
+        b3sum: Null character in path\n\
+        b3sum: WARNING: 1 computed checksum did NOT match\n";
+    assert!(!output.status.success());
+    assert_eq!("", stdout);
+    assert_eq!(expected_stderr, stderr);
+
+    // Check that a Unicode replacement character in the path fails.
+    let output = cmd!(b3sum_exe(), "--check")
+        .stdin_bytes("0000000000000000000000000000000000000000000000000000000000000000  �")
+        .stdout_capture()
+        .stderr_capture()
+        .unchecked()
+        .run()
+        .unwrap();
+    let stdout = std::str::from_utf8(&output.stdout).unwrap();
+    let stderr = std::str::from_utf8(&output.stderr).unwrap();
+    let expected_stderr = "\
+        b3sum: Unicode replacement character in path\n\
+        b3sum: WARNING: 1 computed checksum did NOT match\n";
+    assert!(!output.status.success());
+    assert_eq!("", stdout);
+    assert_eq!(expected_stderr, stderr);
+
+    // Check that an invalid escape sequence in the path fails.
+    let output = cmd!(b3sum_exe(), "--check")
+        .stdin_bytes("\\0000000000000000000000000000000000000000000000000000000000000000  \\a")
+        .stdout_capture()
+        .stderr_capture()
+        .unchecked()
+        .run()
+        .unwrap();
+    let stdout = std::str::from_utf8(&output.stdout).unwrap();
+    let stderr = std::str::from_utf8(&output.stderr).unwrap();
+    let expected_stderr = "\
+        b3sum: Invalid backslash escape\n\
+        b3sum: WARNING: 1 computed checksum did NOT match\n";
+    assert!(!output.status.success());
+    assert_eq!("", stdout);
+    assert_eq!(expected_stderr, stderr);
+
+    // Windows also forbids literal backslashes. Check for that if and only if
+    // we're on Windows.
+    if cfg!(windows) {
+        let output = cmd!(b3sum_exe(), "--check")
+            .stdin_bytes("0000000000000000000000000000000000000000000000000000000000000000  \\")
+            .stdout_capture()
+            .stderr_capture()
+            .unchecked()
+            .run()
+            .unwrap();
+        let stdout = std::str::from_utf8(&output.stdout).unwrap();
+        let stderr = std::str::from_utf8(&output.stderr).unwrap();
+        let expected_stderr = "\
+            b3sum: Backslash in path\n\
+            b3sum: WARNING: 1 computed checksum did NOT match\n";
+        assert!(!output.status.success());
+        assert_eq!("", stdout);
+        assert_eq!(expected_stderr, stderr);
+    }
+}
+
+#[test]
+fn test_globbing() {
+    // On Unix, globbing is provided by the shell. On Windows, globbing is
+    // provided by us, using the `wild` crate.
+    let dir = tempfile::tempdir().unwrap();
+    let file1 = dir.path().join("file1");
+    fs::write(&file1, b"foo").unwrap();
+    let file2 = dir.path().join("file2");
+    fs::write(&file2, b"bar").unwrap();
+
+    let foo_hash = blake3::hash(b"foo");
+    let bar_hash = blake3::hash(b"bar");
+    // NOTE: This assumes that the glob will be expanded in alphabetical order,
+    //       to "file1 file2" rather than "file2 file1". So far, this seems to
+    //       be true (guaranteed?) of Unix shell behavior, and true in practice
+    //       with the `wild` crate on Windows. It's possible that this could
+    //       start failing in the future, though, or on some unknown platform.
+    //       If that ever happens, we'll need to relax this test somehow,
+    //       probably by just testing for both possible outputs. I'm not
+    //       handling that case in advance, though, because I'd prefer to hear
+    //       about it if it comes up.
+    let expected = format!("{}  file1\n{}  file2", foo_hash.to_hex(), bar_hash.to_hex());
+
+    let star_command = format!("{} *", b3sum_exe().to_str().unwrap());
+    let (exe, c_flag) = if cfg!(windows) {
+        ("cmd.exe", "/C")
+    } else {
+        ("/bin/sh", "-c")
+    };
+    let output = cmd!(exe, c_flag, star_command)
+        .dir(dir.path())
+        .read()
+        .unwrap();
+    assert_eq!(expected, output);
+}
diff --git a/thirdparty/blake3/b3sum/what_does_check_do.md b/thirdparty/blake3/b3sum/what_does_check_do.md
new file mode 100644
index 000000000..387c490bc
--- /dev/null
+++ b/thirdparty/blake3/b3sum/what_does_check_do.md
@@ -0,0 +1,176 @@
+# How does `b3sum --check` behave exactly?<br>or: Are filepaths...text?
+
+Most of the time, `b3sum --check` is a drop-in replacement for `md5sum --check`
+and other Coreutils hashing tools. It consumes a checkfile (the output of a
+regular `b3sum` command), re-hashes all the files listed there, and returns
+success if all of those hashes are still correct. What makes this more
+complicated than it might seem, is that representing filepaths as text means we
+need to consider many possible edge cases of unrepresentable filepaths. This
+document describes all of these edge cases in detail.
+
+## The simple case
+
+Here's the result of running `b3sum a b c/d` in a directory that contains
+those three files:
+
+```bash
+$ echo hi > a
+$ echo lo > b
+$ mkdir c
+$ echo stuff > c/d
+$ b3sum a b c/d
+0b8b60248fad7ac6dfac221b7e01a8b91c772421a15b387dd1fb2d6a94aee438  a
+6ae4a57bbba24f79c461d30bcb4db973b9427d9207877e34d2d74528daa84115  b
+2d477356c962e54784f1c5dc5297718d92087006f6ee96b08aeaf7f3cd252377  c/d
+```
+
+If we pipe that output into `b3sum --check`, it will exit with status zero
+(success) and print:
+
+```bash
+$ b3sum a b c/d | b3sum --check
+a: OK
+b: OK
+c/d: OK
+```
+
+If we delete `b` and change the contents of `c/d`, and then use the same
+checkfile as above, `b3sum --check` will exit with a non-zero status (failure)
+and print:
+
+```bash
+$ b3sum a b c/d > checkfile
+$ rm b
+$ echo more stuff >> c/d
+$ b3sum --check checkfile
+a: OK
+b: FAILED (No such file or directory (os error 2))
+c/d: FAILED
+```
+
+In these typical cases, `b3sum` and `md5sum` have identical output for success
+and very similar output for failure.
+
+## Escaping newlines and backslashes
+
+Since the checkfile format (the regular output format of `b3sum`) is
+newline-separated text, we need to worry about what happens when a filepath
+contains a newline, or worse. Suppose we create a file named `x[newline]x`
+(3 characters). One way to create such a file is with a Python one-liner like
+this:
+
+```python
+>>> open("x\nx", "w")
+```
+
+Here's what happens when we hash that file with `b3sum`:
+
+```bash
+$ b3sum x*
+\af1349b9f5f9a1a6a0404dea36dcc9499bcb25c9adc112b7cc9a93cae41f3262  x\nx
+```
+
+Notice two things. First, `b3sum` puts a single `\` character at the front of
+the line. This indicates that the filepath contains escape sequences that
+`b3sum --check` will need to unescape. Then, `b3sum` replaces the newline
+character in the filepath with the two-character escape sequence `\n`.
+Similarly, if the filepath contained carriage returns or backslashes, `b3sum`
+would escape those as `\r` and `\\` in the output. So far, all of this behavior
+is still identical to `md5sum`. (Note: Coreutils [introduced `\r`
+escaping](https://github.com/coreutils/coreutils/commit/ed1c58427d574fb4ff0cb8f915eb0d554000ceeb)
+in v9.0, September 2021.)
+
+## Invalid Unicode
+
+This is where `b3sum` and `md5sum` diverge. Apart from the newline and
+backslash escapes described above, `md5sum` copies all other filepath bytes
+verbatim to its output. That means its output encoding is "ASCII plus whatever
+bytes we got from the command line". This creates two problems:
+
+1. Printing something that isn't UTF-8 is kind of gross.
+2. Windows support.
+
+What's the deal with Windows? To start with, there's a fundamental difference
+in how Unix and Windows represent filepaths. Unix filepaths are "usually UTF-8"
+and Windows filepaths are "usually UTF-16". That means that a file named `abc`
+is typically represented as the bytes `[97, 98, 99]` on Unix and as the bytes
+`[97, 0, 98, 0, 99, 0]` on Windows. The `md5sum` approach won't work if we plan
+on creating a checkfile on Unix and checking it on Windows, or vice versa.
+
+A more portable approach is to convert platform-specific bytes into some
+consistent Unicode encoding. (In practice this is going to be UTF-8, but in
+theory it could be anything.) Then when `--check` needs to open a file, we
+convert the Unicode representation back into platform-specific bytes. This
+makes important common cases like `abc`, and in fact even `abc[newline]def`,
+work as expected. Great!
+
+But...what did we mean above when we said *usually* UTF-8 and *usually* UTF-16?
+It turns out that not every possible sequence of bytes is valid UTF-8, and not
+every possible sequence of 16-bit wide chars is valid UTF-16. For example, the
+byte 0xFF (255) can never appear in any UTF-8 string. If we ask Python to
+decode it, it yells at us:
+
+```python
+>>> b"\xFF".decode("UTF-8")
+UnicodeDecodeError: 'utf-8' codec can't decode byte 0xff in position 0: invalid start byte
+```
+
+However, tragically, we *can* create a file with that byte in its name (on
+Linux at least, though not usually on macOS):
+
+```python
+>>> open(b"y\xFFy", "w")
+```
+
+So some filepaths aren't representable in Unicode at all. Our plan to "convert
+platform-specific bytes into some consistent Unicode encoding" isn't going to
+work for everything. What does `b3sum` do with the file above?
+
+```bash
+$ b3sum y*
+af1349b9f5f9a1a6a0404dea36dcc9499bcb25c9adc112b7cc9a93cae41f3262  y�y
+```
+
+That � in there is a "Unicode replacement character". When we run into
+filepaths that we can't represent in Unicode, we replace the unrepresentable
+parts with these characters. On the checking side, to avoid any possible
+confusion between two different invalid filepaths, we automatically fail if we
+see a replacement character. Together with a few more details covered in the
+next section, this gives us an important set of properties:
+
+1. Any file can be hashed locally.
+2. Any file with a valid Unicode name not containing the � character can be
+   checked.
+3. Checking ambiguous or unrepresentable filepaths always fails.
+4. Checkfiles are always valid UTF-8.
+5. Checkfiles are portable between Unix and Windows.
+
+## Formal Rules
+
+1. When hashing, filepaths are represented in a platform-specific encoding,
+   which can accommodate any filepath on the current platform. In Rust, this is
+   `OsStr`/`OsString`.
+2. In output, filepaths are first converted to UTF-8. Any non-Unicode segments
+   are replaced with Unicode replacement characters (U+FFFD). In Rust, this is
+   `OsStr::to_string_lossy`.
+3. Then, if a filepath contains any backslashes (U+005C) or newlines (U+000A),
+   these characters are escaped as `\\` and `\n` respectively.
+4. Finally, any output line containing an escape sequence is prefixed with a
+   single backslash.
+5. When checking, each line is parsed as UTF-8, separated by a newline
+   (U+000A). Invalid UTF-8 is an error.
+6. Then, if a line begins with a backslash, the filepath component is
+   unescaped. Any escape sequence other than `\\` or `\n` is an error. If a
+   line does not begin with a backslash, unescaping is not performed, and any
+   backslashes in the filepath component are interpreted literally. (`b3sum`
+   output never contains unescaped backslashes, but they can occur in
+   checkfiles assembled by hand.)
+7. Finally, if a filepath contains a Unicode replacement character (U+FFFD) or
+   a null character (U+0000), it is an error.
+
+   **Additionally, on Windows only:**
+
+8. In output, all backslashes (U+005C) are replaced with forward slashes
+   (U+002F).
+9. When checking, after unescaping, if a filepath contains a backslash, it is
+   an error.
diff --git a/thirdparty/blake3/benches/bench.rs b/thirdparty/blake3/benches/bench.rs
new file mode 100644
index 000000000..e89f04c08
--- /dev/null
+++ b/thirdparty/blake3/benches/bench.rs
@@ -0,0 +1,623 @@
+#![feature(test)]
+
+extern crate test;
+
+use arrayref::array_ref;
+use arrayvec::ArrayVec;
+use blake3::platform::{Platform, MAX_SIMD_DEGREE};
+use blake3::OUT_LEN;
+use blake3::{BLOCK_LEN, CHUNK_LEN};
+use rand::prelude::*;
+use test::Bencher;
+
+const KIB: usize = 1024;
+
+// This struct randomizes two things:
+// 1. The actual bytes of input.
+// 2. The page offset the input starts at.
+pub struct RandomInput {
+    buf: Vec<u8>,
+    len: usize,
+    offsets: Vec<usize>,
+    offset_index: usize,
+}
+
+impl RandomInput {
+    pub fn new(b: &mut Bencher, len: usize) -> Self {
+        b.bytes += len as u64;
+        let page_size: usize = page_size::get();
+        let mut buf = vec![0u8; len + page_size];
+        let mut rng = rand::rng();
+        rng.fill_bytes(&mut buf);
+        let mut offsets: Vec<usize> = (0..page_size).collect();
+        offsets.shuffle(&mut rng);
+        Self {
+            buf,
+            len,
+            offsets,
+            offset_index: 0,
+        }
+    }
+
+    pub fn get(&mut self) -> &[u8] {
+        let offset = self.offsets[self.offset_index];
+        self.offset_index += 1;
+        if self.offset_index >= self.offsets.len() {
+            self.offset_index = 0;
+        }
+        &self.buf[offset..][..self.len]
+    }
+}
+
+fn bench_single_compression_fn(b: &mut Bencher, platform: Platform) {
+    let mut state = [1u32; 8];
+    let mut r = RandomInput::new(b, 64);
+    let input = array_ref!(r.get(), 0, 64);
+    b.iter(|| platform.compress_in_place(&mut state, input, 64 as u8, 0, 0));
+}
+
+#[bench]
+fn bench_single_compression_portable(b: &mut Bencher) {
+    bench_single_compression_fn(b, Platform::portable());
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_single_compression_sse2(b: &mut Bencher) {
+    if let Some(platform) = Platform::sse2() {
+        bench_single_compression_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_single_compression_sse41(b: &mut Bencher) {
+    if let Some(platform) = Platform::sse41() {
+        bench_single_compression_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(blake3_avx512_ffi)]
+fn bench_single_compression_avx512(b: &mut Bencher) {
+    if let Some(platform) = Platform::avx512() {
+        bench_single_compression_fn(b, platform);
+    }
+}
+
+fn bench_many_chunks_fn(b: &mut Bencher, platform: Platform) {
+    let degree = platform.simd_degree();
+    let mut inputs = Vec::new();
+    for _ in 0..degree {
+        inputs.push(RandomInput::new(b, CHUNK_LEN));
+    }
+    b.iter(|| {
+        let input_arrays: ArrayVec<&[u8; CHUNK_LEN], MAX_SIMD_DEGREE> = inputs
+            .iter_mut()
+            .take(degree)
+            .map(|i| array_ref!(i.get(), 0, CHUNK_LEN))
+            .collect();
+        let mut out = [0; MAX_SIMD_DEGREE * OUT_LEN];
+        platform.hash_many(
+            &input_arrays[..],
+            &[0; 8],
+            0,
+            blake3::IncrementCounter::Yes,
+            0,
+            0,
+            0,
+            &mut out,
+        );
+    });
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_chunks_sse2(b: &mut Bencher) {
+    if let Some(platform) = Platform::sse2() {
+        bench_many_chunks_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_chunks_sse41(b: &mut Bencher) {
+    if let Some(platform) = Platform::sse41() {
+        bench_many_chunks_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_chunks_avx2(b: &mut Bencher) {
+    if let Some(platform) = Platform::avx2() {
+        bench_many_chunks_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(blake3_avx512_ffi)]
+fn bench_many_chunks_avx512(b: &mut Bencher) {
+    if let Some(platform) = Platform::avx512() {
+        bench_many_chunks_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(blake3_neon)]
+fn bench_many_chunks_neon(b: &mut Bencher) {
+    bench_many_chunks_fn(b, Platform::neon().unwrap());
+}
+
+#[bench]
+#[cfg(blake3_wasm32_simd)]
+fn bench_many_chunks_wasm(b: &mut Bencher) {
+    bench_many_chunks_fn(b, Platform::wasm32_simd().unwrap());
+}
+
+// TODO: When we get const generics we can unify this with the chunks code.
+fn bench_many_parents_fn(b: &mut Bencher, platform: Platform) {
+    let degree = platform.simd_degree();
+    let mut inputs = Vec::new();
+    for _ in 0..degree {
+        inputs.push(RandomInput::new(b, BLOCK_LEN));
+    }
+    b.iter(|| {
+        let input_arrays: ArrayVec<&[u8; BLOCK_LEN], MAX_SIMD_DEGREE> = inputs
+            .iter_mut()
+            .take(degree)
+            .map(|i| array_ref!(i.get(), 0, BLOCK_LEN))
+            .collect();
+        let mut out = [0; MAX_SIMD_DEGREE * OUT_LEN];
+        platform.hash_many(
+            &input_arrays[..],
+            &[0; 8],
+            0,
+            blake3::IncrementCounter::No,
+            0,
+            0,
+            0,
+            &mut out,
+        );
+    });
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_parents_sse2(b: &mut Bencher) {
+    if let Some(platform) = Platform::sse2() {
+        bench_many_parents_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_parents_sse41(b: &mut Bencher) {
+    if let Some(platform) = Platform::sse41() {
+        bench_many_parents_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_parents_avx2(b: &mut Bencher) {
+    if let Some(platform) = Platform::avx2() {
+        bench_many_parents_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(blake3_avx512_ffi)]
+fn bench_many_parents_avx512(b: &mut Bencher) {
+    if let Some(platform) = Platform::avx512() {
+        bench_many_parents_fn(b, platform);
+    }
+}
+
+#[bench]
+#[cfg(blake3_neon)]
+fn bench_many_parents_neon(b: &mut Bencher) {
+    bench_many_parents_fn(b, Platform::neon().unwrap());
+}
+
+#[bench]
+#[cfg(blake3_wasm32_simd)]
+fn bench_many_parents_wasm(b: &mut Bencher) {
+    bench_many_parents_fn(b, Platform::wasm32_simd().unwrap());
+}
+
+fn bench_atonce(b: &mut Bencher, len: usize) {
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| blake3::hash(input.get()));
+}
+
+#[bench]
+fn bench_atonce_0001_block(b: &mut Bencher) {
+    bench_atonce(b, BLOCK_LEN);
+}
+
+#[bench]
+fn bench_atonce_0001_kib(b: &mut Bencher) {
+    bench_atonce(b, 1 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0002_kib(b: &mut Bencher) {
+    bench_atonce(b, 2 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0004_kib(b: &mut Bencher) {
+    bench_atonce(b, 4 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0008_kib(b: &mut Bencher) {
+    bench_atonce(b, 8 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0016_kib(b: &mut Bencher) {
+    bench_atonce(b, 16 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0032_kib(b: &mut Bencher) {
+    bench_atonce(b, 32 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0064_kib(b: &mut Bencher) {
+    bench_atonce(b, 64 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0128_kib(b: &mut Bencher) {
+    bench_atonce(b, 128 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0256_kib(b: &mut Bencher) {
+    bench_atonce(b, 256 * KIB);
+}
+
+#[bench]
+fn bench_atonce_0512_kib(b: &mut Bencher) {
+    bench_atonce(b, 512 * KIB);
+}
+
+#[bench]
+fn bench_atonce_1024_kib(b: &mut Bencher) {
+    bench_atonce(b, 1024 * KIB);
+}
+
+fn bench_incremental(b: &mut Bencher, len: usize) {
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| blake3::Hasher::new().update(input.get()).finalize());
+}
+
+#[bench]
+fn bench_incremental_0001_block(b: &mut Bencher) {
+    bench_incremental(b, BLOCK_LEN);
+}
+
+#[bench]
+fn bench_incremental_0001_kib(b: &mut Bencher) {
+    bench_incremental(b, 1 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0002_kib(b: &mut Bencher) {
+    bench_incremental(b, 2 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0004_kib(b: &mut Bencher) {
+    bench_incremental(b, 4 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0008_kib(b: &mut Bencher) {
+    bench_incremental(b, 8 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0016_kib(b: &mut Bencher) {
+    bench_incremental(b, 16 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0032_kib(b: &mut Bencher) {
+    bench_incremental(b, 32 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0064_kib(b: &mut Bencher) {
+    bench_incremental(b, 64 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0128_kib(b: &mut Bencher) {
+    bench_incremental(b, 128 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0256_kib(b: &mut Bencher) {
+    bench_incremental(b, 256 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0512_kib(b: &mut Bencher) {
+    bench_incremental(b, 512 * KIB);
+}
+
+#[bench]
+fn bench_incremental_1024_kib(b: &mut Bencher) {
+    bench_incremental(b, 1024 * KIB);
+}
+
+fn bench_reference(b: &mut Bencher, len: usize) {
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| {
+        let mut hasher = reference_impl::Hasher::new();
+        hasher.update(input.get());
+        let mut out = [0; 32];
+        hasher.finalize(&mut out);
+        out
+    });
+}
+
+#[bench]
+fn bench_reference_0001_block(b: &mut Bencher) {
+    bench_reference(b, BLOCK_LEN);
+}
+
+#[bench]
+fn bench_reference_0001_kib(b: &mut Bencher) {
+    bench_reference(b, 1 * KIB);
+}
+
+#[bench]
+fn bench_reference_0002_kib(b: &mut Bencher) {
+    bench_reference(b, 2 * KIB);
+}
+
+#[bench]
+fn bench_reference_0004_kib(b: &mut Bencher) {
+    bench_reference(b, 4 * KIB);
+}
+
+#[bench]
+fn bench_reference_0008_kib(b: &mut Bencher) {
+    bench_reference(b, 8 * KIB);
+}
+
+#[bench]
+fn bench_reference_0016_kib(b: &mut Bencher) {
+    bench_reference(b, 16 * KIB);
+}
+
+#[bench]
+fn bench_reference_0032_kib(b: &mut Bencher) {
+    bench_reference(b, 32 * KIB);
+}
+
+#[bench]
+fn bench_reference_0064_kib(b: &mut Bencher) {
+    bench_reference(b, 64 * KIB);
+}
+
+#[bench]
+fn bench_reference_0128_kib(b: &mut Bencher) {
+    bench_reference(b, 128 * KIB);
+}
+
+#[bench]
+fn bench_reference_0256_kib(b: &mut Bencher) {
+    bench_reference(b, 256 * KIB);
+}
+
+#[bench]
+fn bench_reference_0512_kib(b: &mut Bencher) {
+    bench_reference(b, 512 * KIB);
+}
+
+#[bench]
+fn bench_reference_1024_kib(b: &mut Bencher) {
+    bench_reference(b, 1024 * KIB);
+}
+
+#[cfg(feature = "rayon")]
+fn bench_rayon(b: &mut Bencher, len: usize) {
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| blake3::Hasher::new().update_rayon(input.get()).finalize());
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0001_block(b: &mut Bencher) {
+    bench_rayon(b, BLOCK_LEN);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0001_kib(b: &mut Bencher) {
+    bench_rayon(b, 1 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0002_kib(b: &mut Bencher) {
+    bench_rayon(b, 2 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0004_kib(b: &mut Bencher) {
+    bench_rayon(b, 4 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0008_kib(b: &mut Bencher) {
+    bench_rayon(b, 8 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0016_kib(b: &mut Bencher) {
+    bench_rayon(b, 16 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0032_kib(b: &mut Bencher) {
+    bench_rayon(b, 32 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0064_kib(b: &mut Bencher) {
+    bench_rayon(b, 64 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0128_kib(b: &mut Bencher) {
+    bench_rayon(b, 128 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0256_kib(b: &mut Bencher) {
+    bench_rayon(b, 256 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_0512_kib(b: &mut Bencher) {
+    bench_rayon(b, 512 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "rayon")]
+fn bench_rayon_1024_kib(b: &mut Bencher) {
+    bench_rayon(b, 1024 * KIB);
+}
+
+// This checks that update() splits up its input in increasing powers of 2, so
+// that it can recover a high degree of parallelism when the number of bytes
+// hashed so far is uneven. The performance of this benchmark should be
+// reasonably close to bench_incremental_0064_kib, within 80% or so. When we
+// had a bug in this logic (https://github.com/BLAKE3-team/BLAKE3/issues/69),
+// performance was less than half.
+#[bench]
+fn bench_two_updates(b: &mut Bencher) {
+    let len = 65536;
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| {
+        let mut hasher = blake3::Hasher::new();
+        let input = input.get();
+        hasher.update(&input[..1]);
+        hasher.update(&input[1..]);
+        hasher.finalize()
+    });
+}
+
+fn bench_xof(b: &mut Bencher, len: usize) {
+    b.bytes = len as u64;
+    let mut output = [0u8; 64 * BLOCK_LEN];
+    let output_slice = &mut output[..len];
+    let mut xof = blake3::Hasher::new().finalize_xof();
+    b.iter(|| xof.fill(output_slice));
+}
+
+#[bench]
+fn bench_xof_01_block(b: &mut Bencher) {
+    bench_xof(b, 1 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_02_blocks(b: &mut Bencher) {
+    bench_xof(b, 2 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_03_blocks(b: &mut Bencher) {
+    bench_xof(b, 3 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_04_blocks(b: &mut Bencher) {
+    bench_xof(b, 4 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_05_blocks(b: &mut Bencher) {
+    bench_xof(b, 5 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_06_blocks(b: &mut Bencher) {
+    bench_xof(b, 6 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_07_blocks(b: &mut Bencher) {
+    bench_xof(b, 7 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_08_blocks(b: &mut Bencher) {
+    bench_xof(b, 8 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_09_blocks(b: &mut Bencher) {
+    bench_xof(b, 9 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_10_blocks(b: &mut Bencher) {
+    bench_xof(b, 10 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_11_blocks(b: &mut Bencher) {
+    bench_xof(b, 11 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_12_blocks(b: &mut Bencher) {
+    bench_xof(b, 12 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_13_blocks(b: &mut Bencher) {
+    bench_xof(b, 13 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_14_blocks(b: &mut Bencher) {
+    bench_xof(b, 14 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_15_blocks(b: &mut Bencher) {
+    bench_xof(b, 15 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_16_blocks(b: &mut Bencher) {
+    bench_xof(b, 16 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_32_blocks(b: &mut Bencher) {
+    bench_xof(b, 32 * BLOCK_LEN);
+}
+
+#[bench]
+fn bench_xof_64_blocks(b: &mut Bencher) {
+    bench_xof(b, 64 * BLOCK_LEN);
+}
diff --git a/thirdparty/blake3/build.rs b/thirdparty/blake3/build.rs
new file mode 100644
index 000000000..01b692fd7
--- /dev/null
+++ b/thirdparty/blake3/build.rs
@@ -0,0 +1,389 @@
+use std::env;
+
+fn defined(var: &str) -> bool {
+    println!("cargo:rerun-if-env-changed={}", var);
+    env::var_os(var).is_some()
+}
+
+fn is_pure() -> bool {
+    defined("CARGO_FEATURE_PURE")
+}
+
+fn should_prefer_intrinsics() -> bool {
+    defined("CARGO_FEATURE_PREFER_INTRINSICS")
+}
+
+fn is_neon() -> bool {
+    defined("CARGO_FEATURE_NEON")
+}
+
+fn is_no_neon() -> bool {
+    defined("CARGO_FEATURE_NO_NEON")
+}
+
+fn is_wasm32_simd() -> bool {
+    defined("CARGO_FEATURE_WASM32_SIMD")
+}
+
+fn is_ci() -> bool {
+    defined("BLAKE3_CI")
+}
+
+fn warn(warning: &str) {
+    assert!(!warning.contains("\n"));
+    println!("cargo:warning={}", warning);
+    if is_ci() {
+        println!("cargo:warning=Warnings in CI are treated as errors. Build failed.");
+        std::process::exit(1);
+    }
+}
+
+fn target_components() -> Vec<String> {
+    let target = env::var("TARGET").unwrap();
+    target.split("-").map(|s| s.to_string()).collect()
+}
+
+fn is_x86_64() -> bool {
+    target_components()[0] == "x86_64"
+}
+
+fn is_windows_target() -> bool {
+    env::var("CARGO_CFG_TARGET_OS").unwrap() == "windows"
+}
+
+fn use_msvc_asm() -> bool {
+    const MSVC_NAMES: &[&str] = &["", "cl", "cl.exe"];
+    let target_os = env::var("CARGO_CFG_TARGET_OS").unwrap_or_default();
+    let target_env = env::var("CARGO_CFG_TARGET_ENV").unwrap_or_default();
+    let target_windows_msvc = target_os == "windows" && target_env == "msvc";
+    let host_triple = env::var("HOST").unwrap_or_default();
+    let target_triple = env::var("TARGET").unwrap_or_default();
+    let cross_compiling = host_triple != target_triple;
+    let cc = env::var("CC").unwrap_or_default().to_ascii_lowercase();
+    if !target_windows_msvc {
+        // We are not building for Windows with the MSVC toolchain.
+        false
+    } else if !cross_compiling && MSVC_NAMES.contains(&&*cc) {
+        // We are building on Windows with the MSVC toolchain (and not cross-compiling for another architecture or target).
+        true
+    } else {
+        // We are cross-compiling to Windows with the MSVC toolchain.
+        let target_arch = env::var("CARGO_CFG_TARGET_ARCH").unwrap_or_default();
+        let target_vendor = env::var("CARGO_CFG_TARGET_VENDOR").unwrap_or_default();
+        let cc = env::var(format!("CC_{target_arch}_{target_vendor}_windows_msvc"))
+            .unwrap_or_default()
+            .to_ascii_lowercase();
+        // Check if we are using the MSVC compiler.
+        MSVC_NAMES.contains(&&*cc)
+    }
+}
+
+fn is_x86_32() -> bool {
+    let arch = &target_components()[0];
+    arch == "i386" || arch == "i586" || arch == "i686"
+}
+
+fn is_arm() -> bool {
+    is_armv7() || is_aarch64() || target_components()[0] == "arm"
+}
+
+fn is_aarch64() -> bool {
+    target_components()[0] == "aarch64"
+}
+
+fn is_armv7() -> bool {
+    target_components()[0] == "armv7"
+}
+
+fn is_wasm32() -> bool {
+    target_components()[0] == "wasm32"
+}
+
+fn endianness() -> String {
+    let endianness = env::var("CARGO_CFG_TARGET_ENDIAN").unwrap();
+    assert!(endianness == "little" || endianness == "big");
+    endianness
+}
+
+fn is_little_endian() -> bool {
+    endianness() == "little"
+}
+
+fn is_big_endian() -> bool {
+    endianness() == "big"
+}
+
+// Windows targets may be using the MSVC toolchain or the MinGW toolchain. The
+// right compiler flags to use depend on the toolchain. (And we don't want to
+// use flag_if_supported, because we don't want features to be silently
+// disabled by old compilers.)
+fn is_windows_msvc() -> bool {
+    // Some targets are only two components long, so check in steps.
+    let second_component = &target_components()[1];
+    (second_component == "pc" || second_component == "win7")
+        && target_components()[2] == "windows"
+        && target_components()[3] == "msvc"
+}
+
+// MinGW toolchain uses 2 different targets depending on the main compiler.
+// Target for a general MinGW toolchain ends with `-gnu` (GCC is used as C
+// compiler). Target for a LLVM-MinGW toolchain (Clang is used as C compiler)
+// ends with `-gnullvm`.
+fn is_windows_gnu() -> bool {
+    // Some targets are only two components long, so check in steps.
+    let second_component = &target_components()[1];
+    (second_component == "pc" || second_component == "win7")
+        && target_components()[2] == "windows"
+        && target_components()[3] != "msvc"
+}
+
+fn new_build() -> cc::Build {
+    let mut build = cc::Build::new();
+    if !is_windows_msvc() {
+        build.flag("-std=c11");
+    }
+    // Do NOT trigger a rebuild any time the env changes (e.g. $PATH).
+    // This prevents all downstream crates from being rebuilt when `cargo check`
+    // or `cargo build` are run in different environments, like Rust Analyzer
+    // vs. in the terminal vs. in a Git pre-commit hook.
+    build.emit_rerun_if_env_changed(false);
+    build
+}
+
+#[derive(PartialEq)]
+enum CCompilerSupport {
+    NoCompiler,
+    NoAVX512,
+    YesAVX512,
+}
+use CCompilerSupport::*;
+
+fn c_compiler_support() -> CCompilerSupport {
+    let build = new_build();
+    let flags_checked;
+    let support_result: Result<bool, _> = if is_windows_msvc() {
+        flags_checked = "/arch:AVX512";
+        build.is_flag_supported("/arch:AVX512")
+    } else {
+        // Check for both of the flags we use. If -mavx512f works, then -mavx512vl
+        // will probably always work too, but we might as well be thorough.
+        flags_checked = "-mavx512f and -mavx512vl";
+        match build.is_flag_supported("-mavx512f") {
+            Ok(true) => build.is_flag_supported("-mavx512vl"),
+            false_or_error => false_or_error,
+        }
+    };
+    match support_result {
+        Ok(true) => YesAVX512,
+        Ok(false) => {
+            warn(&format!(
+                "The C compiler {:?} does not support {}.",
+                build.get_compiler().path(),
+                flags_checked,
+            ));
+            NoAVX512
+        }
+        Err(e) => {
+            println!("{:?}", e);
+            warn(&format!(
+                "No C compiler {:?} detected.",
+                build.get_compiler().path()
+            ));
+            NoCompiler
+        }
+    }
+}
+
+fn build_sse2_sse41_avx2_rust_intrinsics() {
+    // No C code to compile here. Set the cfg flags that enable the Rust SSE2,
+    // SSE4.1, and AVX2 intrinsics modules. The regular Cargo build will compile
+    // them.
+    println!("cargo:rustc-cfg=blake3_sse2_rust");
+    println!("cargo:rustc-cfg=blake3_sse41_rust");
+    println!("cargo:rustc-cfg=blake3_avx2_rust");
+}
+
+fn build_sse2_sse41_avx2_assembly() {
+    // Build the assembly implementations for SSE4.1 and AVX2. This is
+    // preferred, but it only supports x86_64.
+    assert!(is_x86_64());
+    println!("cargo:rustc-cfg=blake3_sse2_ffi");
+    println!("cargo:rustc-cfg=blake3_sse41_ffi");
+    println!("cargo:rustc-cfg=blake3_avx2_ffi");
+    let mut build = new_build();
+    if is_windows_target() {
+        if use_msvc_asm() {
+            build.file("c/blake3_sse2_x86-64_windows_msvc.asm");
+            build.file("c/blake3_sse41_x86-64_windows_msvc.asm");
+            build.file("c/blake3_avx2_x86-64_windows_msvc.asm");
+        } else {
+            build.file("c/blake3_sse2_x86-64_windows_gnu.S");
+            build.file("c/blake3_sse41_x86-64_windows_gnu.S");
+            build.file("c/blake3_avx2_x86-64_windows_gnu.S");
+        }
+    } else {
+        // All non-Windows implementations are assumed to support
+        // Linux-style assembly. These files do contain a small
+        // explicit workaround for macOS also.
+        build.file("c/blake3_sse2_x86-64_unix.S");
+        build.file("c/blake3_sse41_x86-64_unix.S");
+        build.file("c/blake3_avx2_x86-64_unix.S");
+    }
+    build.compile("blake3_sse2_sse41_avx2_assembly");
+}
+
+fn build_avx512_c_intrinsics() {
+    // This is required on 32-bit x86 targets, since the assembly
+    // implementation doesn't support those.
+    println!("cargo:rustc-cfg=blake3_avx512_ffi");
+    let mut build = new_build();
+    build.file("c/blake3_avx512.c");
+    if is_windows_msvc() {
+        build.flag("/arch:AVX512");
+    } else {
+        build.flag("-mavx512f");
+        build.flag("-mavx512vl");
+    }
+    if is_windows_gnu() {
+        // Workaround for https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65782.
+        build.flag("-fno-asynchronous-unwind-tables");
+    }
+    build.compile("blake3_avx512_intrinsics");
+}
+
+fn build_avx512_assembly() {
+    // Build the assembly implementation for AVX-512. This is preferred, but it
+    // only supports x86_64.
+    assert!(is_x86_64());
+    println!("cargo:rustc-cfg=blake3_avx512_ffi");
+    let mut build = new_build();
+    let mut is_msvc = false;
+    if is_windows_target() {
+        if use_msvc_asm() {
+            build.file("c/blake3_avx512_x86-64_windows_msvc.asm");
+            is_msvc = true;
+        } else {
+            build.file("c/blake3_avx512_x86-64_windows_gnu.S");
+        }
+    } else {
+        build.file("c/blake3_avx512_x86-64_unix.S");
+    }
+
+    // Older versions of Clang require these flags, even for assembly. See
+    // https://github.com/BLAKE3-team/BLAKE3/issues/79.
+    if !is_msvc {
+        build.flag("-mavx512f");
+        build.flag("-mavx512vl");
+    }
+    build.compile("blake3_avx512_assembly");
+}
+
+fn build_neon_c_intrinsics() {
+    let mut build = new_build();
+    // Note that blake3_neon.c normally depends on the blake3_portable.c
+    // for the single-instance compression function, but we expose
+    // portable.rs over FFI instead. See ffi_neon.rs.
+    build.file("c/blake3_neon.c");
+    // ARMv7 platforms that support NEON generally need the following
+    // flags. AArch64 supports NEON by default and does not support -mpfu.
+    if is_armv7() {
+        build.flag("-mfpu=neon-vfpv4");
+        build.flag("-mfloat-abi=hard");
+    }
+    build.compile("blake3_neon");
+}
+
+fn build_wasm32_simd() {
+    assert!(is_wasm32());
+    // No C code to compile here. Set the cfg flags that enable the Wasm SIMD.
+    // The regular Cargo build will compile it.
+    println!("cargo:rustc-cfg=blake3_wasm32_simd");
+}
+
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    // As of Rust 1.80, unrecognized config names are warnings. Give Cargo all of our config names.
+    let all_cfgs = [
+        "blake3_sse2_ffi",
+        "blake3_sse2_rust",
+        "blake3_sse41_ffi",
+        "blake3_sse41_rust",
+        "blake3_avx2_ffi",
+        "blake3_avx2_rust",
+        "blake3_avx512_ffi",
+        "blake3_neon",
+        "blake3_wasm32_simd",
+    ];
+    for cfg_name in all_cfgs {
+        // TODO: Switch this whole file to the new :: syntax when our MSRV reaches 1.77.
+        // https://doc.rust-lang.org/cargo/reference/build-scripts.html#outputs-of-the-build-script
+        println!("cargo:rustc-check-cfg=cfg({cfg_name}, values(none()))");
+    }
+
+    if is_pure() && is_neon() {
+        panic!("It doesn't make sense to enable both \"pure\" and \"neon\".");
+    }
+
+    if is_no_neon() && is_neon() {
+        panic!("It doesn't make sense to enable both \"no_neon\" and \"neon\".");
+    }
+
+    if is_x86_64() || is_x86_32() {
+        let support = c_compiler_support();
+        if is_x86_32() || should_prefer_intrinsics() || is_pure() || support == NoCompiler {
+            build_sse2_sse41_avx2_rust_intrinsics();
+        } else {
+            // We assume that all C compilers can assemble SSE4.1 and AVX2. We
+            // don't explicitly check for support.
+            build_sse2_sse41_avx2_assembly();
+        }
+
+        if is_pure() || support == NoCompiler || support == NoAVX512 {
+            // The binary will not include any AVX-512 code.
+        } else if is_x86_32() || should_prefer_intrinsics() {
+            build_avx512_c_intrinsics();
+        } else {
+            build_avx512_assembly();
+        }
+    }
+
+    if is_neon() && is_big_endian() {
+        panic!("The NEON implementation doesn't support big-endian ARM.")
+    }
+
+    if (is_arm() && is_neon())
+        || (!is_no_neon() && !is_pure() && is_aarch64() && is_little_endian())
+    {
+        println!("cargo:rustc-cfg=blake3_neon");
+        build_neon_c_intrinsics();
+    }
+
+    if is_wasm32() && is_wasm32_simd() {
+        build_wasm32_simd();
+    }
+
+    // The `cc` crate doesn't automatically emit rerun-if directives for the
+    // environment variables it supports, in particular for $CC. We expect to
+    // do a lot of benchmarking across different compilers, so we explicitly
+    // add the variables that we're likely to need.
+    println!("cargo:rerun-if-env-changed=CC");
+    println!("cargo:rerun-if-env-changed=CFLAGS");
+
+    // Ditto for source files, though these shouldn't change as often.
+    for file in std::fs::read_dir("c")? {
+        println!(
+            "cargo:rerun-if-changed={}",
+            file?.path().to_str().expect("utf-8")
+        );
+    }
+
+    // When compiling with clang-cl for windows, it adds .asm files to the root
+    // which we need to delete so cargo doesn't get angry
+    if is_windows_target() && !use_msvc_asm() {
+        let _ = std::fs::remove_file("blake3_avx2_x86-64_windows_gnu.asm");
+        let _ = std::fs::remove_file("blake3_avx512_x86-64_windows_gnu.asm");
+        let _ = std::fs::remove_file("blake3_sse2_x86-64_windows_gnu.asm");
+        let _ = std::fs::remove_file("blake3_sse41_x86-64_windows_gnu.asm");
+    }
+
+    Ok(())
+}
diff --git a/thirdparty/blake3/c/.gitignore b/thirdparty/blake3/c/.gitignore
new file mode 100644
index 000000000..6385bd378
--- /dev/null
+++ b/thirdparty/blake3/c/.gitignore
@@ -0,0 +1,7 @@
+blake3
+example
+example_tbb
+build/
+*.o
+
+CMakeUserPresets.json
diff --git a/thirdparty/blake3/c/Makefile.testing b/thirdparty/blake3/c/Makefile.testing
new file mode 100644
index 000000000..b540528d3
--- /dev/null
+++ b/thirdparty/blake3/c/Makefile.testing
@@ -0,0 +1,82 @@
+# This Makefile is only for testing. C callers should follow the instructions
+# in ./README.md to incorporate these C files into their existing build.
+
+NAME=blake3
+CC=gcc
+CFLAGS=-O3 -Wall -Wextra -std=c11 -pedantic -fstack-protector-strong -D_FORTIFY_SOURCE=2 -fPIE -fvisibility=hidden
+LDFLAGS=-pie -Wl,-z,relro,-z,now
+TARGETS=
+ASM_TARGETS=
+EXTRAFLAGS=-Wa,--noexecstack
+
+ifdef BLAKE3_NO_SSE2
+EXTRAFLAGS += -DBLAKE3_NO_SSE2
+else
+TARGETS += blake3_sse2.o
+ASM_TARGETS += blake3_sse2_x86-64_unix.S
+endif
+
+ifdef BLAKE3_NO_SSE41
+EXTRAFLAGS += -DBLAKE3_NO_SSE41
+else
+TARGETS += blake3_sse41.o
+ASM_TARGETS += blake3_sse41_x86-64_unix.S
+endif
+
+ifdef BLAKE3_NO_AVX2
+EXTRAFLAGS += -DBLAKE3_NO_AVX2
+else
+TARGETS += blake3_avx2.o
+ASM_TARGETS += blake3_avx2_x86-64_unix.S
+endif
+
+ifdef BLAKE3_NO_AVX512
+EXTRAFLAGS += -DBLAKE3_NO_AVX512
+else
+TARGETS += blake3_avx512.o
+ASM_TARGETS += blake3_avx512_x86-64_unix.S
+endif
+
+ifdef BLAKE3_USE_NEON
+EXTRAFLAGS += -DBLAKE3_USE_NEON=1
+TARGETS += blake3_neon.o
+endif
+
+ifdef BLAKE3_NO_NEON
+EXTRAFLAGS += -DBLAKE3_USE_NEON=0
+endif
+
+all: blake3.c blake3_dispatch.c blake3_portable.c main.c $(TARGETS)
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) $^ -o $(NAME) $(LDFLAGS)
+
+blake3_sse2.o: blake3_sse2.c
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) -c $^ -o $@ -msse2
+
+blake3_sse41.o: blake3_sse41.c
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) -c $^ -o $@ -msse4.1
+
+blake3_avx2.o: blake3_avx2.c
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) -c $^ -o $@ -mavx2
+
+blake3_avx512.o: blake3_avx512.c
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) -c $^ -o $@ -mavx512f -mavx512vl
+
+blake3_neon.o: blake3_neon.c
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) -c $^ -o $@
+
+test: CFLAGS += -DBLAKE3_TESTING -fsanitize=address,undefined
+test: all
+	./test.py
+
+asm: blake3.c blake3_dispatch.c blake3_portable.c main.c $(ASM_TARGETS)
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) $^ -o $(NAME) $(LDFLAGS)
+
+test_asm: CFLAGS += -DBLAKE3_TESTING -fsanitize=address,undefined 
+test_asm: asm
+	./test.py
+
+example: example.c blake3.c blake3_dispatch.c blake3_portable.c $(ASM_TARGETS)
+	$(CC) $(CFLAGS) $(EXTRAFLAGS) $^ -o $@ $(LDFLAGS)
+
+clean: 
+	rm -f $(NAME) *.o
diff --git a/thirdparty/blake3/c/README.md b/thirdparty/blake3/c/README.md
new file mode 100644
index 000000000..57260b43b
--- /dev/null
+++ b/thirdparty/blake3/c/README.md
@@ -0,0 +1,403 @@
+The official C implementation of BLAKE3.
+
+# Example
+
+An example program that hashes bytes from standard input and prints the
+result:
+
+```c
+#include "blake3.h"
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+
+int main(void) {
+  // Initialize the hasher.
+  blake3_hasher hasher;
+  blake3_hasher_init(&hasher);
+
+  // Read input bytes from stdin.
+  unsigned char buf[65536];
+  while (1) {
+    ssize_t n = read(STDIN_FILENO, buf, sizeof(buf));
+    if (n > 0) {
+      blake3_hasher_update(&hasher, buf, n);
+    } else if (n == 0) {
+      break; // end of file
+    } else {
+      fprintf(stderr, "read failed: %s\n", strerror(errno));
+      return 1;
+    }
+  }
+
+  // Finalize the hash. BLAKE3_OUT_LEN is the default output length, 32 bytes.
+  uint8_t output[BLAKE3_OUT_LEN];
+  blake3_hasher_finalize(&hasher, output, BLAKE3_OUT_LEN);
+
+  // Print the hash as hexadecimal.
+  for (size_t i = 0; i < BLAKE3_OUT_LEN; i++) {
+    printf("%02x", output[i]);
+  }
+  printf("\n");
+  return 0;
+}
+```
+
+The code above is included in this directory as `example.c`. If you're
+on x86\_64 with a Unix-like OS, you can compile a working binary like
+this:
+
+```bash
+gcc -O3 -o example example.c blake3.c blake3_dispatch.c blake3_portable.c \
+    blake3_sse2_x86-64_unix.S blake3_sse41_x86-64_unix.S blake3_avx2_x86-64_unix.S \
+    blake3_avx512_x86-64_unix.S
+```
+
+# API
+
+## The Struct
+
+```c
+typedef struct {
+  // private fields
+} blake3_hasher;
+```
+
+An incremental BLAKE3 hashing state, which can accept any number of
+updates. This implementation doesn't allocate any heap memory, but
+`sizeof(blake3_hasher)` itself is relatively large, currently 1912 bytes
+on x86-64. This size can be reduced by restricting the maximum input
+length, as described in Section 5.4 of [the BLAKE3
+spec](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf),
+but this implementation doesn't currently support that strategy.
+
+## Common API Functions
+
+```c
+void blake3_hasher_init(
+  blake3_hasher *self);
+```
+
+Initialize a `blake3_hasher` in the default hashing mode.
+
+---
+
+```c
+void blake3_hasher_update(
+  blake3_hasher *self,
+  const void *input,
+  size_t input_len);
+```
+
+Add input to the hasher. This can be called any number of times. This function
+is always single-threaded; for multithreading see `blake3_hasher_update_tbb`
+below.
+
+
+---
+
+```c
+void blake3_hasher_finalize(
+  const blake3_hasher *self,
+  uint8_t *out,
+  size_t out_len);
+```
+
+Finalize the hasher and return an output of any length, given in bytes.
+This doesn't modify the hasher itself, and it's possible to finalize
+again after adding more input. The constant `BLAKE3_OUT_LEN` provides
+the default output length, 32 bytes, which is recommended for most
+callers. See the [Security Notes](#security-notes) below.
+
+## Less Common API Functions
+
+```c
+void blake3_hasher_init_keyed(
+  blake3_hasher *self,
+  const uint8_t key[BLAKE3_KEY_LEN]);
+```
+
+Initialize a `blake3_hasher` in the keyed hashing mode. The key must be
+exactly 32 bytes.
+
+---
+
+```c
+void blake3_hasher_init_derive_key(
+  blake3_hasher *self,
+  const char *context);
+```
+
+Initialize a `blake3_hasher` in the key derivation mode. The context
+string is given as an initialization parameter, and afterwards input key
+material should be given with `blake3_hasher_update`. The context string
+is a null-terminated C string which should be **hardcoded, globally
+unique, and application-specific**. The context string should not
+include any dynamic input like salts, nonces, or identifiers read from a
+database at runtime. A good default format for the context string is
+`"[application] [commit timestamp] [purpose]"`, e.g., `"example.com
+2019-12-25 16:18:03 session tokens v1"`.
+
+This function is intended for application code written in C. For
+language bindings, see `blake3_hasher_init_derive_key_raw` below.
+
+---
+
+```c
+void blake3_hasher_init_derive_key_raw(
+  blake3_hasher *self,
+  const void *context,
+  size_t context_len);
+```
+
+As `blake3_hasher_init_derive_key` above, except that the context string
+is given as a pointer to an array of arbitrary bytes with a provided
+length. This is intended for writing language bindings, where C string
+conversion would add unnecessary overhead and new error cases. Unicode
+strings should be encoded as UTF-8.
+
+Application code in C should prefer `blake3_hasher_init_derive_key`,
+which takes the context as a C string. If you need to use arbitrary
+bytes as a context string in application code, consider whether you're
+violating the requirement that context strings should be hardcoded.
+
+---
+
+```c
+void blake3_hasher_update_tbb(
+  blake3_hasher *self,
+  const void *input,
+  size_t input_len);
+```
+
+Add input to the hasher, using [oneTBB] to process large inputs using multiple
+threads. This can be called any number of times. This gives the same result as
+`blake3_hasher_update` above.
+
+[oneTBB]: https://uxlfoundation.github.io/oneTBB/
+
+NOTE: This function is only enabled when the library is compiled with CMake option `BLAKE3_USE_TBB`
+and when the oneTBB library is detected on the host system. See the building instructions for
+further details.
+
+To get any performance benefit from multithreading, the input buffer needs to
+be large. As a rule of thumb on x86_64, `blake3_hasher_update_tbb` is _slower_
+than `blake3_hasher_update` for inputs under 128 KiB. That threshold varies
+quite a lot across different processors, and it's important to benchmark your
+specific use case.
+
+Hashing large files with this function usually requires
+[memory-mapping](https://en.wikipedia.org/wiki/Memory-mapped_file), since
+reading a file into memory in a single-threaded loop takes longer than hashing
+the resulting buffer. Note that hashing a memory-mapped file with this function
+produces a "random" pattern of disk reads, which can be slow on spinning disks.
+Again it's important to benchmark your specific use case.
+
+This implementation doesn't require configuration of thread resources and will
+use as many cores as possible by default. More fine-grained control of
+resources is possible using the [oneTBB] API.
+
+---
+
+```c
+void blake3_hasher_finalize_seek(
+  const blake3_hasher *self,
+  uint64_t seek,
+  uint8_t *out,
+  size_t out_len);
+```
+
+The same as `blake3_hasher_finalize`, but with an additional `seek`
+parameter for the starting byte position in the output stream. To
+efficiently stream a large output without allocating memory, call this
+function in a loop, incrementing `seek` by the output length each time.
+
+---
+
+```c
+void blake3_hasher_reset(
+  blake3_hasher *self);
+```
+
+Reset the hasher to its initial state, prior to any calls to
+`blake3_hasher_update`. Currently this is no different from calling
+`blake3_hasher_init` or similar again.
+
+# Security Notes
+
+Outputs shorter than the default length of 32 bytes (256 bits) provide less security. An N-bit
+BLAKE3 output is intended to provide N bits of first and second preimage resistance and N/2
+bits of collision resistance, for any N up to 256. Longer outputs don't provide any additional
+security.
+
+Avoid relying on the secrecy of the output offset, that is, the `seek` argument of
+`blake3_hasher_finalize_seek`. [_Block-Cipher-Based Tree Hashing_ by Aldo
+Gunsing](https://eprint.iacr.org/2022/283) shows that an attacker who knows both the message
+and the key (if any) can easily determine the offset of an extended output. For comparison,
+AES-CTR has a similar property: if you know the key, you can decrypt a block from an unknown
+position in the output stream to recover its block index. Callers with strong secret keys
+aren't affected in practice, but secret offsets are a [design
+smell](https://en.wikipedia.org/wiki/Design_smell) in any case.
+
+# Building
+
+The easiest and most complete method of compiling this library is with CMake.
+This is the method described in the next section. Toward the end of the
+building section there are more in depth notes about compiling manually and
+things that are useful to understand if you need to integrate this library with
+another build system.
+
+## CMake
+
+The minimum version of CMake is 3.9. The following invocations will compile and
+install `libblake3`. With recent CMake:
+
+```bash
+cmake -S c -B c/build "-DCMAKE_INSTALL_PREFIX=/usr/local"
+cmake --build c/build --target install
+```
+
+With an older CMake:
+
+```bash
+cd c
+mkdir build
+cd build
+cmake .. "-DCMAKE_INSTALL_PREFIX=/usr/local"
+cmake --build . --target install
+```
+
+The following options are available when compiling with CMake:
+
+- `BLAKE3_USE_TBB`: Enable oneTBB parallelism (Requires a C++20 capable compiler)
+- `BLAKE3_FETCH_TBB`: Allow fetching oneTBB from GitHub (only if not found on system)
+- `BLAKE3_EXAMPLES`: Compile and install example programs
+
+Options can be enabled like this:
+
+```bash
+cmake -S c -B c/build "-DCMAKE_INSTALL_PREFIX=/usr/local" -DBLAKE3_USE_TBB=1 -DBLAKE3_FETCH_TBB=1
+```
+
+## Building manually
+
+We try to keep the build simple enough that you can compile this library "by
+hand", and it's expected that many callers will integrate it with their
+pre-existing build systems. See the `gcc` one-liner in the "Example" section
+above.
+
+### x86
+
+Dynamic dispatch is enabled by default on x86. The implementation will
+query the CPU at runtime to detect SIMD support, and it will use the
+widest instruction set available. By default, `blake3_dispatch.c`
+expects to be linked with code for five different instruction sets:
+portable C, SSE2, SSE4.1, AVX2, and AVX-512.
+
+For each of the x86 SIMD instruction sets, four versions are available:
+three flavors of assembly (Unix, Windows MSVC, and Windows GNU) and one
+version using C intrinsics. The assembly versions are generally
+preferred. They perform better, they perform more consistently across
+different compilers, and they build more quickly. On the other hand, the
+assembly versions are x86\_64-only, and you need to select the right
+flavor for your target platform.
+
+Here's an example of building a shared library on x86\_64 Linux using
+the assembly implementations:
+
+```bash
+gcc -shared -O3 -o libblake3.so blake3.c blake3_dispatch.c blake3_portable.c \
+    blake3_sse2_x86-64_unix.S blake3_sse41_x86-64_unix.S blake3_avx2_x86-64_unix.S \
+    blake3_avx512_x86-64_unix.S
+```
+
+When building the intrinsics-based implementations, you need to build
+each implementation separately, with the corresponding instruction set
+explicitly enabled in the compiler. Here's the same shared library using
+the intrinsics-based implementations:
+
+```bash
+gcc -c -fPIC -O3 -msse2 blake3_sse2.c -o blake3_sse2.o
+gcc -c -fPIC -O3 -msse4.1 blake3_sse41.c -o blake3_sse41.o
+gcc -c -fPIC -O3 -mavx2 blake3_avx2.c -o blake3_avx2.o
+gcc -c -fPIC -O3 -mavx512f -mavx512vl blake3_avx512.c -o blake3_avx512.o
+gcc -shared -O3 -o libblake3.so blake3.c blake3_dispatch.c blake3_portable.c \
+    blake3_avx2.o blake3_avx512.o blake3_sse41.o blake3_sse2.o
+```
+
+Note above that building `blake3_avx512.c` requires both `-mavx512f` and
+`-mavx512vl` under GCC and Clang. Under MSVC, the single `/arch:AVX512`
+flag is sufficient. The MSVC equivalent of `-mavx2` is `/arch:AVX2`.
+MSVC enables SSE2 and SSE4.1 by default, and it doesn't have a
+corresponding flag.
+
+If you want to omit SIMD code entirely, you need to explicitly disable
+each instruction set. Here's an example of building a shared library on
+x86 with only portable code:
+
+```bash
+gcc -shared -O3 -o libblake3.so -DBLAKE3_NO_SSE2 -DBLAKE3_NO_SSE41 -DBLAKE3_NO_AVX2 \
+    -DBLAKE3_NO_AVX512 blake3.c blake3_dispatch.c blake3_portable.c
+```
+
+### ARM NEON
+
+The NEON implementation is enabled by default on AArch64, but not on
+other ARM targets, since not all of them support it. To enable it, set
+`BLAKE3_USE_NEON=1`. Here's an example of building a shared library on
+ARM Linux with NEON support:
+
+```bash
+gcc -shared -O3 -o libblake3.so -DBLAKE3_USE_NEON=1 blake3.c blake3_dispatch.c \
+    blake3_portable.c blake3_neon.c
+```
+
+To explicitiy disable using NEON instructions on AArch64, set
+`BLAKE3_USE_NEON=0`.
+
+```bash
+gcc -shared -O3 -o libblake3.so -DBLAKE3_USE_NEON=0 blake3.c blake3_dispatch.c \
+    blake3_portable.c 
+```
+
+Note that on some targets (ARMv7 in particular), extra flags may be
+required to activate NEON support in the compiler. If you see an error
+like...
+
+```
+/usr/lib/gcc/armv7l-unknown-linux-gnueabihf/9.2.0/include/arm_neon.h:635:1: error: inlining failed
+in call to always_inline ‘vaddq_u32’: target specific option mismatch
+```
+
+...then you may need to add something like `-mfpu=neon-vfpv4
+-mfloat-abi=hard`.
+
+### Other Platforms
+
+The portable implementation should work on most other architectures. For
+example:
+
+```bash
+gcc -shared -O3 -o libblake3.so blake3.c blake3_dispatch.c blake3_portable.c
+```
+
+### Multithreading
+
+Multithreading is available using [oneTBB], by compiling the optional C++
+support file [`blake3_tbb.cpp`](./blake3_tbb.cpp). For an example of using
+`mmap` (non-Windows) and `blake3_hasher_update_tbb` to get large-file
+performance on par with [`b3sum`](../b3sum), see
+[`example_tbb.c`](./example_tbb.c). You can build it like this:
+
+```bash
+g++ -c -O3 -fno-exceptions -fno-rtti -DBLAKE3_USE_TBB -o blake3_tbb.o blake3_tbb.cpp
+gcc -O3 -o example_tbb -lstdc++ -ltbb -DBLAKE3_USE_TBB blake3_tbb.o example_tbb.c blake3.c \
+    blake3_dispatch.c blake3_portable.c blake3_sse2_x86-64_unix.S blake3_sse41_x86-64_unix.S \
+    blake3_avx2_x86-64_unix.S blake3_avx512_x86-64_unix.S
+```
+
+NOTE: `-fno-exceptions` or equivalent is required to compile `blake3_tbb.cpp`,
+and public API methods with external C linkage are marked `noexcept`. Compiling
+that file with exceptions enabled will fail. Compiling with RTTI disabled isn't
+required but is recommended for code size.
diff --git a/thirdparty/blake3/c/blake3-config.cmake.in b/thirdparty/blake3/c/blake3-config.cmake.in
new file mode 100644
index 000000000..92cf395b8
--- /dev/null
+++ b/thirdparty/blake3/c/blake3-config.cmake.in
@@ -0,0 +1,14 @@
+@PACKAGE_INIT@
+
+include(CMakeFindDependencyMacro)
+
+# Remember TBB option state
+set(BLAKE3_USE_TBB @BLAKE3_USE_TBB@)
+
+if(BLAKE3_USE_TBB)
+    find_dependency(TBB @TBB_VERSION@)
+endif()
+
+include("${CMAKE_CURRENT_LIST_DIR}/blake3-targets.cmake")
+
+check_required_components(blake3)
diff --git a/thirdparty/blake3/c/blake3.c b/thirdparty/blake3/c/blake3.c
new file mode 100644
index 000000000..74fc2e73b
--- /dev/null
+++ b/thirdparty/blake3/c/blake3.c
@@ -0,0 +1,650 @@
+#include <assert.h>
+#include <stdbool.h>
+#include <string.h>
+
+#include "blake3.h"
+#include "blake3_impl.h"
+
+const char *blake3_version(void) { return BLAKE3_VERSION_STRING; }
+
+INLINE void chunk_state_init(blake3_chunk_state *self, const uint32_t key[8],
+                             uint8_t flags) {
+  memcpy(self->cv, key, BLAKE3_KEY_LEN);
+  self->chunk_counter = 0;
+  memset(self->buf, 0, BLAKE3_BLOCK_LEN);
+  self->buf_len = 0;
+  self->blocks_compressed = 0;
+  self->flags = flags;
+}
+
+INLINE void chunk_state_reset(blake3_chunk_state *self, const uint32_t key[8],
+                              uint64_t chunk_counter) {
+  memcpy(self->cv, key, BLAKE3_KEY_LEN);
+  self->chunk_counter = chunk_counter;
+  self->blocks_compressed = 0;
+  memset(self->buf, 0, BLAKE3_BLOCK_LEN);
+  self->buf_len = 0;
+}
+
+INLINE size_t chunk_state_len(const blake3_chunk_state *self) {
+  return (BLAKE3_BLOCK_LEN * (size_t)self->blocks_compressed) +
+         ((size_t)self->buf_len);
+}
+
+INLINE size_t chunk_state_fill_buf(blake3_chunk_state *self,
+                                   const uint8_t *input, size_t input_len) {
+  size_t take = BLAKE3_BLOCK_LEN - ((size_t)self->buf_len);
+  if (take > input_len) {
+    take = input_len;
+  }
+  uint8_t *dest = self->buf + ((size_t)self->buf_len);
+  memcpy(dest, input, take);
+  self->buf_len += (uint8_t)take;
+  return take;
+}
+
+INLINE uint8_t chunk_state_maybe_start_flag(const blake3_chunk_state *self) {
+  if (self->blocks_compressed == 0) {
+    return CHUNK_START;
+  } else {
+    return 0;
+  }
+}
+
+typedef struct {
+  uint32_t input_cv[8];
+  uint64_t counter;
+  uint8_t block[BLAKE3_BLOCK_LEN];
+  uint8_t block_len;
+  uint8_t flags;
+} output_t;
+
+INLINE output_t make_output(const uint32_t input_cv[8],
+                            const uint8_t block[BLAKE3_BLOCK_LEN],
+                            uint8_t block_len, uint64_t counter,
+                            uint8_t flags) {
+  output_t ret;
+  memcpy(ret.input_cv, input_cv, 32);
+  memcpy(ret.block, block, BLAKE3_BLOCK_LEN);
+  ret.block_len = block_len;
+  ret.counter = counter;
+  ret.flags = flags;
+  return ret;
+}
+
+// Chaining values within a given chunk (specifically the compress_in_place
+// interface) are represented as words. This avoids unnecessary bytes<->words
+// conversion overhead in the portable implementation. However, the hash_many
+// interface handles both user input and parent node blocks, so it accepts
+// bytes. For that reason, chaining values in the CV stack are represented as
+// bytes.
+INLINE void output_chaining_value(const output_t *self, uint8_t cv[32]) {
+  uint32_t cv_words[8];
+  memcpy(cv_words, self->input_cv, 32);
+  blake3_compress_in_place(cv_words, self->block, self->block_len,
+                           self->counter, self->flags);
+  store_cv_words(cv, cv_words);
+}
+
+INLINE void output_root_bytes(const output_t *self, uint64_t seek, uint8_t *out,
+                              size_t out_len) {
+  if (out_len == 0) {
+      return;
+  }
+  uint64_t output_block_counter = seek / 64;
+  size_t offset_within_block = seek % 64;
+  uint8_t wide_buf[64];
+  if(offset_within_block) {
+    blake3_compress_xof(self->input_cv, self->block, self->block_len, output_block_counter, self->flags | ROOT, wide_buf);
+    const size_t available_bytes = 64 - offset_within_block;
+    const size_t bytes = out_len > available_bytes ? available_bytes : out_len;
+    memcpy(out, wide_buf + offset_within_block, bytes);
+    out += bytes;
+    out_len -= bytes;
+    output_block_counter += 1;
+  }
+  if(out_len / 64) {
+    blake3_xof_many(self->input_cv, self->block, self->block_len, output_block_counter, self->flags | ROOT, out, out_len / 64);
+  }
+  output_block_counter += out_len / 64;
+  out += out_len & -64;
+  out_len -= out_len & -64;
+  if(out_len) {
+    blake3_compress_xof(self->input_cv, self->block, self->block_len, output_block_counter, self->flags | ROOT, wide_buf);
+    memcpy(out, wide_buf, out_len);
+  }
+}
+
+INLINE void chunk_state_update(blake3_chunk_state *self, const uint8_t *input,
+                               size_t input_len) {
+  if (self->buf_len > 0) {
+    size_t take = chunk_state_fill_buf(self, input, input_len);
+    input += take;
+    input_len -= take;
+    if (input_len > 0) {
+      blake3_compress_in_place(
+          self->cv, self->buf, BLAKE3_BLOCK_LEN, self->chunk_counter,
+          self->flags | chunk_state_maybe_start_flag(self));
+      self->blocks_compressed += 1;
+      self->buf_len = 0;
+      memset(self->buf, 0, BLAKE3_BLOCK_LEN);
+    }
+  }
+
+  while (input_len > BLAKE3_BLOCK_LEN) {
+    blake3_compress_in_place(self->cv, input, BLAKE3_BLOCK_LEN,
+                             self->chunk_counter,
+                             self->flags | chunk_state_maybe_start_flag(self));
+    self->blocks_compressed += 1;
+    input += BLAKE3_BLOCK_LEN;
+    input_len -= BLAKE3_BLOCK_LEN;
+  }
+
+  chunk_state_fill_buf(self, input, input_len);
+}
+
+INLINE output_t chunk_state_output(const blake3_chunk_state *self) {
+  uint8_t block_flags =
+      self->flags | chunk_state_maybe_start_flag(self) | CHUNK_END;
+  return make_output(self->cv, self->buf, self->buf_len, self->chunk_counter,
+                     block_flags);
+}
+
+INLINE output_t parent_output(const uint8_t block[BLAKE3_BLOCK_LEN],
+                              const uint32_t key[8], uint8_t flags) {
+  return make_output(key, block, BLAKE3_BLOCK_LEN, 0, flags | PARENT);
+}
+
+// Given some input larger than one chunk, return the number of bytes that
+// should go in the left subtree. This is the largest power-of-2 number of
+// chunks that leaves at least 1 byte for the right subtree.
+INLINE size_t left_subtree_len(size_t input_len) {
+  // Subtract 1 to reserve at least one byte for the right side. input_len
+  // should always be greater than BLAKE3_CHUNK_LEN.
+  size_t full_chunks = (input_len - 1) / BLAKE3_CHUNK_LEN;
+  return round_down_to_power_of_2(full_chunks) * BLAKE3_CHUNK_LEN;
+}
+
+// Use SIMD parallelism to hash up to MAX_SIMD_DEGREE chunks at the same time
+// on a single thread. Write out the chunk chaining values and return the
+// number of chunks hashed. These chunks are never the root and never empty;
+// those cases use a different codepath.
+INLINE size_t compress_chunks_parallel(const uint8_t *input, size_t input_len,
+                                       const uint32_t key[8],
+                                       uint64_t chunk_counter, uint8_t flags,
+                                       uint8_t *out) {
+#if defined(BLAKE3_TESTING)
+  assert(0 < input_len);
+  assert(input_len <= MAX_SIMD_DEGREE * BLAKE3_CHUNK_LEN);
+#endif
+
+  const uint8_t *chunks_array[MAX_SIMD_DEGREE];
+  size_t input_position = 0;
+  size_t chunks_array_len = 0;
+  while (input_len - input_position >= BLAKE3_CHUNK_LEN) {
+    chunks_array[chunks_array_len] = &input[input_position];
+    input_position += BLAKE3_CHUNK_LEN;
+    chunks_array_len += 1;
+  }
+
+  blake3_hash_many(chunks_array, chunks_array_len,
+                   BLAKE3_CHUNK_LEN / BLAKE3_BLOCK_LEN, key, chunk_counter,
+                   true, flags, CHUNK_START, CHUNK_END, out);
+
+  // Hash the remaining partial chunk, if there is one. Note that the empty
+  // chunk (meaning the empty message) is a different codepath.
+  if (input_len > input_position) {
+    uint64_t counter = chunk_counter + (uint64_t)chunks_array_len;
+    blake3_chunk_state chunk_state;
+    chunk_state_init(&chunk_state, key, flags);
+    chunk_state.chunk_counter = counter;
+    chunk_state_update(&chunk_state, &input[input_position],
+                       input_len - input_position);
+    output_t output = chunk_state_output(&chunk_state);
+    output_chaining_value(&output, &out[chunks_array_len * BLAKE3_OUT_LEN]);
+    return chunks_array_len + 1;
+  } else {
+    return chunks_array_len;
+  }
+}
+
+// Use SIMD parallelism to hash up to MAX_SIMD_DEGREE parents at the same time
+// on a single thread. Write out the parent chaining values and return the
+// number of parents hashed. (If there's an odd input chaining value left over,
+// return it as an additional output.) These parents are never the root and
+// never empty; those cases use a different codepath.
+INLINE size_t compress_parents_parallel(const uint8_t *child_chaining_values,
+                                        size_t num_chaining_values,
+                                        const uint32_t key[8], uint8_t flags,
+                                        uint8_t *out) {
+#if defined(BLAKE3_TESTING)
+  assert(2 <= num_chaining_values);
+  assert(num_chaining_values <= 2 * MAX_SIMD_DEGREE_OR_2);
+#endif
+
+  const uint8_t *parents_array[MAX_SIMD_DEGREE_OR_2];
+  size_t parents_array_len = 0;
+  while (num_chaining_values - (2 * parents_array_len) >= 2) {
+    parents_array[parents_array_len] =
+        &child_chaining_values[2 * parents_array_len * BLAKE3_OUT_LEN];
+    parents_array_len += 1;
+  }
+
+  blake3_hash_many(parents_array, parents_array_len, 1, key,
+                   0, // Parents always use counter 0.
+                   false, flags | PARENT,
+                   0, // Parents have no start flags.
+                   0, // Parents have no end flags.
+                   out);
+
+  // If there's an odd child left over, it becomes an output.
+  if (num_chaining_values > 2 * parents_array_len) {
+    memcpy(&out[parents_array_len * BLAKE3_OUT_LEN],
+           &child_chaining_values[2 * parents_array_len * BLAKE3_OUT_LEN],
+           BLAKE3_OUT_LEN);
+    return parents_array_len + 1;
+  } else {
+    return parents_array_len;
+  }
+}
+
+// The wide helper function returns (writes out) an array of chaining values
+// and returns the length of that array. The number of chaining values returned
+// is the dynamically detected SIMD degree, at most MAX_SIMD_DEGREE. Or fewer,
+// if the input is shorter than that many chunks. The reason for maintaining a
+// wide array of chaining values going back up the tree, is to allow the
+// implementation to hash as many parents in parallel as possible.
+//
+// As a special case when the SIMD degree is 1, this function will still return
+// at least 2 outputs. This guarantees that this function doesn't perform the
+// root compression. (If it did, it would use the wrong flags, and also we
+// wouldn't be able to implement extendable output.) Note that this function is
+// not used when the whole input is only 1 chunk long; that's a different
+// codepath.
+//
+// Why not just have the caller split the input on the first update(), instead
+// of implementing this special rule? Because we don't want to limit SIMD or
+// multi-threading parallelism for that update().
+size_t blake3_compress_subtree_wide(const uint8_t *input, size_t input_len,
+                                    const uint32_t key[8],
+                                    uint64_t chunk_counter, uint8_t flags,
+                                    uint8_t *out, bool use_tbb) {
+  // Note that the single chunk case does *not* bump the SIMD degree up to 2
+  // when it is 1. If this implementation adds multi-threading in the future,
+  // this gives us the option of multi-threading even the 2-chunk case, which
+  // can help performance on smaller platforms.
+  if (input_len <= blake3_simd_degree() * BLAKE3_CHUNK_LEN) {
+    return compress_chunks_parallel(input, input_len, key, chunk_counter, flags,
+                                    out);
+  }
+
+  // With more than simd_degree chunks, we need to recurse. Start by dividing
+  // the input into left and right subtrees. (Note that this is only optimal
+  // as long as the SIMD degree is a power of 2. If we ever get a SIMD degree
+  // of 3 or something, we'll need a more complicated strategy.)
+  size_t left_input_len = left_subtree_len(input_len);
+  size_t right_input_len = input_len - left_input_len;
+  const uint8_t *right_input = &input[left_input_len];
+  uint64_t right_chunk_counter =
+      chunk_counter + (uint64_t)(left_input_len / BLAKE3_CHUNK_LEN);
+
+  // Make space for the child outputs. Here we use MAX_SIMD_DEGREE_OR_2 to
+  // account for the special case of returning 2 outputs when the SIMD degree
+  // is 1.
+  uint8_t cv_array[2 * MAX_SIMD_DEGREE_OR_2 * BLAKE3_OUT_LEN];
+  size_t degree = blake3_simd_degree();
+  if (left_input_len > BLAKE3_CHUNK_LEN && degree == 1) {
+    // The special case: We always use a degree of at least two, to make
+    // sure there are two outputs. Except, as noted above, at the chunk
+    // level, where we allow degree=1. (Note that the 1-chunk-input case is
+    // a different codepath.)
+    degree = 2;
+  }
+  uint8_t *right_cvs = &cv_array[degree * BLAKE3_OUT_LEN];
+
+  // Recurse!
+  size_t left_n = -1;
+  size_t right_n = -1;
+
+#if defined(BLAKE3_USE_TBB)
+  blake3_compress_subtree_wide_join_tbb(
+      key, flags, use_tbb,
+      // left-hand side
+      input, left_input_len, chunk_counter, cv_array, &left_n,
+      // right-hand side
+      right_input, right_input_len, right_chunk_counter, right_cvs, &right_n);
+#else
+  left_n = blake3_compress_subtree_wide(
+      input, left_input_len, key, chunk_counter, flags, cv_array, use_tbb);
+  right_n = blake3_compress_subtree_wide(right_input, right_input_len, key,
+                                         right_chunk_counter, flags, right_cvs,
+                                         use_tbb);
+#endif // BLAKE3_USE_TBB
+
+  // The special case again. If simd_degree=1, then we'll have left_n=1 and
+  // right_n=1. Rather than compressing them into a single output, return
+  // them directly, to make sure we always have at least two outputs.
+  if (left_n == 1) {
+    memcpy(out, cv_array, 2 * BLAKE3_OUT_LEN);
+    return 2;
+  }
+
+  // Otherwise, do one layer of parent node compression.
+  size_t num_chaining_values = left_n + right_n;
+  return compress_parents_parallel(cv_array, num_chaining_values, key, flags,
+                                   out);
+}
+
+// Hash a subtree with compress_subtree_wide(), and then condense the resulting
+// list of chaining values down to a single parent node. Don't compress that
+// last parent node, however. Instead, return its message bytes (the
+// concatenated chaining values of its children). This is necessary when the
+// first call to update() supplies a complete subtree, because the topmost
+// parent node of that subtree could end up being the root. It's also necessary
+// for extended output in the general case.
+//
+// As with compress_subtree_wide(), this function is not used on inputs of 1
+// chunk or less. That's a different codepath.
+INLINE void
+compress_subtree_to_parent_node(const uint8_t *input, size_t input_len,
+                                const uint32_t key[8], uint64_t chunk_counter,
+                                uint8_t flags, uint8_t out[2 * BLAKE3_OUT_LEN],
+                                bool use_tbb) {
+#if defined(BLAKE3_TESTING)
+  assert(input_len > BLAKE3_CHUNK_LEN);
+#endif
+
+  uint8_t cv_array[MAX_SIMD_DEGREE_OR_2 * BLAKE3_OUT_LEN];
+  size_t num_cvs = blake3_compress_subtree_wide(input, input_len, key,
+                                                chunk_counter, flags, cv_array, use_tbb);
+  assert(num_cvs <= MAX_SIMD_DEGREE_OR_2);
+  // The following loop never executes when MAX_SIMD_DEGREE_OR_2 is 2, because
+  // as we just asserted, num_cvs will always be <=2 in that case. But GCC
+  // (particularly GCC 8.5) can't tell that it never executes, and if NDEBUG is
+  // set then it emits incorrect warnings here. We tried a few different
+  // hacks to silence these, but in the end our hacks just produced different
+  // warnings (see https://github.com/BLAKE3-team/BLAKE3/pull/380). Out of
+  // desperation, we ifdef out this entire loop when we know it's not needed.
+#if MAX_SIMD_DEGREE_OR_2 > 2
+  // If MAX_SIMD_DEGREE_OR_2 is greater than 2 and there's enough input,
+  // compress_subtree_wide() returns more than 2 chaining values. Condense
+  // them into 2 by forming parent nodes repeatedly.
+  uint8_t out_array[MAX_SIMD_DEGREE_OR_2 * BLAKE3_OUT_LEN / 2];
+  while (num_cvs > 2) {
+    num_cvs =
+        compress_parents_parallel(cv_array, num_cvs, key, flags, out_array);
+    memcpy(cv_array, out_array, num_cvs * BLAKE3_OUT_LEN);
+  }
+#endif
+  memcpy(out, cv_array, 2 * BLAKE3_OUT_LEN);
+}
+
+INLINE void hasher_init_base(blake3_hasher *self, const uint32_t key[8],
+                             uint8_t flags) {
+  memcpy(self->key, key, BLAKE3_KEY_LEN);
+  chunk_state_init(&self->chunk, key, flags);
+  self->cv_stack_len = 0;
+}
+
+void blake3_hasher_init(blake3_hasher *self) { hasher_init_base(self, IV, 0); }
+
+void blake3_hasher_init_keyed(blake3_hasher *self,
+                              const uint8_t key[BLAKE3_KEY_LEN]) {
+  uint32_t key_words[8];
+  load_key_words(key, key_words);
+  hasher_init_base(self, key_words, KEYED_HASH);
+}
+
+void blake3_hasher_init_derive_key_raw(blake3_hasher *self, const void *context,
+                                       size_t context_len) {
+  blake3_hasher context_hasher;
+  hasher_init_base(&context_hasher, IV, DERIVE_KEY_CONTEXT);
+  blake3_hasher_update(&context_hasher, context, context_len);
+  uint8_t context_key[BLAKE3_KEY_LEN];
+  blake3_hasher_finalize(&context_hasher, context_key, BLAKE3_KEY_LEN);
+  uint32_t context_key_words[8];
+  load_key_words(context_key, context_key_words);
+  hasher_init_base(self, context_key_words, DERIVE_KEY_MATERIAL);
+}
+
+void blake3_hasher_init_derive_key(blake3_hasher *self, const char *context) {
+  blake3_hasher_init_derive_key_raw(self, context, strlen(context));
+}
+
+// As described in hasher_push_cv() below, we do "lazy merging", delaying
+// merges until right before the next CV is about to be added. This is
+// different from the reference implementation. Another difference is that we
+// aren't always merging 1 chunk at a time. Instead, each CV might represent
+// any power-of-two number of chunks, as long as the smaller-above-larger stack
+// order is maintained. Instead of the "count the trailing 0-bits" algorithm
+// described in the spec, we use a "count the total number of 1-bits" variant
+// that doesn't require us to retain the subtree size of the CV on top of the
+// stack. The principle is the same: each CV that should remain in the stack is
+// represented by a 1-bit in the total number of chunks (or bytes) so far.
+INLINE void hasher_merge_cv_stack(blake3_hasher *self, uint64_t total_len) {
+  size_t post_merge_stack_len = (size_t)popcnt(total_len);
+  while (self->cv_stack_len > post_merge_stack_len) {
+    uint8_t *parent_node =
+        &self->cv_stack[(self->cv_stack_len - 2) * BLAKE3_OUT_LEN];
+    output_t output = parent_output(parent_node, self->key, self->chunk.flags);
+    output_chaining_value(&output, parent_node);
+    self->cv_stack_len -= 1;
+  }
+}
+
+// In reference_impl.rs, we merge the new CV with existing CVs from the stack
+// before pushing it. We can do that because we know more input is coming, so
+// we know none of the merges are root.
+//
+// This setting is different. We want to feed as much input as possible to
+// compress_subtree_wide(), without setting aside anything for the chunk_state.
+// If the user gives us 64 KiB, we want to parallelize over all 64 KiB at once
+// as a single subtree, if at all possible.
+//
+// This leads to two problems:
+// 1) This 64 KiB input might be the only call that ever gets made to update.
+//    In this case, the root node of the 64 KiB subtree would be the root node
+//    of the whole tree, and it would need to be ROOT finalized. We can't
+//    compress it until we know.
+// 2) This 64 KiB input might complete a larger tree, whose root node is
+//    similarly going to be the root of the whole tree. For example, maybe
+//    we have 196 KiB (that is, 128 + 64) hashed so far. We can't compress the
+//    node at the root of the 256 KiB subtree until we know how to finalize it.
+//
+// The second problem is solved with "lazy merging". That is, when we're about
+// to add a CV to the stack, we don't merge it with anything first, as the
+// reference impl does. Instead we do merges using the *previous* CV that was
+// added, which is sitting on top of the stack, and we put the new CV
+// (unmerged) on top of the stack afterwards. This guarantees that we never
+// merge the root node until finalize().
+//
+// Solving the first problem requires an additional tool,
+// compress_subtree_to_parent_node(). That function always returns the top
+// *two* chaining values of the subtree it's compressing. We then do lazy
+// merging with each of them separately, so that the second CV will always
+// remain unmerged. (That also helps us support extendable output when we're
+// hashing an input all-at-once.)
+INLINE void hasher_push_cv(blake3_hasher *self, uint8_t new_cv[BLAKE3_OUT_LEN],
+                           uint64_t chunk_counter) {
+  hasher_merge_cv_stack(self, chunk_counter);
+  memcpy(&self->cv_stack[self->cv_stack_len * BLAKE3_OUT_LEN], new_cv,
+         BLAKE3_OUT_LEN);
+  self->cv_stack_len += 1;
+}
+
+INLINE void blake3_hasher_update_base(blake3_hasher *self, const void *input,
+                                      size_t input_len, bool use_tbb) {
+  // Explicitly checking for zero avoids causing UB by passing a null pointer
+  // to memcpy. This comes up in practice with things like:
+  //   std::vector<uint8_t> v;
+  //   blake3_hasher_update(&hasher, v.data(), v.size());
+  if (input_len == 0) {
+    return;
+  }
+
+  const uint8_t *input_bytes = (const uint8_t *)input;
+
+  // If we have some partial chunk bytes in the internal chunk_state, we need
+  // to finish that chunk first.
+  if (chunk_state_len(&self->chunk) > 0) {
+    size_t take = BLAKE3_CHUNK_LEN - chunk_state_len(&self->chunk);
+    if (take > input_len) {
+      take = input_len;
+    }
+    chunk_state_update(&self->chunk, input_bytes, take);
+    input_bytes += take;
+    input_len -= take;
+    // If we've filled the current chunk and there's more coming, finalize this
+    // chunk and proceed. In this case we know it's not the root.
+    if (input_len > 0) {
+      output_t output = chunk_state_output(&self->chunk);
+      uint8_t chunk_cv[32];
+      output_chaining_value(&output, chunk_cv);
+      hasher_push_cv(self, chunk_cv, self->chunk.chunk_counter);
+      chunk_state_reset(&self->chunk, self->key, self->chunk.chunk_counter + 1);
+    } else {
+      return;
+    }
+  }
+
+  // Now the chunk_state is clear, and we have more input. If there's more than
+  // a single chunk (so, definitely not the root chunk), hash the largest whole
+  // subtree we can, with the full benefits of SIMD (and maybe in the future,
+  // multi-threading) parallelism. Two restrictions:
+  // - The subtree has to be a power-of-2 number of chunks. Only subtrees along
+  //   the right edge can be incomplete, and we don't know where the right edge
+  //   is going to be until we get to finalize().
+  // - The subtree must evenly divide the total number of chunks up until this
+  //   point (if total is not 0). If the current incomplete subtree is only
+  //   waiting for 1 more chunk, we can't hash a subtree of 4 chunks. We have
+  //   to complete the current subtree first.
+  // Because we might need to break up the input to form powers of 2, or to
+  // evenly divide what we already have, this part runs in a loop.
+  while (input_len > BLAKE3_CHUNK_LEN) {
+    size_t subtree_len = round_down_to_power_of_2(input_len);
+    uint64_t count_so_far = self->chunk.chunk_counter * BLAKE3_CHUNK_LEN;
+    // Shrink the subtree_len until it evenly divides the count so far. We know
+    // that subtree_len itself is a power of 2, so we can use a bitmasking
+    // trick instead of an actual remainder operation. (Note that if the caller
+    // consistently passes power-of-2 inputs of the same size, as is hopefully
+    // typical, this loop condition will always fail, and subtree_len will
+    // always be the full length of the input.)
+    //
+    // An aside: We don't have to shrink subtree_len quite this much. For
+    // example, if count_so_far is 1, we could pass 2 chunks to
+    // compress_subtree_to_parent_node. Since we'll get 2 CVs back, we'll still
+    // get the right answer in the end, and we might get to use 2-way SIMD
+    // parallelism. The problem with this optimization, is that it gets us
+    // stuck always hashing 2 chunks. The total number of chunks will remain
+    // odd, and we'll never graduate to higher degrees of parallelism. See
+    // https://github.com/BLAKE3-team/BLAKE3/issues/69.
+    while ((((uint64_t)(subtree_len - 1)) & count_so_far) != 0) {
+      subtree_len /= 2;
+    }
+    // The shrunken subtree_len might now be 1 chunk long. If so, hash that one
+    // chunk by itself. Otherwise, compress the subtree into a pair of CVs.
+    uint64_t subtree_chunks = subtree_len / BLAKE3_CHUNK_LEN;
+    if (subtree_len <= BLAKE3_CHUNK_LEN) {
+      blake3_chunk_state chunk_state;
+      chunk_state_init(&chunk_state, self->key, self->chunk.flags);
+      chunk_state.chunk_counter = self->chunk.chunk_counter;
+      chunk_state_update(&chunk_state, input_bytes, subtree_len);
+      output_t output = chunk_state_output(&chunk_state);
+      uint8_t cv[BLAKE3_OUT_LEN];
+      output_chaining_value(&output, cv);
+      hasher_push_cv(self, cv, chunk_state.chunk_counter);
+    } else {
+      // This is the high-performance happy path, though getting here depends
+      // on the caller giving us a long enough input.
+      uint8_t cv_pair[2 * BLAKE3_OUT_LEN];
+      compress_subtree_to_parent_node(input_bytes, subtree_len, self->key,
+                                      self->chunk.chunk_counter,
+                                      self->chunk.flags, cv_pair, use_tbb);
+      hasher_push_cv(self, cv_pair, self->chunk.chunk_counter);
+      hasher_push_cv(self, &cv_pair[BLAKE3_OUT_LEN],
+                     self->chunk.chunk_counter + (subtree_chunks / 2));
+    }
+    self->chunk.chunk_counter += subtree_chunks;
+    input_bytes += subtree_len;
+    input_len -= subtree_len;
+  }
+
+  // If there's any remaining input less than a full chunk, add it to the chunk
+  // state. In that case, also do a final merge loop to make sure the subtree
+  // stack doesn't contain any unmerged pairs. The remaining input means we
+  // know these merges are non-root. This merge loop isn't strictly necessary
+  // here, because hasher_push_chunk_cv already does its own merge loop, but it
+  // simplifies blake3_hasher_finalize below.
+  if (input_len > 0) {
+    chunk_state_update(&self->chunk, input_bytes, input_len);
+    hasher_merge_cv_stack(self, self->chunk.chunk_counter);
+  }
+}
+
+void blake3_hasher_update(blake3_hasher *self, const void *input,
+                          size_t input_len) {
+  bool use_tbb = false;
+  blake3_hasher_update_base(self, input, input_len, use_tbb);
+}
+
+#if defined(BLAKE3_USE_TBB)
+void blake3_hasher_update_tbb(blake3_hasher *self, const void *input,
+                              size_t input_len) {
+  bool use_tbb = true;
+  blake3_hasher_update_base(self, input, input_len, use_tbb);
+}
+#endif // BLAKE3_USE_TBB
+
+void blake3_hasher_finalize(const blake3_hasher *self, uint8_t *out,
+                            size_t out_len) {
+  blake3_hasher_finalize_seek(self, 0, out, out_len);
+}
+
+void blake3_hasher_finalize_seek(const blake3_hasher *self, uint64_t seek,
+                                 uint8_t *out, size_t out_len) {
+  // Explicitly checking for zero avoids causing UB by passing a null pointer
+  // to memcpy. This comes up in practice with things like:
+  //   std::vector<uint8_t> v;
+  //   blake3_hasher_finalize(&hasher, v.data(), v.size());
+  if (out_len == 0) {
+    return;
+  }
+
+  // If the subtree stack is empty, then the current chunk is the root.
+  if (self->cv_stack_len == 0) {
+    output_t output = chunk_state_output(&self->chunk);
+    output_root_bytes(&output, seek, out, out_len);
+    return;
+  }
+  // If there are any bytes in the chunk state, finalize that chunk and do a
+  // roll-up merge between that chunk hash and every subtree in the stack. In
+  // this case, the extra merge loop at the end of blake3_hasher_update
+  // guarantees that none of the subtrees in the stack need to be merged with
+  // each other first. Otherwise, if there are no bytes in the chunk state,
+  // then the top of the stack is a chunk hash, and we start the merge from
+  // that.
+  output_t output;
+  size_t cvs_remaining;
+  if (chunk_state_len(&self->chunk) > 0) {
+    cvs_remaining = self->cv_stack_len;
+    output = chunk_state_output(&self->chunk);
+  } else {
+    // There are always at least 2 CVs in the stack in this case.
+    cvs_remaining = self->cv_stack_len - 2;
+    output = parent_output(&self->cv_stack[cvs_remaining * 32], self->key,
+                           self->chunk.flags);
+  }
+  while (cvs_remaining > 0) {
+    cvs_remaining -= 1;
+    uint8_t parent_block[BLAKE3_BLOCK_LEN];
+    memcpy(parent_block, &self->cv_stack[cvs_remaining * 32], 32);
+    output_chaining_value(&output, &parent_block[32]);
+    output = parent_output(parent_block, self->key, self->chunk.flags);
+  }
+  output_root_bytes(&output, seek, out, out_len);
+}
+
+void blake3_hasher_reset(blake3_hasher *self) {
+  chunk_state_reset(&self->chunk, self->key, 0);
+  self->cv_stack_len = 0;
+}
diff --git a/thirdparty/blake3/c/blake3.h b/thirdparty/blake3/c/blake3.h
new file mode 100644
index 000000000..08932954c
--- /dev/null
+++ b/thirdparty/blake3/c/blake3.h
@@ -0,0 +1,86 @@
+#ifndef BLAKE3_H
+#define BLAKE3_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+#if !defined(BLAKE3_API)
+# if defined(_WIN32) || defined(__CYGWIN__)
+#   if defined(BLAKE3_DLL)
+#     if defined(BLAKE3_DLL_EXPORTS)
+#       define BLAKE3_API __declspec(dllexport)
+#     else
+#       define BLAKE3_API __declspec(dllimport)
+#     endif
+#     define BLAKE3_PRIVATE
+#   else
+#     define BLAKE3_API
+#     define BLAKE3_PRIVATE
+#   endif
+# elif __GNUC__ >= 4
+#   define BLAKE3_API __attribute__((visibility("default")))
+#   define BLAKE3_PRIVATE __attribute__((visibility("hidden")))
+# else
+#   define BLAKE3_API
+#   define BLAKE3_PRIVATE
+# endif
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define BLAKE3_VERSION_STRING "1.8.2"
+#define BLAKE3_KEY_LEN 32
+#define BLAKE3_OUT_LEN 32
+#define BLAKE3_BLOCK_LEN 64
+#define BLAKE3_CHUNK_LEN 1024
+#define BLAKE3_MAX_DEPTH 54
+
+// This struct is a private implementation detail. It has to be here because
+// it's part of blake3_hasher below.
+typedef struct {
+  uint32_t cv[8];
+  uint64_t chunk_counter;
+  uint8_t buf[BLAKE3_BLOCK_LEN];
+  uint8_t buf_len;
+  uint8_t blocks_compressed;
+  uint8_t flags;
+} blake3_chunk_state;
+
+typedef struct {
+  uint32_t key[8];
+  blake3_chunk_state chunk;
+  uint8_t cv_stack_len;
+  // The stack size is MAX_DEPTH + 1 because we do lazy merging. For example,
+  // with 7 chunks, we have 3 entries in the stack. Adding an 8th chunk
+  // requires a 4th entry, rather than merging everything down to 1, because we
+  // don't know whether more input is coming. This is different from how the
+  // reference implementation does things.
+  uint8_t cv_stack[(BLAKE3_MAX_DEPTH + 1) * BLAKE3_OUT_LEN];
+} blake3_hasher;
+
+BLAKE3_API const char *blake3_version(void);
+BLAKE3_API void blake3_hasher_init(blake3_hasher *self);
+BLAKE3_API void blake3_hasher_init_keyed(blake3_hasher *self,
+                                         const uint8_t key[BLAKE3_KEY_LEN]);
+BLAKE3_API void blake3_hasher_init_derive_key(blake3_hasher *self, const char *context);
+BLAKE3_API void blake3_hasher_init_derive_key_raw(blake3_hasher *self, const void *context,
+                                                  size_t context_len);
+BLAKE3_API void blake3_hasher_update(blake3_hasher *self, const void *input,
+                                     size_t input_len);
+#if defined(BLAKE3_USE_TBB)
+BLAKE3_API void blake3_hasher_update_tbb(blake3_hasher *self, const void *input,
+                                         size_t input_len);
+#endif // BLAKE3_USE_TBB
+BLAKE3_API void blake3_hasher_finalize(const blake3_hasher *self, uint8_t *out,
+                                       size_t out_len);
+BLAKE3_API void blake3_hasher_finalize_seek(const blake3_hasher *self, uint64_t seek,
+                                            uint8_t *out, size_t out_len);
+BLAKE3_API void blake3_hasher_reset(blake3_hasher *self);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* BLAKE3_H */
diff --git a/thirdparty/blake3/c/blake3_avx2.c b/thirdparty/blake3/c/blake3_avx2.c
new file mode 100644
index 000000000..381e7c422
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx2.c
@@ -0,0 +1,326 @@
+#include "blake3_impl.h"
+
+#include <immintrin.h>
+
+#define DEGREE 8
+
+INLINE __m256i loadu(const uint8_t src[32]) {
+  return _mm256_loadu_si256((const __m256i *)src);
+}
+
+INLINE void storeu(__m256i src, uint8_t dest[16]) {
+  _mm256_storeu_si256((__m256i *)dest, src);
+}
+
+INLINE __m256i addv(__m256i a, __m256i b) { return _mm256_add_epi32(a, b); }
+
+// Note that clang-format doesn't like the name "xor" for some reason.
+INLINE __m256i xorv(__m256i a, __m256i b) { return _mm256_xor_si256(a, b); }
+
+INLINE __m256i set1(uint32_t x) { return _mm256_set1_epi32((int32_t)x); }
+
+INLINE __m256i rot16(__m256i x) {
+  return _mm256_shuffle_epi8(
+      x, _mm256_set_epi8(13, 12, 15, 14, 9, 8, 11, 10, 5, 4, 7, 6, 1, 0, 3, 2,
+                         13, 12, 15, 14, 9, 8, 11, 10, 5, 4, 7, 6, 1, 0, 3, 2));
+}
+
+INLINE __m256i rot12(__m256i x) {
+  return _mm256_or_si256(_mm256_srli_epi32(x, 12), _mm256_slli_epi32(x, 32 - 12));
+}
+
+INLINE __m256i rot8(__m256i x) {
+  return _mm256_shuffle_epi8(
+      x, _mm256_set_epi8(12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1,
+                         12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1));
+}
+
+INLINE __m256i rot7(__m256i x) {
+  return _mm256_or_si256(_mm256_srli_epi32(x, 7), _mm256_slli_epi32(x, 32 - 7));
+}
+
+INLINE void round_fn(__m256i v[16], __m256i m[16], size_t r) {
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
+  v[0] = addv(v[0], v[4]);
+  v[1] = addv(v[1], v[5]);
+  v[2] = addv(v[2], v[6]);
+  v[3] = addv(v[3], v[7]);
+  v[12] = xorv(v[12], v[0]);
+  v[13] = xorv(v[13], v[1]);
+  v[14] = xorv(v[14], v[2]);
+  v[15] = xorv(v[15], v[3]);
+  v[12] = rot16(v[12]);
+  v[13] = rot16(v[13]);
+  v[14] = rot16(v[14]);
+  v[15] = rot16(v[15]);
+  v[8] = addv(v[8], v[12]);
+  v[9] = addv(v[9], v[13]);
+  v[10] = addv(v[10], v[14]);
+  v[11] = addv(v[11], v[15]);
+  v[4] = xorv(v[4], v[8]);
+  v[5] = xorv(v[5], v[9]);
+  v[6] = xorv(v[6], v[10]);
+  v[7] = xorv(v[7], v[11]);
+  v[4] = rot12(v[4]);
+  v[5] = rot12(v[5]);
+  v[6] = rot12(v[6]);
+  v[7] = rot12(v[7]);
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
+  v[0] = addv(v[0], v[4]);
+  v[1] = addv(v[1], v[5]);
+  v[2] = addv(v[2], v[6]);
+  v[3] = addv(v[3], v[7]);
+  v[12] = xorv(v[12], v[0]);
+  v[13] = xorv(v[13], v[1]);
+  v[14] = xorv(v[14], v[2]);
+  v[15] = xorv(v[15], v[3]);
+  v[12] = rot8(v[12]);
+  v[13] = rot8(v[13]);
+  v[14] = rot8(v[14]);
+  v[15] = rot8(v[15]);
+  v[8] = addv(v[8], v[12]);
+  v[9] = addv(v[9], v[13]);
+  v[10] = addv(v[10], v[14]);
+  v[11] = addv(v[11], v[15]);
+  v[4] = xorv(v[4], v[8]);
+  v[5] = xorv(v[5], v[9]);
+  v[6] = xorv(v[6], v[10]);
+  v[7] = xorv(v[7], v[11]);
+  v[4] = rot7(v[4]);
+  v[5] = rot7(v[5]);
+  v[6] = rot7(v[6]);
+  v[7] = rot7(v[7]);
+
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
+  v[0] = addv(v[0], v[5]);
+  v[1] = addv(v[1], v[6]);
+  v[2] = addv(v[2], v[7]);
+  v[3] = addv(v[3], v[4]);
+  v[15] = xorv(v[15], v[0]);
+  v[12] = xorv(v[12], v[1]);
+  v[13] = xorv(v[13], v[2]);
+  v[14] = xorv(v[14], v[3]);
+  v[15] = rot16(v[15]);
+  v[12] = rot16(v[12]);
+  v[13] = rot16(v[13]);
+  v[14] = rot16(v[14]);
+  v[10] = addv(v[10], v[15]);
+  v[11] = addv(v[11], v[12]);
+  v[8] = addv(v[8], v[13]);
+  v[9] = addv(v[9], v[14]);
+  v[5] = xorv(v[5], v[10]);
+  v[6] = xorv(v[6], v[11]);
+  v[7] = xorv(v[7], v[8]);
+  v[4] = xorv(v[4], v[9]);
+  v[5] = rot12(v[5]);
+  v[6] = rot12(v[6]);
+  v[7] = rot12(v[7]);
+  v[4] = rot12(v[4]);
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
+  v[0] = addv(v[0], v[5]);
+  v[1] = addv(v[1], v[6]);
+  v[2] = addv(v[2], v[7]);
+  v[3] = addv(v[3], v[4]);
+  v[15] = xorv(v[15], v[0]);
+  v[12] = xorv(v[12], v[1]);
+  v[13] = xorv(v[13], v[2]);
+  v[14] = xorv(v[14], v[3]);
+  v[15] = rot8(v[15]);
+  v[12] = rot8(v[12]);
+  v[13] = rot8(v[13]);
+  v[14] = rot8(v[14]);
+  v[10] = addv(v[10], v[15]);
+  v[11] = addv(v[11], v[12]);
+  v[8] = addv(v[8], v[13]);
+  v[9] = addv(v[9], v[14]);
+  v[5] = xorv(v[5], v[10]);
+  v[6] = xorv(v[6], v[11]);
+  v[7] = xorv(v[7], v[8]);
+  v[4] = xorv(v[4], v[9]);
+  v[5] = rot7(v[5]);
+  v[6] = rot7(v[6]);
+  v[7] = rot7(v[7]);
+  v[4] = rot7(v[4]);
+}
+
+INLINE void transpose_vecs(__m256i vecs[DEGREE]) {
+  // Interleave 32-bit lanes. The low unpack is lanes 00/11/44/55, and the high
+  // is 22/33/66/77.
+  __m256i ab_0145 = _mm256_unpacklo_epi32(vecs[0], vecs[1]);
+  __m256i ab_2367 = _mm256_unpackhi_epi32(vecs[0], vecs[1]);
+  __m256i cd_0145 = _mm256_unpacklo_epi32(vecs[2], vecs[3]);
+  __m256i cd_2367 = _mm256_unpackhi_epi32(vecs[2], vecs[3]);
+  __m256i ef_0145 = _mm256_unpacklo_epi32(vecs[4], vecs[5]);
+  __m256i ef_2367 = _mm256_unpackhi_epi32(vecs[4], vecs[5]);
+  __m256i gh_0145 = _mm256_unpacklo_epi32(vecs[6], vecs[7]);
+  __m256i gh_2367 = _mm256_unpackhi_epi32(vecs[6], vecs[7]);
+
+  // Interleave 64-bit lanes. The low unpack is lanes 00/22 and the high is
+  // 11/33.
+  __m256i abcd_04 = _mm256_unpacklo_epi64(ab_0145, cd_0145);
+  __m256i abcd_15 = _mm256_unpackhi_epi64(ab_0145, cd_0145);
+  __m256i abcd_26 = _mm256_unpacklo_epi64(ab_2367, cd_2367);
+  __m256i abcd_37 = _mm256_unpackhi_epi64(ab_2367, cd_2367);
+  __m256i efgh_04 = _mm256_unpacklo_epi64(ef_0145, gh_0145);
+  __m256i efgh_15 = _mm256_unpackhi_epi64(ef_0145, gh_0145);
+  __m256i efgh_26 = _mm256_unpacklo_epi64(ef_2367, gh_2367);
+  __m256i efgh_37 = _mm256_unpackhi_epi64(ef_2367, gh_2367);
+
+  // Interleave 128-bit lanes.
+  vecs[0] = _mm256_permute2x128_si256(abcd_04, efgh_04, 0x20);
+  vecs[1] = _mm256_permute2x128_si256(abcd_15, efgh_15, 0x20);
+  vecs[2] = _mm256_permute2x128_si256(abcd_26, efgh_26, 0x20);
+  vecs[3] = _mm256_permute2x128_si256(abcd_37, efgh_37, 0x20);
+  vecs[4] = _mm256_permute2x128_si256(abcd_04, efgh_04, 0x31);
+  vecs[5] = _mm256_permute2x128_si256(abcd_15, efgh_15, 0x31);
+  vecs[6] = _mm256_permute2x128_si256(abcd_26, efgh_26, 0x31);
+  vecs[7] = _mm256_permute2x128_si256(abcd_37, efgh_37, 0x31);
+}
+
+INLINE void transpose_msg_vecs(const uint8_t *const *inputs,
+                               size_t block_offset, __m256i out[16]) {
+  out[0] = loadu(&inputs[0][block_offset + 0 * sizeof(__m256i)]);
+  out[1] = loadu(&inputs[1][block_offset + 0 * sizeof(__m256i)]);
+  out[2] = loadu(&inputs[2][block_offset + 0 * sizeof(__m256i)]);
+  out[3] = loadu(&inputs[3][block_offset + 0 * sizeof(__m256i)]);
+  out[4] = loadu(&inputs[4][block_offset + 0 * sizeof(__m256i)]);
+  out[5] = loadu(&inputs[5][block_offset + 0 * sizeof(__m256i)]);
+  out[6] = loadu(&inputs[6][block_offset + 0 * sizeof(__m256i)]);
+  out[7] = loadu(&inputs[7][block_offset + 0 * sizeof(__m256i)]);
+  out[8] = loadu(&inputs[0][block_offset + 1 * sizeof(__m256i)]);
+  out[9] = loadu(&inputs[1][block_offset + 1 * sizeof(__m256i)]);
+  out[10] = loadu(&inputs[2][block_offset + 1 * sizeof(__m256i)]);
+  out[11] = loadu(&inputs[3][block_offset + 1 * sizeof(__m256i)]);
+  out[12] = loadu(&inputs[4][block_offset + 1 * sizeof(__m256i)]);
+  out[13] = loadu(&inputs[5][block_offset + 1 * sizeof(__m256i)]);
+  out[14] = loadu(&inputs[6][block_offset + 1 * sizeof(__m256i)]);
+  out[15] = loadu(&inputs[7][block_offset + 1 * sizeof(__m256i)]);
+  for (size_t i = 0; i < 8; ++i) {
+    _mm_prefetch((const void *)&inputs[i][block_offset + 256], _MM_HINT_T0);
+  }
+  transpose_vecs(&out[0]);
+  transpose_vecs(&out[8]);
+}
+
+INLINE void load_counters(uint64_t counter, bool increment_counter,
+                          __m256i *out_lo, __m256i *out_hi) {
+  const __m256i mask = _mm256_set1_epi32(-(int32_t)increment_counter);
+  const __m256i add0 = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0);
+  const __m256i add1 = _mm256_and_si256(mask, add0);
+  __m256i l = _mm256_add_epi32(_mm256_set1_epi32((int32_t)counter), add1);
+  __m256i carry = _mm256_cmpgt_epi32(_mm256_xor_si256(add1, _mm256_set1_epi32(0x80000000)), 
+                                     _mm256_xor_si256(   l, _mm256_set1_epi32(0x80000000)));
+  __m256i h = _mm256_sub_epi32(_mm256_set1_epi32((int32_t)(counter >> 32)), carry);
+  *out_lo = l;
+  *out_hi = h;
+}
+
+static
+void blake3_hash8_avx2(const uint8_t *const *inputs, size_t blocks,
+                       const uint32_t key[8], uint64_t counter,
+                       bool increment_counter, uint8_t flags,
+                       uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
+  __m256i h_vecs[8] = {
+      set1(key[0]), set1(key[1]), set1(key[2]), set1(key[3]),
+      set1(key[4]), set1(key[5]), set1(key[6]), set1(key[7]),
+  };
+  __m256i counter_low_vec, counter_high_vec;
+  load_counters(counter, increment_counter, &counter_low_vec,
+                &counter_high_vec);
+  uint8_t block_flags = flags | flags_start;
+
+  for (size_t block = 0; block < blocks; block++) {
+    if (block + 1 == blocks) {
+      block_flags |= flags_end;
+    }
+    __m256i block_len_vec = set1(BLAKE3_BLOCK_LEN);
+    __m256i block_flags_vec = set1(block_flags);
+    __m256i msg_vecs[16];
+    transpose_msg_vecs(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);
+
+    __m256i v[16] = {
+        h_vecs[0],       h_vecs[1],        h_vecs[2],     h_vecs[3],
+        h_vecs[4],       h_vecs[5],        h_vecs[6],     h_vecs[7],
+        set1(IV[0]),     set1(IV[1]),      set1(IV[2]),   set1(IV[3]),
+        counter_low_vec, counter_high_vec, block_len_vec, block_flags_vec,
+    };
+    round_fn(v, msg_vecs, 0);
+    round_fn(v, msg_vecs, 1);
+    round_fn(v, msg_vecs, 2);
+    round_fn(v, msg_vecs, 3);
+    round_fn(v, msg_vecs, 4);
+    round_fn(v, msg_vecs, 5);
+    round_fn(v, msg_vecs, 6);
+    h_vecs[0] = xorv(v[0], v[8]);
+    h_vecs[1] = xorv(v[1], v[9]);
+    h_vecs[2] = xorv(v[2], v[10]);
+    h_vecs[3] = xorv(v[3], v[11]);
+    h_vecs[4] = xorv(v[4], v[12]);
+    h_vecs[5] = xorv(v[5], v[13]);
+    h_vecs[6] = xorv(v[6], v[14]);
+    h_vecs[7] = xorv(v[7], v[15]);
+
+    block_flags = flags;
+  }
+
+  transpose_vecs(h_vecs);
+  storeu(h_vecs[0], &out[0 * sizeof(__m256i)]);
+  storeu(h_vecs[1], &out[1 * sizeof(__m256i)]);
+  storeu(h_vecs[2], &out[2 * sizeof(__m256i)]);
+  storeu(h_vecs[3], &out[3 * sizeof(__m256i)]);
+  storeu(h_vecs[4], &out[4 * sizeof(__m256i)]);
+  storeu(h_vecs[5], &out[5 * sizeof(__m256i)]);
+  storeu(h_vecs[6], &out[6 * sizeof(__m256i)]);
+  storeu(h_vecs[7], &out[7 * sizeof(__m256i)]);
+}
+
+#if !defined(BLAKE3_NO_SSE41)
+void blake3_hash_many_sse41(const uint8_t *const *inputs, size_t num_inputs,
+                            size_t blocks, const uint32_t key[8],
+                            uint64_t counter, bool increment_counter,
+                            uint8_t flags, uint8_t flags_start,
+                            uint8_t flags_end, uint8_t *out);
+#else
+void blake3_hash_many_portable(const uint8_t *const *inputs, size_t num_inputs,
+                               size_t blocks, const uint32_t key[8],
+                               uint64_t counter, bool increment_counter,
+                               uint8_t flags, uint8_t flags_start,
+                               uint8_t flags_end, uint8_t *out);
+#endif
+
+void blake3_hash_many_avx2(const uint8_t *const *inputs, size_t num_inputs,
+                           size_t blocks, const uint32_t key[8],
+                           uint64_t counter, bool increment_counter,
+                           uint8_t flags, uint8_t flags_start,
+                           uint8_t flags_end, uint8_t *out) {
+  while (num_inputs >= DEGREE) {
+    blake3_hash8_avx2(inputs, blocks, key, counter, increment_counter, flags,
+                      flags_start, flags_end, out);
+    if (increment_counter) {
+      counter += DEGREE;
+    }
+    inputs += DEGREE;
+    num_inputs -= DEGREE;
+    out = &out[DEGREE * BLAKE3_OUT_LEN];
+  }
+#if !defined(BLAKE3_NO_SSE41)
+  blake3_hash_many_sse41(inputs, num_inputs, blocks, key, counter,
+                         increment_counter, flags, flags_start, flags_end, out);
+#else
+  blake3_hash_many_portable(inputs, num_inputs, blocks, key, counter,
+                            increment_counter, flags, flags_start, flags_end,
+                            out);
+#endif
+}
diff --git a/thirdparty/blake3/c/blake3_avx2_x86-64_unix.S b/thirdparty/blake3/c/blake3_avx2_x86-64_unix.S
new file mode 100644
index 000000000..812bb8568
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx2_x86-64_unix.S
@@ -0,0 +1,1815 @@
+#if defined(__ELF__) && defined(__linux__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+#if defined(__ELF__) && defined(__CET__) && defined(__has_include)
+#if __has_include(<cet.h>)
+#include <cet.h>
+#endif
+#endif
+
+#if !defined(_CET_ENDBR)
+#define _CET_ENDBR
+#endif
+
+.intel_syntax noprefix
+.global _blake3_hash_many_avx2
+.global blake3_hash_many_avx2
+#ifdef __APPLE__
+.text
+#else
+.section .text
+#endif
+        .p2align  6
+_blake3_hash_many_avx2:
+blake3_hash_many_avx2:
+        _CET_ENDBR
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 680
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        neg     r9d
+        vmovd   xmm0, r9d
+        vpbroadcastd ymm0, xmm0
+        vmovdqa ymmword ptr [rsp+0x280], ymm0
+        vpand   ymm1, ymm0, ymmword ptr [ADD0+rip]
+        vpand   ymm2, ymm0, ymmword ptr [ADD1+rip]
+        vmovdqa ymmword ptr [rsp+0x220], ymm2
+        vmovd   xmm2, r8d
+        vpbroadcastd ymm2, xmm2
+        vpaddd  ymm2, ymm2, ymm1
+        vmovdqa ymmword ptr [rsp+0x240], ymm2
+        vpxor   ymm1, ymm1, ymmword ptr [CMP_MSB_MASK+rip]
+        vpxor   ymm2, ymm2, ymmword ptr [CMP_MSB_MASK+rip]
+        vpcmpgtd ymm2, ymm1, ymm2
+        shr     r8, 32
+        vmovd   xmm3, r8d
+        vpbroadcastd ymm3, xmm3
+        vpsubd  ymm3, ymm3, ymm2
+        vmovdqa ymmword ptr [rsp+0x260], ymm3
+        shl     rdx, 6
+        mov     qword ptr [rsp+0x2A0], rdx
+        cmp     rsi, 8
+        jc      3f
+2:
+        vpbroadcastd ymm0, dword ptr [rcx]
+        vpbroadcastd ymm1, dword ptr [rcx+0x4]
+        vpbroadcastd ymm2, dword ptr [rcx+0x8]
+        vpbroadcastd ymm3, dword ptr [rcx+0xC]
+        vpbroadcastd ymm4, dword ptr [rcx+0x10]
+        vpbroadcastd ymm5, dword ptr [rcx+0x14]
+        vpbroadcastd ymm6, dword ptr [rcx+0x18]
+        vpbroadcastd ymm7, dword ptr [rcx+0x1C]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x20]
+        mov     r13, qword ptr [rdi+0x28]
+        mov     r14, qword ptr [rdi+0x30]
+        mov     r15, qword ptr [rdi+0x38]
+        movzx   eax, byte ptr [rbp+0x38]
+        movzx   ebx, byte ptr [rbp+0x40]
+        or      eax, ebx
+        xor     edx, edx
+.p2align  5
+9:
+        movzx   ebx, byte ptr [rbp+0x48]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+0x2A0]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+0x200], eax
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x40], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x40]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x40], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x40]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x40], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x40]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x40], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0x20], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0x40], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0x60], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x30], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x30], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x30]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x30], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x30]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x30], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+0x80], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0xA0], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0xC0], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0xE0], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x20], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x20]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x20], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x20]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x20], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x20]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x20], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+0x100], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0x120], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0x140], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0x160], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x10], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x10], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x10]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x10], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x10]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x10], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+0x180], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0x1A0], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0x1C0], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0x1E0], ymm11
+        vpbroadcastd ymm15, dword ptr [rsp+0x200]
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r12+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r13+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r14+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        prefetcht0 [r15+rdx+0x80]
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x40]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x80]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm0, ymmword ptr [rsp+0x240]
+        vpxor   ymm13, ymm1, ymmword ptr [rsp+0x260]
+        vpxor   ymm14, ymm2, ymmword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpxor   ymm15, ymm3, ymm15
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [BLAKE3_IV_0+rip]
+        vpaddd  ymm9, ymm13, ymmword ptr [BLAKE3_IV_1+rip]
+        vpaddd  ymm10, ymm14, ymmword ptr [BLAKE3_IV_2+rip]
+        vpaddd  ymm11, ymm15, ymmword ptr [BLAKE3_IV_3+rip]
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x20]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x60]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x100]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x180]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x120]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x40]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x60]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x80]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x20]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x120]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x160]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x60]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x80]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x40]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x120]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x160]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x20]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x140]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x120]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x60]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x80]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x20]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x40]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x100]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x180]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x120]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x140]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x40]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x60]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x20]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x80]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x120]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x100]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x180]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x20]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x40]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x80]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x60]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x160]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x20]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x120]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x40]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x60]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x140]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x80]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vpxor   ymm0, ymm0, ymm8
+        vpxor   ymm1, ymm1, ymm9
+        vpxor   ymm2, ymm2, ymm10
+        vpxor   ymm3, ymm3, ymm11
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpxor   ymm4, ymm4, ymm12
+        vpxor   ymm5, ymm5, ymm13
+        vpxor   ymm6, ymm6, ymm14
+        vpxor   ymm7, ymm7, ymm15
+        movzx   eax, byte ptr [rbp+0x38]
+        jne     9b
+        mov     rbx, qword ptr [rbp+0x50]
+        vunpcklps ymm8, ymm0, ymm1
+        vunpcklps ymm9, ymm2, ymm3
+        vunpckhps ymm10, ymm0, ymm1
+        vunpcklps ymm11, ymm4, ymm5
+        vunpcklps ymm0, ymm6, ymm7
+        vshufps ymm12, ymm8, ymm9, 78
+        vblendps ymm1, ymm8, ymm12, 0xCC
+        vshufps ymm8, ymm11, ymm0, 78
+        vunpckhps ymm13, ymm2, ymm3
+        vblendps ymm2, ymm11, ymm8, 0xCC
+        vblendps ymm3, ymm12, ymm9, 0xCC
+        vperm2f128 ymm12, ymm1, ymm2, 0x20
+        vmovups ymmword ptr [rbx], ymm12
+        vunpckhps ymm14, ymm4, ymm5
+        vblendps ymm4, ymm8, ymm0, 0xCC
+        vunpckhps ymm15, ymm6, ymm7
+        vperm2f128 ymm7, ymm3, ymm4, 0x20
+        vmovups ymmword ptr [rbx+0x20], ymm7
+        vshufps ymm5, ymm10, ymm13, 78
+        vblendps ymm6, ymm5, ymm13, 0xCC
+        vshufps ymm13, ymm14, ymm15, 78
+        vblendps ymm10, ymm10, ymm5, 0xCC
+        vblendps ymm14, ymm14, ymm13, 0xCC
+        vperm2f128 ymm8, ymm10, ymm14, 0x20
+        vmovups ymmword ptr [rbx+0x40], ymm8
+        vblendps ymm15, ymm13, ymm15, 0xCC
+        vperm2f128 ymm13, ymm6, ymm15, 0x20
+        vmovups ymmword ptr [rbx+0x60], ymm13
+        vperm2f128 ymm9, ymm1, ymm2, 0x31
+        vperm2f128 ymm11, ymm3, ymm4, 0x31
+        vmovups ymmword ptr [rbx+0x80], ymm9
+        vperm2f128 ymm14, ymm10, ymm14, 0x31
+        vperm2f128 ymm15, ymm6, ymm15, 0x31
+        vmovups ymmword ptr [rbx+0xA0], ymm11
+        vmovups ymmword ptr [rbx+0xC0], ymm14
+        vmovups ymmword ptr [rbx+0xE0], ymm15
+        vmovdqa ymm0, ymmword ptr [rsp+0x220]
+        vpaddd  ymm1, ymm0, ymmword ptr [rsp+0x240]
+        vmovdqa ymmword ptr [rsp+0x240], ymm1
+        vpxor   ymm0, ymm0, ymmword ptr [CMP_MSB_MASK+rip]
+        vpxor   ymm2, ymm1, ymmword ptr [CMP_MSB_MASK+rip]
+        vpcmpgtd ymm2, ymm0, ymm2
+        vmovdqa ymm0, ymmword ptr [rsp+0x260]
+        vpsubd  ymm2, ymm0, ymm2
+        vmovdqa ymmword ptr [rsp+0x260], ymm2
+        add     rdi, 64
+        add     rbx, 256
+        mov     qword ptr [rbp+0x50], rbx
+        sub     rsi, 8
+        cmp     rsi, 8
+        jnc     2b
+        test    rsi, rsi
+        jnz     3f
+4:
+        vzeroupper
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align  5
+3:
+        mov     rbx, qword ptr [rbp+0x50]
+        mov     r15, qword ptr [rsp+0x2A0]
+        movzx   r13d, byte ptr [rbp+0x38]
+        movzx   r12d, byte ptr [rbp+0x48]
+        test    rsi, 0x4
+        je      3f
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+0x10]
+        vmovdqa ymm8, ymm0
+        vmovdqa ymm9, ymm1
+        vbroadcasti128 ymm12, xmmword ptr [rsp+0x240]
+        vbroadcasti128 ymm13, xmmword ptr [rsp+0x260]
+        vpunpckldq ymm14, ymm12, ymm13
+        vpunpckhdq ymm15, ymm12, ymm13
+        vpermq  ymm14, ymm14, 0x50
+        vpermq  ymm15, ymm15, 0x50
+        vbroadcasti128 ymm12, xmmword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpblendd ymm14, ymm14, ymm12, 0x44
+        vpblendd ymm15, ymm15, ymm12, 0x44
+        vmovdqa ymmword ptr [rsp], ymm14
+        vmovdqa ymmword ptr [rsp+0x20], ymm15
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+.p2align  5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x200], eax
+        vmovups ymm2, ymmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm2, ymm2, xmmword ptr [r9+rdx-0x40], 0x01
+        vmovups ymm3, ymmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm3, ymm3, xmmword ptr [r9+rdx-0x30], 0x01
+        vshufps ymm4, ymm2, ymm3, 136
+        vshufps ymm5, ymm2, ymm3, 221
+        vmovups ymm2, ymmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm2, ymm2, xmmword ptr [r9+rdx-0x20], 0x01
+        vmovups ymm3, ymmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm3, ymm3, xmmword ptr [r9+rdx-0x10], 0x01
+        vshufps ymm6, ymm2, ymm3, 136
+        vshufps ymm7, ymm2, ymm3, 221
+        vpshufd ymm6, ymm6, 0x93
+        vpshufd ymm7, ymm7, 0x93
+        vmovups ymm10, ymmword ptr [r10+rdx-0x40]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r11+rdx-0x40], 0x01
+        vmovups ymm11, ymmword ptr [r10+rdx-0x30]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r11+rdx-0x30], 0x01
+        vshufps ymm12, ymm10, ymm11, 136
+        vshufps ymm13, ymm10, ymm11, 221
+        vmovups ymm10, ymmword ptr [r10+rdx-0x20]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r11+rdx-0x20], 0x01
+        vmovups ymm11, ymmword ptr [r10+rdx-0x10]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r11+rdx-0x10], 0x01
+        vshufps ymm14, ymm10, ymm11, 136
+        vshufps ymm15, ymm10, ymm11, 221
+        vpshufd ymm14, ymm14, 0x93
+        vpshufd ymm15, ymm15, 0x93
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        vpbroadcastd ymm2, dword ptr [rsp+0x200]
+        vmovdqa ymm3, ymmword ptr [rsp]
+        vmovdqa ymm11, ymmword ptr [rsp+0x20]
+        vpblendd ymm3, ymm3, ymm2, 0x88
+        vpblendd ymm11, ymm11, ymm2, 0x88
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovdqa ymm10, ymm2
+        mov     al, 7
+9:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm8, ymm8, ymm12
+        vmovdqa ymmword ptr [rsp+0x40], ymm4
+        nop
+        vmovdqa ymmword ptr [rsp+0x60], ymm12
+        nop
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT16+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 12
+        vpslld  ymm9, ymm9, 20
+        vpor    ymm9, ymm9, ymm4
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vmovdqa ymmword ptr [rsp+0x80], ymm5
+        vmovdqa ymmword ptr [rsp+0xA0], ymm13
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT8+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 7
+        vpslld  ymm9, ymm9, 25
+        vpor    ymm9, ymm9, ymm4
+        vpshufd ymm0, ymm0, 0x93
+        vpshufd ymm8, ymm8, 0x93
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm11, ymm11, 0x4E
+        vpshufd ymm2, ymm2, 0x39
+        vpshufd ymm10, ymm10, 0x39
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm8, ymm8, ymm14
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT16+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 12
+        vpslld  ymm9, ymm9, 20
+        vpor    ymm9, ymm9, ymm4
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm8, ymm8, ymm15
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT8+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 7
+        vpslld  ymm9, ymm9, 25
+        vpor    ymm9, ymm9, ymm4
+        vpshufd ymm0, ymm0, 0x39
+        vpshufd ymm8, ymm8, 0x39
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm11, ymm11, 0x4E
+        vpshufd ymm2, ymm2, 0x93
+        vpshufd ymm10, ymm10, 0x93
+        dec     al
+        je      9f
+        vmovdqa ymm4, ymmword ptr [rsp+0x40]
+        vmovdqa ymm5, ymmword ptr [rsp+0x80]
+        vshufps ymm12, ymm4, ymm5, 214
+        vpshufd ymm13, ymm4, 0x0F
+        vpshufd ymm4, ymm12, 0x39
+        vshufps ymm12, ymm6, ymm7, 250
+        vpblendd ymm13, ymm13, ymm12, 0xAA
+        vpunpcklqdq ymm12, ymm7, ymm5
+        vpblendd ymm12, ymm12, ymm6, 0x88
+        vpshufd ymm12, ymm12, 0x78
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 0x1E
+        vmovdqa ymmword ptr [rsp+0x40], ymm13
+        vmovdqa ymmword ptr [rsp+0x80], ymm12
+        vmovdqa ymm12, ymmword ptr [rsp+0x60]
+        vmovdqa ymm13, ymmword ptr [rsp+0xA0]
+        vshufps ymm5, ymm12, ymm13, 214
+        vpshufd ymm6, ymm12, 0x0F
+        vpshufd ymm12, ymm5, 0x39
+        vshufps ymm5, ymm14, ymm15, 250
+        vpblendd ymm6, ymm6, ymm5, 0xAA
+        vpunpcklqdq ymm5, ymm15, ymm13
+        vpblendd ymm5, ymm5, ymm14, 0x88
+        vpshufd ymm5, ymm5, 0x78
+        vpunpckhdq ymm13, ymm13, ymm15
+        vpunpckldq ymm14, ymm14, ymm13
+        vpshufd ymm15, ymm14, 0x1E
+        vmovdqa ymm13, ymm6
+        vmovdqa ymm14, ymm5
+        vmovdqa ymm5, ymmword ptr [rsp+0x40]
+        vmovdqa ymm6, ymmword ptr [rsp+0x80]
+        jmp     9b
+9:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        vpxor   ymm8, ymm8, ymm10
+        vpxor   ymm9, ymm9, ymm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vmovdqu xmmword ptr [rbx+0x40], xmm8
+        vmovdqu xmmword ptr [rbx+0x50], xmm9
+        vextracti128 xmmword ptr [rbx+0x60], ymm8, 0x01
+        vextracti128 xmmword ptr [rbx+0x70], ymm9, 0x01
+        vmovaps xmm8, xmmword ptr [rsp+0x280]
+        vmovaps xmm0, xmmword ptr [rsp+0x240]
+        vmovaps xmm1, xmmword ptr [rsp+0x250]
+        vmovaps xmm2, xmmword ptr [rsp+0x260]
+        vmovaps xmm3, xmmword ptr [rsp+0x270]
+        vblendvps xmm0, xmm0, xmm1, xmm8
+        vblendvps xmm2, xmm2, xmm3, xmm8
+        vmovaps xmmword ptr [rsp+0x240], xmm0
+        vmovaps xmmword ptr [rsp+0x260], xmm2
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+3:
+        test    rsi, 0x2
+        je      3f
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm13, dword ptr [rsp+0x240]
+        vpinsrd xmm13, xmm13, dword ptr [rsp+0x260], 1
+        vpinsrd xmm13, xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovd   xmm14, dword ptr [rsp+0x244]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+0x264], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vinserti128 ymm13, ymm13, xmm14, 0x01
+        vbroadcasti128 ymm14, xmmword ptr [ROT16+rip]
+        vbroadcasti128 ymm15, xmmword ptr [ROT8+rip]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+.p2align  5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x200], eax
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
+        vpbroadcastd ymm8, dword ptr [rsp+0x200]
+        vpblendd ymm3, ymm13, ymm8, 0x88
+        vmovups ymm8, ymmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x40], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x30], 0x01
+        vshufps ymm4, ymm8, ymm9, 136
+        vshufps ymm5, ymm8, ymm9, 221
+        vmovups ymm8, ymmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x20], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x10], 0x01
+        vshufps ymm6, ymm8, ymm9, 136
+        vshufps ymm7, ymm8, ymm9, 221
+        vpshufd ymm6, ymm6, 0x93
+        vpshufd ymm7, ymm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm14
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm8
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm15
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm8
+        vpshufd ymm0, ymm0, 0x93
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x39
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm14
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm8
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm15
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm8
+        vpshufd ymm0, ymm0, 0x39
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x93
+        dec     al
+        jz      9f
+        vshufps ymm8, ymm4, ymm5, 214
+        vpshufd ymm9, ymm4, 0x0F
+        vpshufd ymm4, ymm8, 0x39
+        vshufps ymm8, ymm6, ymm7, 250
+        vpblendd ymm9, ymm9, ymm8, 0xAA
+        vpunpcklqdq ymm8, ymm7, ymm5
+        vpblendd ymm8, ymm8, ymm6, 0x88
+        vpshufd ymm8, ymm8, 0x78
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 0x1E
+        vmovdqa ymm5, ymm9
+        vmovdqa ymm6, ymm8
+        jmp     9b
+9:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vmovaps ymm8, ymmword ptr [rsp+0x280]
+        vmovaps ymm0, ymmword ptr [rsp+0x240]
+        vmovups ymm1, ymmword ptr [rsp+0x248]
+        vmovaps ymm2, ymmword ptr [rsp+0x260]
+        vmovups ymm3, ymmword ptr [rsp+0x268]
+        vblendvps ymm0, ymm0, ymm1, ymm8
+        vblendvps ymm2, ymm2, ymm3, ymm8
+        vmovaps ymmword ptr [rsp+0x240], ymm0
+        vmovaps ymmword ptr [rsp+0x260], ymm2
+        add     rbx, 64
+        add     rdi, 16
+        sub     rsi, 2
+3:
+        test    rsi, 0x1
+        je      4b
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm3, dword ptr [rsp+0x240]
+        vpinsrd xmm3, xmm3, dword ptr [rsp+0x260], 1
+        vpinsrd xmm13, xmm3, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovdqa xmm14, xmmword ptr [ROT16+rip]
+        vmovdqa xmm15, xmmword ptr [ROT8+rip]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+.p2align  5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        vmovdqa xmm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovdqa xmm3, xmm13
+        vpinsrd xmm3, xmm3, eax, 3
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x30]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x10]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm14
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 12
+        vpslld  xmm1, xmm1, 20
+        vpor    xmm1, xmm1, xmm8
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm15
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 7
+        vpslld  xmm1, xmm1, 25
+        vpor    xmm1, xmm1, xmm8
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm14
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 12
+        vpslld  xmm1, xmm1, 20
+        vpor    xmm1, xmm1, xmm8
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm15
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 7
+        vpslld  xmm1, xmm1, 25
+        vpor    xmm1, xmm1, xmm8
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+
+
+#ifdef __APPLE__
+.static_data
+#else
+.section .rodata
+#endif
+.p2align  6
+ADD0:
+        .long  0, 1, 2, 3, 4, 5, 6, 7
+ADD1:
+        .long  8, 8, 8, 8, 8, 8, 8, 8
+BLAKE3_IV_0:
+        .long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+        .long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+BLAKE3_IV_1:
+        .long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+        .long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+BLAKE3_IV_2:
+        .long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+        .long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+BLAKE3_IV_3:
+        .long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+        .long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+BLAKE3_BLOCK_LEN:
+        .long  0x00000040, 0x00000040, 0x00000040, 0x00000040
+        .long  0x00000040, 0x00000040, 0x00000040, 0x00000040
+ROT16:
+        .byte  2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13
+ROT8:
+        .byte  1, 2, 3, 0, 5, 6, 7, 4, 9, 10, 11, 8, 13, 14, 15, 12
+CMP_MSB_MASK:
+        .long  0x80000000, 0x80000000, 0x80000000, 0x80000000
+        .long  0x80000000, 0x80000000, 0x80000000, 0x80000000
+BLAKE3_IV:
+        .long  0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A
+
diff --git a/thirdparty/blake3/c/blake3_avx2_x86-64_windows_gnu.S b/thirdparty/blake3/c/blake3_avx2_x86-64_windows_gnu.S
new file mode 100644
index 000000000..3d4be4a7d
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx2_x86-64_windows_gnu.S
@@ -0,0 +1,1817 @@
+.intel_syntax noprefix
+.global _blake3_hash_many_avx2
+.global blake3_hash_many_avx2
+.section .text
+        .p2align  6
+_blake3_hash_many_avx2:
+blake3_hash_many_avx2:
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rsi
+        push    rdi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 880
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        vmovdqa xmmword ptr [rsp+0x2D0], xmm6
+        vmovdqa xmmword ptr [rsp+0x2E0], xmm7
+        vmovdqa xmmword ptr [rsp+0x2F0], xmm8
+        vmovdqa xmmword ptr [rsp+0x300], xmm9
+        vmovdqa xmmword ptr [rsp+0x310], xmm10
+        vmovdqa xmmword ptr [rsp+0x320], xmm11
+        vmovdqa xmmword ptr [rsp+0x330], xmm12
+        vmovdqa xmmword ptr [rsp+0x340], xmm13
+        vmovdqa xmmword ptr [rsp+0x350], xmm14
+        vmovdqa xmmword ptr [rsp+0x360], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+0x68]
+        movzx   r9, byte ptr [rbp+0x70]
+        neg     r9d
+        vmovd   xmm0, r9d
+        vpbroadcastd ymm0, xmm0
+        vmovdqa ymmword ptr [rsp+0x260], ymm0
+        vpand   ymm1, ymm0, ymmword ptr [ADD0+rip]
+        vpand   ymm2, ymm0, ymmword ptr [ADD1+rip]
+        vmovdqa ymmword ptr [rsp+0x2A0], ymm2
+        vmovd   xmm2, r8d
+        vpbroadcastd ymm2, xmm2
+        vpaddd  ymm2, ymm2, ymm1
+        vmovdqa ymmword ptr [rsp+0x220], ymm2
+        vpxor   ymm1, ymm1, ymmword ptr [CMP_MSB_MASK+rip]
+        vpxor   ymm2, ymm2, ymmword ptr [CMP_MSB_MASK+rip]
+        vpcmpgtd ymm2, ymm1, ymm2
+        shr     r8, 32
+        vmovd   xmm3, r8d
+        vpbroadcastd ymm3, xmm3
+        vpsubd  ymm3, ymm3, ymm2
+        vmovdqa ymmword ptr [rsp+0x240], ymm3
+        shl     rdx, 6
+        mov     qword ptr [rsp+0x2C0], rdx
+        cmp     rsi, 8
+        jc      3f
+2:
+        vpbroadcastd ymm0, dword ptr [rcx]
+        vpbroadcastd ymm1, dword ptr [rcx+0x4]
+        vpbroadcastd ymm2, dword ptr [rcx+0x8]
+        vpbroadcastd ymm3, dword ptr [rcx+0xC]
+        vpbroadcastd ymm4, dword ptr [rcx+0x10]
+        vpbroadcastd ymm5, dword ptr [rcx+0x14]
+        vpbroadcastd ymm6, dword ptr [rcx+0x18]
+        vpbroadcastd ymm7, dword ptr [rcx+0x1C]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x20]
+        mov     r13, qword ptr [rdi+0x28]
+        mov     r14, qword ptr [rdi+0x30]
+        mov     r15, qword ptr [rdi+0x38]
+        movzx   eax, byte ptr [rbp+0x78]
+        movzx   ebx, byte ptr [rbp+0x80]
+        or      eax, ebx
+        xor     edx, edx
+.p2align 5
+9:
+        movzx   ebx, byte ptr [rbp+0x88]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+0x2C0]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+0x200], eax
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x40], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x40]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x40], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x40]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x40], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x40]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x40], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0x20], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0x40], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0x60], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x30], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x30], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x30]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x30], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x30]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x30], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+0x80], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0xA0], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0xC0], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0xE0], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x20], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x20]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x20], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x20]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x20], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x20]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x20], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+0x100], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0x120], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0x140], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0x160], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x10], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x10], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x10]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x10], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x10]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x10], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+0x180], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0x1A0], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0x1C0], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0x1E0], ymm11
+        vpbroadcastd ymm15, dword ptr [rsp+0x200]
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r12+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r13+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r14+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        prefetcht0 [r15+rdx+0x80]
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x40]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x80]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm0, ymmword ptr [rsp+0x220]
+        vpxor   ymm13, ymm1, ymmword ptr [rsp+0x240]
+        vpxor   ymm14, ymm2, ymmword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpxor   ymm15, ymm3, ymm15
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [BLAKE3_IV_0+rip]
+        vpaddd  ymm9, ymm13, ymmword ptr [BLAKE3_IV_1+rip]
+        vpaddd  ymm10, ymm14, ymmword ptr [BLAKE3_IV_2+rip]
+        vpaddd  ymm11, ymm15, ymmword ptr [BLAKE3_IV_3+rip]
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x20]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x60]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x100]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x180]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x120]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x40]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x60]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x80]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x20]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x120]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x160]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x60]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x80]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x40]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x120]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x160]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x20]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x140]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x120]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x60]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x80]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x20]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x40]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x100]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x180]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x120]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x140]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x40]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x60]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x20]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x80]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x120]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x160]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x100]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x180]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x20]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x40]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x80]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x60]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x140]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x160]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0xA0]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x20]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x100]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1E0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x120]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xC0]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x1C0]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x40]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x60]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0xE0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+0x200], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0x140]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0x180]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0x80]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0x1A0]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8+rip]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+0x200]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vpxor   ymm0, ymm0, ymm8
+        vpxor   ymm1, ymm1, ymm9
+        vpxor   ymm2, ymm2, ymm10
+        vpxor   ymm3, ymm3, ymm11
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpxor   ymm4, ymm4, ymm12
+        vpxor   ymm5, ymm5, ymm13
+        vpxor   ymm6, ymm6, ymm14
+        vpxor   ymm7, ymm7, ymm15
+        movzx   eax, byte ptr [rbp+0x78]
+        jne     9b
+        mov     rbx, qword ptr [rbp+0x90]
+        vunpcklps ymm8, ymm0, ymm1
+        vunpcklps ymm9, ymm2, ymm3
+        vunpckhps ymm10, ymm0, ymm1
+        vunpcklps ymm11, ymm4, ymm5
+        vunpcklps ymm0, ymm6, ymm7
+        vshufps ymm12, ymm8, ymm9, 78
+        vblendps ymm1, ymm8, ymm12, 0xCC
+        vshufps ymm8, ymm11, ymm0, 78
+        vunpckhps ymm13, ymm2, ymm3
+        vblendps ymm2, ymm11, ymm8, 0xCC
+        vblendps ymm3, ymm12, ymm9, 0xCC
+        vperm2f128 ymm12, ymm1, ymm2, 0x20
+        vmovups ymmword ptr [rbx], ymm12
+        vunpckhps ymm14, ymm4, ymm5
+        vblendps ymm4, ymm8, ymm0, 0xCC
+        vunpckhps ymm15, ymm6, ymm7
+        vperm2f128 ymm7, ymm3, ymm4, 0x20
+        vmovups ymmword ptr [rbx+0x20], ymm7
+        vshufps ymm5, ymm10, ymm13, 78
+        vblendps ymm6, ymm5, ymm13, 0xCC
+        vshufps ymm13, ymm14, ymm15, 78
+        vblendps ymm10, ymm10, ymm5, 0xCC
+        vblendps ymm14, ymm14, ymm13, 0xCC
+        vperm2f128 ymm8, ymm10, ymm14, 0x20
+        vmovups ymmword ptr [rbx+0x40], ymm8
+        vblendps ymm15, ymm13, ymm15, 0xCC
+        vperm2f128 ymm13, ymm6, ymm15, 0x20
+        vmovups ymmword ptr [rbx+0x60], ymm13
+        vperm2f128 ymm9, ymm1, ymm2, 0x31
+        vperm2f128 ymm11, ymm3, ymm4, 0x31
+        vmovups ymmword ptr [rbx+0x80], ymm9
+        vperm2f128 ymm14, ymm10, ymm14, 0x31
+        vperm2f128 ymm15, ymm6, ymm15, 0x31
+        vmovups ymmword ptr [rbx+0xA0], ymm11
+        vmovups ymmword ptr [rbx+0xC0], ymm14
+        vmovups ymmword ptr [rbx+0xE0], ymm15
+        vmovdqa ymm0, ymmword ptr [rsp+0x2A0]
+        vpaddd  ymm1, ymm0, ymmword ptr [rsp+0x220]
+        vmovdqa ymmword ptr [rsp+0x220], ymm1
+        vpxor   ymm0, ymm0, ymmword ptr [CMP_MSB_MASK+rip]
+        vpxor   ymm2, ymm1, ymmword ptr [CMP_MSB_MASK+rip]
+        vpcmpgtd ymm2, ymm0, ymm2
+        vmovdqa ymm0, ymmword ptr [rsp+0x240]
+        vpsubd  ymm2, ymm0, ymm2
+        vmovdqa ymmword ptr [rsp+0x240], ymm2
+        add     rdi, 64
+        add     rbx, 256
+        mov     qword ptr [rbp+0x90], rbx
+        sub     rsi, 8
+        cmp     rsi, 8
+        jnc     2b
+        test    rsi, rsi
+        jnz     3f
+4:
+        vzeroupper
+        vmovdqa xmm6, xmmword ptr [rsp+0x2D0]
+        vmovdqa xmm7, xmmword ptr [rsp+0x2E0]
+        vmovdqa xmm8, xmmword ptr [rsp+0x2F0]
+        vmovdqa xmm9, xmmword ptr [rsp+0x300]
+        vmovdqa xmm10, xmmword ptr [rsp+0x310]
+        vmovdqa xmm11, xmmword ptr [rsp+0x320]
+        vmovdqa xmm12, xmmword ptr [rsp+0x330]
+        vmovdqa xmm13, xmmword ptr [rsp+0x340]
+        vmovdqa xmm14, xmmword ptr [rsp+0x350]
+        vmovdqa xmm15, xmmword ptr [rsp+0x360]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rdi
+        pop     rsi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align 5
+3:
+        mov     rbx, qword ptr [rbp+0x90]
+        mov     r15, qword ptr [rsp+0x2C0]
+        movzx   r13d, byte ptr [rbp+0x78]
+        movzx   r12d, byte ptr [rbp+0x88]
+        test    rsi, 0x4
+        je      3f
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+0x10]
+        vmovdqa ymm8, ymm0
+        vmovdqa ymm9, ymm1
+        vbroadcasti128 ymm12, xmmword ptr [rsp+0x220]
+        vbroadcasti128 ymm13, xmmword ptr [rsp+0x240]
+        vpunpckldq ymm14, ymm12, ymm13
+        vpunpckhdq ymm15, ymm12, ymm13
+        vpermq  ymm14, ymm14, 0x50
+        vpermq  ymm15, ymm15, 0x50
+        vbroadcasti128 ymm12, xmmword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpblendd ymm14, ymm14, ymm12, 0x44
+        vpblendd ymm15, ymm15, ymm12, 0x44
+        vmovdqa ymmword ptr [rsp], ymm14
+        vmovdqa ymmword ptr [rsp+0x20], ymm15
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x200], eax
+        vmovups ymm2, ymmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm2, ymm2, xmmword ptr [r9+rdx-0x40], 0x01
+        vmovups ymm3, ymmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm3, ymm3, xmmword ptr [r9+rdx-0x30], 0x01
+        vshufps ymm4, ymm2, ymm3, 136
+        vshufps ymm5, ymm2, ymm3, 221
+        vmovups ymm2, ymmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm2, ymm2, xmmword ptr [r9+rdx-0x20], 0x01
+        vmovups ymm3, ymmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm3, ymm3, xmmword ptr [r9+rdx-0x10], 0x01
+        vshufps ymm6, ymm2, ymm3, 136
+        vshufps ymm7, ymm2, ymm3, 221
+        vpshufd ymm6, ymm6, 0x93
+        vpshufd ymm7, ymm7, 0x93
+        vmovups ymm10, ymmword ptr [r10+rdx-0x40]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r11+rdx-0x40], 0x01
+        vmovups ymm11, ymmword ptr [r10+rdx-0x30]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r11+rdx-0x30], 0x01
+        vshufps ymm12, ymm10, ymm11, 136
+        vshufps ymm13, ymm10, ymm11, 221
+        vmovups ymm10, ymmword ptr [r10+rdx-0x20]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r11+rdx-0x20], 0x01
+        vmovups ymm11, ymmword ptr [r10+rdx-0x10]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r11+rdx-0x10], 0x01
+        vshufps ymm14, ymm10, ymm11, 136
+        vshufps ymm15, ymm10, ymm11, 221
+        vpshufd ymm14, ymm14, 0x93
+        vpshufd ymm15, ymm15, 0x93
+        vpbroadcastd ymm2, dword ptr [rsp+0x200]
+        vmovdqa ymm3, ymmword ptr [rsp]
+        vmovdqa ymm11, ymmword ptr [rsp+0x20]
+        vpblendd ymm3, ymm3, ymm2, 0x88
+        vpblendd ymm11, ymm11, ymm2, 0x88
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovdqa ymm10, ymm2
+        mov     al, 7
+9:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm8, ymm8, ymm12
+        vmovdqa ymmword ptr [rsp+0x40], ymm4
+        nop
+        vmovdqa ymmword ptr [rsp+0x60], ymm12
+        nop
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT16+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 12
+        vpslld  ymm9, ymm9, 20
+        vpor    ymm9, ymm9, ymm4
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vmovdqa ymmword ptr [rsp+0x80], ymm5
+        vmovdqa ymmword ptr [rsp+0xA0], ymm13
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT8+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 7
+        vpslld  ymm9, ymm9, 25
+        vpor    ymm9, ymm9, ymm4
+        vpshufd ymm0, ymm0, 0x93
+        vpshufd ymm8, ymm8, 0x93
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm11, ymm11, 0x4E
+        vpshufd ymm2, ymm2, 0x39
+        vpshufd ymm10, ymm10, 0x39
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm8, ymm8, ymm14
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT16+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 12
+        vpslld  ymm9, ymm9, 20
+        vpor    ymm9, ymm9, ymm4
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm8, ymm8, ymm15
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT8+rip]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 7
+        vpslld  ymm9, ymm9, 25
+        vpor    ymm9, ymm9, ymm4
+        vpshufd ymm0, ymm0, 0x39
+        vpshufd ymm8, ymm8, 0x39
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm11, ymm11, 0x4E
+        vpshufd ymm2, ymm2, 0x93
+        vpshufd ymm10, ymm10, 0x93
+        dec     al
+        je      9f
+        vmovdqa ymm4, ymmword ptr [rsp+0x40]
+        vmovdqa ymm5, ymmword ptr [rsp+0x80]
+        vshufps ymm12, ymm4, ymm5, 214
+        vpshufd ymm13, ymm4, 0x0F
+        vpshufd ymm4, ymm12, 0x39
+        vshufps ymm12, ymm6, ymm7, 250
+        vpblendd ymm13, ymm13, ymm12, 0xAA
+        vpunpcklqdq ymm12, ymm7, ymm5
+        vpblendd ymm12, ymm12, ymm6, 0x88
+        vpshufd ymm12, ymm12, 0x78
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 0x1E
+        vmovdqa ymmword ptr [rsp+0x40], ymm13
+        vmovdqa ymmword ptr [rsp+0x80], ymm12
+        vmovdqa ymm12, ymmword ptr [rsp+0x60]
+        vmovdqa ymm13, ymmword ptr [rsp+0xA0]
+        vshufps ymm5, ymm12, ymm13, 214
+        vpshufd ymm6, ymm12, 0x0F
+        vpshufd ymm12, ymm5, 0x39
+        vshufps ymm5, ymm14, ymm15, 250
+        vpblendd ymm6, ymm6, ymm5, 0xAA
+        vpunpcklqdq ymm5, ymm15, ymm13
+        vpblendd ymm5, ymm5, ymm14, 0x88
+        vpshufd ymm5, ymm5, 0x78
+        vpunpckhdq ymm13, ymm13, ymm15
+        vpunpckldq ymm14, ymm14, ymm13
+        vpshufd ymm15, ymm14, 0x1E
+        vmovdqa ymm13, ymm6
+        vmovdqa ymm14, ymm5
+        vmovdqa ymm5, ymmword ptr [rsp+0x40]
+        vmovdqa ymm6, ymmword ptr [rsp+0x80]
+        jmp     9b
+9:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        vpxor   ymm8, ymm8, ymm10
+        vpxor   ymm9, ymm9, ymm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vmovdqu xmmword ptr [rbx+0x40], xmm8
+        vmovdqu xmmword ptr [rbx+0x50], xmm9
+        vextracti128 xmmword ptr [rbx+0x60], ymm8, 0x01
+        vextracti128 xmmword ptr [rbx+0x70], ymm9, 0x01
+        vmovaps xmm8, xmmword ptr [rsp+0x260]
+        vmovaps xmm0, xmmword ptr [rsp+0x220]
+        vmovaps xmm1, xmmword ptr [rsp+0x230]
+        vmovaps xmm2, xmmword ptr [rsp+0x240]
+        vmovaps xmm3, xmmword ptr [rsp+0x250]
+        vblendvps xmm0, xmm0, xmm1, xmm8
+        vblendvps xmm2, xmm2, xmm3, xmm8
+        vmovaps xmmword ptr [rsp+0x220], xmm0
+        vmovaps xmmword ptr [rsp+0x240], xmm2
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+3:
+        test    rsi, 0x2
+        je      3f
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm13, dword ptr [rsp+0x220]
+        vpinsrd xmm13, xmm13, dword ptr [rsp+0x240], 1
+        vpinsrd xmm13, xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovd   xmm14, dword ptr [rsp+0x224]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+0x244], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vinserti128 ymm13, ymm13, xmm14, 0x01
+        vbroadcasti128 ymm14, xmmword ptr [ROT16+rip]
+        vbroadcasti128 ymm15, xmmword ptr [ROT8+rip]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x200], eax
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
+        vpbroadcastd ymm8, dword ptr [rsp+0x200]
+        vpblendd ymm3, ymm13, ymm8, 0x88
+        vmovups ymm8, ymmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x40], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x30], 0x01
+        vshufps ymm4, ymm8, ymm9, 136
+        vshufps ymm5, ymm8, ymm9, 221
+        vmovups ymm8, ymmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x20], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x10], 0x01
+        vshufps ymm6, ymm8, ymm9, 136
+        vshufps ymm7, ymm8, ymm9, 221
+        vpshufd ymm6, ymm6, 0x93
+        vpshufd ymm7, ymm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm14
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm8
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm15
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm8
+        vpshufd ymm0, ymm0, 0x93
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x39
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm14
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm8
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm15
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm8
+        vpshufd ymm0, ymm0, 0x39
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x93
+        dec     al
+        jz      9f
+        vshufps ymm8, ymm4, ymm5, 214
+        vpshufd ymm9, ymm4, 0x0F
+        vpshufd ymm4, ymm8, 0x39
+        vshufps ymm8, ymm6, ymm7, 250
+        vpblendd ymm9, ymm9, ymm8, 0xAA
+        vpunpcklqdq ymm8, ymm7, ymm5
+        vpblendd ymm8, ymm8, ymm6, 0x88
+        vpshufd ymm8, ymm8, 0x78
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 0x1E
+        vmovdqa ymm5, ymm9
+        vmovdqa ymm6, ymm8
+        jmp     9b
+9:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vmovaps ymm8, ymmword ptr [rsp+0x260]
+        vmovaps ymm0, ymmword ptr [rsp+0x220]
+        vmovups ymm1, ymmword ptr [rsp+0x228]
+        vmovaps ymm2, ymmword ptr [rsp+0x240]
+        vmovups ymm3, ymmword ptr [rsp+0x248]
+        vblendvps ymm0, ymm0, ymm1, ymm8
+        vblendvps ymm2, ymm2, ymm3, ymm8
+        vmovaps ymmword ptr [rsp+0x220], ymm0
+        vmovaps ymmword ptr [rsp+0x240], ymm2
+        add     rbx, 64
+        add     rdi, 16
+        sub     rsi, 2
+3:
+        test    rsi, 0x1
+        je      4b
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm3, dword ptr [rsp+0x220]
+        vpinsrd xmm3, xmm3, dword ptr [rsp+0x240], 1
+        vpinsrd xmm13, xmm3, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovdqa xmm14, xmmword ptr [ROT16+rip]
+        vmovdqa xmm15, xmmword ptr [ROT8+rip]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        vmovdqa xmm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovdqa xmm3, xmm13
+        vpinsrd xmm3, xmm3, eax, 3
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x30]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x10]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm14
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 12
+        vpslld  xmm1, xmm1, 20
+        vpor    xmm1, xmm1, xmm8
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm15
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 7
+        vpslld  xmm1, xmm1, 25
+        vpor    xmm1, xmm1, xmm8
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm14
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 12
+        vpslld  xmm1, xmm1, 20
+        vpor    xmm1, xmm1, xmm8
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm15
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 7
+        vpslld  xmm1, xmm1, 25
+        vpor    xmm1, xmm1, xmm8
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+
+.section .rdata
+.p2align  6
+ADD0:
+        .long  0, 1, 2, 3, 4, 5, 6, 7
+ADD1:
+        .long  8, 8, 8, 8, 8, 8, 8, 8
+BLAKE3_IV_0:
+        .long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+        .long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+BLAKE3_IV_1:
+        .long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+        .long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+BLAKE3_IV_2:
+        .long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+        .long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+BLAKE3_IV_3:
+        .long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+        .long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+BLAKE3_BLOCK_LEN:
+        .long  0x00000040, 0x00000040, 0x00000040, 0x00000040
+        .long  0x00000040, 0x00000040, 0x00000040, 0x00000040
+ROT16:
+        .byte  2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13
+ROT8:
+        .byte  1, 2, 3, 0, 5, 6, 7, 4, 9, 10, 11, 8, 13, 14, 15, 12
+CMP_MSB_MASK:
+        .long  0x80000000, 0x80000000, 0x80000000, 0x80000000
+        .long  0x80000000, 0x80000000, 0x80000000, 0x80000000
+BLAKE3_IV:
+        .long  0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A
+
diff --git a/thirdparty/blake3/c/blake3_avx2_x86-64_windows_msvc.asm b/thirdparty/blake3/c/blake3_avx2_x86-64_windows_msvc.asm
new file mode 100644
index 000000000..352298edd
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx2_x86-64_windows_msvc.asm
@@ -0,0 +1,1828 @@
+public _blake3_hash_many_avx2
+public blake3_hash_many_avx2
+
+_TEXT   SEGMENT ALIGN(16) 'CODE'
+
+ALIGN   16
+blake3_hash_many_avx2 PROC
+_blake3_hash_many_avx2 PROC
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rsi
+        push    rdi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 880
+        and     rsp, 0FFFFFFFFFFFFFFC0H
+        vmovdqa xmmword ptr [rsp+2D0H], xmm6
+        vmovdqa xmmword ptr [rsp+2E0H], xmm7
+        vmovdqa xmmword ptr [rsp+2F0H], xmm8
+        vmovdqa xmmword ptr [rsp+300H], xmm9
+        vmovdqa xmmword ptr [rsp+310H], xmm10
+        vmovdqa xmmword ptr [rsp+320H], xmm11
+        vmovdqa xmmword ptr [rsp+330H], xmm12
+        vmovdqa xmmword ptr [rsp+340H], xmm13
+        vmovdqa xmmword ptr [rsp+350H], xmm14
+        vmovdqa xmmword ptr [rsp+360H], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+68H]
+        movzx   r9, byte ptr [rbp+70H]
+        neg     r9d
+        vmovd   xmm0, r9d
+        vpbroadcastd ymm0, xmm0
+        vmovdqa ymmword ptr [rsp+260H], ymm0
+        vpand   ymm1, ymm0, ymmword ptr [ADD0]
+        vpand   ymm2, ymm0, ymmword ptr [ADD1]
+        vmovdqa ymmword ptr [rsp+2A0H], ymm2
+        vmovd   xmm2, r8d
+        vpbroadcastd ymm2, xmm2
+        vpaddd  ymm2, ymm2, ymm1
+        vmovdqa ymmword ptr [rsp+220H], ymm2
+        vpxor   ymm1, ymm1, ymmword ptr [CMP_MSB_MASK]
+        vpxor   ymm2, ymm2, ymmword ptr [CMP_MSB_MASK]
+        vpcmpgtd ymm2, ymm1, ymm2
+        shr     r8, 32
+        vmovd   xmm3, r8d
+        vpbroadcastd ymm3, xmm3
+        vpsubd  ymm3, ymm3, ymm2
+        vmovdqa ymmword ptr [rsp+240H], ymm3
+        shl     rdx, 6
+        mov     qword ptr [rsp+2C0H], rdx
+        cmp     rsi, 8
+        jc      final7blocks
+outerloop8:
+        vpbroadcastd ymm0, dword ptr [rcx]
+        vpbroadcastd ymm1, dword ptr [rcx+4H]
+        vpbroadcastd ymm2, dword ptr [rcx+8H]
+        vpbroadcastd ymm3, dword ptr [rcx+0CH]
+        vpbroadcastd ymm4, dword ptr [rcx+10H]
+        vpbroadcastd ymm5, dword ptr [rcx+14H]
+        vpbroadcastd ymm6, dword ptr [rcx+18H]
+        vpbroadcastd ymm7, dword ptr [rcx+1CH]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        mov     r12, qword ptr [rdi+20H]
+        mov     r13, qword ptr [rdi+28H]
+        mov     r14, qword ptr [rdi+30H]
+        mov     r15, qword ptr [rdi+38H]
+        movzx   eax, byte ptr [rbp+78H]
+        movzx   ebx, byte ptr [rbp+80H]
+        or      eax, ebx
+        xor     edx, edx
+ALIGN   16
+innerloop8:
+        movzx   ebx, byte ptr [rbp+88H]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+2C0H]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+200H], eax
+        vmovups xmm8, xmmword ptr [r8+rdx-40H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-40H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-40H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-40H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-40H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-40H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-40H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-40H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+20H], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+40H], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+60H], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-30H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-30H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-30H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-30H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-30H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-30H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-30H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-30H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+80H], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+0A0H], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+0C0H], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+0E0H], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-20H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-20H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-20H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-20H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-20H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-20H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-20H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-20H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+100H], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+120H], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+140H], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+160H], ymm11
+        vmovups xmm8, xmmword ptr [r8+rdx-10H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-10H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-10H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-10H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-10H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-10H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-10H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-10H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm8, ymm12, ymm14, 136
+        vmovaps ymmword ptr [rsp+180H], ymm8
+        vshufps ymm9, ymm12, ymm14, 221
+        vmovaps ymmword ptr [rsp+1A0H], ymm9
+        vshufps ymm10, ymm13, ymm15, 136
+        vmovaps ymmword ptr [rsp+1C0H], ymm10
+        vshufps ymm11, ymm13, ymm15, 221
+        vmovaps ymmword ptr [rsp+1E0H], ymm11
+        vpbroadcastd ymm15, dword ptr [rsp+200H]
+        prefetcht0 byte ptr [r8+rdx+80H]
+        prefetcht0 byte ptr [r12+rdx+80H]
+        prefetcht0 byte ptr [r9+rdx+80H]
+        prefetcht0 byte ptr [r13+rdx+80H]
+        prefetcht0 byte ptr [r10+rdx+80H]
+        prefetcht0 byte ptr [r14+rdx+80H]
+        prefetcht0 byte ptr [r11+rdx+80H]
+        prefetcht0 byte ptr [r15+rdx+80H]
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+40H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+80H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0C0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm0, ymmword ptr [rsp+220H]
+        vpxor   ymm13, ymm1, ymmword ptr [rsp+240H]
+        vpxor   ymm14, ymm2, ymmword ptr [BLAKE3_BLOCK_LEN]
+        vpxor   ymm15, ymm3, ymm15
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [BLAKE3_IV_0]
+        vpaddd  ymm9, ymm13, ymmword ptr [BLAKE3_IV_1]
+        vpaddd  ymm10, ymm14, ymmword ptr [BLAKE3_IV_2]
+        vpaddd  ymm11, ymm15, ymmword ptr [BLAKE3_IV_3]
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+20H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+60H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0A0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0E0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+100H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+140H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+180H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1C0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+120H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+160H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+1A0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1E0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+40H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+60H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0E0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+80H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0C0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+140H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1A0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+20H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+180H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+120H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1E0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+160H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0A0H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+1C0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+100H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+60H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+140H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+1A0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0E0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+80H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+180H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+40H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1C0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0C0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+120H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+160H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+100H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0A0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+1E0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+20H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+140H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+180H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+1C0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1A0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0E0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+120H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+60H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1E0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+80H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+160H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0A0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+20H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+40H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+100H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0C0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+180H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+120H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+1E0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1C0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+1A0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+160H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+140H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+100H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+0E0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0A0H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0C0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+40H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+60H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+20H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+80H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+120H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+160H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+100H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1E0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+1C0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0A0H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+180H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+20H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+1A0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+40H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+80H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+60H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+140H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+0C0H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0E0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+160H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+0A0H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+20H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+100H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+1E0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+120H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0C0H]
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxor   ymm12, ymm12, ymm0
+        vpxor   ymm13, ymm13, ymm1
+        vpxor   ymm14, ymm14, ymm2
+        vpxor   ymm15, ymm15, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpshufb ymm15, ymm15, ymm8
+        vpaddd  ymm8, ymm12, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxor   ymm4, ymm4, ymm8
+        vpxor   ymm5, ymm5, ymm9
+        vpxor   ymm6, ymm6, ymm10
+        vpxor   ymm7, ymm7, ymm11
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+1C0H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+40H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+60H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+0E0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT16]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vmovdqa ymmword ptr [rsp+200H], ymm8
+        vpsrld  ymm8, ymm5, 12
+        vpslld  ymm5, ymm5, 20
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 12
+        vpslld  ymm6, ymm6, 20
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 12
+        vpslld  ymm7, ymm7, 20
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 12
+        vpslld  ymm4, ymm4, 20
+        vpor    ymm4, ymm4, ymm8
+        vpaddd  ymm0, ymm0, ymmword ptr [rsp+140H]
+        vpaddd  ymm1, ymm1, ymmword ptr [rsp+180H]
+        vpaddd  ymm2, ymm2, ymmword ptr [rsp+80H]
+        vpaddd  ymm3, ymm3, ymmword ptr [rsp+1A0H]
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxor   ymm15, ymm15, ymm0
+        vpxor   ymm12, ymm12, ymm1
+        vpxor   ymm13, ymm13, ymm2
+        vpxor   ymm14, ymm14, ymm3
+        vbroadcasti128 ymm8, xmmword ptr [ROT8]
+        vpshufb ymm15, ymm15, ymm8
+        vpshufb ymm12, ymm12, ymm8
+        vpshufb ymm13, ymm13, ymm8
+        vpshufb ymm14, ymm14, ymm8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm13, ymmword ptr [rsp+200H]
+        vpaddd  ymm9, ymm9, ymm14
+        vpxor   ymm5, ymm5, ymm10
+        vpxor   ymm6, ymm6, ymm11
+        vpxor   ymm7, ymm7, ymm8
+        vpxor   ymm4, ymm4, ymm9
+        vpxor   ymm0, ymm0, ymm8
+        vpxor   ymm1, ymm1, ymm9
+        vpxor   ymm2, ymm2, ymm10
+        vpxor   ymm3, ymm3, ymm11
+        vpsrld  ymm8, ymm5, 7
+        vpslld  ymm5, ymm5, 25
+        vpor    ymm5, ymm5, ymm8
+        vpsrld  ymm8, ymm6, 7
+        vpslld  ymm6, ymm6, 25
+        vpor    ymm6, ymm6, ymm8
+        vpsrld  ymm8, ymm7, 7
+        vpslld  ymm7, ymm7, 25
+        vpor    ymm7, ymm7, ymm8
+        vpsrld  ymm8, ymm4, 7
+        vpslld  ymm4, ymm4, 25
+        vpor    ymm4, ymm4, ymm8
+        vpxor   ymm4, ymm4, ymm12
+        vpxor   ymm5, ymm5, ymm13
+        vpxor   ymm6, ymm6, ymm14
+        vpxor   ymm7, ymm7, ymm15
+        movzx   eax, byte ptr [rbp+78H]
+        jne     innerloop8
+        mov     rbx, qword ptr [rbp+90H]
+        vunpcklps ymm8, ymm0, ymm1
+        vunpcklps ymm9, ymm2, ymm3
+        vunpckhps ymm10, ymm0, ymm1
+        vunpcklps ymm11, ymm4, ymm5
+        vunpcklps ymm0, ymm6, ymm7
+        vshufps ymm12, ymm8, ymm9, 78
+        vblendps ymm1, ymm8, ymm12, 0CCH
+        vshufps ymm8, ymm11, ymm0, 78
+        vunpckhps ymm13, ymm2, ymm3
+        vblendps ymm2, ymm11, ymm8, 0CCH
+        vblendps ymm3, ymm12, ymm9, 0CCH
+        vperm2f128 ymm12, ymm1, ymm2, 20H
+        vmovups ymmword ptr [rbx], ymm12
+        vunpckhps ymm14, ymm4, ymm5
+        vblendps ymm4, ymm8, ymm0, 0CCH
+        vunpckhps ymm15, ymm6, ymm7
+        vperm2f128 ymm7, ymm3, ymm4, 20H
+        vmovups ymmword ptr [rbx+20H], ymm7
+        vshufps ymm5, ymm10, ymm13, 78
+        vblendps ymm6, ymm5, ymm13, 0CCH
+        vshufps ymm13, ymm14, ymm15, 78
+        vblendps ymm10, ymm10, ymm5, 0CCH
+        vblendps ymm14, ymm14, ymm13, 0CCH
+        vperm2f128 ymm8, ymm10, ymm14, 20H
+        vmovups ymmword ptr [rbx+40H], ymm8
+        vblendps ymm15, ymm13, ymm15, 0CCH
+        vperm2f128 ymm13, ymm6, ymm15, 20H
+        vmovups ymmword ptr [rbx+60H], ymm13
+        vperm2f128 ymm9, ymm1, ymm2, 31H
+        vperm2f128 ymm11, ymm3, ymm4, 31H
+        vmovups ymmword ptr [rbx+80H], ymm9
+        vperm2f128 ymm14, ymm10, ymm14, 31H
+        vperm2f128 ymm15, ymm6, ymm15, 31H
+        vmovups ymmword ptr [rbx+0A0H], ymm11
+        vmovups ymmword ptr [rbx+0C0H], ymm14
+        vmovups ymmword ptr [rbx+0E0H], ymm15
+        vmovdqa ymm0, ymmword ptr [rsp+2A0H]
+        vpaddd  ymm1, ymm0, ymmword ptr [rsp+220H]
+        vmovdqa ymmword ptr [rsp+220H], ymm1
+        vpxor   ymm0, ymm0, ymmword ptr [CMP_MSB_MASK]
+        vpxor   ymm2, ymm1, ymmword ptr [CMP_MSB_MASK]
+        vpcmpgtd ymm2, ymm0, ymm2
+        vmovdqa ymm0, ymmword ptr [rsp+240H]
+        vpsubd  ymm2, ymm0, ymm2
+        vmovdqa ymmword ptr [rsp+240H], ymm2
+        add     rdi, 64
+        add     rbx, 256
+        mov     qword ptr [rbp+90H], rbx
+        sub     rsi, 8
+        cmp     rsi, 8
+        jnc     outerloop8
+        test    rsi, rsi
+        jnz     final7blocks
+unwind:
+        vzeroupper
+        vmovdqa xmm6, xmmword ptr [rsp+2D0H]
+        vmovdqa xmm7, xmmword ptr [rsp+2E0H]
+        vmovdqa xmm8, xmmword ptr [rsp+2F0H]
+        vmovdqa xmm9, xmmword ptr [rsp+300H]
+        vmovdqa xmm10, xmmword ptr [rsp+310H]
+        vmovdqa xmm11, xmmword ptr [rsp+320H]
+        vmovdqa xmm12, xmmword ptr [rsp+330H]
+        vmovdqa xmm13, xmmword ptr [rsp+340H]
+        vmovdqa xmm14, xmmword ptr [rsp+350H]
+        vmovdqa xmm15, xmmword ptr [rsp+360H]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rdi
+        pop     rsi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+ALIGN   16
+final7blocks:
+        mov     rbx, qword ptr [rbp+90H]
+        mov     r15, qword ptr [rsp+2C0H]
+        movzx   r13d, byte ptr [rbp+78H]
+        movzx   r12d, byte ptr [rbp+88H]
+        test    rsi, 4H
+        je      final3blocks
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+10H]
+        vmovdqa ymm8, ymm0
+        vmovdqa ymm9, ymm1
+        vbroadcasti128 ymm12, xmmword ptr [rsp+220H]
+        vbroadcasti128 ymm13, xmmword ptr [rsp+240H]
+        vpunpckldq ymm14, ymm12, ymm13
+        vpunpckhdq ymm15, ymm12, ymm13
+        vpermq  ymm14, ymm14, 50H
+        vpermq  ymm15, ymm15, 50H
+        vbroadcasti128 ymm12, xmmword ptr [BLAKE3_BLOCK_LEN]
+        vpblendd ymm14, ymm14, ymm12, 44H
+        vpblendd ymm15, ymm15, ymm12, 44H
+        vmovdqa ymmword ptr [rsp], ymm14
+        vmovdqa ymmword ptr [rsp+20H], ymm15
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+ALIGN   16
+innerloop4:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+200H], eax
+        vmovups ymm2, ymmword ptr [r8+rdx-40H]
+        vinsertf128 ymm2, ymm2, xmmword ptr [r9+rdx-40H], 01H
+        vmovups ymm3, ymmword ptr [r8+rdx-30H]
+        vinsertf128 ymm3, ymm3, xmmword ptr [r9+rdx-30H], 01H
+        vshufps ymm4, ymm2, ymm3, 136
+        vshufps ymm5, ymm2, ymm3, 221
+        vmovups ymm2, ymmword ptr [r8+rdx-20H]
+        vinsertf128 ymm2, ymm2, xmmword ptr [r9+rdx-20H], 01H
+        vmovups ymm3, ymmword ptr [r8+rdx-10H]
+        vinsertf128 ymm3, ymm3, xmmword ptr [r9+rdx-10H], 01H
+        vshufps ymm6, ymm2, ymm3, 136
+        vshufps ymm7, ymm2, ymm3, 221
+        vpshufd ymm6, ymm6, 93H
+        vpshufd ymm7, ymm7, 93H
+        vmovups ymm10, ymmword ptr [r10+rdx-40H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r11+rdx-40H], 01H
+        vmovups ymm11, ymmword ptr [r10+rdx-30H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r11+rdx-30H], 01H
+        vshufps ymm12, ymm10, ymm11, 136
+        vshufps ymm13, ymm10, ymm11, 221
+        vmovups ymm10, ymmword ptr [r10+rdx-20H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r11+rdx-20H], 01H
+        vmovups ymm11, ymmword ptr [r10+rdx-10H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r11+rdx-10H], 01H
+        vshufps ymm14, ymm10, ymm11, 136
+        vshufps ymm15, ymm10, ymm11, 221
+        vpshufd ymm14, ymm14, 93H
+        vpshufd ymm15, ymm15, 93H
+        vpbroadcastd ymm2, dword ptr [rsp+200H]
+        vmovdqa ymm3, ymmword ptr [rsp]
+        vmovdqa ymm11, ymmword ptr [rsp+20H]
+        vpblendd ymm3, ymm3, ymm2, 88H
+        vpblendd ymm11, ymm11, ymm2, 88H
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV]
+        vmovdqa ymm10, ymm2
+        mov     al, 7
+roundloop4:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm8, ymm8, ymm12
+        vmovdqa ymmword ptr [rsp+40H], ymm4
+        nop
+        vmovdqa ymmword ptr [rsp+60H], ymm12
+        nop
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT16]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 12
+        vpslld  ymm9, ymm9, 20
+        vpor    ymm9, ymm9, ymm4
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vmovdqa ymmword ptr [rsp+80H], ymm5
+        vmovdqa ymmword ptr [rsp+0A0H], ymm13
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT8]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 7
+        vpslld  ymm9, ymm9, 25
+        vpor    ymm9, ymm9, ymm4
+        vpshufd ymm0, ymm0, 93H
+        vpshufd ymm8, ymm8, 93H
+        vpshufd ymm3, ymm3, 4EH
+        vpshufd ymm11, ymm11, 4EH
+        vpshufd ymm2, ymm2, 39H
+        vpshufd ymm10, ymm10, 39H
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm8, ymm8, ymm14
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT16]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 12
+        vpslld  ymm9, ymm9, 20
+        vpor    ymm9, ymm9, ymm4
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm8, ymm8, ymm15
+        vpaddd  ymm0, ymm0, ymm1
+        vpaddd  ymm8, ymm8, ymm9
+        vpxor   ymm3, ymm3, ymm0
+        vpxor   ymm11, ymm11, ymm8
+        vbroadcasti128 ymm4, xmmword ptr [ROT8]
+        vpshufb ymm3, ymm3, ymm4
+        vpshufb ymm11, ymm11, ymm4
+        vpaddd  ymm2, ymm2, ymm3
+        vpaddd  ymm10, ymm10, ymm11
+        vpxor   ymm1, ymm1, ymm2
+        vpxor   ymm9, ymm9, ymm10
+        vpsrld  ymm4, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm4
+        vpsrld  ymm4, ymm9, 7
+        vpslld  ymm9, ymm9, 25
+        vpor    ymm9, ymm9, ymm4
+        vpshufd ymm0, ymm0, 39H
+        vpshufd ymm8, ymm8, 39H
+        vpshufd ymm3, ymm3, 4EH
+        vpshufd ymm11, ymm11, 4EH
+        vpshufd ymm2, ymm2, 93H
+        vpshufd ymm10, ymm10, 93H
+        dec     al
+        je      endroundloop4
+        vmovdqa ymm4, ymmword ptr [rsp+40H]
+        vmovdqa ymm5, ymmword ptr [rsp+80H]
+        vshufps ymm12, ymm4, ymm5, 214
+        vpshufd ymm13, ymm4, 0FH
+        vpshufd ymm4, ymm12, 39H
+        vshufps ymm12, ymm6, ymm7, 250
+        vpblendd ymm13, ymm13, ymm12, 0AAH
+        vpunpcklqdq ymm12, ymm7, ymm5
+        vpblendd ymm12, ymm12, ymm6, 88H
+        vpshufd ymm12, ymm12, 78H
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 1EH
+        vmovdqa ymmword ptr [rsp+40H], ymm13
+        vmovdqa ymmword ptr [rsp+80H], ymm12
+        vmovdqa ymm12, ymmword ptr [rsp+60H]
+        vmovdqa ymm13, ymmword ptr [rsp+0A0H]
+        vshufps ymm5, ymm12, ymm13, 214
+        vpshufd ymm6, ymm12, 0FH
+        vpshufd ymm12, ymm5, 39H
+        vshufps ymm5, ymm14, ymm15, 250
+        vpblendd ymm6, ymm6, ymm5, 0AAH
+        vpunpcklqdq ymm5, ymm15, ymm13
+        vpblendd ymm5, ymm5, ymm14, 88H
+        vpshufd ymm5, ymm5, 78H
+        vpunpckhdq ymm13, ymm13, ymm15
+        vpunpckldq ymm14, ymm14, ymm13
+        vpshufd ymm15, ymm14, 1EH
+        vmovdqa ymm13, ymm6
+        vmovdqa ymm14, ymm5
+        vmovdqa ymm5, ymmword ptr [rsp+40H]
+        vmovdqa ymm6, ymmword ptr [rsp+80H]
+        jmp     roundloop4
+endroundloop4:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        vpxor   ymm8, ymm8, ymm10
+        vpxor   ymm9, ymm9, ymm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop4
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+10H], xmm1
+        vextracti128 xmmword ptr [rbx+20H], ymm0, 01H
+        vextracti128 xmmword ptr [rbx+30H], ymm1, 01H
+        vmovdqu xmmword ptr [rbx+40H], xmm8
+        vmovdqu xmmword ptr [rbx+50H], xmm9
+        vextracti128 xmmword ptr [rbx+60H], ymm8, 01H
+        vextracti128 xmmword ptr [rbx+70H], ymm9, 01H
+        vmovaps xmm8, xmmword ptr [rsp+260H]
+        vmovaps xmm0, xmmword ptr [rsp+220H]
+        vmovaps xmm1, xmmword ptr [rsp+230H]
+        vmovaps xmm2, xmmword ptr [rsp+240H]
+        vmovaps xmm3, xmmword ptr [rsp+250H]
+        vblendvps xmm0, xmm0, xmm1, xmm8
+        vblendvps xmm2, xmm2, xmm3, xmm8
+        vmovaps xmmword ptr [rsp+220H], xmm0
+        vmovaps xmmword ptr [rsp+240H], xmm2
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+final3blocks:
+        test    rsi, 2H
+        je      final1blocks
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+10H]
+        vmovd   xmm13, dword ptr [rsp+220H]
+        vpinsrd xmm13, xmm13, dword ptr [rsp+240H], 1
+        vpinsrd xmm13, xmm13, dword ptr [BLAKE3_BLOCK_LEN], 2
+        vmovd   xmm14, dword ptr [rsp+224H]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+244H], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN], 2
+        vinserti128 ymm13, ymm13, xmm14, 01H
+        vbroadcasti128 ymm14, xmmword ptr [ROT16]
+        vbroadcasti128 ymm15, xmmword ptr [ROT8]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+ALIGN   16
+innerloop2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+200H], eax
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV]
+        vpbroadcastd ymm8, dword ptr [rsp+200H]
+        vpblendd ymm3, ymm13, ymm8, 88H
+        vmovups ymm8, ymmword ptr [r8+rdx-40H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-40H], 01H
+        vmovups ymm9, ymmword ptr [r8+rdx-30H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-30H], 01H
+        vshufps ymm4, ymm8, ymm9, 136
+        vshufps ymm5, ymm8, ymm9, 221
+        vmovups ymm8, ymmword ptr [r8+rdx-20H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-20H], 01H
+        vmovups ymm9, ymmword ptr [r8+rdx-10H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-10H], 01H
+        vshufps ymm6, ymm8, ymm9, 136
+        vshufps ymm7, ymm8, ymm9, 221
+        vpshufd ymm6, ymm6, 93H
+        vpshufd ymm7, ymm7, 93H
+        mov     al, 7
+roundloop2:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm14
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm8
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm15
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm8
+        vpshufd ymm0, ymm0, 93H
+        vpshufd ymm3, ymm3, 4EH
+        vpshufd ymm2, ymm2, 39H
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm14
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 12
+        vpslld  ymm1, ymm1, 20
+        vpor    ymm1, ymm1, ymm8
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm0, ymm0, ymm1
+        vpxor   ymm3, ymm3, ymm0
+        vpshufb ymm3, ymm3, ymm15
+        vpaddd  ymm2, ymm2, ymm3
+        vpxor   ymm1, ymm1, ymm2
+        vpsrld  ymm8, ymm1, 7
+        vpslld  ymm1, ymm1, 25
+        vpor    ymm1, ymm1, ymm8
+        vpshufd ymm0, ymm0, 39H
+        vpshufd ymm3, ymm3, 4EH
+        vpshufd ymm2, ymm2, 93H
+        dec     al
+        jz      endroundloop2
+        vshufps ymm8, ymm4, ymm5, 214
+        vpshufd ymm9, ymm4, 0FH
+        vpshufd ymm4, ymm8, 39H
+        vshufps ymm8, ymm6, ymm7, 250
+        vpblendd ymm9, ymm9, ymm8, 0AAH
+        vpunpcklqdq ymm8, ymm7, ymm5
+        vpblendd ymm8, ymm8, ymm6, 88H
+        vpshufd ymm8, ymm8, 78H
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 1EH
+        vmovdqa ymm5, ymm9
+        vmovdqa ymm6, ymm8
+        jmp     roundloop2
+endroundloop2:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop2
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+10H], xmm1
+        vextracti128 xmmword ptr [rbx+20H], ymm0, 01H
+        vextracti128 xmmword ptr [rbx+30H], ymm1, 01H
+        vmovaps ymm8, ymmword ptr [rsp+260H]
+        vmovaps ymm0, ymmword ptr [rsp+220H]
+        vmovups ymm1, ymmword ptr [rsp+228H]
+        vmovaps ymm2, ymmword ptr [rsp+240H]
+        vmovups ymm3, ymmword ptr [rsp+248H]
+        vblendvps ymm0, ymm0, ymm1, ymm8
+        vblendvps ymm2, ymm2, ymm3, ymm8
+        vmovaps ymmword ptr [rsp+220H], ymm0
+        vmovaps ymmword ptr [rsp+240H], ymm2
+        add     rbx, 64
+        add     rdi, 16
+        sub     rsi, 2
+final1blocks:
+        test    rsi, 1H
+        je      unwind
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+10H]
+        vmovd   xmm3, dword ptr [rsp+220H]
+        vpinsrd xmm3, xmm3, dword ptr [rsp+240H], 1
+        vpinsrd xmm13, xmm3, dword ptr [BLAKE3_BLOCK_LEN], 2
+        vmovdqa xmm14, xmmword ptr [ROT16]
+        vmovdqa xmm15, xmmword ptr [ROT8]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+ALIGN   16
+innerloop1:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        vmovdqa xmm2, xmmword ptr [BLAKE3_IV]
+        vmovdqa xmm3, xmm13
+        vpinsrd xmm3, xmm3, eax, 3
+        vmovups xmm8, xmmword ptr [r8+rdx-40H]
+        vmovups xmm9, xmmword ptr [r8+rdx-30H]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-20H]
+        vmovups xmm9, xmmword ptr [r8+rdx-10H]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 93H
+        vpshufd xmm7, xmm7, 93H
+        mov     al, 7
+roundloop1:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm14
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 12
+        vpslld  xmm1, xmm1, 20
+        vpor    xmm1, xmm1, xmm8
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm15
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 7
+        vpslld  xmm1, xmm1, 25
+        vpor    xmm1, xmm1, xmm8
+        vpshufd xmm0, xmm0, 93H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 39H
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm14
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 12
+        vpslld  xmm1, xmm1, 20
+        vpor    xmm1, xmm1, xmm8
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxor   xmm3, xmm3, xmm0
+        vpshufb xmm3, xmm3, xmm15
+        vpaddd  xmm2, xmm2, xmm3
+        vpxor   xmm1, xmm1, xmm2
+        vpsrld  xmm8, xmm1, 7
+        vpslld  xmm1, xmm1, 25
+        vpor    xmm1, xmm1, xmm8
+        vpshufd xmm0, xmm0, 39H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 93H
+        dec     al
+        jz      endroundloop1
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0FH
+        vpshufd xmm4, xmm8, 39H
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0AAH
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 88H
+        vpshufd xmm8, xmm8, 78H
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 1EH
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     roundloop1
+endroundloop1:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop1
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+10H], xmm1
+        jmp     unwind
+
+_blake3_hash_many_avx2 ENDP
+blake3_hash_many_avx2 ENDP
+_TEXT ENDS
+
+_RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
+ALIGN   64
+ADD0:
+        dd 0, 1, 2, 3, 4, 5, 6, 7
+
+ADD1:
+        dd 8 dup (8)
+
+BLAKE3_IV_0:
+        dd 8 dup (6A09E667H)
+
+BLAKE3_IV_1:
+        dd 8 dup (0BB67AE85H)
+
+BLAKE3_IV_2:
+        dd 8 dup (3C6EF372H)
+
+BLAKE3_IV_3:
+        dd 8 dup (0A54FF53AH)
+
+BLAKE3_BLOCK_LEN:
+        dd 8 dup (64)
+
+ROT16:
+        db 2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13
+
+ROT8:
+        db 1, 2, 3, 0, 5, 6, 7, 4, 9, 10, 11, 8, 13, 14, 15, 12
+
+CMP_MSB_MASK:
+        dd 8 dup(80000000H)
+
+BLAKE3_IV:
+        dd 6A09E667H, 0BB67AE85H, 3C6EF372H, 0A54FF53AH
+
+_RDATA ENDS
+END
diff --git a/thirdparty/blake3/c/blake3_avx512.c b/thirdparty/blake3/c/blake3_avx512.c
new file mode 100644
index 000000000..f88a32d31
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx512.c
@@ -0,0 +1,1388 @@
+#include "blake3_impl.h"
+
+#include <immintrin.h>
+
+#define _mm_shuffle_ps2(a, b, c)                                               \
+  (_mm_castps_si128(                                                           \
+      _mm_shuffle_ps(_mm_castsi128_ps(a), _mm_castsi128_ps(b), (c))))
+
+INLINE __m128i loadu_128(const uint8_t src[16]) {
+  return _mm_loadu_si128((void*)src);
+}
+
+INLINE __m256i loadu_256(const uint8_t src[32]) {
+  return _mm256_loadu_si256((void*)src);
+}
+
+INLINE __m512i loadu_512(const uint8_t src[64]) {
+  return _mm512_loadu_si512((void*)src);
+}
+
+INLINE void storeu_128(__m128i src, uint8_t dest[16]) {
+  _mm_storeu_si128((void*)dest, src);
+}
+
+INLINE void storeu_256(__m256i src, uint8_t dest[16]) {
+  _mm256_storeu_si256((void*)dest, src);
+}
+
+INLINE void storeu_512(__m512i src, uint8_t dest[16]) {
+  _mm512_storeu_si512((void*)dest, src);
+}
+
+INLINE __m128i add_128(__m128i a, __m128i b) { return _mm_add_epi32(a, b); }
+
+INLINE __m256i add_256(__m256i a, __m256i b) { return _mm256_add_epi32(a, b); }
+
+INLINE __m512i add_512(__m512i a, __m512i b) { return _mm512_add_epi32(a, b); }
+
+INLINE __m128i xor_128(__m128i a, __m128i b) { return _mm_xor_si128(a, b); }
+
+INLINE __m256i xor_256(__m256i a, __m256i b) { return _mm256_xor_si256(a, b); }
+
+INLINE __m512i xor_512(__m512i a, __m512i b) { return _mm512_xor_si512(a, b); }
+
+INLINE __m128i set1_128(uint32_t x) { return _mm_set1_epi32((int32_t)x); }
+
+INLINE __m256i set1_256(uint32_t x) { return _mm256_set1_epi32((int32_t)x); }
+
+INLINE __m512i set1_512(uint32_t x) { return _mm512_set1_epi32((int32_t)x); }
+
+INLINE __m128i set4(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
+  return _mm_setr_epi32((int32_t)a, (int32_t)b, (int32_t)c, (int32_t)d);
+}
+
+INLINE __m128i rot16_128(__m128i x) { return _mm_ror_epi32(x, 16); }
+
+INLINE __m256i rot16_256(__m256i x) { return _mm256_ror_epi32(x, 16); }
+
+INLINE __m512i rot16_512(__m512i x) { return _mm512_ror_epi32(x, 16); }
+
+INLINE __m128i rot12_128(__m128i x) { return _mm_ror_epi32(x, 12); }
+
+INLINE __m256i rot12_256(__m256i x) { return _mm256_ror_epi32(x, 12); }
+
+INLINE __m512i rot12_512(__m512i x) { return _mm512_ror_epi32(x, 12); }
+
+INLINE __m128i rot8_128(__m128i x) { return _mm_ror_epi32(x, 8); }
+
+INLINE __m256i rot8_256(__m256i x) { return _mm256_ror_epi32(x, 8); }
+
+INLINE __m512i rot8_512(__m512i x) { return _mm512_ror_epi32(x, 8); }
+
+INLINE __m128i rot7_128(__m128i x) { return _mm_ror_epi32(x, 7); }
+
+INLINE __m256i rot7_256(__m256i x) { return _mm256_ror_epi32(x, 7); }
+
+INLINE __m512i rot7_512(__m512i x) { return _mm512_ror_epi32(x, 7); }
+
+/*
+ * ----------------------------------------------------------------------------
+ * compress_avx512
+ * ----------------------------------------------------------------------------
+ */
+
+INLINE void g1(__m128i *row0, __m128i *row1, __m128i *row2, __m128i *row3,
+               __m128i m) {
+  *row0 = add_128(add_128(*row0, m), *row1);
+  *row3 = xor_128(*row3, *row0);
+  *row3 = rot16_128(*row3);
+  *row2 = add_128(*row2, *row3);
+  *row1 = xor_128(*row1, *row2);
+  *row1 = rot12_128(*row1);
+}
+
+INLINE void g2(__m128i *row0, __m128i *row1, __m128i *row2, __m128i *row3,
+               __m128i m) {
+  *row0 = add_128(add_128(*row0, m), *row1);
+  *row3 = xor_128(*row3, *row0);
+  *row3 = rot8_128(*row3);
+  *row2 = add_128(*row2, *row3);
+  *row1 = xor_128(*row1, *row2);
+  *row1 = rot7_128(*row1);
+}
+
+// Note the optimization here of leaving row1 as the unrotated row, rather than
+// row0. All the message loads below are adjusted to compensate for this. See
+// discussion at https://github.com/sneves/blake2-avx2/pull/4
+INLINE void diagonalize(__m128i *row0, __m128i *row2, __m128i *row3) {
+  *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE(2, 1, 0, 3));
+  *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE(1, 0, 3, 2));
+  *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE(0, 3, 2, 1));
+}
+
+INLINE void undiagonalize(__m128i *row0, __m128i *row2, __m128i *row3) {
+  *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE(0, 3, 2, 1));
+  *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE(1, 0, 3, 2));
+  *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE(2, 1, 0, 3));
+}
+
+INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
+                         const uint8_t block[BLAKE3_BLOCK_LEN],
+                         uint8_t block_len, uint64_t counter, uint8_t flags) {
+  rows[0] = loadu_128((uint8_t *)&cv[0]);
+  rows[1] = loadu_128((uint8_t *)&cv[4]);
+  rows[2] = set4(IV[0], IV[1], IV[2], IV[3]);
+  rows[3] = set4(counter_low(counter), counter_high(counter),
+                 (uint32_t)block_len, (uint32_t)flags);
+
+  __m128i m0 = loadu_128(&block[sizeof(__m128i) * 0]);
+  __m128i m1 = loadu_128(&block[sizeof(__m128i) * 1]);
+  __m128i m2 = loadu_128(&block[sizeof(__m128i) * 2]);
+  __m128i m3 = loadu_128(&block[sizeof(__m128i) * 3]);
+
+  __m128i t0, t1, t2, t3, tt;
+
+  // Round 1. The first round permutes the message words from the original
+  // input order, into the groups that get mixed in parallel.
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(2, 0, 2, 0)); //  6  4  2  0
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 3, 1)); //  7  5  3  1
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(2, 0, 2, 0)); // 14 12 10  8
+  t2 = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2, 1, 0, 3));   // 12 10  8 14
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 1, 3, 1)); // 15 13 11  9
+  t3 = _mm_shuffle_epi32(t3, _MM_SHUFFLE(2, 1, 0, 3));   // 13 11  9 15
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 2. This round and all following rounds apply a fixed permutation
+  // to the message words from the round before.
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 3
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 4
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 5
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 6
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 7
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+}
+
+void blake3_compress_xof_avx512(const uint32_t cv[8],
+                                const uint8_t block[BLAKE3_BLOCK_LEN],
+                                uint8_t block_len, uint64_t counter,
+                                uint8_t flags, uint8_t out[64]) {
+  __m128i rows[4];
+  compress_pre(rows, cv, block, block_len, counter, flags);
+  storeu_128(xor_128(rows[0], rows[2]), &out[0]);
+  storeu_128(xor_128(rows[1], rows[3]), &out[16]);
+  storeu_128(xor_128(rows[2], loadu_128((uint8_t *)&cv[0])), &out[32]);
+  storeu_128(xor_128(rows[3], loadu_128((uint8_t *)&cv[4])), &out[48]);
+}
+
+void blake3_compress_in_place_avx512(uint32_t cv[8],
+                                     const uint8_t block[BLAKE3_BLOCK_LEN],
+                                     uint8_t block_len, uint64_t counter,
+                                     uint8_t flags) {
+  __m128i rows[4];
+  compress_pre(rows, cv, block, block_len, counter, flags);
+  storeu_128(xor_128(rows[0], rows[2]), (uint8_t *)&cv[0]);
+  storeu_128(xor_128(rows[1], rows[3]), (uint8_t *)&cv[4]);
+}
+
+/*
+ * ----------------------------------------------------------------------------
+ * hash4_avx512
+ * ----------------------------------------------------------------------------
+ */
+
+INLINE void round_fn4(__m128i v[16], __m128i m[16], size_t r) {
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
+  v[0] = add_128(v[0], v[4]);
+  v[1] = add_128(v[1], v[5]);
+  v[2] = add_128(v[2], v[6]);
+  v[3] = add_128(v[3], v[7]);
+  v[12] = xor_128(v[12], v[0]);
+  v[13] = xor_128(v[13], v[1]);
+  v[14] = xor_128(v[14], v[2]);
+  v[15] = xor_128(v[15], v[3]);
+  v[12] = rot16_128(v[12]);
+  v[13] = rot16_128(v[13]);
+  v[14] = rot16_128(v[14]);
+  v[15] = rot16_128(v[15]);
+  v[8] = add_128(v[8], v[12]);
+  v[9] = add_128(v[9], v[13]);
+  v[10] = add_128(v[10], v[14]);
+  v[11] = add_128(v[11], v[15]);
+  v[4] = xor_128(v[4], v[8]);
+  v[5] = xor_128(v[5], v[9]);
+  v[6] = xor_128(v[6], v[10]);
+  v[7] = xor_128(v[7], v[11]);
+  v[4] = rot12_128(v[4]);
+  v[5] = rot12_128(v[5]);
+  v[6] = rot12_128(v[6]);
+  v[7] = rot12_128(v[7]);
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
+  v[0] = add_128(v[0], v[4]);
+  v[1] = add_128(v[1], v[5]);
+  v[2] = add_128(v[2], v[6]);
+  v[3] = add_128(v[3], v[7]);
+  v[12] = xor_128(v[12], v[0]);
+  v[13] = xor_128(v[13], v[1]);
+  v[14] = xor_128(v[14], v[2]);
+  v[15] = xor_128(v[15], v[3]);
+  v[12] = rot8_128(v[12]);
+  v[13] = rot8_128(v[13]);
+  v[14] = rot8_128(v[14]);
+  v[15] = rot8_128(v[15]);
+  v[8] = add_128(v[8], v[12]);
+  v[9] = add_128(v[9], v[13]);
+  v[10] = add_128(v[10], v[14]);
+  v[11] = add_128(v[11], v[15]);
+  v[4] = xor_128(v[4], v[8]);
+  v[5] = xor_128(v[5], v[9]);
+  v[6] = xor_128(v[6], v[10]);
+  v[7] = xor_128(v[7], v[11]);
+  v[4] = rot7_128(v[4]);
+  v[5] = rot7_128(v[5]);
+  v[6] = rot7_128(v[6]);
+  v[7] = rot7_128(v[7]);
+
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
+  v[0] = add_128(v[0], v[5]);
+  v[1] = add_128(v[1], v[6]);
+  v[2] = add_128(v[2], v[7]);
+  v[3] = add_128(v[3], v[4]);
+  v[15] = xor_128(v[15], v[0]);
+  v[12] = xor_128(v[12], v[1]);
+  v[13] = xor_128(v[13], v[2]);
+  v[14] = xor_128(v[14], v[3]);
+  v[15] = rot16_128(v[15]);
+  v[12] = rot16_128(v[12]);
+  v[13] = rot16_128(v[13]);
+  v[14] = rot16_128(v[14]);
+  v[10] = add_128(v[10], v[15]);
+  v[11] = add_128(v[11], v[12]);
+  v[8] = add_128(v[8], v[13]);
+  v[9] = add_128(v[9], v[14]);
+  v[5] = xor_128(v[5], v[10]);
+  v[6] = xor_128(v[6], v[11]);
+  v[7] = xor_128(v[7], v[8]);
+  v[4] = xor_128(v[4], v[9]);
+  v[5] = rot12_128(v[5]);
+  v[6] = rot12_128(v[6]);
+  v[7] = rot12_128(v[7]);
+  v[4] = rot12_128(v[4]);
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
+  v[0] = add_128(v[0], v[5]);
+  v[1] = add_128(v[1], v[6]);
+  v[2] = add_128(v[2], v[7]);
+  v[3] = add_128(v[3], v[4]);
+  v[15] = xor_128(v[15], v[0]);
+  v[12] = xor_128(v[12], v[1]);
+  v[13] = xor_128(v[13], v[2]);
+  v[14] = xor_128(v[14], v[3]);
+  v[15] = rot8_128(v[15]);
+  v[12] = rot8_128(v[12]);
+  v[13] = rot8_128(v[13]);
+  v[14] = rot8_128(v[14]);
+  v[10] = add_128(v[10], v[15]);
+  v[11] = add_128(v[11], v[12]);
+  v[8] = add_128(v[8], v[13]);
+  v[9] = add_128(v[9], v[14]);
+  v[5] = xor_128(v[5], v[10]);
+  v[6] = xor_128(v[6], v[11]);
+  v[7] = xor_128(v[7], v[8]);
+  v[4] = xor_128(v[4], v[9]);
+  v[5] = rot7_128(v[5]);
+  v[6] = rot7_128(v[6]);
+  v[7] = rot7_128(v[7]);
+  v[4] = rot7_128(v[4]);
+}
+
+INLINE void transpose_vecs_128(__m128i vecs[4]) {
+  // Interleave 32-bit lanes. The low unpack is lanes 00/11 and the high is
+  // 22/33. Note that this doesn't split the vector into two lanes, as the
+  // AVX2 counterparts do.
+  __m128i ab_01 = _mm_unpacklo_epi32(vecs[0], vecs[1]);
+  __m128i ab_23 = _mm_unpackhi_epi32(vecs[0], vecs[1]);
+  __m128i cd_01 = _mm_unpacklo_epi32(vecs[2], vecs[3]);
+  __m128i cd_23 = _mm_unpackhi_epi32(vecs[2], vecs[3]);
+
+  // Interleave 64-bit lanes.
+  __m128i abcd_0 = _mm_unpacklo_epi64(ab_01, cd_01);
+  __m128i abcd_1 = _mm_unpackhi_epi64(ab_01, cd_01);
+  __m128i abcd_2 = _mm_unpacklo_epi64(ab_23, cd_23);
+  __m128i abcd_3 = _mm_unpackhi_epi64(ab_23, cd_23);
+
+  vecs[0] = abcd_0;
+  vecs[1] = abcd_1;
+  vecs[2] = abcd_2;
+  vecs[3] = abcd_3;
+}
+
+INLINE void transpose_msg_vecs4(const uint8_t *const *inputs,
+                                size_t block_offset, __m128i out[16]) {
+  out[0] = loadu_128(&inputs[0][block_offset + 0 * sizeof(__m128i)]);
+  out[1] = loadu_128(&inputs[1][block_offset + 0 * sizeof(__m128i)]);
+  out[2] = loadu_128(&inputs[2][block_offset + 0 * sizeof(__m128i)]);
+  out[3] = loadu_128(&inputs[3][block_offset + 0 * sizeof(__m128i)]);
+  out[4] = loadu_128(&inputs[0][block_offset + 1 * sizeof(__m128i)]);
+  out[5] = loadu_128(&inputs[1][block_offset + 1 * sizeof(__m128i)]);
+  out[6] = loadu_128(&inputs[2][block_offset + 1 * sizeof(__m128i)]);
+  out[7] = loadu_128(&inputs[3][block_offset + 1 * sizeof(__m128i)]);
+  out[8] = loadu_128(&inputs[0][block_offset + 2 * sizeof(__m128i)]);
+  out[9] = loadu_128(&inputs[1][block_offset + 2 * sizeof(__m128i)]);
+  out[10] = loadu_128(&inputs[2][block_offset + 2 * sizeof(__m128i)]);
+  out[11] = loadu_128(&inputs[3][block_offset + 2 * sizeof(__m128i)]);
+  out[12] = loadu_128(&inputs[0][block_offset + 3 * sizeof(__m128i)]);
+  out[13] = loadu_128(&inputs[1][block_offset + 3 * sizeof(__m128i)]);
+  out[14] = loadu_128(&inputs[2][block_offset + 3 * sizeof(__m128i)]);
+  out[15] = loadu_128(&inputs[3][block_offset + 3 * sizeof(__m128i)]);
+  for (size_t i = 0; i < 4; ++i) {
+    _mm_prefetch((const void *)&inputs[i][block_offset + 256], _MM_HINT_T0);
+  }
+  transpose_vecs_128(&out[0]);
+  transpose_vecs_128(&out[4]);
+  transpose_vecs_128(&out[8]);
+  transpose_vecs_128(&out[12]);
+}
+
+INLINE void load_counters4(uint64_t counter, bool increment_counter,
+                           __m128i *out_lo, __m128i *out_hi) {
+  uint64_t mask = (increment_counter ? ~0 : 0);
+  __m256i mask_vec = _mm256_set1_epi64x(mask);
+  __m256i deltas = _mm256_setr_epi64x(0, 1, 2, 3);
+  deltas = _mm256_and_si256(mask_vec, deltas);
+  __m256i counters =
+      _mm256_add_epi64(_mm256_set1_epi64x((int64_t)counter), deltas);
+  *out_lo = _mm256_cvtepi64_epi32(counters);
+  *out_hi = _mm256_cvtepi64_epi32(_mm256_srli_epi64(counters, 32));
+}
+
+static
+void blake3_hash4_avx512(const uint8_t *const *inputs, size_t blocks,
+                         const uint32_t key[8], uint64_t counter,
+                         bool increment_counter, uint8_t flags,
+                         uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
+  __m128i h_vecs[8] = {
+      set1_128(key[0]), set1_128(key[1]), set1_128(key[2]), set1_128(key[3]),
+      set1_128(key[4]), set1_128(key[5]), set1_128(key[6]), set1_128(key[7]),
+  };
+  __m128i counter_low_vec, counter_high_vec;
+  load_counters4(counter, increment_counter, &counter_low_vec,
+                 &counter_high_vec);
+  uint8_t block_flags = flags | flags_start;
+
+  for (size_t block = 0; block < blocks; block++) {
+    if (block + 1 == blocks) {
+      block_flags |= flags_end;
+    }
+    __m128i block_len_vec = set1_128(BLAKE3_BLOCK_LEN);
+    __m128i block_flags_vec = set1_128(block_flags);
+    __m128i msg_vecs[16];
+    transpose_msg_vecs4(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);
+
+    __m128i v[16] = {
+        h_vecs[0],       h_vecs[1],        h_vecs[2],       h_vecs[3],
+        h_vecs[4],       h_vecs[5],        h_vecs[6],       h_vecs[7],
+        set1_128(IV[0]), set1_128(IV[1]),  set1_128(IV[2]), set1_128(IV[3]),
+        counter_low_vec, counter_high_vec, block_len_vec,   block_flags_vec,
+    };
+    round_fn4(v, msg_vecs, 0);
+    round_fn4(v, msg_vecs, 1);
+    round_fn4(v, msg_vecs, 2);
+    round_fn4(v, msg_vecs, 3);
+    round_fn4(v, msg_vecs, 4);
+    round_fn4(v, msg_vecs, 5);
+    round_fn4(v, msg_vecs, 6);
+    h_vecs[0] = xor_128(v[0], v[8]);
+    h_vecs[1] = xor_128(v[1], v[9]);
+    h_vecs[2] = xor_128(v[2], v[10]);
+    h_vecs[3] = xor_128(v[3], v[11]);
+    h_vecs[4] = xor_128(v[4], v[12]);
+    h_vecs[5] = xor_128(v[5], v[13]);
+    h_vecs[6] = xor_128(v[6], v[14]);
+    h_vecs[7] = xor_128(v[7], v[15]);
+
+    block_flags = flags;
+  }
+
+  transpose_vecs_128(&h_vecs[0]);
+  transpose_vecs_128(&h_vecs[4]);
+  // The first four vecs now contain the first half of each output, and the
+  // second four vecs contain the second half of each output.
+  storeu_128(h_vecs[0], &out[0 * sizeof(__m128i)]);
+  storeu_128(h_vecs[4], &out[1 * sizeof(__m128i)]);
+  storeu_128(h_vecs[1], &out[2 * sizeof(__m128i)]);
+  storeu_128(h_vecs[5], &out[3 * sizeof(__m128i)]);
+  storeu_128(h_vecs[2], &out[4 * sizeof(__m128i)]);
+  storeu_128(h_vecs[6], &out[5 * sizeof(__m128i)]);
+  storeu_128(h_vecs[3], &out[6 * sizeof(__m128i)]);
+  storeu_128(h_vecs[7], &out[7 * sizeof(__m128i)]);
+}
+
+static
+void blake3_xof4_avx512(const uint32_t cv[8],
+                        const uint8_t block[BLAKE3_BLOCK_LEN],
+                        uint8_t block_len, uint64_t counter, uint8_t flags,
+                        uint8_t out[4 * 64]) {
+  __m128i h_vecs[8] = {
+      set1_128(cv[0]), set1_128(cv[1]), set1_128(cv[2]), set1_128(cv[3]),
+      set1_128(cv[4]), set1_128(cv[5]), set1_128(cv[6]), set1_128(cv[7]),
+  };
+  uint32_t block_words[16];
+  load_block_words(block, block_words);
+  __m128i msg_vecs[16];
+  for (size_t i = 0; i < 16; i++) {
+      msg_vecs[i] = set1_128(block_words[i]);
+  }
+  __m128i counter_low_vec, counter_high_vec;
+  load_counters4(counter, true, &counter_low_vec, &counter_high_vec);
+  __m128i block_len_vec = set1_128(block_len);
+  __m128i block_flags_vec = set1_128(flags);
+  __m128i v[16] = {
+      h_vecs[0],       h_vecs[1],        h_vecs[2],       h_vecs[3],
+      h_vecs[4],       h_vecs[5],        h_vecs[6],       h_vecs[7],
+      set1_128(IV[0]), set1_128(IV[1]),  set1_128(IV[2]), set1_128(IV[3]),
+      counter_low_vec, counter_high_vec, block_len_vec,   block_flags_vec,
+  };
+  round_fn4(v, msg_vecs, 0);
+  round_fn4(v, msg_vecs, 1);
+  round_fn4(v, msg_vecs, 2);
+  round_fn4(v, msg_vecs, 3);
+  round_fn4(v, msg_vecs, 4);
+  round_fn4(v, msg_vecs, 5);
+  round_fn4(v, msg_vecs, 6);
+  for (size_t i = 0; i < 8; i++) {
+      v[i] = xor_128(v[i], v[i+8]);
+      v[i+8] = xor_128(v[i+8], h_vecs[i]);
+  }
+  transpose_vecs_128(&v[0]);
+  transpose_vecs_128(&v[4]);
+  transpose_vecs_128(&v[8]);
+  transpose_vecs_128(&v[12]);
+  for (size_t i = 0; i < 4; i++) {
+      storeu_128(v[i+ 0], &out[(4*i+0) * sizeof(__m128i)]);
+      storeu_128(v[i+ 4], &out[(4*i+1) * sizeof(__m128i)]);
+      storeu_128(v[i+ 8], &out[(4*i+2) * sizeof(__m128i)]);
+      storeu_128(v[i+12], &out[(4*i+3) * sizeof(__m128i)]);
+  }
+}
+
+/*
+ * ----------------------------------------------------------------------------
+ * hash8_avx512
+ * ----------------------------------------------------------------------------
+ */
+
+INLINE void round_fn8(__m256i v[16], __m256i m[16], size_t r) {
+  v[0] = add_256(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
+  v[1] = add_256(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
+  v[2] = add_256(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
+  v[3] = add_256(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
+  v[0] = add_256(v[0], v[4]);
+  v[1] = add_256(v[1], v[5]);
+  v[2] = add_256(v[2], v[6]);
+  v[3] = add_256(v[3], v[7]);
+  v[12] = xor_256(v[12], v[0]);
+  v[13] = xor_256(v[13], v[1]);
+  v[14] = xor_256(v[14], v[2]);
+  v[15] = xor_256(v[15], v[3]);
+  v[12] = rot16_256(v[12]);
+  v[13] = rot16_256(v[13]);
+  v[14] = rot16_256(v[14]);
+  v[15] = rot16_256(v[15]);
+  v[8] = add_256(v[8], v[12]);
+  v[9] = add_256(v[9], v[13]);
+  v[10] = add_256(v[10], v[14]);
+  v[11] = add_256(v[11], v[15]);
+  v[4] = xor_256(v[4], v[8]);
+  v[5] = xor_256(v[5], v[9]);
+  v[6] = xor_256(v[6], v[10]);
+  v[7] = xor_256(v[7], v[11]);
+  v[4] = rot12_256(v[4]);
+  v[5] = rot12_256(v[5]);
+  v[6] = rot12_256(v[6]);
+  v[7] = rot12_256(v[7]);
+  v[0] = add_256(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
+  v[1] = add_256(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
+  v[2] = add_256(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
+  v[3] = add_256(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
+  v[0] = add_256(v[0], v[4]);
+  v[1] = add_256(v[1], v[5]);
+  v[2] = add_256(v[2], v[6]);
+  v[3] = add_256(v[3], v[7]);
+  v[12] = xor_256(v[12], v[0]);
+  v[13] = xor_256(v[13], v[1]);
+  v[14] = xor_256(v[14], v[2]);
+  v[15] = xor_256(v[15], v[3]);
+  v[12] = rot8_256(v[12]);
+  v[13] = rot8_256(v[13]);
+  v[14] = rot8_256(v[14]);
+  v[15] = rot8_256(v[15]);
+  v[8] = add_256(v[8], v[12]);
+  v[9] = add_256(v[9], v[13]);
+  v[10] = add_256(v[10], v[14]);
+  v[11] = add_256(v[11], v[15]);
+  v[4] = xor_256(v[4], v[8]);
+  v[5] = xor_256(v[5], v[9]);
+  v[6] = xor_256(v[6], v[10]);
+  v[7] = xor_256(v[7], v[11]);
+  v[4] = rot7_256(v[4]);
+  v[5] = rot7_256(v[5]);
+  v[6] = rot7_256(v[6]);
+  v[7] = rot7_256(v[7]);
+
+  v[0] = add_256(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
+  v[1] = add_256(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
+  v[2] = add_256(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
+  v[3] = add_256(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
+  v[0] = add_256(v[0], v[5]);
+  v[1] = add_256(v[1], v[6]);
+  v[2] = add_256(v[2], v[7]);
+  v[3] = add_256(v[3], v[4]);
+  v[15] = xor_256(v[15], v[0]);
+  v[12] = xor_256(v[12], v[1]);
+  v[13] = xor_256(v[13], v[2]);
+  v[14] = xor_256(v[14], v[3]);
+  v[15] = rot16_256(v[15]);
+  v[12] = rot16_256(v[12]);
+  v[13] = rot16_256(v[13]);
+  v[14] = rot16_256(v[14]);
+  v[10] = add_256(v[10], v[15]);
+  v[11] = add_256(v[11], v[12]);
+  v[8] = add_256(v[8], v[13]);
+  v[9] = add_256(v[9], v[14]);
+  v[5] = xor_256(v[5], v[10]);
+  v[6] = xor_256(v[6], v[11]);
+  v[7] = xor_256(v[7], v[8]);
+  v[4] = xor_256(v[4], v[9]);
+  v[5] = rot12_256(v[5]);
+  v[6] = rot12_256(v[6]);
+  v[7] = rot12_256(v[7]);
+  v[4] = rot12_256(v[4]);
+  v[0] = add_256(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
+  v[1] = add_256(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
+  v[2] = add_256(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
+  v[3] = add_256(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
+  v[0] = add_256(v[0], v[5]);
+  v[1] = add_256(v[1], v[6]);
+  v[2] = add_256(v[2], v[7]);
+  v[3] = add_256(v[3], v[4]);
+  v[15] = xor_256(v[15], v[0]);
+  v[12] = xor_256(v[12], v[1]);
+  v[13] = xor_256(v[13], v[2]);
+  v[14] = xor_256(v[14], v[3]);
+  v[15] = rot8_256(v[15]);
+  v[12] = rot8_256(v[12]);
+  v[13] = rot8_256(v[13]);
+  v[14] = rot8_256(v[14]);
+  v[10] = add_256(v[10], v[15]);
+  v[11] = add_256(v[11], v[12]);
+  v[8] = add_256(v[8], v[13]);
+  v[9] = add_256(v[9], v[14]);
+  v[5] = xor_256(v[5], v[10]);
+  v[6] = xor_256(v[6], v[11]);
+  v[7] = xor_256(v[7], v[8]);
+  v[4] = xor_256(v[4], v[9]);
+  v[5] = rot7_256(v[5]);
+  v[6] = rot7_256(v[6]);
+  v[7] = rot7_256(v[7]);
+  v[4] = rot7_256(v[4]);
+}
+
+INLINE void transpose_vecs_256(__m256i vecs[8]) {
+  // Interleave 32-bit lanes. The low unpack is lanes 00/11/44/55, and the high
+  // is 22/33/66/77.
+  __m256i ab_0145 = _mm256_unpacklo_epi32(vecs[0], vecs[1]);
+  __m256i ab_2367 = _mm256_unpackhi_epi32(vecs[0], vecs[1]);
+  __m256i cd_0145 = _mm256_unpacklo_epi32(vecs[2], vecs[3]);
+  __m256i cd_2367 = _mm256_unpackhi_epi32(vecs[2], vecs[3]);
+  __m256i ef_0145 = _mm256_unpacklo_epi32(vecs[4], vecs[5]);
+  __m256i ef_2367 = _mm256_unpackhi_epi32(vecs[4], vecs[5]);
+  __m256i gh_0145 = _mm256_unpacklo_epi32(vecs[6], vecs[7]);
+  __m256i gh_2367 = _mm256_unpackhi_epi32(vecs[6], vecs[7]);
+
+  // Interleave 64-bit lanes. The low unpack is lanes 00/22 and the high is
+  // 11/33.
+  __m256i abcd_04 = _mm256_unpacklo_epi64(ab_0145, cd_0145);
+  __m256i abcd_15 = _mm256_unpackhi_epi64(ab_0145, cd_0145);
+  __m256i abcd_26 = _mm256_unpacklo_epi64(ab_2367, cd_2367);
+  __m256i abcd_37 = _mm256_unpackhi_epi64(ab_2367, cd_2367);
+  __m256i efgh_04 = _mm256_unpacklo_epi64(ef_0145, gh_0145);
+  __m256i efgh_15 = _mm256_unpackhi_epi64(ef_0145, gh_0145);
+  __m256i efgh_26 = _mm256_unpacklo_epi64(ef_2367, gh_2367);
+  __m256i efgh_37 = _mm256_unpackhi_epi64(ef_2367, gh_2367);
+
+  // Interleave 128-bit lanes.
+  vecs[0] = _mm256_permute2x128_si256(abcd_04, efgh_04, 0x20);
+  vecs[1] = _mm256_permute2x128_si256(abcd_15, efgh_15, 0x20);
+  vecs[2] = _mm256_permute2x128_si256(abcd_26, efgh_26, 0x20);
+  vecs[3] = _mm256_permute2x128_si256(abcd_37, efgh_37, 0x20);
+  vecs[4] = _mm256_permute2x128_si256(abcd_04, efgh_04, 0x31);
+  vecs[5] = _mm256_permute2x128_si256(abcd_15, efgh_15, 0x31);
+  vecs[6] = _mm256_permute2x128_si256(abcd_26, efgh_26, 0x31);
+  vecs[7] = _mm256_permute2x128_si256(abcd_37, efgh_37, 0x31);
+}
+
+INLINE void transpose_msg_vecs8(const uint8_t *const *inputs,
+                                size_t block_offset, __m256i out[16]) {
+  out[0] = loadu_256(&inputs[0][block_offset + 0 * sizeof(__m256i)]);
+  out[1] = loadu_256(&inputs[1][block_offset + 0 * sizeof(__m256i)]);
+  out[2] = loadu_256(&inputs[2][block_offset + 0 * sizeof(__m256i)]);
+  out[3] = loadu_256(&inputs[3][block_offset + 0 * sizeof(__m256i)]);
+  out[4] = loadu_256(&inputs[4][block_offset + 0 * sizeof(__m256i)]);
+  out[5] = loadu_256(&inputs[5][block_offset + 0 * sizeof(__m256i)]);
+  out[6] = loadu_256(&inputs[6][block_offset + 0 * sizeof(__m256i)]);
+  out[7] = loadu_256(&inputs[7][block_offset + 0 * sizeof(__m256i)]);
+  out[8] = loadu_256(&inputs[0][block_offset + 1 * sizeof(__m256i)]);
+  out[9] = loadu_256(&inputs[1][block_offset + 1 * sizeof(__m256i)]);
+  out[10] = loadu_256(&inputs[2][block_offset + 1 * sizeof(__m256i)]);
+  out[11] = loadu_256(&inputs[3][block_offset + 1 * sizeof(__m256i)]);
+  out[12] = loadu_256(&inputs[4][block_offset + 1 * sizeof(__m256i)]);
+  out[13] = loadu_256(&inputs[5][block_offset + 1 * sizeof(__m256i)]);
+  out[14] = loadu_256(&inputs[6][block_offset + 1 * sizeof(__m256i)]);
+  out[15] = loadu_256(&inputs[7][block_offset + 1 * sizeof(__m256i)]);
+  for (size_t i = 0; i < 8; ++i) {
+    _mm_prefetch((const void *)&inputs[i][block_offset + 256], _MM_HINT_T0);
+  }
+  transpose_vecs_256(&out[0]);
+  transpose_vecs_256(&out[8]);
+}
+
+INLINE void load_counters8(uint64_t counter, bool increment_counter,
+                           __m256i *out_lo, __m256i *out_hi) {
+  uint64_t mask = (increment_counter ? ~0 : 0);
+  __m512i mask_vec = _mm512_set1_epi64(mask);
+  __m512i deltas = _mm512_setr_epi64(0, 1, 2, 3, 4, 5, 6, 7);
+  deltas = _mm512_and_si512(mask_vec, deltas);
+  __m512i counters =
+      _mm512_add_epi64(_mm512_set1_epi64((int64_t)counter), deltas);
+  *out_lo = _mm512_cvtepi64_epi32(counters);
+  *out_hi = _mm512_cvtepi64_epi32(_mm512_srli_epi64(counters, 32));
+}
+
+static
+void blake3_hash8_avx512(const uint8_t *const *inputs, size_t blocks,
+                         const uint32_t key[8], uint64_t counter,
+                         bool increment_counter, uint8_t flags,
+                         uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
+  __m256i h_vecs[8] = {
+      set1_256(key[0]), set1_256(key[1]), set1_256(key[2]), set1_256(key[3]),
+      set1_256(key[4]), set1_256(key[5]), set1_256(key[6]), set1_256(key[7]),
+  };
+  __m256i counter_low_vec, counter_high_vec;
+  load_counters8(counter, increment_counter, &counter_low_vec,
+                 &counter_high_vec);
+  uint8_t block_flags = flags | flags_start;
+
+  for (size_t block = 0; block < blocks; block++) {
+    if (block + 1 == blocks) {
+      block_flags |= flags_end;
+    }
+    __m256i block_len_vec = set1_256(BLAKE3_BLOCK_LEN);
+    __m256i block_flags_vec = set1_256(block_flags);
+    __m256i msg_vecs[16];
+    transpose_msg_vecs8(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);
+
+    __m256i v[16] = {
+        h_vecs[0],       h_vecs[1],        h_vecs[2],       h_vecs[3],
+        h_vecs[4],       h_vecs[5],        h_vecs[6],       h_vecs[7],
+        set1_256(IV[0]), set1_256(IV[1]),  set1_256(IV[2]), set1_256(IV[3]),
+        counter_low_vec, counter_high_vec, block_len_vec,   block_flags_vec,
+    };
+    round_fn8(v, msg_vecs, 0);
+    round_fn8(v, msg_vecs, 1);
+    round_fn8(v, msg_vecs, 2);
+    round_fn8(v, msg_vecs, 3);
+    round_fn8(v, msg_vecs, 4);
+    round_fn8(v, msg_vecs, 5);
+    round_fn8(v, msg_vecs, 6);
+    h_vecs[0] = xor_256(v[0], v[8]);
+    h_vecs[1] = xor_256(v[1], v[9]);
+    h_vecs[2] = xor_256(v[2], v[10]);
+    h_vecs[3] = xor_256(v[3], v[11]);
+    h_vecs[4] = xor_256(v[4], v[12]);
+    h_vecs[5] = xor_256(v[5], v[13]);
+    h_vecs[6] = xor_256(v[6], v[14]);
+    h_vecs[7] = xor_256(v[7], v[15]);
+
+    block_flags = flags;
+  }
+
+  transpose_vecs_256(h_vecs);
+  storeu_256(h_vecs[0], &out[0 * sizeof(__m256i)]);
+  storeu_256(h_vecs[1], &out[1 * sizeof(__m256i)]);
+  storeu_256(h_vecs[2], &out[2 * sizeof(__m256i)]);
+  storeu_256(h_vecs[3], &out[3 * sizeof(__m256i)]);
+  storeu_256(h_vecs[4], &out[4 * sizeof(__m256i)]);
+  storeu_256(h_vecs[5], &out[5 * sizeof(__m256i)]);
+  storeu_256(h_vecs[6], &out[6 * sizeof(__m256i)]);
+  storeu_256(h_vecs[7], &out[7 * sizeof(__m256i)]);
+}
+
+static
+void blake3_xof8_avx512(const uint32_t cv[8],
+                        const uint8_t block[BLAKE3_BLOCK_LEN],
+                        uint8_t block_len, uint64_t counter, uint8_t flags,
+                        uint8_t out[8 * 64]) {
+  __m256i h_vecs[8] = {
+      set1_256(cv[0]), set1_256(cv[1]), set1_256(cv[2]), set1_256(cv[3]),
+      set1_256(cv[4]), set1_256(cv[5]), set1_256(cv[6]), set1_256(cv[7]),
+  };
+  uint32_t block_words[16];
+  load_block_words(block, block_words);
+  __m256i msg_vecs[16];
+  for (size_t i = 0; i < 16; i++) {
+      msg_vecs[i] = set1_256(block_words[i]);
+  }
+  __m256i counter_low_vec, counter_high_vec;
+  load_counters8(counter, true, &counter_low_vec, &counter_high_vec);
+  __m256i block_len_vec = set1_256(block_len);
+  __m256i block_flags_vec = set1_256(flags);
+  __m256i v[16] = {
+      h_vecs[0],       h_vecs[1],        h_vecs[2],       h_vecs[3],
+      h_vecs[4],       h_vecs[5],        h_vecs[6],       h_vecs[7],
+      set1_256(IV[0]), set1_256(IV[1]),  set1_256(IV[2]), set1_256(IV[3]),
+      counter_low_vec, counter_high_vec, block_len_vec,   block_flags_vec,
+  };
+  round_fn8(v, msg_vecs, 0);
+  round_fn8(v, msg_vecs, 1);
+  round_fn8(v, msg_vecs, 2);
+  round_fn8(v, msg_vecs, 3);
+  round_fn8(v, msg_vecs, 4);
+  round_fn8(v, msg_vecs, 5);
+  round_fn8(v, msg_vecs, 6);
+  for (size_t i = 0; i < 8; i++) {
+      v[i] = xor_256(v[i], v[i+8]);
+      v[i+8] = xor_256(v[i+8], h_vecs[i]);
+  }
+  transpose_vecs_256(&v[0]);
+  transpose_vecs_256(&v[8]);
+  for (size_t i = 0; i < 8; i++) {
+      storeu_256(v[i+0], &out[(2*i+0) * sizeof(__m256i)]);
+      storeu_256(v[i+8], &out[(2*i+1) * sizeof(__m256i)]);
+  }
+}
+
+/*
+ * ----------------------------------------------------------------------------
+ * hash16_avx512
+ * ----------------------------------------------------------------------------
+ */
+
+INLINE void round_fn16(__m512i v[16], __m512i m[16], size_t r) {
+  v[0] = add_512(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
+  v[1] = add_512(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
+  v[2] = add_512(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
+  v[3] = add_512(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
+  v[0] = add_512(v[0], v[4]);
+  v[1] = add_512(v[1], v[5]);
+  v[2] = add_512(v[2], v[6]);
+  v[3] = add_512(v[3], v[7]);
+  v[12] = xor_512(v[12], v[0]);
+  v[13] = xor_512(v[13], v[1]);
+  v[14] = xor_512(v[14], v[2]);
+  v[15] = xor_512(v[15], v[3]);
+  v[12] = rot16_512(v[12]);
+  v[13] = rot16_512(v[13]);
+  v[14] = rot16_512(v[14]);
+  v[15] = rot16_512(v[15]);
+  v[8] = add_512(v[8], v[12]);
+  v[9] = add_512(v[9], v[13]);
+  v[10] = add_512(v[10], v[14]);
+  v[11] = add_512(v[11], v[15]);
+  v[4] = xor_512(v[4], v[8]);
+  v[5] = xor_512(v[5], v[9]);
+  v[6] = xor_512(v[6], v[10]);
+  v[7] = xor_512(v[7], v[11]);
+  v[4] = rot12_512(v[4]);
+  v[5] = rot12_512(v[5]);
+  v[6] = rot12_512(v[6]);
+  v[7] = rot12_512(v[7]);
+  v[0] = add_512(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
+  v[1] = add_512(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
+  v[2] = add_512(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
+  v[3] = add_512(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
+  v[0] = add_512(v[0], v[4]);
+  v[1] = add_512(v[1], v[5]);
+  v[2] = add_512(v[2], v[6]);
+  v[3] = add_512(v[3], v[7]);
+  v[12] = xor_512(v[12], v[0]);
+  v[13] = xor_512(v[13], v[1]);
+  v[14] = xor_512(v[14], v[2]);
+  v[15] = xor_512(v[15], v[3]);
+  v[12] = rot8_512(v[12]);
+  v[13] = rot8_512(v[13]);
+  v[14] = rot8_512(v[14]);
+  v[15] = rot8_512(v[15]);
+  v[8] = add_512(v[8], v[12]);
+  v[9] = add_512(v[9], v[13]);
+  v[10] = add_512(v[10], v[14]);
+  v[11] = add_512(v[11], v[15]);
+  v[4] = xor_512(v[4], v[8]);
+  v[5] = xor_512(v[5], v[9]);
+  v[6] = xor_512(v[6], v[10]);
+  v[7] = xor_512(v[7], v[11]);
+  v[4] = rot7_512(v[4]);
+  v[5] = rot7_512(v[5]);
+  v[6] = rot7_512(v[6]);
+  v[7] = rot7_512(v[7]);
+
+  v[0] = add_512(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
+  v[1] = add_512(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
+  v[2] = add_512(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
+  v[3] = add_512(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
+  v[0] = add_512(v[0], v[5]);
+  v[1] = add_512(v[1], v[6]);
+  v[2] = add_512(v[2], v[7]);
+  v[3] = add_512(v[3], v[4]);
+  v[15] = xor_512(v[15], v[0]);
+  v[12] = xor_512(v[12], v[1]);
+  v[13] = xor_512(v[13], v[2]);
+  v[14] = xor_512(v[14], v[3]);
+  v[15] = rot16_512(v[15]);
+  v[12] = rot16_512(v[12]);
+  v[13] = rot16_512(v[13]);
+  v[14] = rot16_512(v[14]);
+  v[10] = add_512(v[10], v[15]);
+  v[11] = add_512(v[11], v[12]);
+  v[8] = add_512(v[8], v[13]);
+  v[9] = add_512(v[9], v[14]);
+  v[5] = xor_512(v[5], v[10]);
+  v[6] = xor_512(v[6], v[11]);
+  v[7] = xor_512(v[7], v[8]);
+  v[4] = xor_512(v[4], v[9]);
+  v[5] = rot12_512(v[5]);
+  v[6] = rot12_512(v[6]);
+  v[7] = rot12_512(v[7]);
+  v[4] = rot12_512(v[4]);
+  v[0] = add_512(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
+  v[1] = add_512(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
+  v[2] = add_512(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
+  v[3] = add_512(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
+  v[0] = add_512(v[0], v[5]);
+  v[1] = add_512(v[1], v[6]);
+  v[2] = add_512(v[2], v[7]);
+  v[3] = add_512(v[3], v[4]);
+  v[15] = xor_512(v[15], v[0]);
+  v[12] = xor_512(v[12], v[1]);
+  v[13] = xor_512(v[13], v[2]);
+  v[14] = xor_512(v[14], v[3]);
+  v[15] = rot8_512(v[15]);
+  v[12] = rot8_512(v[12]);
+  v[13] = rot8_512(v[13]);
+  v[14] = rot8_512(v[14]);
+  v[10] = add_512(v[10], v[15]);
+  v[11] = add_512(v[11], v[12]);
+  v[8] = add_512(v[8], v[13]);
+  v[9] = add_512(v[9], v[14]);
+  v[5] = xor_512(v[5], v[10]);
+  v[6] = xor_512(v[6], v[11]);
+  v[7] = xor_512(v[7], v[8]);
+  v[4] = xor_512(v[4], v[9]);
+  v[5] = rot7_512(v[5]);
+  v[6] = rot7_512(v[6]);
+  v[7] = rot7_512(v[7]);
+  v[4] = rot7_512(v[4]);
+}
+
+// 0b10001000, or lanes a0/a2/b0/b2 in little-endian order
+#define LO_IMM8 0x88
+
+INLINE __m512i unpack_lo_128(__m512i a, __m512i b) {
+  return _mm512_shuffle_i32x4(a, b, LO_IMM8);
+}
+
+// 0b11011101, or lanes a1/a3/b1/b3 in little-endian order
+#define HI_IMM8 0xdd
+
+INLINE __m512i unpack_hi_128(__m512i a, __m512i b) {
+  return _mm512_shuffle_i32x4(a, b, HI_IMM8);
+}
+
+INLINE void transpose_vecs_512(__m512i vecs[16]) {
+  // Interleave 32-bit lanes. The _0 unpack is lanes
+  // 0/0/1/1/4/4/5/5/8/8/9/9/12/12/13/13, and the _2 unpack is lanes
+  // 2/2/3/3/6/6/7/7/10/10/11/11/14/14/15/15.
+  __m512i ab_0 = _mm512_unpacklo_epi32(vecs[0], vecs[1]);
+  __m512i ab_2 = _mm512_unpackhi_epi32(vecs[0], vecs[1]);
+  __m512i cd_0 = _mm512_unpacklo_epi32(vecs[2], vecs[3]);
+  __m512i cd_2 = _mm512_unpackhi_epi32(vecs[2], vecs[3]);
+  __m512i ef_0 = _mm512_unpacklo_epi32(vecs[4], vecs[5]);
+  __m512i ef_2 = _mm512_unpackhi_epi32(vecs[4], vecs[5]);
+  __m512i gh_0 = _mm512_unpacklo_epi32(vecs[6], vecs[7]);
+  __m512i gh_2 = _mm512_unpackhi_epi32(vecs[6], vecs[7]);
+  __m512i ij_0 = _mm512_unpacklo_epi32(vecs[8], vecs[9]);
+  __m512i ij_2 = _mm512_unpackhi_epi32(vecs[8], vecs[9]);
+  __m512i kl_0 = _mm512_unpacklo_epi32(vecs[10], vecs[11]);
+  __m512i kl_2 = _mm512_unpackhi_epi32(vecs[10], vecs[11]);
+  __m512i mn_0 = _mm512_unpacklo_epi32(vecs[12], vecs[13]);
+  __m512i mn_2 = _mm512_unpackhi_epi32(vecs[12], vecs[13]);
+  __m512i op_0 = _mm512_unpacklo_epi32(vecs[14], vecs[15]);
+  __m512i op_2 = _mm512_unpackhi_epi32(vecs[14], vecs[15]);
+
+  // Interleave 64-bit lanes. The _0 unpack is lanes
+  // 0/0/0/0/4/4/4/4/8/8/8/8/12/12/12/12, the _1 unpack is lanes
+  // 1/1/1/1/5/5/5/5/9/9/9/9/13/13/13/13, the _2 unpack is lanes
+  // 2/2/2/2/6/6/6/6/10/10/10/10/14/14/14/14, and the _3 unpack is lanes
+  // 3/3/3/3/7/7/7/7/11/11/11/11/15/15/15/15.
+  __m512i abcd_0 = _mm512_unpacklo_epi64(ab_0, cd_0);
+  __m512i abcd_1 = _mm512_unpackhi_epi64(ab_0, cd_0);
+  __m512i abcd_2 = _mm512_unpacklo_epi64(ab_2, cd_2);
+  __m512i abcd_3 = _mm512_unpackhi_epi64(ab_2, cd_2);
+  __m512i efgh_0 = _mm512_unpacklo_epi64(ef_0, gh_0);
+  __m512i efgh_1 = _mm512_unpackhi_epi64(ef_0, gh_0);
+  __m512i efgh_2 = _mm512_unpacklo_epi64(ef_2, gh_2);
+  __m512i efgh_3 = _mm512_unpackhi_epi64(ef_2, gh_2);
+  __m512i ijkl_0 = _mm512_unpacklo_epi64(ij_0, kl_0);
+  __m512i ijkl_1 = _mm512_unpackhi_epi64(ij_0, kl_0);
+  __m512i ijkl_2 = _mm512_unpacklo_epi64(ij_2, kl_2);
+  __m512i ijkl_3 = _mm512_unpackhi_epi64(ij_2, kl_2);
+  __m512i mnop_0 = _mm512_unpacklo_epi64(mn_0, op_0);
+  __m512i mnop_1 = _mm512_unpackhi_epi64(mn_0, op_0);
+  __m512i mnop_2 = _mm512_unpacklo_epi64(mn_2, op_2);
+  __m512i mnop_3 = _mm512_unpackhi_epi64(mn_2, op_2);
+
+  // Interleave 128-bit lanes. The _0 unpack is
+  // 0/0/0/0/8/8/8/8/0/0/0/0/8/8/8/8, the _1 unpack is
+  // 1/1/1/1/9/9/9/9/1/1/1/1/9/9/9/9, and so on.
+  __m512i abcdefgh_0 = unpack_lo_128(abcd_0, efgh_0);
+  __m512i abcdefgh_1 = unpack_lo_128(abcd_1, efgh_1);
+  __m512i abcdefgh_2 = unpack_lo_128(abcd_2, efgh_2);
+  __m512i abcdefgh_3 = unpack_lo_128(abcd_3, efgh_3);
+  __m512i abcdefgh_4 = unpack_hi_128(abcd_0, efgh_0);
+  __m512i abcdefgh_5 = unpack_hi_128(abcd_1, efgh_1);
+  __m512i abcdefgh_6 = unpack_hi_128(abcd_2, efgh_2);
+  __m512i abcdefgh_7 = unpack_hi_128(abcd_3, efgh_3);
+  __m512i ijklmnop_0 = unpack_lo_128(ijkl_0, mnop_0);
+  __m512i ijklmnop_1 = unpack_lo_128(ijkl_1, mnop_1);
+  __m512i ijklmnop_2 = unpack_lo_128(ijkl_2, mnop_2);
+  __m512i ijklmnop_3 = unpack_lo_128(ijkl_3, mnop_3);
+  __m512i ijklmnop_4 = unpack_hi_128(ijkl_0, mnop_0);
+  __m512i ijklmnop_5 = unpack_hi_128(ijkl_1, mnop_1);
+  __m512i ijklmnop_6 = unpack_hi_128(ijkl_2, mnop_2);
+  __m512i ijklmnop_7 = unpack_hi_128(ijkl_3, mnop_3);
+
+  // Interleave 128-bit lanes again for the final outputs.
+  vecs[0] = unpack_lo_128(abcdefgh_0, ijklmnop_0);
+  vecs[1] = unpack_lo_128(abcdefgh_1, ijklmnop_1);
+  vecs[2] = unpack_lo_128(abcdefgh_2, ijklmnop_2);
+  vecs[3] = unpack_lo_128(abcdefgh_3, ijklmnop_3);
+  vecs[4] = unpack_lo_128(abcdefgh_4, ijklmnop_4);
+  vecs[5] = unpack_lo_128(abcdefgh_5, ijklmnop_5);
+  vecs[6] = unpack_lo_128(abcdefgh_6, ijklmnop_6);
+  vecs[7] = unpack_lo_128(abcdefgh_7, ijklmnop_7);
+  vecs[8] = unpack_hi_128(abcdefgh_0, ijklmnop_0);
+  vecs[9] = unpack_hi_128(abcdefgh_1, ijklmnop_1);
+  vecs[10] = unpack_hi_128(abcdefgh_2, ijklmnop_2);
+  vecs[11] = unpack_hi_128(abcdefgh_3, ijklmnop_3);
+  vecs[12] = unpack_hi_128(abcdefgh_4, ijklmnop_4);
+  vecs[13] = unpack_hi_128(abcdefgh_5, ijklmnop_5);
+  vecs[14] = unpack_hi_128(abcdefgh_6, ijklmnop_6);
+  vecs[15] = unpack_hi_128(abcdefgh_7, ijklmnop_7);
+}
+
+INLINE void transpose_msg_vecs16(const uint8_t *const *inputs,
+                                 size_t block_offset, __m512i out[16]) {
+  out[0] = loadu_512(&inputs[0][block_offset]);
+  out[1] = loadu_512(&inputs[1][block_offset]);
+  out[2] = loadu_512(&inputs[2][block_offset]);
+  out[3] = loadu_512(&inputs[3][block_offset]);
+  out[4] = loadu_512(&inputs[4][block_offset]);
+  out[5] = loadu_512(&inputs[5][block_offset]);
+  out[6] = loadu_512(&inputs[6][block_offset]);
+  out[7] = loadu_512(&inputs[7][block_offset]);
+  out[8] = loadu_512(&inputs[8][block_offset]);
+  out[9] = loadu_512(&inputs[9][block_offset]);
+  out[10] = loadu_512(&inputs[10][block_offset]);
+  out[11] = loadu_512(&inputs[11][block_offset]);
+  out[12] = loadu_512(&inputs[12][block_offset]);
+  out[13] = loadu_512(&inputs[13][block_offset]);
+  out[14] = loadu_512(&inputs[14][block_offset]);
+  out[15] = loadu_512(&inputs[15][block_offset]);
+  for (size_t i = 0; i < 16; ++i) {
+    _mm_prefetch((const void *)&inputs[i][block_offset + 256], _MM_HINT_T0);
+  }
+  transpose_vecs_512(out);
+}
+
+INLINE void load_counters16(uint64_t counter, bool increment_counter,
+                            __m512i *out_lo, __m512i *out_hi) {
+  const __m512i mask = _mm512_set1_epi32(-(int32_t)increment_counter);
+  const __m512i deltas = _mm512_set_epi32(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+  const __m512i masked_deltas = _mm512_and_si512(deltas, mask);
+  const __m512i low_words = _mm512_add_epi32(
+    _mm512_set1_epi32((int32_t)counter),
+    masked_deltas);
+  // The carry bit is 1 if the high bit of the word was 1 before addition and is
+  // 0 after.
+  // NOTE: It would be a bit more natural to use _mm512_cmp_epu32_mask to
+  // compute the carry bits here, and originally we did, but that intrinsic is
+  // broken under GCC 5.4. See https://github.com/BLAKE3-team/BLAKE3/issues/271.
+  const __m512i carries = _mm512_srli_epi32(
+    _mm512_andnot_si512(
+        low_words, // 0 after (gets inverted by andnot)
+        _mm512_set1_epi32((int32_t)counter)), // and 1 before
+    31);
+  const __m512i high_words = _mm512_add_epi32(
+    _mm512_set1_epi32((int32_t)(counter >> 32)),
+    carries);
+  *out_lo = low_words;
+  *out_hi = high_words;
+}
+
+static
+void blake3_hash16_avx512(const uint8_t *const *inputs, size_t blocks,
+                          const uint32_t key[8], uint64_t counter,
+                          bool increment_counter, uint8_t flags,
+                          uint8_t flags_start, uint8_t flags_end,
+                          uint8_t *out) {
+  __m512i h_vecs[8] = {
+      set1_512(key[0]), set1_512(key[1]), set1_512(key[2]), set1_512(key[3]),
+      set1_512(key[4]), set1_512(key[5]), set1_512(key[6]), set1_512(key[7]),
+  };
+  __m512i counter_low_vec, counter_high_vec;
+  load_counters16(counter, increment_counter, &counter_low_vec,
+                  &counter_high_vec);
+  uint8_t block_flags = flags | flags_start;
+
+  for (size_t block = 0; block < blocks; block++) {
+    if (block + 1 == blocks) {
+      block_flags |= flags_end;
+    }
+    __m512i block_len_vec = set1_512(BLAKE3_BLOCK_LEN);
+    __m512i block_flags_vec = set1_512(block_flags);
+    __m512i msg_vecs[16];
+    transpose_msg_vecs16(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);
+
+    __m512i v[16] = {
+        h_vecs[0],       h_vecs[1],        h_vecs[2],       h_vecs[3],
+        h_vecs[4],       h_vecs[5],        h_vecs[6],       h_vecs[7],
+        set1_512(IV[0]), set1_512(IV[1]),  set1_512(IV[2]), set1_512(IV[3]),
+        counter_low_vec, counter_high_vec, block_len_vec,   block_flags_vec,
+    };
+    round_fn16(v, msg_vecs, 0);
+    round_fn16(v, msg_vecs, 1);
+    round_fn16(v, msg_vecs, 2);
+    round_fn16(v, msg_vecs, 3);
+    round_fn16(v, msg_vecs, 4);
+    round_fn16(v, msg_vecs, 5);
+    round_fn16(v, msg_vecs, 6);
+    h_vecs[0] = xor_512(v[0], v[8]);
+    h_vecs[1] = xor_512(v[1], v[9]);
+    h_vecs[2] = xor_512(v[2], v[10]);
+    h_vecs[3] = xor_512(v[3], v[11]);
+    h_vecs[4] = xor_512(v[4], v[12]);
+    h_vecs[5] = xor_512(v[5], v[13]);
+    h_vecs[6] = xor_512(v[6], v[14]);
+    h_vecs[7] = xor_512(v[7], v[15]);
+
+    block_flags = flags;
+  }
+
+  // transpose_vecs_512 operates on a 16x16 matrix of words, but we only have 8
+  // state vectors. Pad the matrix with zeros. After transposition, store the
+  // lower half of each vector.
+  __m512i padded[16] = {
+      h_vecs[0],   h_vecs[1],   h_vecs[2],   h_vecs[3],
+      h_vecs[4],   h_vecs[5],   h_vecs[6],   h_vecs[7],
+      set1_512(0), set1_512(0), set1_512(0), set1_512(0),
+      set1_512(0), set1_512(0), set1_512(0), set1_512(0),
+  };
+  transpose_vecs_512(padded);
+  _mm256_mask_storeu_epi32(&out[0 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[0]));
+  _mm256_mask_storeu_epi32(&out[1 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[1]));
+  _mm256_mask_storeu_epi32(&out[2 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[2]));
+  _mm256_mask_storeu_epi32(&out[3 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[3]));
+  _mm256_mask_storeu_epi32(&out[4 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[4]));
+  _mm256_mask_storeu_epi32(&out[5 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[5]));
+  _mm256_mask_storeu_epi32(&out[6 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[6]));
+  _mm256_mask_storeu_epi32(&out[7 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[7]));
+  _mm256_mask_storeu_epi32(&out[8 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[8]));
+  _mm256_mask_storeu_epi32(&out[9 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[9]));
+  _mm256_mask_storeu_epi32(&out[10 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[10]));
+  _mm256_mask_storeu_epi32(&out[11 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[11]));
+  _mm256_mask_storeu_epi32(&out[12 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[12]));
+  _mm256_mask_storeu_epi32(&out[13 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[13]));
+  _mm256_mask_storeu_epi32(&out[14 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[14]));
+  _mm256_mask_storeu_epi32(&out[15 * sizeof(__m256i)], (__mmask8)-1, _mm512_castsi512_si256(padded[15]));
+}
+
+static
+void blake3_xof16_avx512(const uint32_t cv[8],
+                        const uint8_t block[BLAKE3_BLOCK_LEN],
+                        uint8_t block_len, uint64_t counter, uint8_t flags,
+                        uint8_t out[16 * 64]) {
+  __m512i h_vecs[8] = {
+      set1_512(cv[0]), set1_512(cv[1]), set1_512(cv[2]), set1_512(cv[3]),
+      set1_512(cv[4]), set1_512(cv[5]), set1_512(cv[6]), set1_512(cv[7]),
+  };
+  uint32_t block_words[16];
+  load_block_words(block, block_words);
+  __m512i msg_vecs[16];
+  for (size_t i = 0; i < 16; i++) {
+      msg_vecs[i] = set1_512(block_words[i]);
+  }
+  __m512i counter_low_vec, counter_high_vec;
+  load_counters16(counter, true, &counter_low_vec, &counter_high_vec);
+  __m512i block_len_vec = set1_512(block_len);
+  __m512i block_flags_vec = set1_512(flags);
+  __m512i v[16] = {
+      h_vecs[0],       h_vecs[1],        h_vecs[2],       h_vecs[3],
+      h_vecs[4],       h_vecs[5],        h_vecs[6],       h_vecs[7],
+      set1_512(IV[0]), set1_512(IV[1]),  set1_512(IV[2]), set1_512(IV[3]),
+      counter_low_vec, counter_high_vec, block_len_vec,   block_flags_vec,
+  };
+  round_fn16(v, msg_vecs, 0);
+  round_fn16(v, msg_vecs, 1);
+  round_fn16(v, msg_vecs, 2);
+  round_fn16(v, msg_vecs, 3);
+  round_fn16(v, msg_vecs, 4);
+  round_fn16(v, msg_vecs, 5);
+  round_fn16(v, msg_vecs, 6);
+  for (size_t i = 0; i < 8; i++) {
+      v[i] = xor_512(v[i], v[i+8]);
+      v[i+8] = xor_512(v[i+8], h_vecs[i]);
+  }
+  transpose_vecs_512(&v[0]);
+  for (size_t i = 0; i < 16; i++) {
+      storeu_512(v[i], &out[i * sizeof(__m512i)]);
+  }
+}
+
+/*
+ * ----------------------------------------------------------------------------
+ * hash_many_avx512
+ * ----------------------------------------------------------------------------
+ */
+
+INLINE void hash_one_avx512(const uint8_t *input, size_t blocks,
+                            const uint32_t key[8], uint64_t counter,
+                            uint8_t flags, uint8_t flags_start,
+                            uint8_t flags_end, uint8_t out[BLAKE3_OUT_LEN]) {
+  uint32_t cv[8];
+  memcpy(cv, key, BLAKE3_KEY_LEN);
+  uint8_t block_flags = flags | flags_start;
+  while (blocks > 0) {
+    if (blocks == 1) {
+      block_flags |= flags_end;
+    }
+    blake3_compress_in_place_avx512(cv, input, BLAKE3_BLOCK_LEN, counter,
+                                    block_flags);
+    input = &input[BLAKE3_BLOCK_LEN];
+    blocks -= 1;
+    block_flags = flags;
+  }
+  memcpy(out, cv, BLAKE3_OUT_LEN);
+}
+
+void blake3_hash_many_avx512(const uint8_t *const *inputs, size_t num_inputs,
+                             size_t blocks, const uint32_t key[8],
+                             uint64_t counter, bool increment_counter,
+                             uint8_t flags, uint8_t flags_start,
+                             uint8_t flags_end, uint8_t *out) {
+  while (num_inputs >= 16) {
+    blake3_hash16_avx512(inputs, blocks, key, counter, increment_counter, flags,
+                         flags_start, flags_end, out);
+    if (increment_counter) {
+      counter += 16;
+    }
+    inputs += 16;
+    num_inputs -= 16;
+    out = &out[16 * BLAKE3_OUT_LEN];
+  }
+  while (num_inputs >= 8) {
+    blake3_hash8_avx512(inputs, blocks, key, counter, increment_counter, flags,
+                        flags_start, flags_end, out);
+    if (increment_counter) {
+      counter += 8;
+    }
+    inputs += 8;
+    num_inputs -= 8;
+    out = &out[8 * BLAKE3_OUT_LEN];
+  }
+  while (num_inputs >= 4) {
+    blake3_hash4_avx512(inputs, blocks, key, counter, increment_counter, flags,
+                        flags_start, flags_end, out);
+    if (increment_counter) {
+      counter += 4;
+    }
+    inputs += 4;
+    num_inputs -= 4;
+    out = &out[4 * BLAKE3_OUT_LEN];
+  }
+  while (num_inputs > 0) {
+    hash_one_avx512(inputs[0], blocks, key, counter, flags, flags_start,
+                    flags_end, out);
+    if (increment_counter) {
+      counter += 1;
+    }
+    inputs += 1;
+    num_inputs -= 1;
+    out = &out[BLAKE3_OUT_LEN];
+  }
+}
+
+void blake3_xof_many_avx512(const uint32_t cv[8],
+                            const uint8_t block[BLAKE3_BLOCK_LEN],
+                            uint8_t block_len, uint64_t counter, uint8_t flags,
+                            uint8_t* out, size_t outblocks) {
+  while (outblocks >= 16) {
+    blake3_xof16_avx512(cv, block, block_len, counter, flags, out);
+    counter += 16;
+    outblocks -= 16;
+    out += 16 * BLAKE3_BLOCK_LEN;
+  }
+  while (outblocks >= 8) {
+    blake3_xof8_avx512(cv, block, block_len, counter, flags, out);
+    counter += 8;
+    outblocks -= 8;
+    out += 8 * BLAKE3_BLOCK_LEN;
+  }
+  while (outblocks >= 4) {
+    blake3_xof4_avx512(cv, block, block_len, counter, flags, out);
+    counter += 4;
+    outblocks -= 4;
+    out += 4 * BLAKE3_BLOCK_LEN;
+  }
+  while (outblocks > 0) {
+    blake3_compress_xof_avx512(cv, block, block_len, counter, flags, out);
+    counter += 1;
+    outblocks -= 1;
+    out += BLAKE3_BLOCK_LEN;
+  }
+}
diff --git a/thirdparty/blake3/c/blake3_avx512_x86-64_unix.S b/thirdparty/blake3/c/blake3_avx512_x86-64_unix.S
new file mode 100644
index 000000000..9642e413a
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx512_x86-64_unix.S
@@ -0,0 +1,4824 @@
+#if defined(__ELF__) && defined(__linux__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+#if defined(__ELF__) && defined(__CET__) && defined(__has_include)
+#if __has_include(<cet.h>)
+#include <cet.h>
+#endif
+#endif
+
+#if !defined(_CET_ENDBR)
+#define _CET_ENDBR
+#endif
+
+.intel_syntax noprefix
+.global _blake3_hash_many_avx512
+.global blake3_hash_many_avx512
+.global blake3_compress_in_place_avx512
+.global _blake3_compress_in_place_avx512
+.global blake3_compress_xof_avx512
+.global _blake3_compress_xof_avx512
+.global blake3_xof_many_avx512
+.global _blake3_xof_many_avx512
+
+#ifdef __APPLE__
+.text
+#else
+.section .text
+#endif
+.p2align  6
+_blake3_hash_many_avx512:
+blake3_hash_many_avx512:
+        _CET_ENDBR
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 144
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        neg     r9
+        kmovw   k1, r9d
+        vmovd   xmm0, r8d
+        vpbroadcastd ymm0, xmm0
+        shr     r8, 32
+        vmovd   xmm1, r8d
+        vpbroadcastd ymm1, xmm1
+        vmovdqa ymm4, ymm1
+        vmovdqa ymm5, ymm1
+        vpaddd  ymm2, ymm0, ymmword ptr [ADD0+rip]
+        vpaddd  ymm3, ymm0, ymmword ptr [ADD0+32+rip]
+        vpcmpltud k2, ymm2, ymm0
+        vpcmpltud k3, ymm3, ymm0
+        vpaddd  ymm4 {k2}, ymm4, dword ptr [ADD1+rip] {1to8}
+        vpaddd  ymm5 {k3}, ymm5, dword ptr [ADD1+rip] {1to8}
+        knotw   k2, k1
+        vmovdqa32 ymm2 {k2}, ymm0
+        vmovdqa32 ymm3 {k2}, ymm0
+        vmovdqa32 ymm4 {k2}, ymm1
+        vmovdqa32 ymm5 {k2}, ymm1
+        vmovdqa ymmword ptr [rsp], ymm2
+        vmovdqa ymmword ptr [rsp+0x1*0x20], ymm3
+        vmovdqa ymmword ptr [rsp+0x2*0x20], ymm4
+        vmovdqa ymmword ptr [rsp+0x3*0x20], ymm5
+        shl     rdx, 6
+        mov     qword ptr [rsp+0x80], rdx
+        cmp     rsi, 16
+        jc      3f
+2:
+        vpbroadcastd zmm0, dword ptr [rcx]
+        vpbroadcastd zmm1, dword ptr [rcx+0x1*0x4]
+        vpbroadcastd zmm2, dword ptr [rcx+0x2*0x4]
+        vpbroadcastd zmm3, dword ptr [rcx+0x3*0x4]
+        vpbroadcastd zmm4, dword ptr [rcx+0x4*0x4]
+        vpbroadcastd zmm5, dword ptr [rcx+0x5*0x4]
+        vpbroadcastd zmm6, dword ptr [rcx+0x6*0x4]
+        vpbroadcastd zmm7, dword ptr [rcx+0x7*0x4]
+        movzx   eax, byte ptr [rbp+0x38]
+        movzx   ebx, byte ptr [rbp+0x40]
+        or      eax, ebx
+        xor     edx, edx
+.p2align 5
+9:
+        movzx   ebx, byte ptr [rbp+0x48]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+0x80]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+0x88], eax
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x40]
+        mov     r13, qword ptr [rdi+0x48]
+        mov     r14, qword ptr [rdi+0x50]
+        mov     r15, qword ptr [rdi+0x58]
+        vmovdqu32 ymm16, ymmword ptr [rdx+r8-0x2*0x20]
+        vinserti64x4 zmm16, zmm16, ymmword ptr [rdx+r12-0x2*0x20], 0x01
+        vmovdqu32 ymm17, ymmword ptr [rdx+r9-0x2*0x20]
+        vinserti64x4 zmm17, zmm17, ymmword ptr [rdx+r13-0x2*0x20], 0x01
+        vpunpcklqdq zmm8, zmm16, zmm17
+        vpunpckhqdq zmm9, zmm16, zmm17
+        vmovdqu32 ymm18, ymmword ptr [rdx+r10-0x2*0x20]
+        vinserti64x4 zmm18, zmm18, ymmword ptr [rdx+r14-0x2*0x20], 0x01
+        vmovdqu32 ymm19, ymmword ptr [rdx+r11-0x2*0x20]
+        vinserti64x4 zmm19, zmm19, ymmword ptr [rdx+r15-0x2*0x20], 0x01
+        vpunpcklqdq zmm10, zmm18, zmm19
+        vpunpckhqdq zmm11, zmm18, zmm19
+        mov     r8, qword ptr [rdi+0x20]
+        mov     r9, qword ptr [rdi+0x28]
+        mov     r10, qword ptr [rdi+0x30]
+        mov     r11, qword ptr [rdi+0x38]
+        mov     r12, qword ptr [rdi+0x60]
+        mov     r13, qword ptr [rdi+0x68]
+        mov     r14, qword ptr [rdi+0x70]
+        mov     r15, qword ptr [rdi+0x78]
+        vmovdqu32 ymm16, ymmword ptr [rdx+r8-0x2*0x20]
+        vinserti64x4 zmm16, zmm16, ymmword ptr [rdx+r12-0x2*0x20], 0x01
+        vmovdqu32 ymm17, ymmword ptr [rdx+r9-0x2*0x20]
+        vinserti64x4 zmm17, zmm17, ymmword ptr [rdx+r13-0x2*0x20], 0x01
+        vpunpcklqdq zmm12, zmm16, zmm17
+        vpunpckhqdq zmm13, zmm16, zmm17
+        vmovdqu32 ymm18, ymmword ptr [rdx+r10-0x2*0x20]
+        vinserti64x4 zmm18, zmm18, ymmword ptr [rdx+r14-0x2*0x20], 0x01
+        vmovdqu32 ymm19, ymmword ptr [rdx+r11-0x2*0x20]
+        vinserti64x4 zmm19, zmm19, ymmword ptr [rdx+r15-0x2*0x20], 0x01
+        vpunpcklqdq zmm14, zmm18, zmm19
+        vpunpckhqdq zmm15, zmm18, zmm19
+        vmovdqa32 zmm27, zmmword ptr [INDEX0+rip]
+        vmovdqa32 zmm31, zmmword ptr [INDEX1+rip]
+        vshufps zmm16, zmm8, zmm10, 136
+        vshufps zmm17, zmm12, zmm14, 136
+        vmovdqa32 zmm20, zmm16
+        vpermt2d zmm16, zmm27, zmm17
+        vpermt2d zmm20, zmm31, zmm17
+        vshufps zmm17, zmm8, zmm10, 221
+        vshufps zmm30, zmm12, zmm14, 221
+        vmovdqa32 zmm21, zmm17
+        vpermt2d zmm17, zmm27, zmm30
+        vpermt2d zmm21, zmm31, zmm30
+        vshufps zmm18, zmm9, zmm11, 136
+        vshufps zmm8, zmm13, zmm15, 136
+        vmovdqa32 zmm22, zmm18
+        vpermt2d zmm18, zmm27, zmm8
+        vpermt2d zmm22, zmm31, zmm8
+        vshufps zmm19, zmm9, zmm11, 221
+        vshufps zmm8, zmm13, zmm15, 221
+        vmovdqa32 zmm23, zmm19
+        vpermt2d zmm19, zmm27, zmm8
+        vpermt2d zmm23, zmm31, zmm8
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x40]
+        mov     r13, qword ptr [rdi+0x48]
+        mov     r14, qword ptr [rdi+0x50]
+        mov     r15, qword ptr [rdi+0x58]
+        vmovdqu32 ymm24, ymmword ptr [r8+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r12+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r9+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r13+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm8, zmm24, zmm25
+        vpunpckhqdq zmm9, zmm24, zmm25
+        vmovdqu32 ymm24, ymmword ptr [r10+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r14+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r11+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r15+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm10, zmm24, zmm25
+        vpunpckhqdq zmm11, zmm24, zmm25
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r12+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r13+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r14+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        prefetcht0 [r15+rdx+0x80]
+        mov     r8, qword ptr [rdi+0x20]
+        mov     r9, qword ptr [rdi+0x28]
+        mov     r10, qword ptr [rdi+0x30]
+        mov     r11, qword ptr [rdi+0x38]
+        mov     r12, qword ptr [rdi+0x60]
+        mov     r13, qword ptr [rdi+0x68]
+        mov     r14, qword ptr [rdi+0x70]
+        mov     r15, qword ptr [rdi+0x78]
+        vmovdqu32 ymm24, ymmword ptr [r8+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r12+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r9+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r13+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm12, zmm24, zmm25
+        vpunpckhqdq zmm13, zmm24, zmm25
+        vmovdqu32 ymm24, ymmword ptr [r10+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r14+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r11+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r15+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm14, zmm24, zmm25
+        vpunpckhqdq zmm15, zmm24, zmm25
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r12+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r13+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r14+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        prefetcht0 [r15+rdx+0x80]
+        vshufps zmm24, zmm8, zmm10, 136
+        vshufps zmm30, zmm12, zmm14, 136
+        vmovdqa32 zmm28, zmm24
+        vpermt2d zmm24, zmm27, zmm30
+        vpermt2d zmm28, zmm31, zmm30
+        vshufps zmm25, zmm8, zmm10, 221
+        vshufps zmm30, zmm12, zmm14, 221
+        vmovdqa32 zmm29, zmm25
+        vpermt2d zmm25, zmm27, zmm30
+        vpermt2d zmm29, zmm31, zmm30
+        vshufps zmm26, zmm9, zmm11, 136
+        vshufps zmm8, zmm13, zmm15, 136
+        vmovdqa32 zmm30, zmm26
+        vpermt2d zmm26, zmm27, zmm8
+        vpermt2d zmm30, zmm31, zmm8
+        vshufps zmm8, zmm9, zmm11, 221
+        vshufps zmm10, zmm13, zmm15, 221
+        vpermi2d zmm27, zmm8, zmm10
+        vpermi2d zmm31, zmm8, zmm10
+        vpbroadcastd zmm8, dword ptr [BLAKE3_IV_0+rip]
+        vpbroadcastd zmm9, dword ptr [BLAKE3_IV_1+rip]
+        vpbroadcastd zmm10, dword ptr [BLAKE3_IV_2+rip]
+        vpbroadcastd zmm11, dword ptr [BLAKE3_IV_3+rip]
+        vmovdqa32 zmm12, zmmword ptr [rsp]
+        vmovdqa32 zmm13, zmmword ptr [rsp+0x1*0x40]
+        vpbroadcastd zmm14, dword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpbroadcastd zmm15, dword ptr [rsp+0x22*0x4]
+        vpaddd  zmm0, zmm0, zmm16
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm20
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm17
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm21
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm24
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm28
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm25
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm29
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm18
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm23
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm22
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm16
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm17
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm25
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm27
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm30
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm19
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm29
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm20
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm18
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm22
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm27
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm21
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm31
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm26
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm30
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm23
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm19
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm20
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm21
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm16
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm24
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm28
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm31
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm29
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm26
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm23
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm16
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm18
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm17
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm25
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm24
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm30
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm28
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm29
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm18
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm19
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm22
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm27
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm17
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm31
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm25
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm30
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm19
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm26
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm20
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpxord  zmm0, zmm0, zmm8
+        vpxord  zmm1, zmm1, zmm9
+        vpxord  zmm2, zmm2, zmm10
+        vpxord  zmm3, zmm3, zmm11
+        vpxord  zmm4, zmm4, zmm12
+        vpxord  zmm5, zmm5, zmm13
+        vpxord  zmm6, zmm6, zmm14
+        vpxord  zmm7, zmm7, zmm15
+        movzx   eax, byte ptr [rbp+0x38]
+        jne     9b
+        mov     rbx, qword ptr [rbp+0x50]
+        vpunpckldq zmm16, zmm0, zmm1
+        vpunpckhdq zmm17, zmm0, zmm1
+        vpunpckldq zmm18, zmm2, zmm3
+        vpunpckhdq zmm19, zmm2, zmm3
+        vpunpckldq zmm20, zmm4, zmm5
+        vpunpckhdq zmm21, zmm4, zmm5
+        vpunpckldq zmm22, zmm6, zmm7
+        vpunpckhdq zmm23, zmm6, zmm7
+        vpunpcklqdq zmm0, zmm16, zmm18
+        vpunpckhqdq zmm1, zmm16, zmm18
+        vpunpcklqdq zmm2, zmm17, zmm19
+        vpunpckhqdq zmm3, zmm17, zmm19
+        vpunpcklqdq zmm4, zmm20, zmm22
+        vpunpckhqdq zmm5, zmm20, zmm22
+        vpunpcklqdq zmm6, zmm21, zmm23
+        vpunpckhqdq zmm7, zmm21, zmm23
+        vshufi32x4 zmm16, zmm0, zmm4, 0x88
+        vshufi32x4 zmm17, zmm1, zmm5, 0x88
+        vshufi32x4 zmm18, zmm2, zmm6, 0x88
+        vshufi32x4 zmm19, zmm3, zmm7, 0x88
+        vshufi32x4 zmm20, zmm0, zmm4, 0xDD
+        vshufi32x4 zmm21, zmm1, zmm5, 0xDD
+        vshufi32x4 zmm22, zmm2, zmm6, 0xDD
+        vshufi32x4 zmm23, zmm3, zmm7, 0xDD
+        vshufi32x4 zmm0, zmm16, zmm17, 0x88
+        vshufi32x4 zmm1, zmm18, zmm19, 0x88
+        vshufi32x4 zmm2, zmm20, zmm21, 0x88
+        vshufi32x4 zmm3, zmm22, zmm23, 0x88
+        vshufi32x4 zmm4, zmm16, zmm17, 0xDD
+        vshufi32x4 zmm5, zmm18, zmm19, 0xDD
+        vshufi32x4 zmm6, zmm20, zmm21, 0xDD
+        vshufi32x4 zmm7, zmm22, zmm23, 0xDD
+        vmovdqu32 zmmword ptr [rbx], zmm0
+        vmovdqu32 zmmword ptr [rbx+0x1*0x40], zmm1
+        vmovdqu32 zmmword ptr [rbx+0x2*0x40], zmm2
+        vmovdqu32 zmmword ptr [rbx+0x3*0x40], zmm3
+        vmovdqu32 zmmword ptr [rbx+0x4*0x40], zmm4
+        vmovdqu32 zmmword ptr [rbx+0x5*0x40], zmm5
+        vmovdqu32 zmmword ptr [rbx+0x6*0x40], zmm6
+        vmovdqu32 zmmword ptr [rbx+0x7*0x40], zmm7
+        vmovdqa32 zmm0, zmmword ptr [rsp]
+        vmovdqa32 zmm1, zmmword ptr [rsp+0x1*0x40]
+        vmovdqa32 zmm2, zmm0
+        vpaddd  zmm2{k1}, zmm0, dword ptr [ADD16+rip] {1to16}
+        vpcmpltud k2, zmm2, zmm0
+        vpaddd  zmm1 {k2}, zmm1, dword ptr [ADD1+rip] {1to16}
+        vmovdqa32 zmmword ptr [rsp], zmm2
+        vmovdqa32 zmmword ptr [rsp+0x1*0x40], zmm1
+        add     rdi, 128
+        add     rbx, 512
+        mov     qword ptr [rbp+0x50], rbx
+        sub     rsi, 16
+        cmp     rsi, 16
+        jnc     2b
+        test    rsi, rsi
+        jnz     3f
+4:
+        vzeroupper
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align 6
+3:
+        test    esi, 0x8
+        je      3f
+        vpbroadcastd ymm0, dword ptr [rcx]
+        vpbroadcastd ymm1, dword ptr [rcx+0x4]
+        vpbroadcastd ymm2, dword ptr [rcx+0x8]
+        vpbroadcastd ymm3, dword ptr [rcx+0xC]
+        vpbroadcastd ymm4, dword ptr [rcx+0x10]
+        vpbroadcastd ymm5, dword ptr [rcx+0x14]
+        vpbroadcastd ymm6, dword ptr [rcx+0x18]
+        vpbroadcastd ymm7, dword ptr [rcx+0x1C]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x20]
+        mov     r13, qword ptr [rdi+0x28]
+        mov     r14, qword ptr [rdi+0x30]
+        mov     r15, qword ptr [rdi+0x38]
+        movzx   eax, byte ptr [rbp+0x38]
+        movzx   ebx, byte ptr [rbp+0x40]
+        or      eax, ebx
+        xor     edx, edx
+2:
+        movzx   ebx, byte ptr [rbp+0x48]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+0x80]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+0x88], eax
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x40], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x40]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x40], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x40]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x40], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x40]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x40], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm16, ymm12, ymm14, 136
+        vshufps ymm17, ymm12, ymm14, 221
+        vshufps ymm18, ymm13, ymm15, 136
+        vshufps ymm19, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x30], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x30], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x30]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x30], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x30]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x30], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm20, ymm12, ymm14, 136
+        vshufps ymm21, ymm12, ymm14, 221
+        vshufps ymm22, ymm13, ymm15, 136
+        vshufps ymm23, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x20], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x20]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x20], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x20]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x20], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x20]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x20], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm24, ymm12, ymm14, 136
+        vshufps ymm25, ymm12, ymm14, 221
+        vshufps ymm26, ymm13, ymm15, 136
+        vshufps ymm27, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x10], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x10], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x10]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x10], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x10]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x10], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm28, ymm12, ymm14, 136
+        vshufps ymm29, ymm12, ymm14, 221
+        vshufps ymm30, ymm13, ymm15, 136
+        vshufps ymm31, ymm13, ymm15, 221
+        vpbroadcastd ymm8, dword ptr [BLAKE3_IV_0+rip]
+        vpbroadcastd ymm9, dword ptr [BLAKE3_IV_1+rip]
+        vpbroadcastd ymm10, dword ptr [BLAKE3_IV_2+rip]
+        vpbroadcastd ymm11, dword ptr [BLAKE3_IV_3+rip]
+        vmovdqa ymm12, ymmword ptr [rsp]
+        vmovdqa ymm13, ymmword ptr [rsp+0x40]
+        vpbroadcastd ymm14, dword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpbroadcastd ymm15, dword ptr [rsp+0x88]
+        vpaddd  ymm0, ymm0, ymm16
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm20
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm17
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm21
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm24
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm28
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm25
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm29
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm18
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm23
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm22
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm16
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm17
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm25
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm27
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm30
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm19
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm29
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm20
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm18
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm22
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm27
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm21
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm31
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm26
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm30
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm23
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm19
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm20
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm21
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm16
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm24
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm28
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm31
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm29
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm26
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm23
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm16
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm18
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm17
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm25
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm24
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm30
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm28
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm29
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm18
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm19
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm22
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm27
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm17
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm31
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm25
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm30
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm19
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm26
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm20
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpxor   ymm0, ymm0, ymm8
+        vpxor   ymm1, ymm1, ymm9
+        vpxor   ymm2, ymm2, ymm10
+        vpxor   ymm3, ymm3, ymm11
+        vpxor   ymm4, ymm4, ymm12
+        vpxor   ymm5, ymm5, ymm13
+        vpxor   ymm6, ymm6, ymm14
+        vpxor   ymm7, ymm7, ymm15
+        movzx   eax, byte ptr [rbp+0x38]
+        jne     2b
+        mov     rbx, qword ptr [rbp+0x50]
+        vunpcklps ymm8, ymm0, ymm1
+        vunpcklps ymm9, ymm2, ymm3
+        vunpckhps ymm10, ymm0, ymm1
+        vunpcklps ymm11, ymm4, ymm5
+        vunpcklps ymm0, ymm6, ymm7
+        vshufps ymm12, ymm8, ymm9, 78
+        vblendps ymm1, ymm8, ymm12, 0xCC
+        vshufps ymm8, ymm11, ymm0, 78
+        vunpckhps ymm13, ymm2, ymm3
+        vblendps ymm2, ymm11, ymm8, 0xCC
+        vblendps ymm3, ymm12, ymm9, 0xCC
+        vperm2f128 ymm12, ymm1, ymm2, 0x20
+        vmovups ymmword ptr [rbx], ymm12
+        vunpckhps ymm14, ymm4, ymm5
+        vblendps ymm4, ymm8, ymm0, 0xCC
+        vunpckhps ymm15, ymm6, ymm7
+        vperm2f128 ymm7, ymm3, ymm4, 0x20
+        vmovups ymmword ptr [rbx+0x20], ymm7
+        vshufps ymm5, ymm10, ymm13, 78
+        vblendps ymm6, ymm5, ymm13, 0xCC
+        vshufps ymm13, ymm14, ymm15, 78
+        vblendps ymm10, ymm10, ymm5, 0xCC
+        vblendps ymm14, ymm14, ymm13, 0xCC
+        vperm2f128 ymm8, ymm10, ymm14, 0x20
+        vmovups ymmword ptr [rbx+0x40], ymm8
+        vblendps ymm15, ymm13, ymm15, 0xCC
+        vperm2f128 ymm13, ymm6, ymm15, 0x20
+        vmovups ymmword ptr [rbx+0x60], ymm13
+        vperm2f128 ymm9, ymm1, ymm2, 0x31
+        vperm2f128 ymm11, ymm3, ymm4, 0x31
+        vmovups ymmword ptr [rbx+0x80], ymm9
+        vperm2f128 ymm14, ymm10, ymm14, 0x31
+        vperm2f128 ymm15, ymm6, ymm15, 0x31
+        vmovups ymmword ptr [rbx+0xA0], ymm11
+        vmovups ymmword ptr [rbx+0xC0], ymm14
+        vmovups ymmword ptr [rbx+0xE0], ymm15
+        vmovdqa ymm0, ymmword ptr [rsp]
+        vmovdqa ymm2, ymmword ptr [rsp+0x2*0x20]
+        vmovdqa32 ymm0 {k1}, ymmword ptr [rsp+0x1*0x20]
+        vmovdqa32 ymm2 {k1}, ymmword ptr [rsp+0x3*0x20]
+        vmovdqa ymmword ptr [rsp], ymm0
+        vmovdqa ymmword ptr [rsp+0x2*0x20], ymm2
+        add     rbx, 256
+        mov     qword ptr [rbp+0x50], rbx
+        add     rdi, 64
+        sub     rsi, 8
+3:
+        mov     rbx, qword ptr [rbp+0x50]
+        mov     r15, qword ptr [rsp+0x80]
+        movzx   r13, byte ptr [rbp+0x38]
+        movzx   r12, byte ptr [rbp+0x48]
+        test    esi, 0x4
+        je      3f
+        vbroadcasti32x4 zmm0, xmmword ptr [rcx]
+        vbroadcasti32x4 zmm1, xmmword ptr [rcx+0x1*0x10]
+        vmovdqa xmm12, xmmword ptr [rsp]
+        vmovdqa xmm13, xmmword ptr [rsp+0x4*0x10]
+        vpunpckldq xmm14, xmm12, xmm13
+        vpunpckhdq xmm15, xmm12, xmm13
+        vpermq  ymm14, ymm14, 0xDC
+        vpermq  ymm15, ymm15, 0xDC
+        vpbroadcastd zmm12, dword ptr [BLAKE3_BLOCK_LEN+rip]
+        vinserti64x4 zmm13, zmm14, ymm15, 0x01
+        mov     eax, 17476
+        kmovw   k2, eax
+        vpblendmd zmm13 {k2}, zmm13, zmm12
+        vbroadcasti32x4 zmm15, xmmword ptr [BLAKE3_IV+rip]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     eax, 43690
+        kmovw   k3, eax
+        mov     eax, 34952
+        kmovw   k4, eax
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x88], eax
+        vmovdqa32 zmm2, zmm15
+        vpbroadcastd zmm8, dword ptr [rsp+0x22*0x4]
+        vpblendmd zmm3 {k4}, zmm13, zmm8
+        vmovups zmm8, zmmword ptr [r8+rdx-0x1*0x40]
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r9+rdx-0x4*0x10], 0x01
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r10+rdx-0x4*0x10], 0x02
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r11+rdx-0x4*0x10], 0x03
+        vmovups zmm9, zmmword ptr [r8+rdx-0x30]
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r9+rdx-0x3*0x10], 0x01
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r10+rdx-0x3*0x10], 0x02
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r11+rdx-0x3*0x10], 0x03
+        vshufps zmm4, zmm8, zmm9, 136
+        vshufps zmm5, zmm8, zmm9, 221
+        vmovups zmm8, zmmword ptr [r8+rdx-0x20]
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r9+rdx-0x2*0x10], 0x01
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r10+rdx-0x2*0x10], 0x02
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r11+rdx-0x2*0x10], 0x03
+        vmovups zmm9, zmmword ptr [r8+rdx-0x10]
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r9+rdx-0x1*0x10], 0x01
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r10+rdx-0x1*0x10], 0x02
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r11+rdx-0x1*0x10], 0x03
+        vshufps zmm6, zmm8, zmm9, 136
+        vshufps zmm7, zmm8, zmm9, 221
+        vpshufd zmm6, zmm6, 0x93
+        vpshufd zmm7, zmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 16
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 12
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 8
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 7
+        vpshufd zmm0, zmm0, 0x93
+        vpshufd zmm3, zmm3, 0x4E
+        vpshufd zmm2, zmm2, 0x39
+        vpaddd  zmm0, zmm0, zmm6
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 16
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 12
+        vpaddd  zmm0, zmm0, zmm7
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 8
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 7
+        vpshufd zmm0, zmm0, 0x39
+        vpshufd zmm3, zmm3, 0x4E
+        vpshufd zmm2, zmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps zmm8, zmm4, zmm5, 214
+        vpshufd zmm9, zmm4, 0x0F
+        vpshufd zmm4, zmm8, 0x39
+        vshufps zmm8, zmm6, zmm7, 250
+        vpblendmd zmm9 {k3}, zmm9, zmm8
+        vpunpcklqdq zmm8, zmm7, zmm5
+        vpblendmd zmm8 {k4}, zmm8, zmm6
+        vpshufd zmm8, zmm8, 0x78
+        vpunpckhdq zmm5, zmm5, zmm7
+        vpunpckldq zmm6, zmm6, zmm5
+        vpshufd zmm7, zmm6, 0x1E
+        vmovdqa32 zmm5, zmm9
+        vmovdqa32 zmm6, zmm8
+        jmp     9b
+9:
+        vpxord  zmm0, zmm0, zmm2
+        vpxord  zmm1, zmm1, zmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vextracti32x4 xmmword ptr [rbx+0x4*0x10], zmm0, 0x02
+        vextracti32x4 xmmword ptr [rbx+0x5*0x10], zmm1, 0x02
+        vextracti32x4 xmmword ptr [rbx+0x6*0x10], zmm0, 0x03
+        vextracti32x4 xmmword ptr [rbx+0x7*0x10], zmm1, 0x03
+        vmovdqa xmm0, xmmword ptr [rsp]
+        vmovdqa xmm2, xmmword ptr [rsp+0x40]
+        vmovdqa32 xmm0 {k1}, xmmword ptr [rsp+0x1*0x10]
+        vmovdqa32 xmm2 {k1}, xmmword ptr [rsp+0x5*0x10]
+        vmovdqa xmmword ptr [rsp], xmm0
+        vmovdqa xmmword ptr [rsp+0x40], xmm2
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+3:
+        test    esi, 0x2
+        je      3f
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm13, dword ptr [rsp]
+        vpinsrd xmm13, xmm13, dword ptr [rsp+0x40], 1
+        vpinsrd xmm13, xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovd   xmm14, dword ptr [rsp+0x4]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+0x44], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vinserti128 ymm13, ymm13, xmm14, 0x01
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x88], eax
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
+        vpbroadcastd ymm8, dword ptr [rsp+0x88]
+        vpblendd ymm3, ymm13, ymm8, 0x88
+        vmovups ymm8, ymmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x40], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x30], 0x01
+        vshufps ymm4, ymm8, ymm9, 136
+        vshufps ymm5, ymm8, ymm9, 221
+        vmovups ymm8, ymmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x20], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x10], 0x01
+        vshufps ymm6, ymm8, ymm9, 136
+        vshufps ymm7, ymm8, ymm9, 221
+        vpshufd ymm6, ymm6, 0x93
+        vpshufd ymm7, ymm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 16
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 12
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 8
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 7
+        vpshufd ymm0, ymm0, 0x93
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x39
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 16
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 12
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 8
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 7
+        vpshufd ymm0, ymm0, 0x39
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x93
+        dec     al
+        jz      9f
+        vshufps ymm8, ymm4, ymm5, 214
+        vpshufd ymm9, ymm4, 0x0F
+        vpshufd ymm4, ymm8, 0x39
+        vshufps ymm8, ymm6, ymm7, 250
+        vpblendd ymm9, ymm9, ymm8, 0xAA
+        vpunpcklqdq ymm8, ymm7, ymm5
+        vpblendd ymm8, ymm8, ymm6, 0x88
+        vpshufd ymm8, ymm8, 0x78
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 0x1E
+        vmovdqa ymm5, ymm9
+        vmovdqa ymm6, ymm8
+        jmp     9b
+9:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vmovdqa xmm0, xmmword ptr [rsp]
+        vmovdqa xmm2, xmmword ptr [rsp+0x4*0x10]
+        vmovdqu32 xmm0 {k1}, xmmword ptr [rsp+0x8]
+        vmovdqu32 xmm2 {k1}, xmmword ptr [rsp+0x48]
+        vmovdqa xmmword ptr [rsp], xmm0
+        vmovdqa xmmword ptr [rsp+0x4*0x10], xmm2
+        add     rbx, 64
+        add     rdi, 16
+        sub     rsi, 2
+3:
+        test    esi, 0x1
+        je      4b
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm14, dword ptr [rsp]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+0x40], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovdqa xmm15, xmmword ptr [BLAKE3_IV+rip]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        vpinsrd xmm3, xmm14, eax, 3
+        vmovdqa xmm2, xmm15
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x30]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x10]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+.p2align 6
+_blake3_compress_in_place_avx512:
+blake3_compress_in_place_avx512:
+        _CET_ENDBR
+        vmovdqu xmm0, xmmword ptr [rdi]
+        vmovdqu xmm1, xmmword ptr [rdi+0x10]
+        movzx   eax, r8b
+        movzx   edx, dl
+        shl     rax, 32
+        add     rdx, rax
+        vmovq   xmm3, rcx
+        vmovq   xmm4, rdx
+        vpunpcklqdq xmm3, xmm3, xmm4
+        vmovaps xmm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovups xmm8, xmmword ptr [rsi]
+        vmovups xmm9, xmmword ptr [rsi+0x10]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [rsi+0x20]
+        vmovups xmm9, xmmword ptr [rsi+0x30]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        vmovdqu xmmword ptr [rdi], xmm0
+        vmovdqu xmmword ptr [rdi+0x10], xmm1
+        ret
+
+.p2align 6
+_blake3_compress_xof_avx512:
+blake3_compress_xof_avx512:
+        _CET_ENDBR
+        vmovdqu xmm0, xmmword ptr [rdi]
+        vmovdqu xmm1, xmmword ptr [rdi+0x10]
+        movzx   eax, r8b
+        movzx   edx, dl
+        shl     rax, 32
+        add     rdx, rax
+        vmovq   xmm3, rcx
+        vmovq   xmm4, rdx
+        vpunpcklqdq xmm3, xmm3, xmm4
+        vmovaps xmm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovups xmm8, xmmword ptr [rsi]
+        vmovups xmm9, xmmword ptr [rsi+0x10]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [rsi+0x20]
+        vmovups xmm9, xmmword ptr [rsi+0x30]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        vpxor   xmm2, xmm2, [rdi]
+        vpxor   xmm3, xmm3, [rdi+0x10]
+        vmovdqu xmmword ptr [r9], xmm0
+        vmovdqu xmmword ptr [r9+0x10], xmm1
+        vmovdqu xmmword ptr [r9+0x20], xmm2
+        vmovdqu xmmword ptr [r9+0x30], xmm3
+        ret
+
+.p2align 6
+blake3_xof_many_avx512:
+_blake3_xof_many_avx512:
+        _CET_ENDBR
+        mov    r10,QWORD PTR [rsp+0x8]
+        cmp    r10,0x1
+        ja     2f
+        vmovdqu xmm0,XMMWORD PTR [rdi]
+        vmovdqu xmm1,XMMWORD PTR [rdi+0x10]
+        movzx  eax,r8b
+        movzx  edx,dl
+        shl    rax,0x20
+        add    rdx,rax
+        vmovq  xmm3,rcx
+        vmovq  xmm4,rdx
+        vpunpcklqdq xmm3,xmm3,xmm4
+        vmovaps xmm2,XMMWORD PTR [BLAKE3_IV+rip]
+        vmovups xmm8,XMMWORD PTR [rsi]
+        vmovups xmm9,XMMWORD PTR [rsi+0x10]
+        vshufps xmm4,xmm8,xmm9,0x88
+        vshufps xmm5,xmm8,xmm9,0xdd
+        vmovups xmm8,XMMWORD PTR [rsi+0x20]
+        vmovups xmm9,XMMWORD PTR [rsi+0x30]
+        vshufps xmm6,xmm8,xmm9,0x88
+        vshufps xmm7,xmm8,xmm9,0xdd
+        vpshufd xmm6,xmm6,0x93
+        vpshufd xmm7,xmm7,0x93
+        mov    al,0x7
+3:
+        vpaddd xmm0,xmm0,xmm4
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x10
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0xc
+        vpaddd xmm0,xmm0,xmm5
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x8
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0x7
+        vpshufd xmm0,xmm0,0x93
+        vpshufd xmm3,xmm3,0x4e
+        vpshufd xmm2,xmm2,0x39
+        vpaddd xmm0,xmm0,xmm6
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x10
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0xc
+        vpaddd xmm0,xmm0,xmm7
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x8
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0x7
+        vpshufd xmm0,xmm0,0x39
+        vpshufd xmm3,xmm3,0x4e
+        vpshufd xmm2,xmm2,0x93
+        dec    al
+        je     3f
+        vshufps xmm8,xmm4,xmm5,0xd6
+        vpshufd xmm9,xmm4,0xf
+        vpshufd xmm4,xmm8,0x39
+        vshufps xmm8,xmm6,xmm7,0xfa
+        vpblendd xmm9,xmm9,xmm8,0xaa
+        vpunpcklqdq xmm8,xmm7,xmm5
+        vpblendd xmm8,xmm8,xmm6,0x88
+        vpshufd xmm8,xmm8,0x78
+        vpunpckhdq xmm5,xmm5,xmm7
+        vpunpckldq xmm6,xmm6,xmm5
+        vpshufd xmm7,xmm6,0x1e
+        vmovdqa xmm5,xmm9
+        vmovdqa xmm6,xmm8
+        jmp    3b
+3:
+        vpxor  xmm0,xmm0,xmm2
+        vpxor  xmm1,xmm1,xmm3
+        vpxor  xmm2,xmm2,XMMWORD PTR [rdi]
+        vpxor  xmm3,xmm3,XMMWORD PTR [rdi+0x10]
+        vmovdqu XMMWORD PTR [r9],xmm0
+        vmovdqu XMMWORD PTR [r9+0x10],xmm1
+        vmovdqu XMMWORD PTR [r9+0x20],xmm2
+        vmovdqu XMMWORD PTR [r9+0x30],xmm3
+        ret    
+.p2align 6
+2:
+        push   rbp
+        mov    rbp,rsp
+        sub    rsp,0x90
+        and    rsp,0xffffffffffffffc0
+        vpbroadcastd zmm0,ecx
+        shr    rcx,0x20
+        vpbroadcastd zmm1,ecx
+        vpaddd zmm2,zmm0,ZMMWORD PTR [ADD0+rip]
+        vpcmpltud k1,zmm2,zmm0
+        vpaddd zmm1{k1},zmm1,DWORD PTR [ADD1+rip]{1to16}
+        vmovdqa32 ZMMWORD PTR [rsp],zmm2
+        vmovdqa32 ZMMWORD PTR [rsp+0x40],zmm1
+        cmp    r10,0x10
+        jb     2f
+3:
+        vpbroadcastd zmm16,DWORD PTR [rsi]
+        vpbroadcastd zmm17,DWORD PTR [rsi+0x4]
+        vpbroadcastd zmm18,DWORD PTR [rsi+0x8]
+        vpbroadcastd zmm19,DWORD PTR [rsi+0xc]
+        vpbroadcastd zmm20,DWORD PTR [rsi+0x10]
+        vpbroadcastd zmm21,DWORD PTR [rsi+0x14]
+        vpbroadcastd zmm22,DWORD PTR [rsi+0x18]
+        vpbroadcastd zmm23,DWORD PTR [rsi+0x1c]
+        vpbroadcastd zmm24,DWORD PTR [rsi+0x20]
+        vpbroadcastd zmm25,DWORD PTR [rsi+0x24]
+        vpbroadcastd zmm26,DWORD PTR [rsi+0x28]
+        vpbroadcastd zmm27,DWORD PTR [rsi+0x2c]
+        vpbroadcastd zmm28,DWORD PTR [rsi+0x30]
+        vpbroadcastd zmm29,DWORD PTR [rsi+0x34]
+        vpbroadcastd zmm30,DWORD PTR [rsi+0x38]
+        vpbroadcastd zmm31,DWORD PTR [rsi+0x3c]
+        vpbroadcastd zmm0,DWORD PTR [rdi]
+        vpbroadcastd zmm1,DWORD PTR [rdi+0x4]
+        vpbroadcastd zmm2,DWORD PTR [rdi+0x8]
+        vpbroadcastd zmm3,DWORD PTR [rdi+0xc]
+        vpbroadcastd zmm4,DWORD PTR [rdi+0x10]
+        vpbroadcastd zmm5,DWORD PTR [rdi+0x14]
+        vpbroadcastd zmm6,DWORD PTR [rdi+0x18]
+        vpbroadcastd zmm7,DWORD PTR [rdi+0x1c]
+        vpbroadcastd zmm8,DWORD PTR [BLAKE3_IV_0+rip]
+        vpbroadcastd zmm9,DWORD PTR [BLAKE3_IV_1+rip]
+        vpbroadcastd zmm10,DWORD PTR [BLAKE3_IV_2+rip]
+        vpbroadcastd zmm11,DWORD PTR [BLAKE3_IV_3+rip]
+        vmovdqa32 zmm12,ZMMWORD PTR [rsp]
+        vmovdqa32 zmm13,ZMMWORD PTR [rsp+0x40]
+        vpbroadcastd zmm14,edx
+        vpbroadcastd zmm15,r8d
+        vpaddd zmm0,zmm0,zmm16
+        vpaddd zmm1,zmm1,zmm18
+        vpaddd zmm2,zmm2,zmm20
+        vpaddd zmm3,zmm3,zmm22
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vprord zmm15,zmm15,0x10
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0xc
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vpaddd zmm0,zmm0,zmm17
+        vpaddd zmm1,zmm1,zmm19
+        vpaddd zmm2,zmm2,zmm21
+        vpaddd zmm3,zmm3,zmm23
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vprord zmm15,zmm15,0x8
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0x7
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vpaddd zmm0,zmm0,zmm24
+        vpaddd zmm1,zmm1,zmm26
+        vpaddd zmm2,zmm2,zmm28
+        vpaddd zmm3,zmm3,zmm30
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x10
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vprord zmm4,zmm4,0xc
+        vpaddd zmm0,zmm0,zmm25
+        vpaddd zmm1,zmm1,zmm27
+        vpaddd zmm2,zmm2,zmm29
+        vpaddd zmm3,zmm3,zmm31
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x8
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vprord zmm4,zmm4,0x7
+        vpaddd zmm0,zmm0,zmm18
+        vpaddd zmm1,zmm1,zmm19
+        vpaddd zmm2,zmm2,zmm23
+        vpaddd zmm3,zmm3,zmm20
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vprord zmm15,zmm15,0x10
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0xc
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vpaddd zmm0,zmm0,zmm22
+        vpaddd zmm1,zmm1,zmm26
+        vpaddd zmm2,zmm2,zmm16
+        vpaddd zmm3,zmm3,zmm29
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vprord zmm15,zmm15,0x8
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0x7
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vpaddd zmm0,zmm0,zmm17
+        vpaddd zmm1,zmm1,zmm28
+        vpaddd zmm2,zmm2,zmm25
+        vpaddd zmm3,zmm3,zmm31
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x10
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vprord zmm4,zmm4,0xc
+        vpaddd zmm0,zmm0,zmm27
+        vpaddd zmm1,zmm1,zmm21
+        vpaddd zmm2,zmm2,zmm30
+        vpaddd zmm3,zmm3,zmm24
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x8
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vprord zmm4,zmm4,0x7
+        vpaddd zmm0,zmm0,zmm19
+        vpaddd zmm1,zmm1,zmm26
+        vpaddd zmm2,zmm2,zmm29
+        vpaddd zmm3,zmm3,zmm23
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vprord zmm15,zmm15,0x10
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0xc
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vpaddd zmm0,zmm0,zmm20
+        vpaddd zmm1,zmm1,zmm28
+        vpaddd zmm2,zmm2,zmm18
+        vpaddd zmm3,zmm3,zmm30
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vprord zmm15,zmm15,0x8
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0x7
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vpaddd zmm0,zmm0,zmm22
+        vpaddd zmm1,zmm1,zmm25
+        vpaddd zmm2,zmm2,zmm27
+        vpaddd zmm3,zmm3,zmm24
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x10
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vprord zmm4,zmm4,0xc
+        vpaddd zmm0,zmm0,zmm21
+        vpaddd zmm1,zmm1,zmm16
+        vpaddd zmm2,zmm2,zmm31
+        vpaddd zmm3,zmm3,zmm17
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x8
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vprord zmm4,zmm4,0x7
+        vpaddd zmm0,zmm0,zmm26
+        vpaddd zmm1,zmm1,zmm28
+        vpaddd zmm2,zmm2,zmm30
+        vpaddd zmm3,zmm3,zmm29
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vprord zmm15,zmm15,0x10
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0xc
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vpaddd zmm0,zmm0,zmm23
+        vpaddd zmm1,zmm1,zmm25
+        vpaddd zmm2,zmm2,zmm19
+        vpaddd zmm3,zmm3,zmm31
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vprord zmm15,zmm15,0x8
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0x7
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vpaddd zmm0,zmm0,zmm20
+        vpaddd zmm1,zmm1,zmm27
+        vpaddd zmm2,zmm2,zmm21
+        vpaddd zmm3,zmm3,zmm17
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x10
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vprord zmm4,zmm4,0xc
+        vpaddd zmm0,zmm0,zmm16
+        vpaddd zmm1,zmm1,zmm18
+        vpaddd zmm2,zmm2,zmm24
+        vpaddd zmm3,zmm3,zmm22
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x8
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vprord zmm4,zmm4,0x7
+        vpaddd zmm0,zmm0,zmm28
+        vpaddd zmm1,zmm1,zmm25
+        vpaddd zmm2,zmm2,zmm31
+        vpaddd zmm3,zmm3,zmm30
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vprord zmm15,zmm15,0x10
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0xc
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vpaddd zmm0,zmm0,zmm29
+        vpaddd zmm1,zmm1,zmm27
+        vpaddd zmm2,zmm2,zmm26
+        vpaddd zmm3,zmm3,zmm24
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vprord zmm15,zmm15,0x8
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0x7
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vpaddd zmm0,zmm0,zmm23
+        vpaddd zmm1,zmm1,zmm21
+        vpaddd zmm2,zmm2,zmm16
+        vpaddd zmm3,zmm3,zmm22
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x10
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vprord zmm4,zmm4,0xc
+        vpaddd zmm0,zmm0,zmm18
+        vpaddd zmm1,zmm1,zmm19
+        vpaddd zmm2,zmm2,zmm17
+        vpaddd zmm3,zmm3,zmm20
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x8
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vprord zmm4,zmm4,0x7
+        vpaddd zmm0,zmm0,zmm25
+        vpaddd zmm1,zmm1,zmm27
+        vpaddd zmm2,zmm2,zmm24
+        vpaddd zmm3,zmm3,zmm31
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vprord zmm15,zmm15,0x10
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0xc
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vpaddd zmm0,zmm0,zmm30
+        vpaddd zmm1,zmm1,zmm21
+        vpaddd zmm2,zmm2,zmm28
+        vpaddd zmm3,zmm3,zmm17
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vprord zmm15,zmm15,0x8
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0x7
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vpaddd zmm0,zmm0,zmm29
+        vpaddd zmm1,zmm1,zmm16
+        vpaddd zmm2,zmm2,zmm18
+        vpaddd zmm3,zmm3,zmm20
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x10
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vprord zmm4,zmm4,0xc
+        vpaddd zmm0,zmm0,zmm19
+        vpaddd zmm1,zmm1,zmm26
+        vpaddd zmm2,zmm2,zmm22
+        vpaddd zmm3,zmm3,zmm23
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x8
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vprord zmm4,zmm4,0x7
+        vpaddd zmm0,zmm0,zmm27
+        vpaddd zmm1,zmm1,zmm21
+        vpaddd zmm2,zmm2,zmm17
+        vpaddd zmm3,zmm3,zmm24
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vprord zmm15,zmm15,0x10
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0xc
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vpaddd zmm0,zmm0,zmm31
+        vpaddd zmm1,zmm1,zmm16
+        vpaddd zmm2,zmm2,zmm25
+        vpaddd zmm3,zmm3,zmm22
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm1,zmm1,zmm5
+        vpaddd zmm2,zmm2,zmm6
+        vpaddd zmm3,zmm3,zmm7
+        vpxord zmm12,zmm12,zmm0
+        vpxord zmm13,zmm13,zmm1
+        vpxord zmm14,zmm14,zmm2
+        vpxord zmm15,zmm15,zmm3
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vprord zmm15,zmm15,0x8
+        vpaddd zmm8,zmm8,zmm12
+        vpaddd zmm9,zmm9,zmm13
+        vpaddd zmm10,zmm10,zmm14
+        vpaddd zmm11,zmm11,zmm15
+        vpxord zmm4,zmm4,zmm8
+        vpxord zmm5,zmm5,zmm9
+        vpxord zmm6,zmm6,zmm10
+        vpxord zmm7,zmm7,zmm11
+        vprord zmm4,zmm4,0x7
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vpaddd zmm0,zmm0,zmm30
+        vpaddd zmm1,zmm1,zmm18
+        vpaddd zmm2,zmm2,zmm19
+        vpaddd zmm3,zmm3,zmm23
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x10
+        vprord zmm12,zmm12,0x10
+        vprord zmm13,zmm13,0x10
+        vprord zmm14,zmm14,0x10
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0xc
+        vprord zmm6,zmm6,0xc
+        vprord zmm7,zmm7,0xc
+        vprord zmm4,zmm4,0xc
+        vpaddd zmm0,zmm0,zmm26
+        vpaddd zmm1,zmm1,zmm28
+        vpaddd zmm2,zmm2,zmm20
+        vpaddd zmm3,zmm3,zmm29
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm1,zmm1,zmm6
+        vpaddd zmm2,zmm2,zmm7
+        vpaddd zmm3,zmm3,zmm4
+        vpxord zmm15,zmm15,zmm0
+        vpxord zmm12,zmm12,zmm1
+        vpxord zmm13,zmm13,zmm2
+        vpxord zmm14,zmm14,zmm3
+        vprord zmm15,zmm15,0x8
+        vprord zmm12,zmm12,0x8
+        vprord zmm13,zmm13,0x8
+        vprord zmm14,zmm14,0x8
+        vpaddd zmm10,zmm10,zmm15
+        vpaddd zmm11,zmm11,zmm12
+        vpaddd zmm8,zmm8,zmm13
+        vpaddd zmm9,zmm9,zmm14
+        vpxord zmm5,zmm5,zmm10
+        vpxord zmm6,zmm6,zmm11
+        vpxord zmm7,zmm7,zmm8
+        vpxord zmm4,zmm4,zmm9
+        vprord zmm5,zmm5,0x7
+        vprord zmm6,zmm6,0x7
+        vprord zmm7,zmm7,0x7
+        vprord zmm4,zmm4,0x7
+        vpxord zmm0,zmm0,zmm8
+        vpxord zmm1,zmm1,zmm9
+        vpxord zmm2,zmm2,zmm10
+        vpxord zmm3,zmm3,zmm11
+        vpxord zmm4,zmm4,zmm12
+        vpxord zmm5,zmm5,zmm13
+        vpxord zmm6,zmm6,zmm14
+        vpxord zmm7,zmm7,zmm15
+        vpxord zmm8,zmm8,DWORD PTR [rdi]{1to16}
+        vpxord zmm9,zmm9,DWORD PTR [rdi+0x4]{1to16}
+        vpxord zmm10,zmm10,DWORD PTR [rdi+0x8]{1to16}
+        vpxord zmm11,zmm11,DWORD PTR [rdi+0xc]{1to16}
+        vpxord zmm12,zmm12,DWORD PTR [rdi+0x10]{1to16}
+        vpxord zmm13,zmm13,DWORD PTR [rdi+0x14]{1to16}
+        vpxord zmm14,zmm14,DWORD PTR [rdi+0x18]{1to16}
+        vpxord zmm15,zmm15,DWORD PTR [rdi+0x1c]{1to16}
+        vpunpckldq zmm16,zmm0,zmm1
+        vpunpckhdq zmm17,zmm0,zmm1
+        vpunpckldq zmm18,zmm2,zmm3
+        vpunpckhdq zmm19,zmm2,zmm3
+        vpunpckldq zmm20,zmm4,zmm5
+        vpunpckhdq zmm21,zmm4,zmm5
+        vpunpckldq zmm22,zmm6,zmm7
+        vpunpckhdq zmm23,zmm6,zmm7
+        vpunpckldq zmm24,zmm8,zmm9
+        vpunpckhdq zmm25,zmm8,zmm9
+        vpunpckldq zmm26,zmm10,zmm11
+        vpunpckhdq zmm27,zmm10,zmm11
+        vpunpckldq zmm28,zmm12,zmm13
+        vpunpckhdq zmm29,zmm12,zmm13
+        vpunpckldq zmm30,zmm14,zmm15
+        vpunpckhdq zmm31,zmm14,zmm15
+        vpunpcklqdq zmm0,zmm16,zmm18
+        vpunpckhqdq zmm1,zmm16,zmm18
+        vpunpcklqdq zmm2,zmm17,zmm19
+        vpunpckhqdq zmm3,zmm17,zmm19
+        vpunpcklqdq zmm4,zmm20,zmm22
+        vpunpckhqdq zmm5,zmm20,zmm22
+        vpunpcklqdq zmm6,zmm21,zmm23
+        vpunpckhqdq zmm7,zmm21,zmm23
+        vpunpcklqdq zmm8,zmm24,zmm26
+        vpunpckhqdq zmm9,zmm24,zmm26
+        vpunpcklqdq zmm10,zmm25,zmm27
+        vpunpckhqdq zmm11,zmm25,zmm27
+        vpunpcklqdq zmm12,zmm28,zmm30
+        vpunpckhqdq zmm13,zmm28,zmm30
+        vpunpcklqdq zmm14,zmm29,zmm31
+        vpunpckhqdq zmm15,zmm29,zmm31
+        vshufi32x4 zmm16,zmm0,zmm4,0x88
+        vshufi32x4 zmm17,zmm1,zmm5,0x88
+        vshufi32x4 zmm18,zmm2,zmm6,0x88
+        vshufi32x4 zmm19,zmm3,zmm7,0x88
+        vshufi32x4 zmm20,zmm0,zmm4,0xdd
+        vshufi32x4 zmm21,zmm1,zmm5,0xdd
+        vshufi32x4 zmm22,zmm2,zmm6,0xdd
+        vshufi32x4 zmm23,zmm3,zmm7,0xdd
+        vshufi32x4 zmm24,zmm8,zmm12,0x88
+        vshufi32x4 zmm25,zmm9,zmm13,0x88
+        vshufi32x4 zmm26,zmm10,zmm14,0x88
+        vshufi32x4 zmm27,zmm11,zmm15,0x88
+        vshufi32x4 zmm28,zmm8,zmm12,0xdd
+        vshufi32x4 zmm29,zmm9,zmm13,0xdd
+        vshufi32x4 zmm30,zmm10,zmm14,0xdd
+        vshufi32x4 zmm31,zmm11,zmm15,0xdd
+        vshufi32x4 zmm0,zmm16,zmm24,0x88
+        vshufi32x4 zmm1,zmm17,zmm25,0x88
+        vshufi32x4 zmm2,zmm18,zmm26,0x88
+        vshufi32x4 zmm3,zmm19,zmm27,0x88
+        vshufi32x4 zmm4,zmm20,zmm28,0x88
+        vshufi32x4 zmm5,zmm21,zmm29,0x88
+        vshufi32x4 zmm6,zmm22,zmm30,0x88
+        vshufi32x4 zmm7,zmm23,zmm31,0x88
+        vshufi32x4 zmm8,zmm16,zmm24,0xdd
+        vshufi32x4 zmm9,zmm17,zmm25,0xdd
+        vshufi32x4 zmm10,zmm18,zmm26,0xdd
+        vshufi32x4 zmm11,zmm19,zmm27,0xdd
+        vshufi32x4 zmm12,zmm20,zmm28,0xdd
+        vshufi32x4 zmm13,zmm21,zmm29,0xdd
+        vshufi32x4 zmm14,zmm22,zmm30,0xdd
+        vshufi32x4 zmm15,zmm23,zmm31,0xdd
+        vmovdqu32 ZMMWORD PTR [r9],zmm0
+        vmovdqu32 ZMMWORD PTR [r9+0x40],zmm1
+        vmovdqu32 ZMMWORD PTR [r9+0x80],zmm2
+        vmovdqu32 ZMMWORD PTR [r9+0xc0],zmm3
+        vmovdqu32 ZMMWORD PTR [r9+0x100],zmm4
+        vmovdqu32 ZMMWORD PTR [r9+0x140],zmm5
+        vmovdqu32 ZMMWORD PTR [r9+0x180],zmm6
+        vmovdqu32 ZMMWORD PTR [r9+0x1c0],zmm7
+        vmovdqu32 ZMMWORD PTR [r9+0x200],zmm8
+        vmovdqu32 ZMMWORD PTR [r9+0x240],zmm9
+        vmovdqu32 ZMMWORD PTR [r9+0x280],zmm10
+        vmovdqu32 ZMMWORD PTR [r9+0x2c0],zmm11
+        vmovdqu32 ZMMWORD PTR [r9+0x300],zmm12
+        vmovdqu32 ZMMWORD PTR [r9+0x340],zmm13
+        vmovdqu32 ZMMWORD PTR [r9+0x380],zmm14
+        vmovdqu32 ZMMWORD PTR [r9+0x3c0],zmm15
+        vmovdqa32 zmm0,ZMMWORD PTR [rsp]
+        vmovdqa32 zmm1,ZMMWORD PTR [rsp+0x40]
+        vpaddd zmm2,zmm0,DWORD PTR [ADD16+rip]{1to16}
+        vpcmpltud k1,zmm2,zmm0
+        vpaddd zmm1{k1},zmm1,DWORD PTR [ADD1+rip]{1to16}
+        vmovdqa32 ZMMWORD PTR [rsp],zmm2
+        vmovdqa32 ZMMWORD PTR [rsp+0x40],zmm1
+        add    r9,0x400
+        sub    r10,0x10
+        cmp    r10,0x10
+        jae    3b
+        test   r10,r10
+        jne    2f
+9:
+        vzeroupper 
+        mov    rsp,rbp
+        pop    rbp
+        ret    
+2:
+        test   r10,0x8
+        je     2f
+        vpbroadcastd ymm16,DWORD PTR [rsi]
+        vpbroadcastd ymm17,DWORD PTR [rsi+0x4]
+        vpbroadcastd ymm18,DWORD PTR [rsi+0x8]
+        vpbroadcastd ymm19,DWORD PTR [rsi+0xc]
+        vpbroadcastd ymm20,DWORD PTR [rsi+0x10]
+        vpbroadcastd ymm21,DWORD PTR [rsi+0x14]
+        vpbroadcastd ymm22,DWORD PTR [rsi+0x18]
+        vpbroadcastd ymm23,DWORD PTR [rsi+0x1c]
+        vpbroadcastd ymm24,DWORD PTR [rsi+0x20]
+        vpbroadcastd ymm25,DWORD PTR [rsi+0x24]
+        vpbroadcastd ymm26,DWORD PTR [rsi+0x28]
+        vpbroadcastd ymm27,DWORD PTR [rsi+0x2c]
+        vpbroadcastd ymm28,DWORD PTR [rsi+0x30]
+        vpbroadcastd ymm29,DWORD PTR [rsi+0x34]
+        vpbroadcastd ymm30,DWORD PTR [rsi+0x38]
+        vpbroadcastd ymm31,DWORD PTR [rsi+0x3c]
+        vpbroadcastd ymm0,DWORD PTR [rdi]
+        vpbroadcastd ymm1,DWORD PTR [rdi+0x4]
+        vpbroadcastd ymm2,DWORD PTR [rdi+0x8]
+        vpbroadcastd ymm3,DWORD PTR [rdi+0xc]
+        vpbroadcastd ymm4,DWORD PTR [rdi+0x10]
+        vpbroadcastd ymm5,DWORD PTR [rdi+0x14]
+        vpbroadcastd ymm6,DWORD PTR [rdi+0x18]
+        vpbroadcastd ymm7,DWORD PTR [rdi+0x1c]
+        vpbroadcastd ymm8,DWORD PTR [BLAKE3_IV_0+rip]
+        vpbroadcastd ymm9,DWORD PTR [BLAKE3_IV_1+rip]
+        vpbroadcastd ymm10,DWORD PTR [BLAKE3_IV_2+rip]
+        vpbroadcastd ymm11,DWORD PTR [BLAKE3_IV_3+rip]
+        vmovdqa ymm12,YMMWORD PTR [rsp]
+        vmovdqa ymm13,YMMWORD PTR [rsp+0x40]
+        vpbroadcastd ymm14,edx
+        vpbroadcastd ymm15,r8d
+        vpaddd ymm0,ymm0,ymm16
+        vpaddd ymm1,ymm1,ymm18
+        vpaddd ymm2,ymm2,ymm20
+        vpaddd ymm3,ymm3,ymm22
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vprord ymm15,ymm15,0x10
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0xc
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vpaddd ymm0,ymm0,ymm17
+        vpaddd ymm1,ymm1,ymm19
+        vpaddd ymm2,ymm2,ymm21
+        vpaddd ymm3,ymm3,ymm23
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vprord ymm15,ymm15,0x8
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0x7
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vpaddd ymm0,ymm0,ymm24
+        vpaddd ymm1,ymm1,ymm26
+        vpaddd ymm2,ymm2,ymm28
+        vpaddd ymm3,ymm3,ymm30
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x10
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vprord ymm4,ymm4,0xc
+        vpaddd ymm0,ymm0,ymm25
+        vpaddd ymm1,ymm1,ymm27
+        vpaddd ymm2,ymm2,ymm29
+        vpaddd ymm3,ymm3,ymm31
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x8
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vprord ymm4,ymm4,0x7
+        vpaddd ymm0,ymm0,ymm18
+        vpaddd ymm1,ymm1,ymm19
+        vpaddd ymm2,ymm2,ymm23
+        vpaddd ymm3,ymm3,ymm20
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vprord ymm15,ymm15,0x10
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0xc
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vpaddd ymm0,ymm0,ymm22
+        vpaddd ymm1,ymm1,ymm26
+        vpaddd ymm2,ymm2,ymm16
+        vpaddd ymm3,ymm3,ymm29
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vprord ymm15,ymm15,0x8
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0x7
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vpaddd ymm0,ymm0,ymm17
+        vpaddd ymm1,ymm1,ymm28
+        vpaddd ymm2,ymm2,ymm25
+        vpaddd ymm3,ymm3,ymm31
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x10
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vprord ymm4,ymm4,0xc
+        vpaddd ymm0,ymm0,ymm27
+        vpaddd ymm1,ymm1,ymm21
+        vpaddd ymm2,ymm2,ymm30
+        vpaddd ymm3,ymm3,ymm24
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x8
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vprord ymm4,ymm4,0x7
+        vpaddd ymm0,ymm0,ymm19
+        vpaddd ymm1,ymm1,ymm26
+        vpaddd ymm2,ymm2,ymm29
+        vpaddd ymm3,ymm3,ymm23
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vprord ymm15,ymm15,0x10
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0xc
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vpaddd ymm0,ymm0,ymm20
+        vpaddd ymm1,ymm1,ymm28
+        vpaddd ymm2,ymm2,ymm18
+        vpaddd ymm3,ymm3,ymm30
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vprord ymm15,ymm15,0x8
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0x7
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vpaddd ymm0,ymm0,ymm22
+        vpaddd ymm1,ymm1,ymm25
+        vpaddd ymm2,ymm2,ymm27
+        vpaddd ymm3,ymm3,ymm24
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x10
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vprord ymm4,ymm4,0xc
+        vpaddd ymm0,ymm0,ymm21
+        vpaddd ymm1,ymm1,ymm16
+        vpaddd ymm2,ymm2,ymm31
+        vpaddd ymm3,ymm3,ymm17
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x8
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vprord ymm4,ymm4,0x7
+        vpaddd ymm0,ymm0,ymm26
+        vpaddd ymm1,ymm1,ymm28
+        vpaddd ymm2,ymm2,ymm30
+        vpaddd ymm3,ymm3,ymm29
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vprord ymm15,ymm15,0x10
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0xc
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vpaddd ymm0,ymm0,ymm23
+        vpaddd ymm1,ymm1,ymm25
+        vpaddd ymm2,ymm2,ymm19
+        vpaddd ymm3,ymm3,ymm31
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vprord ymm15,ymm15,0x8
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0x7
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vpaddd ymm0,ymm0,ymm20
+        vpaddd ymm1,ymm1,ymm27
+        vpaddd ymm2,ymm2,ymm21
+        vpaddd ymm3,ymm3,ymm17
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x10
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vprord ymm4,ymm4,0xc
+        vpaddd ymm0,ymm0,ymm16
+        vpaddd ymm1,ymm1,ymm18
+        vpaddd ymm2,ymm2,ymm24
+        vpaddd ymm3,ymm3,ymm22
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x8
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vprord ymm4,ymm4,0x7
+        vpaddd ymm0,ymm0,ymm28
+        vpaddd ymm1,ymm1,ymm25
+        vpaddd ymm2,ymm2,ymm31
+        vpaddd ymm3,ymm3,ymm30
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vprord ymm15,ymm15,0x10
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0xc
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vpaddd ymm0,ymm0,ymm29
+        vpaddd ymm1,ymm1,ymm27
+        vpaddd ymm2,ymm2,ymm26
+        vpaddd ymm3,ymm3,ymm24
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vprord ymm15,ymm15,0x8
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0x7
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vpaddd ymm0,ymm0,ymm23
+        vpaddd ymm1,ymm1,ymm21
+        vpaddd ymm2,ymm2,ymm16
+        vpaddd ymm3,ymm3,ymm22
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x10
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vprord ymm4,ymm4,0xc
+        vpaddd ymm0,ymm0,ymm18
+        vpaddd ymm1,ymm1,ymm19
+        vpaddd ymm2,ymm2,ymm17
+        vpaddd ymm3,ymm3,ymm20
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x8
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vprord ymm4,ymm4,0x7
+        vpaddd ymm0,ymm0,ymm25
+        vpaddd ymm1,ymm1,ymm27
+        vpaddd ymm2,ymm2,ymm24
+        vpaddd ymm3,ymm3,ymm31
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vprord ymm15,ymm15,0x10
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0xc
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vpaddd ymm0,ymm0,ymm30
+        vpaddd ymm1,ymm1,ymm21
+        vpaddd ymm2,ymm2,ymm28
+        vpaddd ymm3,ymm3,ymm17
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vprord ymm15,ymm15,0x8
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0x7
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vpaddd ymm0,ymm0,ymm29
+        vpaddd ymm1,ymm1,ymm16
+        vpaddd ymm2,ymm2,ymm18
+        vpaddd ymm3,ymm3,ymm20
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x10
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vprord ymm4,ymm4,0xc
+        vpaddd ymm0,ymm0,ymm19
+        vpaddd ymm1,ymm1,ymm26
+        vpaddd ymm2,ymm2,ymm22
+        vpaddd ymm3,ymm3,ymm23
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x8
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vprord ymm4,ymm4,0x7
+        vpaddd ymm0,ymm0,ymm27
+        vpaddd ymm1,ymm1,ymm21
+        vpaddd ymm2,ymm2,ymm17
+        vpaddd ymm3,ymm3,ymm24
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vprord ymm15,ymm15,0x10
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0xc
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vpaddd ymm0,ymm0,ymm31
+        vpaddd ymm1,ymm1,ymm16
+        vpaddd ymm2,ymm2,ymm25
+        vpaddd ymm3,ymm3,ymm22
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm1,ymm1,ymm5
+        vpaddd ymm2,ymm2,ymm6
+        vpaddd ymm3,ymm3,ymm7
+        vpxord ymm12,ymm12,ymm0
+        vpxord ymm13,ymm13,ymm1
+        vpxord ymm14,ymm14,ymm2
+        vpxord ymm15,ymm15,ymm3
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vprord ymm15,ymm15,0x8
+        vpaddd ymm8,ymm8,ymm12
+        vpaddd ymm9,ymm9,ymm13
+        vpaddd ymm10,ymm10,ymm14
+        vpaddd ymm11,ymm11,ymm15
+        vpxord ymm4,ymm4,ymm8
+        vpxord ymm5,ymm5,ymm9
+        vpxord ymm6,ymm6,ymm10
+        vpxord ymm7,ymm7,ymm11
+        vprord ymm4,ymm4,0x7
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vpaddd ymm0,ymm0,ymm30
+        vpaddd ymm1,ymm1,ymm18
+        vpaddd ymm2,ymm2,ymm19
+        vpaddd ymm3,ymm3,ymm23
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x10
+        vprord ymm12,ymm12,0x10
+        vprord ymm13,ymm13,0x10
+        vprord ymm14,ymm14,0x10
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0xc
+        vprord ymm6,ymm6,0xc
+        vprord ymm7,ymm7,0xc
+        vprord ymm4,ymm4,0xc
+        vpaddd ymm0,ymm0,ymm26
+        vpaddd ymm1,ymm1,ymm28
+        vpaddd ymm2,ymm2,ymm20
+        vpaddd ymm3,ymm3,ymm29
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm1,ymm1,ymm6
+        vpaddd ymm2,ymm2,ymm7
+        vpaddd ymm3,ymm3,ymm4
+        vpxord ymm15,ymm15,ymm0
+        vpxord ymm12,ymm12,ymm1
+        vpxord ymm13,ymm13,ymm2
+        vpxord ymm14,ymm14,ymm3
+        vprord ymm15,ymm15,0x8
+        vprord ymm12,ymm12,0x8
+        vprord ymm13,ymm13,0x8
+        vprord ymm14,ymm14,0x8
+        vpaddd ymm10,ymm10,ymm15
+        vpaddd ymm11,ymm11,ymm12
+        vpaddd ymm8,ymm8,ymm13
+        vpaddd ymm9,ymm9,ymm14
+        vpxord ymm5,ymm5,ymm10
+        vpxord ymm6,ymm6,ymm11
+        vpxord ymm7,ymm7,ymm8
+        vpxord ymm4,ymm4,ymm9
+        vprord ymm5,ymm5,0x7
+        vprord ymm6,ymm6,0x7
+        vprord ymm7,ymm7,0x7
+        vprord ymm4,ymm4,0x7
+        vpxor  ymm0,ymm0,ymm8
+        vpxor  ymm1,ymm1,ymm9
+        vpxor  ymm2,ymm2,ymm10
+        vpxor  ymm3,ymm3,ymm11
+        vpxor  ymm4,ymm4,ymm12
+        vpxor  ymm5,ymm5,ymm13
+        vpxor  ymm6,ymm6,ymm14
+        vpxor  ymm7,ymm7,ymm15
+        vpxord ymm8,ymm8,DWORD PTR [rdi]{1to8}
+        vpxord ymm9,ymm9,DWORD PTR [rdi+0x4]{1to8}
+        vpxord ymm10,ymm10,DWORD PTR [rdi+0x8]{1to8}
+        vpxord ymm11,ymm11,DWORD PTR [rdi+0xc]{1to8}
+        vpxord ymm12,ymm12,DWORD PTR [rdi+0x10]{1to8}
+        vpxord ymm13,ymm13,DWORD PTR [rdi+0x14]{1to8}
+        vpxord ymm14,ymm14,DWORD PTR [rdi+0x18]{1to8}
+        vpxord ymm15,ymm15,DWORD PTR [rdi+0x1c]{1to8}
+        vpunpckldq ymm16,ymm0,ymm1
+        vpunpckhdq ymm17,ymm0,ymm1
+        vpunpckldq ymm18,ymm2,ymm3
+        vpunpckhdq ymm19,ymm2,ymm3
+        vpunpckldq ymm20,ymm4,ymm5
+        vpunpckhdq ymm21,ymm4,ymm5
+        vpunpckldq ymm22,ymm6,ymm7
+        vpunpckhdq ymm23,ymm6,ymm7
+        vpunpckldq ymm24,ymm8,ymm9
+        vpunpckhdq ymm25,ymm8,ymm9
+        vpunpckldq ymm26,ymm10,ymm11
+        vpunpckhdq ymm27,ymm10,ymm11
+        vpunpckldq ymm28,ymm12,ymm13
+        vpunpckhdq ymm29,ymm12,ymm13
+        vpunpckldq ymm30,ymm14,ymm15
+        vpunpckhdq ymm31,ymm14,ymm15
+        vpunpcklqdq ymm0,ymm16,ymm18
+        vpunpckhqdq ymm1,ymm16,ymm18
+        vpunpcklqdq ymm2,ymm17,ymm19
+        vpunpckhqdq ymm3,ymm17,ymm19
+        vpunpcklqdq ymm4,ymm20,ymm22
+        vpunpckhqdq ymm5,ymm20,ymm22
+        vpunpcklqdq ymm6,ymm21,ymm23
+        vpunpckhqdq ymm7,ymm21,ymm23
+        vpunpcklqdq ymm8,ymm24,ymm26
+        vpunpckhqdq ymm9,ymm24,ymm26
+        vpunpcklqdq ymm10,ymm25,ymm27
+        vpunpckhqdq ymm11,ymm25,ymm27
+        vpunpcklqdq ymm12,ymm28,ymm30
+        vpunpckhqdq ymm13,ymm28,ymm30
+        vpunpcklqdq ymm14,ymm29,ymm31
+        vpunpckhqdq ymm15,ymm29,ymm31
+        vshufi32x4 ymm16,ymm0,ymm4,0x0
+        vshufi32x4 ymm17,ymm8,ymm12,0x0
+        vshufi32x4 ymm18,ymm1,ymm5,0x0
+        vshufi32x4 ymm19,ymm9,ymm13,0x0
+        vshufi32x4 ymm20,ymm2,ymm6,0x0
+        vshufi32x4 ymm21,ymm10,ymm14,0x0
+        vshufi32x4 ymm22,ymm3,ymm7,0x0
+        vshufi32x4 ymm23,ymm11,ymm15,0x0
+        vshufi32x4 ymm24,ymm0,ymm4,0x3
+        vshufi32x4 ymm25,ymm8,ymm12,0x3
+        vshufi32x4 ymm26,ymm1,ymm5,0x3
+        vshufi32x4 ymm27,ymm9,ymm13,0x3
+        vshufi32x4 ymm28,ymm2,ymm6,0x3
+        vshufi32x4 ymm29,ymm10,ymm14,0x3
+        vshufi32x4 ymm30,ymm3,ymm7,0x3
+        vshufi32x4 ymm31,ymm11,ymm15,0x3
+        vmovdqu32 YMMWORD PTR [r9],ymm16
+        vmovdqu32 YMMWORD PTR [r9+0x20],ymm17
+        vmovdqu32 YMMWORD PTR [r9+0x40],ymm18
+        vmovdqu32 YMMWORD PTR [r9+0x60],ymm19
+        vmovdqu32 YMMWORD PTR [r9+0x80],ymm20
+        vmovdqu32 YMMWORD PTR [r9+0xa0],ymm21
+        vmovdqu32 YMMWORD PTR [r9+0xc0],ymm22
+        vmovdqu32 YMMWORD PTR [r9+0xe0],ymm23
+        vmovdqu32 YMMWORD PTR [r9+0x100],ymm24
+        vmovdqu32 YMMWORD PTR [r9+0x120],ymm25
+        vmovdqu32 YMMWORD PTR [r9+0x140],ymm26
+        vmovdqu32 YMMWORD PTR [r9+0x160],ymm27
+        vmovdqu32 YMMWORD PTR [r9+0x180],ymm28
+        vmovdqu32 YMMWORD PTR [r9+0x1a0],ymm29
+        vmovdqu32 YMMWORD PTR [r9+0x1c0],ymm30
+        vmovdqu32 YMMWORD PTR [r9+0x1e0],ymm31
+        vmovdqa ymm0,YMMWORD PTR [rsp+0x20]
+        vmovdqa ymm1,YMMWORD PTR [rsp+0x60]
+        vmovdqa YMMWORD PTR [rsp],ymm0
+        vmovdqa YMMWORD PTR [rsp+0x40],ymm1
+        add    r9,0x200
+        sub    r10,0x8
+2:
+        test   r10,0x4
+        je     2f
+        vbroadcasti32x4 zmm0,XMMWORD PTR [rdi]
+        vbroadcasti32x4 zmm1,XMMWORD PTR [rdi+0x10]
+        vbroadcasti32x4 zmm2,XMMWORD PTR [BLAKE3_IV+rip]
+        vmovdqa xmm12,XMMWORD PTR [rsp]
+        vmovdqa xmm13,XMMWORD PTR [rsp+0x40]
+        vpunpckldq xmm14,xmm12,xmm13
+        vpunpckhdq xmm15,xmm12,xmm13
+        vpermq ymm14,ymm14,0xdc
+        vpermq ymm15,ymm15,0xdc
+        vpbroadcastd zmm12,edx
+        vinserti64x4 zmm13,zmm14,ymm15,0x1
+        mov    eax,0x4444
+        kmovw  k2,eax
+        vpblendmd zmm13{k2},zmm13,zmm12
+        vpbroadcastd zmm15,r8d
+        mov    eax,0x8888
+        kmovw  k4,eax
+        vpblendmd zmm3{k4},zmm13,zmm15
+        mov    eax,0xaaaa
+        kmovw  k3,eax
+        vbroadcasti32x4 zmm8,XMMWORD PTR [rsi]
+        vbroadcasti32x4 zmm9,XMMWORD PTR [rsi+0x10]
+        vshufps zmm4,zmm8,zmm9,0x88
+        vshufps zmm5,zmm8,zmm9,0xdd
+        vbroadcasti32x4 zmm8,XMMWORD PTR [rsi+0x20]
+        vbroadcasti32x4 zmm9,XMMWORD PTR [rsi+0x30]
+        vshufps zmm6,zmm8,zmm9,0x88
+        vshufps zmm7,zmm8,zmm9,0xdd
+        vpshufd zmm6,zmm6,0x93
+        vpshufd zmm7,zmm7,0x93
+        mov    al,0x7
+3:
+        vpaddd zmm0,zmm0,zmm4
+        vpaddd zmm0,zmm0,zmm1
+        vpxord zmm3,zmm3,zmm0
+        vprord zmm3,zmm3,0x10
+        vpaddd zmm2,zmm2,zmm3
+        vpxord zmm1,zmm1,zmm2
+        vprord zmm1,zmm1,0xc
+        vpaddd zmm0,zmm0,zmm5
+        vpaddd zmm0,zmm0,zmm1
+        vpxord zmm3,zmm3,zmm0
+        vprord zmm3,zmm3,0x8
+        vpaddd zmm2,zmm2,zmm3
+        vpxord zmm1,zmm1,zmm2
+        vprord zmm1,zmm1,0x7
+        vpshufd zmm0,zmm0,0x93
+        vpshufd zmm3,zmm3,0x4e
+        vpshufd zmm2,zmm2,0x39
+        vpaddd zmm0,zmm0,zmm6
+        vpaddd zmm0,zmm0,zmm1
+        vpxord zmm3,zmm3,zmm0
+        vprord zmm3,zmm3,0x10
+        vpaddd zmm2,zmm2,zmm3
+        vpxord zmm1,zmm1,zmm2
+        vprord zmm1,zmm1,0xc
+        vpaddd zmm0,zmm0,zmm7
+        vpaddd zmm0,zmm0,zmm1
+        vpxord zmm3,zmm3,zmm0
+        vprord zmm3,zmm3,0x8
+        vpaddd zmm2,zmm2,zmm3
+        vpxord zmm1,zmm1,zmm2
+        vprord zmm1,zmm1,0x7
+        vpshufd zmm0,zmm0,0x39
+        vpshufd zmm3,zmm3,0x4e
+        vpshufd zmm2,zmm2,0x93
+        dec    al
+        je     3f
+        vshufps zmm8,zmm4,zmm5,0xd6
+        vpshufd zmm9,zmm4,0xf
+        vpshufd zmm4,zmm8,0x39
+        vshufps zmm8,zmm6,zmm7,0xfa
+        vpblendmd zmm9{k3},zmm9,zmm8
+        vpunpcklqdq zmm8,zmm7,zmm5
+        vpblendmd zmm8{k4},zmm8,zmm6
+        vpshufd zmm8,zmm8,0x78
+        vpunpckhdq zmm5,zmm5,zmm7
+        vpunpckldq zmm6,zmm6,zmm5
+        vpshufd zmm7,zmm6,0x1e
+        vmovdqa32 zmm5,zmm9
+        vmovdqa32 zmm6,zmm8
+        jmp    3b
+3:
+        vpxord zmm0,zmm0,zmm2
+        vpxord zmm1,zmm1,zmm3
+        vbroadcasti32x4 zmm8,XMMWORD PTR [rdi]
+        vbroadcasti32x4 zmm9,XMMWORD PTR [rdi+0x10]
+        vpxord zmm2,zmm2,zmm8
+        vpxord zmm3,zmm3,zmm9
+        vmovdqu XMMWORD PTR [r9],xmm0
+        vmovdqu XMMWORD PTR [r9+0x10],xmm1
+        vmovdqu XMMWORD PTR [r9+0x20],xmm2
+        vmovdqu XMMWORD PTR [r9+0x30],xmm3
+        vextracti128 XMMWORD PTR [r9+0x40],ymm0,0x1
+        vextracti128 XMMWORD PTR [r9+0x50],ymm1,0x1
+        vextracti128 XMMWORD PTR [r9+0x60],ymm2,0x1
+        vextracti128 XMMWORD PTR [r9+0x70],ymm3,0x1
+        vextracti32x4 XMMWORD PTR [r9+0x80],zmm0,0x2
+        vextracti32x4 XMMWORD PTR [r9+0x90],zmm1,0x2
+        vextracti32x4 XMMWORD PTR [r9+0xa0],zmm2,0x2
+        vextracti32x4 XMMWORD PTR [r9+0xb0],zmm3,0x2
+        vextracti32x4 XMMWORD PTR [r9+0xc0],zmm0,0x3
+        vextracti32x4 XMMWORD PTR [r9+0xd0],zmm1,0x3
+        vextracti32x4 XMMWORD PTR [r9+0xe0],zmm2,0x3
+        vextracti32x4 XMMWORD PTR [r9+0xf0],zmm3,0x3
+        vmovdqa xmm0,XMMWORD PTR [rsp+0x10]
+        vmovdqa xmm1,XMMWORD PTR [rsp+0x50]
+        vmovdqa XMMWORD PTR [rsp],xmm0
+        vmovdqa XMMWORD PTR [rsp+0x40],xmm1
+        add    r9,0x100
+        sub    r10,0x4
+2:
+        test   r10,0x2
+        je     2f
+        vbroadcasti128 ymm0,XMMWORD PTR [rdi]
+        vbroadcasti128 ymm1,XMMWORD PTR [rdi+0x10]
+        vmovd  xmm13,DWORD PTR [rsp]
+        vpinsrd xmm13,xmm13,DWORD PTR [rsp+0x40],0x1
+        vpinsrd xmm13,xmm13,edx,0x2
+        vmovd  xmm14,DWORD PTR [rsp+0x4]
+        vpinsrd xmm14,xmm14,DWORD PTR [rsp+0x44],0x1
+        vpinsrd xmm14,xmm14,edx,0x2
+        vinserti128 ymm13,ymm13,xmm14,0x1
+        vbroadcasti128 ymm2,XMMWORD PTR [BLAKE3_IV+rip]
+        vpbroadcastd ymm8,r8d
+        vpblendd ymm3,ymm13,ymm8,0x88
+        vbroadcasti128 ymm8,XMMWORD PTR [rsi]
+        vbroadcasti128 ymm9,XMMWORD PTR [rsi+0x10]
+        vshufps ymm4,ymm8,ymm9,0x88
+        vshufps ymm5,ymm8,ymm9,0xdd
+        vbroadcasti128 ymm8,XMMWORD PTR [rsi+0x20]
+        vbroadcasti128 ymm9,XMMWORD PTR [rsi+0x30]
+        vshufps ymm6,ymm8,ymm9,0x88
+        vshufps ymm7,ymm8,ymm9,0xdd
+        vpshufd ymm6,ymm6,0x93
+        vpshufd ymm7,ymm7,0x93
+        mov    al,0x7
+3:
+        vpaddd ymm0,ymm0,ymm4
+        vpaddd ymm0,ymm0,ymm1
+        vpxord ymm3,ymm3,ymm0
+        vprord ymm3,ymm3,0x10
+        vpaddd ymm2,ymm2,ymm3
+        vpxord ymm1,ymm1,ymm2
+        vprord ymm1,ymm1,0xc
+        vpaddd ymm0,ymm0,ymm5
+        vpaddd ymm0,ymm0,ymm1
+        vpxord ymm3,ymm3,ymm0
+        vprord ymm3,ymm3,0x8
+        vpaddd ymm2,ymm2,ymm3
+        vpxord ymm1,ymm1,ymm2
+        vprord ymm1,ymm1,0x7
+        vpshufd ymm0,ymm0,0x93
+        vpshufd ymm3,ymm3,0x4e
+        vpshufd ymm2,ymm2,0x39
+        vpaddd ymm0,ymm0,ymm6
+        vpaddd ymm0,ymm0,ymm1
+        vpxord ymm3,ymm3,ymm0
+        vprord ymm3,ymm3,0x10
+        vpaddd ymm2,ymm2,ymm3
+        vpxord ymm1,ymm1,ymm2
+        vprord ymm1,ymm1,0xc
+        vpaddd ymm0,ymm0,ymm7
+        vpaddd ymm0,ymm0,ymm1
+        vpxord ymm3,ymm3,ymm0
+        vprord ymm3,ymm3,0x8
+        vpaddd ymm2,ymm2,ymm3
+        vpxord ymm1,ymm1,ymm2
+        vprord ymm1,ymm1,0x7
+        vpshufd ymm0,ymm0,0x39
+        vpshufd ymm3,ymm3,0x4e
+        vpshufd ymm2,ymm2,0x93
+        dec    al
+        je     3f
+        vshufps ymm8,ymm4,ymm5,0xd6
+        vpshufd ymm9,ymm4,0xf
+        vpshufd ymm4,ymm8,0x39
+        vshufps ymm8,ymm6,ymm7,0xfa
+        vpblendd ymm9,ymm9,ymm8,0xaa
+        vpunpcklqdq ymm8,ymm7,ymm5
+        vpblendd ymm8,ymm8,ymm6,0x88
+        vpshufd ymm8,ymm8,0x78
+        vpunpckhdq ymm5,ymm5,ymm7
+        vpunpckldq ymm6,ymm6,ymm5
+        vpshufd ymm7,ymm6,0x1e
+        vmovdqa ymm5,ymm9
+        vmovdqa ymm6,ymm8
+        jmp    3b
+3:
+        vpxor  ymm0,ymm0,ymm2
+        vpxor  ymm1,ymm1,ymm3
+        vbroadcasti128 ymm8,XMMWORD PTR [rdi]
+        vbroadcasti128 ymm9,XMMWORD PTR [rdi+0x10]
+        vpxor  ymm2,ymm2,ymm8
+        vpxor  ymm3,ymm3,ymm9
+        vmovdqu XMMWORD PTR [r9],xmm0
+        vmovdqu XMMWORD PTR [r9+0x10],xmm1
+        vmovdqu XMMWORD PTR [r9+0x20],xmm2
+        vmovdqu XMMWORD PTR [r9+0x30],xmm3
+        vextracti128 XMMWORD PTR [r9+0x40],ymm0,0x1
+        vextracti128 XMMWORD PTR [r9+0x50],ymm1,0x1
+        vextracti128 XMMWORD PTR [r9+0x60],ymm2,0x1
+        vextracti128 XMMWORD PTR [r9+0x70],ymm3,0x1
+        vmovdqu xmm0,XMMWORD PTR [rsp+0x8]
+        vmovdqu xmm1,XMMWORD PTR [rsp+0x48]
+        vmovdqa XMMWORD PTR [rsp],xmm0
+        vmovdqa XMMWORD PTR [rsp+0x40],xmm1
+        add    r9,0x80
+        sub    r10,0x2
+2:
+        test   r10,0x1
+        je     9b
+        vmovdqu xmm0,XMMWORD PTR [rdi]
+        vmovdqu xmm1,XMMWORD PTR [rdi+0x10]
+        vmovd  xmm14,DWORD PTR [rsp]
+        vpinsrd xmm14,xmm14,DWORD PTR [rsp+0x40],0x1
+        vpinsrd xmm14,xmm14,edx,0x2
+        vmovdqa xmm2,XMMWORD PTR [BLAKE3_IV+rip]
+        vpinsrd xmm3,xmm14,r8d,0x3
+        vmovups xmm8,XMMWORD PTR [rsi]
+        vmovups xmm9,XMMWORD PTR [rsi+0x10]
+        vshufps xmm4,xmm8,xmm9,0x88
+        vshufps xmm5,xmm8,xmm9,0xdd
+        vmovups xmm8,XMMWORD PTR [rsi+0x20]
+        vmovups xmm9,XMMWORD PTR [rsi+0x30]
+        vshufps xmm6,xmm8,xmm9,0x88
+        vshufps xmm7,xmm8,xmm9,0xdd
+        vpshufd xmm6,xmm6,0x93
+        vpshufd xmm7,xmm7,0x93
+        mov    al,0x7
+3:
+        vpaddd xmm0,xmm0,xmm4
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x10
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0xc
+        vpaddd xmm0,xmm0,xmm5
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x8
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0x7
+        vpshufd xmm0,xmm0,0x93
+        vpshufd xmm3,xmm3,0x4e
+        vpshufd xmm2,xmm2,0x39
+        vpaddd xmm0,xmm0,xmm6
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x10
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0xc
+        vpaddd xmm0,xmm0,xmm7
+        vpaddd xmm0,xmm0,xmm1
+        vpxord xmm3,xmm3,xmm0
+        vprord xmm3,xmm3,0x8
+        vpaddd xmm2,xmm2,xmm3
+        vpxord xmm1,xmm1,xmm2
+        vprord xmm1,xmm1,0x7
+        vpshufd xmm0,xmm0,0x39
+        vpshufd xmm3,xmm3,0x4e
+        vpshufd xmm2,xmm2,0x93
+        dec    al
+        je     3f
+        vshufps xmm8,xmm4,xmm5,0xd6
+        vpshufd xmm9,xmm4,0xf
+        vpshufd xmm4,xmm8,0x39
+        vshufps xmm8,xmm6,xmm7,0xfa
+        vpblendd xmm9,xmm9,xmm8,0xaa
+        vpunpcklqdq xmm8,xmm7,xmm5
+        vpblendd xmm8,xmm8,xmm6,0x88
+        vpshufd xmm8,xmm8,0x78
+        vpunpckhdq xmm5,xmm5,xmm7
+        vpunpckldq xmm6,xmm6,xmm5
+        vpshufd xmm7,xmm6,0x1e
+        vmovdqa xmm5,xmm9
+        vmovdqa xmm6,xmm8
+        jmp    3b
+3:
+        vpxor  xmm0,xmm0,xmm2
+        vpxor  xmm1,xmm1,xmm3
+        vpxor  xmm2,xmm2,XMMWORD PTR [rdi]
+        vpxor  xmm3,xmm3,XMMWORD PTR [rdi+0x10]
+        vmovdqu XMMWORD PTR [r9],xmm0
+        vmovdqu XMMWORD PTR [r9+0x10],xmm1
+        vmovdqu XMMWORD PTR [r9+0x20],xmm2
+        vmovdqu XMMWORD PTR [r9+0x30],xmm3
+        jmp    9b
+
+
+#ifdef __APPLE__
+.static_data
+#else
+.section .rodata
+#endif
+.p2align  6
+INDEX0:
+        .long    0,  1,  2,  3, 16, 17, 18, 19
+        .long    8,  9, 10, 11, 24, 25, 26, 27
+INDEX1:
+        .long    4,  5,  6,  7, 20, 21, 22, 23
+        .long   12, 13, 14, 15, 28, 29, 30, 31
+ADD0:
+        .long    0,  1,  2,  3,  4,  5,  6,  7
+        .long    8,  9, 10, 11, 12, 13, 14, 15
+ADD1:   .long    1
+
+ADD16:  .long   16
+BLAKE3_BLOCK_LEN:
+        .long   64
+.p2align 6
+BLAKE3_IV:
+BLAKE3_IV_0:
+        .long   0x6A09E667
+BLAKE3_IV_1:
+        .long   0xBB67AE85
+BLAKE3_IV_2:
+        .long   0x3C6EF372
+BLAKE3_IV_3:
+        .long   0xA54FF53A
diff --git a/thirdparty/blake3/c/blake3_avx512_x86-64_windows_gnu.S b/thirdparty/blake3/c/blake3_avx512_x86-64_windows_gnu.S
new file mode 100644
index 000000000..ba4fc5fa9
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx512_x86-64_windows_gnu.S
@@ -0,0 +1,2615 @@
+.intel_syntax noprefix
+
+.global _blake3_hash_many_avx512
+.global blake3_hash_many_avx512
+.global blake3_compress_in_place_avx512
+.global _blake3_compress_in_place_avx512
+.global blake3_compress_xof_avx512
+.global _blake3_compress_xof_avx512
+
+.section .text
+.p2align  6
+_blake3_hash_many_avx512:
+blake3_hash_many_avx512:
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rdi
+        push    rsi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 304
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        vmovdqa xmmword ptr [rsp+0x90], xmm6
+        vmovdqa xmmword ptr [rsp+0xA0], xmm7
+        vmovdqa xmmword ptr [rsp+0xB0], xmm8
+        vmovdqa xmmword ptr [rsp+0xC0], xmm9
+        vmovdqa xmmword ptr [rsp+0xD0], xmm10
+        vmovdqa xmmword ptr [rsp+0xE0], xmm11
+        vmovdqa xmmword ptr [rsp+0xF0], xmm12
+        vmovdqa xmmword ptr [rsp+0x100], xmm13
+        vmovdqa xmmword ptr [rsp+0x110], xmm14
+        vmovdqa xmmword ptr [rsp+0x120], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+0x68]
+        movzx   r9, byte ptr [rbp+0x70]
+        neg     r9
+        kmovw   k1, r9d
+        vmovd   xmm0, r8d
+        vpbroadcastd ymm0, xmm0
+        shr     r8, 32
+        vmovd   xmm1, r8d
+        vpbroadcastd ymm1, xmm1
+        vmovdqa ymm4, ymm1
+        vmovdqa ymm5, ymm1
+        vpaddd  ymm2, ymm0, ymmword ptr [ADD0+rip]
+        vpaddd  ymm3, ymm0, ymmword ptr [ADD0+32+rip]
+        vpcmpltud k2, ymm2, ymm0
+        vpcmpltud k3, ymm3, ymm0
+        vpaddd  ymm4 {k2}, ymm4, dword ptr [ADD1+rip] {1to8}
+        vpaddd  ymm5 {k3}, ymm5, dword ptr [ADD1+rip] {1to8}
+        knotw   k2, k1
+        vmovdqa32 ymm2 {k2}, ymm0
+        vmovdqa32 ymm3 {k2}, ymm0
+        vmovdqa32 ymm4 {k2}, ymm1
+        vmovdqa32 ymm5 {k2}, ymm1
+        vmovdqa ymmword ptr [rsp], ymm2
+        vmovdqa ymmword ptr [rsp+0x20], ymm3
+        vmovdqa ymmword ptr [rsp+0x40], ymm4
+        vmovdqa ymmword ptr [rsp+0x60], ymm5
+        shl     rdx, 6
+        mov     qword ptr [rsp+0x80], rdx
+        cmp     rsi, 16
+        jc      3f
+2:
+        vpbroadcastd zmm0, dword ptr [rcx]
+        vpbroadcastd zmm1, dword ptr [rcx+0x1*0x4]
+        vpbroadcastd zmm2, dword ptr [rcx+0x2*0x4]
+        vpbroadcastd zmm3, dword ptr [rcx+0x3*0x4]
+        vpbroadcastd zmm4, dword ptr [rcx+0x4*0x4]
+        vpbroadcastd zmm5, dword ptr [rcx+0x5*0x4]
+        vpbroadcastd zmm6, dword ptr [rcx+0x6*0x4]
+        vpbroadcastd zmm7, dword ptr [rcx+0x7*0x4]
+        movzx   eax, byte ptr [rbp+0x78]
+        movzx   ebx, byte ptr [rbp+0x80]
+        or      eax, ebx
+        xor     edx, edx
+.p2align 5
+9:
+        movzx   ebx, byte ptr [rbp+0x88]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+0x80]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+0x88], eax
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x40]
+        mov     r13, qword ptr [rdi+0x48]
+        mov     r14, qword ptr [rdi+0x50]
+        mov     r15, qword ptr [rdi+0x58]
+        vmovdqu32 ymm16, ymmword ptr [rdx+r8-0x2*0x20]
+        vinserti64x4 zmm16, zmm16, ymmword ptr [rdx+r12-0x2*0x20], 0x01
+        vmovdqu32 ymm17, ymmword ptr [rdx+r9-0x2*0x20]
+        vinserti64x4 zmm17, zmm17, ymmword ptr [rdx+r13-0x2*0x20], 0x01
+        vpunpcklqdq zmm8, zmm16, zmm17
+        vpunpckhqdq zmm9, zmm16, zmm17
+        vmovdqu32 ymm18, ymmword ptr [rdx+r10-0x2*0x20]
+        vinserti64x4 zmm18, zmm18, ymmword ptr [rdx+r14-0x2*0x20], 0x01
+        vmovdqu32 ymm19, ymmword ptr [rdx+r11-0x2*0x20]
+        vinserti64x4 zmm19, zmm19, ymmword ptr [rdx+r15-0x2*0x20], 0x01
+        vpunpcklqdq zmm10, zmm18, zmm19
+        vpunpckhqdq zmm11, zmm18, zmm19
+        mov     r8, qword ptr [rdi+0x20]
+        mov     r9, qword ptr [rdi+0x28]
+        mov     r10, qword ptr [rdi+0x30]
+        mov     r11, qword ptr [rdi+0x38]
+        mov     r12, qword ptr [rdi+0x60]
+        mov     r13, qword ptr [rdi+0x68]
+        mov     r14, qword ptr [rdi+0x70]
+        mov     r15, qword ptr [rdi+0x78]
+        vmovdqu32 ymm16, ymmword ptr [rdx+r8-0x2*0x20]
+        vinserti64x4 zmm16, zmm16, ymmword ptr [rdx+r12-0x2*0x20], 0x01
+        vmovdqu32 ymm17, ymmword ptr [rdx+r9-0x2*0x20]
+        vinserti64x4 zmm17, zmm17, ymmword ptr [rdx+r13-0x2*0x20], 0x01
+        vpunpcklqdq zmm12, zmm16, zmm17
+        vpunpckhqdq zmm13, zmm16, zmm17
+        vmovdqu32 ymm18, ymmword ptr [rdx+r10-0x2*0x20]
+        vinserti64x4 zmm18, zmm18, ymmword ptr [rdx+r14-0x2*0x20], 0x01
+        vmovdqu32 ymm19, ymmword ptr [rdx+r11-0x2*0x20]
+        vinserti64x4 zmm19, zmm19, ymmword ptr [rdx+r15-0x2*0x20], 0x01
+        vpunpcklqdq zmm14, zmm18, zmm19
+        vpunpckhqdq zmm15, zmm18, zmm19
+        vmovdqa32 zmm27, zmmword ptr [INDEX0+rip]
+        vmovdqa32 zmm31, zmmword ptr [INDEX1+rip]
+        vshufps zmm16, zmm8, zmm10, 136
+        vshufps zmm17, zmm12, zmm14, 136
+        vmovdqa32 zmm20, zmm16
+        vpermt2d zmm16, zmm27, zmm17
+        vpermt2d zmm20, zmm31, zmm17
+        vshufps zmm17, zmm8, zmm10, 221
+        vshufps zmm30, zmm12, zmm14, 221
+        vmovdqa32 zmm21, zmm17
+        vpermt2d zmm17, zmm27, zmm30
+        vpermt2d zmm21, zmm31, zmm30
+        vshufps zmm18, zmm9, zmm11, 136
+        vshufps zmm8, zmm13, zmm15, 136
+        vmovdqa32 zmm22, zmm18
+        vpermt2d zmm18, zmm27, zmm8
+        vpermt2d zmm22, zmm31, zmm8
+        vshufps zmm19, zmm9, zmm11, 221
+        vshufps zmm8, zmm13, zmm15, 221
+        vmovdqa32 zmm23, zmm19
+        vpermt2d zmm19, zmm27, zmm8
+        vpermt2d zmm23, zmm31, zmm8
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x40]
+        mov     r13, qword ptr [rdi+0x48]
+        mov     r14, qword ptr [rdi+0x50]
+        mov     r15, qword ptr [rdi+0x58]
+        vmovdqu32 ymm24, ymmword ptr [r8+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r12+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r9+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r13+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm8, zmm24, zmm25
+        vpunpckhqdq zmm9, zmm24, zmm25
+        vmovdqu32 ymm24, ymmword ptr [r10+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r14+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r11+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r15+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm10, zmm24, zmm25
+        vpunpckhqdq zmm11, zmm24, zmm25
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r12+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r13+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r14+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        prefetcht0 [r15+rdx+0x80]
+        mov     r8, qword ptr [rdi+0x20]
+        mov     r9, qword ptr [rdi+0x28]
+        mov     r10, qword ptr [rdi+0x30]
+        mov     r11, qword ptr [rdi+0x38]
+        mov     r12, qword ptr [rdi+0x60]
+        mov     r13, qword ptr [rdi+0x68]
+        mov     r14, qword ptr [rdi+0x70]
+        mov     r15, qword ptr [rdi+0x78]
+        vmovdqu32 ymm24, ymmword ptr [r8+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r12+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r9+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r13+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm12, zmm24, zmm25
+        vpunpckhqdq zmm13, zmm24, zmm25
+        vmovdqu32 ymm24, ymmword ptr [r10+rdx-0x1*0x20]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r14+rdx-0x1*0x20], 0x01
+        vmovdqu32 ymm25, ymmword ptr [r11+rdx-0x1*0x20]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r15+rdx-0x1*0x20], 0x01
+        vpunpcklqdq zmm14, zmm24, zmm25
+        vpunpckhqdq zmm15, zmm24, zmm25
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r12+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r13+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r14+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        prefetcht0 [r15+rdx+0x80]
+        vshufps zmm24, zmm8, zmm10, 136
+        vshufps zmm30, zmm12, zmm14, 136
+        vmovdqa32 zmm28, zmm24
+        vpermt2d zmm24, zmm27, zmm30
+        vpermt2d zmm28, zmm31, zmm30
+        vshufps zmm25, zmm8, zmm10, 221
+        vshufps zmm30, zmm12, zmm14, 221
+        vmovdqa32 zmm29, zmm25
+        vpermt2d zmm25, zmm27, zmm30
+        vpermt2d zmm29, zmm31, zmm30
+        vshufps zmm26, zmm9, zmm11, 136
+        vshufps zmm8, zmm13, zmm15, 136
+        vmovdqa32 zmm30, zmm26
+        vpermt2d zmm26, zmm27, zmm8
+        vpermt2d zmm30, zmm31, zmm8
+        vshufps zmm8, zmm9, zmm11, 221
+        vshufps zmm10, zmm13, zmm15, 221
+        vpermi2d zmm27, zmm8, zmm10
+        vpermi2d zmm31, zmm8, zmm10
+        vpbroadcastd zmm8, dword ptr [BLAKE3_IV_0+rip]
+        vpbroadcastd zmm9, dword ptr [BLAKE3_IV_1+rip]
+        vpbroadcastd zmm10, dword ptr [BLAKE3_IV_2+rip]
+        vpbroadcastd zmm11, dword ptr [BLAKE3_IV_3+rip]
+        vmovdqa32 zmm12, zmmword ptr [rsp]
+        vmovdqa32 zmm13, zmmword ptr [rsp+0x1*0x40]
+        vpbroadcastd zmm14, dword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpbroadcastd zmm15, dword ptr [rsp+0x22*0x4]
+        vpaddd  zmm0, zmm0, zmm16
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm20
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm17
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm21
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm24
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm28
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm25
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm29
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm18
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm23
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm22
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm16
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm17
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm25
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm27
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm30
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm19
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm29
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm20
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm18
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm22
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm27
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm21
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm31
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm26
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm30
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm23
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm19
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm20
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm21
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm16
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm24
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm28
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm31
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm29
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm26
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm23
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm16
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm18
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm17
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm25
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm24
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm30
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm28
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm29
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm18
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm19
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm22
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm27
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm17
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm31
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm25
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm30
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm19
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm26
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm20
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpxord  zmm0, zmm0, zmm8
+        vpxord  zmm1, zmm1, zmm9
+        vpxord  zmm2, zmm2, zmm10
+        vpxord  zmm3, zmm3, zmm11
+        vpxord  zmm4, zmm4, zmm12
+        vpxord  zmm5, zmm5, zmm13
+        vpxord  zmm6, zmm6, zmm14
+        vpxord  zmm7, zmm7, zmm15
+        movzx   eax, byte ptr [rbp+0x78]
+        jne     9b
+        mov     rbx, qword ptr [rbp+0x90]
+        vpunpckldq zmm16, zmm0, zmm1
+        vpunpckhdq zmm17, zmm0, zmm1
+        vpunpckldq zmm18, zmm2, zmm3
+        vpunpckhdq zmm19, zmm2, zmm3
+        vpunpckldq zmm20, zmm4, zmm5
+        vpunpckhdq zmm21, zmm4, zmm5
+        vpunpckldq zmm22, zmm6, zmm7
+        vpunpckhdq zmm23, zmm6, zmm7
+        vpunpcklqdq zmm0, zmm16, zmm18
+        vpunpckhqdq zmm1, zmm16, zmm18
+        vpunpcklqdq zmm2, zmm17, zmm19
+        vpunpckhqdq zmm3, zmm17, zmm19
+        vpunpcklqdq zmm4, zmm20, zmm22
+        vpunpckhqdq zmm5, zmm20, zmm22
+        vpunpcklqdq zmm6, zmm21, zmm23
+        vpunpckhqdq zmm7, zmm21, zmm23
+        vshufi32x4 zmm16, zmm0, zmm4, 0x88
+        vshufi32x4 zmm17, zmm1, zmm5, 0x88
+        vshufi32x4 zmm18, zmm2, zmm6, 0x88
+        vshufi32x4 zmm19, zmm3, zmm7, 0x88
+        vshufi32x4 zmm20, zmm0, zmm4, 0xDD
+        vshufi32x4 zmm21, zmm1, zmm5, 0xDD
+        vshufi32x4 zmm22, zmm2, zmm6, 0xDD
+        vshufi32x4 zmm23, zmm3, zmm7, 0xDD
+        vshufi32x4 zmm0, zmm16, zmm17, 0x88
+        vshufi32x4 zmm1, zmm18, zmm19, 0x88
+        vshufi32x4 zmm2, zmm20, zmm21, 0x88
+        vshufi32x4 zmm3, zmm22, zmm23, 0x88
+        vshufi32x4 zmm4, zmm16, zmm17, 0xDD
+        vshufi32x4 zmm5, zmm18, zmm19, 0xDD
+        vshufi32x4 zmm6, zmm20, zmm21, 0xDD
+        vshufi32x4 zmm7, zmm22, zmm23, 0xDD
+        vmovdqu32 zmmword ptr [rbx], zmm0
+        vmovdqu32 zmmword ptr [rbx+0x1*0x40], zmm1
+        vmovdqu32 zmmword ptr [rbx+0x2*0x40], zmm2
+        vmovdqu32 zmmword ptr [rbx+0x3*0x40], zmm3
+        vmovdqu32 zmmword ptr [rbx+0x4*0x40], zmm4
+        vmovdqu32 zmmword ptr [rbx+0x5*0x40], zmm5
+        vmovdqu32 zmmword ptr [rbx+0x6*0x40], zmm6
+        vmovdqu32 zmmword ptr [rbx+0x7*0x40], zmm7
+        vmovdqa32 zmm0, zmmword ptr [rsp]
+        vmovdqa32 zmm1, zmmword ptr [rsp+0x1*0x40]
+        vmovdqa32 zmm2, zmm0
+        vpaddd  zmm2{k1}, zmm0, dword ptr [ADD16+rip] {1to16}
+        vpcmpltud k2, zmm2, zmm0
+        vpaddd  zmm1 {k2}, zmm1, dword ptr [ADD1+rip] {1to16}
+        vmovdqa32 zmmword ptr [rsp], zmm2
+        vmovdqa32 zmmword ptr [rsp+0x1*0x40], zmm1
+        add     rdi, 128
+        add     rbx, 512
+        mov     qword ptr [rbp+0x90], rbx
+        sub     rsi, 16
+        cmp     rsi, 16
+        jnc     2b
+        test    rsi, rsi
+        jne     3f
+4:
+        vzeroupper
+        vmovdqa xmm6, xmmword ptr [rsp+0x90]
+        vmovdqa xmm7, xmmword ptr [rsp+0xA0]
+        vmovdqa xmm8, xmmword ptr [rsp+0xB0]
+        vmovdqa xmm9, xmmword ptr [rsp+0xC0]
+        vmovdqa xmm10, xmmword ptr [rsp+0xD0]
+        vmovdqa xmm11, xmmword ptr [rsp+0xE0]
+        vmovdqa xmm12, xmmword ptr [rsp+0xF0]
+        vmovdqa xmm13, xmmword ptr [rsp+0x100]
+        vmovdqa xmm14, xmmword ptr [rsp+0x110]
+        vmovdqa xmm15, xmmword ptr [rsp+0x120]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rsi
+        pop     rdi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align 6
+3:
+        test    esi, 0x8
+        je      3f
+        vpbroadcastd ymm0, dword ptr [rcx]
+        vpbroadcastd ymm1, dword ptr [rcx+0x4]
+        vpbroadcastd ymm2, dword ptr [rcx+0x8]
+        vpbroadcastd ymm3, dword ptr [rcx+0xC]
+        vpbroadcastd ymm4, dword ptr [rcx+0x10]
+        vpbroadcastd ymm5, dword ptr [rcx+0x14]
+        vpbroadcastd ymm6, dword ptr [rcx+0x18]
+        vpbroadcastd ymm7, dword ptr [rcx+0x1C]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     r12, qword ptr [rdi+0x20]
+        mov     r13, qword ptr [rdi+0x28]
+        mov     r14, qword ptr [rdi+0x30]
+        mov     r15, qword ptr [rdi+0x38]
+        movzx   eax, byte ptr [rbp+0x78]
+        movzx   ebx, byte ptr [rbp+0x80]
+        or      eax, ebx
+        xor     edx, edx
+2:
+        movzx   ebx, byte ptr [rbp+0x88]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+0x80]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+0x88], eax
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x40], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x40]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x40], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x40]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x40], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x40]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x40], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm16, ymm12, ymm14, 136
+        vshufps ymm17, ymm12, ymm14, 221
+        vshufps ymm18, ymm13, ymm15, 136
+        vshufps ymm19, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x30], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x30], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x30]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x30], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x30]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x30], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm20, ymm12, ymm14, 136
+        vshufps ymm21, ymm12, ymm14, 221
+        vshufps ymm22, ymm13, ymm15, 136
+        vshufps ymm23, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x20], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x20]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x20], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x20]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x20], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x20]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x20], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm24, ymm12, ymm14, 136
+        vshufps ymm25, ymm12, ymm14, 221
+        vshufps ymm26, ymm13, ymm15, 136
+        vshufps ymm27, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-0x10], 0x01
+        vmovups xmm9, xmmword ptr [r9+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-0x10], 0x01
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-0x10]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-0x10], 0x01
+        vmovups xmm11, xmmword ptr [r11+rdx-0x10]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-0x10], 0x01
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm28, ymm12, ymm14, 136
+        vshufps ymm29, ymm12, ymm14, 221
+        vshufps ymm30, ymm13, ymm15, 136
+        vshufps ymm31, ymm13, ymm15, 221
+        vpbroadcastd ymm8, dword ptr [BLAKE3_IV_0+rip]
+        vpbroadcastd ymm9, dword ptr [BLAKE3_IV_1+rip]
+        vpbroadcastd ymm10, dword ptr [BLAKE3_IV_2+rip]
+        vpbroadcastd ymm11, dword ptr [BLAKE3_IV_3+rip]
+        vmovdqa ymm12, ymmword ptr [rsp]
+        vmovdqa ymm13, ymmword ptr [rsp+0x40]
+        vpbroadcastd ymm14, dword ptr [BLAKE3_BLOCK_LEN+rip]
+        vpbroadcastd ymm15, dword ptr [rsp+0x88]
+        vpaddd  ymm0, ymm0, ymm16
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm20
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm17
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm21
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm24
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm28
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm25
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm29
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm18
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm23
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm22
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm16
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm17
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm25
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm27
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm30
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm19
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm29
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm20
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm18
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm22
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm27
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm21
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm31
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm26
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm30
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm23
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm19
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm20
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm21
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm16
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm24
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm28
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm31
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm29
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm26
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm23
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm16
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm18
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm17
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm25
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm24
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm30
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm28
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm29
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm18
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm19
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm22
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm27
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm17
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm31
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm25
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm30
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm19
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm26
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm20
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpxor   ymm0, ymm0, ymm8
+        vpxor   ymm1, ymm1, ymm9
+        vpxor   ymm2, ymm2, ymm10
+        vpxor   ymm3, ymm3, ymm11
+        vpxor   ymm4, ymm4, ymm12
+        vpxor   ymm5, ymm5, ymm13
+        vpxor   ymm6, ymm6, ymm14
+        vpxor   ymm7, ymm7, ymm15
+        movzx   eax, byte ptr [rbp+0x78]
+        jne     2b
+        mov     rbx, qword ptr [rbp+0x90]
+        vunpcklps ymm8, ymm0, ymm1
+        vunpcklps ymm9, ymm2, ymm3
+        vunpckhps ymm10, ymm0, ymm1
+        vunpcklps ymm11, ymm4, ymm5
+        vunpcklps ymm0, ymm6, ymm7
+        vshufps ymm12, ymm8, ymm9, 78
+        vblendps ymm1, ymm8, ymm12, 0xCC
+        vshufps ymm8, ymm11, ymm0, 78
+        vunpckhps ymm13, ymm2, ymm3
+        vblendps ymm2, ymm11, ymm8, 0xCC
+        vblendps ymm3, ymm12, ymm9, 0xCC
+        vperm2f128 ymm12, ymm1, ymm2, 0x20
+        vmovups ymmword ptr [rbx], ymm12
+        vunpckhps ymm14, ymm4, ymm5
+        vblendps ymm4, ymm8, ymm0, 0xCC
+        vunpckhps ymm15, ymm6, ymm7
+        vperm2f128 ymm7, ymm3, ymm4, 0x20
+        vmovups ymmword ptr [rbx+0x20], ymm7
+        vshufps ymm5, ymm10, ymm13, 78
+        vblendps ymm6, ymm5, ymm13, 0xCC
+        vshufps ymm13, ymm14, ymm15, 78
+        vblendps ymm10, ymm10, ymm5, 0xCC
+        vblendps ymm14, ymm14, ymm13, 0xCC
+        vperm2f128 ymm8, ymm10, ymm14, 0x20
+        vmovups ymmword ptr [rbx+0x40], ymm8
+        vblendps ymm15, ymm13, ymm15, 0xCC
+        vperm2f128 ymm13, ymm6, ymm15, 0x20
+        vmovups ymmword ptr [rbx+0x60], ymm13
+        vperm2f128 ymm9, ymm1, ymm2, 0x31
+        vperm2f128 ymm11, ymm3, ymm4, 0x31
+        vmovups ymmword ptr [rbx+0x80], ymm9
+        vperm2f128 ymm14, ymm10, ymm14, 0x31
+        vperm2f128 ymm15, ymm6, ymm15, 0x31
+        vmovups ymmword ptr [rbx+0xA0], ymm11
+        vmovups ymmword ptr [rbx+0xC0], ymm14
+        vmovups ymmword ptr [rbx+0xE0], ymm15
+        vmovdqa ymm0, ymmword ptr [rsp]
+        vmovdqa ymm2, ymmword ptr [rsp+0x40]
+        vmovdqa32 ymm0 {k1}, ymmword ptr [rsp+0x1*0x20]
+        vmovdqa32 ymm2 {k1}, ymmword ptr [rsp+0x3*0x20]
+        vmovdqa ymmword ptr [rsp], ymm0
+        vmovdqa ymmword ptr [rsp+0x40], ymm2
+        add     rbx, 256
+        mov     qword ptr [rbp+0x90], rbx
+        add     rdi, 64
+        sub     rsi, 8
+3:
+        mov     rbx, qword ptr [rbp+0x90]
+        mov     r15, qword ptr [rsp+0x80]
+        movzx   r13, byte ptr [rbp+0x78]
+        movzx   r12, byte ptr [rbp+0x88]
+        test    esi, 0x4
+        je      3f
+        vbroadcasti32x4 zmm0, xmmword ptr [rcx]
+        vbroadcasti32x4 zmm1, xmmword ptr [rcx+0x1*0x10]
+        vmovdqa xmm12, xmmword ptr [rsp]
+        vmovdqa xmm13, xmmword ptr [rsp+0x40]
+        vpunpckldq xmm14, xmm12, xmm13
+        vpunpckhdq xmm15, xmm12, xmm13
+        vpermq  ymm14, ymm14, 0xDC
+        vpermq  ymm15, ymm15, 0xDC
+        vpbroadcastd zmm12, dword ptr [BLAKE3_BLOCK_LEN+rip]
+        vinserti64x4 zmm13, zmm14, ymm15, 0x01
+        mov     eax, 17476
+        kmovw   k2, eax
+        vpblendmd zmm13 {k2}, zmm13, zmm12
+        vbroadcasti32x4 zmm15, xmmword ptr [BLAKE3_IV+rip]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        mov     eax, 43690
+        kmovw   k3, eax
+        mov     eax, 34952
+        kmovw   k4, eax
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x88], eax
+        vmovdqa32 zmm2, zmm15
+        vpbroadcastd zmm8, dword ptr [rsp+0x22*0x4]
+        vpblendmd zmm3 {k4}, zmm13, zmm8
+        vmovups zmm8, zmmword ptr [r8+rdx-0x1*0x40]
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r9+rdx-0x4*0x10], 0x01
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r10+rdx-0x4*0x10], 0x02
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r11+rdx-0x4*0x10], 0x03
+        vmovups zmm9, zmmword ptr [r8+rdx-0x30]
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r9+rdx-0x3*0x10], 0x01
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r10+rdx-0x3*0x10], 0x02
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r11+rdx-0x3*0x10], 0x03
+        vshufps zmm4, zmm8, zmm9, 136
+        vshufps zmm5, zmm8, zmm9, 221
+        vmovups zmm8, zmmword ptr [r8+rdx-0x20]
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r9+rdx-0x2*0x10], 0x01
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r10+rdx-0x2*0x10], 0x02
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r11+rdx-0x2*0x10], 0x03
+        vmovups zmm9, zmmword ptr [r8+rdx-0x10]
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r9+rdx-0x1*0x10], 0x01
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r10+rdx-0x1*0x10], 0x02
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r11+rdx-0x1*0x10], 0x03
+        vshufps zmm6, zmm8, zmm9, 136
+        vshufps zmm7, zmm8, zmm9, 221
+        vpshufd zmm6, zmm6, 0x93
+        vpshufd zmm7, zmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 16
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 12
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 8
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 7
+        vpshufd zmm0, zmm0, 0x93
+        vpshufd zmm3, zmm3, 0x4E
+        vpshufd zmm2, zmm2, 0x39
+        vpaddd  zmm0, zmm0, zmm6
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 16
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 12
+        vpaddd  zmm0, zmm0, zmm7
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 8
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 7
+        vpshufd zmm0, zmm0, 0x39
+        vpshufd zmm3, zmm3, 0x4E
+        vpshufd zmm2, zmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps zmm8, zmm4, zmm5, 214
+        vpshufd zmm9, zmm4, 0x0F
+        vpshufd zmm4, zmm8, 0x39
+        vshufps zmm8, zmm6, zmm7, 250
+        vpblendmd zmm9 {k3}, zmm9, zmm8
+        vpunpcklqdq zmm8, zmm7, zmm5
+        vpblendmd zmm8 {k4}, zmm8, zmm6
+        vpshufd zmm8, zmm8, 0x78
+        vpunpckhdq zmm5, zmm5, zmm7
+        vpunpckldq zmm6, zmm6, zmm5
+        vpshufd zmm7, zmm6, 0x1E
+        vmovdqa32 zmm5, zmm9
+        vmovdqa32 zmm6, zmm8
+        jmp     9b
+9:
+        vpxord  zmm0, zmm0, zmm2
+        vpxord  zmm1, zmm1, zmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vextracti32x4 xmmword ptr [rbx+0x4*0x10], zmm0, 0x02
+        vextracti32x4 xmmword ptr [rbx+0x5*0x10], zmm1, 0x02
+        vextracti32x4 xmmword ptr [rbx+0x6*0x10], zmm0, 0x03
+        vextracti32x4 xmmword ptr [rbx+0x7*0x10], zmm1, 0x03
+        vmovdqa xmm0, xmmword ptr [rsp]
+        vmovdqa xmm2, xmmword ptr [rsp+0x40]
+        vmovdqa32 xmm0 {k1}, xmmword ptr [rsp+0x1*0x10]
+        vmovdqa32 xmm2 {k1}, xmmword ptr [rsp+0x5*0x10]
+        vmovdqa xmmword ptr [rsp], xmm0
+        vmovdqa xmmword ptr [rsp+0x40], xmm2
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+3:
+        test    esi, 0x2
+        je      3f
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm13, dword ptr [rsp]
+        vpinsrd xmm13, xmm13, dword ptr [rsp+0x40], 1
+        vpinsrd xmm13, xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovd   xmm14, dword ptr [rsp+0x4]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+0x44], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vinserti128 ymm13, ymm13, xmm14, 0x01
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+0x88], eax
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV+rip]
+        vpbroadcastd ymm8, dword ptr [rsp+0x88]
+        vpblendd ymm3, ymm13, ymm8, 0x88
+        vmovups ymm8, ymmword ptr [r8+rdx-0x40]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x40], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x30]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x30], 0x01
+        vshufps ymm4, ymm8, ymm9, 136
+        vshufps ymm5, ymm8, ymm9, 221
+        vmovups ymm8, ymmword ptr [r8+rdx-0x20]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-0x20], 0x01
+        vmovups ymm9, ymmword ptr [r8+rdx-0x10]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-0x10], 0x01
+        vshufps ymm6, ymm8, ymm9, 136
+        vshufps ymm7, ymm8, ymm9, 221
+        vpshufd ymm6, ymm6, 0x93
+        vpshufd ymm7, ymm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 16
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 12
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 8
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 7
+        vpshufd ymm0, ymm0, 0x93
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x39
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 16
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 12
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 8
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 7
+        vpshufd ymm0, ymm0, 0x39
+        vpshufd ymm3, ymm3, 0x4E
+        vpshufd ymm2, ymm2, 0x93
+        dec     al
+        jz      9f
+        vshufps ymm8, ymm4, ymm5, 214
+        vpshufd ymm9, ymm4, 0x0F
+        vpshufd ymm4, ymm8, 0x39
+        vshufps ymm8, ymm6, ymm7, 250
+        vpblendd ymm9, ymm9, ymm8, 0xAA
+        vpunpcklqdq ymm8, ymm7, ymm5
+        vpblendd ymm8, ymm8, ymm6, 0x88
+        vpshufd ymm8, ymm8, 0x78
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 0x1E
+        vmovdqa ymm5, ymm9
+        vmovdqa ymm6, ymm8
+        jmp     9b
+9:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        vextracti128 xmmword ptr [rbx+0x20], ymm0, 0x01
+        vextracti128 xmmword ptr [rbx+0x30], ymm1, 0x01
+        vmovdqa xmm0, xmmword ptr [rsp]
+        vmovdqa xmm2, xmmword ptr [rsp+0x40]
+        vmovdqu32 xmm0 {k1}, xmmword ptr [rsp+0x8]
+        vmovdqu32 xmm2 {k1}, xmmword ptr [rsp+0x48]
+        vmovdqa xmmword ptr [rsp], xmm0
+        vmovdqa xmmword ptr [rsp+0x40], xmm2
+        add     rbx, 64
+        add     rdi, 16
+        sub     rsi, 2
+3:
+        test    esi, 0x1
+        je      4b
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+0x10]
+        vmovd   xmm14, dword ptr [rsp]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+0x40], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        vmovdqa xmm15, xmmword ptr [BLAKE3_IV+rip]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+.p2align 5
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        vpinsrd xmm3, xmm14, eax, 3
+        vmovdqa xmm2, xmm15
+        vmovups xmm8, xmmword ptr [r8+rdx-0x40]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x30]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-0x20]
+        vmovups xmm9, xmmword ptr [r8+rdx-0x10]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+
+
+.p2align 6
+_blake3_compress_in_place_avx512:
+blake3_compress_in_place_avx512:
+        sub     rsp, 72
+        vmovdqa xmmword ptr [rsp], xmm6
+        vmovdqa xmmword ptr [rsp+0x10], xmm7
+        vmovdqa xmmword ptr [rsp+0x20], xmm8
+        vmovdqa xmmword ptr [rsp+0x30], xmm9
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+0x10]
+        movzx   eax, byte ptr [rsp+0x70]
+        movzx   r8d, r8b
+        shl     rax, 32
+        add     r8, rax
+        vmovq   xmm3, r9
+        vmovq   xmm4, r8
+        vpunpcklqdq xmm3, xmm3, xmm4
+        vmovaps xmm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovups xmm8, xmmword ptr [rdx]
+        vmovups xmm9, xmmword ptr [rdx+0x10]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [rdx+0x20]
+        vmovups xmm9, xmmword ptr [rdx+0x30]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        vmovdqu xmmword ptr [rcx], xmm0
+        vmovdqu xmmword ptr [rcx+0x10], xmm1
+        vmovdqa xmm6, xmmword ptr [rsp]
+        vmovdqa xmm7, xmmword ptr [rsp+0x10]
+        vmovdqa xmm8, xmmword ptr [rsp+0x20]
+        vmovdqa xmm9, xmmword ptr [rsp+0x30]
+        add     rsp, 72
+        ret
+
+
+.p2align 6
+_blake3_compress_xof_avx512:
+blake3_compress_xof_avx512:
+        sub     rsp, 72
+        vmovdqa xmmword ptr [rsp], xmm6
+        vmovdqa xmmword ptr [rsp+0x10], xmm7
+        vmovdqa xmmword ptr [rsp+0x20], xmm8
+        vmovdqa xmmword ptr [rsp+0x30], xmm9
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+0x10]
+        movzx   eax, byte ptr [rsp+0x70]
+        movzx   r8d, r8b
+        mov     r10, qword ptr [rsp+0x78]
+        shl     rax, 32
+        add     r8, rax
+        vmovq   xmm3, r9
+        vmovq   xmm4, r8
+        vpunpcklqdq xmm3, xmm3, xmm4
+        vmovaps xmm2, xmmword ptr [BLAKE3_IV+rip]
+        vmovups xmm8, xmmword ptr [rdx]
+        vmovups xmm9, xmmword ptr [rdx+0x10]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [rdx+0x20]
+        vmovups xmm9, xmmword ptr [rdx+0x30]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 0x93
+        vpshufd xmm7, xmm7, 0x93
+        mov     al, 7
+9:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x93
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x39
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 0x39
+        vpshufd xmm3, xmm3, 0x4E
+        vpshufd xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0x0F
+        vpshufd xmm4, xmm8, 0x39
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0xAA
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 0x88
+        vpshufd xmm8, xmm8, 0x78
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 0x1E
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     9b
+9:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        vpxor   xmm2, xmm2, xmmword ptr [rcx]
+        vpxor   xmm3, xmm3, xmmword ptr [rcx+0x10]
+        vmovdqu xmmword ptr [r10], xmm0
+        vmovdqu xmmword ptr [r10+0x10], xmm1
+        vmovdqu xmmword ptr [r10+0x20], xmm2
+        vmovdqu xmmword ptr [r10+0x30], xmm3
+        vmovdqa xmm6, xmmword ptr [rsp]
+        vmovdqa xmm7, xmmword ptr [rsp+0x10]
+        vmovdqa xmm8, xmmword ptr [rsp+0x20]
+        vmovdqa xmm9, xmmword ptr [rsp+0x30]
+        add     rsp, 72
+        ret
+
+.section .rdata
+.p2align  6
+INDEX0:
+        .long    0,  1,  2,  3, 16, 17, 18, 19
+        .long    8,  9, 10, 11, 24, 25, 26, 27
+INDEX1:
+        .long    4,  5,  6,  7, 20, 21, 22, 23
+        .long   12, 13, 14, 15, 28, 29, 30, 31
+ADD0:
+        .long    0,  1,  2,  3,  4,  5,  6,  7
+        .long    8,  9, 10, 11, 12, 13, 14, 15
+ADD1:   .long    1
+
+ADD16:  .long   16
+BLAKE3_BLOCK_LEN:
+        .long   64
+.p2align 6
+BLAKE3_IV:
+BLAKE3_IV_0:
+        .long   0x6A09E667
+BLAKE3_IV_1:
+        .long   0xBB67AE85
+BLAKE3_IV_2:
+        .long   0x3C6EF372
+BLAKE3_IV_3:
+        .long   0xA54FF53A
diff --git a/thirdparty/blake3/c/blake3_avx512_x86-64_windows_msvc.asm b/thirdparty/blake3/c/blake3_avx512_x86-64_windows_msvc.asm
new file mode 100644
index 000000000..b19efbaae
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_avx512_x86-64_windows_msvc.asm
@@ -0,0 +1,2634 @@
+public _blake3_hash_many_avx512
+public blake3_hash_many_avx512
+public blake3_compress_in_place_avx512
+public _blake3_compress_in_place_avx512
+public blake3_compress_xof_avx512
+public _blake3_compress_xof_avx512
+
+_TEXT   SEGMENT ALIGN(16) 'CODE'
+
+ALIGN   16
+blake3_hash_many_avx512 PROC
+_blake3_hash_many_avx512 PROC
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rdi
+        push    rsi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 304
+        and     rsp, 0FFFFFFFFFFFFFFC0H
+        vmovdqa xmmword ptr [rsp+90H], xmm6
+        vmovdqa xmmword ptr [rsp+0A0H], xmm7
+        vmovdqa xmmword ptr [rsp+0B0H], xmm8
+        vmovdqa xmmword ptr [rsp+0C0H], xmm9
+        vmovdqa xmmword ptr [rsp+0D0H], xmm10
+        vmovdqa xmmword ptr [rsp+0E0H], xmm11
+        vmovdqa xmmword ptr [rsp+0F0H], xmm12
+        vmovdqa xmmword ptr [rsp+100H], xmm13
+        vmovdqa xmmword ptr [rsp+110H], xmm14
+        vmovdqa xmmword ptr [rsp+120H], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+68H]
+        movzx   r9, byte ptr [rbp+70H]
+        neg     r9
+        kmovw   k1, r9d
+        vmovd   xmm0, r8d
+        vpbroadcastd ymm0, xmm0
+        shr     r8, 32
+        vmovd   xmm1, r8d
+        vpbroadcastd ymm1, xmm1
+        vmovdqa ymm4, ymm1
+        vmovdqa ymm5, ymm1
+        vpaddd  ymm2, ymm0, ymmword ptr [ADD0]
+        vpaddd  ymm3, ymm0, ymmword ptr [ADD0+32]
+        vpcmpud k2, ymm2, ymm0, 1
+        vpcmpud k3, ymm3, ymm0, 1
+        ; XXX: ml64.exe does not currently understand the syntax. We use a workaround.
+        vpbroadcastd ymm6, dword ptr [ADD1]
+        vpaddd  ymm4 {k2}, ymm4, ymm6
+        vpaddd  ymm5 {k3}, ymm5, ymm6
+        ; vpaddd  ymm4 {k2}, ymm4, dword ptr [ADD1] {1to8}
+        ; vpaddd  ymm5 {k3}, ymm5, dword ptr [ADD1] {1to8}
+        knotw   k2, k1
+        vmovdqa32 ymm2 {k2}, ymm0
+        vmovdqa32 ymm3 {k2}, ymm0
+        vmovdqa32 ymm4 {k2}, ymm1
+        vmovdqa32 ymm5 {k2}, ymm1
+        vmovdqa ymmword ptr [rsp], ymm2
+        vmovdqa ymmword ptr [rsp+20H], ymm3
+        vmovdqa ymmword ptr [rsp+40H], ymm4
+        vmovdqa ymmword ptr [rsp+60H], ymm5
+        shl     rdx, 6
+        mov     qword ptr [rsp+80H], rdx
+        cmp     rsi, 16
+        jc      final15blocks
+outerloop16:
+        vpbroadcastd zmm0, dword ptr [rcx]
+        vpbroadcastd zmm1, dword ptr [rcx+1H*4H]
+        vpbroadcastd zmm2, dword ptr [rcx+2H*4H]
+        vpbroadcastd zmm3, dword ptr [rcx+3H*4H]
+        vpbroadcastd zmm4, dword ptr [rcx+4H*4H]
+        vpbroadcastd zmm5, dword ptr [rcx+5H*4H]
+        vpbroadcastd zmm6, dword ptr [rcx+6H*4H]
+        vpbroadcastd zmm7, dword ptr [rcx+7H*4H]
+        movzx   eax, byte ptr [rbp+78H]
+        movzx   ebx, byte ptr [rbp+80H]
+        or      eax, ebx
+        xor     edx, edx
+ALIGN   16
+innerloop16:
+        movzx   ebx, byte ptr [rbp+88H]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+80H]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+88H], eax
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        mov     r12, qword ptr [rdi+40H]
+        mov     r13, qword ptr [rdi+48H]
+        mov     r14, qword ptr [rdi+50H]
+        mov     r15, qword ptr [rdi+58H]
+        vmovdqu32 ymm16, ymmword ptr [rdx+r8-2H*20H]
+        vinserti64x4 zmm16, zmm16, ymmword ptr [rdx+r12-2H*20H], 01H
+        vmovdqu32 ymm17, ymmword ptr [rdx+r9-2H*20H]
+        vinserti64x4 zmm17, zmm17, ymmword ptr [rdx+r13-2H*20H], 01H
+        vpunpcklqdq zmm8, zmm16, zmm17
+        vpunpckhqdq zmm9, zmm16, zmm17
+        vmovdqu32 ymm18, ymmword ptr [rdx+r10-2H*20H]
+        vinserti64x4 zmm18, zmm18, ymmword ptr [rdx+r14-2H*20H], 01H
+        vmovdqu32 ymm19, ymmword ptr [rdx+r11-2H*20H]
+        vinserti64x4 zmm19, zmm19, ymmword ptr [rdx+r15-2H*20H], 01H
+        vpunpcklqdq zmm10, zmm18, zmm19
+        vpunpckhqdq zmm11, zmm18, zmm19
+        mov     r8, qword ptr [rdi+20H]
+        mov     r9, qword ptr [rdi+28H]
+        mov     r10, qword ptr [rdi+30H]
+        mov     r11, qword ptr [rdi+38H]
+        mov     r12, qword ptr [rdi+60H]
+        mov     r13, qword ptr [rdi+68H]
+        mov     r14, qword ptr [rdi+70H]
+        mov     r15, qword ptr [rdi+78H]
+        vmovdqu32 ymm16, ymmword ptr [rdx+r8-2H*20H]
+        vinserti64x4 zmm16, zmm16, ymmword ptr [rdx+r12-2H*20H], 01H
+        vmovdqu32 ymm17, ymmword ptr [rdx+r9-2H*20H]
+        vinserti64x4 zmm17, zmm17, ymmword ptr [rdx+r13-2H*20H], 01H
+        vpunpcklqdq zmm12, zmm16, zmm17
+        vpunpckhqdq zmm13, zmm16, zmm17
+        vmovdqu32 ymm18, ymmword ptr [rdx+r10-2H*20H]
+        vinserti64x4 zmm18, zmm18, ymmword ptr [rdx+r14-2H*20H], 01H
+        vmovdqu32 ymm19, ymmword ptr [rdx+r11-2H*20H]
+        vinserti64x4 zmm19, zmm19, ymmword ptr [rdx+r15-2H*20H], 01H
+        vpunpcklqdq zmm14, zmm18, zmm19
+        vpunpckhqdq zmm15, zmm18, zmm19
+        vmovdqa32 zmm27, zmmword ptr [INDEX0]
+        vmovdqa32 zmm31, zmmword ptr [INDEX1]
+        vshufps zmm16, zmm8, zmm10, 136
+        vshufps zmm17, zmm12, zmm14, 136
+        vmovdqa32 zmm20, zmm16
+        vpermt2d zmm16, zmm27, zmm17
+        vpermt2d zmm20, zmm31, zmm17
+        vshufps zmm17, zmm8, zmm10, 221
+        vshufps zmm30, zmm12, zmm14, 221
+        vmovdqa32 zmm21, zmm17
+        vpermt2d zmm17, zmm27, zmm30
+        vpermt2d zmm21, zmm31, zmm30
+        vshufps zmm18, zmm9, zmm11, 136
+        vshufps zmm8, zmm13, zmm15, 136
+        vmovdqa32 zmm22, zmm18
+        vpermt2d zmm18, zmm27, zmm8
+        vpermt2d zmm22, zmm31, zmm8
+        vshufps zmm19, zmm9, zmm11, 221
+        vshufps zmm8, zmm13, zmm15, 221
+        vmovdqa32 zmm23, zmm19
+        vpermt2d zmm19, zmm27, zmm8
+        vpermt2d zmm23, zmm31, zmm8
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        mov     r12, qword ptr [rdi+40H]
+        mov     r13, qword ptr [rdi+48H]
+        mov     r14, qword ptr [rdi+50H]
+        mov     r15, qword ptr [rdi+58H]
+        vmovdqu32 ymm24, ymmword ptr [r8+rdx-1H*20H]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r12+rdx-1H*20H], 01H
+        vmovdqu32 ymm25, ymmword ptr [r9+rdx-1H*20H]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r13+rdx-1H*20H], 01H
+        vpunpcklqdq zmm8, zmm24, zmm25
+        vpunpckhqdq zmm9, zmm24, zmm25
+        vmovdqu32 ymm24, ymmword ptr [r10+rdx-1H*20H]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r14+rdx-1H*20H], 01H
+        vmovdqu32 ymm25, ymmword ptr [r11+rdx-1H*20H]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r15+rdx-1H*20H], 01H
+        vpunpcklqdq zmm10, zmm24, zmm25
+        vpunpckhqdq zmm11, zmm24, zmm25
+        prefetcht0 byte ptr [r8+rdx+80H]
+        prefetcht0 byte ptr [r12+rdx+80H]
+        prefetcht0 byte ptr [r9+rdx+80H]
+        prefetcht0 byte ptr [r13+rdx+80H]
+        prefetcht0 byte ptr [r10+rdx+80H]
+        prefetcht0 byte ptr [r14+rdx+80H]
+        prefetcht0 byte ptr [r11+rdx+80H]
+        prefetcht0 byte ptr [r15+rdx+80H]
+        mov     r8, qword ptr [rdi+20H]
+        mov     r9, qword ptr [rdi+28H]
+        mov     r10, qword ptr [rdi+30H]
+        mov     r11, qword ptr [rdi+38H]
+        mov     r12, qword ptr [rdi+60H]
+        mov     r13, qword ptr [rdi+68H]
+        mov     r14, qword ptr [rdi+70H]
+        mov     r15, qword ptr [rdi+78H]
+        vmovdqu32 ymm24, ymmword ptr [r8+rdx-1H*20H]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r12+rdx-1H*20H], 01H
+        vmovdqu32 ymm25, ymmword ptr [r9+rdx-1H*20H]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r13+rdx-1H*20H], 01H
+        vpunpcklqdq zmm12, zmm24, zmm25
+        vpunpckhqdq zmm13, zmm24, zmm25
+        vmovdqu32 ymm24, ymmword ptr [r10+rdx-1H*20H]
+        vinserti64x4 zmm24, zmm24, ymmword ptr [r14+rdx-1H*20H], 01H
+        vmovdqu32 ymm25, ymmword ptr [r11+rdx-1H*20H]
+        vinserti64x4 zmm25, zmm25, ymmword ptr [r15+rdx-1H*20H], 01H
+        vpunpcklqdq zmm14, zmm24, zmm25
+        vpunpckhqdq zmm15, zmm24, zmm25
+        prefetcht0 byte  ptr [r8+rdx+80H]
+        prefetcht0 byte ptr [r12+rdx+80H]
+        prefetcht0 byte ptr [r9+rdx+80H]
+        prefetcht0 byte ptr [r13+rdx+80H]
+        prefetcht0 byte ptr [r10+rdx+80H]
+        prefetcht0 byte ptr [r14+rdx+80H]
+        prefetcht0 byte ptr [r11+rdx+80H]
+        prefetcht0 byte ptr [r15+rdx+80H]
+        vshufps zmm24, zmm8, zmm10, 136
+        vshufps zmm30, zmm12, zmm14, 136
+        vmovdqa32 zmm28, zmm24
+        vpermt2d zmm24, zmm27, zmm30
+        vpermt2d zmm28, zmm31, zmm30
+        vshufps zmm25, zmm8, zmm10, 221
+        vshufps zmm30, zmm12, zmm14, 221
+        vmovdqa32 zmm29, zmm25
+        vpermt2d zmm25, zmm27, zmm30
+        vpermt2d zmm29, zmm31, zmm30
+        vshufps zmm26, zmm9, zmm11, 136
+        vshufps zmm8, zmm13, zmm15, 136
+        vmovdqa32 zmm30, zmm26
+        vpermt2d zmm26, zmm27, zmm8
+        vpermt2d zmm30, zmm31, zmm8
+        vshufps zmm8, zmm9, zmm11, 221
+        vshufps zmm10, zmm13, zmm15, 221
+        vpermi2d zmm27, zmm8, zmm10
+        vpermi2d zmm31, zmm8, zmm10
+        vpbroadcastd zmm8, dword ptr [BLAKE3_IV_0]
+        vpbroadcastd zmm9, dword ptr [BLAKE3_IV_1]
+        vpbroadcastd zmm10, dword ptr [BLAKE3_IV_2]
+        vpbroadcastd zmm11, dword ptr [BLAKE3_IV_3]
+        vmovdqa32 zmm12, zmmword ptr [rsp]
+        vmovdqa32 zmm13, zmmword ptr [rsp+1H*40H]
+        vpbroadcastd zmm14, dword ptr [BLAKE3_BLOCK_LEN]
+        vpbroadcastd zmm15, dword ptr [rsp+22H*4H]
+        vpaddd  zmm0, zmm0, zmm16
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm20
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm17
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm21
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm24
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm28
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm25
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm29
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm18
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm23
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm22
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm16
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm17
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm25
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm27
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm30
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm19
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm29
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm20
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm18
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm22
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm27
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm21
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm31
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm26
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm30
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm23
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm19
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm20
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm21
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm16
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm24
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm28
+        vpaddd  zmm1, zmm1, zmm25
+        vpaddd  zmm2, zmm2, zmm31
+        vpaddd  zmm3, zmm3, zmm30
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm29
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm26
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm23
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm16
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm18
+        vpaddd  zmm1, zmm1, zmm19
+        vpaddd  zmm2, zmm2, zmm17
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm25
+        vpaddd  zmm1, zmm1, zmm27
+        vpaddd  zmm2, zmm2, zmm24
+        vpaddd  zmm3, zmm3, zmm31
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm30
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm28
+        vpaddd  zmm3, zmm3, zmm17
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm29
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm18
+        vpaddd  zmm3, zmm3, zmm20
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm19
+        vpaddd  zmm1, zmm1, zmm26
+        vpaddd  zmm2, zmm2, zmm22
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpaddd  zmm0, zmm0, zmm27
+        vpaddd  zmm1, zmm1, zmm21
+        vpaddd  zmm2, zmm2, zmm17
+        vpaddd  zmm3, zmm3, zmm24
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vprord  zmm15, zmm15, 16
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 12
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vpaddd  zmm0, zmm0, zmm31
+        vpaddd  zmm1, zmm1, zmm16
+        vpaddd  zmm2, zmm2, zmm25
+        vpaddd  zmm3, zmm3, zmm22
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm1, zmm1, zmm5
+        vpaddd  zmm2, zmm2, zmm6
+        vpaddd  zmm3, zmm3, zmm7
+        vpxord  zmm12, zmm12, zmm0
+        vpxord  zmm13, zmm13, zmm1
+        vpxord  zmm14, zmm14, zmm2
+        vpxord  zmm15, zmm15, zmm3
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vprord  zmm15, zmm15, 8
+        vpaddd  zmm8, zmm8, zmm12
+        vpaddd  zmm9, zmm9, zmm13
+        vpaddd  zmm10, zmm10, zmm14
+        vpaddd  zmm11, zmm11, zmm15
+        vpxord  zmm4, zmm4, zmm8
+        vpxord  zmm5, zmm5, zmm9
+        vpxord  zmm6, zmm6, zmm10
+        vpxord  zmm7, zmm7, zmm11
+        vprord  zmm4, zmm4, 7
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vpaddd  zmm0, zmm0, zmm30
+        vpaddd  zmm1, zmm1, zmm18
+        vpaddd  zmm2, zmm2, zmm19
+        vpaddd  zmm3, zmm3, zmm23
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 16
+        vprord  zmm12, zmm12, 16
+        vprord  zmm13, zmm13, 16
+        vprord  zmm14, zmm14, 16
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 12
+        vprord  zmm6, zmm6, 12
+        vprord  zmm7, zmm7, 12
+        vprord  zmm4, zmm4, 12
+        vpaddd  zmm0, zmm0, zmm26
+        vpaddd  zmm1, zmm1, zmm28
+        vpaddd  zmm2, zmm2, zmm20
+        vpaddd  zmm3, zmm3, zmm29
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm1, zmm1, zmm6
+        vpaddd  zmm2, zmm2, zmm7
+        vpaddd  zmm3, zmm3, zmm4
+        vpxord  zmm15, zmm15, zmm0
+        vpxord  zmm12, zmm12, zmm1
+        vpxord  zmm13, zmm13, zmm2
+        vpxord  zmm14, zmm14, zmm3
+        vprord  zmm15, zmm15, 8
+        vprord  zmm12, zmm12, 8
+        vprord  zmm13, zmm13, 8
+        vprord  zmm14, zmm14, 8
+        vpaddd  zmm10, zmm10, zmm15
+        vpaddd  zmm11, zmm11, zmm12
+        vpaddd  zmm8, zmm8, zmm13
+        vpaddd  zmm9, zmm9, zmm14
+        vpxord  zmm5, zmm5, zmm10
+        vpxord  zmm6, zmm6, zmm11
+        vpxord  zmm7, zmm7, zmm8
+        vpxord  zmm4, zmm4, zmm9
+        vprord  zmm5, zmm5, 7
+        vprord  zmm6, zmm6, 7
+        vprord  zmm7, zmm7, 7
+        vprord  zmm4, zmm4, 7
+        vpxord  zmm0, zmm0, zmm8
+        vpxord  zmm1, zmm1, zmm9
+        vpxord  zmm2, zmm2, zmm10
+        vpxord  zmm3, zmm3, zmm11
+        vpxord  zmm4, zmm4, zmm12
+        vpxord  zmm5, zmm5, zmm13
+        vpxord  zmm6, zmm6, zmm14
+        vpxord  zmm7, zmm7, zmm15
+        movzx   eax, byte ptr [rbp+78H]
+        jne     innerloop16
+        mov     rbx, qword ptr [rbp+90H]
+        vpunpckldq zmm16, zmm0, zmm1
+        vpunpckhdq zmm17, zmm0, zmm1
+        vpunpckldq zmm18, zmm2, zmm3
+        vpunpckhdq zmm19, zmm2, zmm3
+        vpunpckldq zmm20, zmm4, zmm5
+        vpunpckhdq zmm21, zmm4, zmm5
+        vpunpckldq zmm22, zmm6, zmm7
+        vpunpckhdq zmm23, zmm6, zmm7
+        vpunpcklqdq zmm0, zmm16, zmm18
+        vpunpckhqdq zmm1, zmm16, zmm18
+        vpunpcklqdq zmm2, zmm17, zmm19
+        vpunpckhqdq zmm3, zmm17, zmm19
+        vpunpcklqdq zmm4, zmm20, zmm22
+        vpunpckhqdq zmm5, zmm20, zmm22
+        vpunpcklqdq zmm6, zmm21, zmm23
+        vpunpckhqdq zmm7, zmm21, zmm23
+        vshufi32x4 zmm16, zmm0, zmm4, 88H
+        vshufi32x4 zmm17, zmm1, zmm5, 88H
+        vshufi32x4 zmm18, zmm2, zmm6, 88H
+        vshufi32x4 zmm19, zmm3, zmm7, 88H
+        vshufi32x4 zmm20, zmm0, zmm4, 0DDH
+        vshufi32x4 zmm21, zmm1, zmm5, 0DDH
+        vshufi32x4 zmm22, zmm2, zmm6, 0DDH
+        vshufi32x4 zmm23, zmm3, zmm7, 0DDH
+        vshufi32x4 zmm0, zmm16, zmm17, 88H
+        vshufi32x4 zmm1, zmm18, zmm19, 88H
+        vshufi32x4 zmm2, zmm20, zmm21, 88H
+        vshufi32x4 zmm3, zmm22, zmm23, 88H
+        vshufi32x4 zmm4, zmm16, zmm17, 0DDH
+        vshufi32x4 zmm5, zmm18, zmm19, 0DDH
+        vshufi32x4 zmm6, zmm20, zmm21, 0DDH
+        vshufi32x4 zmm7, zmm22, zmm23, 0DDH
+        vmovdqu32 zmmword ptr [rbx], zmm0
+        vmovdqu32 zmmword ptr [rbx+1H*40H], zmm1
+        vmovdqu32 zmmword ptr [rbx+2H*40H], zmm2
+        vmovdqu32 zmmword ptr [rbx+3H*40H], zmm3
+        vmovdqu32 zmmword ptr [rbx+4H*40H], zmm4
+        vmovdqu32 zmmword ptr [rbx+5H*40H], zmm5
+        vmovdqu32 zmmword ptr [rbx+6H*40H], zmm6
+        vmovdqu32 zmmword ptr [rbx+7H*40H], zmm7
+        vmovdqa32 zmm0, zmmword ptr [rsp]
+        vmovdqa32 zmm1, zmmword ptr [rsp+1H*40H]
+        vmovdqa32 zmm2, zmm0
+        ; XXX: ml64.exe does not currently understand the syntax. We use a workaround.
+        vpbroadcastd zmm4, dword ptr [ADD16]
+        vpbroadcastd zmm5, dword ptr [ADD1]
+        vpaddd  zmm2{k1}, zmm0, zmm4
+        ; vpaddd  zmm2{k1}, zmm0, dword ptr [ADD16] ; {1to16}
+        vpcmpud k2, zmm2, zmm0, 1
+        vpaddd  zmm1 {k2}, zmm1, zmm5
+        ; vpaddd  zmm1 {k2}, zmm1, dword ptr [ADD1] ; {1to16}
+        vmovdqa32 zmmword ptr [rsp], zmm2
+        vmovdqa32 zmmword ptr [rsp+1H*40H], zmm1
+        add     rdi, 128
+        add     rbx, 512
+        mov     qword ptr [rbp+90H], rbx
+        sub     rsi, 16
+        cmp     rsi, 16
+        jnc     outerloop16
+        test    rsi, rsi
+        jne     final15blocks
+unwind:
+        vzeroupper
+        vmovdqa xmm6, xmmword ptr [rsp+90H]
+        vmovdqa xmm7, xmmword ptr [rsp+0A0H]
+        vmovdqa xmm8, xmmword ptr [rsp+0B0H]
+        vmovdqa xmm9, xmmword ptr [rsp+0C0H]
+        vmovdqa xmm10, xmmword ptr [rsp+0D0H]
+        vmovdqa xmm11, xmmword ptr [rsp+0E0H]
+        vmovdqa xmm12, xmmword ptr [rsp+0F0H]
+        vmovdqa xmm13, xmmword ptr [rsp+100H]
+        vmovdqa xmm14, xmmword ptr [rsp+110H]
+        vmovdqa xmm15, xmmword ptr [rsp+120H]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rsi
+        pop     rdi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+ALIGN   16
+final15blocks:
+        test    esi, 8H
+        je      final7blocks
+        vpbroadcastd ymm0, dword ptr [rcx]
+        vpbroadcastd ymm1, dword ptr [rcx+4H]
+        vpbroadcastd ymm2, dword ptr [rcx+8H]
+        vpbroadcastd ymm3, dword ptr [rcx+0CH]
+        vpbroadcastd ymm4, dword ptr [rcx+10H]
+        vpbroadcastd ymm5, dword ptr [rcx+14H]
+        vpbroadcastd ymm6, dword ptr [rcx+18H]
+        vpbroadcastd ymm7, dword ptr [rcx+1CH]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        mov     r12, qword ptr [rdi+20H]
+        mov     r13, qword ptr [rdi+28H]
+        mov     r14, qword ptr [rdi+30H]
+        mov     r15, qword ptr [rdi+38H]
+        movzx   eax, byte ptr [rbp+78H]
+        movzx   ebx, byte ptr [rbp+80H]
+        or      eax, ebx
+        xor     edx, edx
+innerloop8:
+        movzx   ebx, byte ptr [rbp+88H]
+        or      ebx, eax
+        add     rdx, 64
+        cmp     rdx, qword ptr [rsp+80H]
+        cmove   eax, ebx
+        mov     dword ptr [rsp+88H], eax
+        vmovups xmm8, xmmword ptr [r8+rdx-40H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-40H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-40H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-40H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-40H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-40H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-40H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-40H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm16, ymm12, ymm14, 136
+        vshufps ymm17, ymm12, ymm14, 221
+        vshufps ymm18, ymm13, ymm15, 136
+        vshufps ymm19, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-30H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-30H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-30H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-30H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-30H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-30H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-30H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-30H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm20, ymm12, ymm14, 136
+        vshufps ymm21, ymm12, ymm14, 221
+        vshufps ymm22, ymm13, ymm15, 136
+        vshufps ymm23, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-20H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-20H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-20H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-20H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-20H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-20H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-20H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-20H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm24, ymm12, ymm14, 136
+        vshufps ymm25, ymm12, ymm14, 221
+        vshufps ymm26, ymm13, ymm15, 136
+        vshufps ymm27, ymm13, ymm15, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-10H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r12+rdx-10H], 01H
+        vmovups xmm9, xmmword ptr [r9+rdx-10H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r13+rdx-10H], 01H
+        vunpcklpd ymm12, ymm8, ymm9
+        vunpckhpd ymm13, ymm8, ymm9
+        vmovups xmm10, xmmword ptr [r10+rdx-10H]
+        vinsertf128 ymm10, ymm10, xmmword ptr [r14+rdx-10H], 01H
+        vmovups xmm11, xmmword ptr [r11+rdx-10H]
+        vinsertf128 ymm11, ymm11, xmmword ptr [r15+rdx-10H], 01H
+        vunpcklpd ymm14, ymm10, ymm11
+        vunpckhpd ymm15, ymm10, ymm11
+        vshufps ymm28, ymm12, ymm14, 136
+        vshufps ymm29, ymm12, ymm14, 221
+        vshufps ymm30, ymm13, ymm15, 136
+        vshufps ymm31, ymm13, ymm15, 221
+        vpbroadcastd ymm8, dword ptr [BLAKE3_IV_0]
+        vpbroadcastd ymm9, dword ptr [BLAKE3_IV_1]
+        vpbroadcastd ymm10, dword ptr [BLAKE3_IV_2]
+        vpbroadcastd ymm11, dword ptr [BLAKE3_IV_3]
+        vmovdqa ymm12, ymmword ptr [rsp]
+        vmovdqa ymm13, ymmword ptr [rsp+40H]
+        vpbroadcastd ymm14, dword ptr [BLAKE3_BLOCK_LEN]
+        vpbroadcastd ymm15, dword ptr [rsp+88H]
+        vpaddd  ymm0, ymm0, ymm16
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm20
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm17
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm21
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm24
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm28
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm25
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm29
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm18
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm23
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm22
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm16
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm17
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm25
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm27
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm30
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm19
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm29
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm20
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm18
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm22
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm27
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm21
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm31
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm26
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm30
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm23
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm19
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm20
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm21
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm16
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm24
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm28
+        vpaddd  ymm1, ymm1, ymm25
+        vpaddd  ymm2, ymm2, ymm31
+        vpaddd  ymm3, ymm3, ymm30
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm29
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm26
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm23
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm16
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm18
+        vpaddd  ymm1, ymm1, ymm19
+        vpaddd  ymm2, ymm2, ymm17
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm25
+        vpaddd  ymm1, ymm1, ymm27
+        vpaddd  ymm2, ymm2, ymm24
+        vpaddd  ymm3, ymm3, ymm31
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm30
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm28
+        vpaddd  ymm3, ymm3, ymm17
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm29
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm18
+        vpaddd  ymm3, ymm3, ymm20
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm19
+        vpaddd  ymm1, ymm1, ymm26
+        vpaddd  ymm2, ymm2, ymm22
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpaddd  ymm0, ymm0, ymm27
+        vpaddd  ymm1, ymm1, ymm21
+        vpaddd  ymm2, ymm2, ymm17
+        vpaddd  ymm3, ymm3, ymm24
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vprord  ymm15, ymm15, 16
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 12
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vpaddd  ymm0, ymm0, ymm31
+        vpaddd  ymm1, ymm1, ymm16
+        vpaddd  ymm2, ymm2, ymm25
+        vpaddd  ymm3, ymm3, ymm22
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm1, ymm1, ymm5
+        vpaddd  ymm2, ymm2, ymm6
+        vpaddd  ymm3, ymm3, ymm7
+        vpxord  ymm12, ymm12, ymm0
+        vpxord  ymm13, ymm13, ymm1
+        vpxord  ymm14, ymm14, ymm2
+        vpxord  ymm15, ymm15, ymm3
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vprord  ymm15, ymm15, 8
+        vpaddd  ymm8, ymm8, ymm12
+        vpaddd  ymm9, ymm9, ymm13
+        vpaddd  ymm10, ymm10, ymm14
+        vpaddd  ymm11, ymm11, ymm15
+        vpxord  ymm4, ymm4, ymm8
+        vpxord  ymm5, ymm5, ymm9
+        vpxord  ymm6, ymm6, ymm10
+        vpxord  ymm7, ymm7, ymm11
+        vprord  ymm4, ymm4, 7
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vpaddd  ymm0, ymm0, ymm30
+        vpaddd  ymm1, ymm1, ymm18
+        vpaddd  ymm2, ymm2, ymm19
+        vpaddd  ymm3, ymm3, ymm23
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 16
+        vprord  ymm12, ymm12, 16
+        vprord  ymm13, ymm13, 16
+        vprord  ymm14, ymm14, 16
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 12
+        vprord  ymm6, ymm6, 12
+        vprord  ymm7, ymm7, 12
+        vprord  ymm4, ymm4, 12
+        vpaddd  ymm0, ymm0, ymm26
+        vpaddd  ymm1, ymm1, ymm28
+        vpaddd  ymm2, ymm2, ymm20
+        vpaddd  ymm3, ymm3, ymm29
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm1, ymm1, ymm6
+        vpaddd  ymm2, ymm2, ymm7
+        vpaddd  ymm3, ymm3, ymm4
+        vpxord  ymm15, ymm15, ymm0
+        vpxord  ymm12, ymm12, ymm1
+        vpxord  ymm13, ymm13, ymm2
+        vpxord  ymm14, ymm14, ymm3
+        vprord  ymm15, ymm15, 8
+        vprord  ymm12, ymm12, 8
+        vprord  ymm13, ymm13, 8
+        vprord  ymm14, ymm14, 8
+        vpaddd  ymm10, ymm10, ymm15
+        vpaddd  ymm11, ymm11, ymm12
+        vpaddd  ymm8, ymm8, ymm13
+        vpaddd  ymm9, ymm9, ymm14
+        vpxord  ymm5, ymm5, ymm10
+        vpxord  ymm6, ymm6, ymm11
+        vpxord  ymm7, ymm7, ymm8
+        vpxord  ymm4, ymm4, ymm9
+        vprord  ymm5, ymm5, 7
+        vprord  ymm6, ymm6, 7
+        vprord  ymm7, ymm7, 7
+        vprord  ymm4, ymm4, 7
+        vpxor   ymm0, ymm0, ymm8
+        vpxor   ymm1, ymm1, ymm9
+        vpxor   ymm2, ymm2, ymm10
+        vpxor   ymm3, ymm3, ymm11
+        vpxor   ymm4, ymm4, ymm12
+        vpxor   ymm5, ymm5, ymm13
+        vpxor   ymm6, ymm6, ymm14
+        vpxor   ymm7, ymm7, ymm15
+        movzx   eax, byte ptr [rbp+78H]
+        jne     innerloop8
+        mov     rbx, qword ptr [rbp+90H]
+        vunpcklps ymm8, ymm0, ymm1
+        vunpcklps ymm9, ymm2, ymm3
+        vunpckhps ymm10, ymm0, ymm1
+        vunpcklps ymm11, ymm4, ymm5
+        vunpcklps ymm0, ymm6, ymm7
+        vshufps ymm12, ymm8, ymm9, 78
+        vblendps ymm1, ymm8, ymm12, 0CCH
+        vshufps ymm8, ymm11, ymm0, 78
+        vunpckhps ymm13, ymm2, ymm3
+        vblendps ymm2, ymm11, ymm8, 0CCH
+        vblendps ymm3, ymm12, ymm9, 0CCH
+        vperm2f128 ymm12, ymm1, ymm2, 20H
+        vmovups ymmword ptr [rbx], ymm12
+        vunpckhps ymm14, ymm4, ymm5
+        vblendps ymm4, ymm8, ymm0, 0CCH
+        vunpckhps ymm15, ymm6, ymm7
+        vperm2f128 ymm7, ymm3, ymm4, 20H
+        vmovups ymmword ptr [rbx+20H], ymm7
+        vshufps ymm5, ymm10, ymm13, 78
+        vblendps ymm6, ymm5, ymm13, 0CCH
+        vshufps ymm13, ymm14, ymm15, 78
+        vblendps ymm10, ymm10, ymm5, 0CCH
+        vblendps ymm14, ymm14, ymm13, 0CCH
+        vperm2f128 ymm8, ymm10, ymm14, 20H
+        vmovups ymmword ptr [rbx+40H], ymm8
+        vblendps ymm15, ymm13, ymm15, 0CCH
+        vperm2f128 ymm13, ymm6, ymm15, 20H
+        vmovups ymmword ptr [rbx+60H], ymm13
+        vperm2f128 ymm9, ymm1, ymm2, 31H
+        vperm2f128 ymm11, ymm3, ymm4, 31H
+        vmovups ymmword ptr [rbx+80H], ymm9
+        vperm2f128 ymm14, ymm10, ymm14, 31H
+        vperm2f128 ymm15, ymm6, ymm15, 31H
+        vmovups ymmword ptr [rbx+0A0H], ymm11
+        vmovups ymmword ptr [rbx+0C0H], ymm14
+        vmovups ymmword ptr [rbx+0E0H], ymm15
+        vmovdqa ymm0, ymmword ptr [rsp]
+        vmovdqa ymm2, ymmword ptr [rsp+40H]
+        vmovdqa32 ymm0 {k1}, ymmword ptr [rsp+1H*20H]
+        vmovdqa32 ymm2 {k1}, ymmword ptr [rsp+3H*20H]
+        vmovdqa ymmword ptr [rsp], ymm0
+        vmovdqa ymmword ptr [rsp+40H], ymm2
+        add     rbx, 256
+        mov     qword ptr [rbp+90H], rbx
+        add     rdi, 64
+        sub     rsi, 8
+final7blocks:
+        mov     rbx, qword ptr [rbp+90H]
+        mov     r15, qword ptr [rsp+80H]
+        movzx   r13, byte ptr [rbp+78H]
+        movzx   r12, byte ptr [rbp+88H]
+        test    esi, 4H
+        je      final3blocks
+        vbroadcasti32x4 zmm0, xmmword ptr [rcx]
+        vbroadcasti32x4 zmm1, xmmword ptr [rcx+1H*10H]
+        vmovdqa xmm12, xmmword ptr [rsp]
+        vmovdqa xmm13, xmmword ptr [rsp+40H]
+        vpunpckldq xmm14, xmm12, xmm13
+        vpunpckhdq xmm15, xmm12, xmm13
+        vpermq  ymm14, ymm14, 0DCH
+        vpermq  ymm15, ymm15, 0DCH
+        vpbroadcastd zmm12, dword ptr [BLAKE3_BLOCK_LEN]
+        vinserti64x4 zmm13, zmm14, ymm15, 01H
+        mov     eax, 17476
+        kmovw   k2, eax
+        vpblendmd zmm13 {k2}, zmm13, zmm12
+        vbroadcasti32x4 zmm15, xmmword ptr [BLAKE3_IV]
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        mov     eax, 43690
+        kmovw   k3, eax
+        mov     eax, 34952
+        kmovw   k4, eax
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+ALIGN   16
+innerloop4:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+88H], eax
+        vmovdqa32 zmm2, zmm15
+        vpbroadcastd zmm8, dword ptr [rsp+22H*4H]
+        vpblendmd zmm3 {k4}, zmm13, zmm8
+        vmovups zmm8, zmmword ptr [r8+rdx-1H*40H]
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r9+rdx-4H*10H], 01H
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r10+rdx-4H*10H], 02H
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r11+rdx-4H*10H], 03H
+        vmovups zmm9, zmmword ptr [r8+rdx-30H]
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r9+rdx-3H*10H], 01H
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r10+rdx-3H*10H], 02H
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r11+rdx-3H*10H], 03H
+        vshufps zmm4, zmm8, zmm9, 136
+        vshufps zmm5, zmm8, zmm9, 221
+        vmovups zmm8, zmmword ptr [r8+rdx-20H]
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r9+rdx-2H*10H], 01H
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r10+rdx-2H*10H], 02H
+        vinserti32x4 zmm8, zmm8, xmmword ptr [r11+rdx-2H*10H], 03H
+        vmovups zmm9, zmmword ptr [r8+rdx-10H]
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r9+rdx-1H*10H], 01H
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r10+rdx-1H*10H], 02H
+        vinserti32x4 zmm9, zmm9, xmmword ptr [r11+rdx-1H*10H], 03H
+        vshufps zmm6, zmm8, zmm9, 136
+        vshufps zmm7, zmm8, zmm9, 221
+        vpshufd zmm6, zmm6, 93H
+        vpshufd zmm7, zmm7, 93H
+        mov     al, 7
+roundloop4:
+        vpaddd  zmm0, zmm0, zmm4
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 16
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 12
+        vpaddd  zmm0, zmm0, zmm5
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 8
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 7
+        vpshufd zmm0, zmm0, 93H
+        vpshufd zmm3, zmm3, 4EH
+        vpshufd zmm2, zmm2, 39H
+        vpaddd  zmm0, zmm0, zmm6
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 16
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 12
+        vpaddd  zmm0, zmm0, zmm7
+        vpaddd  zmm0, zmm0, zmm1
+        vpxord  zmm3, zmm3, zmm0
+        vprord  zmm3, zmm3, 8
+        vpaddd  zmm2, zmm2, zmm3
+        vpxord  zmm1, zmm1, zmm2
+        vprord  zmm1, zmm1, 7
+        vpshufd zmm0, zmm0, 39H
+        vpshufd zmm3, zmm3, 4EH
+        vpshufd zmm2, zmm2, 93H
+        dec     al
+        jz      endroundloop4
+        vshufps zmm8, zmm4, zmm5, 214
+        vpshufd zmm9, zmm4, 0FH
+        vpshufd zmm4, zmm8, 39H
+        vshufps zmm8, zmm6, zmm7, 250
+        vpblendmd zmm9 {k3}, zmm9, zmm8
+        vpunpcklqdq zmm8, zmm7, zmm5
+        vpblendmd zmm8 {k4}, zmm8, zmm6
+        vpshufd zmm8, zmm8, 78H
+        vpunpckhdq zmm5, zmm5, zmm7
+        vpunpckldq zmm6, zmm6, zmm5
+        vpshufd zmm7, zmm6, 1EH
+        vmovdqa32 zmm5, zmm9
+        vmovdqa32 zmm6, zmm8
+        jmp     roundloop4
+endroundloop4:
+        vpxord  zmm0, zmm0, zmm2
+        vpxord  zmm1, zmm1, zmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop4
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+10H], xmm1
+        vextracti128 xmmword ptr [rbx+20H], ymm0, 01H
+        vextracti128 xmmword ptr [rbx+30H], ymm1, 01H
+        vextracti32x4 xmmword ptr [rbx+4H*10H], zmm0, 02H
+        vextracti32x4 xmmword ptr [rbx+5H*10H], zmm1, 02H
+        vextracti32x4 xmmword ptr [rbx+6H*10H], zmm0, 03H
+        vextracti32x4 xmmword ptr [rbx+7H*10H], zmm1, 03H
+        vmovdqa xmm0, xmmword ptr [rsp]
+        vmovdqa xmm2, xmmword ptr [rsp+40H]
+        vmovdqa32 xmm0 {k1}, xmmword ptr [rsp+1H*10H]
+        vmovdqa32 xmm2 {k1}, xmmword ptr [rsp+5H*10H]
+        vmovdqa xmmword ptr [rsp], xmm0
+        vmovdqa xmmword ptr [rsp+40H], xmm2
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+final3blocks:
+        test    esi, 2H
+        je      final1block
+        vbroadcasti128 ymm0, xmmword ptr [rcx]
+        vbroadcasti128 ymm1, xmmword ptr [rcx+10H]
+        vmovd   xmm13, dword ptr [rsp]
+        vpinsrd xmm13, xmm13, dword ptr [rsp+40H], 1
+        vpinsrd xmm13, xmm13, dword ptr [BLAKE3_BLOCK_LEN], 2
+        vmovd   xmm14, dword ptr [rsp+4H]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+44H], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN], 2
+        vinserti128 ymm13, ymm13, xmm14, 01H
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+ALIGN   16
+innerloop2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        mov     dword ptr [rsp+88H], eax
+        vbroadcasti128 ymm2, xmmword ptr [BLAKE3_IV]
+        vpbroadcastd ymm8, dword ptr [rsp+88H]
+        vpblendd ymm3, ymm13, ymm8, 88H
+        vmovups ymm8, ymmword ptr [r8+rdx-40H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-40H], 01H
+        vmovups ymm9, ymmword ptr [r8+rdx-30H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-30H], 01H
+        vshufps ymm4, ymm8, ymm9, 136
+        vshufps ymm5, ymm8, ymm9, 221
+        vmovups ymm8, ymmword ptr [r8+rdx-20H]
+        vinsertf128 ymm8, ymm8, xmmword ptr [r9+rdx-20H], 01H
+        vmovups ymm9, ymmword ptr [r8+rdx-10H]
+        vinsertf128 ymm9, ymm9, xmmword ptr [r9+rdx-10H], 01H
+        vshufps ymm6, ymm8, ymm9, 136
+        vshufps ymm7, ymm8, ymm9, 221
+        vpshufd ymm6, ymm6, 93H
+        vpshufd ymm7, ymm7, 93H
+        mov     al, 7
+roundloop2:
+        vpaddd  ymm0, ymm0, ymm4
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 16
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 12
+        vpaddd  ymm0, ymm0, ymm5
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 8
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 7
+        vpshufd ymm0, ymm0, 93H
+        vpshufd ymm3, ymm3, 4EH
+        vpshufd ymm2, ymm2, 39H
+        vpaddd  ymm0, ymm0, ymm6
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 16
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 12
+        vpaddd  ymm0, ymm0, ymm7
+        vpaddd  ymm0, ymm0, ymm1
+        vpxord  ymm3, ymm3, ymm0
+        vprord  ymm3, ymm3, 8
+        vpaddd  ymm2, ymm2, ymm3
+        vpxord  ymm1, ymm1, ymm2
+        vprord  ymm1, ymm1, 7
+        vpshufd ymm0, ymm0, 39H
+        vpshufd ymm3, ymm3, 4EH
+        vpshufd ymm2, ymm2, 93H
+        dec     al
+        jz      endroundloop2
+        vshufps ymm8, ymm4, ymm5, 214
+        vpshufd ymm9, ymm4, 0FH
+        vpshufd ymm4, ymm8, 39H
+        vshufps ymm8, ymm6, ymm7, 250
+        vpblendd ymm9, ymm9, ymm8, 0AAH
+        vpunpcklqdq ymm8, ymm7, ymm5
+        vpblendd ymm8, ymm8, ymm6, 88H
+        vpshufd ymm8, ymm8, 78H
+        vpunpckhdq ymm5, ymm5, ymm7
+        vpunpckldq ymm6, ymm6, ymm5
+        vpshufd ymm7, ymm6, 1EH
+        vmovdqa ymm5, ymm9
+        vmovdqa ymm6, ymm8
+        jmp     roundloop2
+endroundloop2:
+        vpxor   ymm0, ymm0, ymm2
+        vpxor   ymm1, ymm1, ymm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop2
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+10H], xmm1
+        vextracti128 xmmword ptr [rbx+20H], ymm0, 01H
+        vextracti128 xmmword ptr [rbx+30H], ymm1, 01H
+        vmovdqa xmm0, xmmword ptr [rsp]
+        vmovdqa xmm2, xmmword ptr [rsp+40H]
+        vmovdqu32 xmm0 {k1}, xmmword ptr [rsp+8H]
+        vmovdqu32 xmm2 {k1}, xmmword ptr [rsp+48H]
+        vmovdqa xmmword ptr [rsp], xmm0
+        vmovdqa xmmword ptr [rsp+40H], xmm2
+        add     rbx, 64
+        add     rdi, 16
+        sub     rsi, 2
+final1block:
+        test    esi, 1H
+        je      unwind
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+10H]
+        vmovd   xmm14, dword ptr [rsp]
+        vpinsrd xmm14, xmm14, dword ptr [rsp+40H], 1
+        vpinsrd xmm14, xmm14, dword ptr [BLAKE3_BLOCK_LEN], 2
+        vmovdqa xmm15, xmmword ptr [BLAKE3_IV]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+ALIGN   16
+innerloop1:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        vpinsrd xmm3, xmm14, eax, 3
+        vmovdqa xmm2, xmm15
+        vmovups xmm8, xmmword ptr [r8+rdx-40H]
+        vmovups xmm9, xmmword ptr [r8+rdx-30H]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [r8+rdx-20H]
+        vmovups xmm9, xmmword ptr [r8+rdx-10H]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 93H
+        vpshufd xmm7, xmm7, 93H
+        mov     al, 7
+roundloop1:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 93H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 39H
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 39H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 93H
+        dec     al
+        jz      endroundloop1
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0FH
+        vpshufd xmm4, xmm8, 39H
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0AAH
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 88H
+        vpshufd xmm8, xmm8, 78H
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 1EH
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     roundloop1
+endroundloop1:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop1
+        vmovdqu xmmword ptr [rbx], xmm0
+        vmovdqu xmmword ptr [rbx+10H], xmm1
+        jmp     unwind
+
+_blake3_hash_many_avx512 ENDP
+blake3_hash_many_avx512 ENDP
+
+ALIGN 16
+blake3_compress_in_place_avx512 PROC
+_blake3_compress_in_place_avx512 PROC
+        sub     rsp, 72
+        vmovdqa xmmword ptr [rsp], xmm6
+        vmovdqa xmmword ptr [rsp+10H], xmm7
+        vmovdqa xmmword ptr [rsp+20H], xmm8
+        vmovdqa xmmword ptr [rsp+30H], xmm9
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+10H]
+        movzx   eax, byte ptr [rsp+70H]
+        movzx   r8d, r8b
+        shl     rax, 32
+        add     r8, rax
+        vmovq   xmm3, r9
+        vmovq   xmm4, r8
+        vpunpcklqdq xmm3, xmm3, xmm4
+        vmovaps xmm2, xmmword ptr [BLAKE3_IV]
+        vmovups xmm8, xmmword ptr [rdx]
+        vmovups xmm9, xmmword ptr [rdx+10H]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [rdx+20H]
+        vmovups xmm9, xmmword ptr [rdx+30H]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 93H
+        vpshufd xmm7, xmm7, 93H
+        mov     al, 7
+@@:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 93H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 39H
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 39H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 93H
+        dec     al
+        jz      @F
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0FH
+        vpshufd xmm4, xmm8, 39H
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0AAH
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 88H
+        vpshufd xmm8, xmm8, 78H
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 1EH
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     @B
+@@:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        vmovdqu xmmword ptr [rcx], xmm0
+        vmovdqu xmmword ptr [rcx+10H], xmm1
+        vmovdqa xmm6, xmmword ptr [rsp]
+        vmovdqa xmm7, xmmword ptr [rsp+10H]
+        vmovdqa xmm8, xmmword ptr [rsp+20H]
+        vmovdqa xmm9, xmmword ptr [rsp+30H]
+        add     rsp, 72
+        ret
+_blake3_compress_in_place_avx512 ENDP
+blake3_compress_in_place_avx512 ENDP
+
+ALIGN 16
+blake3_compress_xof_avx512 PROC
+_blake3_compress_xof_avx512 PROC
+        sub     rsp, 72
+        vmovdqa xmmword ptr [rsp], xmm6
+        vmovdqa xmmword ptr [rsp+10H], xmm7
+        vmovdqa xmmword ptr [rsp+20H], xmm8
+        vmovdqa xmmword ptr [rsp+30H], xmm9
+        vmovdqu xmm0, xmmword ptr [rcx]
+        vmovdqu xmm1, xmmword ptr [rcx+10H]
+        movzx   eax, byte ptr [rsp+70H]
+        movzx   r8d, r8b
+        mov     r10, qword ptr [rsp+78H]
+        shl     rax, 32
+        add     r8, rax
+        vmovq   xmm3, r9
+        vmovq   xmm4, r8
+        vpunpcklqdq xmm3, xmm3, xmm4
+        vmovaps xmm2, xmmword ptr [BLAKE3_IV]
+        vmovups xmm8, xmmword ptr [rdx]
+        vmovups xmm9, xmmword ptr [rdx+10H]
+        vshufps xmm4, xmm8, xmm9, 136
+        vshufps xmm5, xmm8, xmm9, 221
+        vmovups xmm8, xmmword ptr [rdx+20H]
+        vmovups xmm9, xmmword ptr [rdx+30H]
+        vshufps xmm6, xmm8, xmm9, 136
+        vshufps xmm7, xmm8, xmm9, 221
+        vpshufd xmm6, xmm6, 93H
+        vpshufd xmm7, xmm7, 93H
+        mov     al, 7
+@@:
+        vpaddd  xmm0, xmm0, xmm4
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm5
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 93H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 39H
+        vpaddd  xmm0, xmm0, xmm6
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 16
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 12
+        vpaddd  xmm0, xmm0, xmm7
+        vpaddd  xmm0, xmm0, xmm1
+        vpxord  xmm3, xmm3, xmm0
+        vprord  xmm3, xmm3, 8
+        vpaddd  xmm2, xmm2, xmm3
+        vpxord  xmm1, xmm1, xmm2
+        vprord  xmm1, xmm1, 7
+        vpshufd xmm0, xmm0, 39H
+        vpshufd xmm3, xmm3, 4EH
+        vpshufd xmm2, xmm2, 93H
+        dec     al
+        jz      @F
+        vshufps xmm8, xmm4, xmm5, 214
+        vpshufd xmm9, xmm4, 0FH
+        vpshufd xmm4, xmm8, 39H
+        vshufps xmm8, xmm6, xmm7, 250
+        vpblendd xmm9, xmm9, xmm8, 0AAH
+        vpunpcklqdq xmm8, xmm7, xmm5
+        vpblendd xmm8, xmm8, xmm6, 88H
+        vpshufd xmm8, xmm8, 78H
+        vpunpckhdq xmm5, xmm5, xmm7
+        vpunpckldq xmm6, xmm6, xmm5
+        vpshufd xmm7, xmm6, 1EH
+        vmovdqa xmm5, xmm9
+        vmovdqa xmm6, xmm8
+        jmp     @B
+@@:
+        vpxor   xmm0, xmm0, xmm2
+        vpxor   xmm1, xmm1, xmm3
+        vpxor   xmm2, xmm2, xmmword ptr [rcx]
+        vpxor   xmm3, xmm3, xmmword ptr [rcx+10H]
+        vmovdqu xmmword ptr [r10], xmm0
+        vmovdqu xmmword ptr [r10+10H], xmm1
+        vmovdqu xmmword ptr [r10+20H], xmm2
+        vmovdqu xmmword ptr [r10+30H], xmm3
+        vmovdqa xmm6, xmmword ptr [rsp]
+        vmovdqa xmm7, xmmword ptr [rsp+10H]
+        vmovdqa xmm8, xmmword ptr [rsp+20H]
+        vmovdqa xmm9, xmmword ptr [rsp+30H]
+        add     rsp, 72
+        ret
+_blake3_compress_xof_avx512 ENDP
+blake3_compress_xof_avx512 ENDP
+
+_TEXT ENDS
+
+_RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
+ALIGN   64
+INDEX0:
+        dd    0,  1,  2,  3, 16, 17, 18, 19
+        dd    8,  9, 10, 11, 24, 25, 26, 27
+INDEX1:
+        dd    4,  5,  6,  7, 20, 21, 22, 23
+        dd   12, 13, 14, 15, 28, 29, 30, 31
+ADD0:
+        dd    0,  1,  2,  3,  4,  5,  6,  7
+        dd    8,  9, 10, 11, 12, 13, 14, 15
+ADD1:   
+        dd    1
+ADD16:  
+        dd   16
+BLAKE3_BLOCK_LEN:
+        dd   64
+ALIGN   64
+BLAKE3_IV:
+BLAKE3_IV_0:
+        dd   06A09E667H
+BLAKE3_IV_1:
+        dd   0BB67AE85H
+BLAKE3_IV_2:
+        dd   03C6EF372H
+BLAKE3_IV_3:
+        dd   0A54FF53AH
+
+_RDATA ENDS
+END
diff --git a/thirdparty/blake3/c/blake3_c_rust_bindings/Cargo.toml b/thirdparty/blake3/c/blake3_c_rust_bindings/Cargo.toml
new file mode 100644
index 000000000..4c3a25204
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_c_rust_bindings/Cargo.toml
@@ -0,0 +1,32 @@
+# These are Rust bindings for the C implementation of BLAKE3. As there is a
+# native (and faster) Rust implementation of BLAKE3 provided in this same repo,
+# these bindings are not expected to be used in production. They're intended
+# for testing and benchmarking.
+
+[package]
+name = "blake3_c_rust_bindings"
+version = "0.0.0"
+description = "TESTING ONLY Rust bindings for the BLAKE3 C implementation"
+edition = "2021"
+
+[features]
+# By default the x86-64 build uses assembly implementations. This feature makes
+# the build use the C intrinsics implementations instead.
+prefer_intrinsics = []
+# Activate NEON bindings. We don't currently do any CPU feature detection for
+# this. If this Cargo feature is on, the NEON gets used.
+neon = []
+# Enable TBB-based multithreading.
+tbb = []
+
+[dev-dependencies]
+arrayref = "0.3.5"
+arrayvec = { version = "0.7.0", default-features = false }
+page_size = "0.6.0"
+rand = "0.9.0"
+rand_chacha = "0.9.0"
+reference_impl = { path = "../../reference_impl" }
+
+[build-dependencies]
+cc = "1.0.48"
+ignore = "0.4.23"
diff --git a/thirdparty/blake3/c/blake3_c_rust_bindings/README.md b/thirdparty/blake3/c/blake3_c_rust_bindings/README.md
new file mode 100644
index 000000000..c44726b90
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_c_rust_bindings/README.md
@@ -0,0 +1,4 @@
+These are Rust bindings for the C implementation of BLAKE3. As there is
+a native Rust implementation of BLAKE3 provided in this same repo, these
+bindings are not expected to be used in production. They're intended for
+testing and benchmarking.
diff --git a/thirdparty/blake3/c/blake3_c_rust_bindings/benches/bench.rs b/thirdparty/blake3/c/blake3_c_rust_bindings/benches/bench.rs
new file mode 100644
index 000000000..b6c570444
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_c_rust_bindings/benches/bench.rs
@@ -0,0 +1,477 @@
+#![feature(test)]
+
+extern crate test;
+
+use arrayref::array_ref;
+use arrayvec::ArrayVec;
+use rand::prelude::*;
+use test::Bencher;
+
+const KIB: usize = 1024;
+const MAX_SIMD_DEGREE: usize = 16;
+
+const BLOCK_LEN: usize = 64;
+const CHUNK_LEN: usize = 1024;
+const OUT_LEN: usize = 32;
+
+// This struct randomizes two things:
+// 1. The actual bytes of input.
+// 2. The page offset the input starts at.
+pub struct RandomInput {
+    buf: Vec<u8>,
+    len: usize,
+    offsets: Vec<usize>,
+    offset_index: usize,
+}
+
+impl RandomInput {
+    pub fn new(b: &mut Bencher, len: usize) -> Self {
+        b.bytes += len as u64;
+        let page_size: usize = page_size::get();
+        let mut buf = vec![0u8; len + page_size];
+        let mut rng = rand::rng();
+        rng.fill_bytes(&mut buf);
+        let mut offsets: Vec<usize> = (0..page_size).collect();
+        offsets.shuffle(&mut rng);
+        Self {
+            buf,
+            len,
+            offsets,
+            offset_index: 0,
+        }
+    }
+
+    pub fn get(&mut self) -> &[u8] {
+        let offset = self.offsets[self.offset_index];
+        self.offset_index += 1;
+        if self.offset_index >= self.offsets.len() {
+            self.offset_index = 0;
+        }
+        &self.buf[offset..][..self.len]
+    }
+}
+
+type CompressInPlaceFn =
+    unsafe extern "C" fn(cv: *mut u32, block: *const u8, block_len: u8, counter: u64, flags: u8);
+
+fn bench_single_compression_fn(b: &mut Bencher, f: CompressInPlaceFn) {
+    let mut state = [1u32; 8];
+    let mut r = RandomInput::new(b, 64);
+    let input = array_ref!(r.get(), 0, 64);
+    b.iter(|| unsafe { f(state.as_mut_ptr(), input.as_ptr(), 64, 0, 0) });
+}
+
+#[bench]
+fn bench_single_compression_portable(b: &mut Bencher) {
+    bench_single_compression_fn(
+        b,
+        blake3_c_rust_bindings::ffi::blake3_compress_in_place_portable,
+    );
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_single_compression_sse2(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::sse2_detected() {
+        return;
+    }
+    bench_single_compression_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_compress_in_place_sse2,
+    );
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_single_compression_sse41(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::sse41_detected() {
+        return;
+    }
+    bench_single_compression_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_compress_in_place_sse41,
+    );
+}
+
+#[bench]
+fn bench_single_compression_avx512(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::avx512_detected() {
+        return;
+    }
+    bench_single_compression_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_compress_in_place_avx512,
+    );
+}
+
+type HashManyFn = unsafe extern "C" fn(
+    inputs: *const *const u8,
+    num_inputs: usize,
+    blocks: usize,
+    key: *const u32,
+    counter: u64,
+    increment_counter: bool,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: *mut u8,
+);
+
+fn bench_many_chunks_fn(b: &mut Bencher, f: HashManyFn, degree: usize) {
+    let mut inputs = Vec::new();
+    for _ in 0..degree {
+        inputs.push(RandomInput::new(b, CHUNK_LEN));
+    }
+    b.iter(|| {
+        let input_arrays: ArrayVec<&[u8; CHUNK_LEN], MAX_SIMD_DEGREE> = inputs
+            .iter_mut()
+            .take(degree)
+            .map(|i| array_ref!(i.get(), 0, CHUNK_LEN))
+            .collect();
+        let mut out = [0; MAX_SIMD_DEGREE * OUT_LEN];
+        unsafe {
+            f(
+                input_arrays.as_ptr() as _,
+                input_arrays.len(),
+                CHUNK_LEN / BLOCK_LEN,
+                [0u32; 8].as_ptr(),
+                0,
+                true,
+                0,
+                0,
+                0,
+                out.as_mut_ptr(),
+            )
+        }
+    });
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_chunks_sse2(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::sse2_detected() {
+        return;
+    }
+    bench_many_chunks_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_sse2,
+        4,
+    );
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_chunks_sse41(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::sse41_detected() {
+        return;
+    }
+    bench_many_chunks_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_sse41,
+        4,
+    );
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_chunks_avx2(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::avx2_detected() {
+        return;
+    }
+    bench_many_chunks_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_avx2,
+        8,
+    );
+}
+
+#[bench]
+fn bench_many_chunks_avx512(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::avx512_detected() {
+        return;
+    }
+    bench_many_chunks_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_avx512,
+        16,
+    );
+}
+
+#[bench]
+#[cfg(feature = "neon")]
+fn bench_many_chunks_neon(b: &mut Bencher) {
+    // When "neon" is on, NEON support is assumed.
+    bench_many_chunks_fn(
+        b,
+        blake3_c_rust_bindings::ffi::neon::blake3_hash_many_neon,
+        4,
+    );
+}
+
+// TODO: When we get const generics we can unify this with the chunks code.
+fn bench_many_parents_fn(b: &mut Bencher, f: HashManyFn, degree: usize) {
+    let mut inputs = Vec::new();
+    for _ in 0..degree {
+        inputs.push(RandomInput::new(b, BLOCK_LEN));
+    }
+    b.iter(|| {
+        let input_arrays: ArrayVec<&[u8; BLOCK_LEN], MAX_SIMD_DEGREE> = inputs
+            .iter_mut()
+            .take(degree)
+            .map(|i| array_ref!(i.get(), 0, BLOCK_LEN))
+            .collect();
+        let mut out = [0; MAX_SIMD_DEGREE * OUT_LEN];
+        unsafe {
+            f(
+                input_arrays.as_ptr() as _,
+                input_arrays.len(),
+                1,
+                [0u32; 8].as_ptr(),
+                0,
+                false,
+                0,
+                0,
+                0,
+                out.as_mut_ptr(),
+            )
+        }
+    });
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_parents_sse2(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::sse2_detected() {
+        return;
+    }
+    bench_many_parents_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_sse2,
+        4,
+    );
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_parents_sse41(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::sse41_detected() {
+        return;
+    }
+    bench_many_parents_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_sse41,
+        4,
+    );
+}
+
+#[bench]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn bench_many_parents_avx2(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::avx2_detected() {
+        return;
+    }
+    bench_many_parents_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_avx2,
+        8,
+    );
+}
+
+#[bench]
+fn bench_many_parents_avx512(b: &mut Bencher) {
+    if !blake3_c_rust_bindings::avx512_detected() {
+        return;
+    }
+    bench_many_parents_fn(
+        b,
+        blake3_c_rust_bindings::ffi::x86::blake3_hash_many_avx512,
+        16,
+    );
+}
+
+#[bench]
+#[cfg(feature = "neon")]
+fn bench_many_parents_neon(b: &mut Bencher) {
+    // When "neon" is on, NEON support is assumed.
+    bench_many_parents_fn(
+        b,
+        blake3_c_rust_bindings::ffi::neon::blake3_hash_many_neon,
+        4,
+    );
+}
+
+fn bench_incremental(b: &mut Bencher, len: usize) {
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| {
+        let mut hasher = blake3_c_rust_bindings::Hasher::new();
+        hasher.update(input.get());
+        let mut out = [0; 32];
+        hasher.finalize(&mut out);
+        out
+    });
+}
+
+#[bench]
+fn bench_incremental_0001_block(b: &mut Bencher) {
+    bench_incremental(b, BLOCK_LEN);
+}
+
+#[bench]
+fn bench_incremental_0001_kib(b: &mut Bencher) {
+    bench_incremental(b, 1 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0002_kib(b: &mut Bencher) {
+    bench_incremental(b, 2 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0004_kib(b: &mut Bencher) {
+    bench_incremental(b, 4 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0008_kib(b: &mut Bencher) {
+    bench_incremental(b, 8 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0016_kib(b: &mut Bencher) {
+    bench_incremental(b, 16 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0032_kib(b: &mut Bencher) {
+    bench_incremental(b, 32 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0064_kib(b: &mut Bencher) {
+    bench_incremental(b, 64 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0128_kib(b: &mut Bencher) {
+    bench_incremental(b, 128 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0256_kib(b: &mut Bencher) {
+    bench_incremental(b, 256 * KIB);
+}
+
+#[bench]
+fn bench_incremental_0512_kib(b: &mut Bencher) {
+    bench_incremental(b, 512 * KIB);
+}
+
+#[bench]
+fn bench_incremental_1024_kib(b: &mut Bencher) {
+    bench_incremental(b, 1024 * KIB);
+}
+
+#[cfg(feature = "tbb")]
+fn bench_tbb(b: &mut Bencher, len: usize) {
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| {
+        let mut hasher = blake3_c_rust_bindings::Hasher::new();
+        hasher.update_tbb(input.get());
+        let mut out = [0; 32];
+        hasher.finalize(&mut out);
+        out
+    });
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0001_block(b: &mut Bencher) {
+    bench_tbb(b, BLOCK_LEN);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0001_kib(b: &mut Bencher) {
+    bench_tbb(b, 1 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0002_kib(b: &mut Bencher) {
+    bench_tbb(b, 2 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0004_kib(b: &mut Bencher) {
+    bench_tbb(b, 4 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0008_kib(b: &mut Bencher) {
+    bench_tbb(b, 8 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0016_kib(b: &mut Bencher) {
+    bench_tbb(b, 16 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0032_kib(b: &mut Bencher) {
+    bench_tbb(b, 32 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0064_kib(b: &mut Bencher) {
+    bench_tbb(b, 64 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0128_kib(b: &mut Bencher) {
+    bench_tbb(b, 128 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0256_kib(b: &mut Bencher) {
+    bench_tbb(b, 256 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_0512_kib(b: &mut Bencher) {
+    bench_tbb(b, 512 * KIB);
+}
+
+#[bench]
+#[cfg(feature = "tbb")]
+fn bench_tbb_1024_kib(b: &mut Bencher) {
+    bench_tbb(b, 1024 * KIB);
+}
+
+// This checks that update() splits up its input in increasing powers of 2, so
+// that it can recover a high degree of parallelism when the number of bytes
+// hashed so far is uneven. The performance of this benchmark should be
+// reasonably close to bench_incremental_0064_kib, within 80% or so. When we
+// had a bug in this logic (https://github.com/BLAKE3-team/BLAKE3/issues/69),
+// performance was less than half.
+#[bench]
+fn bench_two_updates(b: &mut Bencher) {
+    let len = 65536;
+    let mut input = RandomInput::new(b, len);
+    b.iter(|| {
+        let mut hasher = blake3_c_rust_bindings::Hasher::new();
+        let input = input.get();
+        hasher.update(&input[..1]);
+        hasher.update(&input[1..]);
+        let mut out = [0; 32];
+        hasher.finalize(&mut out);
+        out
+    });
+}
diff --git a/thirdparty/blake3/c/blake3_c_rust_bindings/build.rs b/thirdparty/blake3/c/blake3_c_rust_bindings/build.rs
new file mode 100644
index 000000000..0236df4b4
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_c_rust_bindings/build.rs
@@ -0,0 +1,253 @@
+use std::env;
+
+fn defined(var: &str) -> bool {
+    env::var_os(var).is_some()
+}
+
+fn target_components() -> Vec<String> {
+    let target = env::var("TARGET").unwrap();
+    target.split("-").map(|s| s.to_string()).collect()
+}
+
+fn is_x86_64() -> bool {
+    target_components()[0] == "x86_64"
+}
+
+fn is_windows_target() -> bool {
+    env::var("CARGO_CFG_TARGET_OS").unwrap() == "windows"
+}
+
+fn use_msvc_asm() -> bool {
+    const MSVC_NAMES: &[&str] = &["", "cl", "cl.exe"];
+    let target_os = env::var("CARGO_CFG_TARGET_OS").unwrap_or_default();
+    let target_env = env::var("CARGO_CFG_TARGET_ENV").unwrap_or_default();
+    let target_windows_msvc = target_os == "windows" && target_env == "msvc";
+    let host_triple = env::var("HOST").unwrap_or_default();
+    let target_triple = env::var("TARGET").unwrap_or_default();
+    let cross_compiling = host_triple != target_triple;
+    let cc = env::var("CC").unwrap_or_default().to_ascii_lowercase();
+    if !target_windows_msvc {
+        // We are not building for Windows with the MSVC toolchain.
+        false
+    } else if !cross_compiling && MSVC_NAMES.contains(&&*cc) {
+        // We are building on Windows with the MSVC toolchain (and not cross-compiling for another architecture or target).
+        true
+    } else {
+        // We are cross-compiling to Windows with the MSVC toolchain.
+        let target_arch = env::var("CARGO_CFG_TARGET_ARCH").unwrap_or_default();
+        let target_vendor = env::var("CARGO_CFG_TARGET_VENDOR").unwrap_or_default();
+        let cc = env::var(format!("CC_{target_arch}_{target_vendor}_windows_msvc"))
+            .unwrap_or_default()
+            .to_ascii_lowercase();
+        // Check if we are using the MSVC compiler.
+        MSVC_NAMES.contains(&&*cc)
+    }
+}
+
+fn is_x86_32() -> bool {
+    let arch = &target_components()[0];
+    arch == "i386" || arch == "i586" || arch == "i686"
+}
+
+fn is_armv7() -> bool {
+    target_components()[0] == "armv7"
+}
+
+fn is_aarch64() -> bool {
+    target_components()[0] == "aarch64"
+}
+
+// Windows targets may be using the MSVC toolchain or the GNU toolchain. The
+// right compiler flags to use depend on the toolchain. (And we don't want to
+// use flag_if_supported, because we don't want features to be silently
+// disabled by old compilers.)
+fn is_windows_msvc() -> bool {
+    // Some targets are only two components long, so check in steps.
+    target_components()[1] == "pc"
+        && target_components()[2] == "windows"
+        && target_components()[3] == "msvc"
+}
+
+fn new_build() -> cc::Build {
+    let mut build = cc::Build::new();
+    if !is_windows_msvc() {
+        build.flag("-std=c11");
+    }
+    build
+}
+
+fn new_cpp_build() -> cc::Build {
+    let mut build = cc::Build::new();
+    build.cpp(true);
+    if is_windows_msvc() {
+        build.flag("/std:c++20");
+        build.flag("/EHs-c-");
+        build.flag("/GR-");
+    } else {
+        build.flag("-std=c++20");
+        build.flag("-fno-exceptions");
+        build.flag("-fno-rtti");
+    }
+    build
+}
+
+fn c_dir_path(filename: &str) -> String {
+    // The `cross` tool doesn't support reading files in parent directories. As a hacky workaround
+    // in `cross_test.sh`, we move the c/ directory around and set BLAKE3_C_DIR_OVERRIDE. Regular
+    // building and testing doesn't require this.
+    if let Ok(c_dir_override) = env::var("BLAKE3_C_DIR_OVERRIDE") {
+        c_dir_override + "/" + filename
+    } else {
+        "../".to_string() + filename
+    }
+}
+
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    let mut base_build = new_build();
+    base_build.file(c_dir_path("blake3.c"));
+    base_build.file(c_dir_path("blake3_dispatch.c"));
+    base_build.file(c_dir_path("blake3_portable.c"));
+    if cfg!(feature = "tbb") {
+        base_build.define("BLAKE3_USE_TBB", "1");
+    }
+    base_build.compile("blake3_base");
+
+    if cfg!(feature = "tbb") {
+        let mut tbb_build = new_cpp_build();
+        tbb_build.define("BLAKE3_USE_TBB", "1");
+        tbb_build.file(c_dir_path("blake3_tbb.cpp"));
+        tbb_build.compile("blake3_tbb");
+        println!("cargo::rustc-link-lib=tbb");
+    }
+
+    if is_x86_64() && !defined("CARGO_FEATURE_PREFER_INTRINSICS") {
+        // On 64-bit, use the assembly implementations, unless the
+        // "prefer_intrinsics" feature is enabled.
+        if is_windows_target() {
+            if use_msvc_asm() {
+                let mut build = new_build();
+                build.file(c_dir_path("blake3_sse2_x86-64_windows_msvc.asm"));
+                build.file(c_dir_path("blake3_sse41_x86-64_windows_msvc.asm"));
+                build.file(c_dir_path("blake3_avx2_x86-64_windows_msvc.asm"));
+                build.file(c_dir_path("blake3_avx512_x86-64_windows_msvc.asm"));
+                build.compile("blake3_asm");
+            } else {
+                let mut build = new_build();
+                build.file(c_dir_path("blake3_sse2_x86-64_windows_gnu.S"));
+                build.file(c_dir_path("blake3_sse41_x86-64_windows_gnu.S"));
+                build.file(c_dir_path("blake3_avx2_x86-64_windows_gnu.S"));
+                build.file(c_dir_path("blake3_avx512_x86-64_windows_gnu.S"));
+                build.compile("blake3_asm");
+            }
+        } else {
+            // All non-Windows implementations are assumed to support
+            // Linux-style assembly. These files do contain a small
+            // explicit workaround for macOS also.
+            let mut build = new_build();
+            build.file(c_dir_path("blake3_sse2_x86-64_unix.S"));
+            build.file(c_dir_path("blake3_sse41_x86-64_unix.S"));
+            build.file(c_dir_path("blake3_avx2_x86-64_unix.S"));
+            build.file(c_dir_path("blake3_avx512_x86-64_unix.S"));
+            build.compile("blake3_asm");
+        }
+    } else if is_x86_64() || is_x86_32() {
+        // Assembly implementations are only for 64-bit. On 32-bit, or if
+        // the "prefer_intrinsics" feature is enabled, use the
+        // intrinsics-based C implementations. These each need to be
+        // compiled separately, with the corresponding instruction set
+        // extension explicitly enabled in the compiler.
+
+        let mut sse2_build = new_build();
+        sse2_build.file(c_dir_path("blake3_sse2.c"));
+        if is_windows_msvc() {
+            // /arch:SSE2 is the default on x86 and undefined on x86_64:
+            // https://docs.microsoft.com/en-us/cpp/build/reference/arch-x86
+            // It also includes SSE4.1 intrinsics:
+            // https://stackoverflow.com/a/32183222/823869
+        } else {
+            sse2_build.flag("-msse2");
+        }
+        sse2_build.compile("blake3_sse2");
+
+        let mut sse41_build = new_build();
+        sse41_build.file(c_dir_path("blake3_sse41.c"));
+        if is_windows_msvc() {
+            // /arch:SSE2 is the default on x86 and undefined on x86_64:
+            // https://docs.microsoft.com/en-us/cpp/build/reference/arch-x86
+            // It also includes SSE4.1 intrinsics:
+            // https://stackoverflow.com/a/32183222/823869
+        } else {
+            sse41_build.flag("-msse4.1");
+        }
+        sse41_build.compile("blake3_sse41");
+
+        let mut avx2_build = new_build();
+        avx2_build.file(c_dir_path("blake3_avx2.c"));
+        if is_windows_msvc() {
+            avx2_build.flag("/arch:AVX2");
+        } else {
+            avx2_build.flag("-mavx2");
+        }
+        avx2_build.compile("blake3_avx2");
+
+        let mut avx512_build = new_build();
+        avx512_build.file(c_dir_path("blake3_avx512.c"));
+        if is_windows_msvc() {
+            // Note that a lot of versions of MSVC don't support /arch:AVX512,
+            // and they'll discard it with a warning, hopefully leading to a
+            // build error.
+            avx512_build.flag("/arch:AVX512");
+        } else {
+            avx512_build.flag("-mavx512f");
+            avx512_build.flag("-mavx512vl");
+        }
+        avx512_build.compile("blake3_avx512");
+    }
+
+    // We only build NEON code here if
+    // 1) it's requested
+    // and 2) the root crate is not already building it.
+    // The only time this will really happen is if you build this
+    // crate by hand with the "neon" feature for some reason.
+    //
+    // In addition, 3) if the target is aarch64, NEON is on by default.
+    if defined("CARGO_FEATURE_NEON") || is_aarch64() {
+        let mut neon_build = new_build();
+        neon_build.file(c_dir_path("blake3_neon.c"));
+        // ARMv7 platforms that support NEON generally need the following
+        // flags. AArch64 supports NEON by default and does not support -mpfu.
+        if is_armv7() {
+            neon_build.flag("-mfpu=neon-vfpv4");
+            neon_build.flag("-mfloat-abi=hard");
+        }
+        neon_build.compile("blake3_neon");
+    }
+
+    // The `cc` crate does not automatically emit rerun-if directives for the
+    // environment variables it supports, in particular for $CC. We expect to
+    // do a lot of benchmarking across different compilers, so we explicitly
+    // add the variables that we're likely to need.
+    println!("cargo:rerun-if-env-changed=CC");
+    println!("cargo:rerun-if-env-changed=CFLAGS");
+
+    // Ditto for source files, though these shouldn't change as often. `ignore::Walk` respects
+    // .gitignore, so this doesn't traverse target/.
+    for result in ignore::Walk::new("..") {
+        let result = result?;
+        let path = result.path();
+        if path.is_file() {
+            println!("cargo:rerun-if-changed={}", path.to_str().unwrap());
+        }
+    }
+
+    // When compiling with clang-cl for windows, it adds .asm files to the root
+    // which we need to delete so cargo doesn't get angry
+    if is_windows_target() && !use_msvc_asm() {
+        let _ = std::fs::remove_file("blake3_avx2_x86-64_windows_gnu.asm");
+        let _ = std::fs::remove_file("blake3_avx512_x86-64_windows_gnu.asm");
+        let _ = std::fs::remove_file("blake3_sse2_x86-64_windows_gnu.asm");
+        let _ = std::fs::remove_file("blake3_sse41_x86-64_windows_gnu.asm");
+    }
+
+    Ok(())
+}
diff --git a/thirdparty/blake3/c/blake3_c_rust_bindings/cross_test.sh b/thirdparty/blake3/c/blake3_c_rust_bindings/cross_test.sh
new file mode 100755
index 000000000..94d50affb
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_c_rust_bindings/cross_test.sh
@@ -0,0 +1,31 @@
+#! /usr/bin/env bash
+
+# This hacky script works around the fact that `cross test` does not support
+# path dependencies. (It uses a docker shared folder to let the guest access
+# project files, so parent directories aren't available.) Solve this problem by
+# copying the entire project to a temp dir and rearranging paths to put "c" and
+# "reference_impl" underneath "blake3_c_rust_bindings", so that everything is
+# accessible. Hopefully this will just run on CI forever and no one will ever
+# read this and discover my deep shame.
+
+set -e -u -o pipefail
+
+project_root="$(realpath "$(dirname "$BASH_SOURCE")/../..")"
+tmpdir="$(mktemp -d)"
+echo "Running cross tests in $tmpdir"
+cd "$tmpdir"
+git clone "$project_root" blake3
+mv blake3/c/blake3_c_rust_bindings .
+mv blake3/reference_impl blake3_c_rust_bindings
+mv blake3/c blake3_c_rust_bindings
+cd blake3_c_rust_bindings
+sed -i 's|reference_impl = { path = "../../reference_impl" }|reference_impl = { path = "reference_impl" }|' Cargo.toml
+
+export BLAKE3_C_DIR_OVERRIDE="./c"
+cat > Cross.toml << EOF
+[build.env]
+passthrough = [
+    "BLAKE3_C_DIR_OVERRIDE",
+]
+EOF
+cross test "$@"
diff --git a/thirdparty/blake3/c/blake3_c_rust_bindings/src/lib.rs b/thirdparty/blake3/c/blake3_c_rust_bindings/src/lib.rs
new file mode 100644
index 000000000..c2b398920
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_c_rust_bindings/src/lib.rs
@@ -0,0 +1,333 @@
+//! These are Rust bindings for the C implementation of BLAKE3. As there is a
+//! native (and faster) Rust implementation of BLAKE3 provided in this same
+//! repo, these bindings are not expected to be used in production. They're
+//! intended for testing and benchmarking.
+
+use std::ffi::{c_void, CString};
+use std::mem::MaybeUninit;
+
+#[cfg(test)]
+mod test;
+
+pub const BLOCK_LEN: usize = 64;
+pub const CHUNK_LEN: usize = 1024;
+pub const OUT_LEN: usize = 32;
+
+// Feature detection functions for tests and benchmarks. Note that the C code
+// does its own feature detection in blake3_dispatch.c.
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+pub fn sse2_detected() -> bool {
+    is_x86_feature_detected!("sse2")
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+pub fn sse41_detected() -> bool {
+    is_x86_feature_detected!("sse4.1")
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+pub fn avx2_detected() -> bool {
+    is_x86_feature_detected!("avx2")
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+pub fn avx512_detected() -> bool {
+    is_x86_feature_detected!("avx512f") && is_x86_feature_detected!("avx512vl")
+}
+
+#[derive(Clone)]
+pub struct Hasher(ffi::blake3_hasher);
+
+impl Hasher {
+    pub fn new() -> Self {
+        let mut c_state = MaybeUninit::uninit();
+        unsafe {
+            ffi::blake3_hasher_init(c_state.as_mut_ptr());
+            Self(c_state.assume_init())
+        }
+    }
+
+    pub fn new_keyed(key: &[u8; 32]) -> Self {
+        let mut c_state = MaybeUninit::uninit();
+        unsafe {
+            ffi::blake3_hasher_init_keyed(c_state.as_mut_ptr(), key.as_ptr());
+            Self(c_state.assume_init())
+        }
+    }
+
+    pub fn new_derive_key(context: &str) -> Self {
+        let mut c_state = MaybeUninit::uninit();
+        let context_c_string = CString::new(context).expect("valid C string, no null bytes");
+        unsafe {
+            ffi::blake3_hasher_init_derive_key(c_state.as_mut_ptr(), context_c_string.as_ptr());
+            Self(c_state.assume_init())
+        }
+    }
+
+    pub fn new_derive_key_raw(context: &[u8]) -> Self {
+        let mut c_state = MaybeUninit::uninit();
+        unsafe {
+            ffi::blake3_hasher_init_derive_key_raw(
+                c_state.as_mut_ptr(),
+                context.as_ptr() as *const _,
+                context.len(),
+            );
+            Self(c_state.assume_init())
+        }
+    }
+
+    pub fn update(&mut self, input: &[u8]) {
+        unsafe {
+            ffi::blake3_hasher_update(&mut self.0, input.as_ptr() as *const c_void, input.len());
+        }
+    }
+
+    #[cfg(feature = "tbb")]
+    pub fn update_tbb(&mut self, input: &[u8]) {
+        unsafe {
+            ffi::blake3_hasher_update_tbb(
+                &mut self.0,
+                input.as_ptr() as *const c_void,
+                input.len(),
+            );
+        }
+    }
+
+    pub fn finalize(&self, output: &mut [u8]) {
+        unsafe {
+            ffi::blake3_hasher_finalize(&self.0, output.as_mut_ptr(), output.len());
+        }
+    }
+
+    pub fn finalize_seek(&self, seek: u64, output: &mut [u8]) {
+        unsafe {
+            ffi::blake3_hasher_finalize_seek(&self.0, seek, output.as_mut_ptr(), output.len());
+        }
+    }
+
+    pub fn reset(&mut self) {
+        unsafe {
+            ffi::blake3_hasher_reset(&mut self.0);
+        }
+    }
+}
+
+pub mod ffi {
+    #[repr(C)]
+    #[derive(Copy, Clone)]
+    pub struct blake3_chunk_state {
+        pub cv: [u32; 8usize],
+        pub chunk_counter: u64,
+        pub buf: [u8; 64usize],
+        pub buf_len: u8,
+        pub blocks_compressed: u8,
+        pub flags: u8,
+    }
+
+    #[repr(C)]
+    #[derive(Copy, Clone)]
+    pub struct blake3_hasher {
+        pub key: [u32; 8usize],
+        pub chunk: blake3_chunk_state,
+        pub cv_stack_len: u8,
+        pub cv_stack: [u8; 1728usize],
+    }
+
+    extern "C" {
+        // public interface
+        pub fn blake3_hasher_init(self_: *mut blake3_hasher);
+        pub fn blake3_hasher_init_keyed(self_: *mut blake3_hasher, key: *const u8);
+        pub fn blake3_hasher_init_derive_key(
+            self_: *mut blake3_hasher,
+            context: *const ::std::os::raw::c_char,
+        );
+        pub fn blake3_hasher_init_derive_key_raw(
+            self_: *mut blake3_hasher,
+            context: *const ::std::os::raw::c_void,
+            context_len: usize,
+        );
+        pub fn blake3_hasher_update(
+            self_: *mut blake3_hasher,
+            input: *const ::std::os::raw::c_void,
+            input_len: usize,
+        );
+        #[cfg(feature = "tbb")]
+        pub fn blake3_hasher_update_tbb(
+            self_: *mut blake3_hasher,
+            input: *const ::std::os::raw::c_void,
+            input_len: usize,
+        );
+        pub fn blake3_hasher_finalize(self_: *const blake3_hasher, out: *mut u8, out_len: usize);
+        pub fn blake3_hasher_finalize_seek(
+            self_: *const blake3_hasher,
+            seek: u64,
+            out: *mut u8,
+            out_len: usize,
+        );
+        pub fn blake3_hasher_reset(self_: *mut blake3_hasher);
+
+        // portable low-level functions
+        pub fn blake3_compress_in_place_portable(
+            cv: *mut u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+        );
+        pub fn blake3_compress_xof_portable(
+            cv: *const u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+            out: *mut u8,
+        );
+        pub fn blake3_hash_many_portable(
+            inputs: *const *const u8,
+            num_inputs: usize,
+            blocks: usize,
+            key: *const u32,
+            counter: u64,
+            increment_counter: bool,
+            flags: u8,
+            flags_start: u8,
+            flags_end: u8,
+            out: *mut u8,
+        );
+    }
+
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    pub mod x86 {
+        extern "C" {
+            // SSE2 low level functions
+            pub fn blake3_compress_in_place_sse2(
+                cv: *mut u32,
+                block: *const u8,
+                block_len: u8,
+                counter: u64,
+                flags: u8,
+            );
+            pub fn blake3_compress_xof_sse2(
+                cv: *const u32,
+                block: *const u8,
+                block_len: u8,
+                counter: u64,
+                flags: u8,
+                out: *mut u8,
+            );
+            pub fn blake3_hash_many_sse2(
+                inputs: *const *const u8,
+                num_inputs: usize,
+                blocks: usize,
+                key: *const u32,
+                counter: u64,
+                increment_counter: bool,
+                flags: u8,
+                flags_start: u8,
+                flags_end: u8,
+                out: *mut u8,
+            );
+
+            // SSE4.1 low level functions
+            pub fn blake3_compress_in_place_sse41(
+                cv: *mut u32,
+                block: *const u8,
+                block_len: u8,
+                counter: u64,
+                flags: u8,
+            );
+            pub fn blake3_compress_xof_sse41(
+                cv: *const u32,
+                block: *const u8,
+                block_len: u8,
+                counter: u64,
+                flags: u8,
+                out: *mut u8,
+            );
+            pub fn blake3_hash_many_sse41(
+                inputs: *const *const u8,
+                num_inputs: usize,
+                blocks: usize,
+                key: *const u32,
+                counter: u64,
+                increment_counter: bool,
+                flags: u8,
+                flags_start: u8,
+                flags_end: u8,
+                out: *mut u8,
+            );
+
+            // AVX2 low level functions
+            pub fn blake3_hash_many_avx2(
+                inputs: *const *const u8,
+                num_inputs: usize,
+                blocks: usize,
+                key: *const u32,
+                counter: u64,
+                increment_counter: bool,
+                flags: u8,
+                flags_start: u8,
+                flags_end: u8,
+                out: *mut u8,
+            );
+
+            // AVX-512 low level functions
+            pub fn blake3_compress_xof_avx512(
+                cv: *const u32,
+                block: *const u8,
+                block_len: u8,
+                counter: u64,
+                flags: u8,
+                out: *mut u8,
+            );
+            pub fn blake3_compress_in_place_avx512(
+                cv: *mut u32,
+                block: *const u8,
+                block_len: u8,
+                counter: u64,
+                flags: u8,
+            );
+            pub fn blake3_hash_many_avx512(
+                inputs: *const *const u8,
+                num_inputs: usize,
+                blocks: usize,
+                key: *const u32,
+                counter: u64,
+                increment_counter: bool,
+                flags: u8,
+                flags_start: u8,
+                flags_end: u8,
+                out: *mut u8,
+            );
+            #[cfg(unix)]
+            pub fn blake3_xof_many_avx512(
+                cv: *const u32,
+                block: *const u8,
+                block_len: u8,
+                counter: u64,
+                flags: u8,
+                out: *mut u8,
+                outblocks: usize,
+            );
+        }
+    }
+
+    #[cfg(feature = "neon")]
+    pub mod neon {
+        extern "C" {
+            // NEON low level functions
+            pub fn blake3_hash_many_neon(
+                inputs: *const *const u8,
+                num_inputs: usize,
+                blocks: usize,
+                key: *const u32,
+                counter: u64,
+                increment_counter: bool,
+                flags: u8,
+                flags_start: u8,
+                flags_end: u8,
+                out: *mut u8,
+            );
+        }
+    }
+}
diff --git a/thirdparty/blake3/c/blake3_c_rust_bindings/src/test.rs b/thirdparty/blake3/c/blake3_c_rust_bindings/src/test.rs
new file mode 100644
index 000000000..6ff61ec6f
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_c_rust_bindings/src/test.rs
@@ -0,0 +1,696 @@
+// Most of this code is duplicated from the root `blake3` crate. Perhaps we
+// could share more of it in the future.
+
+use crate::{BLOCK_LEN, CHUNK_LEN, OUT_LEN};
+use arrayref::{array_mut_ref, array_ref};
+use arrayvec::ArrayVec;
+use core::usize;
+use rand::prelude::*;
+
+const CHUNK_START: u8 = 1 << 0;
+const CHUNK_END: u8 = 1 << 1;
+const PARENT: u8 = 1 << 2;
+const ROOT: u8 = 1 << 3;
+const KEYED_HASH: u8 = 1 << 4;
+// const DERIVE_KEY_CONTEXT: u8 = 1 << 5;
+// const DERIVE_KEY_MATERIAL: u8 = 1 << 6;
+
+// Interesting input lengths to run tests on.
+pub const TEST_CASES: &[usize] = &[
+    0,
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    BLOCK_LEN - 1,
+    BLOCK_LEN,
+    BLOCK_LEN + 1,
+    2 * BLOCK_LEN - 1,
+    2 * BLOCK_LEN,
+    2 * BLOCK_LEN + 1,
+    CHUNK_LEN - 1,
+    CHUNK_LEN,
+    CHUNK_LEN + 1,
+    2 * CHUNK_LEN,
+    2 * CHUNK_LEN + 1,
+    3 * CHUNK_LEN,
+    3 * CHUNK_LEN + 1,
+    4 * CHUNK_LEN,
+    4 * CHUNK_LEN + 1,
+    5 * CHUNK_LEN,
+    5 * CHUNK_LEN + 1,
+    6 * CHUNK_LEN,
+    6 * CHUNK_LEN + 1,
+    7 * CHUNK_LEN,
+    7 * CHUNK_LEN + 1,
+    8 * CHUNK_LEN,
+    8 * CHUNK_LEN + 1,
+    16 * CHUNK_LEN,  // AVX512's bandwidth
+    31 * CHUNK_LEN,  // 16 + 8 + 4 + 2 + 1
+    100 * CHUNK_LEN, // subtrees larger than MAX_SIMD_DEGREE chunks
+];
+
+pub const TEST_CASES_MAX: usize = 100 * CHUNK_LEN;
+
+// There's a test to make sure these two are equal below.
+pub const TEST_KEY: [u8; 32] = *b"whats the Elvish word for friend";
+pub const TEST_KEY_WORDS: [u32; 8] = [
+    1952540791, 1752440947, 1816469605, 1752394102, 1919907616, 1868963940, 1919295602, 1684956521,
+];
+
+// Paint the input with a repeating byte pattern. We use a cycle length of 251,
+// because that's the largest prime number less than 256. This makes it
+// unlikely to swapping any two adjacent input blocks or chunks will give the
+// same answer.
+fn paint_test_input(buf: &mut [u8]) {
+    for (i, b) in buf.iter_mut().enumerate() {
+        *b = (i % 251) as u8;
+    }
+}
+
+#[inline(always)]
+fn le_bytes_from_words_32(words: &[u32; 8]) -> [u8; 32] {
+    let mut out = [0; 32];
+    *array_mut_ref!(out, 0 * 4, 4) = words[0].to_le_bytes();
+    *array_mut_ref!(out, 1 * 4, 4) = words[1].to_le_bytes();
+    *array_mut_ref!(out, 2 * 4, 4) = words[2].to_le_bytes();
+    *array_mut_ref!(out, 3 * 4, 4) = words[3].to_le_bytes();
+    *array_mut_ref!(out, 4 * 4, 4) = words[4].to_le_bytes();
+    *array_mut_ref!(out, 5 * 4, 4) = words[5].to_le_bytes();
+    *array_mut_ref!(out, 6 * 4, 4) = words[6].to_le_bytes();
+    *array_mut_ref!(out, 7 * 4, 4) = words[7].to_le_bytes();
+    out
+}
+
+type CompressInPlaceFn =
+    unsafe extern "C" fn(cv: *mut u32, block: *const u8, block_len: u8, counter: u64, flags: u8);
+
+type CompressXofFn = unsafe extern "C" fn(
+    cv: *const u32,
+    block: *const u8,
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+    out: *mut u8,
+);
+
+// A shared helper function for platform-specific tests.
+pub fn test_compress_fn(compress_in_place_fn: CompressInPlaceFn, compress_xof_fn: CompressXofFn) {
+    let initial_state = TEST_KEY_WORDS;
+    let block_len: u8 = 61;
+    let mut block = [0; BLOCK_LEN];
+    paint_test_input(&mut block[..block_len as usize]);
+    // Use a counter with set bits in both 32-bit words.
+    let counter = (5u64 << 32) + 6;
+    let flags = CHUNK_END | ROOT | KEYED_HASH;
+
+    let mut portable_out = [0; 64];
+    unsafe {
+        crate::ffi::blake3_compress_xof_portable(
+            initial_state.as_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+            portable_out.as_mut_ptr(),
+        );
+    }
+
+    let mut test_state = initial_state;
+    unsafe {
+        compress_in_place_fn(
+            test_state.as_mut_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+        )
+    };
+    let test_state_bytes = le_bytes_from_words_32(&test_state);
+    let mut test_xof = [0; 64];
+    unsafe {
+        compress_xof_fn(
+            initial_state.as_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+            test_xof.as_mut_ptr(),
+        )
+    };
+
+    assert_eq!(&portable_out[..32], &test_state_bytes[..]);
+    assert_eq!(&portable_out[..], &test_xof[..]);
+}
+
+// Testing the portable implementation against itself is circular, but why not.
+#[test]
+fn test_compress_portable() {
+    test_compress_fn(
+        crate::ffi::blake3_compress_in_place_portable,
+        crate::ffi::blake3_compress_xof_portable,
+    );
+}
+
+#[test]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_compress_sse2() {
+    if !crate::sse2_detected() {
+        return;
+    }
+    test_compress_fn(
+        crate::ffi::x86::blake3_compress_in_place_sse2,
+        crate::ffi::x86::blake3_compress_xof_sse2,
+    );
+}
+
+#[test]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_compress_sse41() {
+    if !crate::sse41_detected() {
+        return;
+    }
+    test_compress_fn(
+        crate::ffi::x86::blake3_compress_in_place_sse41,
+        crate::ffi::x86::blake3_compress_xof_sse41,
+    );
+}
+
+#[test]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_compress_avx512() {
+    if !crate::avx512_detected() {
+        return;
+    }
+    test_compress_fn(
+        crate::ffi::x86::blake3_compress_in_place_avx512,
+        crate::ffi::x86::blake3_compress_xof_avx512,
+    );
+}
+
+type HashManyFn = unsafe extern "C" fn(
+    inputs: *const *const u8,
+    num_inputs: usize,
+    blocks: usize,
+    key: *const u32,
+    counter: u64,
+    increment_counter: bool,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: *mut u8,
+);
+
+// A shared helper function for platform-specific tests.
+pub fn test_hash_many_fn(hash_many_fn: HashManyFn) {
+    // Test a few different initial counter values.
+    // - 0: The base case.
+    // - u32::MAX: The low word of the counter overflows for all inputs except the first.
+    // - i32::MAX: *No* overflow. But carry bugs in tricky SIMD code can screw this up, if you XOR
+    //   when you're supposed to ANDNOT...
+    let initial_counters = [0, u32::MAX as u64, i32::MAX as u64];
+    for counter in initial_counters {
+        dbg!(counter);
+
+        // 31 (16 + 8 + 4 + 2 + 1) inputs
+        const NUM_INPUTS: usize = 31;
+        let mut input_buf = [0; CHUNK_LEN * NUM_INPUTS];
+        crate::test::paint_test_input(&mut input_buf);
+
+        // First hash chunks.
+        let mut chunks = ArrayVec::<&[u8; CHUNK_LEN], NUM_INPUTS>::new();
+        for i in 0..NUM_INPUTS {
+            chunks.push(array_ref!(input_buf, i * CHUNK_LEN, CHUNK_LEN));
+        }
+        let mut portable_chunks_out = [0; NUM_INPUTS * OUT_LEN];
+        unsafe {
+            crate::ffi::blake3_hash_many_portable(
+                chunks.as_ptr() as _,
+                chunks.len(),
+                CHUNK_LEN / BLOCK_LEN,
+                TEST_KEY_WORDS.as_ptr(),
+                counter,
+                true,
+                KEYED_HASH,
+                CHUNK_START,
+                CHUNK_END,
+                portable_chunks_out.as_mut_ptr(),
+            );
+        }
+
+        let mut test_chunks_out = [0; NUM_INPUTS * OUT_LEN];
+        unsafe {
+            hash_many_fn(
+                chunks.as_ptr() as _,
+                chunks.len(),
+                CHUNK_LEN / BLOCK_LEN,
+                TEST_KEY_WORDS.as_ptr(),
+                counter,
+                true,
+                KEYED_HASH,
+                CHUNK_START,
+                CHUNK_END,
+                test_chunks_out.as_mut_ptr(),
+            );
+        }
+        for n in 0..NUM_INPUTS {
+            dbg!(n);
+            assert_eq!(
+                &portable_chunks_out[n * OUT_LEN..][..OUT_LEN],
+                &test_chunks_out[n * OUT_LEN..][..OUT_LEN]
+            );
+        }
+
+        // Then hash parents.
+        let mut parents = ArrayVec::<&[u8; 2 * OUT_LEN], NUM_INPUTS>::new();
+        for i in 0..NUM_INPUTS {
+            parents.push(array_ref!(input_buf, i * 2 * OUT_LEN, 2 * OUT_LEN));
+        }
+        let mut portable_parents_out = [0; NUM_INPUTS * OUT_LEN];
+        unsafe {
+            crate::ffi::blake3_hash_many_portable(
+                parents.as_ptr() as _,
+                parents.len(),
+                1,
+                TEST_KEY_WORDS.as_ptr(),
+                counter,
+                false,
+                KEYED_HASH | PARENT,
+                0,
+                0,
+                portable_parents_out.as_mut_ptr(),
+            );
+        }
+
+        let mut test_parents_out = [0; NUM_INPUTS * OUT_LEN];
+        unsafe {
+            hash_many_fn(
+                parents.as_ptr() as _,
+                parents.len(),
+                1,
+                TEST_KEY_WORDS.as_ptr(),
+                counter,
+                false,
+                KEYED_HASH | PARENT,
+                0,
+                0,
+                test_parents_out.as_mut_ptr(),
+            );
+        }
+        for n in 0..NUM_INPUTS {
+            dbg!(n);
+            assert_eq!(
+                &portable_parents_out[n * OUT_LEN..][..OUT_LEN],
+                &test_parents_out[n * OUT_LEN..][..OUT_LEN]
+            );
+        }
+    }
+}
+
+// Testing the portable implementation against itself is circular, but why not.
+#[test]
+fn test_hash_many_portable() {
+    test_hash_many_fn(crate::ffi::blake3_hash_many_portable);
+}
+
+#[test]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_hash_many_sse2() {
+    if !crate::sse2_detected() {
+        return;
+    }
+    test_hash_many_fn(crate::ffi::x86::blake3_hash_many_sse2);
+}
+
+#[test]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_hash_many_sse41() {
+    if !crate::sse41_detected() {
+        return;
+    }
+    test_hash_many_fn(crate::ffi::x86::blake3_hash_many_sse41);
+}
+
+#[test]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_hash_many_avx2() {
+    if !crate::avx2_detected() {
+        return;
+    }
+    test_hash_many_fn(crate::ffi::x86::blake3_hash_many_avx2);
+}
+
+#[test]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_hash_many_avx512() {
+    if !crate::avx512_detected() {
+        return;
+    }
+    test_hash_many_fn(crate::ffi::x86::blake3_hash_many_avx512);
+}
+
+#[test]
+#[cfg(feature = "neon")]
+fn test_hash_many_neon() {
+    test_hash_many_fn(crate::ffi::neon::blake3_hash_many_neon);
+}
+
+#[allow(unused)]
+type XofManyFunction = unsafe extern "C" fn(
+    cv: *const u32,
+    block: *const u8,
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+    out: *mut u8,
+    outblocks: usize,
+);
+
+// A shared helper function for platform-specific tests.
+#[allow(unused)]
+pub fn test_xof_many_fn(xof_many_function: XofManyFunction) {
+    let mut block = [0; BLOCK_LEN];
+    let block_len = 42;
+    crate::test::paint_test_input(&mut block[..block_len]);
+    let cv = [40, 41, 42, 43, 44, 45, 46, 47];
+    let flags = KEYED_HASH;
+
+    // Test a few different initial counter values.
+    // - 0: The base case.
+    // - u32::MAX: The low word of the counter overflows for all inputs except the first.
+    // - i32::MAX: *No* overflow. But carry bugs in tricky SIMD code can screw this up, if you XOR
+    //   when you're supposed to ANDNOT...
+    let initial_counters = [0, u32::MAX as u64, i32::MAX as u64];
+    for counter in initial_counters {
+        dbg!(counter);
+
+        // 31 (16 + 8 + 4 + 2 + 1) outputs
+        const OUTPUT_SIZE: usize = 31 * BLOCK_LEN;
+
+        let mut portable_out = [0u8; OUTPUT_SIZE];
+        for (i, out_block) in portable_out.chunks_exact_mut(BLOCK_LEN).enumerate() {
+            unsafe {
+                crate::ffi::blake3_compress_xof_portable(
+                    cv.as_ptr(),
+                    block.as_ptr(),
+                    block_len as u8,
+                    counter + i as u64,
+                    flags,
+                    out_block.as_mut_ptr(),
+                );
+            }
+        }
+
+        let mut test_out = [0u8; OUTPUT_SIZE];
+        unsafe {
+            xof_many_function(
+                cv.as_ptr(),
+                block.as_ptr(),
+                block_len as u8,
+                counter,
+                flags,
+                test_out.as_mut_ptr(),
+                OUTPUT_SIZE / BLOCK_LEN,
+            );
+        }
+
+        assert_eq!(portable_out, test_out);
+    }
+
+    // Test that xof_many doesn't write more blocks than requested. Note that the current assembly
+    // implementation always outputs at least one block, so we don't test the zero case.
+    for block_count in 1..=32 {
+        let mut array = [0; BLOCK_LEN * 33];
+        let output_start = 17;
+        let output_len = block_count * BLOCK_LEN;
+        let output_end = output_start + output_len;
+        let output = &mut array[output_start..output_end];
+        unsafe {
+            xof_many_function(
+                cv.as_ptr(),
+                block.as_ptr(),
+                block_len as u8,
+                0,
+                flags,
+                output.as_mut_ptr(),
+                block_count,
+            );
+        }
+        for i in 0..array.len() {
+            if i < output_start || output_end <= i {
+                assert_eq!(0, array[i], "index {i}");
+            }
+        }
+    }
+}
+
+#[test]
+#[cfg(unix)]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+fn test_xof_many_avx512() {
+    if !crate::avx512_detected() {
+        return;
+    }
+    test_xof_many_fn(crate::ffi::x86::blake3_xof_many_avx512);
+}
+
+#[test]
+fn test_compare_reference_impl() {
+    const OUT: usize = 303; // more than 64, not a multiple of 4
+    let mut input_buf = [0; TEST_CASES_MAX];
+    paint_test_input(&mut input_buf);
+    for &case in TEST_CASES {
+        let input = &input_buf[..case];
+        dbg!(case);
+
+        // regular
+        {
+            let mut reference_hasher = reference_impl::Hasher::new();
+            reference_hasher.update(input);
+            let mut expected_out = [0; OUT];
+            reference_hasher.finalize(&mut expected_out);
+
+            let mut test_hasher = crate::Hasher::new();
+            test_hasher.update(input);
+            let mut test_out = [0; OUT];
+            test_hasher.finalize(&mut test_out);
+            assert_eq!(test_out[..], expected_out[..]);
+
+            #[cfg(feature = "tbb")]
+            {
+                let mut tbb_hasher = crate::Hasher::new();
+                tbb_hasher.update_tbb(input);
+                let mut tbb_out = [0; OUT];
+                tbb_hasher.finalize(&mut tbb_out);
+                assert_eq!(tbb_out[..], expected_out[..]);
+            }
+        }
+
+        // keyed
+        {
+            let mut reference_hasher = reference_impl::Hasher::new_keyed(&TEST_KEY);
+            reference_hasher.update(input);
+            let mut expected_out = [0; OUT];
+            reference_hasher.finalize(&mut expected_out);
+
+            let mut test_hasher = crate::Hasher::new_keyed(&TEST_KEY);
+            test_hasher.update(input);
+            let mut test_out = [0; OUT];
+            test_hasher.finalize(&mut test_out);
+            assert_eq!(test_out[..], expected_out[..]);
+
+            #[cfg(feature = "tbb")]
+            {
+                let mut tbb_hasher = crate::Hasher::new_keyed(&TEST_KEY);
+                tbb_hasher.update_tbb(input);
+                let mut tbb_out = [0; OUT];
+                tbb_hasher.finalize(&mut tbb_out);
+                assert_eq!(tbb_out[..], expected_out[..]);
+            }
+        }
+
+        // derive_key
+        {
+            let context = "BLAKE3 2019-12-27 16:13:59 example context (not the test vector one)";
+            let mut reference_hasher = reference_impl::Hasher::new_derive_key(context);
+            reference_hasher.update(input);
+            let mut expected_out = [0; OUT];
+            reference_hasher.finalize(&mut expected_out);
+
+            // the regular C string API
+            let mut test_hasher = crate::Hasher::new_derive_key(context);
+            test_hasher.update(input);
+            let mut test_out = [0; OUT];
+            test_hasher.finalize(&mut test_out);
+            assert_eq!(test_out[..], expected_out[..]);
+
+            // the raw bytes API
+            let mut test_hasher_raw = crate::Hasher::new_derive_key_raw(context.as_bytes());
+            test_hasher_raw.update(input);
+            let mut test_out_raw = [0; OUT];
+            test_hasher_raw.finalize(&mut test_out_raw);
+            assert_eq!(test_out_raw[..], expected_out[..]);
+
+            #[cfg(feature = "tbb")]
+            {
+                let mut tbb_hasher = crate::Hasher::new_derive_key(context);
+                tbb_hasher.update_tbb(input);
+                let mut tbb_out = [0; OUT];
+                tbb_hasher.finalize(&mut tbb_out);
+                assert_eq!(tbb_out[..], expected_out[..]);
+            }
+        }
+    }
+}
+
+fn reference_hash(input: &[u8]) -> [u8; OUT_LEN] {
+    let mut hasher = reference_impl::Hasher::new();
+    hasher.update(input);
+    let mut bytes = [0; OUT_LEN];
+    hasher.finalize(&mut bytes);
+    bytes.into()
+}
+
+#[test]
+fn test_compare_update_multiple() {
+    // Don't use all the long test cases here, since that's unnecessarily slow
+    // in debug mode.
+    let mut short_test_cases = TEST_CASES;
+    while *short_test_cases.last().unwrap() > 4 * CHUNK_LEN {
+        short_test_cases = &short_test_cases[..short_test_cases.len() - 1];
+    }
+    assert_eq!(*short_test_cases.last().unwrap(), 4 * CHUNK_LEN);
+
+    let mut input_buf = [0; 2 * TEST_CASES_MAX];
+    paint_test_input(&mut input_buf);
+
+    for &first_update in short_test_cases {
+        dbg!(first_update);
+        let first_input = &input_buf[..first_update];
+        let mut test_hasher = crate::Hasher::new();
+        test_hasher.update(first_input);
+
+        for &second_update in short_test_cases {
+            dbg!(second_update);
+            let second_input = &input_buf[first_update..][..second_update];
+            let total_input = &input_buf[..first_update + second_update];
+
+            // Clone the hasher with first_update bytes already written, so
+            // that the next iteration can reuse it.
+            let mut test_hasher = test_hasher.clone();
+            test_hasher.update(second_input);
+            let mut test_out = [0; OUT_LEN];
+            test_hasher.finalize(&mut test_out);
+
+            let expected = reference_hash(total_input);
+            assert_eq!(expected, test_out);
+        }
+    }
+}
+
+#[test]
+fn test_fuzz_hasher() {
+    const INPUT_MAX: usize = 4 * CHUNK_LEN;
+    let mut input_buf = [0; 3 * INPUT_MAX];
+    paint_test_input(&mut input_buf);
+
+    // Don't do too many iterations in debug mode, to keep the tests under a
+    // second or so. CI should run tests in release mode also. Provide an
+    // environment variable for specifying a larger number of fuzz iterations.
+    let num_tests = if cfg!(debug_assertions) { 100 } else { 10_000 };
+
+    // Use a fixed RNG seed for reproducibility.
+    let mut rng = rand_chacha::ChaCha8Rng::from_seed([1; 32]);
+    for _num_test in 0..num_tests {
+        dbg!(_num_test);
+        let mut hasher = crate::Hasher::new();
+        let mut total_input = 0;
+        // For each test, write 3 inputs of random length.
+        for _ in 0..3 {
+            let input_len = rng.random_range(0..INPUT_MAX + 1);
+            dbg!(input_len);
+            let input = &input_buf[total_input..][..input_len];
+            hasher.update(input);
+            total_input += input_len;
+        }
+        let expected = reference_hash(&input_buf[..total_input]);
+        let mut test_out = [0; 32];
+        hasher.finalize(&mut test_out);
+        assert_eq!(expected, test_out);
+    }
+}
+
+#[test]
+fn test_finalize_seek() {
+    let mut expected = [0; 1000];
+    {
+        let mut reference_hasher = reference_impl::Hasher::new();
+        reference_hasher.update(b"foobarbaz");
+        reference_hasher.finalize(&mut expected);
+    }
+
+    let mut test_hasher = crate::Hasher::new();
+    test_hasher.update(b"foobarbaz");
+
+    let mut out = [0; 103];
+    for &seek in &[0, 1, 7, 59, 63, 64, 65, 501, expected.len() - out.len()] {
+        dbg!(seek);
+        test_hasher.finalize_seek(seek as u64, &mut out);
+        assert_eq!(&expected[seek..][..out.len()], &out[..]);
+    }
+}
+
+#[test]
+fn test_reset() {
+    {
+        let mut hasher = crate::Hasher::new();
+        hasher.update(&[42; 3 * CHUNK_LEN + 7]);
+        hasher.reset();
+        hasher.update(&[42; CHUNK_LEN + 3]);
+        let mut output = [0; 32];
+        hasher.finalize(&mut output);
+
+        let mut reference_hasher = reference_impl::Hasher::new();
+        reference_hasher.update(&[42; CHUNK_LEN + 3]);
+        let mut reference_hash = [0; 32];
+        reference_hasher.finalize(&mut reference_hash);
+
+        assert_eq!(reference_hash, output);
+    }
+    {
+        let key = &[99; 32];
+        let mut hasher = crate::Hasher::new_keyed(key);
+        hasher.update(&[42; 3 * CHUNK_LEN + 7]);
+        hasher.reset();
+        hasher.update(&[42; CHUNK_LEN + 3]);
+        let mut output = [0; 32];
+        hasher.finalize(&mut output);
+
+        let mut reference_hasher = reference_impl::Hasher::new_keyed(key);
+        reference_hasher.update(&[42; CHUNK_LEN + 3]);
+        let mut reference_hash = [0; 32];
+        reference_hasher.finalize(&mut reference_hash);
+
+        assert_eq!(reference_hash, output);
+    }
+    {
+        let context = "BLAKE3 2020-02-12 10:20:58 reset test";
+        let mut hasher = crate::Hasher::new_derive_key(context);
+        hasher.update(&[42; 3 * CHUNK_LEN + 7]);
+        hasher.reset();
+        hasher.update(&[42; CHUNK_LEN + 3]);
+        let mut output = [0; 32];
+        hasher.finalize(&mut output);
+
+        let mut reference_hasher = reference_impl::Hasher::new_derive_key(context);
+        reference_hasher.update(&[42; CHUNK_LEN + 3]);
+        let mut reference_hash = [0; 32];
+        reference_hasher.finalize(&mut reference_hash);
+
+        assert_eq!(reference_hash, output);
+    }
+}
diff --git a/thirdparty/blake3/c/blake3_dispatch.c b/thirdparty/blake3/c/blake3_dispatch.c
new file mode 100644
index 000000000..52db90586
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_dispatch.c
@@ -0,0 +1,332 @@
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+
+#include "blake3_impl.h"
+
+#if defined(_MSC_VER)
+#include <Windows.h>
+#endif
+
+#if defined(IS_X86)
+#if defined(_MSC_VER)
+#include <intrin.h>
+#elif defined(__GNUC__)
+#include <immintrin.h>
+#else
+#undef IS_X86 /* Unimplemented! */
+#endif
+#endif
+
+#if !defined(BLAKE3_ATOMICS)
+#if defined(__has_include)
+#if __has_include(<stdatomic.h>) && !defined(_MSC_VER)
+#define BLAKE3_ATOMICS 1
+#else
+#define BLAKE3_ATOMICS 0
+#endif /* __has_include(<stdatomic.h>) && !defined(_MSC_VER) */
+#else
+#define BLAKE3_ATOMICS 0
+#endif /* defined(__has_include) */
+#endif /* BLAKE3_ATOMICS */
+
+#if BLAKE3_ATOMICS
+#define ATOMIC_INT _Atomic int
+#define ATOMIC_LOAD(x) x
+#define ATOMIC_STORE(x, y) x = y
+#elif defined(_MSC_VER)
+#define ATOMIC_INT LONG
+#define ATOMIC_LOAD(x) InterlockedOr(&x, 0)
+#define ATOMIC_STORE(x, y) InterlockedExchange(&x, y)
+#else
+#define ATOMIC_INT int
+#define ATOMIC_LOAD(x) x
+#define ATOMIC_STORE(x, y) x = y
+#endif
+
+#define MAYBE_UNUSED(x) (void)((x))
+
+#if defined(IS_X86)
+static uint64_t xgetbv(void) {
+#if defined(_MSC_VER)
+  return _xgetbv(0);
+#else
+  uint32_t eax = 0, edx = 0;
+  __asm__ __volatile__("xgetbv\n" : "=a"(eax), "=d"(edx) : "c"(0));
+  return ((uint64_t)edx << 32) | eax;
+#endif
+}
+
+static void cpuid(uint32_t out[4], uint32_t id) {
+#if defined(_MSC_VER)
+  __cpuid((int *)out, id);
+#elif defined(__i386__) || defined(_M_IX86)
+  __asm__ __volatile__("movl %%ebx, %1\n"
+                       "cpuid\n"
+                       "xchgl %1, %%ebx\n"
+                       : "=a"(out[0]), "=r"(out[1]), "=c"(out[2]), "=d"(out[3])
+                       : "a"(id));
+#else
+  __asm__ __volatile__("cpuid\n"
+                       : "=a"(out[0]), "=b"(out[1]), "=c"(out[2]), "=d"(out[3])
+                       : "a"(id));
+#endif
+}
+
+static void cpuidex(uint32_t out[4], uint32_t id, uint32_t sid) {
+#if defined(_MSC_VER)
+  __cpuidex((int *)out, id, sid);
+#elif defined(__i386__) || defined(_M_IX86)
+  __asm__ __volatile__("movl %%ebx, %1\n"
+                       "cpuid\n"
+                       "xchgl %1, %%ebx\n"
+                       : "=a"(out[0]), "=r"(out[1]), "=c"(out[2]), "=d"(out[3])
+                       : "a"(id), "c"(sid));
+#else
+  __asm__ __volatile__("cpuid\n"
+                       : "=a"(out[0]), "=b"(out[1]), "=c"(out[2]), "=d"(out[3])
+                       : "a"(id), "c"(sid));
+#endif
+}
+
+#endif
+
+enum cpu_feature {
+  SSE2 = 1 << 0,
+  SSSE3 = 1 << 1,
+  SSE41 = 1 << 2,
+  AVX = 1 << 3,
+  AVX2 = 1 << 4,
+  AVX512F = 1 << 5,
+  AVX512VL = 1 << 6,
+  /* ... */
+  UNDEFINED = 1 << 30
+};
+
+#if !defined(BLAKE3_TESTING)
+static /* Allow the variable to be controlled manually for testing */
+#endif
+    ATOMIC_INT g_cpu_features = UNDEFINED;
+
+#if !defined(BLAKE3_TESTING)
+static
+#endif
+    enum cpu_feature
+    get_cpu_features(void) {
+
+  /* If TSAN detects a data race here, try compiling with -DBLAKE3_ATOMICS=1 */
+  enum cpu_feature features = ATOMIC_LOAD(g_cpu_features);
+  if (features != UNDEFINED) {
+    return features;
+  } else {
+#if defined(IS_X86)
+    uint32_t regs[4] = {0};
+    uint32_t *eax = &regs[0], *ebx = &regs[1], *ecx = &regs[2], *edx = &regs[3];
+    (void)edx;
+    features = 0;
+    cpuid(regs, 0);
+    const int max_id = *eax;
+    cpuid(regs, 1);
+#if defined(__amd64__) || defined(_M_X64)
+    features |= SSE2;
+#else
+    if (*edx & (1UL << 26))
+      features |= SSE2;
+#endif
+    if (*ecx & (1UL << 9))
+      features |= SSSE3;
+    if (*ecx & (1UL << 19))
+      features |= SSE41;
+
+    if (*ecx & (1UL << 27)) { // OSXSAVE
+      const uint64_t mask = xgetbv();
+      if ((mask & 6) == 6) { // SSE and AVX states
+        if (*ecx & (1UL << 28))
+          features |= AVX;
+        if (max_id >= 7) {
+          cpuidex(regs, 7, 0);
+          if (*ebx & (1UL << 5))
+            features |= AVX2;
+          if ((mask & 224) == 224) { // Opmask, ZMM_Hi256, Hi16_Zmm
+            if (*ebx & (1UL << 31))
+              features |= AVX512VL;
+            if (*ebx & (1UL << 16))
+              features |= AVX512F;
+          }
+        }
+      }
+    }
+    ATOMIC_STORE(g_cpu_features, features);
+    return features;
+#else
+    /* How to detect NEON? */
+    return 0;
+#endif
+  }
+}
+
+void blake3_compress_in_place(uint32_t cv[8],
+                              const uint8_t block[BLAKE3_BLOCK_LEN],
+                              uint8_t block_len, uint64_t counter,
+                              uint8_t flags) {
+#if defined(IS_X86)
+  const enum cpu_feature features = get_cpu_features();
+  MAYBE_UNUSED(features);
+#if !defined(BLAKE3_NO_AVX512)
+  if (features & AVX512VL) {
+    blake3_compress_in_place_avx512(cv, block, block_len, counter, flags);
+    return;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE41)
+  if (features & SSE41) {
+    blake3_compress_in_place_sse41(cv, block, block_len, counter, flags);
+    return;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE2)
+  if (features & SSE2) {
+    blake3_compress_in_place_sse2(cv, block, block_len, counter, flags);
+    return;
+  }
+#endif
+#endif
+  blake3_compress_in_place_portable(cv, block, block_len, counter, flags);
+}
+
+void blake3_compress_xof(const uint32_t cv[8],
+                         const uint8_t block[BLAKE3_BLOCK_LEN],
+                         uint8_t block_len, uint64_t counter, uint8_t flags,
+                         uint8_t out[64]) {
+#if defined(IS_X86)
+  const enum cpu_feature features = get_cpu_features();
+  MAYBE_UNUSED(features);
+#if !defined(BLAKE3_NO_AVX512)
+  if (features & AVX512VL) {
+    blake3_compress_xof_avx512(cv, block, block_len, counter, flags, out);
+    return;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE41)
+  if (features & SSE41) {
+    blake3_compress_xof_sse41(cv, block, block_len, counter, flags, out);
+    return;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE2)
+  if (features & SSE2) {
+    blake3_compress_xof_sse2(cv, block, block_len, counter, flags, out);
+    return;
+  }
+#endif
+#endif
+  blake3_compress_xof_portable(cv, block, block_len, counter, flags, out);
+}
+
+
+void blake3_xof_many(const uint32_t cv[8],
+                     const uint8_t block[BLAKE3_BLOCK_LEN],
+                     uint8_t block_len, uint64_t counter, uint8_t flags,
+                     uint8_t out[64], size_t outblocks) {
+  if (outblocks == 0) {
+    // The current assembly implementation always outputs at least 1 block.
+    return;
+  }
+#if defined(IS_X86)
+  const enum cpu_feature features = get_cpu_features();
+  MAYBE_UNUSED(features);
+#if !defined(_WIN32) && !defined(BLAKE3_NO_AVX512)
+  if (features & AVX512VL) {
+    blake3_xof_many_avx512(cv, block, block_len, counter, flags, out, outblocks);
+    return;
+  }
+#endif
+#endif
+  for(size_t i = 0; i < outblocks; ++i) {
+    blake3_compress_xof(cv, block, block_len, counter + i, flags, out + 64*i);
+  }
+}
+
+void blake3_hash_many(const uint8_t *const *inputs, size_t num_inputs,
+                      size_t blocks, const uint32_t key[8], uint64_t counter,
+                      bool increment_counter, uint8_t flags,
+                      uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
+#if defined(IS_X86)
+  const enum cpu_feature features = get_cpu_features();
+  MAYBE_UNUSED(features);
+#if !defined(BLAKE3_NO_AVX512)
+  if ((features & (AVX512F|AVX512VL)) == (AVX512F|AVX512VL)) {
+    blake3_hash_many_avx512(inputs, num_inputs, blocks, key, counter,
+                            increment_counter, flags, flags_start, flags_end,
+                            out);
+    return;
+  }
+#endif
+#if !defined(BLAKE3_NO_AVX2)
+  if (features & AVX2) {
+    blake3_hash_many_avx2(inputs, num_inputs, blocks, key, counter,
+                          increment_counter, flags, flags_start, flags_end,
+                          out);
+    return;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE41)
+  if (features & SSE41) {
+    blake3_hash_many_sse41(inputs, num_inputs, blocks, key, counter,
+                           increment_counter, flags, flags_start, flags_end,
+                           out);
+    return;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE2)
+  if (features & SSE2) {
+    blake3_hash_many_sse2(inputs, num_inputs, blocks, key, counter,
+                          increment_counter, flags, flags_start, flags_end,
+                          out);
+    return;
+  }
+#endif
+#endif
+
+#if BLAKE3_USE_NEON == 1
+  blake3_hash_many_neon(inputs, num_inputs, blocks, key, counter,
+                        increment_counter, flags, flags_start, flags_end, out);
+  return;
+#endif
+
+  blake3_hash_many_portable(inputs, num_inputs, blocks, key, counter,
+                            increment_counter, flags, flags_start, flags_end,
+                            out);
+}
+
+// The dynamically detected SIMD degree of the current platform.
+size_t blake3_simd_degree(void) {
+#if defined(IS_X86)
+  const enum cpu_feature features = get_cpu_features();
+  MAYBE_UNUSED(features);
+#if !defined(BLAKE3_NO_AVX512)
+  if ((features & (AVX512F|AVX512VL)) == (AVX512F|AVX512VL)) {
+    return 16;
+  }
+#endif
+#if !defined(BLAKE3_NO_AVX2)
+  if (features & AVX2) {
+    return 8;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE41)
+  if (features & SSE41) {
+    return 4;
+  }
+#endif
+#if !defined(BLAKE3_NO_SSE2)
+  if (features & SSE2) {
+    return 4;
+  }
+#endif
+#endif
+#if BLAKE3_USE_NEON == 1
+  return 4;
+#endif
+  return 1;
+}
diff --git a/thirdparty/blake3/c/blake3_impl.h b/thirdparty/blake3/c/blake3_impl.h
new file mode 100644
index 000000000..facd59974
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_impl.h
@@ -0,0 +1,333 @@
+#ifndef BLAKE3_IMPL_H
+#define BLAKE3_IMPL_H
+
+#include <assert.h>
+#include <stdbool.h>
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+#include "blake3.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+// internal flags
+enum blake3_flags {
+  CHUNK_START         = 1 << 0,
+  CHUNK_END           = 1 << 1,
+  PARENT              = 1 << 2,
+  ROOT                = 1 << 3,
+  KEYED_HASH          = 1 << 4,
+  DERIVE_KEY_CONTEXT  = 1 << 5,
+  DERIVE_KEY_MATERIAL = 1 << 6,
+};
+
+// This C implementation tries to support recent versions of GCC, Clang, and
+// MSVC.
+#if defined(_MSC_VER)
+#define INLINE static __forceinline
+#else
+#define INLINE static inline __attribute__((always_inline))
+#endif
+
+#ifdef __cplusplus
+#define NOEXCEPT noexcept
+#else
+#define NOEXCEPT
+#endif
+
+#if (defined(__x86_64__) || defined(_M_X64)) && !defined(_M_ARM64EC)
+#define IS_X86
+#define IS_X86_64
+#endif
+
+#if defined(__i386__) || defined(_M_IX86)
+#define IS_X86
+#define IS_X86_32
+#endif
+
+#if defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
+#define IS_AARCH64
+#endif
+
+#if defined(IS_X86)
+#if defined(_MSC_VER)
+#include <intrin.h>
+#endif
+#endif
+
+#if !defined(BLAKE3_USE_NEON) 
+  // If BLAKE3_USE_NEON not manually set, autodetect based on AArch64ness
+  #if defined(IS_AARCH64)
+    #if defined(__ARM_BIG_ENDIAN)
+      #define BLAKE3_USE_NEON 0
+    #else
+      #define BLAKE3_USE_NEON 1
+    #endif
+  #else
+    #define BLAKE3_USE_NEON 0
+  #endif
+#endif
+
+#if defined(IS_X86)
+#define MAX_SIMD_DEGREE 16
+#elif BLAKE3_USE_NEON == 1
+#define MAX_SIMD_DEGREE 4
+#else
+#define MAX_SIMD_DEGREE 1
+#endif
+
+// There are some places where we want a static size that's equal to the
+// MAX_SIMD_DEGREE, but also at least 2.
+#define MAX_SIMD_DEGREE_OR_2 (MAX_SIMD_DEGREE > 2 ? MAX_SIMD_DEGREE : 2)
+
+static const uint32_t IV[8] = {0x6A09E667UL, 0xBB67AE85UL, 0x3C6EF372UL,
+                               0xA54FF53AUL, 0x510E527FUL, 0x9B05688CUL,
+                               0x1F83D9ABUL, 0x5BE0CD19UL};
+
+static const uint8_t MSG_SCHEDULE[7][16] = {
+    {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15},
+    {2, 6, 3, 10, 7, 0, 4, 13, 1, 11, 12, 5, 9, 14, 15, 8},
+    {3, 4, 10, 12, 13, 2, 7, 14, 6, 5, 9, 0, 11, 15, 8, 1},
+    {10, 7, 12, 9, 14, 3, 13, 15, 4, 0, 11, 2, 5, 8, 1, 6},
+    {12, 13, 9, 11, 15, 10, 14, 8, 7, 2, 5, 3, 0, 1, 6, 4},
+    {9, 14, 11, 5, 8, 12, 15, 1, 13, 3, 0, 10, 2, 6, 4, 7},
+    {11, 15, 5, 0, 1, 9, 8, 6, 14, 10, 2, 12, 3, 4, 7, 13},
+};
+
+/* Find index of the highest set bit */
+/* x is assumed to be nonzero.       */
+static unsigned int highest_one(uint64_t x) {
+#if defined(__GNUC__) || defined(__clang__)
+  return 63 ^ (unsigned int)__builtin_clzll(x);
+#elif defined(_MSC_VER) && defined(IS_X86_64)
+  unsigned long index;
+  _BitScanReverse64(&index, x);
+  return index;
+#elif defined(_MSC_VER) && defined(IS_X86_32)
+  if(x >> 32) {
+    unsigned long index;
+    _BitScanReverse(&index, (unsigned long)(x >> 32));
+    return 32 + index;
+  } else {
+    unsigned long index;
+    _BitScanReverse(&index, (unsigned long)x);
+    return index;
+  }
+#else
+  unsigned int c = 0;
+  if(x & 0xffffffff00000000ULL) { x >>= 32; c += 32; }
+  if(x & 0x00000000ffff0000ULL) { x >>= 16; c += 16; }
+  if(x & 0x000000000000ff00ULL) { x >>=  8; c +=  8; }
+  if(x & 0x00000000000000f0ULL) { x >>=  4; c +=  4; }
+  if(x & 0x000000000000000cULL) { x >>=  2; c +=  2; }
+  if(x & 0x0000000000000002ULL) {           c +=  1; }
+  return c;
+#endif
+}
+
+// Count the number of 1 bits.
+INLINE unsigned int popcnt(uint64_t x) {
+#if defined(__GNUC__) || defined(__clang__)
+  return (unsigned int)__builtin_popcountll(x);
+#else
+  unsigned int count = 0;
+  while (x != 0) {
+    count += 1;
+    x &= x - 1;
+  }
+  return count;
+#endif
+}
+
+// Largest power of two less than or equal to x. As a special case, returns 1
+// when x is 0. 
+INLINE uint64_t round_down_to_power_of_2(uint64_t x) {
+  return 1ULL << highest_one(x | 1);
+}
+
+INLINE uint32_t counter_low(uint64_t counter) { return (uint32_t)counter; }
+
+INLINE uint32_t counter_high(uint64_t counter) {
+  return (uint32_t)(counter >> 32);
+}
+
+INLINE uint32_t load32(const void *src) {
+  const uint8_t *p = (const uint8_t *)src;
+  return ((uint32_t)(p[0]) << 0) | ((uint32_t)(p[1]) << 8) |
+         ((uint32_t)(p[2]) << 16) | ((uint32_t)(p[3]) << 24);
+}
+
+INLINE void load_key_words(const uint8_t key[BLAKE3_KEY_LEN],
+                           uint32_t key_words[8]) {
+  key_words[0] = load32(&key[0 * 4]);
+  key_words[1] = load32(&key[1 * 4]);
+  key_words[2] = load32(&key[2 * 4]);
+  key_words[3] = load32(&key[3 * 4]);
+  key_words[4] = load32(&key[4 * 4]);
+  key_words[5] = load32(&key[5 * 4]);
+  key_words[6] = load32(&key[6 * 4]);
+  key_words[7] = load32(&key[7 * 4]);
+}
+
+INLINE void load_block_words(const uint8_t block[BLAKE3_BLOCK_LEN],
+                             uint32_t block_words[16]) {
+  for (size_t i = 0; i < 16; i++) {
+      block_words[i] = load32(&block[i * 4]);
+  }
+}
+
+INLINE void store32(void *dst, uint32_t w) {
+  uint8_t *p = (uint8_t *)dst;
+  p[0] = (uint8_t)(w >> 0);
+  p[1] = (uint8_t)(w >> 8);
+  p[2] = (uint8_t)(w >> 16);
+  p[3] = (uint8_t)(w >> 24);
+}
+
+INLINE void store_cv_words(uint8_t bytes_out[32], uint32_t cv_words[8]) {
+  store32(&bytes_out[0 * 4], cv_words[0]);
+  store32(&bytes_out[1 * 4], cv_words[1]);
+  store32(&bytes_out[2 * 4], cv_words[2]);
+  store32(&bytes_out[3 * 4], cv_words[3]);
+  store32(&bytes_out[4 * 4], cv_words[4]);
+  store32(&bytes_out[5 * 4], cv_words[5]);
+  store32(&bytes_out[6 * 4], cv_words[6]);
+  store32(&bytes_out[7 * 4], cv_words[7]);
+}
+
+void blake3_compress_in_place(uint32_t cv[8],
+                              const uint8_t block[BLAKE3_BLOCK_LEN],
+                              uint8_t block_len, uint64_t counter,
+                              uint8_t flags);
+
+void blake3_compress_xof(const uint32_t cv[8],
+                         const uint8_t block[BLAKE3_BLOCK_LEN],
+                         uint8_t block_len, uint64_t counter, uint8_t flags,
+                         uint8_t out[64]);
+
+void blake3_xof_many(const uint32_t cv[8],
+                     const uint8_t block[BLAKE3_BLOCK_LEN],
+                     uint8_t block_len, uint64_t counter, uint8_t flags,
+                     uint8_t out[64], size_t outblocks);
+
+void blake3_hash_many(const uint8_t *const *inputs, size_t num_inputs,
+                      size_t blocks, const uint32_t key[8], uint64_t counter,
+                      bool increment_counter, uint8_t flags,
+                      uint8_t flags_start, uint8_t flags_end, uint8_t *out);
+
+size_t blake3_simd_degree(void);
+
+BLAKE3_PRIVATE size_t blake3_compress_subtree_wide(const uint8_t *input, size_t input_len,
+                                                   const uint32_t key[8],
+                                                   uint64_t chunk_counter, uint8_t flags,
+                                                   uint8_t *out, bool use_tbb);
+
+#if defined(BLAKE3_USE_TBB)
+BLAKE3_PRIVATE void blake3_compress_subtree_wide_join_tbb(
+    // shared params
+    const uint32_t key[8], uint8_t flags, bool use_tbb,
+    // left-hand side params
+    const uint8_t *l_input, size_t l_input_len, uint64_t l_chunk_counter,
+    uint8_t *l_cvs, size_t *l_n,
+    // right-hand side params
+    const uint8_t *r_input, size_t r_input_len, uint64_t r_chunk_counter,
+    uint8_t *r_cvs, size_t *r_n) NOEXCEPT;
+#endif
+
+// Declarations for implementation-specific functions.
+void blake3_compress_in_place_portable(uint32_t cv[8],
+                                       const uint8_t block[BLAKE3_BLOCK_LEN],
+                                       uint8_t block_len, uint64_t counter,
+                                       uint8_t flags);
+
+void blake3_compress_xof_portable(const uint32_t cv[8],
+                                  const uint8_t block[BLAKE3_BLOCK_LEN],
+                                  uint8_t block_len, uint64_t counter,
+                                  uint8_t flags, uint8_t out[64]);
+
+void blake3_hash_many_portable(const uint8_t *const *inputs, size_t num_inputs,
+                               size_t blocks, const uint32_t key[8],
+                               uint64_t counter, bool increment_counter,
+                               uint8_t flags, uint8_t flags_start,
+                               uint8_t flags_end, uint8_t *out);
+
+#if defined(IS_X86)
+#if !defined(BLAKE3_NO_SSE2)
+void blake3_compress_in_place_sse2(uint32_t cv[8],
+                                   const uint8_t block[BLAKE3_BLOCK_LEN],
+                                   uint8_t block_len, uint64_t counter,
+                                   uint8_t flags);
+void blake3_compress_xof_sse2(const uint32_t cv[8],
+                              const uint8_t block[BLAKE3_BLOCK_LEN],
+                              uint8_t block_len, uint64_t counter,
+                              uint8_t flags, uint8_t out[64]);
+void blake3_hash_many_sse2(const uint8_t *const *inputs, size_t num_inputs,
+                           size_t blocks, const uint32_t key[8],
+                           uint64_t counter, bool increment_counter,
+                           uint8_t flags, uint8_t flags_start,
+                           uint8_t flags_end, uint8_t *out);
+#endif
+#if !defined(BLAKE3_NO_SSE41)
+void blake3_compress_in_place_sse41(uint32_t cv[8],
+                                    const uint8_t block[BLAKE3_BLOCK_LEN],
+                                    uint8_t block_len, uint64_t counter,
+                                    uint8_t flags);
+void blake3_compress_xof_sse41(const uint32_t cv[8],
+                               const uint8_t block[BLAKE3_BLOCK_LEN],
+                               uint8_t block_len, uint64_t counter,
+                               uint8_t flags, uint8_t out[64]);
+void blake3_hash_many_sse41(const uint8_t *const *inputs, size_t num_inputs,
+                            size_t blocks, const uint32_t key[8],
+                            uint64_t counter, bool increment_counter,
+                            uint8_t flags, uint8_t flags_start,
+                            uint8_t flags_end, uint8_t *out);
+#endif
+#if !defined(BLAKE3_NO_AVX2)
+void blake3_hash_many_avx2(const uint8_t *const *inputs, size_t num_inputs,
+                           size_t blocks, const uint32_t key[8],
+                           uint64_t counter, bool increment_counter,
+                           uint8_t flags, uint8_t flags_start,
+                           uint8_t flags_end, uint8_t *out);
+#endif
+#if !defined(BLAKE3_NO_AVX512)
+void blake3_compress_in_place_avx512(uint32_t cv[8],
+                                     const uint8_t block[BLAKE3_BLOCK_LEN],
+                                     uint8_t block_len, uint64_t counter,
+                                     uint8_t flags);
+
+void blake3_compress_xof_avx512(const uint32_t cv[8],
+                                const uint8_t block[BLAKE3_BLOCK_LEN],
+                                uint8_t block_len, uint64_t counter,
+                                uint8_t flags, uint8_t out[64]);
+
+void blake3_hash_many_avx512(const uint8_t *const *inputs, size_t num_inputs,
+                             size_t blocks, const uint32_t key[8],
+                             uint64_t counter, bool increment_counter,
+                             uint8_t flags, uint8_t flags_start,
+                             uint8_t flags_end, uint8_t *out);
+
+#if !defined(_WIN32)
+void blake3_xof_many_avx512(const uint32_t cv[8],
+                            const uint8_t block[BLAKE3_BLOCK_LEN],
+                            uint8_t block_len, uint64_t counter, uint8_t flags,
+                            uint8_t* out, size_t outblocks);
+#endif
+#endif
+#endif
+
+#if BLAKE3_USE_NEON == 1
+void blake3_hash_many_neon(const uint8_t *const *inputs, size_t num_inputs,
+                           size_t blocks, const uint32_t key[8],
+                           uint64_t counter, bool increment_counter,
+                           uint8_t flags, uint8_t flags_start,
+                           uint8_t flags_end, uint8_t *out);
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* BLAKE3_IMPL_H */
diff --git a/thirdparty/blake3/c/blake3_neon.c b/thirdparty/blake3/c/blake3_neon.c
new file mode 100644
index 000000000..53ce83c3b
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_neon.c
@@ -0,0 +1,366 @@
+#include "blake3_impl.h"
+
+#include <arm_neon.h>
+
+#ifdef __ARM_BIG_ENDIAN
+#error "This implementation only supports little-endian ARM."
+// It might be that all we need for big-endian support here is to get the loads
+// and stores right, but step zero would be finding a way to test it in CI.
+#endif
+
+INLINE uint32x4_t loadu_128(const uint8_t src[16]) {
+  // vld1q_u32 has alignment requirements. Don't use it.
+  return vreinterpretq_u32_u8(vld1q_u8(src));
+}
+
+INLINE void storeu_128(uint32x4_t src, uint8_t dest[16]) {
+  // vst1q_u32 has alignment requirements. Don't use it.
+  vst1q_u8(dest, vreinterpretq_u8_u32(src));
+}
+
+INLINE uint32x4_t add_128(uint32x4_t a, uint32x4_t b) {
+  return vaddq_u32(a, b);
+}
+
+INLINE uint32x4_t xor_128(uint32x4_t a, uint32x4_t b) {
+  return veorq_u32(a, b);
+}
+
+INLINE uint32x4_t set1_128(uint32_t x) { return vld1q_dup_u32(&x); }
+
+INLINE uint32x4_t set4(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
+  uint32_t array[4] = {a, b, c, d};
+  return vld1q_u32(array);
+}
+
+INLINE uint32x4_t rot16_128(uint32x4_t x) {
+  // The straightforward implementation would be two shifts and an or, but that's
+  // slower on microarchitectures we've tested. See
+  // https://github.com/BLAKE3-team/BLAKE3/pull/319.
+  // return vorrq_u32(vshrq_n_u32(x, 16), vshlq_n_u32(x, 32 - 16));
+  return vreinterpretq_u32_u16(vrev32q_u16(vreinterpretq_u16_u32(x)));
+}
+
+INLINE uint32x4_t rot12_128(uint32x4_t x) {
+  // See comment in rot16_128.
+  // return vorrq_u32(vshrq_n_u32(x, 12), vshlq_n_u32(x, 32 - 12));
+  return vsriq_n_u32(vshlq_n_u32(x, 32-12), x, 12);
+}
+
+INLINE uint32x4_t rot8_128(uint32x4_t x) {
+  // See comment in rot16_128.
+  // return vorrq_u32(vshrq_n_u32(x, 8), vshlq_n_u32(x, 32 - 8));
+#if defined(__clang__)
+  return vreinterpretq_u32_u8(__builtin_shufflevector(vreinterpretq_u8_u32(x), vreinterpretq_u8_u32(x), 1,2,3,0,5,6,7,4,9,10,11,8,13,14,15,12));
+#elif __GNUC__ * 10000 + __GNUC_MINOR__ * 100 >=40700
+  static const uint8x16_t r8 = {1,2,3,0,5,6,7,4,9,10,11,8,13,14,15,12};
+  return vreinterpretq_u32_u8(__builtin_shuffle(vreinterpretq_u8_u32(x), vreinterpretq_u8_u32(x), r8));
+#else 
+  return vsriq_n_u32(vshlq_n_u32(x, 32-8), x, 8);
+#endif
+}
+
+INLINE uint32x4_t rot7_128(uint32x4_t x) {
+  // See comment in rot16_128.
+  // return vorrq_u32(vshrq_n_u32(x, 7), vshlq_n_u32(x, 32 - 7));
+  return vsriq_n_u32(vshlq_n_u32(x, 32-7), x, 7);
+}
+
+// TODO: compress_neon
+
+// TODO: hash2_neon
+
+/*
+ * ----------------------------------------------------------------------------
+ * hash4_neon
+ * ----------------------------------------------------------------------------
+ */
+
+INLINE void round_fn4(uint32x4_t v[16], uint32x4_t m[16], size_t r) {
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
+  v[0] = add_128(v[0], v[4]);
+  v[1] = add_128(v[1], v[5]);
+  v[2] = add_128(v[2], v[6]);
+  v[3] = add_128(v[3], v[7]);
+  v[12] = xor_128(v[12], v[0]);
+  v[13] = xor_128(v[13], v[1]);
+  v[14] = xor_128(v[14], v[2]);
+  v[15] = xor_128(v[15], v[3]);
+  v[12] = rot16_128(v[12]);
+  v[13] = rot16_128(v[13]);
+  v[14] = rot16_128(v[14]);
+  v[15] = rot16_128(v[15]);
+  v[8] = add_128(v[8], v[12]);
+  v[9] = add_128(v[9], v[13]);
+  v[10] = add_128(v[10], v[14]);
+  v[11] = add_128(v[11], v[15]);
+  v[4] = xor_128(v[4], v[8]);
+  v[5] = xor_128(v[5], v[9]);
+  v[6] = xor_128(v[6], v[10]);
+  v[7] = xor_128(v[7], v[11]);
+  v[4] = rot12_128(v[4]);
+  v[5] = rot12_128(v[5]);
+  v[6] = rot12_128(v[6]);
+  v[7] = rot12_128(v[7]);
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
+  v[0] = add_128(v[0], v[4]);
+  v[1] = add_128(v[1], v[5]);
+  v[2] = add_128(v[2], v[6]);
+  v[3] = add_128(v[3], v[7]);
+  v[12] = xor_128(v[12], v[0]);
+  v[13] = xor_128(v[13], v[1]);
+  v[14] = xor_128(v[14], v[2]);
+  v[15] = xor_128(v[15], v[3]);
+  v[12] = rot8_128(v[12]);
+  v[13] = rot8_128(v[13]);
+  v[14] = rot8_128(v[14]);
+  v[15] = rot8_128(v[15]);
+  v[8] = add_128(v[8], v[12]);
+  v[9] = add_128(v[9], v[13]);
+  v[10] = add_128(v[10], v[14]);
+  v[11] = add_128(v[11], v[15]);
+  v[4] = xor_128(v[4], v[8]);
+  v[5] = xor_128(v[5], v[9]);
+  v[6] = xor_128(v[6], v[10]);
+  v[7] = xor_128(v[7], v[11]);
+  v[4] = rot7_128(v[4]);
+  v[5] = rot7_128(v[5]);
+  v[6] = rot7_128(v[6]);
+  v[7] = rot7_128(v[7]);
+
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
+  v[0] = add_128(v[0], v[5]);
+  v[1] = add_128(v[1], v[6]);
+  v[2] = add_128(v[2], v[7]);
+  v[3] = add_128(v[3], v[4]);
+  v[15] = xor_128(v[15], v[0]);
+  v[12] = xor_128(v[12], v[1]);
+  v[13] = xor_128(v[13], v[2]);
+  v[14] = xor_128(v[14], v[3]);
+  v[15] = rot16_128(v[15]);
+  v[12] = rot16_128(v[12]);
+  v[13] = rot16_128(v[13]);
+  v[14] = rot16_128(v[14]);
+  v[10] = add_128(v[10], v[15]);
+  v[11] = add_128(v[11], v[12]);
+  v[8] = add_128(v[8], v[13]);
+  v[9] = add_128(v[9], v[14]);
+  v[5] = xor_128(v[5], v[10]);
+  v[6] = xor_128(v[6], v[11]);
+  v[7] = xor_128(v[7], v[8]);
+  v[4] = xor_128(v[4], v[9]);
+  v[5] = rot12_128(v[5]);
+  v[6] = rot12_128(v[6]);
+  v[7] = rot12_128(v[7]);
+  v[4] = rot12_128(v[4]);
+  v[0] = add_128(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
+  v[1] = add_128(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
+  v[2] = add_128(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
+  v[3] = add_128(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
+  v[0] = add_128(v[0], v[5]);
+  v[1] = add_128(v[1], v[6]);
+  v[2] = add_128(v[2], v[7]);
+  v[3] = add_128(v[3], v[4]);
+  v[15] = xor_128(v[15], v[0]);
+  v[12] = xor_128(v[12], v[1]);
+  v[13] = xor_128(v[13], v[2]);
+  v[14] = xor_128(v[14], v[3]);
+  v[15] = rot8_128(v[15]);
+  v[12] = rot8_128(v[12]);
+  v[13] = rot8_128(v[13]);
+  v[14] = rot8_128(v[14]);
+  v[10] = add_128(v[10], v[15]);
+  v[11] = add_128(v[11], v[12]);
+  v[8] = add_128(v[8], v[13]);
+  v[9] = add_128(v[9], v[14]);
+  v[5] = xor_128(v[5], v[10]);
+  v[6] = xor_128(v[6], v[11]);
+  v[7] = xor_128(v[7], v[8]);
+  v[4] = xor_128(v[4], v[9]);
+  v[5] = rot7_128(v[5]);
+  v[6] = rot7_128(v[6]);
+  v[7] = rot7_128(v[7]);
+  v[4] = rot7_128(v[4]);
+}
+
+INLINE void transpose_vecs_128(uint32x4_t vecs[4]) {
+  // Individually transpose the four 2x2 sub-matrices in each corner.
+  uint32x4x2_t rows01 = vtrnq_u32(vecs[0], vecs[1]);
+  uint32x4x2_t rows23 = vtrnq_u32(vecs[2], vecs[3]);
+
+  // Swap the top-right and bottom-left 2x2s (which just got transposed).
+  vecs[0] =
+      vcombine_u32(vget_low_u32(rows01.val[0]), vget_low_u32(rows23.val[0]));
+  vecs[1] =
+      vcombine_u32(vget_low_u32(rows01.val[1]), vget_low_u32(rows23.val[1]));
+  vecs[2] =
+      vcombine_u32(vget_high_u32(rows01.val[0]), vget_high_u32(rows23.val[0]));
+  vecs[3] =
+      vcombine_u32(vget_high_u32(rows01.val[1]), vget_high_u32(rows23.val[1]));
+}
+
+INLINE void transpose_msg_vecs4(const uint8_t *const *inputs,
+                                size_t block_offset, uint32x4_t out[16]) {
+  out[0] = loadu_128(&inputs[0][block_offset + 0 * sizeof(uint32x4_t)]);
+  out[1] = loadu_128(&inputs[1][block_offset + 0 * sizeof(uint32x4_t)]);
+  out[2] = loadu_128(&inputs[2][block_offset + 0 * sizeof(uint32x4_t)]);
+  out[3] = loadu_128(&inputs[3][block_offset + 0 * sizeof(uint32x4_t)]);
+  out[4] = loadu_128(&inputs[0][block_offset + 1 * sizeof(uint32x4_t)]);
+  out[5] = loadu_128(&inputs[1][block_offset + 1 * sizeof(uint32x4_t)]);
+  out[6] = loadu_128(&inputs[2][block_offset + 1 * sizeof(uint32x4_t)]);
+  out[7] = loadu_128(&inputs[3][block_offset + 1 * sizeof(uint32x4_t)]);
+  out[8] = loadu_128(&inputs[0][block_offset + 2 * sizeof(uint32x4_t)]);
+  out[9] = loadu_128(&inputs[1][block_offset + 2 * sizeof(uint32x4_t)]);
+  out[10] = loadu_128(&inputs[2][block_offset + 2 * sizeof(uint32x4_t)]);
+  out[11] = loadu_128(&inputs[3][block_offset + 2 * sizeof(uint32x4_t)]);
+  out[12] = loadu_128(&inputs[0][block_offset + 3 * sizeof(uint32x4_t)]);
+  out[13] = loadu_128(&inputs[1][block_offset + 3 * sizeof(uint32x4_t)]);
+  out[14] = loadu_128(&inputs[2][block_offset + 3 * sizeof(uint32x4_t)]);
+  out[15] = loadu_128(&inputs[3][block_offset + 3 * sizeof(uint32x4_t)]);
+  transpose_vecs_128(&out[0]);
+  transpose_vecs_128(&out[4]);
+  transpose_vecs_128(&out[8]);
+  transpose_vecs_128(&out[12]);
+}
+
+INLINE void load_counters4(uint64_t counter, bool increment_counter,
+                           uint32x4_t *out_low, uint32x4_t *out_high) {
+  uint64_t mask = (increment_counter ? ~0 : 0);
+  *out_low = set4(
+      counter_low(counter + (mask & 0)), counter_low(counter + (mask & 1)),
+      counter_low(counter + (mask & 2)), counter_low(counter + (mask & 3)));
+  *out_high = set4(
+      counter_high(counter + (mask & 0)), counter_high(counter + (mask & 1)),
+      counter_high(counter + (mask & 2)), counter_high(counter + (mask & 3)));
+}
+
+void blake3_hash4_neon(const uint8_t *const *inputs, size_t blocks,
+                       const uint32_t key[8], uint64_t counter,
+                       bool increment_counter, uint8_t flags,
+                       uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
+  uint32x4_t h_vecs[8] = {
+      set1_128(key[0]), set1_128(key[1]), set1_128(key[2]), set1_128(key[3]),
+      set1_128(key[4]), set1_128(key[5]), set1_128(key[6]), set1_128(key[7]),
+  };
+  uint32x4_t counter_low_vec, counter_high_vec;
+  load_counters4(counter, increment_counter, &counter_low_vec,
+                 &counter_high_vec);
+  uint8_t block_flags = flags | flags_start;
+
+  for (size_t block = 0; block < blocks; block++) {
+    if (block + 1 == blocks) {
+      block_flags |= flags_end;
+    }
+    uint32x4_t block_len_vec = set1_128(BLAKE3_BLOCK_LEN);
+    uint32x4_t block_flags_vec = set1_128(block_flags);
+    uint32x4_t msg_vecs[16];
+    transpose_msg_vecs4(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);
+
+    uint32x4_t v[16] = {
+        h_vecs[0],       h_vecs[1],        h_vecs[2],       h_vecs[3],
+        h_vecs[4],       h_vecs[5],        h_vecs[6],       h_vecs[7],
+        set1_128(IV[0]), set1_128(IV[1]),  set1_128(IV[2]), set1_128(IV[3]),
+        counter_low_vec, counter_high_vec, block_len_vec,   block_flags_vec,
+    };
+    round_fn4(v, msg_vecs, 0);
+    round_fn4(v, msg_vecs, 1);
+    round_fn4(v, msg_vecs, 2);
+    round_fn4(v, msg_vecs, 3);
+    round_fn4(v, msg_vecs, 4);
+    round_fn4(v, msg_vecs, 5);
+    round_fn4(v, msg_vecs, 6);
+    h_vecs[0] = xor_128(v[0], v[8]);
+    h_vecs[1] = xor_128(v[1], v[9]);
+    h_vecs[2] = xor_128(v[2], v[10]);
+    h_vecs[3] = xor_128(v[3], v[11]);
+    h_vecs[4] = xor_128(v[4], v[12]);
+    h_vecs[5] = xor_128(v[5], v[13]);
+    h_vecs[6] = xor_128(v[6], v[14]);
+    h_vecs[7] = xor_128(v[7], v[15]);
+
+    block_flags = flags;
+  }
+
+  transpose_vecs_128(&h_vecs[0]);
+  transpose_vecs_128(&h_vecs[4]);
+  // The first four vecs now contain the first half of each output, and the
+  // second four vecs contain the second half of each output.
+  storeu_128(h_vecs[0], &out[0 * sizeof(uint32x4_t)]);
+  storeu_128(h_vecs[4], &out[1 * sizeof(uint32x4_t)]);
+  storeu_128(h_vecs[1], &out[2 * sizeof(uint32x4_t)]);
+  storeu_128(h_vecs[5], &out[3 * sizeof(uint32x4_t)]);
+  storeu_128(h_vecs[2], &out[4 * sizeof(uint32x4_t)]);
+  storeu_128(h_vecs[6], &out[5 * sizeof(uint32x4_t)]);
+  storeu_128(h_vecs[3], &out[6 * sizeof(uint32x4_t)]);
+  storeu_128(h_vecs[7], &out[7 * sizeof(uint32x4_t)]);
+}
+
+/*
+ * ----------------------------------------------------------------------------
+ * hash_many_neon
+ * ----------------------------------------------------------------------------
+ */
+
+void blake3_compress_in_place_portable(uint32_t cv[8],
+                                       const uint8_t block[BLAKE3_BLOCK_LEN],
+                                       uint8_t block_len, uint64_t counter,
+                                       uint8_t flags);
+
+INLINE void hash_one_neon(const uint8_t *input, size_t blocks,
+                          const uint32_t key[8], uint64_t counter,
+                          uint8_t flags, uint8_t flags_start, uint8_t flags_end,
+                          uint8_t out[BLAKE3_OUT_LEN]) {
+  uint32_t cv[8];
+  memcpy(cv, key, BLAKE3_KEY_LEN);
+  uint8_t block_flags = flags | flags_start;
+  while (blocks > 0) {
+    if (blocks == 1) {
+      block_flags |= flags_end;
+    }
+    // TODO: Implement compress_neon. However note that according to
+    // https://github.com/BLAKE2/BLAKE2/commit/7965d3e6e1b4193438b8d3a656787587d2579227,
+    // compress_neon might not be any faster than compress_portable.
+    blake3_compress_in_place_portable(cv, input, BLAKE3_BLOCK_LEN, counter,
+                                      block_flags);
+    input = &input[BLAKE3_BLOCK_LEN];
+    blocks -= 1;
+    block_flags = flags;
+  }
+  memcpy(out, cv, BLAKE3_OUT_LEN);
+}
+
+void blake3_hash_many_neon(const uint8_t *const *inputs, size_t num_inputs,
+                           size_t blocks, const uint32_t key[8],
+                           uint64_t counter, bool increment_counter,
+                           uint8_t flags, uint8_t flags_start,
+                           uint8_t flags_end, uint8_t *out) {
+  while (num_inputs >= 4) {
+    blake3_hash4_neon(inputs, blocks, key, counter, increment_counter, flags,
+                      flags_start, flags_end, out);
+    if (increment_counter) {
+      counter += 4;
+    }
+    inputs += 4;
+    num_inputs -= 4;
+    out = &out[4 * BLAKE3_OUT_LEN];
+  }
+  while (num_inputs > 0) {
+    hash_one_neon(inputs[0], blocks, key, counter, flags, flags_start,
+                  flags_end, out);
+    if (increment_counter) {
+      counter += 1;
+    }
+    inputs += 1;
+    num_inputs -= 1;
+    out = &out[BLAKE3_OUT_LEN];
+  }
+}
diff --git a/thirdparty/blake3/c/blake3_portable.c b/thirdparty/blake3/c/blake3_portable.c
new file mode 100644
index 000000000..062dd1b47
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_portable.c
@@ -0,0 +1,160 @@
+#include "blake3_impl.h"
+#include <string.h>
+
+INLINE uint32_t rotr32(uint32_t w, uint32_t c) {
+  return (w >> c) | (w << (32 - c));
+}
+
+INLINE void g(uint32_t *state, size_t a, size_t b, size_t c, size_t d,
+              uint32_t x, uint32_t y) {
+  state[a] = state[a] + state[b] + x;
+  state[d] = rotr32(state[d] ^ state[a], 16);
+  state[c] = state[c] + state[d];
+  state[b] = rotr32(state[b] ^ state[c], 12);
+  state[a] = state[a] + state[b] + y;
+  state[d] = rotr32(state[d] ^ state[a], 8);
+  state[c] = state[c] + state[d];
+  state[b] = rotr32(state[b] ^ state[c], 7);
+}
+
+INLINE void round_fn(uint32_t state[16], const uint32_t *msg, size_t round) {
+  // Select the message schedule based on the round.
+  const uint8_t *schedule = MSG_SCHEDULE[round];
+
+  // Mix the columns.
+  g(state, 0, 4, 8, 12, msg[schedule[0]], msg[schedule[1]]);
+  g(state, 1, 5, 9, 13, msg[schedule[2]], msg[schedule[3]]);
+  g(state, 2, 6, 10, 14, msg[schedule[4]], msg[schedule[5]]);
+  g(state, 3, 7, 11, 15, msg[schedule[6]], msg[schedule[7]]);
+
+  // Mix the rows.
+  g(state, 0, 5, 10, 15, msg[schedule[8]], msg[schedule[9]]);
+  g(state, 1, 6, 11, 12, msg[schedule[10]], msg[schedule[11]]);
+  g(state, 2, 7, 8, 13, msg[schedule[12]], msg[schedule[13]]);
+  g(state, 3, 4, 9, 14, msg[schedule[14]], msg[schedule[15]]);
+}
+
+INLINE void compress_pre(uint32_t state[16], const uint32_t cv[8],
+                         const uint8_t block[BLAKE3_BLOCK_LEN],
+                         uint8_t block_len, uint64_t counter, uint8_t flags) {
+  uint32_t block_words[16];
+  block_words[0] = load32(block + 4 * 0);
+  block_words[1] = load32(block + 4 * 1);
+  block_words[2] = load32(block + 4 * 2);
+  block_words[3] = load32(block + 4 * 3);
+  block_words[4] = load32(block + 4 * 4);
+  block_words[5] = load32(block + 4 * 5);
+  block_words[6] = load32(block + 4 * 6);
+  block_words[7] = load32(block + 4 * 7);
+  block_words[8] = load32(block + 4 * 8);
+  block_words[9] = load32(block + 4 * 9);
+  block_words[10] = load32(block + 4 * 10);
+  block_words[11] = load32(block + 4 * 11);
+  block_words[12] = load32(block + 4 * 12);
+  block_words[13] = load32(block + 4 * 13);
+  block_words[14] = load32(block + 4 * 14);
+  block_words[15] = load32(block + 4 * 15);
+
+  state[0] = cv[0];
+  state[1] = cv[1];
+  state[2] = cv[2];
+  state[3] = cv[3];
+  state[4] = cv[4];
+  state[5] = cv[5];
+  state[6] = cv[6];
+  state[7] = cv[7];
+  state[8] = IV[0];
+  state[9] = IV[1];
+  state[10] = IV[2];
+  state[11] = IV[3];
+  state[12] = counter_low(counter);
+  state[13] = counter_high(counter);
+  state[14] = (uint32_t)block_len;
+  state[15] = (uint32_t)flags;
+
+  round_fn(state, &block_words[0], 0);
+  round_fn(state, &block_words[0], 1);
+  round_fn(state, &block_words[0], 2);
+  round_fn(state, &block_words[0], 3);
+  round_fn(state, &block_words[0], 4);
+  round_fn(state, &block_words[0], 5);
+  round_fn(state, &block_words[0], 6);
+}
+
+void blake3_compress_in_place_portable(uint32_t cv[8],
+                                       const uint8_t block[BLAKE3_BLOCK_LEN],
+                                       uint8_t block_len, uint64_t counter,
+                                       uint8_t flags) {
+  uint32_t state[16];
+  compress_pre(state, cv, block, block_len, counter, flags);
+  cv[0] = state[0] ^ state[8];
+  cv[1] = state[1] ^ state[9];
+  cv[2] = state[2] ^ state[10];
+  cv[3] = state[3] ^ state[11];
+  cv[4] = state[4] ^ state[12];
+  cv[5] = state[5] ^ state[13];
+  cv[6] = state[6] ^ state[14];
+  cv[7] = state[7] ^ state[15];
+}
+
+void blake3_compress_xof_portable(const uint32_t cv[8],
+                                  const uint8_t block[BLAKE3_BLOCK_LEN],
+                                  uint8_t block_len, uint64_t counter,
+                                  uint8_t flags, uint8_t out[64]) {
+  uint32_t state[16];
+  compress_pre(state, cv, block, block_len, counter, flags);
+
+  store32(&out[0 * 4], state[0] ^ state[8]);
+  store32(&out[1 * 4], state[1] ^ state[9]);
+  store32(&out[2 * 4], state[2] ^ state[10]);
+  store32(&out[3 * 4], state[3] ^ state[11]);
+  store32(&out[4 * 4], state[4] ^ state[12]);
+  store32(&out[5 * 4], state[5] ^ state[13]);
+  store32(&out[6 * 4], state[6] ^ state[14]);
+  store32(&out[7 * 4], state[7] ^ state[15]);
+  store32(&out[8 * 4], state[8] ^ cv[0]);
+  store32(&out[9 * 4], state[9] ^ cv[1]);
+  store32(&out[10 * 4], state[10] ^ cv[2]);
+  store32(&out[11 * 4], state[11] ^ cv[3]);
+  store32(&out[12 * 4], state[12] ^ cv[4]);
+  store32(&out[13 * 4], state[13] ^ cv[5]);
+  store32(&out[14 * 4], state[14] ^ cv[6]);
+  store32(&out[15 * 4], state[15] ^ cv[7]);
+}
+
+INLINE void hash_one_portable(const uint8_t *input, size_t blocks,
+                              const uint32_t key[8], uint64_t counter,
+                              uint8_t flags, uint8_t flags_start,
+                              uint8_t flags_end, uint8_t out[BLAKE3_OUT_LEN]) {
+  uint32_t cv[8];
+  memcpy(cv, key, BLAKE3_KEY_LEN);
+  uint8_t block_flags = flags | flags_start;
+  while (blocks > 0) {
+    if (blocks == 1) {
+      block_flags |= flags_end;
+    }
+    blake3_compress_in_place_portable(cv, input, BLAKE3_BLOCK_LEN, counter,
+                                      block_flags);
+    input = &input[BLAKE3_BLOCK_LEN];
+    blocks -= 1;
+    block_flags = flags;
+  }
+  store_cv_words(out, cv);
+}
+
+void blake3_hash_many_portable(const uint8_t *const *inputs, size_t num_inputs,
+                               size_t blocks, const uint32_t key[8],
+                               uint64_t counter, bool increment_counter,
+                               uint8_t flags, uint8_t flags_start,
+                               uint8_t flags_end, uint8_t *out) {
+  while (num_inputs > 0) {
+    hash_one_portable(inputs[0], blocks, key, counter, flags, flags_start,
+                      flags_end, out);
+    if (increment_counter) {
+      counter += 1;
+    }
+    inputs += 1;
+    num_inputs -= 1;
+    out = &out[BLAKE3_OUT_LEN];
+  }
+}
diff --git a/thirdparty/blake3/c/blake3_sse2.c b/thirdparty/blake3/c/blake3_sse2.c
new file mode 100644
index 000000000..691e1c680
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse2.c
@@ -0,0 +1,566 @@
+#include "blake3_impl.h"
+
+#include <immintrin.h>
+
+#define DEGREE 4
+
+#define _mm_shuffle_ps2(a, b, c)                                               \
+  (_mm_castps_si128(                                                           \
+      _mm_shuffle_ps(_mm_castsi128_ps(a), _mm_castsi128_ps(b), (c))))
+
+INLINE __m128i loadu(const uint8_t src[16]) {
+  return _mm_loadu_si128((const __m128i *)src);
+}
+
+INLINE void storeu(__m128i src, uint8_t dest[16]) {
+  _mm_storeu_si128((__m128i *)dest, src);
+}
+
+INLINE __m128i addv(__m128i a, __m128i b) { return _mm_add_epi32(a, b); }
+
+// Note that clang-format doesn't like the name "xor" for some reason.
+INLINE __m128i xorv(__m128i a, __m128i b) { return _mm_xor_si128(a, b); }
+
+INLINE __m128i set1(uint32_t x) { return _mm_set1_epi32((int32_t)x); }
+
+INLINE __m128i set4(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
+  return _mm_setr_epi32((int32_t)a, (int32_t)b, (int32_t)c, (int32_t)d);
+}
+
+INLINE __m128i rot16(__m128i x) {
+  return _mm_shufflehi_epi16(_mm_shufflelo_epi16(x, 0xB1), 0xB1);
+}
+
+INLINE __m128i rot12(__m128i x) {
+  return xorv(_mm_srli_epi32(x, 12), _mm_slli_epi32(x, 32 - 12));
+}
+
+INLINE __m128i rot8(__m128i x) {
+  return xorv(_mm_srli_epi32(x, 8), _mm_slli_epi32(x, 32 - 8));
+}
+
+INLINE __m128i rot7(__m128i x) {
+  return xorv(_mm_srli_epi32(x, 7), _mm_slli_epi32(x, 32 - 7));
+}
+
+INLINE void g1(__m128i *row0, __m128i *row1, __m128i *row2, __m128i *row3,
+               __m128i m) {
+  *row0 = addv(addv(*row0, m), *row1);
+  *row3 = xorv(*row3, *row0);
+  *row3 = rot16(*row3);
+  *row2 = addv(*row2, *row3);
+  *row1 = xorv(*row1, *row2);
+  *row1 = rot12(*row1);
+}
+
+INLINE void g2(__m128i *row0, __m128i *row1, __m128i *row2, __m128i *row3,
+               __m128i m) {
+  *row0 = addv(addv(*row0, m), *row1);
+  *row3 = xorv(*row3, *row0);
+  *row3 = rot8(*row3);
+  *row2 = addv(*row2, *row3);
+  *row1 = xorv(*row1, *row2);
+  *row1 = rot7(*row1);
+}
+
+// Note the optimization here of leaving row1 as the unrotated row, rather than
+// row0. All the message loads below are adjusted to compensate for this. See
+// discussion at https://github.com/sneves/blake2-avx2/pull/4
+INLINE void diagonalize(__m128i *row0, __m128i *row2, __m128i *row3) {
+  *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE(2, 1, 0, 3));
+  *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE(1, 0, 3, 2));
+  *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE(0, 3, 2, 1));
+}
+
+INLINE void undiagonalize(__m128i *row0, __m128i *row2, __m128i *row3) {
+  *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE(0, 3, 2, 1));
+  *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE(1, 0, 3, 2));
+  *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE(2, 1, 0, 3));
+}
+
+INLINE __m128i blend_epi16(__m128i a, __m128i b, const int16_t imm8) {
+  const __m128i bits = _mm_set_epi16(0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01);
+  __m128i mask = _mm_set1_epi16(imm8);
+  mask = _mm_and_si128(mask, bits);
+  mask = _mm_cmpeq_epi16(mask, bits);
+  return _mm_or_si128(_mm_and_si128(mask, b), _mm_andnot_si128(mask, a));
+}
+
+INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
+                         const uint8_t block[BLAKE3_BLOCK_LEN],
+                         uint8_t block_len, uint64_t counter, uint8_t flags) {
+  rows[0] = loadu((uint8_t *)&cv[0]);
+  rows[1] = loadu((uint8_t *)&cv[4]);
+  rows[2] = set4(IV[0], IV[1], IV[2], IV[3]);
+  rows[3] = set4(counter_low(counter), counter_high(counter),
+                 (uint32_t)block_len, (uint32_t)flags);
+
+  __m128i m0 = loadu(&block[sizeof(__m128i) * 0]);
+  __m128i m1 = loadu(&block[sizeof(__m128i) * 1]);
+  __m128i m2 = loadu(&block[sizeof(__m128i) * 2]);
+  __m128i m3 = loadu(&block[sizeof(__m128i) * 3]);
+
+  __m128i t0, t1, t2, t3, tt;
+
+  // Round 1. The first round permutes the message words from the original
+  // input order, into the groups that get mixed in parallel.
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(2, 0, 2, 0)); //  6  4  2  0
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 3, 1)); //  7  5  3  1
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(2, 0, 2, 0)); // 14 12 10  8
+  t2 = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2, 1, 0, 3));   // 12 10  8 14
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 1, 3, 1)); // 15 13 11  9
+  t3 = _mm_shuffle_epi32(t3, _MM_SHUFFLE(2, 1, 0, 3));   // 13 11  9 15
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 2. This round and all following rounds apply a fixed permutation
+  // to the message words from the round before.
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 3
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 4
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 5
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 6
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 7
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+}
+
+void blake3_compress_in_place_sse2(uint32_t cv[8],
+                                   const uint8_t block[BLAKE3_BLOCK_LEN],
+                                   uint8_t block_len, uint64_t counter,
+                                   uint8_t flags) {
+  __m128i rows[4];
+  compress_pre(rows, cv, block, block_len, counter, flags);
+  storeu(xorv(rows[0], rows[2]), (uint8_t *)&cv[0]);
+  storeu(xorv(rows[1], rows[3]), (uint8_t *)&cv[4]);
+}
+
+void blake3_compress_xof_sse2(const uint32_t cv[8],
+                              const uint8_t block[BLAKE3_BLOCK_LEN],
+                              uint8_t block_len, uint64_t counter,
+                              uint8_t flags, uint8_t out[64]) {
+  __m128i rows[4];
+  compress_pre(rows, cv, block, block_len, counter, flags);
+  storeu(xorv(rows[0], rows[2]), &out[0]);
+  storeu(xorv(rows[1], rows[3]), &out[16]);
+  storeu(xorv(rows[2], loadu((uint8_t *)&cv[0])), &out[32]);
+  storeu(xorv(rows[3], loadu((uint8_t *)&cv[4])), &out[48]);
+}
+
+INLINE void round_fn(__m128i v[16], __m128i m[16], size_t r) {
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
+  v[0] = addv(v[0], v[4]);
+  v[1] = addv(v[1], v[5]);
+  v[2] = addv(v[2], v[6]);
+  v[3] = addv(v[3], v[7]);
+  v[12] = xorv(v[12], v[0]);
+  v[13] = xorv(v[13], v[1]);
+  v[14] = xorv(v[14], v[2]);
+  v[15] = xorv(v[15], v[3]);
+  v[12] = rot16(v[12]);
+  v[13] = rot16(v[13]);
+  v[14] = rot16(v[14]);
+  v[15] = rot16(v[15]);
+  v[8] = addv(v[8], v[12]);
+  v[9] = addv(v[9], v[13]);
+  v[10] = addv(v[10], v[14]);
+  v[11] = addv(v[11], v[15]);
+  v[4] = xorv(v[4], v[8]);
+  v[5] = xorv(v[5], v[9]);
+  v[6] = xorv(v[6], v[10]);
+  v[7] = xorv(v[7], v[11]);
+  v[4] = rot12(v[4]);
+  v[5] = rot12(v[5]);
+  v[6] = rot12(v[6]);
+  v[7] = rot12(v[7]);
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
+  v[0] = addv(v[0], v[4]);
+  v[1] = addv(v[1], v[5]);
+  v[2] = addv(v[2], v[6]);
+  v[3] = addv(v[3], v[7]);
+  v[12] = xorv(v[12], v[0]);
+  v[13] = xorv(v[13], v[1]);
+  v[14] = xorv(v[14], v[2]);
+  v[15] = xorv(v[15], v[3]);
+  v[12] = rot8(v[12]);
+  v[13] = rot8(v[13]);
+  v[14] = rot8(v[14]);
+  v[15] = rot8(v[15]);
+  v[8] = addv(v[8], v[12]);
+  v[9] = addv(v[9], v[13]);
+  v[10] = addv(v[10], v[14]);
+  v[11] = addv(v[11], v[15]);
+  v[4] = xorv(v[4], v[8]);
+  v[5] = xorv(v[5], v[9]);
+  v[6] = xorv(v[6], v[10]);
+  v[7] = xorv(v[7], v[11]);
+  v[4] = rot7(v[4]);
+  v[5] = rot7(v[5]);
+  v[6] = rot7(v[6]);
+  v[7] = rot7(v[7]);
+
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
+  v[0] = addv(v[0], v[5]);
+  v[1] = addv(v[1], v[6]);
+  v[2] = addv(v[2], v[7]);
+  v[3] = addv(v[3], v[4]);
+  v[15] = xorv(v[15], v[0]);
+  v[12] = xorv(v[12], v[1]);
+  v[13] = xorv(v[13], v[2]);
+  v[14] = xorv(v[14], v[3]);
+  v[15] = rot16(v[15]);
+  v[12] = rot16(v[12]);
+  v[13] = rot16(v[13]);
+  v[14] = rot16(v[14]);
+  v[10] = addv(v[10], v[15]);
+  v[11] = addv(v[11], v[12]);
+  v[8] = addv(v[8], v[13]);
+  v[9] = addv(v[9], v[14]);
+  v[5] = xorv(v[5], v[10]);
+  v[6] = xorv(v[6], v[11]);
+  v[7] = xorv(v[7], v[8]);
+  v[4] = xorv(v[4], v[9]);
+  v[5] = rot12(v[5]);
+  v[6] = rot12(v[6]);
+  v[7] = rot12(v[7]);
+  v[4] = rot12(v[4]);
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
+  v[0] = addv(v[0], v[5]);
+  v[1] = addv(v[1], v[6]);
+  v[2] = addv(v[2], v[7]);
+  v[3] = addv(v[3], v[4]);
+  v[15] = xorv(v[15], v[0]);
+  v[12] = xorv(v[12], v[1]);
+  v[13] = xorv(v[13], v[2]);
+  v[14] = xorv(v[14], v[3]);
+  v[15] = rot8(v[15]);
+  v[12] = rot8(v[12]);
+  v[13] = rot8(v[13]);
+  v[14] = rot8(v[14]);
+  v[10] = addv(v[10], v[15]);
+  v[11] = addv(v[11], v[12]);
+  v[8] = addv(v[8], v[13]);
+  v[9] = addv(v[9], v[14]);
+  v[5] = xorv(v[5], v[10]);
+  v[6] = xorv(v[6], v[11]);
+  v[7] = xorv(v[7], v[8]);
+  v[4] = xorv(v[4], v[9]);
+  v[5] = rot7(v[5]);
+  v[6] = rot7(v[6]);
+  v[7] = rot7(v[7]);
+  v[4] = rot7(v[4]);
+}
+
+INLINE void transpose_vecs(__m128i vecs[DEGREE]) {
+  // Interleave 32-bit lanes. The low unpack is lanes 00/11 and the high is
+  // 22/33. Note that this doesn't split the vector into two lanes, as the
+  // AVX2 counterparts do.
+  __m128i ab_01 = _mm_unpacklo_epi32(vecs[0], vecs[1]);
+  __m128i ab_23 = _mm_unpackhi_epi32(vecs[0], vecs[1]);
+  __m128i cd_01 = _mm_unpacklo_epi32(vecs[2], vecs[3]);
+  __m128i cd_23 = _mm_unpackhi_epi32(vecs[2], vecs[3]);
+
+  // Interleave 64-bit lanes.
+  __m128i abcd_0 = _mm_unpacklo_epi64(ab_01, cd_01);
+  __m128i abcd_1 = _mm_unpackhi_epi64(ab_01, cd_01);
+  __m128i abcd_2 = _mm_unpacklo_epi64(ab_23, cd_23);
+  __m128i abcd_3 = _mm_unpackhi_epi64(ab_23, cd_23);
+
+  vecs[0] = abcd_0;
+  vecs[1] = abcd_1;
+  vecs[2] = abcd_2;
+  vecs[3] = abcd_3;
+}
+
+INLINE void transpose_msg_vecs(const uint8_t *const *inputs,
+                               size_t block_offset, __m128i out[16]) {
+  out[0] = loadu(&inputs[0][block_offset + 0 * sizeof(__m128i)]);
+  out[1] = loadu(&inputs[1][block_offset + 0 * sizeof(__m128i)]);
+  out[2] = loadu(&inputs[2][block_offset + 0 * sizeof(__m128i)]);
+  out[3] = loadu(&inputs[3][block_offset + 0 * sizeof(__m128i)]);
+  out[4] = loadu(&inputs[0][block_offset + 1 * sizeof(__m128i)]);
+  out[5] = loadu(&inputs[1][block_offset + 1 * sizeof(__m128i)]);
+  out[6] = loadu(&inputs[2][block_offset + 1 * sizeof(__m128i)]);
+  out[7] = loadu(&inputs[3][block_offset + 1 * sizeof(__m128i)]);
+  out[8] = loadu(&inputs[0][block_offset + 2 * sizeof(__m128i)]);
+  out[9] = loadu(&inputs[1][block_offset + 2 * sizeof(__m128i)]);
+  out[10] = loadu(&inputs[2][block_offset + 2 * sizeof(__m128i)]);
+  out[11] = loadu(&inputs[3][block_offset + 2 * sizeof(__m128i)]);
+  out[12] = loadu(&inputs[0][block_offset + 3 * sizeof(__m128i)]);
+  out[13] = loadu(&inputs[1][block_offset + 3 * sizeof(__m128i)]);
+  out[14] = loadu(&inputs[2][block_offset + 3 * sizeof(__m128i)]);
+  out[15] = loadu(&inputs[3][block_offset + 3 * sizeof(__m128i)]);
+  for (size_t i = 0; i < 4; ++i) {
+    _mm_prefetch((const void *)&inputs[i][block_offset + 256], _MM_HINT_T0);
+  }
+  transpose_vecs(&out[0]);
+  transpose_vecs(&out[4]);
+  transpose_vecs(&out[8]);
+  transpose_vecs(&out[12]);
+}
+
+INLINE void load_counters(uint64_t counter, bool increment_counter,
+                          __m128i *out_lo, __m128i *out_hi) {
+  const __m128i mask = _mm_set1_epi32(-(int32_t)increment_counter);
+  const __m128i add0 = _mm_set_epi32(3, 2, 1, 0);
+  const __m128i add1 = _mm_and_si128(mask, add0);
+  __m128i l = _mm_add_epi32(_mm_set1_epi32((int32_t)counter), add1);
+  __m128i carry = _mm_cmpgt_epi32(_mm_xor_si128(add1, _mm_set1_epi32(0x80000000)), 
+                                  _mm_xor_si128(   l, _mm_set1_epi32(0x80000000)));
+  __m128i h = _mm_sub_epi32(_mm_set1_epi32((int32_t)(counter >> 32)), carry);
+  *out_lo = l;
+  *out_hi = h;
+}
+
+static
+void blake3_hash4_sse2(const uint8_t *const *inputs, size_t blocks,
+                       const uint32_t key[8], uint64_t counter,
+                       bool increment_counter, uint8_t flags,
+                       uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
+  __m128i h_vecs[8] = {
+      set1(key[0]), set1(key[1]), set1(key[2]), set1(key[3]),
+      set1(key[4]), set1(key[5]), set1(key[6]), set1(key[7]),
+  };
+  __m128i counter_low_vec, counter_high_vec;
+  load_counters(counter, increment_counter, &counter_low_vec,
+                &counter_high_vec);
+  uint8_t block_flags = flags | flags_start;
+
+  for (size_t block = 0; block < blocks; block++) {
+    if (block + 1 == blocks) {
+      block_flags |= flags_end;
+    }
+    __m128i block_len_vec = set1(BLAKE3_BLOCK_LEN);
+    __m128i block_flags_vec = set1(block_flags);
+    __m128i msg_vecs[16];
+    transpose_msg_vecs(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);
+
+    __m128i v[16] = {
+        h_vecs[0],       h_vecs[1],        h_vecs[2],     h_vecs[3],
+        h_vecs[4],       h_vecs[5],        h_vecs[6],     h_vecs[7],
+        set1(IV[0]),     set1(IV[1]),      set1(IV[2]),   set1(IV[3]),
+        counter_low_vec, counter_high_vec, block_len_vec, block_flags_vec,
+    };
+    round_fn(v, msg_vecs, 0);
+    round_fn(v, msg_vecs, 1);
+    round_fn(v, msg_vecs, 2);
+    round_fn(v, msg_vecs, 3);
+    round_fn(v, msg_vecs, 4);
+    round_fn(v, msg_vecs, 5);
+    round_fn(v, msg_vecs, 6);
+    h_vecs[0] = xorv(v[0], v[8]);
+    h_vecs[1] = xorv(v[1], v[9]);
+    h_vecs[2] = xorv(v[2], v[10]);
+    h_vecs[3] = xorv(v[3], v[11]);
+    h_vecs[4] = xorv(v[4], v[12]);
+    h_vecs[5] = xorv(v[5], v[13]);
+    h_vecs[6] = xorv(v[6], v[14]);
+    h_vecs[7] = xorv(v[7], v[15]);
+
+    block_flags = flags;
+  }
+
+  transpose_vecs(&h_vecs[0]);
+  transpose_vecs(&h_vecs[4]);
+  // The first four vecs now contain the first half of each output, and the
+  // second four vecs contain the second half of each output.
+  storeu(h_vecs[0], &out[0 * sizeof(__m128i)]);
+  storeu(h_vecs[4], &out[1 * sizeof(__m128i)]);
+  storeu(h_vecs[1], &out[2 * sizeof(__m128i)]);
+  storeu(h_vecs[5], &out[3 * sizeof(__m128i)]);
+  storeu(h_vecs[2], &out[4 * sizeof(__m128i)]);
+  storeu(h_vecs[6], &out[5 * sizeof(__m128i)]);
+  storeu(h_vecs[3], &out[6 * sizeof(__m128i)]);
+  storeu(h_vecs[7], &out[7 * sizeof(__m128i)]);
+}
+
+INLINE void hash_one_sse2(const uint8_t *input, size_t blocks,
+                          const uint32_t key[8], uint64_t counter,
+                          uint8_t flags, uint8_t flags_start,
+                          uint8_t flags_end, uint8_t out[BLAKE3_OUT_LEN]) {
+  uint32_t cv[8];
+  memcpy(cv, key, BLAKE3_KEY_LEN);
+  uint8_t block_flags = flags | flags_start;
+  while (blocks > 0) {
+    if (blocks == 1) {
+      block_flags |= flags_end;
+    }
+    blake3_compress_in_place_sse2(cv, input, BLAKE3_BLOCK_LEN, counter,
+                                  block_flags);
+    input = &input[BLAKE3_BLOCK_LEN];
+    blocks -= 1;
+    block_flags = flags;
+  }
+  memcpy(out, cv, BLAKE3_OUT_LEN);
+}
+
+void blake3_hash_many_sse2(const uint8_t *const *inputs, size_t num_inputs,
+                           size_t blocks, const uint32_t key[8],
+                           uint64_t counter, bool increment_counter,
+                           uint8_t flags, uint8_t flags_start,
+                           uint8_t flags_end, uint8_t *out) {
+  while (num_inputs >= DEGREE) {
+    blake3_hash4_sse2(inputs, blocks, key, counter, increment_counter, flags,
+                      flags_start, flags_end, out);
+    if (increment_counter) {
+      counter += DEGREE;
+    }
+    inputs += DEGREE;
+    num_inputs -= DEGREE;
+    out = &out[DEGREE * BLAKE3_OUT_LEN];
+  }
+  while (num_inputs > 0) {
+    hash_one_sse2(inputs[0], blocks, key, counter, flags, flags_start,
+                  flags_end, out);
+    if (increment_counter) {
+      counter += 1;
+    }
+    inputs += 1;
+    num_inputs -= 1;
+    out = &out[BLAKE3_OUT_LEN];
+  }
+}
diff --git a/thirdparty/blake3/c/blake3_sse2_x86-64_unix.S b/thirdparty/blake3/c/blake3_sse2_x86-64_unix.S
new file mode 100644
index 000000000..99f033fef
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse2_x86-64_unix.S
@@ -0,0 +1,2291 @@
+#if defined(__ELF__) && defined(__linux__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+#if defined(__ELF__) && defined(__CET__) && defined(__has_include)
+#if __has_include(<cet.h>)
+#include <cet.h>
+#endif
+#endif
+
+#if !defined(_CET_ENDBR)
+#define _CET_ENDBR
+#endif
+
+.intel_syntax noprefix
+.global blake3_hash_many_sse2
+.global _blake3_hash_many_sse2
+.global blake3_compress_in_place_sse2
+.global _blake3_compress_in_place_sse2
+.global blake3_compress_xof_sse2
+.global _blake3_compress_xof_sse2
+#ifdef __APPLE__
+.text
+#else
+.section .text
+#endif
+        .p2align  6
+_blake3_hash_many_sse2:
+blake3_hash_many_sse2:
+        _CET_ENDBR
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 360
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        neg     r9d
+        movd    xmm0, r9d
+        pshufd  xmm0, xmm0, 0x00
+        movdqa  xmmword ptr [rsp+0x130], xmm0
+        movdqa  xmm1, xmm0
+        pand    xmm1, xmmword ptr [ADD0+rip]
+        pand    xmm0, xmmword ptr [ADD1+rip]
+        movdqa  xmmword ptr [rsp+0x150], xmm0
+        movd    xmm0, r8d
+        pshufd  xmm0, xmm0, 0x00
+        paddd   xmm0, xmm1
+        movdqa  xmmword ptr [rsp+0x110], xmm0
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm1, xmm0
+        shr     r8, 32
+        movd    xmm2, r8d
+        pshufd  xmm2, xmm2, 0x00
+        psubd   xmm2, xmm1
+        movdqa  xmmword ptr [rsp+0x120], xmm2
+        mov     rbx, qword ptr [rbp+0x50]
+        mov     r15, rdx
+        shl     r15, 6
+        movzx   r13d, byte ptr [rbp+0x38]
+        movzx   r12d, byte ptr [rbp+0x48]
+        cmp     rsi, 4
+        jc      3f
+2:
+        movdqu  xmm3, xmmword ptr [rcx]
+        pshufd  xmm0, xmm3, 0x00
+        pshufd  xmm1, xmm3, 0x55
+        pshufd  xmm2, xmm3, 0xAA
+        pshufd  xmm3, xmm3, 0xFF
+        movdqu  xmm7, xmmword ptr [rcx+0x10]
+        pshufd  xmm4, xmm7, 0x00
+        pshufd  xmm5, xmm7, 0x55
+        pshufd  xmm6, xmm7, 0xAA
+        pshufd  xmm7, xmm7, 0xFF
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+9:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x40]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x40]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x40]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x40]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp], xmm8
+        movdqa  xmmword ptr [rsp+0x10], xmm9
+        movdqa  xmmword ptr [rsp+0x20], xmm12
+        movdqa  xmmword ptr [rsp+0x30], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x30]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x30]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x30]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x30]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x40], xmm8
+        movdqa  xmmword ptr [rsp+0x50], xmm9
+        movdqa  xmmword ptr [rsp+0x60], xmm12
+        movdqa  xmmword ptr [rsp+0x70], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x20]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x20]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x20]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x20]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x80], xmm8
+        movdqa  xmmword ptr [rsp+0x90], xmm9
+        movdqa  xmmword ptr [rsp+0xA0], xmm12
+        movdqa  xmmword ptr [rsp+0xB0], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x10]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x10]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x10]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x10]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0xC0], xmm8
+        movdqa  xmmword ptr [rsp+0xD0], xmm9
+        movdqa  xmmword ptr [rsp+0xE0], xmm12
+        movdqa  xmmword ptr [rsp+0xF0], xmm13
+        movdqa  xmm9, xmmword ptr [BLAKE3_IV_1+rip]
+        movdqa  xmm10, xmmword ptr [BLAKE3_IV_2+rip]
+        movdqa  xmm11, xmmword ptr [BLAKE3_IV_3+rip]
+        movdqa  xmm12, xmmword ptr [rsp+0x110]
+        movdqa  xmm13, xmmword ptr [rsp+0x120]
+        movdqa  xmm14, xmmword ptr [BLAKE3_BLOCK_LEN+rip]
+        movd    xmm15, eax
+        pshufd  xmm15, xmm15, 0x00
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [BLAKE3_IV_0+rip]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x80]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x70]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xB0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x50]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xC0]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xA0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0x60]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xF0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        pxor    xmm0, xmm8
+        pxor    xmm1, xmm9
+        pxor    xmm2, xmm10
+        pxor    xmm3, xmm11
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        pxor    xmm4, xmm12
+        pxor    xmm5, xmm13
+        pxor    xmm6, xmm14
+        pxor    xmm7, xmm15
+        mov     eax, r13d
+        jne     9b
+        movdqa  xmm9, xmm0
+        punpckldq xmm0, xmm1
+        punpckhdq xmm9, xmm1
+        movdqa  xmm11, xmm2
+        punpckldq xmm2, xmm3
+        punpckhdq xmm11, xmm3
+        movdqa  xmm1, xmm0
+        punpcklqdq xmm0, xmm2
+        punpckhqdq xmm1, xmm2
+        movdqa  xmm3, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm3, xmm11
+        movdqu  xmmword ptr [rbx], xmm0
+        movdqu  xmmword ptr [rbx+0x20], xmm1
+        movdqu  xmmword ptr [rbx+0x40], xmm9
+        movdqu  xmmword ptr [rbx+0x60], xmm3
+        movdqa  xmm9, xmm4
+        punpckldq xmm4, xmm5
+        punpckhdq xmm9, xmm5
+        movdqa  xmm11, xmm6
+        punpckldq xmm6, xmm7
+        punpckhdq xmm11, xmm7
+        movdqa  xmm5, xmm4
+        punpcklqdq xmm4, xmm6
+        punpckhqdq xmm5, xmm6
+        movdqa  xmm7, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm7, xmm11
+        movdqu  xmmword ptr [rbx+0x10], xmm4
+        movdqu  xmmword ptr [rbx+0x30], xmm5
+        movdqu  xmmword ptr [rbx+0x50], xmm9
+        movdqu  xmmword ptr [rbx+0x70], xmm7
+        movdqa  xmm1, xmmword ptr [rsp+0x110]
+        movdqa  xmm0, xmm1
+        paddd   xmm1, xmmword ptr [rsp+0x150]
+        movdqa  xmmword ptr [rsp+0x110], xmm1
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm0, xmm1
+        movdqa  xmm1, xmmword ptr [rsp+0x120]
+        psubd   xmm1, xmm0
+        movdqa  xmmword ptr [rsp+0x120], xmm1
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+        cmp     rsi, 4
+        jnc     2b
+        test    rsi, rsi
+        jnz     3f
+4:
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align 5
+3:
+        test    esi, 0x2
+        je      3f
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm8, xmm0
+        movaps  xmm9, xmm1
+        movd    xmm13, dword ptr [rsp+0x110]
+        movd    xmm14, dword ptr [rsp+0x120]
+        punpckldq xmm13, xmm14
+        movaps  xmmword ptr [rsp], xmm13
+        movd    xmm14, dword ptr [rsp+0x114]
+        movd    xmm13, dword ptr [rsp+0x124]
+        punpckldq xmm14, xmm13
+        movaps  xmmword ptr [rsp+0x10], xmm14
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movaps  xmm10, xmm2
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm3, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm3, xmm5, 221
+        movaps  xmm5, xmm3
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm3, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm3, xmm7, 221
+        pshufd  xmm7, xmm3, 0x93
+        movups  xmm12, xmmword ptr [r9+rdx-0x40]
+        movups  xmm13, xmmword ptr [r9+rdx-0x30]
+        movaps  xmm11, xmm12
+        shufps  xmm12, xmm13, 136
+        shufps  xmm11, xmm13, 221
+        movaps  xmm13, xmm11
+        movups  xmm14, xmmword ptr [r9+rdx-0x20]
+        movups  xmm15, xmmword ptr [r9+rdx-0x10]
+        movaps  xmm11, xmm14
+        shufps  xmm14, xmm15, 136
+        pshufd  xmm14, xmm14, 0x93
+        shufps  xmm11, xmm15, 221
+        pshufd  xmm15, xmm11, 0x93
+        shl     rax, 0x20
+        or      rax, 0x40
+        movq    xmm3, rax
+        movdqa  xmmword ptr [rsp+0x20], xmm3
+        movaps  xmm3, xmmword ptr [rsp]
+        movaps  xmm11, xmmword ptr [rsp+0x10]
+        punpcklqdq xmm3, xmmword ptr [rsp+0x20]
+        punpcklqdq xmm11, xmmword ptr [rsp+0x20]
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm8, xmm12
+        movaps  xmmword ptr [rsp+0x20], xmm4
+        movaps  xmmword ptr [rsp+0x30], xmm12
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        pshuflw xmm11, xmm11, 0xB1
+        pshufhw xmm11, xmm11, 0xB1
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm5
+        paddd   xmm8, xmm13
+        movaps  xmmword ptr [rsp+0x40], xmm5
+        movaps  xmmword ptr [rsp+0x50], xmm13
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movdqa  xmm13, xmm3
+        psrld   xmm3, 8
+        pslld   xmm13, 24
+        pxor    xmm3, xmm13
+        movdqa  xmm13, xmm11
+        psrld   xmm11, 8
+        pslld   xmm13, 24
+        pxor    xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm8, xmm8, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        pshufd  xmm10, xmm10, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm8, xmm14
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        pshuflw xmm11, xmm11, 0xB1
+        pshufhw xmm11, xmm11, 0xB1
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm7
+        paddd   xmm8, xmm15
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movdqa  xmm13, xmm3
+        psrld   xmm3, 8
+        pslld   xmm13, 24
+        pxor    xmm3, xmm13
+        movdqa  xmm13, xmm11
+        psrld   xmm11, 8
+        pslld   xmm13, 24
+        pxor    xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm8, xmm8, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        pshufd  xmm10, xmm10, 0x93
+        dec     al
+        je      9f
+        movdqa  xmm12, xmmword ptr [rsp+0x20]
+        movdqa  xmm5, xmmword ptr [rsp+0x40]
+        pshufd  xmm13, xmm12, 0x0F
+        shufps  xmm12, xmm5, 214
+        pshufd  xmm4, xmm12, 0x39
+        movdqa  xmm12, xmm6
+        shufps  xmm12, xmm7, 250
+        pand    xmm13, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm12, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm13, xmm12
+        movdqa  xmmword ptr [rsp+0x20], xmm13
+        movdqa  xmm12, xmm7
+        punpcklqdq xmm12, xmm5
+        movdqa  xmm13, xmm6
+        pand    xmm12, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm13, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm12, xmm13
+        pshufd  xmm12, xmm12, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmmword ptr [rsp+0x40], xmm12
+        movdqa  xmm5, xmmword ptr [rsp+0x30]
+        movdqa  xmm13, xmmword ptr [rsp+0x50]
+        pshufd  xmm6, xmm5, 0x0F
+        shufps  xmm5, xmm13, 214
+        pshufd  xmm12, xmm5, 0x39
+        movdqa  xmm5, xmm14
+        shufps  xmm5, xmm15, 250
+        pand    xmm6, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm5, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm6, xmm5
+        movdqa  xmm5, xmm15
+        punpcklqdq xmm5, xmm13
+        movdqa  xmmword ptr [rsp+0x30], xmm2
+        movdqa  xmm2, xmm14
+        pand    xmm5, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm2, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm5, xmm2
+        movdqa  xmm2, xmmword ptr [rsp+0x30]
+        pshufd  xmm5, xmm5, 0x78
+        punpckhdq xmm13, xmm15
+        punpckldq xmm14, xmm13
+        pshufd  xmm15, xmm14, 0x1E
+        movdqa  xmm13, xmm6
+        movdqa  xmm14, xmm5
+        movdqa  xmm5, xmmword ptr [rsp+0x20]
+        movdqa  xmm6, xmmword ptr [rsp+0x40]
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm8, xmm10
+        pxor    xmm9, xmm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        movups  xmmword ptr [rbx+0x20], xmm8
+        movups  xmmword ptr [rbx+0x30], xmm9
+        mov     eax, dword ptr [rsp+0x130]
+        neg     eax
+        mov    r10d, dword ptr [rsp+0x110+8*rax]
+        mov    r11d, dword ptr [rsp+0x120+8*rax]
+        mov dword ptr [rsp+0x110], r10d
+        mov dword ptr [rsp+0x120], r11d
+        add     rdi, 16
+        add     rbx, 64
+        sub     rsi, 2
+3:
+        test    esi, 0x1
+        je      4b
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movd    xmm13, dword ptr [rsp+0x110]
+        movd    xmm14, dword ptr [rsp+0x120]
+        punpckldq xmm13, xmm14
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        shl     rax, 32
+        or      rax, 64
+        movq    xmm12, rax
+        movdqa  xmm3, xmm13
+        punpcklqdq xmm3, xmm12
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm10, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm10, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm8, xmm10
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+
+.p2align 6
+blake3_compress_in_place_sse2:
+_blake3_compress_in_place_sse2:
+        _CET_ENDBR
+        movups  xmm0, xmmword ptr [rdi]
+        movups  xmm1, xmmword ptr [rdi+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        shl     r8, 32
+        add     rdx, r8
+        movq    xmm3, rcx
+        movq    xmm4, rdx
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rsi]
+        movups  xmm5, xmmword ptr [rsi+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rsi+0x20]
+        movups  xmm7, xmmword ptr [rsi+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm10, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm10, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm8, xmm10
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        movups  xmmword ptr [rdi], xmm0
+        movups  xmmword ptr [rdi+0x10], xmm1
+        ret
+
+.p2align 6
+blake3_compress_xof_sse2:
+_blake3_compress_xof_sse2:
+        _CET_ENDBR
+        movups  xmm0, xmmword ptr [rdi]
+        movups  xmm1, xmmword ptr [rdi+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movzx   eax, r8b
+        movzx   edx, dl
+        shl     rax, 32
+        add     rdx, rax
+        movq    xmm3, rcx
+        movq    xmm4, rdx
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rsi]
+        movups  xmm5, xmmword ptr [rsi+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rsi+0x20]
+        movups  xmm7, xmmword ptr [rsi+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm10, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm10, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm8, xmm10
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        movdqu  xmm4, xmmword ptr [rdi]
+        movdqu  xmm5, xmmword ptr [rdi+0x10]
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm2, xmm4
+        pxor    xmm3, xmm5
+        movups  xmmword ptr [r9], xmm0
+        movups  xmmword ptr [r9+0x10], xmm1
+        movups  xmmword ptr [r9+0x20], xmm2
+        movups  xmmword ptr [r9+0x30], xmm3
+        ret
+
+
+#ifdef __APPLE__
+.static_data
+#else
+.section .rodata
+#endif
+.p2align  6
+BLAKE3_IV:
+        .long  0x6A09E667, 0xBB67AE85
+        .long  0x3C6EF372, 0xA54FF53A
+ADD0:	
+        .long  0, 1, 2, 3
+ADD1:
+	.long  4, 4, 4, 4
+BLAKE3_IV_0:
+	.long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+BLAKE3_IV_1:
+	.long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+BLAKE3_IV_2:
+	.long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+BLAKE3_IV_3:
+	.long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+BLAKE3_BLOCK_LEN:
+	.long  64, 64, 64, 64
+CMP_MSB_MASK:
+	.long  0x80000000, 0x80000000, 0x80000000, 0x80000000
+PBLENDW_0x33_MASK:
+	.long  0xFFFFFFFF, 0x00000000, 0xFFFFFFFF, 0x00000000
+PBLENDW_0xCC_MASK:
+	.long  0x00000000, 0xFFFFFFFF, 0x00000000, 0xFFFFFFFF
+PBLENDW_0x3F_MASK:
+	.long  0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000
+PBLENDW_0xC0_MASK:
+	.long  0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF
diff --git a/thirdparty/blake3/c/blake3_sse2_x86-64_windows_gnu.S b/thirdparty/blake3/c/blake3_sse2_x86-64_windows_gnu.S
new file mode 100644
index 000000000..4facb50e7
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse2_x86-64_windows_gnu.S
@@ -0,0 +1,2332 @@
+.intel_syntax noprefix
+.global blake3_hash_many_sse2
+.global _blake3_hash_many_sse2
+.global blake3_compress_in_place_sse2
+.global _blake3_compress_in_place_sse2
+.global blake3_compress_xof_sse2
+.global _blake3_compress_xof_sse2
+.section .text
+        .p2align  6
+_blake3_hash_many_sse2:
+blake3_hash_many_sse2:
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rsi
+        push    rdi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 528
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        movdqa  xmmword ptr [rsp+0x170], xmm6
+        movdqa  xmmword ptr [rsp+0x180], xmm7
+        movdqa  xmmword ptr [rsp+0x190], xmm8
+        movdqa  xmmword ptr [rsp+0x1A0], xmm9
+        movdqa  xmmword ptr [rsp+0x1B0], xmm10
+        movdqa  xmmword ptr [rsp+0x1C0], xmm11
+        movdqa  xmmword ptr [rsp+0x1D0], xmm12
+        movdqa  xmmword ptr [rsp+0x1E0], xmm13
+        movdqa  xmmword ptr [rsp+0x1F0], xmm14
+        movdqa  xmmword ptr [rsp+0x200], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+0x68]
+        movzx   r9, byte ptr [rbp+0x70]
+        neg     r9d
+        movd    xmm0, r9d
+        pshufd  xmm0, xmm0, 0x00
+        movdqa  xmmword ptr [rsp+0x130], xmm0
+        movdqa  xmm1, xmm0
+        pand    xmm1, xmmword ptr [ADD0+rip]
+        pand    xmm0, xmmword ptr [ADD1+rip]
+        movdqa  xmmword ptr [rsp+0x150], xmm0
+        movd    xmm0, r8d
+        pshufd  xmm0, xmm0, 0x00
+        paddd   xmm0, xmm1
+        movdqa  xmmword ptr [rsp+0x110], xmm0
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm1, xmm0
+        shr     r8, 32
+        movd    xmm2, r8d
+        pshufd  xmm2, xmm2, 0x00
+        psubd   xmm2, xmm1
+        movdqa  xmmword ptr [rsp+0x120], xmm2
+        mov     rbx, qword ptr [rbp+0x90]
+        mov     r15, rdx
+        shl     r15, 6
+        movzx   r13d, byte ptr [rbp+0x78]
+        movzx   r12d, byte ptr [rbp+0x88]
+        cmp     rsi, 4
+        jc      3f
+2:
+        movdqu  xmm3, xmmword ptr [rcx]
+        pshufd  xmm0, xmm3, 0x00
+        pshufd  xmm1, xmm3, 0x55
+        pshufd  xmm2, xmm3, 0xAA
+        pshufd  xmm3, xmm3, 0xFF
+        movdqu  xmm7, xmmword ptr [rcx+0x10]
+        pshufd  xmm4, xmm7, 0x00
+        pshufd  xmm5, xmm7, 0x55
+        pshufd  xmm6, xmm7, 0xAA
+        pshufd  xmm7, xmm7, 0xFF
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+9:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x40]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x40]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x40]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x40]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp], xmm8
+        movdqa  xmmword ptr [rsp+0x10], xmm9
+        movdqa  xmmword ptr [rsp+0x20], xmm12
+        movdqa  xmmword ptr [rsp+0x30], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x30]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x30]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x30]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x30]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x40], xmm8
+        movdqa  xmmword ptr [rsp+0x50], xmm9
+        movdqa  xmmword ptr [rsp+0x60], xmm12
+        movdqa  xmmword ptr [rsp+0x70], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x20]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x20]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x20]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x20]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x80], xmm8
+        movdqa  xmmword ptr [rsp+0x90], xmm9
+        movdqa  xmmword ptr [rsp+0xA0], xmm12
+        movdqa  xmmword ptr [rsp+0xB0], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x10]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x10]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x10]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x10]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0xC0], xmm8
+        movdqa  xmmword ptr [rsp+0xD0], xmm9
+        movdqa  xmmword ptr [rsp+0xE0], xmm12
+        movdqa  xmmword ptr [rsp+0xF0], xmm13
+        movdqa  xmm9, xmmword ptr [BLAKE3_IV_1+rip]
+        movdqa  xmm10, xmmword ptr [BLAKE3_IV_2+rip]
+        movdqa  xmm11, xmmword ptr [BLAKE3_IV_3+rip]
+        movdqa  xmm12, xmmword ptr [rsp+0x110]
+        movdqa  xmm13, xmmword ptr [rsp+0x120]
+        movdqa  xmm14, xmmword ptr [BLAKE3_BLOCK_LEN+rip]
+        movd    xmm15, eax
+        pshufd  xmm15, xmm15, 0x00
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [BLAKE3_IV_0+rip]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x80]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x70]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xB0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x50]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xC0]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xA0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0x60]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xF0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0xB1
+        pshufhw xmm15, xmm15, 0xB1
+        pshuflw xmm12, xmm12, 0xB1
+        pshufhw xmm12, xmm12, 0xB1
+        pshuflw xmm13, xmm13, 0xB1
+        pshufhw xmm13, xmm13, 0xB1
+        pshuflw xmm14, xmm14, 0xB1
+        pshufhw xmm14, xmm14, 0xB1
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        pxor    xmm0, xmm8
+        pxor    xmm1, xmm9
+        pxor    xmm2, xmm10
+        pxor    xmm3, xmm11
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        pxor    xmm4, xmm12
+        pxor    xmm5, xmm13
+        pxor    xmm6, xmm14
+        pxor    xmm7, xmm15
+        mov     eax, r13d
+        jne     9b
+        movdqa  xmm9, xmm0
+        punpckldq xmm0, xmm1
+        punpckhdq xmm9, xmm1
+        movdqa  xmm11, xmm2
+        punpckldq xmm2, xmm3
+        punpckhdq xmm11, xmm3
+        movdqa  xmm1, xmm0
+        punpcklqdq xmm0, xmm2
+        punpckhqdq xmm1, xmm2
+        movdqa  xmm3, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm3, xmm11
+        movdqu  xmmword ptr [rbx], xmm0
+        movdqu  xmmword ptr [rbx+0x20], xmm1
+        movdqu  xmmword ptr [rbx+0x40], xmm9
+        movdqu  xmmword ptr [rbx+0x60], xmm3
+        movdqa  xmm9, xmm4
+        punpckldq xmm4, xmm5
+        punpckhdq xmm9, xmm5
+        movdqa  xmm11, xmm6
+        punpckldq xmm6, xmm7
+        punpckhdq xmm11, xmm7
+        movdqa  xmm5, xmm4
+        punpcklqdq xmm4, xmm6
+        punpckhqdq xmm5, xmm6
+        movdqa  xmm7, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm7, xmm11
+        movdqu  xmmword ptr [rbx+0x10], xmm4
+        movdqu  xmmword ptr [rbx+0x30], xmm5
+        movdqu  xmmword ptr [rbx+0x50], xmm9
+        movdqu  xmmword ptr [rbx+0x70], xmm7
+        movdqa  xmm1, xmmword ptr [rsp+0x110]
+        movdqa  xmm0, xmm1
+        paddd   xmm1, xmmword ptr [rsp+0x150]
+        movdqa  xmmword ptr [rsp+0x110], xmm1
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm0, xmm1
+        movdqa  xmm1, xmmword ptr [rsp+0x120]
+        psubd   xmm1, xmm0
+        movdqa  xmmword ptr [rsp+0x120], xmm1
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+        cmp     rsi, 4
+        jnc     2b
+        test    rsi, rsi
+        jne     3f
+4:
+        movdqa  xmm6, xmmword ptr [rsp+0x170]
+        movdqa  xmm7, xmmword ptr [rsp+0x180]
+        movdqa  xmm8, xmmword ptr [rsp+0x190]
+        movdqa  xmm9, xmmword ptr [rsp+0x1A0]
+        movdqa  xmm10, xmmword ptr [rsp+0x1B0]
+        movdqa  xmm11, xmmword ptr [rsp+0x1C0]
+        movdqa  xmm12, xmmword ptr [rsp+0x1D0]
+        movdqa  xmm13, xmmword ptr [rsp+0x1E0]
+        movdqa  xmm14, xmmword ptr [rsp+0x1F0]
+        movdqa  xmm15, xmmword ptr [rsp+0x200]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rdi
+        pop     rsi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align 5
+3:
+        test    esi, 0x2
+        je      3f
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm8, xmm0
+        movaps  xmm9, xmm1
+        movd    xmm13, dword ptr [rsp+0x110]
+        movd    xmm14, dword ptr [rsp+0x120]
+        punpckldq xmm13, xmm14
+        movaps  xmmword ptr [rsp], xmm13
+        movd    xmm14, dword ptr [rsp+0x114]
+        movd    xmm13, dword ptr [rsp+0x124]
+        punpckldq xmm14, xmm13
+        movaps  xmmword ptr [rsp+0x10], xmm14
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movaps  xmm10, xmm2
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm3, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm3, xmm5, 221
+        movaps  xmm5, xmm3
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm3, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm3, xmm7, 221
+        pshufd  xmm7, xmm3, 0x93
+        movups  xmm12, xmmword ptr [r9+rdx-0x40]
+        movups  xmm13, xmmword ptr [r9+rdx-0x30]
+        movaps  xmm11, xmm12
+        shufps  xmm12, xmm13, 136
+        shufps  xmm11, xmm13, 221
+        movaps  xmm13, xmm11
+        movups  xmm14, xmmword ptr [r9+rdx-0x20]
+        movups  xmm15, xmmword ptr [r9+rdx-0x10]
+        movaps  xmm11, xmm14
+        shufps  xmm14, xmm15, 136
+        pshufd  xmm14, xmm14, 0x93
+        shufps  xmm11, xmm15, 221
+        pshufd  xmm15, xmm11, 0x93
+        shl     rax, 0x20
+        or      rax, 0x40
+        movq    xmm3, rax
+        movdqa  xmmword ptr [rsp+0x20], xmm3
+        movaps  xmm3, xmmword ptr [rsp]
+        movaps  xmm11, xmmword ptr [rsp+0x10]
+        punpcklqdq xmm3, xmmword ptr [rsp+0x20]
+        punpcklqdq xmm11, xmmword ptr [rsp+0x20]
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm8, xmm12
+        movaps  xmmword ptr [rsp+0x20], xmm4
+        movaps  xmmword ptr [rsp+0x30], xmm12
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        pshuflw xmm11, xmm11, 0xB1
+        pshufhw xmm11, xmm11, 0xB1
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm5
+        paddd   xmm8, xmm13
+        movaps  xmmword ptr [rsp+0x40], xmm5
+        movaps  xmmword ptr [rsp+0x50], xmm13
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movdqa  xmm13, xmm3
+        psrld   xmm3, 8
+        pslld   xmm13, 24
+        pxor    xmm3, xmm13
+        movdqa  xmm13, xmm11
+        psrld   xmm11, 8
+        pslld   xmm13, 24
+        pxor    xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm8, xmm8, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        pshufd  xmm10, xmm10, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm8, xmm14
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        pshuflw xmm11, xmm11, 0xB1
+        pshufhw xmm11, xmm11, 0xB1
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm7
+        paddd   xmm8, xmm15
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movdqa  xmm13, xmm3
+        psrld   xmm3, 8
+        pslld   xmm13, 24
+        pxor    xmm3, xmm13
+        movdqa  xmm13, xmm11
+        psrld   xmm11, 8
+        pslld   xmm13, 24
+        pxor    xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm8, xmm8, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        pshufd  xmm10, xmm10, 0x93
+        dec     al
+        je      9f
+        movdqa  xmm12, xmmword ptr [rsp+0x20]
+        movdqa  xmm5, xmmword ptr [rsp+0x40]
+        pshufd  xmm13, xmm12, 0x0F
+        shufps  xmm12, xmm5, 214
+        pshufd  xmm4, xmm12, 0x39
+        movdqa  xmm12, xmm6
+        shufps  xmm12, xmm7, 250
+        pand    xmm13, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm12, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm13, xmm12
+        movdqa  xmmword ptr [rsp+0x20], xmm13
+        movdqa  xmm12, xmm7
+        punpcklqdq xmm12, xmm5
+        movdqa  xmm13, xmm6
+        pand    xmm12, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm13, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm12, xmm13
+        pshufd  xmm12, xmm12, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmmword ptr [rsp+0x40], xmm12
+        movdqa  xmm5, xmmword ptr [rsp+0x30]
+        movdqa  xmm13, xmmword ptr [rsp+0x50]
+        pshufd  xmm6, xmm5, 0x0F
+        shufps  xmm5, xmm13, 214
+        pshufd  xmm12, xmm5, 0x39
+        movdqa  xmm5, xmm14
+        shufps  xmm5, xmm15, 250
+        pand    xmm6, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm5, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm6, xmm5
+        movdqa  xmm5, xmm15
+        punpcklqdq xmm5, xmm13
+        movdqa  xmmword ptr [rsp+0x30], xmm2
+        movdqa  xmm2, xmm14
+        pand    xmm5, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm2, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm5, xmm2
+        movdqa  xmm2, xmmword ptr [rsp+0x30]
+        pshufd  xmm5, xmm5, 0x78
+        punpckhdq xmm13, xmm15
+        punpckldq xmm14, xmm13
+        pshufd  xmm15, xmm14, 0x1E
+        movdqa  xmm13, xmm6
+        movdqa  xmm14, xmm5
+        movdqa  xmm5, xmmword ptr [rsp+0x20]
+        movdqa  xmm6, xmmword ptr [rsp+0x40]
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm8, xmm10
+        pxor    xmm9, xmm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        movups  xmmword ptr [rbx+0x20], xmm8
+        movups  xmmword ptr [rbx+0x30], xmm9
+        mov     eax, dword ptr [rsp+0x130]
+        neg     eax
+        mov    r10d, dword ptr [rsp+0x110+8*rax]
+        mov    r11d, dword ptr [rsp+0x120+8*rax]
+        mov dword ptr [rsp+0x110], r10d
+        mov dword ptr [rsp+0x120], r11d
+        add     rdi, 16
+        add     rbx, 64
+        sub     rsi, 2
+3:
+        test    esi, 0x1
+        je      4b
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movd    xmm13, dword ptr [rsp+0x110]
+        movd    xmm14, dword ptr [rsp+0x120]
+        punpckldq xmm13, xmm14
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        shl     rax, 32
+        or      rax, 64
+        movq    xmm12, rax
+        movdqa  xmm3, xmm13
+        punpcklqdq xmm3, xmm12
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm10, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm10, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm8, xmm10
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+
+.p2align 6
+blake3_compress_in_place_sse2:
+_blake3_compress_in_place_sse2:
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+0x10], xmm7
+        movdqa  xmmword ptr [rsp+0x20], xmm8
+        movdqa  xmmword ptr [rsp+0x30], xmm9
+        movdqa  xmmword ptr [rsp+0x40], xmm11
+        movdqa  xmmword ptr [rsp+0x50], xmm14
+        movdqa  xmmword ptr [rsp+0x60], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movzx   eax, byte ptr [rsp+0xA0]
+        movzx   r8d, r8b
+        shl     rax, 32
+        add     r8, rax
+        movq    xmm3, r9
+        movq    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+0x20]
+        movups  xmm7, xmmword ptr [rdx+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm14, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm14, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm8, xmm14
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        movups  xmmword ptr [rcx], xmm0
+        movups  xmmword ptr [rcx+0x10], xmm1
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+0x10]
+        movdqa  xmm8, xmmword ptr [rsp+0x20]
+        movdqa  xmm9, xmmword ptr [rsp+0x30]
+        movdqa  xmm11, xmmword ptr [rsp+0x40]
+        movdqa  xmm14, xmmword ptr [rsp+0x50]
+        movdqa  xmm15, xmmword ptr [rsp+0x60]
+        add     rsp, 120
+        ret
+
+
+.p2align 6
+_blake3_compress_xof_sse2:
+blake3_compress_xof_sse2:
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+0x10], xmm7
+        movdqa  xmmword ptr [rsp+0x20], xmm8
+        movdqa  xmmword ptr [rsp+0x30], xmm9
+        movdqa  xmmword ptr [rsp+0x40], xmm11
+        movdqa  xmmword ptr [rsp+0x50], xmm14
+        movdqa  xmmword ptr [rsp+0x60], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movzx   eax, byte ptr [rsp+0xA0]
+        movzx   r8d, r8b
+        mov     r10, qword ptr [rsp+0xA8]
+        shl     rax, 32
+        add     r8, rax
+        movq    xmm3, r9
+        movq    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+0x20]
+        movups  xmm7, xmmword ptr [rdx+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0xB1
+        pshufhw xmm3, xmm3, 0xB1
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK+rip]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK+rip]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm14, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK+rip]
+        pand    xmm14, xmmword ptr [PBLENDW_0xC0_MASK+rip]
+        por     xmm8, xmm14
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        movdqu  xmm4, xmmword ptr [rcx]
+        movdqu  xmm5, xmmword ptr [rcx+0x10]
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm2, xmm4
+        pxor    xmm3, xmm5
+        movups  xmmword ptr [r10], xmm0
+        movups  xmmword ptr [r10+0x10], xmm1
+        movups  xmmword ptr [r10+0x20], xmm2
+        movups  xmmword ptr [r10+0x30], xmm3
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+0x10]
+        movdqa  xmm8, xmmword ptr [rsp+0x20]
+        movdqa  xmm9, xmmword ptr [rsp+0x30]
+        movdqa  xmm11, xmmword ptr [rsp+0x40]
+        movdqa  xmm14, xmmword ptr [rsp+0x50]
+        movdqa  xmm15, xmmword ptr [rsp+0x60]
+        add     rsp, 120
+        ret
+
+
+.section .rdata
+.p2align  6
+BLAKE3_IV:
+        .long  0x6A09E667, 0xBB67AE85
+        .long  0x3C6EF372, 0xA54FF53A
+ADD0:   
+        .long  0, 1, 2, 3
+ADD1:
+        .long  4, 4, 4, 4
+BLAKE3_IV_0:
+        .long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+BLAKE3_IV_1:
+        .long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+BLAKE3_IV_2:
+        .long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+BLAKE3_IV_3:
+        .long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+BLAKE3_BLOCK_LEN:
+        .long  64, 64, 64, 64
+CMP_MSB_MASK:
+        .long  0x80000000, 0x80000000, 0x80000000, 0x80000000
+PBLENDW_0x33_MASK:
+        .long  0xFFFFFFFF, 0x00000000, 0xFFFFFFFF, 0x00000000
+PBLENDW_0xCC_MASK:
+        .long  0x00000000, 0xFFFFFFFF, 0x00000000, 0xFFFFFFFF
+PBLENDW_0x3F_MASK:
+        .long  0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0x00000000
+PBLENDW_0xC0_MASK:
+        .long  0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF
diff --git a/thirdparty/blake3/c/blake3_sse2_x86-64_windows_msvc.asm b/thirdparty/blake3/c/blake3_sse2_x86-64_windows_msvc.asm
new file mode 100644
index 000000000..507502f11
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse2_x86-64_windows_msvc.asm
@@ -0,0 +1,2350 @@
+public _blake3_hash_many_sse2
+public blake3_hash_many_sse2
+public blake3_compress_in_place_sse2
+public _blake3_compress_in_place_sse2
+public blake3_compress_xof_sse2
+public _blake3_compress_xof_sse2
+
+_TEXT   SEGMENT ALIGN(16) 'CODE'
+
+ALIGN   16
+blake3_hash_many_sse2 PROC
+_blake3_hash_many_sse2 PROC
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rsi
+        push    rdi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 528
+        and     rsp, 0FFFFFFFFFFFFFFC0H
+        movdqa  xmmword ptr [rsp+170H], xmm6
+        movdqa  xmmword ptr [rsp+180H], xmm7
+        movdqa  xmmword ptr [rsp+190H], xmm8
+        movdqa  xmmword ptr [rsp+1A0H], xmm9
+        movdqa  xmmword ptr [rsp+1B0H], xmm10
+        movdqa  xmmword ptr [rsp+1C0H], xmm11
+        movdqa  xmmword ptr [rsp+1D0H], xmm12
+        movdqa  xmmword ptr [rsp+1E0H], xmm13
+        movdqa  xmmword ptr [rsp+1F0H], xmm14
+        movdqa  xmmword ptr [rsp+200H], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+68H]
+        movzx   r9, byte ptr [rbp+70H]
+        neg     r9d
+        movd    xmm0, r9d
+        pshufd  xmm0, xmm0, 00H
+        movdqa  xmmword ptr [rsp+130H], xmm0
+        movdqa  xmm1, xmm0
+        pand    xmm1, xmmword ptr [ADD0]
+        pand    xmm0, xmmword ptr [ADD1]
+        movdqa  xmmword ptr [rsp+150H], xmm0
+        movd    xmm0, r8d
+        pshufd  xmm0, xmm0, 00H
+        paddd   xmm0, xmm1
+        movdqa  xmmword ptr [rsp+110H], xmm0
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK]
+        pcmpgtd xmm1, xmm0
+        shr     r8, 32
+        movd    xmm2, r8d
+        pshufd  xmm2, xmm2, 00H
+        psubd   xmm2, xmm1
+        movdqa  xmmword ptr [rsp+120H], xmm2
+        mov     rbx, qword ptr [rbp+90H]
+        mov     r15, rdx
+        shl     r15, 6
+        movzx   r13d, byte ptr [rbp+78H]
+        movzx   r12d, byte ptr [rbp+88H]
+        cmp     rsi, 4
+        jc      final3blocks
+outerloop4:
+        movdqu  xmm3, xmmword ptr [rcx]
+        pshufd  xmm0, xmm3, 00H
+        pshufd  xmm1, xmm3, 55H
+        pshufd  xmm2, xmm3, 0AAH
+        pshufd  xmm3, xmm3, 0FFH
+        movdqu  xmm7, xmmword ptr [rcx+10H]
+        pshufd  xmm4, xmm7, 00H
+        pshufd  xmm5, xmm7, 55H
+        pshufd  xmm6, xmm7, 0AAH
+        pshufd  xmm7, xmm7, 0FFH
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+innerloop4:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movdqu  xmm8, xmmword ptr [r8+rdx-40H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-40H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-40H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-40H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp], xmm8
+        movdqa  xmmword ptr [rsp+10H], xmm9
+        movdqa  xmmword ptr [rsp+20H], xmm12
+        movdqa  xmmword ptr [rsp+30H], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-30H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-30H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-30H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-30H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+40H], xmm8
+        movdqa  xmmword ptr [rsp+50H], xmm9
+        movdqa  xmmword ptr [rsp+60H], xmm12
+        movdqa  xmmword ptr [rsp+70H], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-20H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-20H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-20H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-20H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+80H], xmm8
+        movdqa  xmmword ptr [rsp+90H], xmm9
+        movdqa  xmmword ptr [rsp+0A0H], xmm12
+        movdqa  xmmword ptr [rsp+0B0H], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-10H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-10H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-10H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-10H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0C0H], xmm8
+        movdqa  xmmword ptr [rsp+0D0H], xmm9
+        movdqa  xmmword ptr [rsp+0E0H], xmm12
+        movdqa  xmmword ptr [rsp+0F0H], xmm13
+        movdqa  xmm9, xmmword ptr [BLAKE3_IV_1]
+        movdqa  xmm10, xmmword ptr [BLAKE3_IV_2]
+        movdqa  xmm11, xmmword ptr [BLAKE3_IV_3]
+        movdqa  xmm12, xmmword ptr [rsp+110H]
+        movdqa  xmm13, xmmword ptr [rsp+120H]
+        movdqa  xmm14, xmmword ptr [BLAKE3_BLOCK_LEN]
+        movd    xmm15, eax
+        pshufd  xmm15, xmm15, 00H
+        prefetcht0 byte ptr [r8+rdx+80H]
+        prefetcht0 byte ptr [r9+rdx+80H]
+        prefetcht0 byte ptr [r10+rdx+80H]
+        prefetcht0 byte ptr [r11+rdx+80H]
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+20H]
+        paddd   xmm2, xmmword ptr [rsp+40H]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        movdqa  xmm8, xmmword ptr [BLAKE3_IV_0]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+10H]
+        paddd   xmm1, xmmword ptr [rsp+30H]
+        paddd   xmm2, xmmword ptr [rsp+50H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+80H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp+0C0H]
+        paddd   xmm3, xmmword ptr [rsp+0E0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+90H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+0D0H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+20H]
+        paddd   xmm1, xmmword ptr [rsp+30H]
+        paddd   xmm2, xmmword ptr [rsp+70H]
+        paddd   xmm3, xmmword ptr [rsp+40H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+60H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0D0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+10H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+90H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0B0H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp+0E0H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+30H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp+0D0H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+40H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+20H]
+        paddd   xmm3, xmmword ptr [rsp+0E0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+60H]
+        paddd   xmm1, xmmword ptr [rsp+90H]
+        paddd   xmm2, xmmword ptr [rsp+0B0H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+50H]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0F0H]
+        paddd   xmm3, xmmword ptr [rsp+10H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0A0H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+0E0H]
+        paddd   xmm3, xmmword ptr [rsp+0D0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+70H]
+        paddd   xmm1, xmmword ptr [rsp+90H]
+        paddd   xmm2, xmmword ptr [rsp+30H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+40H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+50H]
+        paddd   xmm3, xmmword ptr [rsp+10H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+20H]
+        paddd   xmm2, xmmword ptr [rsp+80H]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0C0H]
+        paddd   xmm1, xmmword ptr [rsp+90H]
+        paddd   xmm2, xmmword ptr [rsp+0F0H]
+        paddd   xmm3, xmmword ptr [rsp+0E0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0D0H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+0A0H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+70H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+20H]
+        paddd   xmm1, xmmword ptr [rsp+30H]
+        paddd   xmm2, xmmword ptr [rsp+10H]
+        paddd   xmm3, xmmword ptr [rsp+40H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+90H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+80H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0E0H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp+0C0H]
+        paddd   xmm3, xmmword ptr [rsp+10H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0D0H]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+20H]
+        paddd   xmm3, xmmword ptr [rsp+40H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+30H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp+60H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0B0H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp+10H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0F0H]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+90H]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0E0H]
+        paddd   xmm1, xmmword ptr [rsp+20H]
+        paddd   xmm2, xmmword ptr [rsp+30H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        pshuflw xmm15, xmm15, 0B1H
+        pshufhw xmm15, xmm15, 0B1H
+        pshuflw xmm12, xmm12, 0B1H
+        pshufhw xmm12, xmm12, 0B1H
+        pshuflw xmm13, xmm13, 0B1H
+        pshufhw xmm13, xmm13, 0B1H
+        pshuflw xmm14, xmm14, 0B1H
+        pshufhw xmm14, xmm14, 0B1H
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0A0H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+40H]
+        paddd   xmm3, xmmword ptr [rsp+0D0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmm15
+        psrld   xmm15, 8
+        pslld   xmm8, 24
+        pxor    xmm15, xmm8
+        movdqa  xmm8, xmm12
+        psrld   xmm12, 8
+        pslld   xmm8, 24
+        pxor    xmm12, xmm8
+        movdqa  xmm8, xmm13
+        psrld   xmm13, 8
+        pslld   xmm8, 24
+        pxor    xmm13, xmm8
+        movdqa  xmm8, xmm14
+        psrld   xmm14, 8
+        pslld   xmm8, 24
+        pxor    xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        pxor    xmm0, xmm8
+        pxor    xmm1, xmm9
+        pxor    xmm2, xmm10
+        pxor    xmm3, xmm11
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        pxor    xmm4, xmm12
+        pxor    xmm5, xmm13
+        pxor    xmm6, xmm14
+        pxor    xmm7, xmm15
+        mov     eax, r13d
+        jne     innerloop4
+        movdqa  xmm9, xmm0
+        punpckldq xmm0, xmm1
+        punpckhdq xmm9, xmm1
+        movdqa  xmm11, xmm2
+        punpckldq xmm2, xmm3
+        punpckhdq xmm11, xmm3
+        movdqa  xmm1, xmm0
+        punpcklqdq xmm0, xmm2
+        punpckhqdq xmm1, xmm2
+        movdqa  xmm3, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm3, xmm11
+        movdqu  xmmword ptr [rbx], xmm0
+        movdqu  xmmword ptr [rbx+20H], xmm1
+        movdqu  xmmword ptr [rbx+40H], xmm9
+        movdqu  xmmword ptr [rbx+60H], xmm3
+        movdqa  xmm9, xmm4
+        punpckldq xmm4, xmm5
+        punpckhdq xmm9, xmm5
+        movdqa  xmm11, xmm6
+        punpckldq xmm6, xmm7
+        punpckhdq xmm11, xmm7
+        movdqa  xmm5, xmm4
+        punpcklqdq xmm4, xmm6
+        punpckhqdq xmm5, xmm6
+        movdqa  xmm7, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm7, xmm11
+        movdqu  xmmword ptr [rbx+10H], xmm4
+        movdqu  xmmword ptr [rbx+30H], xmm5
+        movdqu  xmmword ptr [rbx+50H], xmm9
+        movdqu  xmmword ptr [rbx+70H], xmm7
+        movdqa  xmm1, xmmword ptr [rsp+110H]
+        movdqa  xmm0, xmm1
+        paddd   xmm1, xmmword ptr [rsp+150H]
+        movdqa  xmmword ptr [rsp+110H], xmm1
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK]
+        pcmpgtd xmm0, xmm1
+        movdqa  xmm1, xmmword ptr [rsp+120H]
+        psubd   xmm1, xmm0
+        movdqa  xmmword ptr [rsp+120H], xmm1
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+        cmp     rsi, 4
+        jnc     outerloop4
+        test    rsi, rsi
+        jne     final3blocks
+unwind:
+        movdqa  xmm6, xmmword ptr [rsp+170H]
+        movdqa  xmm7, xmmword ptr [rsp+180H]
+        movdqa  xmm8, xmmword ptr [rsp+190H]
+        movdqa  xmm9, xmmword ptr [rsp+1A0H]
+        movdqa  xmm10, xmmword ptr [rsp+1B0H]
+        movdqa  xmm11, xmmword ptr [rsp+1C0H]
+        movdqa  xmm12, xmmword ptr [rsp+1D0H]
+        movdqa  xmm13, xmmword ptr [rsp+1E0H]
+        movdqa  xmm14, xmmword ptr [rsp+1F0H]
+        movdqa  xmm15, xmmword ptr [rsp+200H]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rdi
+        pop     rsi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+ALIGN   16
+final3blocks:
+        test    esi, 2H
+        je      final1block
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movaps  xmm8, xmm0
+        movaps  xmm9, xmm1
+        movd    xmm13, dword ptr [rsp+110H]
+        movd    xmm14, dword ptr [rsp+120H]
+        punpckldq xmm13, xmm14
+        movaps  xmmword ptr [rsp], xmm13
+        movd    xmm14, dword ptr [rsp+114H]
+        movd    xmm13, dword ptr [rsp+124H]
+        punpckldq xmm14, xmm13
+        movaps  xmmword ptr [rsp+10H], xmm14
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+innerloop2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        movaps  xmm10, xmm2
+        movups  xmm4, xmmword ptr [r8+rdx-40H]
+        movups  xmm5, xmmword ptr [r8+rdx-30H]
+        movaps  xmm3, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm3, xmm5, 221
+        movaps  xmm5, xmm3
+        movups  xmm6, xmmword ptr [r8+rdx-20H]
+        movups  xmm7, xmmword ptr [r8+rdx-10H]
+        movaps  xmm3, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm3, xmm7, 221
+        pshufd  xmm7, xmm3, 93H
+        movups  xmm12, xmmword ptr [r9+rdx-40H]
+        movups  xmm13, xmmword ptr [r9+rdx-30H]
+        movaps  xmm11, xmm12
+        shufps  xmm12, xmm13, 136
+        shufps  xmm11, xmm13, 221
+        movaps  xmm13, xmm11
+        movups  xmm14, xmmword ptr [r9+rdx-20H]
+        movups  xmm15, xmmword ptr [r9+rdx-10H]
+        movaps  xmm11, xmm14
+        shufps  xmm14, xmm15, 136
+        pshufd  xmm14, xmm14, 93H
+        shufps  xmm11, xmm15, 221
+        pshufd  xmm15, xmm11, 93H
+        shl     rax, 20H
+        or      rax, 40H
+        movd    xmm3, rax
+        movdqa  xmmword ptr [rsp+20H], xmm3
+        movaps  xmm3, xmmword ptr [rsp]
+        movaps  xmm11, xmmword ptr [rsp+10H]
+        punpcklqdq xmm3, xmmword ptr [rsp+20H]
+        punpcklqdq xmm11, xmmword ptr [rsp+20H]
+        mov     al, 7
+roundloop2:
+        paddd   xmm0, xmm4
+        paddd   xmm8, xmm12
+        movaps  xmmword ptr [rsp+20H], xmm4
+        movaps  xmmword ptr [rsp+30H], xmm12
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        pshuflw xmm11, xmm11, 0B1H
+        pshufhw xmm11, xmm11, 0B1H
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm5
+        paddd   xmm8, xmm13
+        movaps  xmmword ptr [rsp+40H], xmm5
+        movaps  xmmword ptr [rsp+50H], xmm13
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movdqa  xmm13, xmm3
+        psrld   xmm3, 8
+        pslld   xmm13, 24
+        pxor    xmm3, xmm13
+        movdqa  xmm13, xmm11
+        psrld   xmm11, 8
+        pslld   xmm13, 24
+        pxor    xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm8, xmm8, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm11, xmm11, 4EH
+        pshufd  xmm2, xmm2, 39H
+        pshufd  xmm10, xmm10, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm8, xmm14
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        pshuflw xmm11, xmm11, 0B1H
+        pshufhw xmm11, xmm11, 0B1H
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm7
+        paddd   xmm8, xmm15
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movdqa  xmm13, xmm3
+        psrld   xmm3, 8
+        pslld   xmm13, 24
+        pxor    xmm3, xmm13
+        movdqa  xmm13, xmm11
+        psrld   xmm11, 8
+        pslld   xmm13, 24
+        pxor    xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm8, xmm8, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm11, xmm11, 4EH
+        pshufd  xmm2, xmm2, 93H
+        pshufd  xmm10, xmm10, 93H
+        dec     al
+        je      endroundloop2
+        movdqa  xmm12, xmmword ptr [rsp+20H]
+        movdqa  xmm5, xmmword ptr [rsp+40H]
+        pshufd  xmm13, xmm12, 0FH
+        shufps  xmm12, xmm5, 214
+        pshufd  xmm4, xmm12, 39H
+        movdqa  xmm12, xmm6
+        shufps  xmm12, xmm7, 250
+        pand    xmm13, xmmword ptr [PBLENDW_0x33_MASK]
+        pand    xmm12, xmmword ptr [PBLENDW_0xCC_MASK]
+        por     xmm13, xmm12
+        movdqa  xmmword ptr [rsp+20H], xmm13
+        movdqa  xmm12, xmm7
+        punpcklqdq xmm12, xmm5
+        movdqa  xmm13, xmm6
+        pand    xmm12, xmmword ptr [PBLENDW_0x3F_MASK]
+        pand    xmm13, xmmword ptr [PBLENDW_0xC0_MASK]
+        por     xmm12, xmm13
+        pshufd  xmm12, xmm12, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmmword ptr [rsp+40H], xmm12
+        movdqa  xmm5, xmmword ptr [rsp+30H]
+        movdqa  xmm13, xmmword ptr [rsp+50H]
+        pshufd  xmm6, xmm5, 0FH
+        shufps  xmm5, xmm13, 214
+        pshufd  xmm12, xmm5, 39H
+        movdqa  xmm5, xmm14
+        shufps  xmm5, xmm15, 250
+        pand    xmm6, xmmword ptr [PBLENDW_0x33_MASK]
+        pand    xmm5, xmmword ptr [PBLENDW_0xCC_MASK]
+        por     xmm6, xmm5
+        movdqa  xmm5, xmm15
+        punpcklqdq xmm5, xmm13
+        movdqa  xmmword ptr [rsp+30H], xmm2
+        movdqa  xmm2, xmm14
+        pand    xmm5, xmmword ptr [PBLENDW_0x3F_MASK]
+        pand    xmm2, xmmword ptr [PBLENDW_0xC0_MASK]
+        por     xmm5, xmm2
+        movdqa  xmm2, xmmword ptr [rsp+30H]
+        pshufd  xmm5, xmm5, 78H
+        punpckhdq xmm13, xmm15
+        punpckldq xmm14, xmm13
+        pshufd  xmm15, xmm14, 1EH
+        movdqa  xmm13, xmm6
+        movdqa  xmm14, xmm5
+        movdqa  xmm5, xmmword ptr [rsp+20H]
+        movdqa  xmm6, xmmword ptr [rsp+40H]
+        jmp     roundloop2
+endroundloop2:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm8, xmm10
+        pxor    xmm9, xmm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop2
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+10H], xmm1
+        movups  xmmword ptr [rbx+20H], xmm8
+        movups  xmmword ptr [rbx+30H], xmm9
+        mov     eax, dword ptr [rsp+130H]
+        neg     eax
+        mov    r10d, dword ptr [rsp+110H+8*rax]
+        mov    r11d, dword ptr [rsp+120H+8*rax]
+        mov dword ptr [rsp+110H], r10d
+        mov dword ptr [rsp+120H], r11d
+        add     rdi, 16
+        add     rbx, 64
+        sub     rsi, 2
+final1block:
+        test    esi, 1H
+        je      unwind
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movd    xmm13, dword ptr [rsp+110H]
+        movd    xmm14, dword ptr [rsp+120H]
+        punpckldq xmm13, xmm14
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+innerloop1:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        shl     rax, 32
+        or      rax, 64
+        movd    xmm12, rax
+        movdqa  xmm3, xmm13
+        punpcklqdq xmm3, xmm12
+        movups  xmm4, xmmword ptr [r8+rdx-40H]
+        movups  xmm5, xmmword ptr [r8+rdx-30H]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [r8+rdx-20H]
+        movups  xmm7, xmmword ptr [r8+rdx-10H]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 93H
+        mov     al, 7
+roundloop1:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 93H
+        dec     al
+        jz      endroundloop1
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0FH
+        pshufd  xmm4, xmm8, 39H
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm10, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK]
+        pand    xmm10, xmmword ptr [PBLENDW_0xC0_MASK]
+        por     xmm8, xmm10
+        pshufd  xmm8, xmm8, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     roundloop1
+endroundloop1:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop1
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+10H], xmm1
+        jmp     unwind
+_blake3_hash_many_sse2 ENDP
+blake3_hash_many_sse2 ENDP
+
+blake3_compress_in_place_sse2 PROC
+_blake3_compress_in_place_sse2 PROC
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+10H], xmm7
+        movdqa  xmmword ptr [rsp+20H], xmm8
+        movdqa  xmmword ptr [rsp+30H], xmm9
+        movdqa  xmmword ptr [rsp+40H], xmm11
+        movdqa  xmmword ptr [rsp+50H], xmm14
+        movdqa  xmmword ptr [rsp+60H], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        movzx   eax, byte ptr [rsp+0A0H]
+        movzx   r8d, r8b
+        shl     rax, 32
+        add     r8, rax
+        movd    xmm3, r9
+        movd    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+10H]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+20H]
+        movups  xmm7, xmmword ptr [rdx+30H]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 93H
+        mov     al, 7
+@@:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 93H
+        dec     al
+        jz      @F
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0FH
+        pshufd  xmm4, xmm8, 39H
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm14, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK]
+        pand    xmm14, xmmword ptr [PBLENDW_0xC0_MASK]
+        por     xmm8, xmm14
+        pshufd  xmm8, xmm8, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     @B
+@@:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        movups  xmmword ptr [rcx], xmm0
+        movups  xmmword ptr [rcx+10H], xmm1
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+10H]
+        movdqa  xmm8, xmmword ptr [rsp+20H]
+        movdqa  xmm9, xmmword ptr [rsp+30H]
+        movdqa  xmm11, xmmword ptr [rsp+40H]
+        movdqa  xmm14, xmmword ptr [rsp+50H]
+        movdqa  xmm15, xmmword ptr [rsp+60H]
+        add     rsp, 120
+        ret
+_blake3_compress_in_place_sse2 ENDP
+blake3_compress_in_place_sse2 ENDP
+
+ALIGN 16
+blake3_compress_xof_sse2 PROC
+_blake3_compress_xof_sse2 PROC
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+10H], xmm7
+        movdqa  xmmword ptr [rsp+20H], xmm8
+        movdqa  xmmword ptr [rsp+30H], xmm9
+        movdqa  xmmword ptr [rsp+40H], xmm11
+        movdqa  xmmword ptr [rsp+50H], xmm14
+        movdqa  xmmword ptr [rsp+60H], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        movzx   eax, byte ptr [rsp+0A0H]
+        movzx   r8d, r8b
+        mov     r10, qword ptr [rsp+0A8H]
+        shl     rax, 32
+        add     r8, rax
+        movd    xmm3, r9
+        movd    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+10H]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+20H]
+        movups  xmm7, xmmword ptr [rdx+30H]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 93H
+        mov     al, 7
+@@:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshuflw xmm3, xmm3, 0B1H
+        pshufhw xmm3, xmm3, 0B1H
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        movdqa  xmm14, xmm3
+        psrld   xmm3, 8
+        pslld   xmm14, 24
+        pxor    xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 93H
+        dec     al
+        jz      @F
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0FH
+        pshufd  xmm4, xmm8, 39H
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pand    xmm9, xmmword ptr [PBLENDW_0x33_MASK]
+        pand    xmm8, xmmword ptr [PBLENDW_0xCC_MASK]
+        por     xmm9, xmm8
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        movdqa  xmm14, xmm6
+        pand    xmm8, xmmword ptr [PBLENDW_0x3F_MASK]
+        pand    xmm14, xmmword ptr [PBLENDW_0xC0_MASK]
+        por     xmm8, xmm14
+        pshufd  xmm8, xmm8, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     @B
+@@:
+        movdqu  xmm4, xmmword ptr [rcx]
+        movdqu  xmm5, xmmword ptr [rcx+10H]
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm2, xmm4
+        pxor    xmm3, xmm5
+        movups  xmmword ptr [r10], xmm0
+        movups  xmmword ptr [r10+10H], xmm1
+        movups  xmmword ptr [r10+20H], xmm2
+        movups  xmmword ptr [r10+30H], xmm3
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+10H]
+        movdqa  xmm8, xmmword ptr [rsp+20H]
+        movdqa  xmm9, xmmword ptr [rsp+30H]
+        movdqa  xmm11, xmmword ptr [rsp+40H]
+        movdqa  xmm14, xmmword ptr [rsp+50H]
+        movdqa  xmm15, xmmword ptr [rsp+60H]
+        add     rsp, 120
+        ret
+_blake3_compress_xof_sse2 ENDP
+blake3_compress_xof_sse2 ENDP
+
+_TEXT ENDS
+
+
+_RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
+ALIGN   64
+BLAKE3_IV:
+        dd 6A09E667H, 0BB67AE85H, 3C6EF372H, 0A54FF53AH
+
+ADD0:
+        dd 0, 1, 2, 3
+
+ADD1:
+        dd 4 dup (4)
+
+BLAKE3_IV_0:
+        dd 4 dup (6A09E667H)
+
+BLAKE3_IV_1:
+        dd 4 dup (0BB67AE85H)
+
+BLAKE3_IV_2:
+        dd 4 dup (3C6EF372H)
+
+BLAKE3_IV_3:
+        dd 4 dup (0A54FF53AH)
+
+BLAKE3_BLOCK_LEN:
+        dd 4 dup (64)
+
+CMP_MSB_MASK:
+        dd 8 dup(80000000H)
+
+PBLENDW_0x33_MASK:
+       dd 0FFFFFFFFH, 000000000H, 0FFFFFFFFH, 000000000H
+PBLENDW_0xCC_MASK:
+       dd 000000000H, 0FFFFFFFFH, 000000000H, 0FFFFFFFFH
+PBLENDW_0x3F_MASK:
+	dd 0FFFFFFFFH, 0FFFFFFFFH, 0FFFFFFFFH, 000000000H
+PBLENDW_0xC0_MASK:
+       dd 000000000H, 000000000H, 000000000H, 0FFFFFFFFH
+
+_RDATA ENDS
+END
diff --git a/thirdparty/blake3/c/blake3_sse41.c b/thirdparty/blake3/c/blake3_sse41.c
new file mode 100644
index 000000000..4653a856f
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse41.c
@@ -0,0 +1,560 @@
+#include "blake3_impl.h"
+
+#include <immintrin.h>
+
+#define DEGREE 4
+
+#define _mm_shuffle_ps2(a, b, c)                                               \
+  (_mm_castps_si128(                                                           \
+      _mm_shuffle_ps(_mm_castsi128_ps(a), _mm_castsi128_ps(b), (c))))
+
+INLINE __m128i loadu(const uint8_t src[16]) {
+  return _mm_loadu_si128((const __m128i *)src);
+}
+
+INLINE void storeu(__m128i src, uint8_t dest[16]) {
+  _mm_storeu_si128((__m128i *)dest, src);
+}
+
+INLINE __m128i addv(__m128i a, __m128i b) { return _mm_add_epi32(a, b); }
+
+// Note that clang-format doesn't like the name "xor" for some reason.
+INLINE __m128i xorv(__m128i a, __m128i b) { return _mm_xor_si128(a, b); }
+
+INLINE __m128i set1(uint32_t x) { return _mm_set1_epi32((int32_t)x); }
+
+INLINE __m128i set4(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
+  return _mm_setr_epi32((int32_t)a, (int32_t)b, (int32_t)c, (int32_t)d);
+}
+
+INLINE __m128i rot16(__m128i x) {
+  return _mm_shuffle_epi8(
+      x, _mm_set_epi8(13, 12, 15, 14, 9, 8, 11, 10, 5, 4, 7, 6, 1, 0, 3, 2));
+}
+
+INLINE __m128i rot12(__m128i x) {
+  return xorv(_mm_srli_epi32(x, 12), _mm_slli_epi32(x, 32 - 12));
+}
+
+INLINE __m128i rot8(__m128i x) {
+  return _mm_shuffle_epi8(
+      x, _mm_set_epi8(12, 15, 14, 13, 8, 11, 10, 9, 4, 7, 6, 5, 0, 3, 2, 1));
+}
+
+INLINE __m128i rot7(__m128i x) {
+  return xorv(_mm_srli_epi32(x, 7), _mm_slli_epi32(x, 32 - 7));
+}
+
+INLINE void g1(__m128i *row0, __m128i *row1, __m128i *row2, __m128i *row3,
+               __m128i m) {
+  *row0 = addv(addv(*row0, m), *row1);
+  *row3 = xorv(*row3, *row0);
+  *row3 = rot16(*row3);
+  *row2 = addv(*row2, *row3);
+  *row1 = xorv(*row1, *row2);
+  *row1 = rot12(*row1);
+}
+
+INLINE void g2(__m128i *row0, __m128i *row1, __m128i *row2, __m128i *row3,
+               __m128i m) {
+  *row0 = addv(addv(*row0, m), *row1);
+  *row3 = xorv(*row3, *row0);
+  *row3 = rot8(*row3);
+  *row2 = addv(*row2, *row3);
+  *row1 = xorv(*row1, *row2);
+  *row1 = rot7(*row1);
+}
+
+// Note the optimization here of leaving row1 as the unrotated row, rather than
+// row0. All the message loads below are adjusted to compensate for this. See
+// discussion at https://github.com/sneves/blake2-avx2/pull/4
+INLINE void diagonalize(__m128i *row0, __m128i *row2, __m128i *row3) {
+  *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE(2, 1, 0, 3));
+  *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE(1, 0, 3, 2));
+  *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE(0, 3, 2, 1));
+}
+
+INLINE void undiagonalize(__m128i *row0, __m128i *row2, __m128i *row3) {
+  *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE(0, 3, 2, 1));
+  *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE(1, 0, 3, 2));
+  *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE(2, 1, 0, 3));
+}
+
+INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
+                         const uint8_t block[BLAKE3_BLOCK_LEN],
+                         uint8_t block_len, uint64_t counter, uint8_t flags) {
+  rows[0] = loadu((uint8_t *)&cv[0]);
+  rows[1] = loadu((uint8_t *)&cv[4]);
+  rows[2] = set4(IV[0], IV[1], IV[2], IV[3]);
+  rows[3] = set4(counter_low(counter), counter_high(counter),
+                 (uint32_t)block_len, (uint32_t)flags);
+
+  __m128i m0 = loadu(&block[sizeof(__m128i) * 0]);
+  __m128i m1 = loadu(&block[sizeof(__m128i) * 1]);
+  __m128i m2 = loadu(&block[sizeof(__m128i) * 2]);
+  __m128i m3 = loadu(&block[sizeof(__m128i) * 3]);
+
+  __m128i t0, t1, t2, t3, tt;
+
+  // Round 1. The first round permutes the message words from the original
+  // input order, into the groups that get mixed in parallel.
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(2, 0, 2, 0)); //  6  4  2  0
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 3, 1)); //  7  5  3  1
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(2, 0, 2, 0)); // 14 12 10  8
+  t2 = _mm_shuffle_epi32(t2, _MM_SHUFFLE(2, 1, 0, 3));   // 12 10  8 14
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 1, 3, 1)); // 15 13 11  9
+  t3 = _mm_shuffle_epi32(t3, _MM_SHUFFLE(2, 1, 0, 3));   // 13 11  9 15
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 2. This round and all following rounds apply a fixed permutation
+  // to the message words from the round before.
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 3
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 4
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 5
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 6
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+  m0 = t0;
+  m1 = t1;
+  m2 = t2;
+  m3 = t3;
+
+  // Round 7
+  t0 = _mm_shuffle_ps2(m0, m1, _MM_SHUFFLE(3, 1, 1, 2));
+  t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE(0, 3, 2, 1));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
+  t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
+  tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
+  t1 = _mm_blend_epi16(tt, t1, 0xCC);
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
+  diagonalize(&rows[0], &rows[2], &rows[3]);
+  t2 = _mm_unpacklo_epi64(m3, m1);
+  tt = _mm_blend_epi16(t2, m2, 0xC0);
+  t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
+  g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
+  t3 = _mm_unpackhi_epi32(m1, m3);
+  tt = _mm_unpacklo_epi32(m2, t3);
+  t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(0, 1, 3, 2));
+  g2(&rows[0], &rows[1], &rows[2], &rows[3], t3);
+  undiagonalize(&rows[0], &rows[2], &rows[3]);
+}
+
+void blake3_compress_in_place_sse41(uint32_t cv[8],
+                                    const uint8_t block[BLAKE3_BLOCK_LEN],
+                                    uint8_t block_len, uint64_t counter,
+                                    uint8_t flags) {
+  __m128i rows[4];
+  compress_pre(rows, cv, block, block_len, counter, flags);
+  storeu(xorv(rows[0], rows[2]), (uint8_t *)&cv[0]);
+  storeu(xorv(rows[1], rows[3]), (uint8_t *)&cv[4]);
+}
+
+void blake3_compress_xof_sse41(const uint32_t cv[8],
+                               const uint8_t block[BLAKE3_BLOCK_LEN],
+                               uint8_t block_len, uint64_t counter,
+                               uint8_t flags, uint8_t out[64]) {
+  __m128i rows[4];
+  compress_pre(rows, cv, block, block_len, counter, flags);
+  storeu(xorv(rows[0], rows[2]), &out[0]);
+  storeu(xorv(rows[1], rows[3]), &out[16]);
+  storeu(xorv(rows[2], loadu((uint8_t *)&cv[0])), &out[32]);
+  storeu(xorv(rows[3], loadu((uint8_t *)&cv[4])), &out[48]);
+}
+
+INLINE void round_fn(__m128i v[16], __m128i m[16], size_t r) {
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][0]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][2]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][4]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][6]]);
+  v[0] = addv(v[0], v[4]);
+  v[1] = addv(v[1], v[5]);
+  v[2] = addv(v[2], v[6]);
+  v[3] = addv(v[3], v[7]);
+  v[12] = xorv(v[12], v[0]);
+  v[13] = xorv(v[13], v[1]);
+  v[14] = xorv(v[14], v[2]);
+  v[15] = xorv(v[15], v[3]);
+  v[12] = rot16(v[12]);
+  v[13] = rot16(v[13]);
+  v[14] = rot16(v[14]);
+  v[15] = rot16(v[15]);
+  v[8] = addv(v[8], v[12]);
+  v[9] = addv(v[9], v[13]);
+  v[10] = addv(v[10], v[14]);
+  v[11] = addv(v[11], v[15]);
+  v[4] = xorv(v[4], v[8]);
+  v[5] = xorv(v[5], v[9]);
+  v[6] = xorv(v[6], v[10]);
+  v[7] = xorv(v[7], v[11]);
+  v[4] = rot12(v[4]);
+  v[5] = rot12(v[5]);
+  v[6] = rot12(v[6]);
+  v[7] = rot12(v[7]);
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][1]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][3]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][5]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][7]]);
+  v[0] = addv(v[0], v[4]);
+  v[1] = addv(v[1], v[5]);
+  v[2] = addv(v[2], v[6]);
+  v[3] = addv(v[3], v[7]);
+  v[12] = xorv(v[12], v[0]);
+  v[13] = xorv(v[13], v[1]);
+  v[14] = xorv(v[14], v[2]);
+  v[15] = xorv(v[15], v[3]);
+  v[12] = rot8(v[12]);
+  v[13] = rot8(v[13]);
+  v[14] = rot8(v[14]);
+  v[15] = rot8(v[15]);
+  v[8] = addv(v[8], v[12]);
+  v[9] = addv(v[9], v[13]);
+  v[10] = addv(v[10], v[14]);
+  v[11] = addv(v[11], v[15]);
+  v[4] = xorv(v[4], v[8]);
+  v[5] = xorv(v[5], v[9]);
+  v[6] = xorv(v[6], v[10]);
+  v[7] = xorv(v[7], v[11]);
+  v[4] = rot7(v[4]);
+  v[5] = rot7(v[5]);
+  v[6] = rot7(v[6]);
+  v[7] = rot7(v[7]);
+
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][8]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][10]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][12]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][14]]);
+  v[0] = addv(v[0], v[5]);
+  v[1] = addv(v[1], v[6]);
+  v[2] = addv(v[2], v[7]);
+  v[3] = addv(v[3], v[4]);
+  v[15] = xorv(v[15], v[0]);
+  v[12] = xorv(v[12], v[1]);
+  v[13] = xorv(v[13], v[2]);
+  v[14] = xorv(v[14], v[3]);
+  v[15] = rot16(v[15]);
+  v[12] = rot16(v[12]);
+  v[13] = rot16(v[13]);
+  v[14] = rot16(v[14]);
+  v[10] = addv(v[10], v[15]);
+  v[11] = addv(v[11], v[12]);
+  v[8] = addv(v[8], v[13]);
+  v[9] = addv(v[9], v[14]);
+  v[5] = xorv(v[5], v[10]);
+  v[6] = xorv(v[6], v[11]);
+  v[7] = xorv(v[7], v[8]);
+  v[4] = xorv(v[4], v[9]);
+  v[5] = rot12(v[5]);
+  v[6] = rot12(v[6]);
+  v[7] = rot12(v[7]);
+  v[4] = rot12(v[4]);
+  v[0] = addv(v[0], m[(size_t)MSG_SCHEDULE[r][9]]);
+  v[1] = addv(v[1], m[(size_t)MSG_SCHEDULE[r][11]]);
+  v[2] = addv(v[2], m[(size_t)MSG_SCHEDULE[r][13]]);
+  v[3] = addv(v[3], m[(size_t)MSG_SCHEDULE[r][15]]);
+  v[0] = addv(v[0], v[5]);
+  v[1] = addv(v[1], v[6]);
+  v[2] = addv(v[2], v[7]);
+  v[3] = addv(v[3], v[4]);
+  v[15] = xorv(v[15], v[0]);
+  v[12] = xorv(v[12], v[1]);
+  v[13] = xorv(v[13], v[2]);
+  v[14] = xorv(v[14], v[3]);
+  v[15] = rot8(v[15]);
+  v[12] = rot8(v[12]);
+  v[13] = rot8(v[13]);
+  v[14] = rot8(v[14]);
+  v[10] = addv(v[10], v[15]);
+  v[11] = addv(v[11], v[12]);
+  v[8] = addv(v[8], v[13]);
+  v[9] = addv(v[9], v[14]);
+  v[5] = xorv(v[5], v[10]);
+  v[6] = xorv(v[6], v[11]);
+  v[7] = xorv(v[7], v[8]);
+  v[4] = xorv(v[4], v[9]);
+  v[5] = rot7(v[5]);
+  v[6] = rot7(v[6]);
+  v[7] = rot7(v[7]);
+  v[4] = rot7(v[4]);
+}
+
+INLINE void transpose_vecs(__m128i vecs[DEGREE]) {
+  // Interleave 32-bit lanes. The low unpack is lanes 00/11 and the high is
+  // 22/33. Note that this doesn't split the vector into two lanes, as the
+  // AVX2 counterparts do.
+  __m128i ab_01 = _mm_unpacklo_epi32(vecs[0], vecs[1]);
+  __m128i ab_23 = _mm_unpackhi_epi32(vecs[0], vecs[1]);
+  __m128i cd_01 = _mm_unpacklo_epi32(vecs[2], vecs[3]);
+  __m128i cd_23 = _mm_unpackhi_epi32(vecs[2], vecs[3]);
+
+  // Interleave 64-bit lanes.
+  __m128i abcd_0 = _mm_unpacklo_epi64(ab_01, cd_01);
+  __m128i abcd_1 = _mm_unpackhi_epi64(ab_01, cd_01);
+  __m128i abcd_2 = _mm_unpacklo_epi64(ab_23, cd_23);
+  __m128i abcd_3 = _mm_unpackhi_epi64(ab_23, cd_23);
+
+  vecs[0] = abcd_0;
+  vecs[1] = abcd_1;
+  vecs[2] = abcd_2;
+  vecs[3] = abcd_3;
+}
+
+INLINE void transpose_msg_vecs(const uint8_t *const *inputs,
+                               size_t block_offset, __m128i out[16]) {
+  out[0] = loadu(&inputs[0][block_offset + 0 * sizeof(__m128i)]);
+  out[1] = loadu(&inputs[1][block_offset + 0 * sizeof(__m128i)]);
+  out[2] = loadu(&inputs[2][block_offset + 0 * sizeof(__m128i)]);
+  out[3] = loadu(&inputs[3][block_offset + 0 * sizeof(__m128i)]);
+  out[4] = loadu(&inputs[0][block_offset + 1 * sizeof(__m128i)]);
+  out[5] = loadu(&inputs[1][block_offset + 1 * sizeof(__m128i)]);
+  out[6] = loadu(&inputs[2][block_offset + 1 * sizeof(__m128i)]);
+  out[7] = loadu(&inputs[3][block_offset + 1 * sizeof(__m128i)]);
+  out[8] = loadu(&inputs[0][block_offset + 2 * sizeof(__m128i)]);
+  out[9] = loadu(&inputs[1][block_offset + 2 * sizeof(__m128i)]);
+  out[10] = loadu(&inputs[2][block_offset + 2 * sizeof(__m128i)]);
+  out[11] = loadu(&inputs[3][block_offset + 2 * sizeof(__m128i)]);
+  out[12] = loadu(&inputs[0][block_offset + 3 * sizeof(__m128i)]);
+  out[13] = loadu(&inputs[1][block_offset + 3 * sizeof(__m128i)]);
+  out[14] = loadu(&inputs[2][block_offset + 3 * sizeof(__m128i)]);
+  out[15] = loadu(&inputs[3][block_offset + 3 * sizeof(__m128i)]);
+  for (size_t i = 0; i < 4; ++i) {
+    _mm_prefetch((const void *)&inputs[i][block_offset + 256], _MM_HINT_T0);
+  }
+  transpose_vecs(&out[0]);
+  transpose_vecs(&out[4]);
+  transpose_vecs(&out[8]);
+  transpose_vecs(&out[12]);
+}
+
+INLINE void load_counters(uint64_t counter, bool increment_counter,
+                          __m128i *out_lo, __m128i *out_hi) {
+  const __m128i mask = _mm_set1_epi32(-(int32_t)increment_counter);
+  const __m128i add0 = _mm_set_epi32(3, 2, 1, 0);
+  const __m128i add1 = _mm_and_si128(mask, add0);
+  __m128i l = _mm_add_epi32(_mm_set1_epi32((int32_t)counter), add1);
+  __m128i carry = _mm_cmpgt_epi32(_mm_xor_si128(add1, _mm_set1_epi32(0x80000000)), 
+                                  _mm_xor_si128(   l, _mm_set1_epi32(0x80000000)));
+  __m128i h = _mm_sub_epi32(_mm_set1_epi32((int32_t)(counter >> 32)), carry);
+  *out_lo = l;
+  *out_hi = h;
+}
+
+static
+void blake3_hash4_sse41(const uint8_t *const *inputs, size_t blocks,
+                        const uint32_t key[8], uint64_t counter,
+                        bool increment_counter, uint8_t flags,
+                        uint8_t flags_start, uint8_t flags_end, uint8_t *out) {
+  __m128i h_vecs[8] = {
+      set1(key[0]), set1(key[1]), set1(key[2]), set1(key[3]),
+      set1(key[4]), set1(key[5]), set1(key[6]), set1(key[7]),
+  };
+  __m128i counter_low_vec, counter_high_vec;
+  load_counters(counter, increment_counter, &counter_low_vec,
+                &counter_high_vec);
+  uint8_t block_flags = flags | flags_start;
+
+  for (size_t block = 0; block < blocks; block++) {
+    if (block + 1 == blocks) {
+      block_flags |= flags_end;
+    }
+    __m128i block_len_vec = set1(BLAKE3_BLOCK_LEN);
+    __m128i block_flags_vec = set1(block_flags);
+    __m128i msg_vecs[16];
+    transpose_msg_vecs(inputs, block * BLAKE3_BLOCK_LEN, msg_vecs);
+
+    __m128i v[16] = {
+        h_vecs[0],       h_vecs[1],        h_vecs[2],     h_vecs[3],
+        h_vecs[4],       h_vecs[5],        h_vecs[6],     h_vecs[7],
+        set1(IV[0]),     set1(IV[1]),      set1(IV[2]),   set1(IV[3]),
+        counter_low_vec, counter_high_vec, block_len_vec, block_flags_vec,
+    };
+    round_fn(v, msg_vecs, 0);
+    round_fn(v, msg_vecs, 1);
+    round_fn(v, msg_vecs, 2);
+    round_fn(v, msg_vecs, 3);
+    round_fn(v, msg_vecs, 4);
+    round_fn(v, msg_vecs, 5);
+    round_fn(v, msg_vecs, 6);
+    h_vecs[0] = xorv(v[0], v[8]);
+    h_vecs[1] = xorv(v[1], v[9]);
+    h_vecs[2] = xorv(v[2], v[10]);
+    h_vecs[3] = xorv(v[3], v[11]);
+    h_vecs[4] = xorv(v[4], v[12]);
+    h_vecs[5] = xorv(v[5], v[13]);
+    h_vecs[6] = xorv(v[6], v[14]);
+    h_vecs[7] = xorv(v[7], v[15]);
+
+    block_flags = flags;
+  }
+
+  transpose_vecs(&h_vecs[0]);
+  transpose_vecs(&h_vecs[4]);
+  // The first four vecs now contain the first half of each output, and the
+  // second four vecs contain the second half of each output.
+  storeu(h_vecs[0], &out[0 * sizeof(__m128i)]);
+  storeu(h_vecs[4], &out[1 * sizeof(__m128i)]);
+  storeu(h_vecs[1], &out[2 * sizeof(__m128i)]);
+  storeu(h_vecs[5], &out[3 * sizeof(__m128i)]);
+  storeu(h_vecs[2], &out[4 * sizeof(__m128i)]);
+  storeu(h_vecs[6], &out[5 * sizeof(__m128i)]);
+  storeu(h_vecs[3], &out[6 * sizeof(__m128i)]);
+  storeu(h_vecs[7], &out[7 * sizeof(__m128i)]);
+}
+
+INLINE void hash_one_sse41(const uint8_t *input, size_t blocks,
+                           const uint32_t key[8], uint64_t counter,
+                           uint8_t flags, uint8_t flags_start,
+                           uint8_t flags_end, uint8_t out[BLAKE3_OUT_LEN]) {
+  uint32_t cv[8];
+  memcpy(cv, key, BLAKE3_KEY_LEN);
+  uint8_t block_flags = flags | flags_start;
+  while (blocks > 0) {
+    if (blocks == 1) {
+      block_flags |= flags_end;
+    }
+    blake3_compress_in_place_sse41(cv, input, BLAKE3_BLOCK_LEN, counter,
+                                   block_flags);
+    input = &input[BLAKE3_BLOCK_LEN];
+    blocks -= 1;
+    block_flags = flags;
+  }
+  memcpy(out, cv, BLAKE3_OUT_LEN);
+}
+
+void blake3_hash_many_sse41(const uint8_t *const *inputs, size_t num_inputs,
+                            size_t blocks, const uint32_t key[8],
+                            uint64_t counter, bool increment_counter,
+                            uint8_t flags, uint8_t flags_start,
+                            uint8_t flags_end, uint8_t *out) {
+  while (num_inputs >= DEGREE) {
+    blake3_hash4_sse41(inputs, blocks, key, counter, increment_counter, flags,
+                       flags_start, flags_end, out);
+    if (increment_counter) {
+      counter += DEGREE;
+    }
+    inputs += DEGREE;
+    num_inputs -= DEGREE;
+    out = &out[DEGREE * BLAKE3_OUT_LEN];
+  }
+  while (num_inputs > 0) {
+    hash_one_sse41(inputs[0], blocks, key, counter, flags, flags_start,
+                   flags_end, out);
+    if (increment_counter) {
+      counter += 1;
+    }
+    inputs += 1;
+    num_inputs -= 1;
+    out = &out[BLAKE3_OUT_LEN];
+  }
+}
diff --git a/thirdparty/blake3/c/blake3_sse41_x86-64_unix.S b/thirdparty/blake3/c/blake3_sse41_x86-64_unix.S
new file mode 100644
index 000000000..a3ff64269
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse41_x86-64_unix.S
@@ -0,0 +1,2028 @@
+#if defined(__ELF__) && defined(__linux__)
+.section .note.GNU-stack,"",%progbits
+#endif
+
+#if defined(__ELF__) && defined(__CET__) && defined(__has_include)
+#if __has_include(<cet.h>)
+#include <cet.h>
+#endif
+#endif
+
+#if !defined(_CET_ENDBR)
+#define _CET_ENDBR
+#endif
+
+.intel_syntax noprefix
+.global blake3_hash_many_sse41
+.global _blake3_hash_many_sse41
+.global blake3_compress_in_place_sse41
+.global _blake3_compress_in_place_sse41
+.global blake3_compress_xof_sse41
+.global _blake3_compress_xof_sse41
+#ifdef __APPLE__
+.text
+#else
+.section .text
+#endif
+        .p2align  6
+_blake3_hash_many_sse41:
+blake3_hash_many_sse41:
+        _CET_ENDBR
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 360
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        neg     r9d
+        movd    xmm0, r9d
+        pshufd  xmm0, xmm0, 0x00
+        movdqa  xmmword ptr [rsp+0x130], xmm0
+        movdqa  xmm1, xmm0
+        pand    xmm1, xmmword ptr [ADD0+rip]
+        pand    xmm0, xmmword ptr [ADD1+rip]
+        movdqa  xmmword ptr [rsp+0x150], xmm0
+        movd    xmm0, r8d
+        pshufd  xmm0, xmm0, 0x00
+        paddd   xmm0, xmm1
+        movdqa  xmmword ptr [rsp+0x110], xmm0
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm1, xmm0
+        shr     r8, 32
+        movd    xmm2, r8d
+        pshufd  xmm2, xmm2, 0x00
+        psubd   xmm2, xmm1
+        movdqa  xmmword ptr [rsp+0x120], xmm2
+        mov     rbx, qword ptr [rbp+0x50]
+        mov     r15, rdx
+        shl     r15, 6
+        movzx   r13d, byte ptr [rbp+0x38]
+        movzx   r12d, byte ptr [rbp+0x48]
+        cmp     rsi, 4
+        jc      3f
+2:
+        movdqu  xmm3, xmmword ptr [rcx]
+        pshufd  xmm0, xmm3, 0x00
+        pshufd  xmm1, xmm3, 0x55
+        pshufd  xmm2, xmm3, 0xAA
+        pshufd  xmm3, xmm3, 0xFF
+        movdqu  xmm7, xmmword ptr [rcx+0x10]
+        pshufd  xmm4, xmm7, 0x00
+        pshufd  xmm5, xmm7, 0x55
+        pshufd  xmm6, xmm7, 0xAA
+        pshufd  xmm7, xmm7, 0xFF
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+9:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x40]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x40]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x40]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x40]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp], xmm8
+        movdqa  xmmword ptr [rsp+0x10], xmm9
+        movdqa  xmmword ptr [rsp+0x20], xmm12
+        movdqa  xmmword ptr [rsp+0x30], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x30]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x30]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x30]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x30]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x40], xmm8
+        movdqa  xmmword ptr [rsp+0x50], xmm9
+        movdqa  xmmword ptr [rsp+0x60], xmm12
+        movdqa  xmmword ptr [rsp+0x70], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x20]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x20]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x20]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x20]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x80], xmm8
+        movdqa  xmmword ptr [rsp+0x90], xmm9
+        movdqa  xmmword ptr [rsp+0xA0], xmm12
+        movdqa  xmmword ptr [rsp+0xB0], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x10]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x10]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x10]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x10]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0xC0], xmm8
+        movdqa  xmmword ptr [rsp+0xD0], xmm9
+        movdqa  xmmword ptr [rsp+0xE0], xmm12
+        movdqa  xmmword ptr [rsp+0xF0], xmm13
+        movdqa  xmm9, xmmword ptr [BLAKE3_IV_1+rip]
+        movdqa  xmm10, xmmword ptr [BLAKE3_IV_2+rip]
+        movdqa  xmm11, xmmword ptr [BLAKE3_IV_3+rip]
+        movdqa  xmm12, xmmword ptr [rsp+0x110]
+        movdqa  xmm13, xmmword ptr [rsp+0x120]
+        movdqa  xmm14, xmmword ptr [BLAKE3_BLOCK_LEN+rip]
+        movd    xmm15, eax
+        pshufd  xmm15, xmm15, 0x00
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [BLAKE3_IV_0+rip]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x80]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x70]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xB0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x50]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xC0]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xA0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0x60]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xF0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        pxor    xmm0, xmm8
+        pxor    xmm1, xmm9
+        pxor    xmm2, xmm10
+        pxor    xmm3, xmm11
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        pxor    xmm4, xmm12
+        pxor    xmm5, xmm13
+        pxor    xmm6, xmm14
+        pxor    xmm7, xmm15
+        mov     eax, r13d
+        jne     9b
+        movdqa  xmm9, xmm0
+        punpckldq xmm0, xmm1
+        punpckhdq xmm9, xmm1
+        movdqa  xmm11, xmm2
+        punpckldq xmm2, xmm3
+        punpckhdq xmm11, xmm3
+        movdqa  xmm1, xmm0
+        punpcklqdq xmm0, xmm2
+        punpckhqdq xmm1, xmm2
+        movdqa  xmm3, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm3, xmm11
+        movdqu  xmmword ptr [rbx], xmm0
+        movdqu  xmmword ptr [rbx+0x20], xmm1
+        movdqu  xmmword ptr [rbx+0x40], xmm9
+        movdqu  xmmword ptr [rbx+0x60], xmm3
+        movdqa  xmm9, xmm4
+        punpckldq xmm4, xmm5
+        punpckhdq xmm9, xmm5
+        movdqa  xmm11, xmm6
+        punpckldq xmm6, xmm7
+        punpckhdq xmm11, xmm7
+        movdqa  xmm5, xmm4
+        punpcklqdq xmm4, xmm6
+        punpckhqdq xmm5, xmm6
+        movdqa  xmm7, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm7, xmm11
+        movdqu  xmmword ptr [rbx+0x10], xmm4
+        movdqu  xmmword ptr [rbx+0x30], xmm5
+        movdqu  xmmword ptr [rbx+0x50], xmm9
+        movdqu  xmmword ptr [rbx+0x70], xmm7
+        movdqa  xmm1, xmmword ptr [rsp+0x110]
+        movdqa  xmm0, xmm1
+        paddd   xmm1, xmmword ptr [rsp+0x150]
+        movdqa  xmmword ptr [rsp+0x110], xmm1
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm0, xmm1
+        movdqa  xmm1, xmmword ptr [rsp+0x120]
+        psubd   xmm1, xmm0
+        movdqa  xmmword ptr [rsp+0x120], xmm1
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+        cmp     rsi, 4
+        jnc     2b
+        test    rsi, rsi
+        jnz     3f
+4:
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align 5
+3:
+        test    esi, 0x2
+        je      3f
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm8, xmm0
+        movaps  xmm9, xmm1
+        movd    xmm13, dword ptr [rsp+0x110]
+        pinsrd  xmm13, dword ptr [rsp+0x120], 1
+        pinsrd  xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        movaps  xmmword ptr [rsp], xmm13
+        movd    xmm14, dword ptr [rsp+0x114]
+        pinsrd  xmm14, dword ptr [rsp+0x124], 1
+        pinsrd  xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        movaps  xmmword ptr [rsp+0x10], xmm14
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movaps  xmm10, xmm2
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm3, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm3, xmm5, 221
+        movaps  xmm5, xmm3
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm3, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm3, xmm7, 221
+        pshufd  xmm7, xmm3, 0x93
+        movups  xmm12, xmmword ptr [r9+rdx-0x40]
+        movups  xmm13, xmmword ptr [r9+rdx-0x30]
+        movaps  xmm11, xmm12
+        shufps  xmm12, xmm13, 136
+        shufps  xmm11, xmm13, 221
+        movaps  xmm13, xmm11
+        movups  xmm14, xmmword ptr [r9+rdx-0x20]
+        movups  xmm15, xmmword ptr [r9+rdx-0x10]
+        movaps  xmm11, xmm14
+        shufps  xmm14, xmm15, 136
+        pshufd  xmm14, xmm14, 0x93
+        shufps  xmm11, xmm15, 221
+        pshufd  xmm15, xmm11, 0x93
+        movaps  xmm3, xmmword ptr [rsp]
+        movaps  xmm11, xmmword ptr [rsp+0x10]
+        pinsrd  xmm3, eax, 3
+        pinsrd  xmm11, eax, 3
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm8, xmm12
+        movaps  xmmword ptr [rsp+0x20], xmm4
+        movaps  xmmword ptr [rsp+0x30], xmm12
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movaps  xmm12, xmmword ptr [ROT16+rip]
+        pshufb  xmm3, xmm12
+        pshufb  xmm11, xmm12
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm5
+        paddd   xmm8, xmm13
+        movaps  xmmword ptr [rsp+0x40], xmm5
+        movaps  xmmword ptr [rsp+0x50], xmm13
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movaps  xmm13, xmmword ptr [ROT8+rip]
+        pshufb  xmm3, xmm13
+        pshufb  xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm8, xmm8, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        pshufd  xmm10, xmm10, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm8, xmm14
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshufb  xmm3, xmm12
+        pshufb  xmm11, xmm12
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm7
+        paddd   xmm8, xmm15
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshufb  xmm3, xmm13
+        pshufb  xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm8, xmm8, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        pshufd  xmm10, xmm10, 0x93
+        dec     al
+        je      9f
+        movdqa  xmm12, xmmword ptr [rsp+0x20]
+        movdqa  xmm5, xmmword ptr [rsp+0x40]
+        pshufd  xmm13, xmm12, 0x0F
+        shufps  xmm12, xmm5, 214
+        pshufd  xmm4, xmm12, 0x39
+        movdqa  xmm12, xmm6
+        shufps  xmm12, xmm7, 250
+        pblendw xmm13, xmm12, 0xCC
+        movdqa  xmm12, xmm7
+        punpcklqdq xmm12, xmm5
+        pblendw xmm12, xmm6, 0xC0
+        pshufd  xmm12, xmm12, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmmword ptr [rsp+0x20], xmm13
+        movdqa  xmmword ptr [rsp+0x40], xmm12
+        movdqa  xmm5, xmmword ptr [rsp+0x30]
+        movdqa  xmm13, xmmword ptr [rsp+0x50]
+        pshufd  xmm6, xmm5, 0x0F
+        shufps  xmm5, xmm13, 214
+        pshufd  xmm12, xmm5, 0x39
+        movdqa  xmm5, xmm14
+        shufps  xmm5, xmm15, 250
+        pblendw xmm6, xmm5, 0xCC
+        movdqa  xmm5, xmm15
+        punpcklqdq xmm5, xmm13
+        pblendw xmm5, xmm14, 0xC0
+        pshufd  xmm5, xmm5, 0x78
+        punpckhdq xmm13, xmm15
+        punpckldq xmm14, xmm13
+        pshufd  xmm15, xmm14, 0x1E
+        movdqa  xmm13, xmm6
+        movdqa  xmm14, xmm5
+        movdqa  xmm5, xmmword ptr [rsp+0x20]
+        movdqa  xmm6, xmmword ptr [rsp+0x40]
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm8, xmm10
+        pxor    xmm9, xmm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        movups  xmmword ptr [rbx+0x20], xmm8
+        movups  xmmword ptr [rbx+0x30], xmm9
+        movdqa  xmm0, xmmword ptr [rsp+0x130]
+        movdqa  xmm1, xmmword ptr [rsp+0x110]
+        movdqa  xmm2, xmmword ptr [rsp+0x120]
+        movdqu  xmm3, xmmword ptr [rsp+0x118]
+        movdqu  xmm4, xmmword ptr [rsp+0x128]
+        blendvps xmm1, xmm3, xmm0
+        blendvps xmm2, xmm4, xmm0
+        movdqa  xmmword ptr [rsp+0x110], xmm1
+        movdqa  xmmword ptr [rsp+0x120], xmm2
+        add     rdi, 16
+        add     rbx, 64
+        sub     rsi, 2
+3:
+        test    esi, 0x1
+        je      4b
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movd    xmm13, dword ptr [rsp+0x110]
+        pinsrd  xmm13, dword ptr [rsp+0x120], 1
+        pinsrd  xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        movaps  xmm14, xmmword ptr [ROT8+rip]
+        movaps  xmm15, xmmword ptr [ROT16+rip]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x40]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movaps  xmm3, xmm13
+        pinsrd  xmm3, eax, 3
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0xCC
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0xC0
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+
+.p2align 6
+blake3_compress_in_place_sse41:
+_blake3_compress_in_place_sse41:
+        _CET_ENDBR
+        movups  xmm0, xmmword ptr [rdi]
+        movups  xmm1, xmmword ptr [rdi+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        shl     r8, 32
+        add     rdx, r8
+        movq    xmm3, rcx
+        movq    xmm4, rdx
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rsi]
+        movups  xmm5, xmmword ptr [rsi+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rsi+0x20]
+        movups  xmm7, xmmword ptr [rsi+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        movaps  xmm14, xmmword ptr [ROT8+rip]
+        movaps  xmm15, xmmword ptr [ROT16+rip]
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0xCC
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0xC0
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        movups  xmmword ptr [rdi], xmm0
+        movups  xmmword ptr [rdi+0x10], xmm1
+        ret
+
+.p2align 6
+blake3_compress_xof_sse41:
+_blake3_compress_xof_sse41:
+        _CET_ENDBR
+        movups  xmm0, xmmword ptr [rdi]
+        movups  xmm1, xmmword ptr [rdi+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movzx   eax, r8b
+        movzx   edx, dl
+        shl     rax, 32
+        add     rdx, rax
+        movq    xmm3, rcx
+        movq    xmm4, rdx
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rsi]
+        movups  xmm5, xmmword ptr [rsi+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rsi+0x20]
+        movups  xmm7, xmmword ptr [rsi+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        movaps  xmm14, xmmword ptr [ROT8+rip]
+        movaps  xmm15, xmmword ptr [ROT16+rip]
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0xCC
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0xC0
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        movdqu  xmm4, xmmword ptr [rdi]
+        movdqu  xmm5, xmmword ptr [rdi+0x10]
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm2, xmm4
+        pxor    xmm3, xmm5
+        movups  xmmword ptr [r9], xmm0
+        movups  xmmword ptr [r9+0x10], xmm1
+        movups  xmmword ptr [r9+0x20], xmm2
+        movups  xmmword ptr [r9+0x30], xmm3
+        ret
+
+
+#ifdef __APPLE__
+.static_data
+#else
+.section .rodata
+#endif
+.p2align  6
+BLAKE3_IV:
+        .long  0x6A09E667, 0xBB67AE85
+        .long  0x3C6EF372, 0xA54FF53A
+ROT16:
+        .byte  2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13
+ROT8:
+        .byte  1, 2, 3, 0, 5, 6, 7, 4, 9, 10, 11, 8, 13, 14, 15, 12
+ADD0:	
+        .long  0, 1, 2, 3
+ADD1:
+	.long  4, 4, 4, 4
+BLAKE3_IV_0:
+	.long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+BLAKE3_IV_1:
+	.long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+BLAKE3_IV_2:
+	.long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+BLAKE3_IV_3:
+	.long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+BLAKE3_BLOCK_LEN:
+	.long  64, 64, 64, 64
+CMP_MSB_MASK:
+	.long  0x80000000, 0x80000000, 0x80000000, 0x80000000
diff --git a/thirdparty/blake3/c/blake3_sse41_x86-64_windows_gnu.S b/thirdparty/blake3/c/blake3_sse41_x86-64_windows_gnu.S
new file mode 100644
index 000000000..02083f9d5
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse41_x86-64_windows_gnu.S
@@ -0,0 +1,2069 @@
+.intel_syntax noprefix
+.global blake3_hash_many_sse41
+.global _blake3_hash_many_sse41
+.global blake3_compress_in_place_sse41
+.global _blake3_compress_in_place_sse41
+.global blake3_compress_xof_sse41
+.global _blake3_compress_xof_sse41
+.section .text
+        .p2align  6
+_blake3_hash_many_sse41:
+blake3_hash_many_sse41:
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rsi
+        push    rdi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 528
+        and     rsp, 0xFFFFFFFFFFFFFFC0
+        movdqa  xmmword ptr [rsp+0x170], xmm6
+        movdqa  xmmword ptr [rsp+0x180], xmm7
+        movdqa  xmmword ptr [rsp+0x190], xmm8
+        movdqa  xmmword ptr [rsp+0x1A0], xmm9
+        movdqa  xmmword ptr [rsp+0x1B0], xmm10
+        movdqa  xmmword ptr [rsp+0x1C0], xmm11
+        movdqa  xmmword ptr [rsp+0x1D0], xmm12
+        movdqa  xmmword ptr [rsp+0x1E0], xmm13
+        movdqa  xmmword ptr [rsp+0x1F0], xmm14
+        movdqa  xmmword ptr [rsp+0x200], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+0x68]
+        movzx   r9, byte ptr [rbp+0x70]
+        neg     r9d
+        movd    xmm0, r9d
+        pshufd  xmm0, xmm0, 0x00
+        movdqa  xmmword ptr [rsp+0x130], xmm0
+        movdqa  xmm1, xmm0
+        pand    xmm1, xmmword ptr [ADD0+rip]
+        pand    xmm0, xmmword ptr [ADD1+rip]
+        movdqa  xmmword ptr [rsp+0x150], xmm0
+        movd    xmm0, r8d
+        pshufd  xmm0, xmm0, 0x00
+        paddd   xmm0, xmm1
+        movdqa  xmmword ptr [rsp+0x110], xmm0
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm1, xmm0
+        shr     r8, 32
+        movd    xmm2, r8d
+        pshufd  xmm2, xmm2, 0x00
+        psubd   xmm2, xmm1
+        movdqa  xmmword ptr [rsp+0x120], xmm2
+        mov     rbx, qword ptr [rbp+0x90]
+        mov     r15, rdx
+        shl     r15, 6
+        movzx   r13d, byte ptr [rbp+0x78]
+        movzx   r12d, byte ptr [rbp+0x88]
+        cmp     rsi, 4
+        jc      3f
+2:
+        movdqu  xmm3, xmmword ptr [rcx]
+        pshufd  xmm0, xmm3, 0x00
+        pshufd  xmm1, xmm3, 0x55
+        pshufd  xmm2, xmm3, 0xAA
+        pshufd  xmm3, xmm3, 0xFF
+        movdqu  xmm7, xmmword ptr [rcx+0x10]
+        pshufd  xmm4, xmm7, 0x00
+        pshufd  xmm5, xmm7, 0x55
+        pshufd  xmm6, xmm7, 0xAA
+        pshufd  xmm7, xmm7, 0xFF
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        mov     r10, qword ptr [rdi+0x10]
+        mov     r11, qword ptr [rdi+0x18]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+9:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x40]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x40]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x40]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x40]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp], xmm8
+        movdqa  xmmword ptr [rsp+0x10], xmm9
+        movdqa  xmmword ptr [rsp+0x20], xmm12
+        movdqa  xmmword ptr [rsp+0x30], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x30]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x30]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x30]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x30]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x40], xmm8
+        movdqa  xmmword ptr [rsp+0x50], xmm9
+        movdqa  xmmword ptr [rsp+0x60], xmm12
+        movdqa  xmmword ptr [rsp+0x70], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x20]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x20]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x20]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x20]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0x80], xmm8
+        movdqa  xmmword ptr [rsp+0x90], xmm9
+        movdqa  xmmword ptr [rsp+0xA0], xmm12
+        movdqa  xmmword ptr [rsp+0xB0], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-0x10]
+        movdqu  xmm9, xmmword ptr [r9+rdx-0x10]
+        movdqu  xmm10, xmmword ptr [r10+rdx-0x10]
+        movdqu  xmm11, xmmword ptr [r11+rdx-0x10]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0xC0], xmm8
+        movdqa  xmmword ptr [rsp+0xD0], xmm9
+        movdqa  xmmword ptr [rsp+0xE0], xmm12
+        movdqa  xmmword ptr [rsp+0xF0], xmm13
+        movdqa  xmm9, xmmword ptr [BLAKE3_IV_1+rip]
+        movdqa  xmm10, xmmword ptr [BLAKE3_IV_2+rip]
+        movdqa  xmm11, xmmword ptr [BLAKE3_IV_3+rip]
+        movdqa  xmm12, xmmword ptr [rsp+0x110]
+        movdqa  xmm13, xmmword ptr [rsp+0x120]
+        movdqa  xmm14, xmmword ptr [BLAKE3_BLOCK_LEN+rip]
+        movd    xmm15, eax
+        pshufd  xmm15, xmm15, 0x00
+        prefetcht0 [r8+rdx+0x80]
+        prefetcht0 [r9+rdx+0x80]
+        prefetcht0 [r10+rdx+0x80]
+        prefetcht0 [r11+rdx+0x80]
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [BLAKE3_IV_0+rip]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x80]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x70]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x10]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0xD0]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x60]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xB0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x50]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0xE0]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x40]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x50]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xC0]
+        paddd   xmm1, xmmword ptr [rsp+0x90]
+        paddd   xmm2, xmmword ptr [rsp+0xF0]
+        paddd   xmm3, xmmword ptr [rsp+0xE0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0xA0]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x70]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x20]
+        paddd   xmm1, xmmword ptr [rsp+0x30]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x90]
+        paddd   xmm1, xmmword ptr [rsp+0xB0]
+        paddd   xmm2, xmmword ptr [rsp+0x80]
+        paddd   xmm3, xmmword ptr [rsp+0xF0]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0xC0]
+        paddd   xmm3, xmmword ptr [rsp+0x10]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xD0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x20]
+        paddd   xmm3, xmmword ptr [rsp+0x40]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0x30]
+        paddd   xmm1, xmmword ptr [rsp+0xA0]
+        paddd   xmm2, xmmword ptr [rsp+0x60]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xB0]
+        paddd   xmm1, xmmword ptr [rsp+0x50]
+        paddd   xmm2, xmmword ptr [rsp+0x10]
+        paddd   xmm3, xmmword ptr [rsp+0x80]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xF0]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0x90]
+        paddd   xmm3, xmmword ptr [rsp+0x60]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xE0]
+        paddd   xmm1, xmmword ptr [rsp+0x20]
+        paddd   xmm2, xmmword ptr [rsp+0x30]
+        paddd   xmm3, xmmword ptr [rsp+0x70]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+0x100], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0xA0]
+        paddd   xmm1, xmmword ptr [rsp+0xC0]
+        paddd   xmm2, xmmword ptr [rsp+0x40]
+        paddd   xmm3, xmmword ptr [rsp+0xD0]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8+rip]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+0x100]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        pxor    xmm0, xmm8
+        pxor    xmm1, xmm9
+        pxor    xmm2, xmm10
+        pxor    xmm3, xmm11
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        pxor    xmm4, xmm12
+        pxor    xmm5, xmm13
+        pxor    xmm6, xmm14
+        pxor    xmm7, xmm15
+        mov     eax, r13d
+        jne     9b
+        movdqa  xmm9, xmm0
+        punpckldq xmm0, xmm1
+        punpckhdq xmm9, xmm1
+        movdqa  xmm11, xmm2
+        punpckldq xmm2, xmm3
+        punpckhdq xmm11, xmm3
+        movdqa  xmm1, xmm0
+        punpcklqdq xmm0, xmm2
+        punpckhqdq xmm1, xmm2
+        movdqa  xmm3, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm3, xmm11
+        movdqu  xmmword ptr [rbx], xmm0
+        movdqu  xmmword ptr [rbx+0x20], xmm1
+        movdqu  xmmword ptr [rbx+0x40], xmm9
+        movdqu  xmmword ptr [rbx+0x60], xmm3
+        movdqa  xmm9, xmm4
+        punpckldq xmm4, xmm5
+        punpckhdq xmm9, xmm5
+        movdqa  xmm11, xmm6
+        punpckldq xmm6, xmm7
+        punpckhdq xmm11, xmm7
+        movdqa  xmm5, xmm4
+        punpcklqdq xmm4, xmm6
+        punpckhqdq xmm5, xmm6
+        movdqa  xmm7, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm7, xmm11
+        movdqu  xmmword ptr [rbx+0x10], xmm4
+        movdqu  xmmword ptr [rbx+0x30], xmm5
+        movdqu  xmmword ptr [rbx+0x50], xmm9
+        movdqu  xmmword ptr [rbx+0x70], xmm7
+        movdqa  xmm1, xmmword ptr [rsp+0x110]
+        movdqa  xmm0, xmm1
+        paddd   xmm1, xmmword ptr [rsp+0x150]
+        movdqa  xmmword ptr [rsp+0x110], xmm1
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK+rip]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK+rip]
+        pcmpgtd xmm0, xmm1
+        movdqa  xmm1, xmmword ptr [rsp+0x120]
+        psubd   xmm1, xmm0
+        movdqa  xmmword ptr [rsp+0x120], xmm1
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+        cmp     rsi, 4
+        jnc     2b
+        test    rsi, rsi
+        jne     3f
+4:
+        movdqa  xmm6, xmmword ptr [rsp+0x170]
+        movdqa  xmm7, xmmword ptr [rsp+0x180]
+        movdqa  xmm8, xmmword ptr [rsp+0x190]
+        movdqa  xmm9, xmmword ptr [rsp+0x1A0]
+        movdqa  xmm10, xmmword ptr [rsp+0x1B0]
+        movdqa  xmm11, xmmword ptr [rsp+0x1C0]
+        movdqa  xmm12, xmmword ptr [rsp+0x1D0]
+        movdqa  xmm13, xmmword ptr [rsp+0x1E0]
+        movdqa  xmm14, xmmword ptr [rsp+0x1F0]
+        movdqa  xmm15, xmmword ptr [rsp+0x200]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rdi
+        pop     rsi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+.p2align 5
+3:
+        test    esi, 0x2
+        je      3f
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm8, xmm0
+        movaps  xmm9, xmm1
+        movd    xmm13, dword ptr [rsp+0x110]
+        pinsrd  xmm13, dword ptr [rsp+0x120], 1
+        pinsrd  xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        movaps  xmmword ptr [rsp], xmm13
+        movd    xmm14, dword ptr [rsp+0x114]
+        pinsrd  xmm14, dword ptr [rsp+0x124], 1
+        pinsrd  xmm14, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        movaps  xmmword ptr [rsp+0x10], xmm14
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+0x8]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movaps  xmm10, xmm2
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm3, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm3, xmm5, 221
+        movaps  xmm5, xmm3
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm3, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm3, xmm7, 221
+        pshufd  xmm7, xmm3, 0x93
+        movups  xmm12, xmmword ptr [r9+rdx-0x40]
+        movups  xmm13, xmmword ptr [r9+rdx-0x30]
+        movaps  xmm11, xmm12
+        shufps  xmm12, xmm13, 136
+        shufps  xmm11, xmm13, 221
+        movaps  xmm13, xmm11
+        movups  xmm14, xmmword ptr [r9+rdx-0x20]
+        movups  xmm15, xmmword ptr [r9+rdx-0x10]
+        movaps  xmm11, xmm14
+        shufps  xmm14, xmm15, 136
+        pshufd  xmm14, xmm14, 0x93
+        shufps  xmm11, xmm15, 221
+        pshufd  xmm15, xmm11, 0x93
+        movaps  xmm3, xmmword ptr [rsp]
+        movaps  xmm11, xmmword ptr [rsp+0x10]
+        pinsrd  xmm3, eax, 3
+        pinsrd  xmm11, eax, 3
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm8, xmm12
+        movaps  xmmword ptr [rsp+0x20], xmm4
+        movaps  xmmword ptr [rsp+0x30], xmm12
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movaps  xmm12, xmmword ptr [ROT16+rip]
+        pshufb  xmm3, xmm12
+        pshufb  xmm11, xmm12
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm5
+        paddd   xmm8, xmm13
+        movaps  xmmword ptr [rsp+0x40], xmm5
+        movaps  xmmword ptr [rsp+0x50], xmm13
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movaps  xmm13, xmmword ptr [ROT8+rip]
+        pshufb  xmm3, xmm13
+        pshufb  xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm8, xmm8, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        pshufd  xmm10, xmm10, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm8, xmm14
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshufb  xmm3, xmm12
+        pshufb  xmm11, xmm12
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm7
+        paddd   xmm8, xmm15
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshufb  xmm3, xmm13
+        pshufb  xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm8, xmm8, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm11, xmm11, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        pshufd  xmm10, xmm10, 0x93
+        dec     al
+        je      9f
+        movdqa  xmm12, xmmword ptr [rsp+0x20]
+        movdqa  xmm5, xmmword ptr [rsp+0x40]
+        pshufd  xmm13, xmm12, 0x0F
+        shufps  xmm12, xmm5, 214
+        pshufd  xmm4, xmm12, 0x39
+        movdqa  xmm12, xmm6
+        shufps  xmm12, xmm7, 250
+        pblendw xmm13, xmm12, 0xCC
+        movdqa  xmm12, xmm7
+        punpcklqdq xmm12, xmm5
+        pblendw xmm12, xmm6, 0xC0
+        pshufd  xmm12, xmm12, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmmword ptr [rsp+0x20], xmm13
+        movdqa  xmmword ptr [rsp+0x40], xmm12
+        movdqa  xmm5, xmmword ptr [rsp+0x30]
+        movdqa  xmm13, xmmword ptr [rsp+0x50]
+        pshufd  xmm6, xmm5, 0x0F
+        shufps  xmm5, xmm13, 214
+        pshufd  xmm12, xmm5, 0x39
+        movdqa  xmm5, xmm14
+        shufps  xmm5, xmm15, 250
+        pblendw xmm6, xmm5, 0xCC
+        movdqa  xmm5, xmm15
+        punpcklqdq xmm5, xmm13
+        pblendw xmm5, xmm14, 0xC0
+        pshufd  xmm5, xmm5, 0x78
+        punpckhdq xmm13, xmm15
+        punpckldq xmm14, xmm13
+        pshufd  xmm15, xmm14, 0x1E
+        movdqa  xmm13, xmm6
+        movdqa  xmm14, xmm5
+        movdqa  xmm5, xmmword ptr [rsp+0x20]
+        movdqa  xmm6, xmmword ptr [rsp+0x40]
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm8, xmm10
+        pxor    xmm9, xmm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        movups  xmmword ptr [rbx+0x20], xmm8
+        movups  xmmword ptr [rbx+0x30], xmm9
+        movdqa  xmm0, xmmword ptr [rsp+0x130]
+        movdqa  xmm1, xmmword ptr [rsp+0x110]
+        movdqa  xmm2, xmmword ptr [rsp+0x120]
+        movdqu  xmm3, xmmword ptr [rsp+0x118]
+        movdqu  xmm4, xmmword ptr [rsp+0x128]
+        blendvps xmm1, xmm3, xmm0
+        blendvps xmm2, xmm4, xmm0
+        movdqa  xmmword ptr [rsp+0x110], xmm1
+        movdqa  xmmword ptr [rsp+0x120], xmm2
+        add     rdi, 16
+        add     rbx, 64
+        sub     rsi, 2
+3:
+        test    esi, 0x1
+        je      4b
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movd    xmm13, dword ptr [rsp+0x110]
+        pinsrd  xmm13, dword ptr [rsp+0x120], 1
+        pinsrd  xmm13, dword ptr [BLAKE3_BLOCK_LEN+rip], 2
+        movaps  xmm14, xmmword ptr [ROT8+rip]
+        movaps  xmm15, xmmword ptr [ROT16+rip]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+0x80]
+        or      eax, r13d
+        xor     edx, edx
+2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movaps  xmm3, xmm13
+        pinsrd  xmm3, eax, 3
+        movups  xmm4, xmmword ptr [r8+rdx-0x40]
+        movups  xmm5, xmmword ptr [r8+rdx-0x30]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [r8+rdx-0x20]
+        movups  xmm7, xmmword ptr [r8+rdx-0x10]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0xCC
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0xC0
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     2b
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+0x10], xmm1
+        jmp     4b
+
+.p2align 6
+blake3_compress_in_place_sse41:
+_blake3_compress_in_place_sse41:
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+0x10], xmm7
+        movdqa  xmmword ptr [rsp+0x20], xmm8
+        movdqa  xmmword ptr [rsp+0x30], xmm9
+        movdqa  xmmword ptr [rsp+0x40], xmm11
+        movdqa  xmmword ptr [rsp+0x50], xmm14
+        movdqa  xmmword ptr [rsp+0x60], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movzx   eax, byte ptr [rsp+0xA0]
+        movzx   r8d, r8b
+        shl     rax, 32
+        add     r8, rax
+        movq    xmm3, r9
+        movq    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+0x20]
+        movups  xmm7, xmmword ptr [rdx+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        movaps  xmm14, xmmword ptr [ROT8+rip]
+        movaps  xmm15, xmmword ptr [ROT16+rip]
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0xCC
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0xC0
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        movups  xmmword ptr [rcx], xmm0
+        movups  xmmword ptr [rcx+0x10], xmm1
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+0x10]
+        movdqa  xmm8, xmmword ptr [rsp+0x20]
+        movdqa  xmm9, xmmword ptr [rsp+0x30]
+        movdqa  xmm11, xmmword ptr [rsp+0x40]
+        movdqa  xmm14, xmmword ptr [rsp+0x50]
+        movdqa  xmm15, xmmword ptr [rsp+0x60]
+        add     rsp, 120
+        ret
+
+
+.p2align 6
+_blake3_compress_xof_sse41:
+blake3_compress_xof_sse41:
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+0x10], xmm7
+        movdqa  xmmword ptr [rsp+0x20], xmm8
+        movdqa  xmmword ptr [rsp+0x30], xmm9
+        movdqa  xmmword ptr [rsp+0x40], xmm11
+        movdqa  xmmword ptr [rsp+0x50], xmm14
+        movdqa  xmmword ptr [rsp+0x60], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+0x10]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV+rip]
+        movzx   eax, byte ptr [rsp+0xA0]
+        movzx   r8d, r8b
+        mov     r10, qword ptr [rsp+0xA8]
+        shl     rax, 32
+        add     r8, rax
+        movq    xmm3, r9
+        movq    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+0x10]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+0x20]
+        movups  xmm7, xmmword ptr [rdx+0x30]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 0x93
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 0x93
+        movaps  xmm14, xmmword ptr [ROT8+rip]
+        movaps  xmm15, xmmword ptr [ROT16+rip]
+        mov     al, 7
+9:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x93
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x39
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 0x39
+        pshufd  xmm3, xmm3, 0x4E
+        pshufd  xmm2, xmm2, 0x93
+        dec     al
+        jz      9f
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0x0F
+        pshufd  xmm4, xmm8, 0x39
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0xCC
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0xC0
+        pshufd  xmm8, xmm8, 0x78
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 0x1E
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     9b
+9:
+        movdqu  xmm4, xmmword ptr [rcx]
+        movdqu  xmm5, xmmword ptr [rcx+0x10]
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm2, xmm4
+        pxor    xmm3, xmm5
+        movups  xmmword ptr [r10], xmm0
+        movups  xmmword ptr [r10+0x10], xmm1
+        movups  xmmword ptr [r10+0x20], xmm2
+        movups  xmmword ptr [r10+0x30], xmm3
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+0x10]
+        movdqa  xmm8, xmmword ptr [rsp+0x20]
+        movdqa  xmm9, xmmword ptr [rsp+0x30]
+        movdqa  xmm11, xmmword ptr [rsp+0x40]
+        movdqa  xmm14, xmmword ptr [rsp+0x50]
+        movdqa  xmm15, xmmword ptr [rsp+0x60]
+        add     rsp, 120
+        ret
+
+
+.section .rdata
+.p2align  6
+BLAKE3_IV:
+        .long  0x6A09E667, 0xBB67AE85
+        .long  0x3C6EF372, 0xA54FF53A
+ROT16:
+        .byte  2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13
+ROT8:
+        .byte  1, 2, 3, 0, 5, 6, 7, 4, 9, 10, 11, 8, 13, 14, 15, 12
+ADD0:   
+        .long  0, 1, 2, 3
+ADD1:
+        .long  4, 4, 4, 4
+BLAKE3_IV_0:
+        .long  0x6A09E667, 0x6A09E667, 0x6A09E667, 0x6A09E667
+BLAKE3_IV_1:
+        .long  0xBB67AE85, 0xBB67AE85, 0xBB67AE85, 0xBB67AE85
+BLAKE3_IV_2:
+        .long  0x3C6EF372, 0x3C6EF372, 0x3C6EF372, 0x3C6EF372
+BLAKE3_IV_3:
+        .long  0xA54FF53A, 0xA54FF53A, 0xA54FF53A, 0xA54FF53A
+BLAKE3_BLOCK_LEN:
+        .long  64, 64, 64, 64
+CMP_MSB_MASK:
+        .long  0x80000000, 0x80000000, 0x80000000, 0x80000000
diff --git a/thirdparty/blake3/c/blake3_sse41_x86-64_windows_msvc.asm b/thirdparty/blake3/c/blake3_sse41_x86-64_windows_msvc.asm
new file mode 100644
index 000000000..8966c7b84
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_sse41_x86-64_windows_msvc.asm
@@ -0,0 +1,2089 @@
+public _blake3_hash_many_sse41
+public blake3_hash_many_sse41
+public blake3_compress_in_place_sse41
+public _blake3_compress_in_place_sse41
+public blake3_compress_xof_sse41
+public _blake3_compress_xof_sse41
+
+_TEXT   SEGMENT ALIGN(16) 'CODE'
+
+ALIGN   16
+blake3_hash_many_sse41 PROC
+_blake3_hash_many_sse41 PROC
+        push    r15
+        push    r14
+        push    r13
+        push    r12
+        push    rsi
+        push    rdi
+        push    rbx
+        push    rbp
+        mov     rbp, rsp
+        sub     rsp, 528
+        and     rsp, 0FFFFFFFFFFFFFFC0H
+        movdqa  xmmword ptr [rsp+170H], xmm6
+        movdqa  xmmword ptr [rsp+180H], xmm7
+        movdqa  xmmword ptr [rsp+190H], xmm8
+        movdqa  xmmword ptr [rsp+1A0H], xmm9
+        movdqa  xmmword ptr [rsp+1B0H], xmm10
+        movdqa  xmmword ptr [rsp+1C0H], xmm11
+        movdqa  xmmword ptr [rsp+1D0H], xmm12
+        movdqa  xmmword ptr [rsp+1E0H], xmm13
+        movdqa  xmmword ptr [rsp+1F0H], xmm14
+        movdqa  xmmword ptr [rsp+200H], xmm15
+        mov     rdi, rcx
+        mov     rsi, rdx
+        mov     rdx, r8
+        mov     rcx, r9
+        mov     r8, qword ptr [rbp+68H]
+        movzx   r9, byte ptr [rbp+70H]
+        neg     r9d
+        movd    xmm0, r9d
+        pshufd  xmm0, xmm0, 00H
+        movdqa  xmmword ptr [rsp+130H], xmm0
+        movdqa  xmm1, xmm0
+        pand    xmm1, xmmword ptr [ADD0]
+        pand    xmm0, xmmword ptr [ADD1]
+        movdqa  xmmword ptr [rsp+150H], xmm0
+        movd    xmm0, r8d
+        pshufd  xmm0, xmm0, 00H
+        paddd   xmm0, xmm1
+        movdqa  xmmword ptr [rsp+110H], xmm0
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK]
+        pcmpgtd xmm1, xmm0
+        shr     r8, 32
+        movd    xmm2, r8d
+        pshufd  xmm2, xmm2, 00H
+        psubd   xmm2, xmm1
+        movdqa  xmmword ptr [rsp+120H], xmm2
+        mov     rbx, qword ptr [rbp+90H]
+        mov     r15, rdx
+        shl     r15, 6
+        movzx   r13d, byte ptr [rbp+78H]
+        movzx   r12d, byte ptr [rbp+88H]
+        cmp     rsi, 4
+        jc      final3blocks
+outerloop4:
+        movdqu  xmm3, xmmword ptr [rcx]
+        pshufd  xmm0, xmm3, 00H
+        pshufd  xmm1, xmm3, 55H
+        pshufd  xmm2, xmm3, 0AAH
+        pshufd  xmm3, xmm3, 0FFH
+        movdqu  xmm7, xmmword ptr [rcx+10H]
+        pshufd  xmm4, xmm7, 00H
+        pshufd  xmm5, xmm7, 55H
+        pshufd  xmm6, xmm7, 0AAH
+        pshufd  xmm7, xmm7, 0FFH
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        mov     r10, qword ptr [rdi+10H]
+        mov     r11, qword ptr [rdi+18H]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+innerloop4:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movdqu  xmm8, xmmword ptr [r8+rdx-40H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-40H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-40H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-40H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp], xmm8
+        movdqa  xmmword ptr [rsp+10H], xmm9
+        movdqa  xmmword ptr [rsp+20H], xmm12
+        movdqa  xmmword ptr [rsp+30H], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-30H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-30H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-30H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-30H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+40H], xmm8
+        movdqa  xmmword ptr [rsp+50H], xmm9
+        movdqa  xmmword ptr [rsp+60H], xmm12
+        movdqa  xmmword ptr [rsp+70H], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-20H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-20H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-20H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-20H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+80H], xmm8
+        movdqa  xmmword ptr [rsp+90H], xmm9
+        movdqa  xmmword ptr [rsp+0A0H], xmm12
+        movdqa  xmmword ptr [rsp+0B0H], xmm13
+        movdqu  xmm8, xmmword ptr [r8+rdx-10H]
+        movdqu  xmm9, xmmword ptr [r9+rdx-10H]
+        movdqu  xmm10, xmmword ptr [r10+rdx-10H]
+        movdqu  xmm11, xmmword ptr [r11+rdx-10H]
+        movdqa  xmm12, xmm8
+        punpckldq xmm8, xmm9
+        punpckhdq xmm12, xmm9
+        movdqa  xmm14, xmm10
+        punpckldq xmm10, xmm11
+        punpckhdq xmm14, xmm11
+        movdqa  xmm9, xmm8
+        punpcklqdq xmm8, xmm10
+        punpckhqdq xmm9, xmm10
+        movdqa  xmm13, xmm12
+        punpcklqdq xmm12, xmm14
+        punpckhqdq xmm13, xmm14
+        movdqa  xmmword ptr [rsp+0C0H], xmm8
+        movdqa  xmmword ptr [rsp+0D0H], xmm9
+        movdqa  xmmword ptr [rsp+0E0H], xmm12
+        movdqa  xmmword ptr [rsp+0F0H], xmm13
+        movdqa  xmm9, xmmword ptr [BLAKE3_IV_1]
+        movdqa  xmm10, xmmword ptr [BLAKE3_IV_2]
+        movdqa  xmm11, xmmword ptr [BLAKE3_IV_3]
+        movdqa  xmm12, xmmword ptr [rsp+110H]
+        movdqa  xmm13, xmmword ptr [rsp+120H]
+        movdqa  xmm14, xmmword ptr [BLAKE3_BLOCK_LEN]
+        movd    xmm15, eax
+        pshufd  xmm15, xmm15, 00H
+        prefetcht0 byte ptr [r8+rdx+80H]
+        prefetcht0 byte ptr [r9+rdx+80H]
+        prefetcht0 byte ptr [r10+rdx+80H]
+        prefetcht0 byte ptr [r11+rdx+80H]
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+20H]
+        paddd   xmm2, xmmword ptr [rsp+40H]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [BLAKE3_IV_0]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+10H]
+        paddd   xmm1, xmmword ptr [rsp+30H]
+        paddd   xmm2, xmmword ptr [rsp+50H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+80H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp+0C0H]
+        paddd   xmm3, xmmword ptr [rsp+0E0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+90H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+0D0H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+20H]
+        paddd   xmm1, xmmword ptr [rsp+30H]
+        paddd   xmm2, xmmword ptr [rsp+70H]
+        paddd   xmm3, xmmword ptr [rsp+40H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+60H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+0D0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+10H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+90H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0B0H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp+0E0H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+30H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp+0D0H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+40H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+20H]
+        paddd   xmm3, xmmword ptr [rsp+0E0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+60H]
+        paddd   xmm1, xmmword ptr [rsp+90H]
+        paddd   xmm2, xmmword ptr [rsp+0B0H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+50H]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+0F0H]
+        paddd   xmm3, xmmword ptr [rsp+10H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0A0H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+0E0H]
+        paddd   xmm3, xmmword ptr [rsp+0D0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+70H]
+        paddd   xmm1, xmmword ptr [rsp+90H]
+        paddd   xmm2, xmmword ptr [rsp+30H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+40H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+50H]
+        paddd   xmm3, xmmword ptr [rsp+10H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp]
+        paddd   xmm1, xmmword ptr [rsp+20H]
+        paddd   xmm2, xmmword ptr [rsp+80H]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0C0H]
+        paddd   xmm1, xmmword ptr [rsp+90H]
+        paddd   xmm2, xmmword ptr [rsp+0F0H]
+        paddd   xmm3, xmmword ptr [rsp+0E0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0D0H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+0A0H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+70H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+20H]
+        paddd   xmm1, xmmword ptr [rsp+30H]
+        paddd   xmm2, xmmword ptr [rsp+10H]
+        paddd   xmm3, xmmword ptr [rsp+40H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+90H]
+        paddd   xmm1, xmmword ptr [rsp+0B0H]
+        paddd   xmm2, xmmword ptr [rsp+80H]
+        paddd   xmm3, xmmword ptr [rsp+0F0H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0E0H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp+0C0H]
+        paddd   xmm3, xmmword ptr [rsp+10H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0D0H]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+20H]
+        paddd   xmm3, xmmword ptr [rsp+40H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+30H]
+        paddd   xmm1, xmmword ptr [rsp+0A0H]
+        paddd   xmm2, xmmword ptr [rsp+60H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0B0H]
+        paddd   xmm1, xmmword ptr [rsp+50H]
+        paddd   xmm2, xmmword ptr [rsp+10H]
+        paddd   xmm3, xmmword ptr [rsp+80H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0F0H]
+        paddd   xmm1, xmmword ptr [rsp]
+        paddd   xmm2, xmmword ptr [rsp+90H]
+        paddd   xmm3, xmmword ptr [rsp+60H]
+        paddd   xmm0, xmm4
+        paddd   xmm1, xmm5
+        paddd   xmm2, xmm6
+        paddd   xmm3, xmm7
+        pxor    xmm12, xmm0
+        pxor    xmm13, xmm1
+        pxor    xmm14, xmm2
+        pxor    xmm15, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        pshufb  xmm15, xmm8
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm12
+        paddd   xmm9, xmm13
+        paddd   xmm10, xmm14
+        paddd   xmm11, xmm15
+        pxor    xmm4, xmm8
+        pxor    xmm5, xmm9
+        pxor    xmm6, xmm10
+        pxor    xmm7, xmm11
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0E0H]
+        paddd   xmm1, xmmword ptr [rsp+20H]
+        paddd   xmm2, xmmword ptr [rsp+30H]
+        paddd   xmm3, xmmword ptr [rsp+70H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT16]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        movdqa  xmmword ptr [rsp+100H], xmm8
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 12
+        pslld   xmm5, 20
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 12
+        pslld   xmm6, 20
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 12
+        pslld   xmm7, 20
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 12
+        pslld   xmm4, 20
+        por     xmm4, xmm8
+        paddd   xmm0, xmmword ptr [rsp+0A0H]
+        paddd   xmm1, xmmword ptr [rsp+0C0H]
+        paddd   xmm2, xmmword ptr [rsp+40H]
+        paddd   xmm3, xmmword ptr [rsp+0D0H]
+        paddd   xmm0, xmm5
+        paddd   xmm1, xmm6
+        paddd   xmm2, xmm7
+        paddd   xmm3, xmm4
+        pxor    xmm15, xmm0
+        pxor    xmm12, xmm1
+        pxor    xmm13, xmm2
+        pxor    xmm14, xmm3
+        movdqa  xmm8, xmmword ptr [ROT8]
+        pshufb  xmm15, xmm8
+        pshufb  xmm12, xmm8
+        pshufb  xmm13, xmm8
+        pshufb  xmm14, xmm8
+        paddd   xmm10, xmm15
+        paddd   xmm11, xmm12
+        movdqa  xmm8, xmmword ptr [rsp+100H]
+        paddd   xmm8, xmm13
+        paddd   xmm9, xmm14
+        pxor    xmm5, xmm10
+        pxor    xmm6, xmm11
+        pxor    xmm7, xmm8
+        pxor    xmm4, xmm9
+        pxor    xmm0, xmm8
+        pxor    xmm1, xmm9
+        pxor    xmm2, xmm10
+        pxor    xmm3, xmm11
+        movdqa  xmm8, xmm5
+        psrld   xmm8, 7
+        pslld   xmm5, 25
+        por     xmm5, xmm8
+        movdqa  xmm8, xmm6
+        psrld   xmm8, 7
+        pslld   xmm6, 25
+        por     xmm6, xmm8
+        movdqa  xmm8, xmm7
+        psrld   xmm8, 7
+        pslld   xmm7, 25
+        por     xmm7, xmm8
+        movdqa  xmm8, xmm4
+        psrld   xmm8, 7
+        pslld   xmm4, 25
+        por     xmm4, xmm8
+        pxor    xmm4, xmm12
+        pxor    xmm5, xmm13
+        pxor    xmm6, xmm14
+        pxor    xmm7, xmm15
+        mov     eax, r13d
+        jne     innerloop4
+        movdqa  xmm9, xmm0
+        punpckldq xmm0, xmm1
+        punpckhdq xmm9, xmm1
+        movdqa  xmm11, xmm2
+        punpckldq xmm2, xmm3
+        punpckhdq xmm11, xmm3
+        movdqa  xmm1, xmm0
+        punpcklqdq xmm0, xmm2
+        punpckhqdq xmm1, xmm2
+        movdqa  xmm3, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm3, xmm11
+        movdqu  xmmword ptr [rbx], xmm0
+        movdqu  xmmword ptr [rbx+20H], xmm1
+        movdqu  xmmword ptr [rbx+40H], xmm9
+        movdqu  xmmword ptr [rbx+60H], xmm3
+        movdqa  xmm9, xmm4
+        punpckldq xmm4, xmm5
+        punpckhdq xmm9, xmm5
+        movdqa  xmm11, xmm6
+        punpckldq xmm6, xmm7
+        punpckhdq xmm11, xmm7
+        movdqa  xmm5, xmm4
+        punpcklqdq xmm4, xmm6
+        punpckhqdq xmm5, xmm6
+        movdqa  xmm7, xmm9
+        punpcklqdq xmm9, xmm11
+        punpckhqdq xmm7, xmm11
+        movdqu  xmmword ptr [rbx+10H], xmm4
+        movdqu  xmmword ptr [rbx+30H], xmm5
+        movdqu  xmmword ptr [rbx+50H], xmm9
+        movdqu  xmmword ptr [rbx+70H], xmm7
+        movdqa  xmm1, xmmword ptr [rsp+110H]
+        movdqa  xmm0, xmm1
+        paddd   xmm1, xmmword ptr [rsp+150H]
+        movdqa  xmmword ptr [rsp+110H], xmm1
+        pxor    xmm0, xmmword ptr [CMP_MSB_MASK]
+        pxor    xmm1, xmmword ptr [CMP_MSB_MASK]
+        pcmpgtd xmm0, xmm1
+        movdqa  xmm1, xmmword ptr [rsp+120H]
+        psubd   xmm1, xmm0
+        movdqa  xmmword ptr [rsp+120H], xmm1
+        add     rbx, 128
+        add     rdi, 32
+        sub     rsi, 4
+        cmp     rsi, 4
+        jnc     outerloop4
+        test    rsi, rsi
+        jne     final3blocks
+unwind:
+        movdqa  xmm6, xmmword ptr [rsp+170H]
+        movdqa  xmm7, xmmword ptr [rsp+180H]
+        movdqa  xmm8, xmmword ptr [rsp+190H]
+        movdqa  xmm9, xmmword ptr [rsp+1A0H]
+        movdqa  xmm10, xmmword ptr [rsp+1B0H]
+        movdqa  xmm11, xmmword ptr [rsp+1C0H]
+        movdqa  xmm12, xmmword ptr [rsp+1D0H]
+        movdqa  xmm13, xmmword ptr [rsp+1E0H]
+        movdqa  xmm14, xmmword ptr [rsp+1F0H]
+        movdqa  xmm15, xmmword ptr [rsp+200H]
+        mov     rsp, rbp
+        pop     rbp
+        pop     rbx
+        pop     rdi
+        pop     rsi
+        pop     r12
+        pop     r13
+        pop     r14
+        pop     r15
+        ret
+ALIGN   16
+final3blocks:
+        test    esi, 2H
+        je      final1block
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movaps  xmm8, xmm0
+        movaps  xmm9, xmm1
+        movd    xmm13, dword ptr [rsp+110H]
+        pinsrd  xmm13, dword ptr [rsp+120H], 1
+        pinsrd  xmm13, dword ptr [BLAKE3_BLOCK_LEN], 2
+        movaps  xmmword ptr [rsp], xmm13
+        movd    xmm14, dword ptr [rsp+114H]
+        pinsrd  xmm14, dword ptr [rsp+124H], 1
+        pinsrd  xmm14, dword ptr [BLAKE3_BLOCK_LEN], 2
+        movaps  xmmword ptr [rsp+10H], xmm14
+        mov     r8, qword ptr [rdi]
+        mov     r9, qword ptr [rdi+8H]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+innerloop2:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        movaps  xmm10, xmm2
+        movups  xmm4, xmmword ptr [r8+rdx-40H]
+        movups  xmm5, xmmword ptr [r8+rdx-30H]
+        movaps  xmm3, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm3, xmm5, 221
+        movaps  xmm5, xmm3
+        movups  xmm6, xmmword ptr [r8+rdx-20H]
+        movups  xmm7, xmmword ptr [r8+rdx-10H]
+        movaps  xmm3, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm3, xmm7, 221
+        pshufd  xmm7, xmm3, 93H
+        movups  xmm12, xmmword ptr [r9+rdx-40H]
+        movups  xmm13, xmmword ptr [r9+rdx-30H]
+        movaps  xmm11, xmm12
+        shufps  xmm12, xmm13, 136
+        shufps  xmm11, xmm13, 221
+        movaps  xmm13, xmm11
+        movups  xmm14, xmmword ptr [r9+rdx-20H]
+        movups  xmm15, xmmword ptr [r9+rdx-10H]
+        movaps  xmm11, xmm14
+        shufps  xmm14, xmm15, 136
+        pshufd  xmm14, xmm14, 93H
+        shufps  xmm11, xmm15, 221
+        pshufd  xmm15, xmm11, 93H
+        movaps  xmm3, xmmword ptr [rsp]
+        movaps  xmm11, xmmword ptr [rsp+10H]
+        pinsrd  xmm3, eax, 3
+        pinsrd  xmm11, eax, 3
+        mov     al, 7
+roundloop2:
+        paddd   xmm0, xmm4
+        paddd   xmm8, xmm12
+        movaps  xmmword ptr [rsp+20H], xmm4
+        movaps  xmmword ptr [rsp+30H], xmm12
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movaps  xmm12, xmmword ptr [ROT16]
+        pshufb  xmm3, xmm12
+        pshufb  xmm11, xmm12
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm5
+        paddd   xmm8, xmm13
+        movaps  xmmword ptr [rsp+40H], xmm5
+        movaps  xmmword ptr [rsp+50H], xmm13
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        movaps  xmm13, xmmword ptr [ROT8]
+        pshufb  xmm3, xmm13
+        pshufb  xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm8, xmm8, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm11, xmm11, 4EH
+        pshufd  xmm2, xmm2, 39H
+        pshufd  xmm10, xmm10, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm8, xmm14
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshufb  xmm3, xmm12
+        pshufb  xmm11, xmm12
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 20
+        psrld   xmm4, 12
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 20
+        psrld   xmm4, 12
+        por     xmm9, xmm4
+        paddd   xmm0, xmm7
+        paddd   xmm8, xmm15
+        paddd   xmm0, xmm1
+        paddd   xmm8, xmm9
+        pxor    xmm3, xmm0
+        pxor    xmm11, xmm8
+        pshufb  xmm3, xmm13
+        pshufb  xmm11, xmm13
+        paddd   xmm2, xmm3
+        paddd   xmm10, xmm11
+        pxor    xmm1, xmm2
+        pxor    xmm9, xmm10
+        movdqa  xmm4, xmm1
+        pslld   xmm1, 25
+        psrld   xmm4, 7
+        por     xmm1, xmm4
+        movdqa  xmm4, xmm9
+        pslld   xmm9, 25
+        psrld   xmm4, 7
+        por     xmm9, xmm4
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm8, xmm8, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm11, xmm11, 4EH
+        pshufd  xmm2, xmm2, 93H
+        pshufd  xmm10, xmm10, 93H
+        dec     al
+        je      endroundloop2
+        movdqa  xmm12, xmmword ptr [rsp+20H]
+        movdqa  xmm5, xmmword ptr [rsp+40H]
+        pshufd  xmm13, xmm12, 0FH
+        shufps  xmm12, xmm5, 214
+        pshufd  xmm4, xmm12, 39H
+        movdqa  xmm12, xmm6
+        shufps  xmm12, xmm7, 250
+        pblendw xmm13, xmm12, 0CCH
+        movdqa  xmm12, xmm7
+        punpcklqdq xmm12, xmm5
+        pblendw xmm12, xmm6, 0C0H
+        pshufd  xmm12, xmm12, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmmword ptr [rsp+20H], xmm13
+        movdqa  xmmword ptr [rsp+40H], xmm12
+        movdqa  xmm5, xmmword ptr [rsp+30H]
+        movdqa  xmm13, xmmword ptr [rsp+50H]
+        pshufd  xmm6, xmm5, 0FH
+        shufps  xmm5, xmm13, 214
+        pshufd  xmm12, xmm5, 39H
+        movdqa  xmm5, xmm14
+        shufps  xmm5, xmm15, 250
+        pblendw xmm6, xmm5, 0CCH
+        movdqa  xmm5, xmm15
+        punpcklqdq xmm5, xmm13
+        pblendw xmm5, xmm14, 0C0H
+        pshufd  xmm5, xmm5, 78H
+        punpckhdq xmm13, xmm15
+        punpckldq xmm14, xmm13
+        pshufd  xmm15, xmm14, 1EH
+        movdqa  xmm13, xmm6
+        movdqa  xmm14, xmm5
+        movdqa  xmm5, xmmword ptr [rsp+20H]
+        movdqa  xmm6, xmmword ptr [rsp+40H]
+        jmp     roundloop2
+endroundloop2:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm8, xmm10
+        pxor    xmm9, xmm11
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop2
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+10H], xmm1
+        movups  xmmword ptr [rbx+20H], xmm8
+        movups  xmmword ptr [rbx+30H], xmm9
+        movdqa  xmm0, xmmword ptr [rsp+130H]
+        movdqa  xmm1, xmmword ptr [rsp+110H]
+        movdqa  xmm2, xmmword ptr [rsp+120H]
+        movdqu  xmm3, xmmword ptr [rsp+118H]
+        movdqu  xmm4, xmmword ptr [rsp+128H]
+        blendvps xmm1, xmm3, xmm0
+        blendvps xmm2, xmm4, xmm0
+        movdqa  xmmword ptr [rsp+110H], xmm1
+        movdqa  xmmword ptr [rsp+120H], xmm2
+        add     rdi, 16
+        add     rbx, 64
+        sub     rsi, 2
+final1block:
+        test    esi, 1H
+        je      unwind
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movd    xmm13, dword ptr [rsp+110H]
+        pinsrd  xmm13, dword ptr [rsp+120H], 1
+        pinsrd  xmm13, dword ptr [BLAKE3_BLOCK_LEN], 2
+        movaps  xmm14, xmmword ptr [ROT8]
+        movaps  xmm15, xmmword ptr [ROT16]
+        mov     r8, qword ptr [rdi]
+        movzx   eax, byte ptr [rbp+80H]
+        or      eax, r13d
+        xor     edx, edx
+innerloop1:
+        mov     r14d, eax
+        or      eax, r12d
+        add     rdx, 64
+        cmp     rdx, r15
+        cmovne  eax, r14d
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        movaps  xmm3, xmm13
+        pinsrd  xmm3, eax, 3
+        movups  xmm4, xmmword ptr [r8+rdx-40H]
+        movups  xmm5, xmmword ptr [r8+rdx-30H]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [r8+rdx-20H]
+        movups  xmm7, xmmword ptr [r8+rdx-10H]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 93H
+        mov     al, 7
+roundloop1:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 93H
+        dec     al
+        jz      endroundloop1
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0FH
+        pshufd  xmm4, xmm8, 39H
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0CCH
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0C0H
+        pshufd  xmm8, xmm8, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     roundloop1
+endroundloop1:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        mov     eax, r13d
+        cmp     rdx, r15
+        jne     innerloop1
+        movups  xmmword ptr [rbx], xmm0
+        movups  xmmword ptr [rbx+10H], xmm1
+        jmp     unwind
+_blake3_hash_many_sse41 ENDP
+blake3_hash_many_sse41 ENDP
+
+blake3_compress_in_place_sse41 PROC
+_blake3_compress_in_place_sse41 PROC
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+10H], xmm7
+        movdqa  xmmword ptr [rsp+20H], xmm8
+        movdqa  xmmword ptr [rsp+30H], xmm9
+        movdqa  xmmword ptr [rsp+40H], xmm11
+        movdqa  xmmword ptr [rsp+50H], xmm14
+        movdqa  xmmword ptr [rsp+60H], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        movzx   eax, byte ptr [rsp+0A0H]
+        movzx   r8d, r8b
+        shl     rax, 32
+        add     r8, rax
+        movd    xmm3, r9
+        movd    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+10H]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+20H]
+        movups  xmm7, xmmword ptr [rdx+30H]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 93H
+        movaps  xmm14, xmmword ptr [ROT8]
+        movaps  xmm15, xmmword ptr [ROT16]
+        mov     al, 7
+@@:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 93H
+        dec     al
+        jz      @F
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0FH
+        pshufd  xmm4, xmm8, 39H
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0CCH
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0C0H
+        pshufd  xmm8, xmm8, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     @B
+@@:
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        movups  xmmword ptr [rcx], xmm0
+        movups  xmmword ptr [rcx+10H], xmm1
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+10H]
+        movdqa  xmm8, xmmword ptr [rsp+20H]
+        movdqa  xmm9, xmmword ptr [rsp+30H]
+        movdqa  xmm11, xmmword ptr [rsp+40H]
+        movdqa  xmm14, xmmword ptr [rsp+50H]
+        movdqa  xmm15, xmmword ptr [rsp+60H]
+        add     rsp, 120
+        ret
+_blake3_compress_in_place_sse41 ENDP
+blake3_compress_in_place_sse41 ENDP
+
+ALIGN 16
+blake3_compress_xof_sse41 PROC
+_blake3_compress_xof_sse41 PROC
+        sub     rsp, 120
+        movdqa  xmmword ptr [rsp], xmm6
+        movdqa  xmmword ptr [rsp+10H], xmm7
+        movdqa  xmmword ptr [rsp+20H], xmm8
+        movdqa  xmmword ptr [rsp+30H], xmm9
+        movdqa  xmmword ptr [rsp+40H], xmm11
+        movdqa  xmmword ptr [rsp+50H], xmm14
+        movdqa  xmmword ptr [rsp+60H], xmm15
+        movups  xmm0, xmmword ptr [rcx]
+        movups  xmm1, xmmword ptr [rcx+10H]
+        movaps  xmm2, xmmword ptr [BLAKE3_IV]
+        movzx   eax, byte ptr [rsp+0A0H]
+        movzx   r8d, r8b
+        mov     r10, qword ptr [rsp+0A8H]
+        shl     rax, 32
+        add     r8, rax
+        movd    xmm3, r9
+        movd    xmm4, r8
+        punpcklqdq xmm3, xmm4
+        movups  xmm4, xmmword ptr [rdx]
+        movups  xmm5, xmmword ptr [rdx+10H]
+        movaps  xmm8, xmm4
+        shufps  xmm4, xmm5, 136
+        shufps  xmm8, xmm5, 221
+        movaps  xmm5, xmm8
+        movups  xmm6, xmmword ptr [rdx+20H]
+        movups  xmm7, xmmword ptr [rdx+30H]
+        movaps  xmm8, xmm6
+        shufps  xmm6, xmm7, 136
+        pshufd  xmm6, xmm6, 93H
+        shufps  xmm8, xmm7, 221
+        pshufd  xmm7, xmm8, 93H
+        movaps  xmm14, xmmword ptr [ROT8]
+        movaps  xmm15, xmmword ptr [ROT16]
+        mov     al, 7
+@@:
+        paddd   xmm0, xmm4
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm5
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 93H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 39H
+        paddd   xmm0, xmm6
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm15
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 20
+        psrld   xmm11, 12
+        por     xmm1, xmm11
+        paddd   xmm0, xmm7
+        paddd   xmm0, xmm1
+        pxor    xmm3, xmm0
+        pshufb  xmm3, xmm14
+        paddd   xmm2, xmm3
+        pxor    xmm1, xmm2
+        movdqa  xmm11, xmm1
+        pslld   xmm1, 25
+        psrld   xmm11, 7
+        por     xmm1, xmm11
+        pshufd  xmm0, xmm0, 39H
+        pshufd  xmm3, xmm3, 4EH
+        pshufd  xmm2, xmm2, 93H
+        dec     al
+        jz      @F
+        movdqa  xmm8, xmm4
+        shufps  xmm8, xmm5, 214
+        pshufd  xmm9, xmm4, 0FH
+        pshufd  xmm4, xmm8, 39H
+        movdqa  xmm8, xmm6
+        shufps  xmm8, xmm7, 250
+        pblendw xmm9, xmm8, 0CCH
+        movdqa  xmm8, xmm7
+        punpcklqdq xmm8, xmm5
+        pblendw xmm8, xmm6, 0C0H
+        pshufd  xmm8, xmm8, 78H
+        punpckhdq xmm5, xmm7
+        punpckldq xmm6, xmm5
+        pshufd  xmm7, xmm6, 1EH
+        movdqa  xmm5, xmm9
+        movdqa  xmm6, xmm8
+        jmp     @B
+@@:
+        movdqu  xmm4, xmmword ptr [rcx]
+        movdqu  xmm5, xmmword ptr [rcx+10H]
+        pxor    xmm0, xmm2
+        pxor    xmm1, xmm3
+        pxor    xmm2, xmm4
+        pxor    xmm3, xmm5
+        movups  xmmword ptr [r10], xmm0
+        movups  xmmword ptr [r10+10H], xmm1
+        movups  xmmword ptr [r10+20H], xmm2
+        movups  xmmword ptr [r10+30H], xmm3
+        movdqa  xmm6, xmmword ptr [rsp]
+        movdqa  xmm7, xmmword ptr [rsp+10H]
+        movdqa  xmm8, xmmword ptr [rsp+20H]
+        movdqa  xmm9, xmmword ptr [rsp+30H]
+        movdqa  xmm11, xmmword ptr [rsp+40H]
+        movdqa  xmm14, xmmword ptr [rsp+50H]
+        movdqa  xmm15, xmmword ptr [rsp+60H]
+        add     rsp, 120
+        ret
+_blake3_compress_xof_sse41 ENDP
+blake3_compress_xof_sse41 ENDP
+
+_TEXT ENDS
+
+
+_RDATA SEGMENT READONLY PAGE ALIAS(".rdata") 'CONST'
+ALIGN   64
+BLAKE3_IV:
+        dd 6A09E667H, 0BB67AE85H, 3C6EF372H, 0A54FF53AH
+
+ADD0:
+        dd 0, 1, 2, 3
+
+ADD1:
+        dd 4 dup (4)
+
+BLAKE3_IV_0:
+        dd 4 dup (6A09E667H)
+
+BLAKE3_IV_1:
+        dd 4 dup (0BB67AE85H)
+
+BLAKE3_IV_2:
+        dd 4 dup (3C6EF372H)
+
+BLAKE3_IV_3:
+        dd 4 dup (0A54FF53AH)
+
+BLAKE3_BLOCK_LEN:
+        dd 4 dup (64)
+
+ROT16:
+        db 2, 3, 0, 1, 6, 7, 4, 5, 10, 11, 8, 9, 14, 15, 12, 13
+
+ROT8:
+        db 1, 2, 3, 0, 5, 6, 7, 4, 9, 10, 11, 8, 13, 14, 15, 12
+
+CMP_MSB_MASK:
+        dd 8 dup(80000000H)
+
+_RDATA ENDS
+END
+
diff --git a/thirdparty/blake3/c/blake3_tbb.cpp b/thirdparty/blake3/c/blake3_tbb.cpp
new file mode 100644
index 000000000..c2bc2dbed
--- /dev/null
+++ b/thirdparty/blake3/c/blake3_tbb.cpp
@@ -0,0 +1,37 @@
+#include <cstddef>
+#include <cstdint>
+
+#include <oneapi/tbb/parallel_invoke.h>
+
+#include "blake3_impl.h"
+
+static_assert(TBB_USE_EXCEPTIONS == 0,
+              "This file should be compiled with C++ exceptions disabled.");
+
+extern "C" void blake3_compress_subtree_wide_join_tbb(
+    // shared params
+    const uint32_t key[8], uint8_t flags, bool use_tbb,
+    // left-hand side params
+    const uint8_t *l_input, size_t l_input_len, uint64_t l_chunk_counter,
+    uint8_t *l_cvs, size_t *l_n,
+    // right-hand side params
+    const uint8_t *r_input, size_t r_input_len, uint64_t r_chunk_counter,
+    uint8_t *r_cvs, size_t *r_n) noexcept {
+  if (!use_tbb) {
+    *l_n = blake3_compress_subtree_wide(l_input, l_input_len, key,
+                                        l_chunk_counter, flags, l_cvs, use_tbb);
+    *r_n = blake3_compress_subtree_wide(r_input, r_input_len, key,
+                                        r_chunk_counter, flags, r_cvs, use_tbb);
+    return;
+  }
+
+  oneapi::tbb::parallel_invoke(
+      [=]() {
+        *l_n = blake3_compress_subtree_wide(
+            l_input, l_input_len, key, l_chunk_counter, flags, l_cvs, use_tbb);
+      },
+      [=]() {
+        *r_n = blake3_compress_subtree_wide(
+            r_input, r_input_len, key, r_chunk_counter, flags, r_cvs, use_tbb);
+      });
+}
diff --git a/thirdparty/blake3/c/cmake/BLAKE3/ContinuousIntegration.cmake b/thirdparty/blake3/c/cmake/BLAKE3/ContinuousIntegration.cmake
new file mode 100644
index 000000000..57a1f1810
--- /dev/null
+++ b/thirdparty/blake3/c/cmake/BLAKE3/ContinuousIntegration.cmake
@@ -0,0 +1,235 @@
+cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
+
+if(BUILD_SHARED_LIBS)
+  message(FATAL_ERROR "BUILD_SHARED_LIBS is incompatible with BLAKE3_TESTING_CI")
+endif()
+
+include(CTest)
+
+# Declare a testing specific variant of the `blake3` library target.
+#
+# We use a separate library target in order to be able to perform compilation with various
+# combinations of features which are too noisy to specify in the main CMake config as options for
+# the normal `blake3` target.
+#
+# Initially this target has no properties but eventually we will populate them by copying all of the
+# relevant properties from the normal `blake3` target.
+add_library(blake3-testing
+  blake3.c
+  blake3_dispatch.c
+  blake3_portable.c
+)
+
+if(BLAKE3_USE_TBB AND TBB_FOUND)
+  target_sources(blake3-testing
+    PRIVATE
+      blake3_tbb.cpp)
+endif()
+
+if(BLAKE3_SIMD_TYPE STREQUAL "amd64-asm")
+  # Conditionally add amd64 asm files to `blake3-testing` sources
+  if(MSVC)
+    if(NOT BLAKE3_NO_AVX2)
+      list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_avx2_x86-64_windows_msvc.asm)
+    endif()
+    if(NOT BLAKE3_NO_AVX512)
+      list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_avx512_x86-64_windows_msvc.asm)
+    endif()
+    if(NOT BLAKE3_NO_SSE2)
+      list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_sse2_x86-64_windows_msvc.asm)
+    endif()
+    if(NOT BLAKE3_NO_SSE41)
+      list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_sse41_x86-64_windows_msvc.asm)
+    endif()
+  elseif(CMAKE_C_COMPILER_ID STREQUAL "GNU"
+        OR CMAKE_C_COMPILER_ID STREQUAL "Clang"
+        OR CMAKE_C_COMPILER_ID STREQUAL "AppleClang")
+    if (WIN32)
+      if(NOT BLAKE3_NO_AVX2)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_avx2_x86-64_windows_gnu.S)
+      endif()
+      if(NOT BLAKE3_NO_AVX512)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_avx512_x86-64_windows_gnu.S)
+      endif()
+      if(NOT BLAKE3_NO_SSE2)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_sse2_x86-64_windows_gnu.S)
+      endif()
+      if(NOT BLAKE3_NO_SSE41)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_sse41_x86-64_windows_gnu.S)
+      endif()
+    elseif(UNIX)
+      if(NOT BLAKE3_NO_AVX2)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_avx2_x86-64_unix.S)
+      endif()
+      if(NOT BLAKE3_NO_AVX512)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_avx512_x86-64_unix.S)
+      endif()
+      if(NOT BLAKE3_NO_SSE2)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_sse2_x86-64_unix.S)
+      endif()
+      if(NOT BLAKE3_NO_SSE41)
+        list(APPEND BLAKE3_TESTING_AMD64_ASM_SOURCES blake3_sse41_x86-64_unix.S)
+      endif()
+    endif()
+  endif()
+  target_sources(blake3-testing PRIVATE ${BLAKE3_AMD64_ASM_SOURCES})
+elseif(BLAKE3_SIMD_TYPE STREQUAL "x86-intrinsics")
+  # Conditionally add amd64 C files to `blake3-testing` sources
+  if (NOT DEFINED BLAKE3_CFLAGS_SSE2
+      OR NOT DEFINED BLAKE3_CFLAGS_SSE4.1
+      OR NOT DEFINED BLAKE3_CFLAGS_AVX2
+      OR NOT DEFINED BLAKE3_CFLAGS_AVX512)
+    message(WARNING "BLAKE3_SIMD_TYPE is set to 'x86-intrinsics' but no compiler flags are available for the target architecture.")
+  else()
+    set(BLAKE3_SIMD_X86_INTRINSICS ON)
+  endif()
+
+  if(NOT BLAKE3_NO_AVX2)
+    target_sources(blake3-testing PRIVATE blake3_avx2.c)
+    set_source_files_properties(blake3_avx2.c PROPERTIES COMPILE_FLAGS "${BLAKE3_CFLAGS_AVX2}")
+  endif()
+  if(NOT BLAKE3_NO_AVX512)
+    target_sources(blake3-testing PRIVATE blake3_avx512.c)
+    set_source_files_properties(blake3_avx512.c PROPERTIES COMPILE_FLAGS "${BLAKE3_CFLAGS_AVX512}")
+  endif()
+  if(NOT BLAKE3_NO_SSE2)
+    target_sources(blake3-testing PRIVATE blake3_sse2.c)
+    set_source_files_properties(blake3_sse2.c PROPERTIES COMPILE_FLAGS "${BLAKE3_CFLAGS_SSE2}")
+  endif()
+  if(NOT BLAKE3_NO_SSE41)
+    target_sources(blake3-testing PRIVATE blake3_sse41.c)
+    set_source_files_properties(blake3_sse41.c PROPERTIES COMPILE_FLAGS "${BLAKE3_CFLAGS_SSE4.1}")
+  endif()
+
+elseif(BLAKE3_SIMD_TYPE STREQUAL "neon-intrinsics")
+  # Conditionally add neon C files to `blake3-testing` sources
+
+  target_sources(blake3-testing PRIVATE
+    blake3_neon.c
+  )
+  target_compile_definitions(blake3-testing PRIVATE
+    BLAKE3_USE_NEON=1
+  )
+
+  if (DEFINED BLAKE3_CFLAGS_NEON)
+    set_source_files_properties(blake3_neon.c PROPERTIES COMPILE_FLAGS "${BLAKE3_CFLAGS_NEON}")
+  endif()
+
+elseif(BLAKE3_SIMD_TYPE STREQUAL "none")
+  # Disable neon if simd type is "none". We check for individual amd64 features further below.
+
+  target_compile_definitions(blake3-testing PRIVATE
+    BLAKE3_USE_NEON=0
+  )
+
+endif()
+
+if(BLAKE3_NO_AVX2)
+  target_compile_definitions(blake3-testing PRIVATE BLAKE3_NO_AVX2)
+endif()
+if(BLAKE3_NO_AVX512)
+  target_compile_definitions(blake3-testing PRIVATE BLAKE3_NO_AVX512)
+endif()
+if(BLAKE3_NO_SSE2)
+  target_compile_definitions(blake3-testing PRIVATE BLAKE3_NO_SSE2)
+endif()
+if(BLAKE3_NO_SSE41)
+  target_compile_definitions(blake3-testing PRIVATE BLAKE3_NO_SSE41)
+endif()
+
+target_compile_definitions(blake3-testing PUBLIC BLAKE3_TESTING)
+
+get_target_property(BLAKE3_COMPILE_DEFINITIONS blake3 COMPILE_DEFINITIONS)
+if(BLAKE3_COMPILE_DEFINITIONS)
+  target_compile_definitions(blake3-testing PUBLIC
+    ${BLAKE3_COMPILE_DEFINITIONS})
+endif()
+
+get_target_property(BLAKE3_COMPILE_OPTIONS blake3 COMPILE_OPTIONS)
+if(BLAKE3_COMPILE_OPTIONS)
+  target_compile_options(blake3-testing PRIVATE
+    ${BLAKE3_COMPILE_OPTIONS}
+    -O3
+    -Wall
+    -Wextra
+    -pedantic
+    -fstack-protector-strong
+    -D_FORTIFY_SOURCE=2
+    -fPIE
+    -fvisibility=hidden
+    -fsanitize=address,undefined
+  )
+endif()
+
+get_target_property(BLAKE3_INCLUDE_DIRECTORIES blake3 INCLUDE_DIRECTORIES)
+if(BLAKE3_INCLUDE_DIRECTORIES)
+  target_include_directories(blake3-testing PUBLIC
+    $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>
+    $<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>
+  )
+endif()
+
+get_target_property(BLAKE3_LINK_LIBRARIES blake3 LINK_LIBRARIES)
+if(BLAKE3_LINK_LIBRARIES)
+  target_link_libraries(blake3-testing PRIVATE ${BLAKE3_LINK_LIBRARIES})
+endif()
+
+get_target_property(BLAKE3_LINK_OPTIONS blake3 LINK_OPTIONS)
+if(BLAKE3_LINK_OPTIONS)
+  target_link_options(blake3-testing PRIVATE
+    ${BLAKE3_LINK_OPTIONS}
+    -fsanitize=address,undefined
+    -pie
+    -Wl,-z,relro,-z,now
+  )
+endif()
+
+# test asm target
+add_executable(blake3-asm-test
+  main.c
+)
+set_target_properties(blake3-asm-test PROPERTIES
+  OUTPUT_NAME blake3
+  RUNTIME_OUTPUT_DIRECTORY ${CMAKE_SOURCE_DIR})
+target_link_libraries(blake3-asm-test PRIVATE blake3-testing)
+target_compile_definitions(blake3-asm-test PRIVATE BLAKE3_TESTING)
+target_compile_options(blake3-asm-test PRIVATE
+  -O3
+  -Wall
+  -Wextra
+  -pedantic
+  -fstack-protector-strong
+  -D_FORTIFY_SOURCE=2
+  -fPIE
+  -fvisibility=hidden
+  -fsanitize=address,undefined
+)
+target_link_options(blake3-asm-test PRIVATE
+  -fsanitize=address,undefined
+  -pie
+  -Wl,-z,relro,-z,now
+)
+
+add_test(NAME blake3-testing
+  COMMAND "${CMAKE_CTEST_COMMAND}"
+    --verbose
+    --extra-verbose
+    --build-and-test "${CMAKE_SOURCE_DIR}" "${CMAKE_BINARY_DIR}"
+    --build-generator "${CMAKE_GENERATOR}"
+    --build-makeprogram "${CMAKE_MAKE_PROGRAM}"
+    --build-project libblake3
+    --build-target blake3-asm-test
+    --build-options
+      --fresh
+      "-DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBS}"
+      "-DBLAKE3_TESTING=${BLAKE3_TESTING}"
+      "-DBLAKE3_TESTING_CI=${BLAKE3_TESTING_CI}"
+      "-DBLAKE3_USE_TBB=${BLAKE3_USE_TBB}"
+      "-DBLAKE3_SIMD_TYPE=${BLAKE3_SIMD_TYPE}"
+      "-DBLAKE3_NO_SSE2=${BLAKE3_NO_SSE2}"
+      "-DBLAKE3_NO_SSE41=${BLAKE3_NO_SSE41}"
+      "-DBLAKE3_NO_AVX2=${BLAKE3_NO_AVX2}"
+      "-DBLAKE3_NO_AVX512=${BLAKE3_NO_AVX512}"
+    --test-command
+      "${CMAKE_SOURCE_DIR}/test.py"
+  )
diff --git a/thirdparty/blake3/c/cmake/BLAKE3/Examples.cmake b/thirdparty/blake3/c/cmake/BLAKE3/Examples.cmake
new file mode 100644
index 000000000..8f0edc578
--- /dev/null
+++ b/thirdparty/blake3/c/cmake/BLAKE3/Examples.cmake
@@ -0,0 +1,13 @@
+if(NOT WIN32)
+  add_executable(blake3-example
+    example.c)
+  target_link_libraries(blake3-example PRIVATE blake3)
+  install(TARGETS blake3-example)
+
+  if(BLAKE3_USE_TBB)
+    add_executable(blake3-example-tbb
+      example_tbb.c)
+    target_link_libraries(blake3-example-tbb PRIVATE blake3)
+    install(TARGETS blake3-example-tbb)
+  endif()
+endif()
diff --git a/thirdparty/blake3/c/cmake/BLAKE3/Testing.cmake b/thirdparty/blake3/c/cmake/BLAKE3/Testing.cmake
new file mode 100644
index 000000000..9beb16a69
--- /dev/null
+++ b/thirdparty/blake3/c/cmake/BLAKE3/Testing.cmake
@@ -0,0 +1,3 @@
+if(BLAKE3_TESTING_CI)
+  include(BLAKE3/ContinuousIntegration)
+endif()
diff --git a/thirdparty/blake3/c/example.c b/thirdparty/blake3/c/example.c
new file mode 100644
index 000000000..41a707ea1
--- /dev/null
+++ b/thirdparty/blake3/c/example.c
@@ -0,0 +1,36 @@
+#include "blake3.h"
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+
+int main(void) {
+  // Initialize the hasher.
+  blake3_hasher hasher;
+  blake3_hasher_init(&hasher);
+
+  // Read input bytes from stdin.
+  unsigned char buf[65536];
+  while (1) {
+    ssize_t n = read(STDIN_FILENO, buf, sizeof(buf));
+    if (n > 0) {
+      blake3_hasher_update(&hasher, buf, n);
+    } else if (n == 0) {
+      break; // end of file
+    } else {
+      fprintf(stderr, "read failed: %s\n", strerror(errno));
+      return 1;
+    }
+  }
+
+  // Finalize the hash. BLAKE3_OUT_LEN is the default output length, 32 bytes.
+  uint8_t output[BLAKE3_OUT_LEN];
+  blake3_hasher_finalize(&hasher, output, BLAKE3_OUT_LEN);
+
+  // Print the hash as hexadecimal.
+  for (size_t i = 0; i < BLAKE3_OUT_LEN; i++) {
+    printf("%02x", output[i]);
+  }
+  printf("\n");
+  return 0;
+}
diff --git a/thirdparty/blake3/c/example_tbb.c b/thirdparty/blake3/c/example_tbb.c
new file mode 100644
index 000000000..88c4348f2
--- /dev/null
+++ b/thirdparty/blake3/c/example_tbb.c
@@ -0,0 +1,57 @@
+#include "blake3.h"
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <string.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+int main(int argc, char **argv) {
+  // For each filepath argument, memory map it and hash it.
+  for (int i = 1; i < argc; i++) {
+    // Open and memory map the file.
+    int fd = open(argv[i], O_RDONLY);
+    if (fd == -1) {
+      fprintf(stderr, "open failed: %s\n", strerror(errno));
+      return 1;
+    }
+    struct stat statbuf;
+    if (fstat(fd, &statbuf) == -1) {
+      fprintf(stderr, "stat failed: %s\n", strerror(errno));
+      return 1;
+    }
+    void *mapped = mmap(NULL, statbuf.st_size, PROT_READ, MAP_PRIVATE, fd, 0);
+    if (mapped == MAP_FAILED) {
+      fprintf(stderr, "mmap failed: %s\n", strerror(errno));
+      return 1;
+    }
+
+    // Initialize the hasher.
+    blake3_hasher hasher;
+    blake3_hasher_init(&hasher);
+
+    // Hash the mapped file using multiple threads.
+    blake3_hasher_update_tbb(&hasher, mapped, statbuf.st_size);
+
+    // Unmap and close the file.
+    if (munmap(mapped, statbuf.st_size) == -1) {
+      fprintf(stderr, "munmap failed: %s\n", strerror(errno));
+      return 1;
+    }
+    if (close(fd) == -1) {
+      fprintf(stderr, "close failed: %s\n", strerror(errno));
+      return 1;
+    }
+
+    // Finalize the hash. BLAKE3_OUT_LEN is the default output length, 32 bytes.
+    uint8_t output[BLAKE3_OUT_LEN];
+    blake3_hasher_finalize(&hasher, output, BLAKE3_OUT_LEN);
+
+    // Print the hash as hexadecimal.
+    for (size_t i = 0; i < BLAKE3_OUT_LEN; i++) {
+      printf("%02x", output[i]);
+    }
+    printf("\n");
+  }
+}
diff --git a/thirdparty/blake3/c/libblake3.pc.in b/thirdparty/blake3/c/libblake3.pc.in
new file mode 100644
index 000000000..460dc90b1
--- /dev/null
+++ b/thirdparty/blake3/c/libblake3.pc.in
@@ -0,0 +1,12 @@
+prefix="@CMAKE_INSTALL_PREFIX@"
+exec_prefix="${prefix}"
+libdir="@PKG_CONFIG_INSTALL_LIBDIR@"
+includedir="@PKG_CONFIG_INSTALL_INCLUDEDIR@"
+
+Name: @PROJECT_NAME@
+Description: @PROJECT_DESCRIPTION@
+Version: @PROJECT_VERSION@
+
+Requires: @PKG_CONFIG_REQUIRES@
+Libs: -L"${libdir}" -lblake3 @PKG_CONFIG_LIBS@
+Cflags: -I"${includedir}" @PKG_CONFIG_CFLAGS@
diff --git a/thirdparty/blake3/c/main.c b/thirdparty/blake3/c/main.c
new file mode 100644
index 000000000..77cab58f8
--- /dev/null
+++ b/thirdparty/blake3/c/main.c
@@ -0,0 +1,166 @@
+/*
+ * This main file is intended for testing via `make test`. It does not build in
+ * other settings. See README.md in this directory for examples of how to build
+ * C code.
+ */
+
+#include <assert.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "blake3.h"
+#include "blake3_impl.h"
+
+#define HASH_MODE 0
+#define KEYED_HASH_MODE 1
+#define DERIVE_KEY_MODE 2
+
+static void hex_char_value(uint8_t c, uint8_t *value, bool *valid) {
+  if ('0' <= c && c <= '9') {
+    *value = c - '0';
+    *valid = true;
+  } else if ('a' <= c && c <= 'f') {
+    *value = 10 + c - 'a';
+    *valid = true;
+  } else {
+    *valid = false;
+  }
+}
+
+static int parse_key(char *hex_key, uint8_t out[BLAKE3_KEY_LEN]) {
+  size_t hex_len = strlen(hex_key);
+  if (hex_len != 64) {
+    fprintf(stderr, "Expected a 64-char hexadecimal key, got %zu chars.\n",
+            hex_len);
+    return 1;
+  }
+  for (size_t i = 0; i < 64; i++) {
+    uint8_t value;
+    bool valid;
+    hex_char_value(hex_key[i], &value, &valid);
+    if (!valid) {
+      fprintf(stderr, "Invalid hex char.\n");
+      return 1;
+    }
+    if (i % 2 == 0) {
+      out[i / 2] = 0;
+      value <<= 4;
+    }
+    out[i / 2] += value;
+  }
+  return 0;
+}
+
+/* A little repetition here */
+enum cpu_feature {
+  SSE2 = 1 << 0,
+  SSSE3 = 1 << 1,
+  SSE41 = 1 << 2,
+  AVX = 1 << 3,
+  AVX2 = 1 << 4,
+  AVX512F = 1 << 5,
+  AVX512VL = 1 << 6,
+  /* ... */
+  UNDEFINED = 1 << 30
+};
+
+extern enum cpu_feature g_cpu_features;
+enum cpu_feature get_cpu_features(void);
+
+int main(int argc, char **argv) {
+  size_t out_len = BLAKE3_OUT_LEN;
+  uint8_t key[BLAKE3_KEY_LEN];
+  char *context = "";
+  uint8_t mode = HASH_MODE;
+  while (argc > 1) {
+    if (argc <= 2) {
+      fprintf(stderr, "Odd number of arguments.\n");
+      return 1;
+    }
+    if (strcmp("--length", argv[1]) == 0) {
+      char *endptr = NULL;
+      errno = 0;
+      unsigned long long out_len_ll = strtoull(argv[2], &endptr, 10);
+      if (errno != 0 || out_len_ll > SIZE_MAX || endptr == argv[2] ||
+          *endptr != 0) {
+        fprintf(stderr, "Bad length argument.\n");
+        return 1;
+      }
+      out_len = (size_t)out_len_ll;
+    } else if (strcmp("--keyed", argv[1]) == 0) {
+      mode = KEYED_HASH_MODE;
+      int ret = parse_key(argv[2], key);
+      if (ret != 0) {
+        return ret;
+      }
+    } else if (strcmp("--derive-key", argv[1]) == 0) {
+      mode = DERIVE_KEY_MODE;
+      context = argv[2];
+    } else {
+      fprintf(stderr, "Unknown flag.\n");
+      return 1;
+    }
+    argc -= 2;
+    argv += 2;
+  }
+
+  /*
+   * We're going to hash the input multiple times, so we need to buffer it all.
+   * This is just for test cases, so go ahead and assume that the input is less
+   * than 1 MiB.
+   */
+  size_t buf_capacity = 1 << 20;
+  uint8_t *buf = malloc(buf_capacity);
+  assert(buf != NULL);
+  size_t buf_len = 0;
+  while (1) {
+    size_t n = fread(&buf[buf_len], 1, buf_capacity - buf_len, stdin);
+    if (n == 0) {
+      break;
+    }
+    buf_len += n;
+    assert(buf_len < buf_capacity);
+  }
+
+  const int mask = get_cpu_features();
+  int feature = 0;
+  do {
+    fprintf(stderr, "Testing 0x%08X\n", feature);
+    g_cpu_features = feature;
+    blake3_hasher hasher;
+    switch (mode) {
+    case HASH_MODE:
+      blake3_hasher_init(&hasher);
+      break;
+    case KEYED_HASH_MODE:
+      blake3_hasher_init_keyed(&hasher, key);
+      break;
+    case DERIVE_KEY_MODE:
+      blake3_hasher_init_derive_key(&hasher, context);
+      break;
+    default:
+      abort();
+    }
+
+    blake3_hasher_update(&hasher, buf, buf_len);
+
+    /* TODO: An incremental output reader API to avoid this allocation. */
+    uint8_t *out = malloc(out_len);
+    if (out_len > 0 && out == NULL) {
+      fprintf(stderr, "malloc() failed.\n");
+      return 1;
+    }
+    blake3_hasher_finalize(&hasher, out, out_len);
+    for (size_t i = 0; i < out_len; i++) {
+      printf("%02x", out[i]);
+    }
+    printf("\n");
+    free(out);
+    feature = (feature - mask) & mask;
+  } while (feature != 0);
+  free(buf);
+  return 0;
+}
diff --git a/thirdparty/blake3/c/test.py b/thirdparty/blake3/c/test.py
new file mode 100755
index 000000000..98b1c3dff
--- /dev/null
+++ b/thirdparty/blake3/c/test.py
@@ -0,0 +1,97 @@
+#! /usr/bin/env python3
+
+from binascii import hexlify
+import json
+from os import path
+import subprocess
+
+HERE = path.dirname(__file__)
+TEST_VECTORS_PATH = path.join(HERE, "..", "test_vectors", "test_vectors.json")
+TEST_VECTORS = json.load(open(TEST_VECTORS_PATH))
+
+
+def run_blake3(args, input):
+    output = subprocess.run([path.join(HERE, "blake3")] + args,
+                            input=input,
+                            stdout=subprocess.PIPE,
+                            check=True)
+    return output.stdout.decode().strip()
+
+
+# Fill the input with a repeating byte pattern. We use a cycle length of 251,
+# because that's the largest prime number less than 256. This makes it unlikely
+# to swapping any two adjacent input blocks or chunks will give the same
+# answer.
+def make_test_input(length):
+    i = 0
+    buf = bytearray()
+    while len(buf) < length:
+        buf.append(i)
+        i = (i + 1) % 251
+    return buf
+
+
+def main():
+    for case in TEST_VECTORS["cases"]:
+        input_len = case["input_len"]
+        input = make_test_input(input_len)
+        hex_key = hexlify(TEST_VECTORS["key"].encode())
+        context_string = TEST_VECTORS["context_string"]
+        expected_hash_xof = case["hash"]
+        expected_hash = expected_hash_xof[:64]
+        expected_keyed_hash_xof = case["keyed_hash"]
+        expected_keyed_hash = expected_keyed_hash_xof[:64]
+        expected_derive_key_xof = case["derive_key"]
+        expected_derive_key = expected_derive_key_xof[:64]
+
+        # Test the default hash.
+        test_hash = run_blake3([], input)
+        for line in test_hash.splitlines():
+            assert expected_hash == line, \
+                "hash({}): {} != {}".format(input_len, expected_hash, line)
+
+        # Test the extended hash.
+        xof_len = len(expected_hash_xof) // 2
+        test_hash_xof = run_blake3(["--length", str(xof_len)], input)
+        for line in test_hash_xof.splitlines():
+            assert expected_hash_xof == line, \
+                "hash_xof({}): {} != {}".format(
+                    input_len, expected_hash_xof, line)
+
+        # Test the default keyed hash.
+        test_keyed_hash = run_blake3(["--keyed", hex_key], input)
+        for line in test_keyed_hash.splitlines():
+            assert expected_keyed_hash == line, \
+                "keyed_hash({}): {} != {}".format(
+                    input_len, expected_keyed_hash, line)
+
+        # Test the extended keyed hash.
+        xof_len = len(expected_keyed_hash_xof) // 2
+        test_keyed_hash_xof = run_blake3(
+            ["--keyed", hex_key, "--length",
+             str(xof_len)], input)
+        for line in test_keyed_hash_xof.splitlines():
+            assert expected_keyed_hash_xof == line, \
+                "keyed_hash_xof({}): {} != {}".format(
+                    input_len, expected_keyed_hash_xof, line)
+
+        # Test the default derive key.
+        test_derive_key = run_blake3(["--derive-key", context_string], input)
+        for line in test_derive_key.splitlines():
+            assert expected_derive_key == line, \
+                "derive_key({}): {} != {}".format(
+                    input_len, expected_derive_key, line)
+
+        # Test the extended derive key.
+        xof_len = len(expected_derive_key_xof) // 2
+        test_derive_key_xof = run_blake3(
+            ["--derive-key", context_string, "--length",
+             str(xof_len)], input)
+        for line in test_derive_key_xof.splitlines():
+            assert expected_derive_key_xof == line, \
+                "derive_key_xof({}): {} != {}".format(
+                    input_len, expected_derive_key_xof, line)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/thirdparty/blake3/media/B3.svg b/thirdparty/blake3/media/B3.svg
new file mode 100644
index 000000000..a50da0ce9
--- /dev/null
+++ b/thirdparty/blake3/media/B3.svg
@@ -0,0 +1,70 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Created with Inkscape (http://www.inkscape.org/) -->
+
+<svg
+   xmlns:dc="http://purl.org/dc/elements/1.1/"
+   xmlns:cc="http://creativecommons.org/ns#"
+   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+   xmlns:svg="http://www.w3.org/2000/svg"
+   xmlns="http://www.w3.org/2000/svg"
+   xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+   xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+   width="13.356165mm"
+   height="7.1437497mm"
+   viewBox="0 0 13.356165 7.1437497"
+   version="1.1"
+   id="svg8"
+   sodipodi:docname="B3.svg"
+   inkscape:version="0.92.4 5da689c313, 2019-01-14">
+  <defs
+     id="defs2" />
+  <sodipodi:namedview
+     id="base"
+     pagecolor="#ffffff"
+     bordercolor="#666666"
+     borderopacity="1.0"
+     inkscape:pageopacity="0.0"
+     inkscape:pageshadow="2"
+     inkscape:zoom="4"
+     inkscape:cx="72.73328"
+     inkscape:cy="-34.835127"
+     inkscape:document-units="mm"
+     inkscape:current-layer="layer1"
+     showgrid="false"
+     inkscape:window-width="1920"
+     inkscape:window-height="1016"
+     inkscape:window-x="0"
+     inkscape:window-y="27"
+     inkscape:window-maximized="1" />
+  <metadata
+     id="metadata5">
+    <rdf:RDF>
+      <cc:Work
+         rdf:about="">
+        <dc:format>image/svg+xml</dc:format>
+        <dc:type
+           rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+        <dc:title />
+      </cc:Work>
+    </rdf:RDF>
+  </metadata>
+  <g
+     inkscape:label="Layer 1"
+     inkscape:groupmode="layer"
+     id="layer1"
+     transform="translate(-24.441005,-113.52518)">
+    <g
+       aria-label="B3"
+       style="font-style:normal;font-weight:normal;font-size:10.58333302px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;fill:#000000;fill-opacity:1;stroke:none;stroke-width:0.26458332"
+       id="text868">
+      <path
+         d="m 28.176921,113.52518 q 0.635,0 1.0795,0.14817 0.455084,0.13758 0.740834,0.40216 0.296333,0.254 0.433916,0.61384 0.137584,0.35983 0.137584,0.79375 0,0.62441 -0.264584,1.00541 -0.254,0.381 -0.762,0.58209 0.508,0.21166 0.783167,0.61383 0.275167,0.39158 0.275167,1.016 0,0.43392 -0.137584,0.79375 -0.137583,0.35983 -0.433916,0.62442 -0.28575,0.254 -0.740834,0.40216 -0.4445,0.14817 -1.0795,0.14817 h -3.174999 q -0.592667,0 -0.592667,-0.58208 v -5.97959 q 0,-0.58208 0.592667,-0.58208 z m -2.508249,5.78908 q 0,0.11642 0.137583,0.11642 h 2.434166 q 0.5715,0 0.836084,-0.24342 0.264583,-0.24341 0.264583,-0.68791 0,-0.92075 -1.100667,-0.92075 h -2.571749 z m 0,-2.77283 h 2.539999 q 1.100667,0 1.100667,-0.85725 0,-0.42333 -0.264583,-0.67733 -0.254,-0.254 -0.8255,-0.254 h -2.413 q -0.137583,0 -0.137583,0.127 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path814" />
+      <path
+         d="m 35.38417,113.52518 q 0.635,0 1.0795,0.14817 0.455083,0.13758 0.740833,0.40216 0.296333,0.254 0.433917,0.60325 0.137583,0.34925 0.137583,0.762 0,0.635 -0.264583,1.03717 -0.254,0.39158 -0.751417,0.60325 0.508,0.21167 0.772583,0.62442 0.264584,0.40216 0.264584,1.04775 0,0.40216 -0.137584,0.75141 -0.137583,0.34925 -0.423333,0.61384 -0.28575,0.254 -0.740833,0.40216 -0.4445,0.14817 -1.0795,0.14817 h -3.608917 v -1.24883 h 3.608917 q 0.550333,0 0.814917,-0.23284 0.264583,-0.24341 0.264583,-0.67733 0,-0.85725 -1.090083,-0.85725 h -2.201334 v -1.13242 h 2.169584 q 0.550333,0 0.814916,-0.20108 0.275167,-0.21167 0.275167,-0.65617 0,-0.40216 -0.254,-0.64558 -0.254,-0.24342 -0.8255,-0.24342 h -3.566583 v -1.24883 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path816" />
+    </g>
+  </g>
+</svg>
diff --git a/thirdparty/blake3/media/BLAKE3.svg b/thirdparty/blake3/media/BLAKE3.svg
new file mode 100644
index 000000000..2d50c2d3b
--- /dev/null
+++ b/thirdparty/blake3/media/BLAKE3.svg
@@ -0,0 +1,85 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Created with Inkscape (http://www.inkscape.org/) -->
+
+<svg
+   xmlns:dc="http://purl.org/dc/elements/1.1/"
+   xmlns:cc="http://creativecommons.org/ns#"
+   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+   xmlns:svg="http://www.w3.org/2000/svg"
+   xmlns="http://www.w3.org/2000/svg"
+   xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+   xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+   width="43.063534mm"
+   height="7.2707496mm"
+   viewBox="0 0 43.063534 7.2707496"
+   version="1.1"
+   id="svg8"
+   sodipodi:docname="BLAKE3.svg"
+   inkscape:version="0.92.4 5da689c313, 2019-01-14">
+  <defs
+     id="defs2" />
+  <sodipodi:namedview
+     id="base"
+     pagecolor="#ffffff"
+     bordercolor="#666666"
+     borderopacity="1.0"
+     inkscape:pageopacity="0.0"
+     inkscape:pageshadow="2"
+     inkscape:zoom="4"
+     inkscape:cx="72.73328"
+     inkscape:cy="-34.835127"
+     inkscape:document-units="mm"
+     inkscape:current-layer="layer1"
+     showgrid="false"
+     inkscape:window-width="1920"
+     inkscape:window-height="1016"
+     inkscape:window-x="0"
+     inkscape:window-y="27"
+     inkscape:window-maximized="1" />
+  <metadata
+     id="metadata5">
+    <rdf:RDF>
+      <cc:Work
+         rdf:about="">
+        <dc:format>image/svg+xml</dc:format>
+        <dc:type
+           rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+      </cc:Work>
+    </rdf:RDF>
+  </metadata>
+  <g
+     inkscape:label="Layer 1"
+     inkscape:groupmode="layer"
+     id="layer1"
+     transform="translate(-24.441005,-113.39818)">
+    <g
+       aria-label="BLAKE3"
+       style="font-style:normal;font-weight:normal;font-size:10.58333302px;line-height:1.25;font-family:sans-serif;letter-spacing:0px;word-spacing:0px;fill:#000000;fill-opacity:1;stroke:none;stroke-width:0.26458332"
+       id="text868">
+      <path
+         d="m 28.176921,113.52518 q 0.635,0 1.0795,0.14817 0.455084,0.13758 0.740834,0.40216 0.296333,0.254 0.433916,0.61384 0.137584,0.35983 0.137584,0.79375 0,0.62441 -0.264584,1.00541 -0.254,0.381 -0.762,0.58209 0.508,0.21166 0.783167,0.61383 0.275167,0.39158 0.275167,1.016 0,0.43392 -0.137584,0.79375 -0.137583,0.35983 -0.433916,0.62442 -0.28575,0.254 -0.740834,0.40216 -0.4445,0.14817 -1.0795,0.14817 h -3.174999 q -0.592667,0 -0.592667,-0.58208 v -5.97959 q 0,-0.58208 0.592667,-0.58208 z m -2.508249,5.78908 q 0,0.11642 0.137583,0.11642 h 2.434166 q 0.5715,0 0.836084,-0.24342 0.264583,-0.24341 0.264583,-0.68791 0,-0.92075 -1.100667,-0.92075 h -2.571749 z m 0,-2.77283 h 2.539999 q 1.100667,0 1.100667,-0.85725 0,-0.42333 -0.264583,-0.67733 -0.254,-0.254 -0.8255,-0.254 h -2.413 q -0.137583,0 -0.137583,0.127 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path814" />
+      <path
+         d="m 33.22517,113.52518 v 4.66725 q 0,0.254 0.0635,0.48683 0.07408,0.22225 0.243417,0.39159 0.169333,0.15875 0.4445,0.254 0.28575,0.0953 0.709083,0.0953 h 2.772833 v 1.24883 h -2.846916 q -0.709084,0 -1.217084,-0.17992 -0.497416,-0.1905 -0.814916,-0.51858 -0.3175,-0.32808 -0.465667,-0.77258 -0.137583,-0.45509 -0.137583,-0.99484 v -4.67783 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path816" />
+      <path
+         d="M 39.342334,120.66893 H 37.9665 l 2.50825,-6.35 q 0.201084,-0.508 0.560917,-0.70908 0.370417,-0.21167 0.941917,-0.21167 0.560916,0 0.92075,0.21167 0.370416,0.20108 0.560916,0.70908 l 2.413,6.35 h -1.386416 l -2.169584,-5.74675 q -0.09525,-0.24342 -0.34925,-0.24342 -0.254,0 -0.359833,0.24342 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path818" />
+      <path
+         d="m 48.179401,113.52518 v 3.02683 h 0.687917 q 0.455083,0 0.740833,-0.0212 0.296333,-0.0318 0.486833,-0.127 0.1905,-0.0953 0.3175,-0.26459 0.137584,-0.17991 0.28575,-0.47625 l 1.090084,-2.13783 h 1.344083 l -1.121833,2.2225 q -0.243417,0.47625 -0.518584,0.79375 -0.275166,0.3175 -0.719666,0.508 0.254,0.0635 0.4445,0.17992 0.1905,0.10583 0.34925,0.27516 0.169333,0.15875 0.3175,0.39159 0.148166,0.22225 0.306916,0.52916 l 1.153584,2.24367 h -1.397 l -1.090084,-2.11667 q -0.148166,-0.28575 -0.28575,-0.45508 -0.137583,-0.16933 -0.34925,-0.26458 -0.211666,-0.0952 -0.529166,-0.11642 -0.3175,-0.0317 -0.8255,-0.0317 h -0.687917 v 2.9845 h -1.248833 v -7.14375 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path820" />
+      <path
+         d="m 60.127965,113.52518 v 1.24883 h -3.577166 q -0.5715,0 -0.8255,0.24342 -0.254,0.24342 -0.254,0.65617 0,0.84666 1.090083,0.84666 h 3.513667 v 1.13242 h -3.545417 q -1.090083,0 -1.090083,0.86783 0,0.42334 0.264583,0.66675 0.264583,0.23284 0.814917,0.23284 h 3.6195 v 1.24883 h -3.6195 q -0.635,0 -1.090083,-0.14817 -0.4445,-0.14816 -0.740834,-0.40216 -0.28575,-0.26459 -0.423333,-0.62442 -0.127,-0.35983 -0.127,-0.77258 0,-0.61384 0.264583,-1.016 0.264584,-0.41275 0.762,-0.62442 -1.005416,-0.41275 -1.005416,-1.60867 0,-0.42333 0.137583,-0.78316 0.137583,-0.35984 0.423333,-0.61384 0.296334,-0.26458 0.740834,-0.40216 0.455083,-0.14817 1.090083,-0.14817 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path822" />
+      <path
+         d="m 65.091539,113.52518 q 0.635,0 1.0795,0.14817 0.455083,0.13758 0.740833,0.40216 0.296333,0.254 0.433917,0.60325 0.137583,0.34925 0.137583,0.762 0,0.635 -0.264583,1.03717 -0.254,0.39158 -0.751417,0.60325 0.508,0.21167 0.772583,0.62442 0.264584,0.40216 0.264584,1.04775 0,0.40216 -0.137584,0.75141 -0.137583,0.34925 -0.423333,0.61384 -0.28575,0.254 -0.740833,0.40216 -0.4445,0.14817 -1.0795,0.14817 h -3.608917 v -1.24883 h 3.608917 q 0.550333,0 0.814916,-0.23284 0.264584,-0.24341 0.264584,-0.67733 0,-0.85725 -1.090084,-0.85725 h -2.201333 v -1.13242 h 2.169583 q 0.550334,0 0.814917,-0.20108 0.275167,-0.21167 0.275167,-0.65617 0,-0.40216 -0.254,-0.64558 -0.254,-0.24342 -0.8255,-0.24342 h -3.566583 v -1.24883 z"
+         style="font-style:normal;font-variant:normal;font-weight:normal;font-stretch:normal;font-family:Nasalization;-inkscape-font-specification:Nasalization;stroke-width:0.26458332"
+         id="path824" />
+    </g>
+  </g>
+</svg>
diff --git a/thirdparty/blake3/media/speed.svg b/thirdparty/blake3/media/speed.svg
new file mode 100644
index 000000000..7bd65ca3c
--- /dev/null
+++ b/thirdparty/blake3/media/speed.svg
@@ -0,0 +1,1474 @@
+<?xml version="1.0" encoding="utf-8" standalone="no"?>
+<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN"
+  "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
+<!-- Created with matplotlib (https://matplotlib.org/) -->
+<svg height="331.389812pt" version="1.1" viewBox="0 0 449.761406 331.389812" width="449.761406pt" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
+ <defs>
+  <style type="text/css">
+*{stroke-linecap:butt;stroke-linejoin:round;}
+  </style>
+ </defs>
+ <g id="figure_1">
+  <g id="patch_1">
+   <path d="M 0 331.389812 
+L 449.761406 331.389812 
+L 449.761406 0 
+L 0 0 
+z
+" style="fill:#ffffff;"/>
+  </g>
+  <g id="axes_1">
+   <g id="patch_2">
+    <path d="M 71.443906 288.430125 
+L 428.563906 288.430125 
+L 428.563906 22.318125 
+L 71.443906 22.318125 
+z
+" style="fill:#ffffff;"/>
+   </g>
+   <g id="matplotlib.axis_1">
+    <g id="xtick_1">
+     <g id="line2d_1">
+      <defs>
+       <path d="M 0 0 
+L 0 6 
+" id="me95d5351a6" style="stroke:#262626;stroke-width:1.25;"/>
+      </defs>
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_1">
+      <!-- 0 -->
+      <defs>
+       <path d="M 31.78125 66.40625 
+Q 24.171875 66.40625 20.328125 58.90625 
+Q 16.5 51.421875 16.5 36.375 
+Q 16.5 21.390625 20.328125 13.890625 
+Q 24.171875 6.390625 31.78125 6.390625 
+Q 39.453125 6.390625 43.28125 13.890625 
+Q 47.125 21.390625 47.125 36.375 
+Q 47.125 51.421875 43.28125 58.90625 
+Q 39.453125 66.40625 31.78125 66.40625 
+z
+M 31.78125 74.21875 
+Q 44.046875 74.21875 50.515625 64.515625 
+Q 56.984375 54.828125 56.984375 36.375 
+Q 56.984375 17.96875 50.515625 8.265625 
+Q 44.046875 -1.421875 31.78125 -1.421875 
+Q 19.53125 -1.421875 13.0625 8.265625 
+Q 6.59375 17.96875 6.59375 36.375 
+Q 6.59375 54.828125 13.0625 64.515625 
+Q 19.53125 74.21875 31.78125 74.21875 
+z
+" id="DejaVuSans-48"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(67.944531 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_2">
+     <g id="line2d_2">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="116.083906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_2">
+      <!-- 1000 -->
+      <defs>
+       <path d="M 12.40625 8.296875 
+L 28.515625 8.296875 
+L 28.515625 63.921875 
+L 10.984375 60.40625 
+L 10.984375 69.390625 
+L 28.421875 72.90625 
+L 38.28125 72.90625 
+L 38.28125 8.296875 
+L 54.390625 8.296875 
+L 54.390625 0 
+L 12.40625 0 
+z
+" id="DejaVuSans-49"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(102.086406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-49"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_3">
+     <g id="line2d_3">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="160.723906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_3">
+      <!-- 2000 -->
+      <defs>
+       <path d="M 19.1875 8.296875 
+L 53.609375 8.296875 
+L 53.609375 0 
+L 7.328125 0 
+L 7.328125 8.296875 
+Q 12.9375 14.109375 22.625 23.890625 
+Q 32.328125 33.6875 34.8125 36.53125 
+Q 39.546875 41.84375 41.421875 45.53125 
+Q 43.3125 49.21875 43.3125 52.78125 
+Q 43.3125 58.59375 39.234375 62.25 
+Q 35.15625 65.921875 28.609375 65.921875 
+Q 23.96875 65.921875 18.8125 64.3125 
+Q 13.671875 62.703125 7.8125 59.421875 
+L 7.8125 69.390625 
+Q 13.765625 71.78125 18.9375 73 
+Q 24.125 74.21875 28.421875 74.21875 
+Q 39.75 74.21875 46.484375 68.546875 
+Q 53.21875 62.890625 53.21875 53.421875 
+Q 53.21875 48.921875 51.53125 44.890625 
+Q 49.859375 40.875 45.40625 35.40625 
+Q 44.1875 33.984375 37.640625 27.21875 
+Q 31.109375 20.453125 19.1875 8.296875 
+z
+" id="DejaVuSans-50"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(146.726406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-50"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_4">
+     <g id="line2d_4">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="205.363906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_4">
+      <!-- 3000 -->
+      <defs>
+       <path d="M 40.578125 39.3125 
+Q 47.65625 37.796875 51.625 33 
+Q 55.609375 28.21875 55.609375 21.1875 
+Q 55.609375 10.40625 48.1875 4.484375 
+Q 40.765625 -1.421875 27.09375 -1.421875 
+Q 22.515625 -1.421875 17.65625 -0.515625 
+Q 12.796875 0.390625 7.625 2.203125 
+L 7.625 11.71875 
+Q 11.71875 9.328125 16.59375 8.109375 
+Q 21.484375 6.890625 26.8125 6.890625 
+Q 36.078125 6.890625 40.9375 10.546875 
+Q 45.796875 14.203125 45.796875 21.1875 
+Q 45.796875 27.640625 41.28125 31.265625 
+Q 36.765625 34.90625 28.71875 34.90625 
+L 20.21875 34.90625 
+L 20.21875 43.015625 
+L 29.109375 43.015625 
+Q 36.375 43.015625 40.234375 45.921875 
+Q 44.09375 48.828125 44.09375 54.296875 
+Q 44.09375 59.90625 40.109375 62.90625 
+Q 36.140625 65.921875 28.71875 65.921875 
+Q 24.65625 65.921875 20.015625 65.03125 
+Q 15.375 64.15625 9.8125 62.3125 
+L 9.8125 71.09375 
+Q 15.4375 72.65625 20.34375 73.4375 
+Q 25.25 74.21875 29.59375 74.21875 
+Q 40.828125 74.21875 47.359375 69.109375 
+Q 53.90625 64.015625 53.90625 55.328125 
+Q 53.90625 49.265625 50.4375 45.09375 
+Q 46.96875 40.921875 40.578125 39.3125 
+z
+" id="DejaVuSans-51"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(191.366406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-51"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_5">
+     <g id="line2d_5">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="250.003906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_5">
+      <!-- 4000 -->
+      <defs>
+       <path d="M 37.796875 64.3125 
+L 12.890625 25.390625 
+L 37.796875 25.390625 
+z
+M 35.203125 72.90625 
+L 47.609375 72.90625 
+L 47.609375 25.390625 
+L 58.015625 25.390625 
+L 58.015625 17.1875 
+L 47.609375 17.1875 
+L 47.609375 0 
+L 37.796875 0 
+L 37.796875 17.1875 
+L 4.890625 17.1875 
+L 4.890625 26.703125 
+z
+" id="DejaVuSans-52"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(236.006406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-52"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_6">
+     <g id="line2d_6">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="294.643906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_6">
+      <!-- 5000 -->
+      <defs>
+       <path d="M 10.796875 72.90625 
+L 49.515625 72.90625 
+L 49.515625 64.59375 
+L 19.828125 64.59375 
+L 19.828125 46.734375 
+Q 21.96875 47.46875 24.109375 47.828125 
+Q 26.265625 48.1875 28.421875 48.1875 
+Q 40.625 48.1875 47.75 41.5 
+Q 54.890625 34.8125 54.890625 23.390625 
+Q 54.890625 11.625 47.5625 5.09375 
+Q 40.234375 -1.421875 26.90625 -1.421875 
+Q 22.3125 -1.421875 17.546875 -0.640625 
+Q 12.796875 0.140625 7.71875 1.703125 
+L 7.71875 11.625 
+Q 12.109375 9.234375 16.796875 8.0625 
+Q 21.484375 6.890625 26.703125 6.890625 
+Q 35.15625 6.890625 40.078125 11.328125 
+Q 45.015625 15.765625 45.015625 23.390625 
+Q 45.015625 31 40.078125 35.4375 
+Q 35.15625 39.890625 26.703125 39.890625 
+Q 22.75 39.890625 18.8125 39.015625 
+Q 14.890625 38.140625 10.796875 36.28125 
+z
+" id="DejaVuSans-53"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(280.646406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-53"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_7">
+     <g id="line2d_7">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="339.283906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_7">
+      <!-- 6000 -->
+      <defs>
+       <path d="M 33.015625 40.375 
+Q 26.375 40.375 22.484375 35.828125 
+Q 18.609375 31.296875 18.609375 23.390625 
+Q 18.609375 15.53125 22.484375 10.953125 
+Q 26.375 6.390625 33.015625 6.390625 
+Q 39.65625 6.390625 43.53125 10.953125 
+Q 47.40625 15.53125 47.40625 23.390625 
+Q 47.40625 31.296875 43.53125 35.828125 
+Q 39.65625 40.375 33.015625 40.375 
+z
+M 52.59375 71.296875 
+L 52.59375 62.3125 
+Q 48.875 64.0625 45.09375 64.984375 
+Q 41.3125 65.921875 37.59375 65.921875 
+Q 27.828125 65.921875 22.671875 59.328125 
+Q 17.53125 52.734375 16.796875 39.40625 
+Q 19.671875 43.65625 24.015625 45.921875 
+Q 28.375 48.1875 33.59375 48.1875 
+Q 44.578125 48.1875 50.953125 41.515625 
+Q 57.328125 34.859375 57.328125 23.390625 
+Q 57.328125 12.15625 50.6875 5.359375 
+Q 44.046875 -1.421875 33.015625 -1.421875 
+Q 20.359375 -1.421875 13.671875 8.265625 
+Q 6.984375 17.96875 6.984375 36.375 
+Q 6.984375 53.65625 15.1875 63.9375 
+Q 23.390625 74.21875 37.203125 74.21875 
+Q 40.921875 74.21875 44.703125 73.484375 
+Q 48.484375 72.75 52.59375 71.296875 
+z
+" id="DejaVuSans-54"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(325.286406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-54"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_8">
+     <g id="line2d_8">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="383.923906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_8">
+      <!-- 7000 -->
+      <defs>
+       <path d="M 8.203125 72.90625 
+L 55.078125 72.90625 
+L 55.078125 68.703125 
+L 28.609375 0 
+L 18.3125 0 
+L 43.21875 64.59375 
+L 8.203125 64.59375 
+z
+" id="DejaVuSans-55"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(369.926406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-55"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="xtick_9">
+     <g id="line2d_9">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="428.563906" xlink:href="#me95d5351a6" y="288.430125"/>
+      </g>
+     </g>
+     <g id="text_9">
+      <!-- 8000 -->
+      <defs>
+       <path d="M 31.78125 34.625 
+Q 24.75 34.625 20.71875 30.859375 
+Q 16.703125 27.09375 16.703125 20.515625 
+Q 16.703125 13.921875 20.71875 10.15625 
+Q 24.75 6.390625 31.78125 6.390625 
+Q 38.8125 6.390625 42.859375 10.171875 
+Q 46.921875 13.96875 46.921875 20.515625 
+Q 46.921875 27.09375 42.890625 30.859375 
+Q 38.875 34.625 31.78125 34.625 
+z
+M 21.921875 38.8125 
+Q 15.578125 40.375 12.03125 44.71875 
+Q 8.5 49.078125 8.5 55.328125 
+Q 8.5 64.0625 14.71875 69.140625 
+Q 20.953125 74.21875 31.78125 74.21875 
+Q 42.671875 74.21875 48.875 69.140625 
+Q 55.078125 64.0625 55.078125 55.328125 
+Q 55.078125 49.078125 51.53125 44.71875 
+Q 48 40.375 41.703125 38.8125 
+Q 48.828125 37.15625 52.796875 32.3125 
+Q 56.78125 27.484375 56.78125 20.515625 
+Q 56.78125 9.90625 50.3125 4.234375 
+Q 43.84375 -1.421875 31.78125 -1.421875 
+Q 19.734375 -1.421875 13.25 4.234375 
+Q 6.78125 9.90625 6.78125 20.515625 
+Q 6.78125 27.484375 10.78125 32.3125 
+Q 14.796875 37.15625 21.921875 38.8125 
+z
+M 18.3125 54.390625 
+Q 18.3125 48.734375 21.84375 45.5625 
+Q 25.390625 42.390625 31.78125 42.390625 
+Q 38.140625 42.390625 41.71875 45.5625 
+Q 45.3125 48.734375 45.3125 54.390625 
+Q 45.3125 60.0625 41.71875 63.234375 
+Q 38.140625 66.40625 31.78125 66.40625 
+Q 25.390625 66.40625 21.84375 63.234375 
+Q 18.3125 60.0625 18.3125 54.390625 
+z
+" id="DejaVuSans-56"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(414.566406 306.288406)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-56"/>
+       <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+       <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+       <use x="190.869141" xlink:href="#DejaVuSans-48"/>
+      </g>
+     </g>
+    </g>
+    <g id="text_10">
+     <!-- Speed (MiB/s) -->
+     <defs>
+      <path d="M 53.515625 70.515625 
+L 53.515625 60.890625 
+Q 47.90625 63.578125 42.921875 64.890625 
+Q 37.9375 66.21875 33.296875 66.21875 
+Q 25.25 66.21875 20.875 63.09375 
+Q 16.5 59.96875 16.5 54.203125 
+Q 16.5 49.359375 19.40625 46.890625 
+Q 22.3125 44.4375 30.421875 42.921875 
+L 36.375 41.703125 
+Q 47.40625 39.59375 52.65625 34.296875 
+Q 57.90625 29 57.90625 20.125 
+Q 57.90625 9.515625 50.796875 4.046875 
+Q 43.703125 -1.421875 29.984375 -1.421875 
+Q 24.8125 -1.421875 18.96875 -0.25 
+Q 13.140625 0.921875 6.890625 3.21875 
+L 6.890625 13.375 
+Q 12.890625 10.015625 18.65625 8.296875 
+Q 24.421875 6.59375 29.984375 6.59375 
+Q 38.421875 6.59375 43.015625 9.90625 
+Q 47.609375 13.234375 47.609375 19.390625 
+Q 47.609375 24.75 44.3125 27.78125 
+Q 41.015625 30.8125 33.5 32.328125 
+L 27.484375 33.5 
+Q 16.453125 35.6875 11.515625 40.375 
+Q 6.59375 45.0625 6.59375 53.421875 
+Q 6.59375 63.09375 13.40625 68.65625 
+Q 20.21875 74.21875 32.171875 74.21875 
+Q 37.3125 74.21875 42.625 73.28125 
+Q 47.953125 72.359375 53.515625 70.515625 
+z
+" id="DejaVuSans-83"/>
+      <path d="M 18.109375 8.203125 
+L 18.109375 -20.796875 
+L 9.078125 -20.796875 
+L 9.078125 54.6875 
+L 18.109375 54.6875 
+L 18.109375 46.390625 
+Q 20.953125 51.265625 25.265625 53.625 
+Q 29.59375 56 35.59375 56 
+Q 45.5625 56 51.78125 48.09375 
+Q 58.015625 40.1875 58.015625 27.296875 
+Q 58.015625 14.40625 51.78125 6.484375 
+Q 45.5625 -1.421875 35.59375 -1.421875 
+Q 29.59375 -1.421875 25.265625 0.953125 
+Q 20.953125 3.328125 18.109375 8.203125 
+z
+M 48.6875 27.296875 
+Q 48.6875 37.203125 44.609375 42.84375 
+Q 40.53125 48.484375 33.40625 48.484375 
+Q 26.265625 48.484375 22.1875 42.84375 
+Q 18.109375 37.203125 18.109375 27.296875 
+Q 18.109375 17.390625 22.1875 11.75 
+Q 26.265625 6.109375 33.40625 6.109375 
+Q 40.53125 6.109375 44.609375 11.75 
+Q 48.6875 17.390625 48.6875 27.296875 
+z
+" id="DejaVuSans-112"/>
+      <path d="M 56.203125 29.59375 
+L 56.203125 25.203125 
+L 14.890625 25.203125 
+Q 15.484375 15.921875 20.484375 11.0625 
+Q 25.484375 6.203125 34.421875 6.203125 
+Q 39.59375 6.203125 44.453125 7.46875 
+Q 49.3125 8.734375 54.109375 11.28125 
+L 54.109375 2.78125 
+Q 49.265625 0.734375 44.1875 -0.34375 
+Q 39.109375 -1.421875 33.890625 -1.421875 
+Q 20.796875 -1.421875 13.15625 6.1875 
+Q 5.515625 13.8125 5.515625 26.8125 
+Q 5.515625 40.234375 12.765625 48.109375 
+Q 20.015625 56 32.328125 56 
+Q 43.359375 56 49.78125 48.890625 
+Q 56.203125 41.796875 56.203125 29.59375 
+z
+M 47.21875 32.234375 
+Q 47.125 39.59375 43.09375 43.984375 
+Q 39.0625 48.390625 32.421875 48.390625 
+Q 24.90625 48.390625 20.390625 44.140625 
+Q 15.875 39.890625 15.1875 32.171875 
+z
+" id="DejaVuSans-101"/>
+      <path d="M 45.40625 46.390625 
+L 45.40625 75.984375 
+L 54.390625 75.984375 
+L 54.390625 0 
+L 45.40625 0 
+L 45.40625 8.203125 
+Q 42.578125 3.328125 38.25 0.953125 
+Q 33.9375 -1.421875 27.875 -1.421875 
+Q 17.96875 -1.421875 11.734375 6.484375 
+Q 5.515625 14.40625 5.515625 27.296875 
+Q 5.515625 40.1875 11.734375 48.09375 
+Q 17.96875 56 27.875 56 
+Q 33.9375 56 38.25 53.625 
+Q 42.578125 51.265625 45.40625 46.390625 
+z
+M 14.796875 27.296875 
+Q 14.796875 17.390625 18.875 11.75 
+Q 22.953125 6.109375 30.078125 6.109375 
+Q 37.203125 6.109375 41.296875 11.75 
+Q 45.40625 17.390625 45.40625 27.296875 
+Q 45.40625 37.203125 41.296875 42.84375 
+Q 37.203125 48.484375 30.078125 48.484375 
+Q 22.953125 48.484375 18.875 42.84375 
+Q 14.796875 37.203125 14.796875 27.296875 
+z
+" id="DejaVuSans-100"/>
+      <path id="DejaVuSans-32"/>
+      <path d="M 31 75.875 
+Q 24.46875 64.65625 21.28125 53.65625 
+Q 18.109375 42.671875 18.109375 31.390625 
+Q 18.109375 20.125 21.3125 9.0625 
+Q 24.515625 -2 31 -13.1875 
+L 23.1875 -13.1875 
+Q 15.875 -1.703125 12.234375 9.375 
+Q 8.59375 20.453125 8.59375 31.390625 
+Q 8.59375 42.28125 12.203125 53.3125 
+Q 15.828125 64.359375 23.1875 75.875 
+z
+" id="DejaVuSans-40"/>
+      <path d="M 9.8125 72.90625 
+L 24.515625 72.90625 
+L 43.109375 23.296875 
+L 61.8125 72.90625 
+L 76.515625 72.90625 
+L 76.515625 0 
+L 66.890625 0 
+L 66.890625 64.015625 
+L 48.09375 14.015625 
+L 38.1875 14.015625 
+L 19.390625 64.015625 
+L 19.390625 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-77"/>
+      <path d="M 9.421875 54.6875 
+L 18.40625 54.6875 
+L 18.40625 0 
+L 9.421875 0 
+z
+M 9.421875 75.984375 
+L 18.40625 75.984375 
+L 18.40625 64.59375 
+L 9.421875 64.59375 
+z
+" id="DejaVuSans-105"/>
+      <path d="M 19.671875 34.8125 
+L 19.671875 8.109375 
+L 35.5 8.109375 
+Q 43.453125 8.109375 47.28125 11.40625 
+Q 51.125 14.703125 51.125 21.484375 
+Q 51.125 28.328125 47.28125 31.5625 
+Q 43.453125 34.8125 35.5 34.8125 
+z
+M 19.671875 64.796875 
+L 19.671875 42.828125 
+L 34.28125 42.828125 
+Q 41.5 42.828125 45.03125 45.53125 
+Q 48.578125 48.25 48.578125 53.8125 
+Q 48.578125 59.328125 45.03125 62.0625 
+Q 41.5 64.796875 34.28125 64.796875 
+z
+M 9.8125 72.90625 
+L 35.015625 72.90625 
+Q 46.296875 72.90625 52.390625 68.21875 
+Q 58.5 63.53125 58.5 54.890625 
+Q 58.5 48.1875 55.375 44.234375 
+Q 52.25 40.28125 46.1875 39.3125 
+Q 53.46875 37.75 57.5 32.78125 
+Q 61.53125 27.828125 61.53125 20.40625 
+Q 61.53125 10.640625 54.890625 5.3125 
+Q 48.25 0 35.984375 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-66"/>
+      <path d="M 25.390625 72.90625 
+L 33.6875 72.90625 
+L 8.296875 -9.28125 
+L 0 -9.28125 
+z
+" id="DejaVuSans-47"/>
+      <path d="M 44.28125 53.078125 
+L 44.28125 44.578125 
+Q 40.484375 46.53125 36.375 47.5 
+Q 32.28125 48.484375 27.875 48.484375 
+Q 21.1875 48.484375 17.84375 46.4375 
+Q 14.5 44.390625 14.5 40.28125 
+Q 14.5 37.15625 16.890625 35.375 
+Q 19.28125 33.59375 26.515625 31.984375 
+L 29.59375 31.296875 
+Q 39.15625 29.25 43.1875 25.515625 
+Q 47.21875 21.78125 47.21875 15.09375 
+Q 47.21875 7.46875 41.1875 3.015625 
+Q 35.15625 -1.421875 24.609375 -1.421875 
+Q 20.21875 -1.421875 15.453125 -0.5625 
+Q 10.6875 0.296875 5.421875 2 
+L 5.421875 11.28125 
+Q 10.40625 8.6875 15.234375 7.390625 
+Q 20.0625 6.109375 24.8125 6.109375 
+Q 31.15625 6.109375 34.5625 8.28125 
+Q 37.984375 10.453125 37.984375 14.40625 
+Q 37.984375 18.0625 35.515625 20.015625 
+Q 33.0625 21.96875 24.703125 23.78125 
+L 21.578125 24.515625 
+Q 13.234375 26.265625 9.515625 29.90625 
+Q 5.8125 33.546875 5.8125 39.890625 
+Q 5.8125 47.609375 11.28125 51.796875 
+Q 16.75 56 26.8125 56 
+Q 31.78125 56 36.171875 55.265625 
+Q 40.578125 54.546875 44.28125 53.078125 
+z
+" id="DejaVuSans-115"/>
+      <path d="M 8.015625 75.875 
+L 15.828125 75.875 
+Q 23.140625 64.359375 26.78125 53.3125 
+Q 30.421875 42.28125 30.421875 31.390625 
+Q 30.421875 20.453125 26.78125 9.375 
+Q 23.140625 -1.703125 15.828125 -13.1875 
+L 8.015625 -13.1875 
+Q 14.5 -2 17.703125 9.0625 
+Q 20.90625 20.125 20.90625 31.390625 
+Q 20.90625 42.671875 17.703125 53.65625 
+Q 14.5 64.65625 8.015625 75.875 
+z
+" id="DejaVuSans-41"/>
+     </defs>
+     <g style="fill:#262626;" transform="translate(208.497031 321.694187)scale(0.12 -0.12)">
+      <use xlink:href="#DejaVuSans-83"/>
+      <use x="63.476562" xlink:href="#DejaVuSans-112"/>
+      <use x="126.953125" xlink:href="#DejaVuSans-101"/>
+      <use x="188.476562" xlink:href="#DejaVuSans-101"/>
+      <use x="250" xlink:href="#DejaVuSans-100"/>
+      <use x="313.476562" xlink:href="#DejaVuSans-32"/>
+      <use x="345.263672" xlink:href="#DejaVuSans-40"/>
+      <use x="384.277344" xlink:href="#DejaVuSans-77"/>
+      <use x="470.556641" xlink:href="#DejaVuSans-105"/>
+      <use x="498.339844" xlink:href="#DejaVuSans-66"/>
+      <use x="566.943359" xlink:href="#DejaVuSans-47"/>
+      <use x="600.634766" xlink:href="#DejaVuSans-115"/>
+      <use x="652.734375" xlink:href="#DejaVuSans-41"/>
+     </g>
+    </g>
+   </g>
+   <g id="matplotlib.axis_2">
+    <g id="ytick_1">
+     <g id="line2d_10">
+      <defs>
+       <path d="M 0 0 
+L -6 0 
+" id="m7d1bb602a9" style="stroke:#262626;stroke-width:1.25;"/>
+      </defs>
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="38.950125"/>
+      </g>
+     </g>
+     <g id="text_11">
+      <!-- BLAKE3 -->
+      <defs>
+       <path d="M 9.8125 72.90625 
+L 19.671875 72.90625 
+L 19.671875 8.296875 
+L 55.171875 8.296875 
+L 55.171875 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-76"/>
+       <path d="M 34.1875 63.1875 
+L 20.796875 26.90625 
+L 47.609375 26.90625 
+z
+M 28.609375 72.90625 
+L 39.796875 72.90625 
+L 67.578125 0 
+L 57.328125 0 
+L 50.6875 18.703125 
+L 17.828125 18.703125 
+L 11.1875 0 
+L 0.78125 0 
+z
+" id="DejaVuSans-65"/>
+       <path d="M 9.8125 72.90625 
+L 19.671875 72.90625 
+L 19.671875 42.09375 
+L 52.390625 72.90625 
+L 65.09375 72.90625 
+L 28.90625 38.921875 
+L 67.671875 0 
+L 54.6875 0 
+L 19.671875 35.109375 
+L 19.671875 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-75"/>
+       <path d="M 9.8125 72.90625 
+L 55.90625 72.90625 
+L 55.90625 64.59375 
+L 19.671875 64.59375 
+L 19.671875 43.015625 
+L 54.390625 43.015625 
+L 54.390625 34.71875 
+L 19.671875 34.71875 
+L 19.671875 8.296875 
+L 56.78125 8.296875 
+L 56.78125 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-69"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(19.576719 43.129266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-66"/>
+       <use x="68.603516" xlink:href="#DejaVuSans-76"/>
+       <use x="124.347656" xlink:href="#DejaVuSans-65"/>
+       <use x="192.755859" xlink:href="#DejaVuSans-75"/>
+       <use x="258.332031" xlink:href="#DejaVuSans-69"/>
+       <use x="321.515625" xlink:href="#DejaVuSans-51"/>
+      </g>
+     </g>
+    </g>
+    <g id="ytick_2">
+     <g id="line2d_11">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="72.214125"/>
+      </g>
+     </g>
+     <g id="text_12">
+      <!-- BLAKE2b -->
+      <defs>
+       <path d="M 48.6875 27.296875 
+Q 48.6875 37.203125 44.609375 42.84375 
+Q 40.53125 48.484375 33.40625 48.484375 
+Q 26.265625 48.484375 22.1875 42.84375 
+Q 18.109375 37.203125 18.109375 27.296875 
+Q 18.109375 17.390625 22.1875 11.75 
+Q 26.265625 6.109375 33.40625 6.109375 
+Q 40.53125 6.109375 44.609375 11.75 
+Q 48.6875 17.390625 48.6875 27.296875 
+z
+M 18.109375 46.390625 
+Q 20.953125 51.265625 25.265625 53.625 
+Q 29.59375 56 35.59375 56 
+Q 45.5625 56 51.78125 48.09375 
+Q 58.015625 40.1875 58.015625 27.296875 
+Q 58.015625 14.40625 51.78125 6.484375 
+Q 45.5625 -1.421875 35.59375 -1.421875 
+Q 29.59375 -1.421875 25.265625 0.953125 
+Q 20.953125 3.328125 18.109375 8.203125 
+L 18.109375 0 
+L 9.078125 0 
+L 9.078125 75.984375 
+L 18.109375 75.984375 
+z
+" id="DejaVuSans-98"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(12.593438 76.393266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-66"/>
+       <use x="68.603516" xlink:href="#DejaVuSans-76"/>
+       <use x="124.347656" xlink:href="#DejaVuSans-65"/>
+       <use x="192.755859" xlink:href="#DejaVuSans-75"/>
+       <use x="258.332031" xlink:href="#DejaVuSans-69"/>
+       <use x="321.515625" xlink:href="#DejaVuSans-50"/>
+       <use x="385.138672" xlink:href="#DejaVuSans-98"/>
+      </g>
+     </g>
+    </g>
+    <g id="ytick_3">
+     <g id="line2d_12">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="105.478125"/>
+      </g>
+     </g>
+     <g id="text_13">
+      <!-- SHA-1 -->
+      <defs>
+       <path d="M 9.8125 72.90625 
+L 19.671875 72.90625 
+L 19.671875 43.015625 
+L 55.515625 43.015625 
+L 55.515625 72.90625 
+L 65.375 72.90625 
+L 65.375 0 
+L 55.515625 0 
+L 55.515625 34.71875 
+L 19.671875 34.71875 
+L 19.671875 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-72"/>
+       <path d="M 4.890625 31.390625 
+L 31.203125 31.390625 
+L 31.203125 23.390625 
+L 4.890625 23.390625 
+z
+" id="DejaVuSans-45"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(28.199687 109.657266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-83"/>
+       <use x="63.476562" xlink:href="#DejaVuSans-72"/>
+       <use x="138.671875" xlink:href="#DejaVuSans-65"/>
+       <use x="207.048828" xlink:href="#DejaVuSans-45"/>
+       <use x="243.132812" xlink:href="#DejaVuSans-49"/>
+      </g>
+     </g>
+    </g>
+    <g id="ytick_4">
+     <g id="line2d_13">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="138.742125"/>
+      </g>
+     </g>
+     <g id="text_14">
+      <!-- BLAKE2s -->
+      <g style="fill:#262626;" transform="translate(13.846406 142.921266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-66"/>
+       <use x="68.603516" xlink:href="#DejaVuSans-76"/>
+       <use x="124.347656" xlink:href="#DejaVuSans-65"/>
+       <use x="192.755859" xlink:href="#DejaVuSans-75"/>
+       <use x="258.332031" xlink:href="#DejaVuSans-69"/>
+       <use x="321.515625" xlink:href="#DejaVuSans-50"/>
+       <use x="385.138672" xlink:href="#DejaVuSans-115"/>
+      </g>
+     </g>
+    </g>
+    <g id="ytick_5">
+     <g id="line2d_14">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="172.006125"/>
+      </g>
+     </g>
+     <g id="text_15">
+      <!-- MD5 -->
+      <defs>
+       <path d="M 19.671875 64.796875 
+L 19.671875 8.109375 
+L 31.59375 8.109375 
+Q 46.6875 8.109375 53.6875 14.9375 
+Q 60.6875 21.78125 60.6875 36.53125 
+Q 60.6875 51.171875 53.6875 57.984375 
+Q 46.6875 64.796875 31.59375 64.796875 
+z
+M 9.8125 72.90625 
+L 30.078125 72.90625 
+Q 51.265625 72.90625 61.171875 64.09375 
+Q 71.09375 55.28125 71.09375 36.53125 
+Q 71.09375 17.671875 61.125 8.828125 
+Q 51.171875 0 30.078125 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-68"/>
+      </defs>
+      <g style="fill:#262626;" transform="translate(36.984219 176.185266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-77"/>
+       <use x="86.279297" xlink:href="#DejaVuSans-68"/>
+       <use x="163.28125" xlink:href="#DejaVuSans-53"/>
+      </g>
+     </g>
+    </g>
+    <g id="ytick_6">
+     <g id="line2d_15">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="205.270125"/>
+      </g>
+     </g>
+     <g id="text_16">
+      <!-- SHA-512 -->
+      <g style="fill:#262626;" transform="translate(14.202188 209.449266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-83"/>
+       <use x="63.476562" xlink:href="#DejaVuSans-72"/>
+       <use x="138.671875" xlink:href="#DejaVuSans-65"/>
+       <use x="207.048828" xlink:href="#DejaVuSans-45"/>
+       <use x="243.132812" xlink:href="#DejaVuSans-53"/>
+       <use x="306.755859" xlink:href="#DejaVuSans-49"/>
+       <use x="370.378906" xlink:href="#DejaVuSans-50"/>
+      </g>
+     </g>
+    </g>
+    <g id="ytick_7">
+     <g id="line2d_16">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="238.534125"/>
+      </g>
+     </g>
+     <g id="text_17">
+      <!-- SHA-256 -->
+      <g style="fill:#262626;" transform="translate(14.202188 242.713266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-83"/>
+       <use x="63.476562" xlink:href="#DejaVuSans-72"/>
+       <use x="138.671875" xlink:href="#DejaVuSans-65"/>
+       <use x="207.048828" xlink:href="#DejaVuSans-45"/>
+       <use x="243.132812" xlink:href="#DejaVuSans-50"/>
+       <use x="306.755859" xlink:href="#DejaVuSans-53"/>
+       <use x="370.378906" xlink:href="#DejaVuSans-54"/>
+      </g>
+     </g>
+    </g>
+    <g id="ytick_8">
+     <g id="line2d_17">
+      <g>
+       <use style="fill:#262626;stroke:#262626;stroke-width:1.25;" x="71.443906" xlink:href="#m7d1bb602a9" y="271.798125"/>
+      </g>
+     </g>
+     <g id="text_18">
+      <!-- SHA3-256 -->
+      <g style="fill:#262626;" transform="translate(7.2 275.977266)scale(0.11 -0.11)">
+       <use xlink:href="#DejaVuSans-83"/>
+       <use x="63.476562" xlink:href="#DejaVuSans-72"/>
+       <use x="138.671875" xlink:href="#DejaVuSans-65"/>
+       <use x="207.080078" xlink:href="#DejaVuSans-51"/>
+       <use x="270.703125" xlink:href="#DejaVuSans-45"/>
+       <use x="306.787109" xlink:href="#DejaVuSans-50"/>
+       <use x="370.410156" xlink:href="#DejaVuSans-53"/>
+       <use x="434.033203" xlink:href="#DejaVuSans-54"/>
+      </g>
+     </g>
+    </g>
+   </g>
+   <g id="patch_3">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 25.644525 
+L 377.942146 25.644525 
+L 377.942146 52.255725 
+L 71.443906 52.255725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="patch_4">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 58.908525 
+L 130.011586 58.908525 
+L 130.011586 85.519725 
+L 71.443906 85.519725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="patch_5">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 92.172525 
+L 117.289186 92.172525 
+L 117.289186 118.783725 
+L 71.443906 118.783725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="patch_6">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 125.436525 
+L 110.548546 125.436525 
+L 110.548546 152.047725 
+L 71.443906 152.047725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="patch_7">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 158.700525 
+L 104.477506 158.700525 
+L 104.477506 185.311725 
+L 71.443906 185.311725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="patch_8">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 191.964525 
+L 103.584706 191.964525 
+L 103.584706 218.575725 
+L 71.443906 218.575725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="patch_9">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 225.228525 
+L 93.049666 225.228525 
+L 93.049666 251.839725 
+L 71.443906 251.839725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="patch_10">
+    <path clip-path="url(#p6091bd3d0f)" d="M 71.443906 258.492525 
+L 89.032066 258.492525 
+L 89.032066 285.103725 
+L 71.443906 285.103725 
+z
+" style="fill:#df2020;stroke:#000000;stroke-linejoin:miter;"/>
+   </g>
+   <g id="line2d_18">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="line2d_19">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="line2d_20">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="line2d_21">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="line2d_22">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="line2d_23">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="line2d_24">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="line2d_25">
+    <path clip-path="url(#p6091bd3d0f)" d="M 0 0 
+" style="fill:none;stroke:#424242;stroke-linecap:round;stroke-width:2.7;"/>
+   </g>
+   <g id="patch_11">
+    <path d="M 71.443906 288.430125 
+L 71.443906 22.318125 
+" style="fill:none;stroke:#262626;stroke-linecap:square;stroke-linejoin:miter;stroke-width:1.25;"/>
+   </g>
+   <g id="patch_12">
+    <path d="M 428.563906 288.430125 
+L 428.563906 22.318125 
+" style="fill:none;stroke:#262626;stroke-linecap:square;stroke-linejoin:miter;stroke-width:1.25;"/>
+   </g>
+   <g id="patch_13">
+    <path d="M 71.443906 288.430125 
+L 428.563906 288.430125 
+" style="fill:none;stroke:#262626;stroke-linecap:square;stroke-linejoin:miter;stroke-width:1.25;"/>
+   </g>
+   <g id="patch_14">
+    <path d="M 71.443906 22.318125 
+L 428.563906 22.318125 
+" style="fill:none;stroke:#262626;stroke-linecap:square;stroke-linejoin:miter;stroke-width:1.25;"/>
+   </g>
+   <g id="text_19">
+    <!-- 6866 -->
+    <g style="fill:#262626;" transform="translate(382.406146 43.939725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-54"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-56"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-54"/>
+     <use x="190.869141" xlink:href="#DejaVuSans-54"/>
+    </g>
+   </g>
+   <g id="text_20">
+    <!-- 1312 -->
+    <g style="fill:#262626;" transform="translate(134.475586 77.203725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-49"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-51"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-49"/>
+     <use x="190.869141" xlink:href="#DejaVuSans-50"/>
+    </g>
+   </g>
+   <g id="text_21">
+    <!-- 1027 -->
+    <g style="fill:#262626;" transform="translate(121.753186 110.467725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-49"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-48"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-50"/>
+     <use x="190.869141" xlink:href="#DejaVuSans-55"/>
+    </g>
+   </g>
+   <g id="text_22">
+    <!-- 876 -->
+    <g style="fill:#262626;" transform="translate(115.012546 143.731725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-56"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-55"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-54"/>
+    </g>
+   </g>
+   <g id="text_23">
+    <!-- 740 -->
+    <g style="fill:#262626;" transform="translate(108.941506 176.995725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-55"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-52"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+    </g>
+   </g>
+   <g id="text_24">
+    <!-- 720 -->
+    <g style="fill:#262626;" transform="translate(108.048706 210.259725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-55"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-50"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-48"/>
+    </g>
+   </g>
+   <g id="text_25">
+    <!-- 484 -->
+    <g style="fill:#262626;" transform="translate(97.513666 243.523725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-52"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-56"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-52"/>
+    </g>
+   </g>
+   <g id="text_26">
+    <!-- 394 -->
+    <defs>
+     <path d="M 10.984375 1.515625 
+L 10.984375 10.5 
+Q 14.703125 8.734375 18.5 7.8125 
+Q 22.3125 6.890625 25.984375 6.890625 
+Q 35.75 6.890625 40.890625 13.453125 
+Q 46.046875 20.015625 46.78125 33.40625 
+Q 43.953125 29.203125 39.59375 26.953125 
+Q 35.25 24.703125 29.984375 24.703125 
+Q 19.046875 24.703125 12.671875 31.3125 
+Q 6.296875 37.9375 6.296875 49.421875 
+Q 6.296875 60.640625 12.9375 67.421875 
+Q 19.578125 74.21875 30.609375 74.21875 
+Q 43.265625 74.21875 49.921875 64.515625 
+Q 56.59375 54.828125 56.59375 36.375 
+Q 56.59375 19.140625 48.40625 8.859375 
+Q 40.234375 -1.421875 26.421875 -1.421875 
+Q 22.703125 -1.421875 18.890625 -0.6875 
+Q 15.09375 0.046875 10.984375 1.515625 
+z
+M 30.609375 32.421875 
+Q 37.25 32.421875 41.125 36.953125 
+Q 45.015625 41.5 45.015625 49.421875 
+Q 45.015625 57.28125 41.125 61.84375 
+Q 37.25 66.40625 30.609375 66.40625 
+Q 23.96875 66.40625 20.09375 61.84375 
+Q 16.21875 57.28125 16.21875 49.421875 
+Q 16.21875 41.5 20.09375 36.953125 
+Q 23.96875 32.421875 30.609375 32.421875 
+z
+" id="DejaVuSans-57"/>
+    </defs>
+    <g style="fill:#262626;" transform="translate(93.496066 276.787725)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-51"/>
+     <use x="63.623047" xlink:href="#DejaVuSans-57"/>
+     <use x="127.246094" xlink:href="#DejaVuSans-52"/>
+    </g>
+   </g>
+   <g id="text_27">
+    <!-- Performance on AWS c5.metal, 16 KiB input, 1 thread -->
+    <defs>
+     <path d="M 19.671875 64.796875 
+L 19.671875 37.40625 
+L 32.078125 37.40625 
+Q 38.96875 37.40625 42.71875 40.96875 
+Q 46.484375 44.53125 46.484375 51.125 
+Q 46.484375 57.671875 42.71875 61.234375 
+Q 38.96875 64.796875 32.078125 64.796875 
+z
+M 9.8125 72.90625 
+L 32.078125 72.90625 
+Q 44.34375 72.90625 50.609375 67.359375 
+Q 56.890625 61.8125 56.890625 51.125 
+Q 56.890625 40.328125 50.609375 34.8125 
+Q 44.34375 29.296875 32.078125 29.296875 
+L 19.671875 29.296875 
+L 19.671875 0 
+L 9.8125 0 
+z
+" id="DejaVuSans-80"/>
+     <path d="M 41.109375 46.296875 
+Q 39.59375 47.171875 37.8125 47.578125 
+Q 36.03125 48 33.890625 48 
+Q 26.265625 48 22.1875 43.046875 
+Q 18.109375 38.09375 18.109375 28.8125 
+L 18.109375 0 
+L 9.078125 0 
+L 9.078125 54.6875 
+L 18.109375 54.6875 
+L 18.109375 46.1875 
+Q 20.953125 51.171875 25.484375 53.578125 
+Q 30.03125 56 36.53125 56 
+Q 37.453125 56 38.578125 55.875 
+Q 39.703125 55.765625 41.0625 55.515625 
+z
+" id="DejaVuSans-114"/>
+     <path d="M 37.109375 75.984375 
+L 37.109375 68.5 
+L 28.515625 68.5 
+Q 23.6875 68.5 21.796875 66.546875 
+Q 19.921875 64.59375 19.921875 59.515625 
+L 19.921875 54.6875 
+L 34.71875 54.6875 
+L 34.71875 47.703125 
+L 19.921875 47.703125 
+L 19.921875 0 
+L 10.890625 0 
+L 10.890625 47.703125 
+L 2.296875 47.703125 
+L 2.296875 54.6875 
+L 10.890625 54.6875 
+L 10.890625 58.5 
+Q 10.890625 67.625 15.140625 71.796875 
+Q 19.390625 75.984375 28.609375 75.984375 
+z
+" id="DejaVuSans-102"/>
+     <path d="M 30.609375 48.390625 
+Q 23.390625 48.390625 19.1875 42.75 
+Q 14.984375 37.109375 14.984375 27.296875 
+Q 14.984375 17.484375 19.15625 11.84375 
+Q 23.34375 6.203125 30.609375 6.203125 
+Q 37.796875 6.203125 41.984375 11.859375 
+Q 46.1875 17.53125 46.1875 27.296875 
+Q 46.1875 37.015625 41.984375 42.703125 
+Q 37.796875 48.390625 30.609375 48.390625 
+z
+M 30.609375 56 
+Q 42.328125 56 49.015625 48.375 
+Q 55.71875 40.765625 55.71875 27.296875 
+Q 55.71875 13.875 49.015625 6.21875 
+Q 42.328125 -1.421875 30.609375 -1.421875 
+Q 18.84375 -1.421875 12.171875 6.21875 
+Q 5.515625 13.875 5.515625 27.296875 
+Q 5.515625 40.765625 12.171875 48.375 
+Q 18.84375 56 30.609375 56 
+z
+" id="DejaVuSans-111"/>
+     <path d="M 52 44.1875 
+Q 55.375 50.25 60.0625 53.125 
+Q 64.75 56 71.09375 56 
+Q 79.640625 56 84.28125 50.015625 
+Q 88.921875 44.046875 88.921875 33.015625 
+L 88.921875 0 
+L 79.890625 0 
+L 79.890625 32.71875 
+Q 79.890625 40.578125 77.09375 44.375 
+Q 74.3125 48.1875 68.609375 48.1875 
+Q 61.625 48.1875 57.5625 43.546875 
+Q 53.515625 38.921875 53.515625 30.90625 
+L 53.515625 0 
+L 44.484375 0 
+L 44.484375 32.71875 
+Q 44.484375 40.625 41.703125 44.40625 
+Q 38.921875 48.1875 33.109375 48.1875 
+Q 26.21875 48.1875 22.15625 43.53125 
+Q 18.109375 38.875 18.109375 30.90625 
+L 18.109375 0 
+L 9.078125 0 
+L 9.078125 54.6875 
+L 18.109375 54.6875 
+L 18.109375 46.1875 
+Q 21.1875 51.21875 25.484375 53.609375 
+Q 29.78125 56 35.6875 56 
+Q 41.65625 56 45.828125 52.96875 
+Q 50 49.953125 52 44.1875 
+z
+" id="DejaVuSans-109"/>
+     <path d="M 34.28125 27.484375 
+Q 23.390625 27.484375 19.1875 25 
+Q 14.984375 22.515625 14.984375 16.5 
+Q 14.984375 11.71875 18.140625 8.90625 
+Q 21.296875 6.109375 26.703125 6.109375 
+Q 34.1875 6.109375 38.703125 11.40625 
+Q 43.21875 16.703125 43.21875 25.484375 
+L 43.21875 27.484375 
+z
+M 52.203125 31.203125 
+L 52.203125 0 
+L 43.21875 0 
+L 43.21875 8.296875 
+Q 40.140625 3.328125 35.546875 0.953125 
+Q 30.953125 -1.421875 24.3125 -1.421875 
+Q 15.921875 -1.421875 10.953125 3.296875 
+Q 6 8.015625 6 15.921875 
+Q 6 25.140625 12.171875 29.828125 
+Q 18.359375 34.515625 30.609375 34.515625 
+L 43.21875 34.515625 
+L 43.21875 35.40625 
+Q 43.21875 41.609375 39.140625 45 
+Q 35.0625 48.390625 27.6875 48.390625 
+Q 23 48.390625 18.546875 47.265625 
+Q 14.109375 46.140625 10.015625 43.890625 
+L 10.015625 52.203125 
+Q 14.9375 54.109375 19.578125 55.046875 
+Q 24.21875 56 28.609375 56 
+Q 40.484375 56 46.34375 49.84375 
+Q 52.203125 43.703125 52.203125 31.203125 
+z
+" id="DejaVuSans-97"/>
+     <path d="M 54.890625 33.015625 
+L 54.890625 0 
+L 45.90625 0 
+L 45.90625 32.71875 
+Q 45.90625 40.484375 42.875 44.328125 
+Q 39.84375 48.1875 33.796875 48.1875 
+Q 26.515625 48.1875 22.3125 43.546875 
+Q 18.109375 38.921875 18.109375 30.90625 
+L 18.109375 0 
+L 9.078125 0 
+L 9.078125 54.6875 
+L 18.109375 54.6875 
+L 18.109375 46.1875 
+Q 21.34375 51.125 25.703125 53.5625 
+Q 30.078125 56 35.796875 56 
+Q 45.21875 56 50.046875 50.171875 
+Q 54.890625 44.34375 54.890625 33.015625 
+z
+" id="DejaVuSans-110"/>
+     <path d="M 48.78125 52.59375 
+L 48.78125 44.1875 
+Q 44.96875 46.296875 41.140625 47.34375 
+Q 37.3125 48.390625 33.40625 48.390625 
+Q 24.65625 48.390625 19.8125 42.84375 
+Q 14.984375 37.3125 14.984375 27.296875 
+Q 14.984375 17.28125 19.8125 11.734375 
+Q 24.65625 6.203125 33.40625 6.203125 
+Q 37.3125 6.203125 41.140625 7.25 
+Q 44.96875 8.296875 48.78125 10.40625 
+L 48.78125 2.09375 
+Q 45.015625 0.34375 40.984375 -0.53125 
+Q 36.96875 -1.421875 32.421875 -1.421875 
+Q 20.0625 -1.421875 12.78125 6.34375 
+Q 5.515625 14.109375 5.515625 27.296875 
+Q 5.515625 40.671875 12.859375 48.328125 
+Q 20.21875 56 33.015625 56 
+Q 37.15625 56 41.109375 55.140625 
+Q 45.0625 54.296875 48.78125 52.59375 
+z
+" id="DejaVuSans-99"/>
+     <path d="M 3.328125 72.90625 
+L 13.28125 72.90625 
+L 28.609375 11.28125 
+L 43.890625 72.90625 
+L 54.984375 72.90625 
+L 70.3125 11.28125 
+L 85.59375 72.90625 
+L 95.609375 72.90625 
+L 77.296875 0 
+L 64.890625 0 
+L 49.515625 63.28125 
+L 33.984375 0 
+L 21.578125 0 
+z
+" id="DejaVuSans-87"/>
+     <path d="M 10.6875 12.40625 
+L 21 12.40625 
+L 21 0 
+L 10.6875 0 
+z
+" id="DejaVuSans-46"/>
+     <path d="M 18.3125 70.21875 
+L 18.3125 54.6875 
+L 36.8125 54.6875 
+L 36.8125 47.703125 
+L 18.3125 47.703125 
+L 18.3125 18.015625 
+Q 18.3125 11.328125 20.140625 9.421875 
+Q 21.96875 7.515625 27.59375 7.515625 
+L 36.8125 7.515625 
+L 36.8125 0 
+L 27.59375 0 
+Q 17.1875 0 13.234375 3.875 
+Q 9.28125 7.765625 9.28125 18.015625 
+L 9.28125 47.703125 
+L 2.6875 47.703125 
+L 2.6875 54.6875 
+L 9.28125 54.6875 
+L 9.28125 70.21875 
+z
+" id="DejaVuSans-116"/>
+     <path d="M 9.421875 75.984375 
+L 18.40625 75.984375 
+L 18.40625 0 
+L 9.421875 0 
+z
+" id="DejaVuSans-108"/>
+     <path d="M 11.71875 12.40625 
+L 22.015625 12.40625 
+L 22.015625 4 
+L 14.015625 -11.625 
+L 7.71875 -11.625 
+L 11.71875 4 
+z
+" id="DejaVuSans-44"/>
+     <path d="M 8.5 21.578125 
+L 8.5 54.6875 
+L 17.484375 54.6875 
+L 17.484375 21.921875 
+Q 17.484375 14.15625 20.5 10.265625 
+Q 23.53125 6.390625 29.59375 6.390625 
+Q 36.859375 6.390625 41.078125 11.03125 
+Q 45.3125 15.671875 45.3125 23.6875 
+L 45.3125 54.6875 
+L 54.296875 54.6875 
+L 54.296875 0 
+L 45.3125 0 
+L 45.3125 8.40625 
+Q 42.046875 3.421875 37.71875 1 
+Q 33.40625 -1.421875 27.6875 -1.421875 
+Q 18.265625 -1.421875 13.375 4.4375 
+Q 8.5 10.296875 8.5 21.578125 
+z
+M 31.109375 56 
+z
+" id="DejaVuSans-117"/>
+     <path d="M 54.890625 33.015625 
+L 54.890625 0 
+L 45.90625 0 
+L 45.90625 32.71875 
+Q 45.90625 40.484375 42.875 44.328125 
+Q 39.84375 48.1875 33.796875 48.1875 
+Q 26.515625 48.1875 22.3125 43.546875 
+Q 18.109375 38.921875 18.109375 30.90625 
+L 18.109375 0 
+L 9.078125 0 
+L 9.078125 75.984375 
+L 18.109375 75.984375 
+L 18.109375 46.1875 
+Q 21.34375 51.125 25.703125 53.5625 
+Q 30.078125 56 35.796875 56 
+Q 45.21875 56 50.046875 50.171875 
+Q 54.890625 44.34375 54.890625 33.015625 
+z
+" id="DejaVuSans-104"/>
+    </defs>
+    <g style="fill:#262626;" transform="translate(88.626406 16.318125)scale(0.12 -0.12)">
+     <use xlink:href="#DejaVuSans-80"/>
+     <use x="60.255859" xlink:href="#DejaVuSans-101"/>
+     <use x="121.779297" xlink:href="#DejaVuSans-114"/>
+     <use x="162.892578" xlink:href="#DejaVuSans-102"/>
+     <use x="198.097656" xlink:href="#DejaVuSans-111"/>
+     <use x="259.279297" xlink:href="#DejaVuSans-114"/>
+     <use x="300.376953" xlink:href="#DejaVuSans-109"/>
+     <use x="397.789062" xlink:href="#DejaVuSans-97"/>
+     <use x="459.068359" xlink:href="#DejaVuSans-110"/>
+     <use x="522.447266" xlink:href="#DejaVuSans-99"/>
+     <use x="577.427734" xlink:href="#DejaVuSans-101"/>
+     <use x="638.951172" xlink:href="#DejaVuSans-32"/>
+     <use x="670.738281" xlink:href="#DejaVuSans-111"/>
+     <use x="731.919922" xlink:href="#DejaVuSans-110"/>
+     <use x="795.298828" xlink:href="#DejaVuSans-32"/>
+     <use x="827.085938" xlink:href="#DejaVuSans-65"/>
+     <use x="895.416016" xlink:href="#DejaVuSans-87"/>
+     <use x="994.292969" xlink:href="#DejaVuSans-83"/>
+     <use x="1057.769531" xlink:href="#DejaVuSans-32"/>
+     <use x="1089.556641" xlink:href="#DejaVuSans-99"/>
+     <use x="1144.537109" xlink:href="#DejaVuSans-53"/>
+     <use x="1208.160156" xlink:href="#DejaVuSans-46"/>
+     <use x="1239.947266" xlink:href="#DejaVuSans-109"/>
+     <use x="1337.359375" xlink:href="#DejaVuSans-101"/>
+     <use x="1398.882812" xlink:href="#DejaVuSans-116"/>
+     <use x="1438.091797" xlink:href="#DejaVuSans-97"/>
+     <use x="1499.371094" xlink:href="#DejaVuSans-108"/>
+     <use x="1527.154297" xlink:href="#DejaVuSans-44"/>
+     <use x="1558.941406" xlink:href="#DejaVuSans-32"/>
+     <use x="1590.728516" xlink:href="#DejaVuSans-49"/>
+     <use x="1654.351562" xlink:href="#DejaVuSans-54"/>
+     <use x="1717.974609" xlink:href="#DejaVuSans-32"/>
+     <use x="1749.761719" xlink:href="#DejaVuSans-75"/>
+     <use x="1815.337891" xlink:href="#DejaVuSans-105"/>
+     <use x="1843.121094" xlink:href="#DejaVuSans-66"/>
+     <use x="1911.724609" xlink:href="#DejaVuSans-32"/>
+     <use x="1943.511719" xlink:href="#DejaVuSans-105"/>
+     <use x="1971.294922" xlink:href="#DejaVuSans-110"/>
+     <use x="2034.673828" xlink:href="#DejaVuSans-112"/>
+     <use x="2098.150391" xlink:href="#DejaVuSans-117"/>
+     <use x="2161.529297" xlink:href="#DejaVuSans-116"/>
+     <use x="2200.738281" xlink:href="#DejaVuSans-44"/>
+     <use x="2232.525391" xlink:href="#DejaVuSans-32"/>
+     <use x="2264.3125" xlink:href="#DejaVuSans-49"/>
+     <use x="2327.935547" xlink:href="#DejaVuSans-32"/>
+     <use x="2359.722656" xlink:href="#DejaVuSans-116"/>
+     <use x="2398.931641" xlink:href="#DejaVuSans-104"/>
+     <use x="2462.310547" xlink:href="#DejaVuSans-114"/>
+     <use x="2503.392578" xlink:href="#DejaVuSans-101"/>
+     <use x="2564.916016" xlink:href="#DejaVuSans-97"/>
+     <use x="2626.195312" xlink:href="#DejaVuSans-100"/>
+    </g>
+   </g>
+  </g>
+ </g>
+ <defs>
+  <clipPath id="p6091bd3d0f">
+   <rect height="266.112" width="357.12" x="71.443906" y="22.318125"/>
+  </clipPath>
+ </defs>
+</svg>
diff --git a/thirdparty/blake3/reference_impl/Cargo.toml b/thirdparty/blake3/reference_impl/Cargo.toml
new file mode 100644
index 000000000..e269fe9d2
--- /dev/null
+++ b/thirdparty/blake3/reference_impl/Cargo.toml
@@ -0,0 +1,8 @@
+[package]
+name = "reference_impl"
+version = "0.0.0"
+edition = "2021"
+
+[lib]
+name = "reference_impl"
+path = "reference_impl.rs"
diff --git a/thirdparty/blake3/reference_impl/README.md b/thirdparty/blake3/reference_impl/README.md
new file mode 100644
index 000000000..6d123f3f5
--- /dev/null
+++ b/thirdparty/blake3/reference_impl/README.md
@@ -0,0 +1,14 @@
+This is the reference implementation of BLAKE3. It is used for testing and
+as a readable example of the algorithms involved. Section 5.1 of [the BLAKE3
+spec](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf)
+discusses this implementation. You can render docs for this implementation
+by running `cargo doc --open` in this directory.
+
+This implementation is a single file
+([`reference_impl.rs`](reference_impl.rs)) with no dependencies. It is
+not optimized for performance.
+
+There are ports of this reference implementation to other languages:
+
+- [C](https://github.com/oconnor663/blake3_reference_impl_c)
+- [Python](https://github.com/oconnor663/pure_python_blake3)
diff --git a/thirdparty/blake3/reference_impl/reference_impl.rs b/thirdparty/blake3/reference_impl/reference_impl.rs
new file mode 100644
index 000000000..bc6138350
--- /dev/null
+++ b/thirdparty/blake3/reference_impl/reference_impl.rs
@@ -0,0 +1,374 @@
+//! This is the reference implementation of BLAKE3. It is used for testing and
+//! as a readable example of the algorithms involved. Section 5.1 of [the BLAKE3
+//! spec](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf)
+//! discusses this implementation. You can render docs for this implementation
+//! by running `cargo doc --open` in this directory.
+//!
+//! # Example
+//!
+//! ```
+//! let mut hasher = reference_impl::Hasher::new();
+//! hasher.update(b"abc");
+//! hasher.update(b"def");
+//! let mut hash = [0; 32];
+//! hasher.finalize(&mut hash);
+//! let mut extended_hash = [0; 500];
+//! hasher.finalize(&mut extended_hash);
+//! assert_eq!(hash, extended_hash[..32]);
+//! ```
+
+use core::cmp::min;
+
+const OUT_LEN: usize = 32;
+const KEY_LEN: usize = 32;
+const BLOCK_LEN: usize = 64;
+const CHUNK_LEN: usize = 1024;
+
+const CHUNK_START: u32 = 1 << 0;
+const CHUNK_END: u32 = 1 << 1;
+const PARENT: u32 = 1 << 2;
+const ROOT: u32 = 1 << 3;
+const KEYED_HASH: u32 = 1 << 4;
+const DERIVE_KEY_CONTEXT: u32 = 1 << 5;
+const DERIVE_KEY_MATERIAL: u32 = 1 << 6;
+
+const IV: [u32; 8] = [
+    0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A, 0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19,
+];
+
+const MSG_PERMUTATION: [usize; 16] = [2, 6, 3, 10, 7, 0, 4, 13, 1, 11, 12, 5, 9, 14, 15, 8];
+
+// The mixing function, G, which mixes either a column or a diagonal.
+fn g(state: &mut [u32; 16], a: usize, b: usize, c: usize, d: usize, mx: u32, my: u32) {
+    state[a] = state[a].wrapping_add(state[b]).wrapping_add(mx);
+    state[d] = (state[d] ^ state[a]).rotate_right(16);
+    state[c] = state[c].wrapping_add(state[d]);
+    state[b] = (state[b] ^ state[c]).rotate_right(12);
+    state[a] = state[a].wrapping_add(state[b]).wrapping_add(my);
+    state[d] = (state[d] ^ state[a]).rotate_right(8);
+    state[c] = state[c].wrapping_add(state[d]);
+    state[b] = (state[b] ^ state[c]).rotate_right(7);
+}
+
+fn round(state: &mut [u32; 16], m: &[u32; 16]) {
+    // Mix the columns.
+    g(state, 0, 4, 8, 12, m[0], m[1]);
+    g(state, 1, 5, 9, 13, m[2], m[3]);
+    g(state, 2, 6, 10, 14, m[4], m[5]);
+    g(state, 3, 7, 11, 15, m[6], m[7]);
+    // Mix the diagonals.
+    g(state, 0, 5, 10, 15, m[8], m[9]);
+    g(state, 1, 6, 11, 12, m[10], m[11]);
+    g(state, 2, 7, 8, 13, m[12], m[13]);
+    g(state, 3, 4, 9, 14, m[14], m[15]);
+}
+
+fn permute(m: &mut [u32; 16]) {
+    let mut permuted = [0; 16];
+    for i in 0..16 {
+        permuted[i] = m[MSG_PERMUTATION[i]];
+    }
+    *m = permuted;
+}
+
+fn compress(
+    chaining_value: &[u32; 8],
+    block_words: &[u32; 16],
+    counter: u64,
+    block_len: u32,
+    flags: u32,
+) -> [u32; 16] {
+    let counter_low = counter as u32;
+    let counter_high = (counter >> 32) as u32;
+    #[rustfmt::skip]
+    let mut state = [
+        chaining_value[0], chaining_value[1], chaining_value[2], chaining_value[3],
+        chaining_value[4], chaining_value[5], chaining_value[6], chaining_value[7],
+        IV[0],             IV[1],             IV[2],             IV[3],
+        counter_low,       counter_high,      block_len,         flags,
+    ];
+    let mut block = *block_words;
+
+    round(&mut state, &block); // round 1
+    permute(&mut block);
+    round(&mut state, &block); // round 2
+    permute(&mut block);
+    round(&mut state, &block); // round 3
+    permute(&mut block);
+    round(&mut state, &block); // round 4
+    permute(&mut block);
+    round(&mut state, &block); // round 5
+    permute(&mut block);
+    round(&mut state, &block); // round 6
+    permute(&mut block);
+    round(&mut state, &block); // round 7
+
+    for i in 0..8 {
+        state[i] ^= state[i + 8];
+        state[i + 8] ^= chaining_value[i];
+    }
+    state
+}
+
+fn first_8_words(compression_output: [u32; 16]) -> [u32; 8] {
+    compression_output[0..8].try_into().unwrap()
+}
+
+fn words_from_little_endian_bytes(bytes: &[u8], words: &mut [u32]) {
+    debug_assert_eq!(bytes.len(), 4 * words.len());
+    for (four_bytes, word) in bytes.chunks_exact(4).zip(words) {
+        *word = u32::from_le_bytes(four_bytes.try_into().unwrap());
+    }
+}
+
+// Each chunk or parent node can produce either an 8-word chaining value or, by
+// setting the ROOT flag, any number of final output bytes. The Output struct
+// captures the state just prior to choosing between those two possibilities.
+struct Output {
+    input_chaining_value: [u32; 8],
+    block_words: [u32; 16],
+    counter: u64,
+    block_len: u32,
+    flags: u32,
+}
+
+impl Output {
+    fn chaining_value(&self) -> [u32; 8] {
+        first_8_words(compress(
+            &self.input_chaining_value,
+            &self.block_words,
+            self.counter,
+            self.block_len,
+            self.flags,
+        ))
+    }
+
+    fn root_output_bytes(&self, out_slice: &mut [u8]) {
+        let mut output_block_counter = 0;
+        for out_block in out_slice.chunks_mut(2 * OUT_LEN) {
+            let words = compress(
+                &self.input_chaining_value,
+                &self.block_words,
+                output_block_counter,
+                self.block_len,
+                self.flags | ROOT,
+            );
+            // The output length might not be a multiple of 4.
+            for (word, out_word) in words.iter().zip(out_block.chunks_mut(4)) {
+                out_word.copy_from_slice(&word.to_le_bytes()[..out_word.len()]);
+            }
+            output_block_counter += 1;
+        }
+    }
+}
+
+struct ChunkState {
+    chaining_value: [u32; 8],
+    chunk_counter: u64,
+    block: [u8; BLOCK_LEN],
+    block_len: u8,
+    blocks_compressed: u8,
+    flags: u32,
+}
+
+impl ChunkState {
+    fn new(key_words: [u32; 8], chunk_counter: u64, flags: u32) -> Self {
+        Self {
+            chaining_value: key_words,
+            chunk_counter,
+            block: [0; BLOCK_LEN],
+            block_len: 0,
+            blocks_compressed: 0,
+            flags,
+        }
+    }
+
+    fn len(&self) -> usize {
+        BLOCK_LEN * self.blocks_compressed as usize + self.block_len as usize
+    }
+
+    fn start_flag(&self) -> u32 {
+        if self.blocks_compressed == 0 {
+            CHUNK_START
+        } else {
+            0
+        }
+    }
+
+    fn update(&mut self, mut input: &[u8]) {
+        while !input.is_empty() {
+            // If the block buffer is full, compress it and clear it. More
+            // input is coming, so this compression is not CHUNK_END.
+            if self.block_len as usize == BLOCK_LEN {
+                let mut block_words = [0; 16];
+                words_from_little_endian_bytes(&self.block, &mut block_words);
+                self.chaining_value = first_8_words(compress(
+                    &self.chaining_value,
+                    &block_words,
+                    self.chunk_counter,
+                    BLOCK_LEN as u32,
+                    self.flags | self.start_flag(),
+                ));
+                self.blocks_compressed += 1;
+                self.block = [0; BLOCK_LEN];
+                self.block_len = 0;
+            }
+
+            // Copy input bytes into the block buffer.
+            let want = BLOCK_LEN - self.block_len as usize;
+            let take = min(want, input.len());
+            self.block[self.block_len as usize..][..take].copy_from_slice(&input[..take]);
+            self.block_len += take as u8;
+            input = &input[take..];
+        }
+    }
+
+    fn output(&self) -> Output {
+        let mut block_words = [0; 16];
+        words_from_little_endian_bytes(&self.block, &mut block_words);
+        Output {
+            input_chaining_value: self.chaining_value,
+            block_words,
+            counter: self.chunk_counter,
+            block_len: self.block_len as u32,
+            flags: self.flags | self.start_flag() | CHUNK_END,
+        }
+    }
+}
+
+fn parent_output(
+    left_child_cv: [u32; 8],
+    right_child_cv: [u32; 8],
+    key_words: [u32; 8],
+    flags: u32,
+) -> Output {
+    let mut block_words = [0; 16];
+    block_words[..8].copy_from_slice(&left_child_cv);
+    block_words[8..].copy_from_slice(&right_child_cv);
+    Output {
+        input_chaining_value: key_words,
+        block_words,
+        counter: 0,                  // Always 0 for parent nodes.
+        block_len: BLOCK_LEN as u32, // Always BLOCK_LEN (64) for parent nodes.
+        flags: PARENT | flags,
+    }
+}
+
+fn parent_cv(
+    left_child_cv: [u32; 8],
+    right_child_cv: [u32; 8],
+    key_words: [u32; 8],
+    flags: u32,
+) -> [u32; 8] {
+    parent_output(left_child_cv, right_child_cv, key_words, flags).chaining_value()
+}
+
+/// An incremental hasher that can accept any number of writes.
+pub struct Hasher {
+    chunk_state: ChunkState,
+    key_words: [u32; 8],
+    cv_stack: [[u32; 8]; 54], // Space for 54 subtree chaining values:
+    cv_stack_len: u8,         // 2^54 * CHUNK_LEN = 2^64
+    flags: u32,
+}
+
+impl Hasher {
+    fn new_internal(key_words: [u32; 8], flags: u32) -> Self {
+        Self {
+            chunk_state: ChunkState::new(key_words, 0, flags),
+            key_words,
+            cv_stack: [[0; 8]; 54],
+            cv_stack_len: 0,
+            flags,
+        }
+    }
+
+    /// Construct a new `Hasher` for the regular hash function.
+    pub fn new() -> Self {
+        Self::new_internal(IV, 0)
+    }
+
+    /// Construct a new `Hasher` for the keyed hash function.
+    pub fn new_keyed(key: &[u8; KEY_LEN]) -> Self {
+        let mut key_words = [0; 8];
+        words_from_little_endian_bytes(key, &mut key_words);
+        Self::new_internal(key_words, KEYED_HASH)
+    }
+
+    /// Construct a new `Hasher` for the key derivation function. The context
+    /// string should be hardcoded, globally unique, and application-specific.
+    pub fn new_derive_key(context: &str) -> Self {
+        let mut context_hasher = Self::new_internal(IV, DERIVE_KEY_CONTEXT);
+        context_hasher.update(context.as_bytes());
+        let mut context_key = [0; KEY_LEN];
+        context_hasher.finalize(&mut context_key);
+        let mut context_key_words = [0; 8];
+        words_from_little_endian_bytes(&context_key, &mut context_key_words);
+        Self::new_internal(context_key_words, DERIVE_KEY_MATERIAL)
+    }
+
+    fn push_stack(&mut self, cv: [u32; 8]) {
+        self.cv_stack[self.cv_stack_len as usize] = cv;
+        self.cv_stack_len += 1;
+    }
+
+    fn pop_stack(&mut self) -> [u32; 8] {
+        self.cv_stack_len -= 1;
+        self.cv_stack[self.cv_stack_len as usize]
+    }
+
+    // Section 5.1.2 of the BLAKE3 spec explains this algorithm in more detail.
+    fn add_chunk_chaining_value(&mut self, mut new_cv: [u32; 8], mut total_chunks: u64) {
+        // This chunk might complete some subtrees. For each completed subtree,
+        // its left child will be the current top entry in the CV stack, and
+        // its right child will be the current value of `new_cv`. Pop each left
+        // child off the stack, merge it with `new_cv`, and overwrite `new_cv`
+        // with the result. After all these merges, push the final value of
+        // `new_cv` onto the stack. The number of completed subtrees is given
+        // by the number of trailing 0-bits in the new total number of chunks.
+        while total_chunks & 1 == 0 {
+            new_cv = parent_cv(self.pop_stack(), new_cv, self.key_words, self.flags);
+            total_chunks >>= 1;
+        }
+        self.push_stack(new_cv);
+    }
+
+    /// Add input to the hash state. This can be called any number of times.
+    pub fn update(&mut self, mut input: &[u8]) {
+        while !input.is_empty() {
+            // If the current chunk is complete, finalize it and reset the
+            // chunk state. More input is coming, so this chunk is not ROOT.
+            if self.chunk_state.len() == CHUNK_LEN {
+                let chunk_cv = self.chunk_state.output().chaining_value();
+                let total_chunks = self.chunk_state.chunk_counter + 1;
+                self.add_chunk_chaining_value(chunk_cv, total_chunks);
+                self.chunk_state = ChunkState::new(self.key_words, total_chunks, self.flags);
+            }
+
+            // Compress input bytes into the current chunk state.
+            let want = CHUNK_LEN - self.chunk_state.len();
+            let take = min(want, input.len());
+            self.chunk_state.update(&input[..take]);
+            input = &input[take..];
+        }
+    }
+
+    /// Finalize the hash and write any number of output bytes.
+    pub fn finalize(&self, out_slice: &mut [u8]) {
+        // Starting with the Output from the current chunk, compute all the
+        // parent chaining values along the right edge of the tree, until we
+        // have the root Output.
+        let mut output = self.chunk_state.output();
+        let mut parent_nodes_remaining = self.cv_stack_len as usize;
+        while parent_nodes_remaining > 0 {
+            parent_nodes_remaining -= 1;
+            output = parent_output(
+                self.cv_stack[parent_nodes_remaining],
+                output.chaining_value(),
+                self.key_words,
+                self.flags,
+            );
+        }
+        output.root_output_bytes(out_slice);
+    }
+}
diff --git a/thirdparty/blake3/src/ffi_avx2.rs b/thirdparty/blake3/src/ffi_avx2.rs
new file mode 100644
index 000000000..43bf1504a
--- /dev/null
+++ b/thirdparty/blake3/src/ffi_avx2.rs
@@ -0,0 +1,65 @@
+use crate::{CVWords, IncrementCounter, BLOCK_LEN, OUT_LEN};
+
+// Note that there is no AVX2 implementation of compress_in_place or
+// compress_xof.
+
+// Unsafe because this may only be called on platforms supporting AVX2.
+pub unsafe fn hash_many<const N: usize>(
+    inputs: &[&[u8; N]],
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8],
+) {
+    unsafe {
+        // The Rust hash_many implementations do bounds checking on the `out`
+        // array, but the C implementations don't. Even though this is an unsafe
+        // function, assert the bounds here.
+        assert!(out.len() >= inputs.len() * OUT_LEN);
+        ffi::blake3_hash_many_avx2(
+            inputs.as_ptr() as *const *const u8,
+            inputs.len(),
+            N / BLOCK_LEN,
+            key.as_ptr(),
+            counter,
+            increment_counter.yes(),
+            flags,
+            flags_start,
+            flags_end,
+            out.as_mut_ptr(),
+        )
+    }
+}
+
+pub mod ffi {
+    extern "C" {
+        pub fn blake3_hash_many_avx2(
+            inputs: *const *const u8,
+            num_inputs: usize,
+            blocks: usize,
+            key: *const u32,
+            counter: u64,
+            increment_counter: bool,
+            flags: u8,
+            flags_start: u8,
+            flags_end: u8,
+            out: *mut u8,
+        );
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_hash_many() {
+        if !crate::platform::avx2_detected() {
+            return;
+        }
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/ffi_avx512.rs b/thirdparty/blake3/src/ffi_avx512.rs
new file mode 100644
index 000000000..e648edaf2
--- /dev/null
+++ b/thirdparty/blake3/src/ffi_avx512.rs
@@ -0,0 +1,169 @@
+use crate::{CVWords, IncrementCounter, BLOCK_LEN, OUT_LEN};
+
+// Unsafe because this may only be called on platforms supporting AVX-512.
+pub unsafe fn compress_in_place(
+    cv: &mut CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    unsafe {
+        ffi::blake3_compress_in_place_avx512(
+            cv.as_mut_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+        )
+    }
+}
+
+// Unsafe because this may only be called on platforms supporting AVX-512.
+pub unsafe fn compress_xof(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64] {
+    unsafe {
+        let mut out = [0u8; 64];
+        ffi::blake3_compress_xof_avx512(
+            cv.as_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+            out.as_mut_ptr(),
+        );
+        out
+    }
+}
+
+// Unsafe because this may only be called on platforms supporting AVX-512.
+pub unsafe fn hash_many<const N: usize>(
+    inputs: &[&[u8; N]],
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8],
+) {
+    unsafe {
+        // The Rust hash_many implementations do bounds checking on the `out`
+        // array, but the C implementations don't. Even though this is an unsafe
+        // function, assert the bounds here.
+        assert!(out.len() >= inputs.len() * OUT_LEN);
+        ffi::blake3_hash_many_avx512(
+            inputs.as_ptr() as *const *const u8,
+            inputs.len(),
+            N / BLOCK_LEN,
+            key.as_ptr(),
+            counter,
+            increment_counter.yes(),
+            flags,
+            flags_start,
+            flags_end,
+            out.as_mut_ptr(),
+        )
+    }
+}
+
+// Unsafe because this may only be called on platforms supporting AVX-512.
+#[cfg(unix)]
+pub unsafe fn xof_many(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+    out: &mut [u8],
+) {
+    unsafe {
+        debug_assert_eq!(0, out.len() % BLOCK_LEN, "whole blocks only");
+        ffi::blake3_xof_many_avx512(
+            cv.as_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+            out.as_mut_ptr(),
+            out.len() / BLOCK_LEN,
+        );
+    }
+}
+
+pub mod ffi {
+    extern "C" {
+        pub fn blake3_compress_in_place_avx512(
+            cv: *mut u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+        );
+        pub fn blake3_compress_xof_avx512(
+            cv: *const u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+            out: *mut u8,
+        );
+        pub fn blake3_hash_many_avx512(
+            inputs: *const *const u8,
+            num_inputs: usize,
+            blocks: usize,
+            key: *const u32,
+            counter: u64,
+            increment_counter: bool,
+            flags: u8,
+            flags_start: u8,
+            flags_end: u8,
+            out: *mut u8,
+        );
+        #[cfg(unix)]
+        pub fn blake3_xof_many_avx512(
+            cv: *const u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+            out: *mut u8,
+            outblocks: usize,
+        );
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_compress() {
+        if !crate::platform::avx512_detected() {
+            return;
+        }
+        crate::test::test_compress_fn(compress_in_place, compress_xof);
+    }
+
+    #[test]
+    fn test_hash_many() {
+        if !crate::platform::avx512_detected() {
+            return;
+        }
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+
+    #[cfg(unix)]
+    #[test]
+    fn test_xof_many() {
+        if !crate::platform::avx512_detected() {
+            return;
+        }
+        crate::test::test_xof_many_fn(xof_many);
+    }
+}
diff --git a/thirdparty/blake3/src/ffi_neon.rs b/thirdparty/blake3/src/ffi_neon.rs
new file mode 100644
index 000000000..54d07a4de
--- /dev/null
+++ b/thirdparty/blake3/src/ffi_neon.rs
@@ -0,0 +1,82 @@
+use crate::{CVWords, IncrementCounter, BLOCK_LEN, OUT_LEN};
+
+// Unsafe because this may only be called on platforms supporting NEON.
+pub unsafe fn hash_many<const N: usize>(
+    inputs: &[&[u8; N]],
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8],
+) {
+    // The Rust hash_many implementations do bounds checking on the `out`
+    // array, but the C implementations don't. Even though this is an unsafe
+    // function, assert the bounds here.
+    assert!(out.len() >= inputs.len() * OUT_LEN);
+    ffi::blake3_hash_many_neon(
+        inputs.as_ptr() as *const *const u8,
+        inputs.len(),
+        N / BLOCK_LEN,
+        key.as_ptr(),
+        counter,
+        increment_counter.yes(),
+        flags,
+        flags_start,
+        flags_end,
+        out.as_mut_ptr(),
+    )
+}
+
+// blake3_neon.c normally depends on blake3_portable.c, because the NEON
+// implementation only provides 4x compression, and it relies on the portable
+// implementation for 1x compression. However, we expose the portable Rust
+// implementation here instead, to avoid linking in unnecessary code.
+#[no_mangle]
+pub extern "C" fn blake3_compress_in_place_portable(
+    cv: *mut u32,
+    block: *const u8,
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    unsafe {
+        crate::portable::compress_in_place(
+            &mut *(cv as *mut [u32; 8]),
+            &*(block as *const [u8; 64]),
+            block_len,
+            counter,
+            flags,
+        )
+    }
+}
+
+pub mod ffi {
+    extern "C" {
+        pub fn blake3_hash_many_neon(
+            inputs: *const *const u8,
+            num_inputs: usize,
+            blocks: usize,
+            key: *const u32,
+            counter: u64,
+            increment_counter: bool,
+            flags: u8,
+            flags_start: u8,
+            flags_end: u8,
+            out: *mut u8,
+        );
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_hash_many() {
+        // This entire file is gated on feature="neon", so NEON support is
+        // assumed here.
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/ffi_sse2.rs b/thirdparty/blake3/src/ffi_sse2.rs
new file mode 100644
index 000000000..8dafd6983
--- /dev/null
+++ b/thirdparty/blake3/src/ffi_sse2.rs
@@ -0,0 +1,126 @@
+use crate::{CVWords, IncrementCounter, BLOCK_LEN, OUT_LEN};
+
+// Unsafe because this may only be called on platforms supporting SSE2.
+pub unsafe fn compress_in_place(
+    cv: &mut CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    unsafe {
+        ffi::blake3_compress_in_place_sse2(
+            cv.as_mut_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+        )
+    }
+}
+
+// Unsafe because this may only be called on platforms supporting SSE2.
+pub unsafe fn compress_xof(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64] {
+    unsafe {
+        let mut out = [0u8; 64];
+        ffi::blake3_compress_xof_sse2(
+            cv.as_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+            out.as_mut_ptr(),
+        );
+        out
+    }
+}
+
+// Unsafe because this may only be called on platforms supporting SSE2.
+pub unsafe fn hash_many<const N: usize>(
+    inputs: &[&[u8; N]],
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8],
+) {
+    unsafe {
+        // The Rust hash_many implementations do bounds checking on the `out`
+        // array, but the C implementations don't. Even though this is an unsafe
+        // function, assert the bounds here.
+        assert!(out.len() >= inputs.len() * OUT_LEN);
+        ffi::blake3_hash_many_sse2(
+            inputs.as_ptr() as *const *const u8,
+            inputs.len(),
+            N / BLOCK_LEN,
+            key.as_ptr(),
+            counter,
+            increment_counter.yes(),
+            flags,
+            flags_start,
+            flags_end,
+            out.as_mut_ptr(),
+        )
+    }
+}
+
+pub mod ffi {
+    extern "C" {
+        pub fn blake3_compress_in_place_sse2(
+            cv: *mut u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+        );
+        pub fn blake3_compress_xof_sse2(
+            cv: *const u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+            out: *mut u8,
+        );
+        pub fn blake3_hash_many_sse2(
+            inputs: *const *const u8,
+            num_inputs: usize,
+            blocks: usize,
+            key: *const u32,
+            counter: u64,
+            increment_counter: bool,
+            flags: u8,
+            flags_start: u8,
+            flags_end: u8,
+            out: *mut u8,
+        );
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_compress() {
+        if !crate::platform::sse2_detected() {
+            return;
+        }
+        crate::test::test_compress_fn(compress_in_place, compress_xof);
+    }
+
+    #[test]
+    fn test_hash_many() {
+        if !crate::platform::sse2_detected() {
+            return;
+        }
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/ffi_sse41.rs b/thirdparty/blake3/src/ffi_sse41.rs
new file mode 100644
index 000000000..f851ca153
--- /dev/null
+++ b/thirdparty/blake3/src/ffi_sse41.rs
@@ -0,0 +1,126 @@
+use crate::{CVWords, IncrementCounter, BLOCK_LEN, OUT_LEN};
+
+// Unsafe because this may only be called on platforms supporting SSE4.1.
+pub unsafe fn compress_in_place(
+    cv: &mut CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    unsafe {
+        ffi::blake3_compress_in_place_sse41(
+            cv.as_mut_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+        )
+    }
+}
+
+// Unsafe because this may only be called on platforms supporting SSE4.1.
+pub unsafe fn compress_xof(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64] {
+    unsafe {
+        let mut out = [0u8; 64];
+        ffi::blake3_compress_xof_sse41(
+            cv.as_ptr(),
+            block.as_ptr(),
+            block_len,
+            counter,
+            flags,
+            out.as_mut_ptr(),
+        );
+        out
+    }
+}
+
+// Unsafe because this may only be called on platforms supporting SSE4.1.
+pub unsafe fn hash_many<const N: usize>(
+    inputs: &[&[u8; N]],
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8],
+) {
+    unsafe {
+        // The Rust hash_many implementations do bounds checking on the `out`
+        // array, but the C implementations don't. Even though this is an unsafe
+        // function, assert the bounds here.
+        assert!(out.len() >= inputs.len() * OUT_LEN);
+        ffi::blake3_hash_many_sse41(
+            inputs.as_ptr() as *const *const u8,
+            inputs.len(),
+            N / BLOCK_LEN,
+            key.as_ptr(),
+            counter,
+            increment_counter.yes(),
+            flags,
+            flags_start,
+            flags_end,
+            out.as_mut_ptr(),
+        )
+    }
+}
+
+pub mod ffi {
+    extern "C" {
+        pub fn blake3_compress_in_place_sse41(
+            cv: *mut u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+        );
+        pub fn blake3_compress_xof_sse41(
+            cv: *const u32,
+            block: *const u8,
+            block_len: u8,
+            counter: u64,
+            flags: u8,
+            out: *mut u8,
+        );
+        pub fn blake3_hash_many_sse41(
+            inputs: *const *const u8,
+            num_inputs: usize,
+            blocks: usize,
+            key: *const u32,
+            counter: u64,
+            increment_counter: bool,
+            flags: u8,
+            flags_start: u8,
+            flags_end: u8,
+            out: *mut u8,
+        );
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_compress() {
+        if !crate::platform::sse41_detected() {
+            return;
+        }
+        crate::test::test_compress_fn(compress_in_place, compress_xof);
+    }
+
+    #[test]
+    fn test_hash_many() {
+        if !crate::platform::sse41_detected() {
+            return;
+        }
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/guts.rs b/thirdparty/blake3/src/guts.rs
new file mode 100644
index 000000000..6bbf5a571
--- /dev/null
+++ b/thirdparty/blake3/src/guts.rs
@@ -0,0 +1,60 @@
+//! Deprecated in favor of [`hazmat`](crate::hazmat)
+
+pub use crate::{BLOCK_LEN, CHUNK_LEN};
+
+#[derive(Clone, Debug)]
+pub struct ChunkState(crate::ChunkState);
+
+impl ChunkState {
+    // Currently this type only supports the regular hash mode. If an
+    // incremental user needs keyed_hash or derive_key, we can add that.
+    pub fn new(chunk_counter: u64) -> Self {
+        Self(crate::ChunkState::new(
+            crate::IV,
+            chunk_counter,
+            0,
+            crate::platform::Platform::detect(),
+        ))
+    }
+
+    #[inline]
+    pub fn len(&self) -> usize {
+        self.0.count()
+    }
+
+    #[inline]
+    pub fn update(&mut self, input: &[u8]) -> &mut Self {
+        self.0.update(input);
+        self
+    }
+
+    pub fn finalize(&self, is_root: bool) -> crate::Hash {
+        let output = self.0.output();
+        if is_root {
+            output.root_hash()
+        } else {
+            output.chaining_value().into()
+        }
+    }
+}
+
+// As above, this currently assumes the regular hash mode. If an incremental
+// user needs keyed_hash or derive_key, we can add that.
+pub fn parent_cv(
+    left_child: &crate::Hash,
+    right_child: &crate::Hash,
+    is_root: bool,
+) -> crate::Hash {
+    let output = crate::parent_node_output(
+        left_child.as_bytes(),
+        right_child.as_bytes(),
+        crate::IV,
+        0,
+        crate::platform::Platform::detect(),
+    );
+    if is_root {
+        output.root_hash()
+    } else {
+        output.chaining_value().into()
+    }
+}
diff --git a/thirdparty/blake3/src/hazmat.rs b/thirdparty/blake3/src/hazmat.rs
new file mode 100644
index 000000000..2fd2449db
--- /dev/null
+++ b/thirdparty/blake3/src/hazmat.rs
@@ -0,0 +1,704 @@
+//! Low-level tree manipulations and other sharp tools
+//!
+//! The target audience for this module is projects like [Bao](https://github.com/oconnor663/bao),
+//! which work directly with the interior hashes ("chaining values") of BLAKE3 chunks and subtrees.
+//! For example, you could use these functions to implement a BitTorrent-like protocol using the
+//! BLAKE3 tree structure, or to hash an input that's distributed across different machines. These
+//! use cases are advanced, and most applications don't need this module. Also:
+//!
+//! <div class="warning">
+//!
+//! **Warning:** This module is *hazardous material*. If you've heard folks say *don't roll your
+//! own crypto,* this is the sort of thing they're talking about. These functions have complicated
+//! requirements, and any mistakes will give you garbage output and/or break the security
+//! properties that BLAKE3 is supposed to have. Read section 2.1 of [the BLAKE3
+//! paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf) to understand the
+//! tree structure you need to maintain. Test your code against [`blake3::hash`](../fn.hash.html)
+//! and make sure you can get the same outputs for [lots of different
+//! inputs](https://github.com/BLAKE3-team/BLAKE3/blob/master/test_vectors/test_vectors.json).
+//!
+//! </div>
+//!
+//! On the other hand:
+//!
+//! <div class="warning">
+//!
+//! **Encouragement:** Playing with these functions is a great way to learn how BLAKE3 works on the
+//! inside. Have fun!
+//!
+//! </div>
+//!
+//! The main entrypoint for this module is the [`HasherExt`] trait, particularly the
+//! [`set_input_offset`](HasherExt::set_input_offset) and
+//! [`finalize_non_root`](HasherExt::finalize_non_root) methods. These let you compute the chaining
+//! values of individual chunks or subtrees. You then combine these chaining values into larger
+//! subtrees using [`merge_subtrees_non_root`] and finally (once at the very top)
+//! [`merge_subtrees_root`] or [`merge_subtrees_root_xof`].
+//!
+//! # Examples
+//!
+//! Here's an example of computing all the interior hashes in a 3-chunk tree:
+//!
+//! ```text
+//!            root
+//!          /      \
+//!      parent      \
+//!    /       \      \
+//! chunk0  chunk1  chunk2
+//! ```
+//!
+//! ```
+//! # fn main() {
+//! use blake3::{Hasher, CHUNK_LEN};
+//! use blake3::hazmat::{merge_subtrees_non_root, merge_subtrees_root, Mode};
+//! use blake3::hazmat::HasherExt; // an extension trait for Hasher
+//!
+//! let chunk0 = [b'a'; CHUNK_LEN];
+//! let chunk1 = [b'b'; CHUNK_LEN];
+//! let chunk2 = [b'c'; 42]; // The final chunk can be short.
+//!
+//! // Compute the non-root hashes ("chaining values") of all three chunks. Chunks or subtrees
+//! // that don't begin at the start of the input use `set_input_offset` to say where they begin.
+//! let chunk0_cv = Hasher::new()
+//!     // .set_input_offset(0) is the default.
+//!     .update(&chunk0)
+//!     .finalize_non_root();
+//! let chunk1_cv = Hasher::new()
+//!     .set_input_offset(CHUNK_LEN as u64)
+//!     .update(&chunk1)
+//!     .finalize_non_root();
+//! let chunk2_cv = Hasher::new()
+//!     .set_input_offset(2 * CHUNK_LEN as u64)
+//!     .update(&chunk2)
+//!     .finalize_non_root();
+//!
+//! // Join the first two chunks with a non-root parent node and compute its chaining value.
+//! let parent_cv = merge_subtrees_non_root(&chunk0_cv, &chunk1_cv, Mode::Hash);
+//!
+//! // Join that parent node and the third chunk with a root parent node and compute the hash.
+//! let root_hash = merge_subtrees_root(&parent_cv, &chunk2_cv, Mode::Hash);
+//!
+//! // Double check that we got the right answer.
+//! let mut combined_input = Vec::new();
+//! combined_input.extend_from_slice(&chunk0);
+//! combined_input.extend_from_slice(&chunk1);
+//! combined_input.extend_from_slice(&chunk2);
+//! assert_eq!(root_hash, blake3::hash(&combined_input));
+//! # }
+//! ```
+//!
+//! Hashing many chunks together is important for performance, because it allows the implementation
+//! to use SIMD parallelism internally. ([AVX-512](https://en.wikipedia.org/wiki/AVX-512) for
+//! example needs 16 chunks to really get going.) We can reproduce `parent_cv` by hashing `chunk0`
+//! and `chunk1` at the same time:
+//!
+//! ```
+//! # fn main() {
+//! # use blake3::{Hasher, CHUNK_LEN};
+//! # use blake3::hazmat::{Mode, HasherExt, merge_subtrees_non_root, merge_subtrees_root};
+//! # let chunk0 = [b'a'; CHUNK_LEN];
+//! # let chunk1 = [b'b'; CHUNK_LEN];
+//! # let chunk0_cv = Hasher::new().update(&chunk0).finalize_non_root();
+//! # let chunk1_cv = Hasher::new().set_input_offset(CHUNK_LEN as u64).update(&chunk1).finalize_non_root();
+//! # let parent_cv = merge_subtrees_non_root(&chunk0_cv, &chunk1_cv, Mode::Hash);
+//! # let mut combined_input = Vec::new();
+//! # combined_input.extend_from_slice(&chunk0);
+//! # combined_input.extend_from_slice(&chunk1);
+//! let left_subtree_cv = Hasher::new()
+//!     // .set_input_offset(0) is the default.
+//!     .update(&combined_input[..2 * CHUNK_LEN])
+//!     .finalize_non_root();
+//! assert_eq!(left_subtree_cv, parent_cv);
+//!
+//! // Using multiple updates gives the same answer, though it's not as efficient.
+//! let mut subtree_hasher = Hasher::new();
+//! // Again, .set_input_offset(0) is the default.
+//! subtree_hasher.update(&chunk0);
+//! subtree_hasher.update(&chunk1);
+//! assert_eq!(left_subtree_cv, subtree_hasher.finalize_non_root());
+//! # }
+//! ```
+//!
+//! However, hashing multiple chunks together **must** respect the overall tree structure. Hashing
+//! `chunk0` and `chunk1` together is valid, but hashing `chunk1` and `chunk2` together is
+//! incorrect and gives a garbage result that will never match a standard BLAKE3 hash. The
+//! implementation includes a few best-effort asserts to catch some of these mistakes, but these
+//! checks aren't guaranteed. For example, this second call to `update` currently panics:
+//!
+//! ```should_panic
+//! # fn main() {
+//! # use blake3::{Hasher, CHUNK_LEN};
+//! # use blake3::hazmat::HasherExt;
+//! # let chunk0 = [b'a'; CHUNK_LEN];
+//! # let chunk1 = [b'b'; CHUNK_LEN];
+//! # let chunk2 = [b'c'; 42];
+//! let oops = Hasher::new()
+//!     .set_input_offset(CHUNK_LEN as u64)
+//!     .update(&chunk1)
+//!     // PANIC: "the subtree starting at 1024 contains at most 1024 bytes"
+//!     .update(&chunk2)
+//!     .finalize_non_root();
+//! # }
+//! ```
+//!
+//! For more on valid tree structures, see the docs for and [`left_subtree_len`] and
+//! [`max_subtree_len`], and see section 2.1 of [the BLAKE3
+//! paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf). Note that the
+//! merging functions ([`merge_subtrees_root`] and friends) don't know the shape of the left and
+//! right subtrees you're giving them, and they can't help you catch mistakes. The best way to
+//! catch mistakes with these is to compare your root output to the [`blake3::hash`](crate::hash)
+//! of the same input.
+
+use crate::platform::Platform;
+use crate::{CVWords, Hasher, CHUNK_LEN, IV, KEY_LEN, OUT_LEN};
+
+/// Extension methods for [`Hasher`]. This is the main entrypoint to the `hazmat` module.
+pub trait HasherExt {
+    /// Similar to [`Hasher::new_derive_key`] but using a pre-hashed [`ContextKey`] from
+    /// [`hash_derive_key_context`].
+    ///
+    /// The [`hash_derive_key_context`] function is _only_ valid source of the [`ContextKey`]
+    ///
+    /// # Example
+    ///
+    /// ```
+    /// use blake3::Hasher;
+    /// use blake3::hazmat::HasherExt;
+    ///
+    /// let context_key = blake3::hazmat::hash_derive_key_context("foo");
+    /// let mut hasher = Hasher::new_from_context_key(&context_key);
+    /// hasher.update(b"bar");
+    /// let derived_key = *hasher.finalize().as_bytes();
+    ///
+    /// assert_eq!(derived_key, blake3::derive_key("foo", b"bar"));
+    /// ```
+    fn new_from_context_key(context_key: &ContextKey) -> Self;
+
+    /// Configure the `Hasher` to process a chunk or subtree starting at `offset` bytes into the
+    /// whole input.
+    ///
+    /// You must call this function before processing any input with [`update`](Hasher::update) or
+    /// similar. This step isn't required for the first chunk, or for a subtree that includes the
+    /// first chunk (i.e. when the `offset` is zero), but it's required for all other chunks and
+    /// subtrees.
+    ///
+    /// The starting input offset of a subtree implies a maximum possible length for that subtree.
+    /// See [`max_subtree_len`] and section 2.1 of [the BLAKE3
+    /// paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf). Note that only
+    /// subtrees along the right edge of the whole tree can have a length less than their maximum
+    /// possible length.
+    ///
+    /// See the [module level examples](index.html#examples).
+    ///
+    /// # Panics
+    ///
+    /// This function panics if the `Hasher` has already accepted any input with
+    /// [`update`](Hasher::update) or similar.
+    ///
+    /// This should always be paired with [`finalize_non_root`](HasherExt::finalize_non_root). It's
+    /// never correct to use a non-zero input offset with [`finalize`](Hasher::finalize) or
+    /// [`finalize_xof`](Hasher::finalize_xof). The `offset` must also be a multiple of
+    /// `CHUNK_LEN`. Violating either of these rules will currently fail an assertion and panic,
+    /// but this is not guaranteed.
+    fn set_input_offset(&mut self, offset: u64) -> &mut Self;
+
+    /// Finalize the non-root hash ("chaining value") of the current chunk or subtree.
+    ///
+    /// Afterwards you can merge subtree chaining values into parent nodes using
+    /// [`merge_subtrees_non_root`] and ultimately into the root node with either
+    /// [`merge_subtrees_root`] (similar to [`Hasher::finalize`]) or [`merge_subtrees_root_xof`]
+    /// (similar to [`Hasher::finalize_xof`]).
+    ///
+    /// See the [module level examples](index.html#examples), particularly the discussion of valid
+    /// tree structures.
+    fn finalize_non_root(&self) -> ChainingValue;
+}
+
+impl HasherExt for Hasher {
+    fn new_from_context_key(context_key: &[u8; KEY_LEN]) -> Hasher {
+        let context_key_words = crate::platform::words_from_le_bytes_32(context_key);
+        Hasher::new_internal(&context_key_words, crate::DERIVE_KEY_MATERIAL)
+    }
+
+    fn set_input_offset(&mut self, offset: u64) -> &mut Hasher {
+        assert_eq!(self.count(), 0, "hasher has already accepted input");
+        assert_eq!(
+            offset % CHUNK_LEN as u64,
+            0,
+            "offset ({offset}) must be a chunk boundary (divisible by {CHUNK_LEN})",
+        );
+        let counter = offset / CHUNK_LEN as u64;
+        self.chunk_state.chunk_counter = counter;
+        self.initial_chunk_counter = counter;
+        self
+    }
+
+    fn finalize_non_root(&self) -> ChainingValue {
+        assert_ne!(self.count(), 0, "empty subtrees are never valid");
+        self.final_output().chaining_value()
+    }
+}
+
+/// The maximum length of a subtree in bytes, given its starting offset in bytes
+///
+/// If you try to hash more than this many bytes as one subtree, you'll end up merging parent nodes
+/// that shouldn't be merged, and your output will be garbage. [`Hasher::update`] will currently
+/// panic in this case, but this is not guaranteed.
+///
+/// For input offset zero (the default), there is no maximum length, and this function returns
+/// `None`. For all other offsets it returns `Some`. Note that valid offsets must be a multiple of
+/// [`CHUNK_LEN`] (1024); it's not possible to start hashing a chunk in the middle.
+///
+/// In the example tree below, chunks are numbered by their _0-based index_. The subtree that
+/// _starts_ with chunk 3, i.e. `input_offset = 3 * CHUNK_LEN`, includes only that one chunk, so
+/// its max length is `Some(CHUNK_LEN)`. The subtree that starts with chunk 6 includes chunk 7 but
+/// not chunk 8, so its max length is `Some(2 * CHUNK_LEN)`. The subtree that starts with chunk 12
+/// includes chunks 13, 14, and 15, but if the tree were bigger it would not include chunk 16, so
+/// its max length is `Some(4 * CHUNK_LEN)`. One way to think about the rule here is that, if you
+/// go beyond the max subtree length from a given starting offset, you start dealing with subtrees
+/// that include chunks _to the left_ of where you started.
+///
+/// ```text
+///                           root
+///                 /                       \
+///              .                             .
+///        /           \                 /           \
+///       .             .               .             .
+///    /    \         /    \         /    \         /    \
+///   .      .       .      .       .      .       .      .
+///  / \    / \     / \    / \     / \    / \     / \    / \
+/// 0  1   2  3    4  5   6  7    8  9   10 11   12 13  14 15
+/// ```
+///
+/// The general rule turns out to be that for a subtree starting at a 0-based chunk index N greater
+/// than zero, the maximum number of chunks in that subtree is the largest power-of-two that
+/// divides N, which is given by `1 << N.trailing_zeros()`.
+///
+/// This function can be useful for writing tests or debug assertions, but it's actually rare to
+/// use this for real control flow. Callers who split their input recursively using
+/// [`left_subtree_len`] will automatically satisfy the `max_subtree_len` bound and don't
+/// necessarily need to check. It's also common to choose some fixed power-of-two subtree size, say
+/// 64 chunks, and divide your input up into slices of that fixed length (with the final slice
+/// possibly short). This approach also automatically satisfies the `max_subtree_len` bound and
+/// doesn't need to check. Proving that this is true can be an interesting exercise. Note that
+/// chunks 0, 4, 8, and 12 all begin subtrees of at least 4 chunks in the example tree above.
+///
+/// # Panics
+///
+/// This function currently panics if `input_offset` is not a multiple of `CHUNK_LEN`. This is not
+/// guaranteed.
+#[inline(always)]
+pub fn max_subtree_len(input_offset: u64) -> Option<u64> {
+    if input_offset == 0 {
+        return None;
+    }
+    assert_eq!(input_offset % CHUNK_LEN as u64, 0);
+    let counter = input_offset / CHUNK_LEN as u64;
+    let max_chunks = 1 << counter.trailing_zeros();
+    Some(max_chunks * CHUNK_LEN as u64)
+}
+
+#[test]
+fn test_max_subtree_len() {
+    assert_eq!(max_subtree_len(0), None);
+    // (chunk index, max chunks)
+    let cases = [
+        (1, 1),
+        (2, 2),
+        (3, 1),
+        (4, 4),
+        (5, 1),
+        (6, 2),
+        (7, 1),
+        (8, 8),
+    ];
+    for (chunk_index, max_chunks) in cases {
+        let input_offset = chunk_index * CHUNK_LEN as u64;
+        assert_eq!(
+            max_subtree_len(input_offset),
+            Some(max_chunks * CHUNK_LEN as u64),
+        );
+    }
+}
+
+/// Given the length in bytes of either a complete input or a subtree input, return the number of
+/// bytes that belong to its left child subtree. The rest belong to its right child subtree.
+///
+/// Concretely, this function returns the largest power-of-two number of bytes that's strictly less
+/// than `input_len`. This leads to a tree where all left subtrees are "complete" and at least as
+/// large as their sibling right subtrees, as specified in section 2.1 of [the BLAKE3
+/// paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf). For example, if an
+/// input is exactly two chunks, its left and right subtrees both get one chunk. But if an input is
+/// two chunks plus one more byte, then its left subtree gets two chunks, and its right subtree
+/// only gets one byte.
+///
+/// This function isn't meaningful for one chunk of input, because chunks don't have children. It
+/// currently panics in debug mode if `input_len <= CHUNK_LEN`.
+///
+/// # Example
+///
+/// Hash a input of random length as two subtrees:
+///
+/// ```
+/// # #[cfg(feature = "std")] {
+/// use blake3::hazmat::{left_subtree_len, merge_subtrees_root, HasherExt, Mode};
+/// use blake3::{Hasher, CHUNK_LEN};
+///
+/// // Generate a random-length input. Note that to be split into two subtrees, the input length
+/// // must be greater than CHUNK_LEN.
+/// let input_len = rand::random_range(CHUNK_LEN + 1..1_000_000);
+/// let mut input = vec![0; input_len];
+/// rand::fill(&mut input[..]);
+///
+/// // Compute the left and right subtree hashes and then the root hash. left_subtree_len() tells
+/// // us exactly where to split the input. Any other split would either panic (if we're lucky) or
+/// // lead to an incorrect root hash.
+/// let left_len = left_subtree_len(input_len as u64) as usize;
+/// let left_subtree_cv = Hasher::new()
+///     .update(&input[..left_len])
+///     .finalize_non_root();
+/// let right_subtree_cv = Hasher::new()
+///     .set_input_offset(left_len as u64)
+///     .update(&input[left_len..])
+///     .finalize_non_root();
+/// let root_hash = merge_subtrees_root(&left_subtree_cv, &right_subtree_cv, Mode::Hash);
+///
+/// // Double check the answer.
+/// assert_eq!(root_hash, blake3::hash(&input));
+/// # }
+/// ```
+#[inline(always)]
+pub fn left_subtree_len(input_len: u64) -> u64 {
+    debug_assert!(input_len > CHUNK_LEN as u64);
+    // Note that .next_power_of_two() is greater than *or equal*.
+    ((input_len + 1) / 2).next_power_of_two()
+}
+
+#[test]
+fn test_left_subtree_len() {
+    assert_eq!(left_subtree_len(1025), 1024);
+    for boundary_case in [2, 4, 8, 16, 32, 64] {
+        let input_len = boundary_case * CHUNK_LEN as u64;
+        assert_eq!(left_subtree_len(input_len - 1), input_len / 2);
+        assert_eq!(left_subtree_len(input_len), input_len / 2);
+        assert_eq!(left_subtree_len(input_len + 1), input_len);
+    }
+}
+
+/// The `mode` argument to [`merge_subtrees_root`] and friends
+///
+/// See the [module level examples](index.html#examples).
+#[derive(Copy, Clone, Debug)]
+pub enum Mode<'a> {
+    /// Corresponding to [`hash`](crate::hash)
+    Hash,
+
+    /// Corresponding to [`keyed_hash`](crate::hash)
+    KeyedHash(&'a [u8; KEY_LEN]),
+
+    /// Corresponding to [`derive_key`](crate::hash)
+    ///
+    /// The [`ContextKey`] comes from [`hash_derive_key_context`].
+    DeriveKeyMaterial(&'a ContextKey),
+}
+
+impl<'a> Mode<'a> {
+    fn key_words(&self) -> CVWords {
+        match self {
+            Mode::Hash => *IV,
+            Mode::KeyedHash(key) => crate::platform::words_from_le_bytes_32(key),
+            Mode::DeriveKeyMaterial(cx_key) => crate::platform::words_from_le_bytes_32(cx_key),
+        }
+    }
+
+    fn flags_byte(&self) -> u8 {
+        match self {
+            Mode::Hash => 0,
+            Mode::KeyedHash(_) => crate::KEYED_HASH,
+            Mode::DeriveKeyMaterial(_) => crate::DERIVE_KEY_MATERIAL,
+        }
+    }
+}
+
+/// "Chaining value" is the academic term for a non-root or non-final hash.
+///
+/// Besides just sounding fancy, it turns out there are [security
+/// reasons](https://jacko.io/tree_hashing.html) to be careful about the difference between
+/// (root/final) hashes and (non-root/non-final) chaining values.
+pub type ChainingValue = [u8; OUT_LEN];
+
+fn merge_subtrees_inner(
+    left_child: &ChainingValue,
+    right_child: &ChainingValue,
+    mode: Mode,
+) -> crate::Output {
+    crate::parent_node_output(
+        &left_child,
+        &right_child,
+        &mode.key_words(),
+        mode.flags_byte(),
+        Platform::detect(),
+    )
+}
+
+/// Compute a non-root parent node chaining value from two child chaining values.
+///
+/// See the [module level examples](index.html#examples), particularly the discussion of valid tree
+/// structures. The left and right child chaining values can come from either
+/// [`Hasher::finalize_non_root`](HasherExt::finalize_non_root) or other calls to
+/// `merge_subtrees_non_root`. "Chaining value" is the academic term for a non-root or non-final
+/// hash.
+pub fn merge_subtrees_non_root(
+    left_child: &ChainingValue,
+    right_child: &ChainingValue,
+    mode: Mode,
+) -> ChainingValue {
+    merge_subtrees_inner(left_child, right_child, mode).chaining_value()
+}
+
+/// Compute a root hash from two child chaining values.
+///
+/// See the [module level examples](index.html#examples), particularly the discussion of valid tree
+/// structures. The left and right child chaining values can come from either
+/// [`Hasher::finalize_non_root`](HasherExt::finalize_non_root) or [`merge_subtrees_non_root`].
+/// "Chaining value" is the academic term for a non-root or non-final hash.
+///
+/// Note that inputs of [`CHUNK_LEN`] or less don't produce any parent nodes and can't be hashed
+/// using this function. In that case you must get the root hash from [`Hasher::finalize`] (or just
+/// [`blake3::hash`](crate::hash)).
+pub fn merge_subtrees_root(
+    left_child: &ChainingValue,
+    right_child: &ChainingValue,
+    mode: Mode,
+) -> crate::Hash {
+    merge_subtrees_inner(left_child, right_child, mode).root_hash()
+}
+
+/// Build a root [`OutputReader`](crate::OutputReader) from two child chaining values.
+///
+/// See also the [module level examples](index.html#examples), particularly the discussion of valid
+/// tree structures. The left and right child chaining values can come from either
+/// [`Hasher::finalize_non_root`](HasherExt::finalize_non_root) or [`merge_subtrees_non_root`].
+/// "Chaining value" is the academic term for a non-root or non-final hash.
+///
+/// Note that inputs of [`CHUNK_LEN`] or less don't produce any parent nodes and can't be hashed
+/// using this function. In that case you must get the `OutputReader` from
+/// [`Hasher::finalize_xof`].
+///
+/// # Example
+///
+/// ```
+/// use blake3::hazmat::{merge_subtrees_root_xof, HasherExt, Mode};
+/// use blake3::{Hasher, CHUNK_LEN};
+///
+/// // Hash a 2-chunk subtree in steps. Note that only
+/// // the final chunk can be shorter than CHUNK_LEN.
+/// let chunk0 = &[42; CHUNK_LEN];
+/// let chunk1 = b"hello world";
+/// let chunk0_cv = Hasher::new()
+///     .update(chunk0)
+///     .finalize_non_root();
+/// let chunk1_cv = Hasher::new()
+///     .set_input_offset(CHUNK_LEN as u64)
+///     .update(chunk1)
+///     .finalize_non_root();
+///
+/// // Obtain a blake3::OutputReader at the root and extract 1000 bytes.
+/// let mut output_reader = merge_subtrees_root_xof(&chunk0_cv, &chunk1_cv, Mode::Hash);
+/// let mut output_bytes = [0; 1_000];
+/// output_reader.fill(&mut output_bytes);
+///
+/// // Double check the answer.
+/// let mut hasher = Hasher::new();
+/// hasher.update(chunk0);
+/// hasher.update(chunk1);
+/// let mut expected = [0; 1_000];
+/// hasher.finalize_xof().fill(&mut expected);
+/// assert_eq!(output_bytes, expected);
+/// ```
+pub fn merge_subtrees_root_xof(
+    left_child: &ChainingValue,
+    right_child: &ChainingValue,
+    mode: Mode,
+) -> crate::OutputReader {
+    crate::OutputReader::new(merge_subtrees_inner(left_child, right_child, mode))
+}
+
+/// An alias to distinguish [`hash_derive_key_context`] outputs from other keys.
+pub type ContextKey = [u8; KEY_LEN];
+
+/// Hash a [`derive_key`](crate::derive_key) context string and return a [`ContextKey`].
+///
+/// The _only_ valid uses for the returned [`ContextKey`] are [`Hasher::new_from_context_key`] and
+/// [`Mode::DeriveKeyMaterial`] (together with the merge subtree functions).
+///
+/// # Example
+///
+/// ```
+/// use blake3::Hasher;
+/// use blake3::hazmat::HasherExt;
+///
+/// let context_key = blake3::hazmat::hash_derive_key_context("foo");
+/// let mut hasher = Hasher::new_from_context_key(&context_key);
+/// hasher.update(b"bar");
+/// let derived_key = *hasher.finalize().as_bytes();
+///
+/// assert_eq!(derived_key, blake3::derive_key("foo", b"bar"));
+/// ```
+pub fn hash_derive_key_context(context: &str) -> ContextKey {
+    crate::hash_all_at_once::<crate::join::SerialJoin>(
+        context.as_bytes(),
+        IV,
+        crate::DERIVE_KEY_CONTEXT,
+    )
+    .root_hash()
+    .0
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    #[should_panic]
+    fn test_empty_subtree_should_panic() {
+        Hasher::new().finalize_non_root();
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_unaligned_offset_should_panic() {
+        Hasher::new().set_input_offset(1);
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_hasher_already_accepted_input_should_panic() {
+        Hasher::new().update(b"x").set_input_offset(0);
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_too_much_input_should_panic() {
+        Hasher::new()
+            .set_input_offset(CHUNK_LEN as u64)
+            .update(&[0; CHUNK_LEN + 1]);
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_set_input_offset_cant_finalize() {
+        Hasher::new().set_input_offset(CHUNK_LEN as u64).finalize();
+    }
+
+    #[test]
+    #[should_panic]
+    fn test_set_input_offset_cant_finalize_xof() {
+        Hasher::new()
+            .set_input_offset(CHUNK_LEN as u64)
+            .finalize_xof();
+    }
+
+    #[test]
+    fn test_grouped_hash() {
+        const MAX_CHUNKS: usize = (crate::test::TEST_CASES_MAX + 1) / CHUNK_LEN;
+        let mut input_buf = [0; crate::test::TEST_CASES_MAX];
+        crate::test::paint_test_input(&mut input_buf);
+        for subtree_chunks in [1, 2, 4, 8, 16, 32] {
+            #[cfg(feature = "std")]
+            dbg!(subtree_chunks);
+            let subtree_len = subtree_chunks * CHUNK_LEN;
+            for &case in crate::test::TEST_CASES {
+                if case <= subtree_len {
+                    continue;
+                }
+                #[cfg(feature = "std")]
+                dbg!(case);
+                let input = &input_buf[..case];
+                let expected_hash = crate::hash(input);
+
+                // Collect all the group chaining values.
+                let mut chaining_values = arrayvec::ArrayVec::<ChainingValue, MAX_CHUNKS>::new();
+                let mut subtree_offset = 0;
+                while subtree_offset < input.len() {
+                    let take = core::cmp::min(subtree_len, input.len() - subtree_offset);
+                    let subtree_input = &input[subtree_offset..][..take];
+                    let subtree_cv = Hasher::new()
+                        .set_input_offset(subtree_offset as u64)
+                        .update(subtree_input)
+                        .finalize_non_root();
+                    chaining_values.push(subtree_cv);
+                    subtree_offset += take;
+                }
+
+                // Compress all the chaining_values together, layer by layer.
+                assert!(chaining_values.len() >= 2);
+                while chaining_values.len() > 2 {
+                    let n = chaining_values.len();
+                    // Merge each side-by-side pair in place, overwriting the front half of the
+                    // array with the merged results. This moves us "up one level" in the tree.
+                    for i in 0..(n / 2) {
+                        chaining_values[i] = merge_subtrees_non_root(
+                            &chaining_values[2 * i],
+                            &chaining_values[2 * i + 1],
+                            Mode::Hash,
+                        );
+                    }
+                    // If there's an odd CV out, it moves up.
+                    if n % 2 == 1 {
+                        chaining_values[n / 2] = chaining_values[n - 1];
+                    }
+                    chaining_values.truncate(n / 2 + n % 2);
+                }
+                assert_eq!(chaining_values.len(), 2);
+                let root_hash =
+                    merge_subtrees_root(&chaining_values[0], &chaining_values[1], Mode::Hash);
+                assert_eq!(expected_hash, root_hash);
+            }
+        }
+    }
+
+    #[test]
+    fn test_keyed_hash_xof() {
+        let group0 = &[42; 4096];
+        let group1 = &[43; 4095];
+        let mut input = [0; 8191];
+        input[..4096].copy_from_slice(group0);
+        input[4096..].copy_from_slice(group1);
+        let key = &[44; 32];
+
+        let mut expected_output = [0; 100];
+        Hasher::new_keyed(&key)
+            .update(&input)
+            .finalize_xof()
+            .fill(&mut expected_output);
+
+        let mut hazmat_output = [0; 100];
+        let left = Hasher::new_keyed(key).update(group0).finalize_non_root();
+        let right = Hasher::new_keyed(key)
+            .set_input_offset(group0.len() as u64)
+            .update(group1)
+            .finalize_non_root();
+        merge_subtrees_root_xof(&left, &right, Mode::KeyedHash(&key)).fill(&mut hazmat_output);
+        assert_eq!(expected_output, hazmat_output);
+    }
+
+    #[test]
+    fn test_derive_key() {
+        let context = "foo";
+        let mut input = [0; 1025];
+        crate::test::paint_test_input(&mut input);
+        let expected = crate::derive_key(context, &input);
+
+        let cx_key = hash_derive_key_context(context);
+        let left = Hasher::new_from_context_key(&cx_key)
+            .update(&input[..1024])
+            .finalize_non_root();
+        let right = Hasher::new_from_context_key(&cx_key)
+            .set_input_offset(1024)
+            .update(&input[1024..])
+            .finalize_non_root();
+        let derived_key = merge_subtrees_root(&left, &right, Mode::DeriveKeyMaterial(&cx_key)).0;
+        assert_eq!(expected, derived_key);
+    }
+}
diff --git a/thirdparty/blake3/src/io.rs b/thirdparty/blake3/src/io.rs
new file mode 100644
index 000000000..7e8e154f7
--- /dev/null
+++ b/thirdparty/blake3/src/io.rs
@@ -0,0 +1,64 @@
+//! Helper functions for efficient IO.
+
+#[cfg(feature = "std")]
+pub(crate) fn copy_wide(
+    mut reader: impl std::io::Read,
+    hasher: &mut crate::Hasher,
+) -> std::io::Result<u64> {
+    let mut buffer = [0; 65536];
+    let mut total = 0;
+    loop {
+        match reader.read(&mut buffer) {
+            Ok(0) => return Ok(total),
+            Ok(n) => {
+                hasher.update(&buffer[..n]);
+                total += n as u64;
+            }
+            // see test_update_reader_interrupted
+            Err(e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
+            Err(e) => return Err(e),
+        }
+    }
+}
+
+// Mmap a file, if it looks like a good idea. Return None in cases where we know mmap will fail, or
+// if the file is short enough that mmapping isn't worth it. However, if we do try to mmap and it
+// fails, return the error.
+//
+// SAFETY: Mmaps are fundamentally unsafe, because you can call invariant-checking functions like
+// str::from_utf8 on them and then have them change out from under you. Letting a safe caller get
+// their hands on an mmap, or even a &[u8] that's backed by an mmap, is unsound. However, because
+// this function is crate-private, we can guarantee that all can ever happen in the event of a race
+// condition is that we either hash nonsense bytes or crash with SIGBUS or similar, neither of
+// which should risk memory corruption in a safe caller.
+//
+// PARANOIA: But a data race...is a data race...is a data race...right? Even if we know that no
+// platform in the "real world" is ever going to do anything other than compute the "wrong answer"
+// if we race on this mmap while we hash it, aren't we still supposed to feel bad about doing this?
+// Well, maybe. This is IO, and IO gets special carve-outs in the memory model. Consider a
+// memory-mapped register that returns random 32-bit words. (This is actually realistic if you have
+// a hardware RNG.) It's probably sound to construct a *const i32 pointing to that register and do
+// some raw pointer reads from it. Those reads should be volatile if you don't want the compiler to
+// coalesce them, but either way the compiler isn't allowed to just _go nuts_ and insert
+// should-never-happen branches to wipe your hard drive if two adjacent reads happen to give
+// different values. As far as I'm aware, there's no such thing as a read that's allowed if it's
+// volatile but prohibited if it's not (unlike atomics). As mentioned above, it's not ok to
+// construct a safe &i32 to the register if you're going to leak that reference to unknown callers.
+// But if you "know what you're doing," I don't think *const i32 and &i32 are fundamentally
+// different here. Feedback needed.
+#[cfg(feature = "mmap")]
+pub(crate) fn maybe_mmap_file(file: &std::fs::File) -> std::io::Result<Option<memmap2::Mmap>> {
+    let metadata = file.metadata()?;
+    let file_size = metadata.len();
+    if !metadata.is_file() {
+        // Not a real file.
+        Ok(None)
+    } else if file_size < 16 * 1024 {
+        // Mapping small files is not worth it, and some special files that can't be mapped report
+        // a size of zero.
+        Ok(None)
+    } else {
+        let map = unsafe { memmap2::Mmap::map(file)? };
+        Ok(Some(map))
+    }
+}
diff --git a/thirdparty/blake3/src/join.rs b/thirdparty/blake3/src/join.rs
new file mode 100644
index 000000000..862ebcf9a
--- /dev/null
+++ b/thirdparty/blake3/src/join.rs
@@ -0,0 +1,92 @@
+//! The multi-threading abstractions used by `Hasher::update_with_join`.
+//!
+//! Different implementations of the `Join` trait determine whether
+//! `Hasher::update_with_join` performs multi-threading on sufficiently large
+//! inputs. The `SerialJoin` implementation is single-threaded, and the
+//! `RayonJoin` implementation (gated by the `rayon` feature) is multi-threaded.
+//! Interfaces other than `Hasher::update_with_join`, like [`hash`](crate::hash)
+//! and [`Hasher::update`](crate::Hasher::update), always use `SerialJoin`
+//! internally.
+//!
+//! The `Join` trait is an almost exact copy of the [`rayon::join`] API, and
+//! `RayonJoin` is the only non-trivial implementation. Previously this trait
+//! was public, but currently it's been re-privatized, as it's both 1) of no
+//! value to most callers and 2) a pretty big implementation detail to commit
+//! to.
+//!
+//! [`rayon::join`]: https://docs.rs/rayon/1.3.0/rayon/fn.join.html
+
+/// The trait that abstracts over single-threaded and multi-threaded recursion.
+///
+/// See the [`join` module docs](index.html) for more details.
+pub trait Join {
+    fn join<A, B, RA, RB>(oper_a: A, oper_b: B) -> (RA, RB)
+    where
+        A: FnOnce() -> RA + Send,
+        B: FnOnce() -> RB + Send,
+        RA: Send,
+        RB: Send;
+}
+
+/// The trivial, serial implementation of `Join`. The left and right sides are
+/// executed one after the other, on the calling thread. The standalone hashing
+/// functions and the `Hasher::update` method use this implementation
+/// internally.
+///
+/// See the [`join` module docs](index.html) for more details.
+pub enum SerialJoin {}
+
+impl Join for SerialJoin {
+    #[inline]
+    fn join<A, B, RA, RB>(oper_a: A, oper_b: B) -> (RA, RB)
+    where
+        A: FnOnce() -> RA + Send,
+        B: FnOnce() -> RB + Send,
+        RA: Send,
+        RB: Send,
+    {
+        (oper_a(), oper_b())
+    }
+}
+
+/// The Rayon-based implementation of `Join`. The left and right sides are
+/// executed on the Rayon thread pool, potentially in parallel. This
+/// implementation is gated by the `rayon` feature, which is off by default.
+///
+/// See the [`join` module docs](index.html) for more details.
+#[cfg(feature = "rayon")]
+pub enum RayonJoin {}
+
+#[cfg(feature = "rayon")]
+impl Join for RayonJoin {
+    #[inline]
+    fn join<A, B, RA, RB>(oper_a: A, oper_b: B) -> (RA, RB)
+    where
+        A: FnOnce() -> RA + Send,
+        B: FnOnce() -> RB + Send,
+        RA: Send,
+        RB: Send,
+    {
+        rayon_core::join(oper_a, oper_b)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_serial_join() {
+        let oper_a = || 1 + 1;
+        let oper_b = || 2 + 2;
+        assert_eq!((2, 4), SerialJoin::join(oper_a, oper_b));
+    }
+
+    #[test]
+    #[cfg(feature = "rayon")]
+    fn test_rayon_join() {
+        let oper_a = || 1 + 1;
+        let oper_b = || 2 + 2;
+        assert_eq!((2, 4), RayonJoin::join(oper_a, oper_b));
+    }
+}
diff --git a/thirdparty/blake3/src/lib.rs b/thirdparty/blake3/src/lib.rs
new file mode 100644
index 000000000..d777896f4
--- /dev/null
+++ b/thirdparty/blake3/src/lib.rs
@@ -0,0 +1,1835 @@
+//! The official Rust implementation of the [BLAKE3] cryptographic hash
+//! function.
+//!
+//! # Examples
+//!
+//! ```
+//! # fn main() -> Result<(), Box<dyn std::error::Error>> {
+//! // Hash an input all at once.
+//! let hash1 = blake3::hash(b"foobarbaz");
+//!
+//! // Hash an input incrementally.
+//! let mut hasher = blake3::Hasher::new();
+//! hasher.update(b"foo");
+//! hasher.update(b"bar");
+//! hasher.update(b"baz");
+//! let hash2 = hasher.finalize();
+//! assert_eq!(hash1, hash2);
+//!
+//! // Extended output. OutputReader also implements Read and Seek.
+//! # #[cfg(feature = "std")] {
+//! let mut output = [0; 1000];
+//! let mut output_reader = hasher.finalize_xof();
+//! output_reader.fill(&mut output);
+//! assert_eq!(hash1, output[..32]);
+//! # }
+//!
+//! // Print a hash as hex.
+//! println!("{}", hash1);
+//! # Ok(())
+//! # }
+//! ```
+//!
+//! # Cargo Features
+//!
+//! The `std` feature (the only feature enabled by default) is required for
+//! implementations of the [`Write`] and [`Seek`] traits, the
+//! [`update_reader`](Hasher::update_reader) helper method, and runtime CPU
+//! feature detection on x86. If this feature is disabled, the only way to use
+//! the x86 SIMD implementations is to enable the corresponding instruction sets
+//! globally, with e.g. `RUSTFLAGS="-C target-cpu=native"`. The resulting binary
+//! will not be portable to other machines.
+//!
+//! The `rayon` feature (disabled by default, but enabled for [docs.rs]) adds
+//! the [`update_rayon`](Hasher::update_rayon) and (in combination with `mmap`
+//! below) [`update_mmap_rayon`](Hasher::update_mmap_rayon) methods, for
+//! multithreaded hashing. However, even if this feature is enabled, all other
+//! APIs remain single-threaded.
+//!
+//! The `mmap` feature (disabled by default, but enabled for [docs.rs]) adds the
+//! [`update_mmap`](Hasher::update_mmap) and (in combination with `rayon` above)
+//! [`update_mmap_rayon`](Hasher::update_mmap_rayon) helper methods for
+//! memory-mapped IO.
+//!
+//! The `zeroize` feature (disabled by default, but enabled for [docs.rs])
+//! implements
+//! [`Zeroize`](https://docs.rs/zeroize/latest/zeroize/trait.Zeroize.html) for
+//! this crate's types.
+//!
+//! The `serde` feature (disabled by default, but enabled for [docs.rs]) implements
+//! [`serde::Serialize`](https://docs.rs/serde/latest/serde/trait.Serialize.html) and
+//! [`serde::Deserialize`](https://docs.rs/serde/latest/serde/trait.Deserialize.html)
+//! for [`Hash`](struct@Hash).
+//!
+//! The NEON implementation is enabled by default for AArch64 but requires the
+//! `neon` feature for other ARM targets. Not all ARMv7 CPUs support NEON, and
+//! enabling this feature will produce a binary that's not portable to CPUs
+//! without NEON support.
+//!
+//! The `wasm32_simd` feature enables the WASM SIMD implementation for all `wasm32-`
+//! targets. Similar to the `neon` feature, if `wasm32_simd` is enabled, WASM SIMD
+//! support is assumed. This may become the default in the future.
+//!
+//! The `traits-preview` feature enables implementations of traits from the
+//! RustCrypto [`digest`] crate, and re-exports that crate as `traits::digest`.
+//! However, the traits aren't stable, and they're expected to change in
+//! incompatible ways before that crate reaches 1.0. For that reason, this crate
+//! makes no SemVer guarantees for this feature, and callers who use it should
+//! expect breaking changes between patch versions. (The "-preview" feature name
+//! follows the conventions of the RustCrypto [`signature`] crate.)
+//!
+//! [`Hasher::update_rayon`]: struct.Hasher.html#method.update_rayon
+//! [BLAKE3]: https://blake3.io
+//! [Rayon]: https://github.com/rayon-rs/rayon
+//! [docs.rs]: https://docs.rs/
+//! [`Write`]: https://doc.rust-lang.org/std/io/trait.Write.html
+//! [`Seek`]: https://doc.rust-lang.org/std/io/trait.Seek.html
+//! [`digest`]: https://crates.io/crates/digest
+//! [`signature`]: https://crates.io/crates/signature
+
+#![cfg_attr(not(feature = "std"), no_std)]
+
+#[cfg(test)]
+mod test;
+
+#[doc(hidden)]
+#[deprecated(since = "1.8.0", note = "use the hazmat module instead")]
+pub mod guts;
+
+pub mod hazmat;
+
+/// Undocumented and unstable, for benchmarks only.
+#[doc(hidden)]
+pub mod platform;
+
+// Platform-specific implementations of the compression function. These
+// BLAKE3-specific cfg flags are set in build.rs.
+#[cfg(blake3_avx2_rust)]
+#[path = "rust_avx2.rs"]
+mod avx2;
+#[cfg(blake3_avx2_ffi)]
+#[path = "ffi_avx2.rs"]
+mod avx2;
+#[cfg(blake3_avx512_ffi)]
+#[path = "ffi_avx512.rs"]
+mod avx512;
+#[cfg(blake3_neon)]
+#[path = "ffi_neon.rs"]
+mod neon;
+mod portable;
+#[cfg(blake3_sse2_rust)]
+#[path = "rust_sse2.rs"]
+mod sse2;
+#[cfg(blake3_sse2_ffi)]
+#[path = "ffi_sse2.rs"]
+mod sse2;
+#[cfg(blake3_sse41_rust)]
+#[path = "rust_sse41.rs"]
+mod sse41;
+#[cfg(blake3_sse41_ffi)]
+#[path = "ffi_sse41.rs"]
+mod sse41;
+
+#[cfg(blake3_wasm32_simd)]
+#[path = "wasm32_simd.rs"]
+mod wasm32_simd;
+
+#[cfg(feature = "traits-preview")]
+pub mod traits;
+
+mod io;
+mod join;
+
+use arrayref::{array_mut_ref, array_ref};
+use arrayvec::{ArrayString, ArrayVec};
+use core::cmp;
+use core::fmt;
+use platform::{Platform, MAX_SIMD_DEGREE, MAX_SIMD_DEGREE_OR_2};
+#[cfg(feature = "zeroize")]
+use zeroize::Zeroize;
+
+/// The number of bytes in a [`Hash`](struct.Hash.html), 32.
+pub const OUT_LEN: usize = 32;
+
+/// The number of bytes in a key, 32.
+pub const KEY_LEN: usize = 32;
+
+/// The number of bytes in a block, 64.
+///
+/// You don't usually need to think about this number. One case where it matters is calling
+/// [`OutputReader::fill`] in a loop, where using a `buf` argument that's a multiple of `BLOCK_LEN`
+/// avoids repeating work.
+pub const BLOCK_LEN: usize = 64;
+
+/// The number of bytes in a chunk, 1024.
+///
+/// You don't usually need to think about this number, but it often comes up in benchmarks, because
+/// the maximum degree of parallelism used by the implementation equals the number of chunks.
+pub const CHUNK_LEN: usize = 1024;
+
+const MAX_DEPTH: usize = 54; // 2^54 * CHUNK_LEN = 2^64
+
+// While iterating the compression function within a chunk, the CV is
+// represented as words, to avoid doing two extra endianness conversions for
+// each compression in the portable implementation. But the hash_many interface
+// needs to hash both input bytes and parent nodes, so its better for its
+// output CVs to be represented as bytes.
+type CVWords = [u32; 8];
+type CVBytes = [u8; 32]; // little-endian
+
+const IV: &CVWords = &[
+    0x6A09E667, 0xBB67AE85, 0x3C6EF372, 0xA54FF53A, 0x510E527F, 0x9B05688C, 0x1F83D9AB, 0x5BE0CD19,
+];
+
+const MSG_SCHEDULE: [[usize; 16]; 7] = [
+    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15],
+    [2, 6, 3, 10, 7, 0, 4, 13, 1, 11, 12, 5, 9, 14, 15, 8],
+    [3, 4, 10, 12, 13, 2, 7, 14, 6, 5, 9, 0, 11, 15, 8, 1],
+    [10, 7, 12, 9, 14, 3, 13, 15, 4, 0, 11, 2, 5, 8, 1, 6],
+    [12, 13, 9, 11, 15, 10, 14, 8, 7, 2, 5, 3, 0, 1, 6, 4],
+    [9, 14, 11, 5, 8, 12, 15, 1, 13, 3, 0, 10, 2, 6, 4, 7],
+    [11, 15, 5, 0, 1, 9, 8, 6, 14, 10, 2, 12, 3, 4, 7, 13],
+];
+
+// These are the internal flags that we use to domain separate root/non-root,
+// chunk/parent, and chunk beginning/middle/end. These get set at the high end
+// of the block flags word in the compression function, so their values start
+// high and go down.
+const CHUNK_START: u8 = 1 << 0;
+const CHUNK_END: u8 = 1 << 1;
+const PARENT: u8 = 1 << 2;
+const ROOT: u8 = 1 << 3;
+const KEYED_HASH: u8 = 1 << 4;
+const DERIVE_KEY_CONTEXT: u8 = 1 << 5;
+const DERIVE_KEY_MATERIAL: u8 = 1 << 6;
+
+#[inline]
+fn counter_low(counter: u64) -> u32 {
+    counter as u32
+}
+
+#[inline]
+fn counter_high(counter: u64) -> u32 {
+    (counter >> 32) as u32
+}
+
+/// An output of the default size, 32 bytes, which provides constant-time
+/// equality checking.
+///
+/// `Hash` implements [`From`] and [`Into`] for `[u8; 32]`, and it provides
+/// [`from_bytes`] and [`as_bytes`] for explicit conversions between itself and
+/// `[u8; 32]`. However, byte arrays and slices don't provide constant-time
+/// equality checking, which is often a security requirement in software that
+/// handles private data. `Hash` doesn't implement [`Deref`] or [`AsRef`], to
+/// avoid situations where a type conversion happens implicitly and the
+/// constant-time property is accidentally lost.
+///
+/// `Hash` provides the [`to_hex`] and [`from_hex`] methods for converting to
+/// and from hexadecimal. It also implements [`Display`] and [`FromStr`].
+///
+/// [`From`]: https://doc.rust-lang.org/std/convert/trait.From.html
+/// [`Into`]: https://doc.rust-lang.org/std/convert/trait.Into.html
+/// [`as_bytes`]: #method.as_bytes
+/// [`from_bytes`]: #method.from_bytes
+/// [`Deref`]: https://doc.rust-lang.org/stable/std/ops/trait.Deref.html
+/// [`AsRef`]: https://doc.rust-lang.org/std/convert/trait.AsRef.html
+/// [`to_hex`]: #method.to_hex
+/// [`from_hex`]: #method.from_hex
+/// [`Display`]: https://doc.rust-lang.org/std/fmt/trait.Display.html
+/// [`FromStr`]: https://doc.rust-lang.org/std/str/trait.FromStr.html
+#[cfg_attr(feature = "serde", derive(serde::Deserialize, serde::Serialize))]
+#[derive(Clone, Copy, Hash, Eq)]
+pub struct Hash([u8; OUT_LEN]);
+
+impl Hash {
+    /// The raw bytes of the `Hash`. Note that byte arrays don't provide
+    /// constant-time equality checking, so if  you need to compare hashes,
+    /// prefer the `Hash` type.
+    #[inline]
+    pub const fn as_bytes(&self) -> &[u8; OUT_LEN] {
+        &self.0
+    }
+
+    /// Create a `Hash` from its raw bytes representation.
+    pub const fn from_bytes(bytes: [u8; OUT_LEN]) -> Self {
+        Self(bytes)
+    }
+
+    /// Create a `Hash` from its raw bytes representation as a slice.
+    ///
+    /// Returns an error if the slice is not exactly 32 bytes long.
+    pub fn from_slice(bytes: &[u8]) -> Result<Self, core::array::TryFromSliceError> {
+        Ok(Self::from_bytes(bytes.try_into()?))
+    }
+
+    /// Encode a `Hash` in lowercase hexadecimal.
+    ///
+    /// The returned [`ArrayString`] is a fixed size and doesn't allocate memory
+    /// on the heap. Note that [`ArrayString`] doesn't provide constant-time
+    /// equality checking, so if you need to compare hashes, prefer the `Hash`
+    /// type.
+    ///
+    /// [`ArrayString`]: https://docs.rs/arrayvec/0.5.1/arrayvec/struct.ArrayString.html
+    pub fn to_hex(&self) -> ArrayString<{ 2 * OUT_LEN }> {
+        let mut s = ArrayString::new();
+        let table = b"0123456789abcdef";
+        for &b in self.0.iter() {
+            s.push(table[(b >> 4) as usize] as char);
+            s.push(table[(b & 0xf) as usize] as char);
+        }
+        s
+    }
+
+    /// Decode a `Hash` from hexadecimal. Both uppercase and lowercase ASCII
+    /// bytes are supported.
+    ///
+    /// Any byte outside the ranges `'0'...'9'`, `'a'...'f'`, and `'A'...'F'`
+    /// results in an error. An input length other than 64 also results in an
+    /// error.
+    ///
+    /// Note that `Hash` also implements `FromStr`, so `Hash::from_hex("...")`
+    /// is equivalent to `"...".parse()`.
+    pub fn from_hex(hex: impl AsRef<[u8]>) -> Result<Self, HexError> {
+        fn hex_val(byte: u8) -> Result<u8, HexError> {
+            match byte {
+                b'A'..=b'F' => Ok(byte - b'A' + 10),
+                b'a'..=b'f' => Ok(byte - b'a' + 10),
+                b'0'..=b'9' => Ok(byte - b'0'),
+                _ => Err(HexError(HexErrorInner::InvalidByte(byte))),
+            }
+        }
+        let hex_bytes: &[u8] = hex.as_ref();
+        if hex_bytes.len() != OUT_LEN * 2 {
+            return Err(HexError(HexErrorInner::InvalidLen(hex_bytes.len())));
+        }
+        let mut hash_bytes: [u8; OUT_LEN] = [0; OUT_LEN];
+        for i in 0..OUT_LEN {
+            hash_bytes[i] = 16 * hex_val(hex_bytes[2 * i])? + hex_val(hex_bytes[2 * i + 1])?;
+        }
+        Ok(Hash::from(hash_bytes))
+    }
+}
+
+impl From<[u8; OUT_LEN]> for Hash {
+    #[inline]
+    fn from(bytes: [u8; OUT_LEN]) -> Self {
+        Self::from_bytes(bytes)
+    }
+}
+
+impl From<Hash> for [u8; OUT_LEN] {
+    #[inline]
+    fn from(hash: Hash) -> Self {
+        hash.0
+    }
+}
+
+impl core::str::FromStr for Hash {
+    type Err = HexError;
+
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        Hash::from_hex(s)
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Zeroize for Hash {
+    fn zeroize(&mut self) {
+        // Destructuring to trigger compile error as a reminder to update this impl.
+        let Self(bytes) = self;
+        bytes.zeroize();
+    }
+}
+
+/// This implementation is constant-time.
+impl PartialEq for Hash {
+    #[inline]
+    fn eq(&self, other: &Hash) -> bool {
+        constant_time_eq::constant_time_eq_32(&self.0, &other.0)
+    }
+}
+
+/// This implementation is constant-time.
+impl PartialEq<[u8; OUT_LEN]> for Hash {
+    #[inline]
+    fn eq(&self, other: &[u8; OUT_LEN]) -> bool {
+        constant_time_eq::constant_time_eq_32(&self.0, other)
+    }
+}
+
+/// This implementation is constant-time if the target is 32 bytes long.
+impl PartialEq<[u8]> for Hash {
+    #[inline]
+    fn eq(&self, other: &[u8]) -> bool {
+        constant_time_eq::constant_time_eq(&self.0, other)
+    }
+}
+
+impl fmt::Display for Hash {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        // Formatting field as `&str` to reduce code size since the `Debug`
+        // dynamic dispatch table for `&str` is likely needed elsewhere already,
+        // but that for `ArrayString<[u8; 64]>` is not.
+        let hex = self.to_hex();
+        let hex: &str = hex.as_str();
+
+        f.write_str(hex)
+    }
+}
+
+impl fmt::Debug for Hash {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        // Formatting field as `&str` to reduce code size since the `Debug`
+        // dynamic dispatch table for `&str` is likely needed elsewhere already,
+        // but that for `ArrayString<[u8; 64]>` is not.
+        let hex = self.to_hex();
+        let hex: &str = hex.as_str();
+
+        f.debug_tuple("Hash").field(&hex).finish()
+    }
+}
+
+/// The error type for [`Hash::from_hex`].
+///
+/// The `.to_string()` representation of this error currently distinguishes between bad length
+/// errors and bad character errors. This is to help with logging and debugging, but it isn't a
+/// stable API detail, and it may change at any time.
+#[derive(Clone, Debug)]
+pub struct HexError(HexErrorInner);
+
+#[derive(Clone, Debug)]
+enum HexErrorInner {
+    InvalidByte(u8),
+    InvalidLen(usize),
+}
+
+impl fmt::Display for HexError {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        match self.0 {
+            HexErrorInner::InvalidByte(byte) => {
+                if byte < 128 {
+                    write!(f, "invalid hex character: {:?}", byte as char)
+                } else {
+                    write!(f, "invalid hex character: 0x{:x}", byte)
+                }
+            }
+            HexErrorInner::InvalidLen(len) => {
+                write!(f, "expected 64 hex bytes, received {}", len)
+            }
+        }
+    }
+}
+
+#[cfg(feature = "std")]
+impl std::error::Error for HexError {}
+
+// Each chunk or parent node can produce either a 32-byte chaining value or, by
+// setting the ROOT flag, any number of final output bytes. The Output struct
+// captures the state just prior to choosing between those two possibilities.
+#[derive(Clone)]
+struct Output {
+    input_chaining_value: CVWords,
+    block: [u8; 64],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+    platform: Platform,
+}
+
+impl Output {
+    fn chaining_value(&self) -> CVBytes {
+        let mut cv = self.input_chaining_value;
+        self.platform.compress_in_place(
+            &mut cv,
+            &self.block,
+            self.block_len,
+            self.counter,
+            self.flags,
+        );
+        platform::le_bytes_from_words_32(&cv)
+    }
+
+    fn root_hash(&self) -> Hash {
+        debug_assert_eq!(self.counter, 0);
+        let mut cv = self.input_chaining_value;
+        self.platform
+            .compress_in_place(&mut cv, &self.block, self.block_len, 0, self.flags | ROOT);
+        Hash(platform::le_bytes_from_words_32(&cv))
+    }
+
+    fn root_output_block(&self) -> [u8; 2 * OUT_LEN] {
+        self.platform.compress_xof(
+            &self.input_chaining_value,
+            &self.block,
+            self.block_len,
+            self.counter,
+            self.flags | ROOT,
+        )
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Zeroize for Output {
+    fn zeroize(&mut self) {
+        // Destructuring to trigger compile error as a reminder to update this impl.
+        let Self {
+            input_chaining_value,
+            block,
+            block_len,
+            counter,
+            flags,
+            platform: _,
+        } = self;
+
+        input_chaining_value.zeroize();
+        block.zeroize();
+        block_len.zeroize();
+        counter.zeroize();
+        flags.zeroize();
+    }
+}
+
+#[derive(Clone)]
+struct ChunkState {
+    cv: CVWords,
+    chunk_counter: u64,
+    buf: [u8; BLOCK_LEN],
+    buf_len: u8,
+    blocks_compressed: u8,
+    flags: u8,
+    platform: Platform,
+}
+
+impl ChunkState {
+    fn new(key: &CVWords, chunk_counter: u64, flags: u8, platform: Platform) -> Self {
+        Self {
+            cv: *key,
+            chunk_counter,
+            buf: [0; BLOCK_LEN],
+            buf_len: 0,
+            blocks_compressed: 0,
+            flags,
+            platform,
+        }
+    }
+
+    fn count(&self) -> usize {
+        BLOCK_LEN * self.blocks_compressed as usize + self.buf_len as usize
+    }
+
+    fn fill_buf(&mut self, input: &mut &[u8]) {
+        let want = BLOCK_LEN - self.buf_len as usize;
+        let take = cmp::min(want, input.len());
+        self.buf[self.buf_len as usize..][..take].copy_from_slice(&input[..take]);
+        self.buf_len += take as u8;
+        *input = &input[take..];
+    }
+
+    fn start_flag(&self) -> u8 {
+        if self.blocks_compressed == 0 {
+            CHUNK_START
+        } else {
+            0
+        }
+    }
+
+    // Try to avoid buffering as much as possible, by compressing directly from
+    // the input slice when full blocks are available.
+    fn update(&mut self, mut input: &[u8]) -> &mut Self {
+        if self.buf_len > 0 {
+            self.fill_buf(&mut input);
+            if !input.is_empty() {
+                debug_assert_eq!(self.buf_len as usize, BLOCK_LEN);
+                let block_flags = self.flags | self.start_flag(); // borrowck
+                self.platform.compress_in_place(
+                    &mut self.cv,
+                    &self.buf,
+                    BLOCK_LEN as u8,
+                    self.chunk_counter,
+                    block_flags,
+                );
+                self.buf_len = 0;
+                self.buf = [0; BLOCK_LEN];
+                self.blocks_compressed += 1;
+            }
+        }
+
+        while input.len() > BLOCK_LEN {
+            debug_assert_eq!(self.buf_len, 0);
+            let block_flags = self.flags | self.start_flag(); // borrowck
+            self.platform.compress_in_place(
+                &mut self.cv,
+                array_ref!(input, 0, BLOCK_LEN),
+                BLOCK_LEN as u8,
+                self.chunk_counter,
+                block_flags,
+            );
+            self.blocks_compressed += 1;
+            input = &input[BLOCK_LEN..];
+        }
+
+        self.fill_buf(&mut input);
+        debug_assert!(input.is_empty());
+        debug_assert!(self.count() <= CHUNK_LEN);
+        self
+    }
+
+    fn output(&self) -> Output {
+        let block_flags = self.flags | self.start_flag() | CHUNK_END;
+        Output {
+            input_chaining_value: self.cv,
+            block: self.buf,
+            block_len: self.buf_len,
+            counter: self.chunk_counter,
+            flags: block_flags,
+            platform: self.platform,
+        }
+    }
+}
+
+// Don't derive(Debug), because the state may be secret.
+impl fmt::Debug for ChunkState {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        f.debug_struct("ChunkState")
+            .field("count", &self.count())
+            .field("chunk_counter", &self.chunk_counter)
+            .field("flags", &self.flags)
+            .field("platform", &self.platform)
+            .finish()
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Zeroize for ChunkState {
+    fn zeroize(&mut self) {
+        // Destructuring to trigger compile error as a reminder to update this impl.
+        let Self {
+            cv,
+            chunk_counter,
+            buf,
+            buf_len,
+            blocks_compressed,
+            flags,
+            platform: _,
+        } = self;
+
+        cv.zeroize();
+        chunk_counter.zeroize();
+        buf.zeroize();
+        buf_len.zeroize();
+        blocks_compressed.zeroize();
+        flags.zeroize();
+    }
+}
+
+// IMPLEMENTATION NOTE
+// ===================
+// The recursive function compress_subtree_wide(), implemented below, is the
+// basis of high-performance BLAKE3. We use it both for all-at-once hashing,
+// and for the incremental input with Hasher (though we have to be careful with
+// subtree boundaries in the incremental case). compress_subtree_wide() applies
+// several optimizations at the same time:
+// - Multithreading with Rayon.
+// - Parallel chunk hashing with SIMD.
+// - Parallel parent hashing with SIMD. Note that while SIMD chunk hashing
+//   maxes out at MAX_SIMD_DEGREE*CHUNK_LEN, parallel parent hashing continues
+//   to benefit from larger inputs, because more levels of the tree benefit can
+//   use full-width SIMD vectors for parent hashing. Without parallel parent
+//   hashing, we lose about 10% of overall throughput on AVX2 and AVX-512.
+
+/// Undocumented and unstable, for benchmarks only.
+#[doc(hidden)]
+#[derive(Clone, Copy)]
+pub enum IncrementCounter {
+    Yes,
+    No,
+}
+
+impl IncrementCounter {
+    #[inline]
+    fn yes(&self) -> bool {
+        match self {
+            IncrementCounter::Yes => true,
+            IncrementCounter::No => false,
+        }
+    }
+}
+
+// The largest power of two less than or equal to `n`, used in Hasher::update(). This is similar to
+// left_subtree_len(n), but note that left_subtree_len(n) is strictly less than `n`.
+fn largest_power_of_two_leq(n: usize) -> usize {
+    ((n / 2) + 1).next_power_of_two()
+}
+
+// Use SIMD parallelism to hash up to MAX_SIMD_DEGREE chunks at the same time
+// on a single thread. Write out the chunk chaining values and return the
+// number of chunks hashed. These chunks are never the root and never empty;
+// those cases use a different codepath.
+fn compress_chunks_parallel(
+    input: &[u8],
+    key: &CVWords,
+    chunk_counter: u64,
+    flags: u8,
+    platform: Platform,
+    out: &mut [u8],
+) -> usize {
+    debug_assert!(!input.is_empty(), "empty chunks below the root");
+    debug_assert!(input.len() <= MAX_SIMD_DEGREE * CHUNK_LEN);
+
+    let mut chunks_exact = input.chunks_exact(CHUNK_LEN);
+    let mut chunks_array = ArrayVec::<&[u8; CHUNK_LEN], MAX_SIMD_DEGREE>::new();
+    for chunk in &mut chunks_exact {
+        chunks_array.push(array_ref!(chunk, 0, CHUNK_LEN));
+    }
+    platform.hash_many(
+        &chunks_array,
+        key,
+        chunk_counter,
+        IncrementCounter::Yes,
+        flags,
+        CHUNK_START,
+        CHUNK_END,
+        out,
+    );
+
+    // Hash the remaining partial chunk, if there is one. Note that the empty
+    // chunk (meaning the empty message) is a different codepath.
+    let chunks_so_far = chunks_array.len();
+    if !chunks_exact.remainder().is_empty() {
+        let counter = chunk_counter + chunks_so_far as u64;
+        let mut chunk_state = ChunkState::new(key, counter, flags, platform);
+        chunk_state.update(chunks_exact.remainder());
+        *array_mut_ref!(out, chunks_so_far * OUT_LEN, OUT_LEN) =
+            chunk_state.output().chaining_value();
+        chunks_so_far + 1
+    } else {
+        chunks_so_far
+    }
+}
+
+// Use SIMD parallelism to hash up to MAX_SIMD_DEGREE parents at the same time
+// on a single thread. Write out the parent chaining values and return the
+// number of parents hashed. (If there's an odd input chaining value left over,
+// return it as an additional output.) These parents are never the root and
+// never empty; those cases use a different codepath.
+fn compress_parents_parallel(
+    child_chaining_values: &[u8],
+    key: &CVWords,
+    flags: u8,
+    platform: Platform,
+    out: &mut [u8],
+) -> usize {
+    debug_assert_eq!(child_chaining_values.len() % OUT_LEN, 0, "wacky hash bytes");
+    let num_children = child_chaining_values.len() / OUT_LEN;
+    debug_assert!(num_children >= 2, "not enough children");
+    debug_assert!(num_children <= 2 * MAX_SIMD_DEGREE_OR_2, "too many");
+
+    let mut parents_exact = child_chaining_values.chunks_exact(BLOCK_LEN);
+    // Use MAX_SIMD_DEGREE_OR_2 rather than MAX_SIMD_DEGREE here, because of
+    // the requirements of compress_subtree_wide().
+    let mut parents_array = ArrayVec::<&[u8; BLOCK_LEN], MAX_SIMD_DEGREE_OR_2>::new();
+    for parent in &mut parents_exact {
+        parents_array.push(array_ref!(parent, 0, BLOCK_LEN));
+    }
+    platform.hash_many(
+        &parents_array,
+        key,
+        0, // Parents always use counter 0.
+        IncrementCounter::No,
+        flags | PARENT,
+        0, // Parents have no start flags.
+        0, // Parents have no end flags.
+        out,
+    );
+
+    // If there's an odd child left over, it becomes an output.
+    let parents_so_far = parents_array.len();
+    if !parents_exact.remainder().is_empty() {
+        out[parents_so_far * OUT_LEN..][..OUT_LEN].copy_from_slice(parents_exact.remainder());
+        parents_so_far + 1
+    } else {
+        parents_so_far
+    }
+}
+
+// The wide helper function returns (writes out) an array of chaining values
+// and returns the length of that array. The number of chaining values returned
+// is the dynamically detected SIMD degree, at most MAX_SIMD_DEGREE. Or fewer,
+// if the input is shorter than that many chunks. The reason for maintaining a
+// wide array of chaining values going back up the tree, is to allow the
+// implementation to hash as many parents in parallel as possible.
+//
+// As a special case when the SIMD degree is 1, this function will still return
+// at least 2 outputs. This guarantees that this function doesn't perform the
+// root compression. (If it did, it would use the wrong flags, and also we
+// wouldn't be able to implement extendable output.) Note that this function is
+// not used when the whole input is only 1 chunk long; that's a different
+// codepath.
+//
+// Why not just have the caller split the input on the first update(), instead
+// of implementing this special rule? Because we don't want to limit SIMD or
+// multithreading parallelism for that update().
+fn compress_subtree_wide<J: join::Join>(
+    input: &[u8],
+    key: &CVWords,
+    chunk_counter: u64,
+    flags: u8,
+    platform: Platform,
+    out: &mut [u8],
+) -> usize {
+    // Note that the single chunk case does *not* bump the SIMD degree up to 2
+    // when it is 1. This allows Rayon the option of multithreading even the
+    // 2-chunk case, which can help performance on smaller platforms.
+    if input.len() <= platform.simd_degree() * CHUNK_LEN {
+        return compress_chunks_parallel(input, key, chunk_counter, flags, platform, out);
+    }
+
+    // With more than simd_degree chunks, we need to recurse. Start by dividing
+    // the input into left and right subtrees. (Note that this is only optimal
+    // as long as the SIMD degree is a power of 2. If we ever get a SIMD degree
+    // of 3 or something, we'll need a more complicated strategy.)
+    debug_assert_eq!(platform.simd_degree().count_ones(), 1, "power of 2");
+    let (left, right) = input.split_at(hazmat::left_subtree_len(input.len() as u64) as usize);
+    let right_chunk_counter = chunk_counter + (left.len() / CHUNK_LEN) as u64;
+
+    // Make space for the child outputs. Here we use MAX_SIMD_DEGREE_OR_2 to
+    // account for the special case of returning 2 outputs when the SIMD degree
+    // is 1.
+    let mut cv_array = [0; 2 * MAX_SIMD_DEGREE_OR_2 * OUT_LEN];
+    let degree = if left.len() == CHUNK_LEN {
+        // The "simd_degree=1 and we're at the leaf nodes" case.
+        debug_assert_eq!(platform.simd_degree(), 1);
+        1
+    } else {
+        cmp::max(platform.simd_degree(), 2)
+    };
+    let (left_out, right_out) = cv_array.split_at_mut(degree * OUT_LEN);
+
+    // Recurse! For update_rayon(), this is where we take advantage of RayonJoin and use multiple
+    // threads.
+    let (left_n, right_n) = J::join(
+        || compress_subtree_wide::<J>(left, key, chunk_counter, flags, platform, left_out),
+        || compress_subtree_wide::<J>(right, key, right_chunk_counter, flags, platform, right_out),
+    );
+
+    // The special case again. If simd_degree=1, then we'll have left_n=1 and
+    // right_n=1. Rather than compressing them into a single output, return
+    // them directly, to make sure we always have at least two outputs.
+    debug_assert_eq!(left_n, degree);
+    debug_assert!(right_n >= 1 && right_n <= left_n);
+    if left_n == 1 {
+        out[..2 * OUT_LEN].copy_from_slice(&cv_array[..2 * OUT_LEN]);
+        return 2;
+    }
+
+    // Otherwise, do one layer of parent node compression.
+    let num_children = left_n + right_n;
+    compress_parents_parallel(
+        &cv_array[..num_children * OUT_LEN],
+        key,
+        flags,
+        platform,
+        out,
+    )
+}
+
+// Hash a subtree with compress_subtree_wide(), and then condense the resulting
+// list of chaining values down to a single parent node. Don't compress that
+// last parent node, however. Instead, return its message bytes (the
+// concatenated chaining values of its children). This is necessary when the
+// first call to update() supplies a complete subtree, because the topmost
+// parent node of that subtree could end up being the root. It's also necessary
+// for extended output in the general case.
+//
+// As with compress_subtree_wide(), this function is not used on inputs of 1
+// chunk or less. That's a different codepath.
+fn compress_subtree_to_parent_node<J: join::Join>(
+    input: &[u8],
+    key: &CVWords,
+    chunk_counter: u64,
+    flags: u8,
+    platform: Platform,
+) -> [u8; BLOCK_LEN] {
+    debug_assert!(input.len() > CHUNK_LEN);
+    let mut cv_array = [0; MAX_SIMD_DEGREE_OR_2 * OUT_LEN];
+    let mut num_cvs =
+        compress_subtree_wide::<J>(input, &key, chunk_counter, flags, platform, &mut cv_array);
+    debug_assert!(num_cvs >= 2);
+
+    // If MAX_SIMD_DEGREE is greater than 2 and there's enough input,
+    // compress_subtree_wide() returns more than 2 chaining values. Condense
+    // them into 2 by forming parent nodes repeatedly.
+    let mut out_array = [0; MAX_SIMD_DEGREE_OR_2 * OUT_LEN / 2];
+    while num_cvs > 2 {
+        let cv_slice = &cv_array[..num_cvs * OUT_LEN];
+        num_cvs = compress_parents_parallel(cv_slice, key, flags, platform, &mut out_array);
+        cv_array[..num_cvs * OUT_LEN].copy_from_slice(&out_array[..num_cvs * OUT_LEN]);
+    }
+    *array_ref!(cv_array, 0, 2 * OUT_LEN)
+}
+
+// Hash a complete input all at once. Unlike compress_subtree_wide() and
+// compress_subtree_to_parent_node(), this function handles the 1 chunk case.
+fn hash_all_at_once<J: join::Join>(input: &[u8], key: &CVWords, flags: u8) -> Output {
+    let platform = Platform::detect();
+
+    // If the whole subtree is one chunk, hash it directly with a ChunkState.
+    if input.len() <= CHUNK_LEN {
+        return ChunkState::new(key, 0, flags, platform)
+            .update(input)
+            .output();
+    }
+
+    // Otherwise construct an Output object from the parent node returned by
+    // compress_subtree_to_parent_node().
+    Output {
+        input_chaining_value: *key,
+        block: compress_subtree_to_parent_node::<J>(input, key, 0, flags, platform),
+        block_len: BLOCK_LEN as u8,
+        counter: 0,
+        flags: flags | PARENT,
+        platform,
+    }
+}
+
+/// The default hash function.
+///
+/// For an incremental version that accepts multiple writes, see [`Hasher::new`],
+/// [`Hasher::update`], and [`Hasher::finalize`]. These two lines are equivalent:
+///
+/// ```
+/// let hash = blake3::hash(b"foo");
+/// # let hash1 = hash;
+///
+/// let hash = blake3::Hasher::new().update(b"foo").finalize();
+/// # let hash2 = hash;
+/// # assert_eq!(hash1, hash2);
+/// ```
+///
+/// For output sizes other than 32 bytes, see [`Hasher::finalize_xof`] and
+/// [`OutputReader`].
+///
+/// This function is always single-threaded. For multithreading support, see
+/// [`Hasher::update_rayon`](struct.Hasher.html#method.update_rayon).
+pub fn hash(input: &[u8]) -> Hash {
+    hash_all_at_once::<join::SerialJoin>(input, IV, 0).root_hash()
+}
+
+/// The keyed hash function.
+///
+/// This is suitable for use as a message authentication code, for example to
+/// replace an HMAC instance. In that use case, the constant-time equality
+/// checking provided by [`Hash`](struct.Hash.html) is almost always a security
+/// requirement, and callers need to be careful not to compare MACs as raw
+/// bytes.
+///
+/// For an incremental version that accepts multiple writes, see [`Hasher::new_keyed`],
+/// [`Hasher::update`], and [`Hasher::finalize`]. These two lines are equivalent:
+///
+/// ```
+/// # const KEY: &[u8; 32] = &[0; 32];
+/// let mac = blake3::keyed_hash(KEY, b"foo");
+/// # let mac1 = mac;
+///
+/// let mac = blake3::Hasher::new_keyed(KEY).update(b"foo").finalize();
+/// # let mac2 = mac;
+/// # assert_eq!(mac1, mac2);
+/// ```
+///
+/// For output sizes other than 32 bytes, see [`Hasher::finalize_xof`], and [`OutputReader`].
+///
+/// This function is always single-threaded. For multithreading support, see
+/// [`Hasher::update_rayon`](struct.Hasher.html#method.update_rayon).
+pub fn keyed_hash(key: &[u8; KEY_LEN], input: &[u8]) -> Hash {
+    let key_words = platform::words_from_le_bytes_32(key);
+    hash_all_at_once::<join::SerialJoin>(input, &key_words, KEYED_HASH).root_hash()
+}
+
+/// The key derivation function.
+///
+/// Given cryptographic key material of any length and a context string of any
+/// length, this function outputs a 32-byte derived subkey. **The context string
+/// should be hardcoded, globally unique, and application-specific.** A good
+/// default format for such strings is `"[application] [commit timestamp]
+/// [purpose]"`, e.g., `"example.com 2019-12-25 16:18:03 session tokens v1"`.
+///
+/// Key derivation is important when you want to use the same key in multiple
+/// algorithms or use cases. Using the same key with different cryptographic
+/// algorithms is generally forbidden, and deriving a separate subkey for each
+/// use case protects you from bad interactions. Derived keys also mitigate the
+/// damage from one part of your application accidentally leaking its key.
+///
+/// As a rare exception to that general rule, however, it is possible to use
+/// `derive_key` itself with key material that you are already using with
+/// another algorithm. You might need to do this if you're adding features to
+/// an existing application, which does not yet use key derivation internally.
+/// However, you still must not share key material with algorithms that forbid
+/// key reuse entirely, like a one-time pad. For more on this, see sections 6.2
+/// and 7.8 of the [BLAKE3 paper](https://github.com/BLAKE3-team/BLAKE3-specs/blob/master/blake3.pdf).
+///
+/// Note that BLAKE3 is not a password hash, and **`derive_key` should never be
+/// used with passwords.** Instead, use a dedicated password hash like
+/// [Argon2]. Password hashes are entirely different from generic hash
+/// functions, with opposite design requirements.
+///
+/// For an incremental version that accepts multiple writes, see [`Hasher::new_derive_key`],
+/// [`Hasher::update`], and [`Hasher::finalize`]. These two statements are equivalent:
+///
+/// ```
+/// # const CONTEXT: &str = "example.com 2019-12-25 16:18:03 session tokens v1";
+/// let key = blake3::derive_key(CONTEXT, b"key material, not a password");
+/// # let key1 = key;
+///
+/// let key: [u8; 32] = blake3::Hasher::new_derive_key(CONTEXT)
+///     .update(b"key material, not a password")
+///     .finalize()
+///     .into();
+/// # let key2 = key;
+/// # assert_eq!(key1, key2);
+/// ```
+///
+/// For output sizes other than 32 bytes, see [`Hasher::finalize_xof`], and [`OutputReader`].
+///
+/// This function is always single-threaded. For multithreading support, see
+/// [`Hasher::update_rayon`](struct.Hasher.html#method.update_rayon).
+///
+/// [Argon2]: https://en.wikipedia.org/wiki/Argon2
+pub fn derive_key(context: &str, key_material: &[u8]) -> [u8; OUT_LEN] {
+    let context_key = hazmat::hash_derive_key_context(context);
+    let context_key_words = platform::words_from_le_bytes_32(&context_key);
+    hash_all_at_once::<join::SerialJoin>(key_material, &context_key_words, DERIVE_KEY_MATERIAL)
+        .root_hash()
+        .0
+}
+
+fn parent_node_output(
+    left_child: &CVBytes,
+    right_child: &CVBytes,
+    key: &CVWords,
+    flags: u8,
+    platform: Platform,
+) -> Output {
+    let mut block = [0; BLOCK_LEN];
+    block[..32].copy_from_slice(left_child);
+    block[32..].copy_from_slice(right_child);
+    Output {
+        input_chaining_value: *key,
+        block,
+        block_len: BLOCK_LEN as u8,
+        counter: 0,
+        flags: flags | PARENT,
+        platform,
+    }
+}
+
+/// An incremental hash state that can accept any number of writes.
+///
+/// The `rayon` and `mmap` Cargo features enable additional methods on this
+/// type related to multithreading and memory-mapped IO.
+///
+/// When the `traits-preview` Cargo feature is enabled, this type implements
+/// several commonly used traits from the
+/// [`digest`](https://crates.io/crates/digest) crate. However, those
+/// traits aren't stable, and they're expected to change in incompatible ways
+/// before that crate reaches 1.0. For that reason, this crate makes no SemVer
+/// guarantees for this feature, and callers who use it should expect breaking
+/// changes between patch versions.
+///
+/// # Examples
+///
+/// ```
+/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
+/// // Hash an input incrementally.
+/// let mut hasher = blake3::Hasher::new();
+/// hasher.update(b"foo");
+/// hasher.update(b"bar");
+/// hasher.update(b"baz");
+/// assert_eq!(hasher.finalize(), blake3::hash(b"foobarbaz"));
+///
+/// // Extended output. OutputReader also implements Read and Seek.
+/// # #[cfg(feature = "std")] {
+/// let mut output = [0; 1000];
+/// let mut output_reader = hasher.finalize_xof();
+/// output_reader.fill(&mut output);
+/// assert_eq!(&output[..32], blake3::hash(b"foobarbaz").as_bytes());
+/// # }
+/// # Ok(())
+/// # }
+/// ```
+#[derive(Clone)]
+pub struct Hasher {
+    key: CVWords,
+    chunk_state: ChunkState,
+    initial_chunk_counter: u64,
+    // The stack size is MAX_DEPTH + 1 because we do lazy merging. For example,
+    // with 7 chunks, we have 3 entries in the stack. Adding an 8th chunk
+    // requires a 4th entry, rather than merging everything down to 1, because
+    // we don't know whether more input is coming. This is different from how
+    // the reference implementation does things.
+    cv_stack: ArrayVec<CVBytes, { MAX_DEPTH + 1 }>,
+}
+
+impl Hasher {
+    fn new_internal(key: &CVWords, flags: u8) -> Self {
+        Self {
+            key: *key,
+            chunk_state: ChunkState::new(key, 0, flags, Platform::detect()),
+            initial_chunk_counter: 0,
+            cv_stack: ArrayVec::new(),
+        }
+    }
+
+    /// Construct a new `Hasher` for the regular hash function.
+    pub fn new() -> Self {
+        Self::new_internal(IV, 0)
+    }
+
+    /// Construct a new `Hasher` for the keyed hash function. See
+    /// [`keyed_hash`].
+    ///
+    /// [`keyed_hash`]: fn.keyed_hash.html
+    pub fn new_keyed(key: &[u8; KEY_LEN]) -> Self {
+        let key_words = platform::words_from_le_bytes_32(key);
+        Self::new_internal(&key_words, KEYED_HASH)
+    }
+
+    /// Construct a new `Hasher` for the key derivation function. See
+    /// [`derive_key`]. The context string should be hardcoded, globally
+    /// unique, and application-specific.
+    ///
+    /// [`derive_key`]: fn.derive_key.html
+    pub fn new_derive_key(context: &str) -> Self {
+        let context_key = hazmat::hash_derive_key_context(context);
+        let context_key_words = platform::words_from_le_bytes_32(&context_key);
+        Self::new_internal(&context_key_words, DERIVE_KEY_MATERIAL)
+    }
+
+    /// Reset the `Hasher` to its initial state.
+    ///
+    /// This is functionally the same as overwriting the `Hasher` with a new
+    /// one, using the same key or context string if any.
+    pub fn reset(&mut self) -> &mut Self {
+        self.chunk_state = ChunkState::new(
+            &self.key,
+            0,
+            self.chunk_state.flags,
+            self.chunk_state.platform,
+        );
+        self.cv_stack.clear();
+        self
+    }
+
+    // As described in push_cv() below, we do "lazy merging", delaying merges
+    // until right before the next CV is about to be added. This is different
+    // from the reference implementation. Another difference is that we aren't
+    // always merging 1 chunk at a time. Instead, each CV might represent any
+    // power-of-two number of chunks, as long as the smaller-above-larger stack
+    // order is maintained. Instead of the "count the trailing 0-bits"
+    // algorithm described in the spec (which assumes you're adding one chunk
+    // at a time), we use a "count the total number of 1-bits" variant (which
+    // doesn't assume that). The principle is the same: each CV that should
+    // remain in the stack is represented by a 1-bit in the total number of
+    // chunks (or bytes) so far.
+    fn merge_cv_stack(&mut self, chunk_counter: u64) {
+        // Account for non-zero cases of Hasher::set_input_offset, where there are no prior
+        // subtrees in the stack. Note that initial_chunk_counter is always 0 for callers who don't
+        // use the hazmat module.
+        let post_merge_stack_len =
+            (chunk_counter - self.initial_chunk_counter).count_ones() as usize;
+        while self.cv_stack.len() > post_merge_stack_len {
+            let right_child = self.cv_stack.pop().unwrap();
+            let left_child = self.cv_stack.pop().unwrap();
+            let parent_output = parent_node_output(
+                &left_child,
+                &right_child,
+                &self.key,
+                self.chunk_state.flags,
+                self.chunk_state.platform,
+            );
+            self.cv_stack.push(parent_output.chaining_value());
+        }
+    }
+
+    // In reference_impl.rs, we merge the new CV with existing CVs from the
+    // stack before pushing it. We can do that because we know more input is
+    // coming, so we know none of the merges are root.
+    //
+    // This setting is different. We want to feed as much input as possible to
+    // compress_subtree_wide(), without setting aside anything for the
+    // chunk_state. If the user gives us 64 KiB, we want to parallelize over
+    // all 64 KiB at once as a single subtree, if at all possible.
+    //
+    // This leads to two problems:
+    // 1) This 64 KiB input might be the only call that ever gets made to
+    //    update. In this case, the root node of the 64 KiB subtree would be
+    //    the root node of the whole tree, and it would need to be ROOT
+    //    finalized. We can't compress it until we know.
+    // 2) This 64 KiB input might complete a larger tree, whose root node is
+    //    similarly going to be the root of the whole tree. For example,
+    //    maybe we have 196 KiB (that is, 128 + 64) hashed so far. We can't
+    //    compress the node at the root of the 256 KiB subtree until we know
+    //    how to finalize it.
+    //
+    // The second problem is solved with "lazy merging". That is, when we're
+    // about to add a CV to the stack, we don't merge it with anything first,
+    // as the reference impl does. Instead we do merges using the *previous* CV
+    // that was added, which is sitting on top of the stack, and we put the new
+    // CV (unmerged) on top of the stack afterwards. This guarantees that we
+    // never merge the root node until finalize().
+    //
+    // Solving the first problem requires an additional tool,
+    // compress_subtree_to_parent_node(). That function always returns the top
+    // *two* chaining values of the subtree it's compressing. We then do lazy
+    // merging with each of them separately, so that the second CV will always
+    // remain unmerged. (That also helps us support extendable output when
+    // we're hashing an input all-at-once.)
+    fn push_cv(&mut self, new_cv: &CVBytes, chunk_counter: u64) {
+        self.merge_cv_stack(chunk_counter);
+        self.cv_stack.push(*new_cv);
+    }
+
+    /// Add input bytes to the hash state. You can call this any number of times.
+    ///
+    /// This method is always single-threaded. For multithreading support, see
+    /// [`update_rayon`](#method.update_rayon) (enabled with the `rayon` Cargo feature).
+    ///
+    /// Note that the degree of SIMD parallelism that `update` can use is limited by the size of
+    /// this input buffer. See [`update_reader`](#method.update_reader).
+    pub fn update(&mut self, input: &[u8]) -> &mut Self {
+        self.update_with_join::<join::SerialJoin>(input)
+    }
+
+    fn update_with_join<J: join::Join>(&mut self, mut input: &[u8]) -> &mut Self {
+        let input_offset = self.initial_chunk_counter * CHUNK_LEN as u64;
+        if let Some(max) = hazmat::max_subtree_len(input_offset) {
+            let remaining = max - self.count();
+            assert!(
+                input.len() as u64 <= remaining,
+                "the subtree starting at {} contains at most {} bytes (found {})",
+                CHUNK_LEN as u64 * self.initial_chunk_counter,
+                max,
+                input.len(),
+            );
+        }
+        // If we have some partial chunk bytes in the internal chunk_state, we
+        // need to finish that chunk first.
+        if self.chunk_state.count() > 0 {
+            let want = CHUNK_LEN - self.chunk_state.count();
+            let take = cmp::min(want, input.len());
+            self.chunk_state.update(&input[..take]);
+            input = &input[take..];
+            if !input.is_empty() {
+                // We've filled the current chunk, and there's more input
+                // coming, so we know it's not the root and we can finalize it.
+                // Then we'll proceed to hashing whole chunks below.
+                debug_assert_eq!(self.chunk_state.count(), CHUNK_LEN);
+                let chunk_cv = self.chunk_state.output().chaining_value();
+                self.push_cv(&chunk_cv, self.chunk_state.chunk_counter);
+                self.chunk_state = ChunkState::new(
+                    &self.key,
+                    self.chunk_state.chunk_counter + 1,
+                    self.chunk_state.flags,
+                    self.chunk_state.platform,
+                );
+            } else {
+                return self;
+            }
+        }
+
+        // Now the chunk_state is clear, and we have more input. If there's
+        // more than a single chunk (so, definitely not the root chunk), hash
+        // the largest whole subtree we can, with the full benefits of SIMD and
+        // multithreading parallelism. Two restrictions:
+        // - The subtree has to be a power-of-2 number of chunks. Only subtrees
+        //   along the right edge can be incomplete, and we don't know where
+        //   the right edge is going to be until we get to finalize().
+        // - The subtree must evenly divide the total number of chunks up until
+        //   this point (if total is not 0). If the current incomplete subtree
+        //   is only waiting for 1 more chunk, we can't hash a subtree of 4
+        //   chunks. We have to complete the current subtree first.
+        // Because we might need to break up the input to form powers of 2, or
+        // to evenly divide what we already have, this part runs in a loop.
+        while input.len() > CHUNK_LEN {
+            debug_assert_eq!(self.chunk_state.count(), 0, "no partial chunk data");
+            debug_assert_eq!(CHUNK_LEN.count_ones(), 1, "power of 2 chunk len");
+            let mut subtree_len = largest_power_of_two_leq(input.len());
+            let count_so_far = self.chunk_state.chunk_counter * CHUNK_LEN as u64;
+            // Shrink the subtree_len until it evenly divides the count so far.
+            // We know that subtree_len itself is a power of 2, so we can use a
+            // bitmasking trick instead of an actual remainder operation. (Note
+            // that if the caller consistently passes power-of-2 inputs of the
+            // same size, as is hopefully typical, this loop condition will
+            // always fail, and subtree_len will always be the full length of
+            // the input.)
+            //
+            // An aside: We don't have to shrink subtree_len quite this much.
+            // For example, if count_so_far is 1, we could pass 2 chunks to
+            // compress_subtree_to_parent_node. Since we'll get 2 CVs back,
+            // we'll still get the right answer in the end, and we might get to
+            // use 2-way SIMD parallelism. The problem with this optimization,
+            // is that it gets us stuck always hashing 2 chunks. The total
+            // number of chunks will remain odd, and we'll never graduate to
+            // higher degrees of parallelism. See
+            // https://github.com/BLAKE3-team/BLAKE3/issues/69.
+            while (subtree_len - 1) as u64 & count_so_far != 0 {
+                subtree_len /= 2;
+            }
+            // The shrunken subtree_len might now be 1 chunk long. If so, hash
+            // that one chunk by itself. Otherwise, compress the subtree into a
+            // pair of CVs.
+            let subtree_chunks = (subtree_len / CHUNK_LEN) as u64;
+            if subtree_len <= CHUNK_LEN {
+                debug_assert_eq!(subtree_len, CHUNK_LEN);
+                self.push_cv(
+                    &ChunkState::new(
+                        &self.key,
+                        self.chunk_state.chunk_counter,
+                        self.chunk_state.flags,
+                        self.chunk_state.platform,
+                    )
+                    .update(&input[..subtree_len])
+                    .output()
+                    .chaining_value(),
+                    self.chunk_state.chunk_counter,
+                );
+            } else {
+                // This is the high-performance happy path, though getting here
+                // depends on the caller giving us a long enough input.
+                let cv_pair = compress_subtree_to_parent_node::<J>(
+                    &input[..subtree_len],
+                    &self.key,
+                    self.chunk_state.chunk_counter,
+                    self.chunk_state.flags,
+                    self.chunk_state.platform,
+                );
+                let left_cv = array_ref!(cv_pair, 0, 32);
+                let right_cv = array_ref!(cv_pair, 32, 32);
+                // Push the two CVs we received into the CV stack in order. Because
+                // the stack merges lazily, this guarantees we aren't merging the
+                // root.
+                self.push_cv(left_cv, self.chunk_state.chunk_counter);
+                self.push_cv(
+                    right_cv,
+                    self.chunk_state.chunk_counter + (subtree_chunks / 2),
+                );
+            }
+            self.chunk_state.chunk_counter += subtree_chunks;
+            input = &input[subtree_len..];
+        }
+
+        // What remains is 1 chunk or less. Add it to the chunk state.
+        debug_assert!(input.len() <= CHUNK_LEN);
+        if !input.is_empty() {
+            self.chunk_state.update(input);
+            // Having added some input to the chunk_state, we know what's in
+            // the CV stack won't become the root node, and we can do an extra
+            // merge. This simplifies finalize().
+            self.merge_cv_stack(self.chunk_state.chunk_counter);
+        }
+
+        self
+    }
+
+    fn final_output(&self) -> Output {
+        // If the current chunk is the only chunk, that makes it the root node
+        // also. Convert it directly into an Output. Otherwise, we need to
+        // merge subtrees below.
+        if self.cv_stack.is_empty() {
+            debug_assert_eq!(self.chunk_state.chunk_counter, self.initial_chunk_counter);
+            return self.chunk_state.output();
+        }
+
+        // If there are any bytes in the ChunkState, finalize that chunk and
+        // merge its CV with everything in the CV stack. In that case, the work
+        // we did at the end of update() above guarantees that the stack
+        // doesn't contain any unmerged subtrees that need to be merged first.
+        // (This is important, because if there were two chunk hashes sitting
+        // on top of the stack, they would need to merge with each other, and
+        // merging a new chunk hash into them would be incorrect.)
+        //
+        // If there are no bytes in the ChunkState, we'll merge what's already
+        // in the stack. In this case it's fine if there are unmerged chunks on
+        // top, because we'll merge them with each other. Note that the case of
+        // the empty chunk is taken care of above.
+        let mut output: Output;
+        let mut num_cvs_remaining = self.cv_stack.len();
+        if self.chunk_state.count() > 0 {
+            debug_assert_eq!(
+                self.cv_stack.len(),
+                (self.chunk_state.chunk_counter - self.initial_chunk_counter).count_ones() as usize,
+                "cv stack does not need a merge",
+            );
+            output = self.chunk_state.output();
+        } else {
+            debug_assert!(self.cv_stack.len() >= 2);
+            output = parent_node_output(
+                &self.cv_stack[num_cvs_remaining - 2],
+                &self.cv_stack[num_cvs_remaining - 1],
+                &self.key,
+                self.chunk_state.flags,
+                self.chunk_state.platform,
+            );
+            num_cvs_remaining -= 2;
+        }
+        while num_cvs_remaining > 0 {
+            output = parent_node_output(
+                &self.cv_stack[num_cvs_remaining - 1],
+                &output.chaining_value(),
+                &self.key,
+                self.chunk_state.flags,
+                self.chunk_state.platform,
+            );
+            num_cvs_remaining -= 1;
+        }
+        output
+    }
+
+    /// Finalize the hash state and return the [`Hash`](struct.Hash.html) of
+    /// the input.
+    ///
+    /// This method is idempotent. Calling it twice will give the same result.
+    /// You can also add more input and finalize again.
+    pub fn finalize(&self) -> Hash {
+        assert_eq!(
+            self.initial_chunk_counter, 0,
+            "set_input_offset must be used with finalize_non_root",
+        );
+        self.final_output().root_hash()
+    }
+
+    /// Finalize the hash state and return an [`OutputReader`], which can
+    /// supply any number of output bytes.
+    ///
+    /// This method is idempotent. Calling it twice will give the same result.
+    /// You can also add more input and finalize again.
+    ///
+    /// [`OutputReader`]: struct.OutputReader.html
+    pub fn finalize_xof(&self) -> OutputReader {
+        assert_eq!(
+            self.initial_chunk_counter, 0,
+            "set_input_offset must be used with finalize_non_root",
+        );
+        OutputReader::new(self.final_output())
+    }
+
+    /// Return the total number of bytes hashed so far.
+    ///
+    /// [`hazmat::HasherExt::set_input_offset`] does not affect this value. This only counts bytes
+    /// passed to [`update`](Hasher::update).
+    pub fn count(&self) -> u64 {
+        // Account for non-zero cases of Hasher::set_input_offset. Note that initial_chunk_counter
+        // is always 0 for callers who don't use the hazmat module.
+        (self.chunk_state.chunk_counter - self.initial_chunk_counter) * CHUNK_LEN as u64
+            + self.chunk_state.count() as u64
+    }
+
+    /// As [`update`](Hasher::update), but reading from a
+    /// [`std::io::Read`](https://doc.rust-lang.org/std/io/trait.Read.html) implementation.
+    ///
+    /// [`Hasher`] implements
+    /// [`std::io::Write`](https://doc.rust-lang.org/std/io/trait.Write.html), so it's possible to
+    /// use [`std::io::copy`](https://doc.rust-lang.org/std/io/fn.copy.html) to update a [`Hasher`]
+    /// from any reader. Unfortunately, this standard approach can limit performance, because
+    /// `copy` currently uses an internal 8 KiB buffer that isn't big enough to take advantage of
+    /// all SIMD instruction sets. (In particular, [AVX-512](https://en.wikipedia.org/wiki/AVX-512)
+    /// needs a 16 KiB buffer.) `update_reader` avoids this performance problem and is slightly
+    /// more convenient.
+    ///
+    /// The internal buffer size this method uses may change at any time, and it may be different
+    /// for different targets. The only guarantee is that it will be large enough for all of this
+    /// crate's SIMD implementations on the current platform.
+    ///
+    /// The most common implementer of
+    /// [`std::io::Read`](https://doc.rust-lang.org/std/io/trait.Read.html) might be
+    /// [`std::fs::File`](https://doc.rust-lang.org/std/fs/struct.File.html), but note that memory
+    /// mapping can be faster than this method for hashing large files. See
+    /// [`update_mmap`](Hasher::update_mmap) and [`update_mmap_rayon`](Hasher::update_mmap_rayon),
+    /// which require the `mmap` and (for the latter) `rayon` Cargo features.
+    ///
+    /// This method requires the `std` Cargo feature, which is enabled by default.
+    ///
+    /// # Example
+    ///
+    /// ```no_run
+    /// # use std::fs::File;
+    /// # use std::io;
+    /// # fn main() -> io::Result<()> {
+    /// // Hash standard input.
+    /// let mut hasher = blake3::Hasher::new();
+    /// hasher.update_reader(std::io::stdin().lock())?;
+    /// println!("{}", hasher.finalize());
+    /// # Ok(())
+    /// # }
+    /// ```
+    #[cfg(feature = "std")]
+    pub fn update_reader(&mut self, reader: impl std::io::Read) -> std::io::Result<&mut Self> {
+        io::copy_wide(reader, self)?;
+        Ok(self)
+    }
+
+    /// As [`update`](Hasher::update), but using Rayon-based multithreading
+    /// internally.
+    ///
+    /// This method is gated by the `rayon` Cargo feature, which is disabled by
+    /// default but enabled on [docs.rs](https://docs.rs).
+    ///
+    /// To get any performance benefit from multithreading, the input buffer
+    /// needs to be large. As a rule of thumb on x86_64, `update_rayon` is
+    /// _slower_ than `update` for inputs under 128 KiB. That threshold varies
+    /// quite a lot across different processors, and it's important to benchmark
+    /// your specific use case. See also the performance warning associated with
+    /// [`update_mmap_rayon`](Hasher::update_mmap_rayon).
+    ///
+    /// If you already have a large buffer in memory, and you want to hash it
+    /// with multiple threads, this method is a good option. However, reading a
+    /// file into memory just to call this method can be a performance mistake,
+    /// both because it requires lots of memory and because single-threaded
+    /// reads can be slow. For hashing whole files, see
+    /// [`update_mmap_rayon`](Hasher::update_mmap_rayon), which is gated by both
+    /// the `rayon` and `mmap` Cargo features.
+    #[cfg(feature = "rayon")]
+    pub fn update_rayon(&mut self, input: &[u8]) -> &mut Self {
+        self.update_with_join::<join::RayonJoin>(input)
+    }
+
+    /// As [`update`](Hasher::update), but reading the contents of a file using memory mapping.
+    ///
+    /// Not all files can be memory mapped, and memory mapping small files can be slower than
+    /// reading them the usual way. In those cases, this method will fall back to standard file IO.
+    /// The heuristic for whether to use memory mapping is currently very simple (file size >=
+    /// 16 KiB), and it might change at any time.
+    ///
+    /// Like [`update`](Hasher::update), this method is single-threaded. In this author's
+    /// experience, memory mapping improves single-threaded performance by ~10% for large files
+    /// that are already in cache. This probably varies between platforms, and as always it's a
+    /// good idea to benchmark your own use case. In comparison, the multithreaded
+    /// [`update_mmap_rayon`](Hasher::update_mmap_rayon) method can have a much larger impact on
+    /// performance.
+    ///
+    /// There's a correctness reason that this method takes
+    /// [`Path`](https://doc.rust-lang.org/stable/std/path/struct.Path.html) instead of
+    /// [`File`](https://doc.rust-lang.org/std/fs/struct.File.html): reading from a memory-mapped
+    /// file ignores the seek position of the original file handle (it neither respects the current
+    /// position nor updates the position). This difference in behavior would've caused
+    /// `update_mmap` and [`update_reader`](Hasher::update_reader) to give different answers and
+    /// have different side effects in some cases. Taking a
+    /// [`Path`](https://doc.rust-lang.org/stable/std/path/struct.Path.html) avoids this problem by
+    /// making it clear that a new [`File`](https://doc.rust-lang.org/std/fs/struct.File.html) is
+    /// opened internally.
+    ///
+    /// This method requires the `mmap` Cargo feature, which is disabled by default but enabled on
+    /// [docs.rs](https://docs.rs).
+    ///
+    /// # Example
+    ///
+    /// ```no_run
+    /// # use std::io;
+    /// # use std::path::Path;
+    /// # fn main() -> io::Result<()> {
+    /// let path = Path::new("file.dat");
+    /// let mut hasher = blake3::Hasher::new();
+    /// hasher.update_mmap(path)?;
+    /// println!("{}", hasher.finalize());
+    /// # Ok(())
+    /// # }
+    /// ```
+    #[cfg(feature = "mmap")]
+    pub fn update_mmap(&mut self, path: impl AsRef<std::path::Path>) -> std::io::Result<&mut Self> {
+        let file = std::fs::File::open(path.as_ref())?;
+        if let Some(mmap) = io::maybe_mmap_file(&file)? {
+            self.update(&mmap);
+        } else {
+            io::copy_wide(&file, self)?;
+        }
+        Ok(self)
+    }
+
+    /// As [`update_rayon`](Hasher::update_rayon), but reading the contents of a file using
+    /// memory mapping. This is the default behavior of `b3sum`.
+    ///
+    /// For large files that are likely to be in cache, this can be much faster than
+    /// single-threaded hashing. When benchmarks report that BLAKE3 is 10x or 20x faster than other
+    /// cryptographic hashes, this is usually what they're measuring. However...
+    ///
+    /// **Performance Warning:** There are cases where multithreading hurts performance. The worst
+    /// case is [a large file on a spinning disk](https://github.com/BLAKE3-team/BLAKE3/issues/31),
+    /// where simultaneous reads from multiple threads can cause "thrashing" (i.e. the disk spends
+    /// more time seeking around than reading data). Windows tends to be somewhat worse about this,
+    /// in part because it's less likely than Linux to keep very large files in cache. More
+    /// generally, if your CPU cores are already busy, then multithreading will add overhead
+    /// without improving performance. If your code runs in different environments that you don't
+    /// control and can't measure, then unfortunately there's no one-size-fits-all answer for
+    /// whether multithreading is a good idea.
+    ///
+    /// The memory mapping behavior of this function is the same as
+    /// [`update_mmap`](Hasher::update_mmap), and the heuristic for when to fall back to standard
+    /// file IO might change at any time.
+    ///
+    /// This method requires both the `mmap` and `rayon` Cargo features, which are disabled by
+    /// default but enabled on [docs.rs](https://docs.rs).
+    ///
+    /// # Example
+    ///
+    /// ```no_run
+    /// # use std::io;
+    /// # use std::path::Path;
+    /// # fn main() -> io::Result<()> {
+    /// # #[cfg(feature = "rayon")]
+    /// # {
+    /// let path = Path::new("big_file.dat");
+    /// let mut hasher = blake3::Hasher::new();
+    /// hasher.update_mmap_rayon(path)?;
+    /// println!("{}", hasher.finalize());
+    /// # }
+    /// # Ok(())
+    /// # }
+    /// ```
+    #[cfg(feature = "mmap")]
+    #[cfg(feature = "rayon")]
+    pub fn update_mmap_rayon(
+        &mut self,
+        path: impl AsRef<std::path::Path>,
+    ) -> std::io::Result<&mut Self> {
+        let file = std::fs::File::open(path.as_ref())?;
+        if let Some(mmap) = io::maybe_mmap_file(&file)? {
+            self.update_rayon(&mmap);
+        } else {
+            io::copy_wide(&file, self)?;
+        }
+        Ok(self)
+    }
+}
+
+// Don't derive(Debug), because the state may be secret.
+impl fmt::Debug for Hasher {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        f.debug_struct("Hasher")
+            .field("flags", &self.chunk_state.flags)
+            .field("platform", &self.chunk_state.platform)
+            .finish()
+    }
+}
+
+impl Default for Hasher {
+    #[inline]
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+#[cfg(feature = "std")]
+impl std::io::Write for Hasher {
+    /// This is equivalent to [`update`](#method.update).
+    #[inline]
+    fn write(&mut self, input: &[u8]) -> std::io::Result<usize> {
+        self.update(input);
+        Ok(input.len())
+    }
+
+    #[inline]
+    fn flush(&mut self) -> std::io::Result<()> {
+        Ok(())
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Zeroize for Hasher {
+    fn zeroize(&mut self) {
+        // Destructuring to trigger compile error as a reminder to update this impl.
+        let Self {
+            key,
+            chunk_state,
+            initial_chunk_counter,
+            cv_stack,
+        } = self;
+
+        key.zeroize();
+        chunk_state.zeroize();
+        initial_chunk_counter.zeroize();
+        cv_stack.zeroize();
+    }
+}
+
+/// An incremental reader for extended output, returned by
+/// [`Hasher::finalize_xof`](struct.Hasher.html#method.finalize_xof).
+///
+/// Shorter BLAKE3 outputs are prefixes of longer ones, and explicitly requesting a short output is
+/// equivalent to truncating the default-length output. Note that this is a difference between
+/// BLAKE2 and BLAKE3.
+///
+/// # Security notes
+///
+/// Outputs shorter than the default length of 32 bytes (256 bits) provide less security. An N-bit
+/// BLAKE3 output is intended to provide N bits of first and second preimage resistance and N/2
+/// bits of collision resistance, for any N up to 256. Longer outputs don't provide any additional
+/// security.
+///
+/// Avoid relying on the secrecy of the output offset, that is, the number of output bytes read or
+/// the arguments to [`seek`](struct.OutputReader.html#method.seek) or
+/// [`set_position`](struct.OutputReader.html#method.set_position). [_Block-Cipher-Based Tree
+/// Hashing_ by Aldo Gunsing](https://eprint.iacr.org/2022/283) shows that an attacker who knows
+/// both the message and the key (if any) can easily determine the offset of an extended output.
+/// For comparison, AES-CTR has a similar property: if you know the key, you can decrypt a block
+/// from an unknown position in the output stream to recover its block index. Callers with strong
+/// secret keys aren't affected in practice, but secret offsets are a [design
+/// smell](https://en.wikipedia.org/wiki/Design_smell) in any case.
+#[derive(Clone)]
+pub struct OutputReader {
+    inner: Output,
+    position_within_block: u8,
+}
+
+impl OutputReader {
+    fn new(inner: Output) -> Self {
+        Self {
+            inner,
+            position_within_block: 0,
+        }
+    }
+
+    // This helper function handles both the case where the output buffer is
+    // shorter than one block, and the case where our position_within_block is
+    // non-zero.
+    fn fill_one_block(&mut self, buf: &mut &mut [u8]) {
+        let output_block: [u8; BLOCK_LEN] = self.inner.root_output_block();
+        let output_bytes = &output_block[self.position_within_block as usize..];
+        let take = cmp::min(buf.len(), output_bytes.len());
+        buf[..take].copy_from_slice(&output_bytes[..take]);
+        self.position_within_block += take as u8;
+        if self.position_within_block == BLOCK_LEN as u8 {
+            self.inner.counter += 1;
+            self.position_within_block = 0;
+        }
+        // Advance the dest buffer. mem::take() is a borrowck workaround.
+        *buf = &mut core::mem::take(buf)[take..];
+    }
+
+    /// Fill a buffer with output bytes and advance the position of the
+    /// `OutputReader`. This is equivalent to [`Read::read`], except that it
+    /// doesn't return a `Result`. Both methods always fill the entire buffer.
+    ///
+    /// Note that `OutputReader` doesn't buffer output bytes internally, so
+    /// calling `fill` repeatedly with a short-length or odd-length slice will
+    /// end up performing the same compression multiple times. If you're
+    /// reading output in a loop, prefer a slice length that's a multiple of
+    /// [`BLOCK_LEN`] (64 bytes).
+    ///
+    /// The maximum output size of BLAKE3 is 2<sup>64</sup>-1 bytes. If you try
+    /// to extract more than that, for example by seeking near the end and
+    /// reading further, the behavior is unspecified.
+    ///
+    /// [`Read::read`]: #method.read
+    pub fn fill(&mut self, mut buf: &mut [u8]) {
+        if buf.is_empty() {
+            return;
+        }
+
+        // If we're partway through a block, try to get to a block boundary.
+        if self.position_within_block != 0 {
+            self.fill_one_block(&mut buf);
+        }
+
+        let full_blocks = buf.len() / BLOCK_LEN;
+        let full_blocks_len = full_blocks * BLOCK_LEN;
+        if full_blocks > 0 {
+            debug_assert_eq!(0, self.position_within_block);
+            self.inner.platform.xof_many(
+                &self.inner.input_chaining_value,
+                &self.inner.block,
+                self.inner.block_len,
+                self.inner.counter,
+                self.inner.flags | ROOT,
+                &mut buf[..full_blocks_len],
+            );
+            self.inner.counter += full_blocks as u64;
+            buf = &mut buf[full_blocks * BLOCK_LEN..];
+        }
+
+        if !buf.is_empty() {
+            debug_assert!(buf.len() < BLOCK_LEN);
+            self.fill_one_block(&mut buf);
+            debug_assert!(buf.is_empty());
+        }
+    }
+
+    /// Return the current read position in the output stream. This is
+    /// equivalent to [`Seek::stream_position`], except that it doesn't return
+    /// a `Result`. The position of a new `OutputReader` starts at 0, and each
+    /// call to [`fill`] or [`Read::read`] moves the position forward by the
+    /// number of bytes read.
+    ///
+    /// [`Seek::stream_position`]: #method.stream_position
+    /// [`fill`]: #method.fill
+    /// [`Read::read`]: #method.read
+    pub fn position(&self) -> u64 {
+        self.inner.counter * BLOCK_LEN as u64 + self.position_within_block as u64
+    }
+
+    /// Seek to a new read position in the output stream. This is equivalent to
+    /// calling [`Seek::seek`] with [`SeekFrom::Start`], except that it doesn't
+    /// return a `Result`.
+    ///
+    /// [`Seek::seek`]: #method.seek
+    /// [`SeekFrom::Start`]: https://doc.rust-lang.org/std/io/enum.SeekFrom.html
+    pub fn set_position(&mut self, position: u64) {
+        self.position_within_block = (position % BLOCK_LEN as u64) as u8;
+        self.inner.counter = position / BLOCK_LEN as u64;
+    }
+}
+
+// Don't derive(Debug), because the state may be secret.
+impl fmt::Debug for OutputReader {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        f.debug_struct("OutputReader")
+            .field("position", &self.position())
+            .finish()
+    }
+}
+
+#[cfg(feature = "std")]
+impl std::io::Read for OutputReader {
+    #[inline]
+    fn read(&mut self, buf: &mut [u8]) -> std::io::Result<usize> {
+        self.fill(buf);
+        Ok(buf.len())
+    }
+}
+
+#[cfg(feature = "std")]
+impl std::io::Seek for OutputReader {
+    fn seek(&mut self, pos: std::io::SeekFrom) -> std::io::Result<u64> {
+        let max_position = u64::max_value() as i128;
+        let target_position: i128 = match pos {
+            std::io::SeekFrom::Start(x) => x as i128,
+            std::io::SeekFrom::Current(x) => self.position() as i128 + x as i128,
+            std::io::SeekFrom::End(_) => {
+                return Err(std::io::Error::new(
+                    std::io::ErrorKind::InvalidInput,
+                    "seek from end not supported",
+                ));
+            }
+        };
+        if target_position < 0 {
+            return Err(std::io::Error::new(
+                std::io::ErrorKind::InvalidInput,
+                "seek before start",
+            ));
+        }
+        self.set_position(cmp::min(target_position, max_position) as u64);
+        Ok(self.position())
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Zeroize for OutputReader {
+    fn zeroize(&mut self) {
+        // Destructuring to trigger compile error as a reminder to update this impl.
+        let Self {
+            inner,
+            position_within_block,
+        } = self;
+
+        inner.zeroize();
+        position_within_block.zeroize();
+    }
+}
diff --git a/thirdparty/blake3/src/platform.rs b/thirdparty/blake3/src/platform.rs
new file mode 100644
index 000000000..51b3b7b17
--- /dev/null
+++ b/thirdparty/blake3/src/platform.rs
@@ -0,0 +1,587 @@
+use crate::{portable, CVWords, IncrementCounter, BLOCK_LEN};
+use arrayref::{array_mut_ref, array_ref};
+
+cfg_if::cfg_if! {
+    if #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] {
+        cfg_if::cfg_if! {
+            if #[cfg(blake3_avx512_ffi)] {
+                pub const MAX_SIMD_DEGREE: usize = 16;
+            } else {
+                pub const MAX_SIMD_DEGREE: usize = 8;
+            }
+        }
+    } else if #[cfg(blake3_neon)] {
+        pub const MAX_SIMD_DEGREE: usize = 4;
+    } else if #[cfg(blake3_wasm32_simd)] {
+        pub const MAX_SIMD_DEGREE: usize = 4;
+    } else {
+        pub const MAX_SIMD_DEGREE: usize = 1;
+    }
+}
+
+// There are some places where we want a static size that's equal to the
+// MAX_SIMD_DEGREE, but also at least 2. Constant contexts aren't currently
+// allowed to use cmp::max, so we have to hardcode this additional constant
+// value. Get rid of this once cmp::max is a const fn.
+cfg_if::cfg_if! {
+    if #[cfg(any(target_arch = "x86", target_arch = "x86_64"))] {
+        cfg_if::cfg_if! {
+            if #[cfg(blake3_avx512_ffi)] {
+                pub const MAX_SIMD_DEGREE_OR_2: usize = 16;
+            } else {
+                pub const MAX_SIMD_DEGREE_OR_2: usize = 8;
+            }
+        }
+    } else if #[cfg(blake3_neon)] {
+        pub const MAX_SIMD_DEGREE_OR_2: usize = 4;
+    } else if #[cfg(blake3_wasm32_simd)] {
+        pub const MAX_SIMD_DEGREE_OR_2: usize = 4;
+    } else {
+        pub const MAX_SIMD_DEGREE_OR_2: usize = 2;
+    }
+}
+
+#[derive(Clone, Copy, Debug)]
+pub enum Platform {
+    Portable,
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    SSE2,
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    SSE41,
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    AVX2,
+    #[cfg(blake3_avx512_ffi)]
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    AVX512,
+    #[cfg(blake3_neon)]
+    NEON,
+    #[cfg(blake3_wasm32_simd)]
+    #[allow(non_camel_case_types)]
+    WASM32_SIMD,
+}
+
+impl Platform {
+    #[allow(unreachable_code)]
+    pub fn detect() -> Self {
+        #[cfg(miri)]
+        {
+            return Platform::Portable;
+        }
+
+        #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+        {
+            #[cfg(blake3_avx512_ffi)]
+            {
+                if avx512_detected() {
+                    return Platform::AVX512;
+                }
+            }
+            if avx2_detected() {
+                return Platform::AVX2;
+            }
+            if sse41_detected() {
+                return Platform::SSE41;
+            }
+            if sse2_detected() {
+                return Platform::SSE2;
+            }
+        }
+        // We don't use dynamic feature detection for NEON. If the "neon"
+        // feature is on, NEON is assumed to be supported.
+        #[cfg(blake3_neon)]
+        {
+            return Platform::NEON;
+        }
+        #[cfg(blake3_wasm32_simd)]
+        {
+            return Platform::WASM32_SIMD;
+        }
+        Platform::Portable
+    }
+
+    pub fn simd_degree(&self) -> usize {
+        let degree = match self {
+            Platform::Portable => 1,
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE2 => 4,
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE41 => 4,
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::AVX2 => 8,
+            #[cfg(blake3_avx512_ffi)]
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::AVX512 => 16,
+            #[cfg(blake3_neon)]
+            Platform::NEON => 4,
+            #[cfg(blake3_wasm32_simd)]
+            Platform::WASM32_SIMD => 4,
+        };
+        debug_assert!(degree <= MAX_SIMD_DEGREE);
+        degree
+    }
+
+    pub fn compress_in_place(
+        &self,
+        cv: &mut CVWords,
+        block: &[u8; BLOCK_LEN],
+        block_len: u8,
+        counter: u64,
+        flags: u8,
+    ) {
+        match self {
+            Platform::Portable => portable::compress_in_place(cv, block, block_len, counter, flags),
+            // Safe because detect() checked for platform support.
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE2 => unsafe {
+                crate::sse2::compress_in_place(cv, block, block_len, counter, flags)
+            },
+            // Safe because detect() checked for platform support.
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE41 | Platform::AVX2 => unsafe {
+                crate::sse41::compress_in_place(cv, block, block_len, counter, flags)
+            },
+            // Safe because detect() checked for platform support.
+            #[cfg(blake3_avx512_ffi)]
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::AVX512 => unsafe {
+                crate::avx512::compress_in_place(cv, block, block_len, counter, flags)
+            },
+            // No NEON compress_in_place() implementation yet.
+            #[cfg(blake3_neon)]
+            Platform::NEON => portable::compress_in_place(cv, block, block_len, counter, flags),
+            #[cfg(blake3_wasm32_simd)]
+            Platform::WASM32_SIMD => {
+                crate::wasm32_simd::compress_in_place(cv, block, block_len, counter, flags)
+            }
+        }
+    }
+
+    pub fn compress_xof(
+        &self,
+        cv: &CVWords,
+        block: &[u8; BLOCK_LEN],
+        block_len: u8,
+        counter: u64,
+        flags: u8,
+    ) -> [u8; 64] {
+        match self {
+            Platform::Portable => portable::compress_xof(cv, block, block_len, counter, flags),
+            // Safe because detect() checked for platform support.
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE2 => unsafe {
+                crate::sse2::compress_xof(cv, block, block_len, counter, flags)
+            },
+            // Safe because detect() checked for platform support.
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE41 | Platform::AVX2 => unsafe {
+                crate::sse41::compress_xof(cv, block, block_len, counter, flags)
+            },
+            // Safe because detect() checked for platform support.
+            #[cfg(blake3_avx512_ffi)]
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::AVX512 => unsafe {
+                crate::avx512::compress_xof(cv, block, block_len, counter, flags)
+            },
+            // No NEON compress_xof() implementation yet.
+            #[cfg(blake3_neon)]
+            Platform::NEON => portable::compress_xof(cv, block, block_len, counter, flags),
+            #[cfg(blake3_wasm32_simd)]
+            Platform::WASM32_SIMD => {
+                crate::wasm32_simd::compress_xof(cv, block, block_len, counter, flags)
+            }
+        }
+    }
+
+    // IMPLEMENTATION NOTE
+    // ===================
+    // hash_many() applies two optimizations. The critically important
+    // optimization is the high-performance parallel SIMD hashing mode,
+    // described in detail in the spec. This more than doubles throughput per
+    // thread. Another optimization is keeping the state vectors transposed
+    // from block to block within a chunk. When state vectors are transposed
+    // after every block, there's a small but measurable performance loss.
+    // Compressing chunks with a dedicated loop avoids this.
+
+    pub fn hash_many<const N: usize>(
+        &self,
+        inputs: &[&[u8; N]],
+        key: &CVWords,
+        counter: u64,
+        increment_counter: IncrementCounter,
+        flags: u8,
+        flags_start: u8,
+        flags_end: u8,
+        out: &mut [u8],
+    ) {
+        match self {
+            Platform::Portable => portable::hash_many(
+                inputs,
+                key,
+                counter,
+                increment_counter,
+                flags,
+                flags_start,
+                flags_end,
+                out,
+            ),
+            // Safe because detect() checked for platform support.
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE2 => unsafe {
+                crate::sse2::hash_many(
+                    inputs,
+                    key,
+                    counter,
+                    increment_counter,
+                    flags,
+                    flags_start,
+                    flags_end,
+                    out,
+                )
+            },
+            // Safe because detect() checked for platform support.
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::SSE41 => unsafe {
+                crate::sse41::hash_many(
+                    inputs,
+                    key,
+                    counter,
+                    increment_counter,
+                    flags,
+                    flags_start,
+                    flags_end,
+                    out,
+                )
+            },
+            // Safe because detect() checked for platform support.
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::AVX2 => unsafe {
+                crate::avx2::hash_many(
+                    inputs,
+                    key,
+                    counter,
+                    increment_counter,
+                    flags,
+                    flags_start,
+                    flags_end,
+                    out,
+                )
+            },
+            // Safe because detect() checked for platform support.
+            #[cfg(blake3_avx512_ffi)]
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::AVX512 => unsafe {
+                crate::avx512::hash_many(
+                    inputs,
+                    key,
+                    counter,
+                    increment_counter,
+                    flags,
+                    flags_start,
+                    flags_end,
+                    out,
+                )
+            },
+            // Assumed to be safe if the "neon" feature is on.
+            #[cfg(blake3_neon)]
+            Platform::NEON => unsafe {
+                crate::neon::hash_many(
+                    inputs,
+                    key,
+                    counter,
+                    increment_counter,
+                    flags,
+                    flags_start,
+                    flags_end,
+                    out,
+                )
+            },
+            // Assumed to be safe if the "wasm32_simd" feature is on.
+            #[cfg(blake3_wasm32_simd)]
+            Platform::WASM32_SIMD => unsafe {
+                crate::wasm32_simd::hash_many(
+                    inputs,
+                    key,
+                    counter,
+                    increment_counter,
+                    flags,
+                    flags_start,
+                    flags_end,
+                    out,
+                )
+            },
+        }
+    }
+
+    pub fn xof_many(
+        &self,
+        cv: &CVWords,
+        block: &[u8; BLOCK_LEN],
+        block_len: u8,
+        mut counter: u64,
+        flags: u8,
+        out: &mut [u8],
+    ) {
+        debug_assert_eq!(0, out.len() % BLOCK_LEN, "whole blocks only");
+        if out.is_empty() {
+            // The current assembly implementation always outputs at least 1 block.
+            return;
+        }
+        match self {
+            // Safe because detect() checked for platform support.
+            #[cfg(blake3_avx512_ffi)]
+            #[cfg(unix)]
+            #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+            Platform::AVX512 => unsafe {
+                crate::avx512::xof_many(cv, block, block_len, counter, flags, out)
+            },
+            _ => {
+                // For platforms without an optimized xof_many, fall back to a loop over
+                // compress_xof. This is still faster than portable code.
+                for out_block in out.chunks_exact_mut(BLOCK_LEN) {
+                    // TODO: Use array_chunks_mut here once that's stable.
+                    let out_array: &mut [u8; BLOCK_LEN] = out_block.try_into().unwrap();
+                    *out_array = self.compress_xof(cv, block, block_len, counter, flags);
+                    counter += 1;
+                }
+            }
+        }
+    }
+
+    // Explicit platform constructors, for benchmarks.
+
+    pub fn portable() -> Self {
+        Self::Portable
+    }
+
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    pub fn sse2() -> Option<Self> {
+        if sse2_detected() {
+            Some(Self::SSE2)
+        } else {
+            None
+        }
+    }
+
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    pub fn sse41() -> Option<Self> {
+        if sse41_detected() {
+            Some(Self::SSE41)
+        } else {
+            None
+        }
+    }
+
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    pub fn avx2() -> Option<Self> {
+        if avx2_detected() {
+            Some(Self::AVX2)
+        } else {
+            None
+        }
+    }
+
+    #[cfg(blake3_avx512_ffi)]
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    pub fn avx512() -> Option<Self> {
+        if avx512_detected() {
+            Some(Self::AVX512)
+        } else {
+            None
+        }
+    }
+
+    #[cfg(blake3_neon)]
+    pub fn neon() -> Option<Self> {
+        // Assumed to be safe if the "neon" feature is on.
+        Some(Self::NEON)
+    }
+
+    #[cfg(blake3_wasm32_simd)]
+    pub fn wasm32_simd() -> Option<Self> {
+        // Assumed to be safe if the "wasm32_simd" feature is on.
+        Some(Self::WASM32_SIMD)
+    }
+}
+
+// Note that AVX-512 is divided into multiple featuresets, and we use two of
+// them, F and VL.
+#[cfg(blake3_avx512_ffi)]
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[inline(always)]
+#[allow(unreachable_code)]
+pub fn avx512_detected() -> bool {
+    if cfg!(miri) {
+        return false;
+    }
+
+    // A testing-only short-circuit.
+    if cfg!(feature = "no_avx512") {
+        return false;
+    }
+    // Static check, e.g. for building with target-cpu=native.
+    #[cfg(all(target_feature = "avx512f", target_feature = "avx512vl"))]
+    {
+        return true;
+    }
+    // Dynamic check, if std is enabled.
+    #[cfg(feature = "std")]
+    {
+        if is_x86_feature_detected!("avx512f") && is_x86_feature_detected!("avx512vl") {
+            return true;
+        }
+    }
+    false
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[inline(always)]
+#[allow(unreachable_code)]
+pub fn avx2_detected() -> bool {
+    if cfg!(miri) {
+        return false;
+    }
+
+    // A testing-only short-circuit.
+    if cfg!(feature = "no_avx2") {
+        return false;
+    }
+    // Static check, e.g. for building with target-cpu=native.
+    #[cfg(target_feature = "avx2")]
+    {
+        return true;
+    }
+    // Dynamic check, if std is enabled.
+    #[cfg(feature = "std")]
+    {
+        if is_x86_feature_detected!("avx2") {
+            return true;
+        }
+    }
+    false
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[inline(always)]
+#[allow(unreachable_code)]
+pub fn sse41_detected() -> bool {
+    if cfg!(miri) {
+        return false;
+    }
+
+    // A testing-only short-circuit.
+    if cfg!(feature = "no_sse41") {
+        return false;
+    }
+    // Static check, e.g. for building with target-cpu=native.
+    #[cfg(target_feature = "sse4.1")]
+    {
+        return true;
+    }
+    // Dynamic check, if std is enabled.
+    #[cfg(feature = "std")]
+    {
+        if is_x86_feature_detected!("sse4.1") {
+            return true;
+        }
+    }
+    false
+}
+
+#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+#[inline(always)]
+#[allow(unreachable_code)]
+pub fn sse2_detected() -> bool {
+    if cfg!(miri) {
+        return false;
+    }
+
+    // A testing-only short-circuit.
+    if cfg!(feature = "no_sse2") {
+        return false;
+    }
+    // Static check, e.g. for building with target-cpu=native.
+    #[cfg(target_feature = "sse2")]
+    {
+        return true;
+    }
+    // Dynamic check, if std is enabled.
+    #[cfg(feature = "std")]
+    {
+        if is_x86_feature_detected!("sse2") {
+            return true;
+        }
+    }
+    false
+}
+
+#[inline(always)]
+pub fn words_from_le_bytes_32(bytes: &[u8; 32]) -> [u32; 8] {
+    let mut out = [0; 8];
+    out[0] = u32::from_le_bytes(*array_ref!(bytes, 0 * 4, 4));
+    out[1] = u32::from_le_bytes(*array_ref!(bytes, 1 * 4, 4));
+    out[2] = u32::from_le_bytes(*array_ref!(bytes, 2 * 4, 4));
+    out[3] = u32::from_le_bytes(*array_ref!(bytes, 3 * 4, 4));
+    out[4] = u32::from_le_bytes(*array_ref!(bytes, 4 * 4, 4));
+    out[5] = u32::from_le_bytes(*array_ref!(bytes, 5 * 4, 4));
+    out[6] = u32::from_le_bytes(*array_ref!(bytes, 6 * 4, 4));
+    out[7] = u32::from_le_bytes(*array_ref!(bytes, 7 * 4, 4));
+    out
+}
+
+#[inline(always)]
+pub fn words_from_le_bytes_64(bytes: &[u8; 64]) -> [u32; 16] {
+    let mut out = [0; 16];
+    out[0] = u32::from_le_bytes(*array_ref!(bytes, 0 * 4, 4));
+    out[1] = u32::from_le_bytes(*array_ref!(bytes, 1 * 4, 4));
+    out[2] = u32::from_le_bytes(*array_ref!(bytes, 2 * 4, 4));
+    out[3] = u32::from_le_bytes(*array_ref!(bytes, 3 * 4, 4));
+    out[4] = u32::from_le_bytes(*array_ref!(bytes, 4 * 4, 4));
+    out[5] = u32::from_le_bytes(*array_ref!(bytes, 5 * 4, 4));
+    out[6] = u32::from_le_bytes(*array_ref!(bytes, 6 * 4, 4));
+    out[7] = u32::from_le_bytes(*array_ref!(bytes, 7 * 4, 4));
+    out[8] = u32::from_le_bytes(*array_ref!(bytes, 8 * 4, 4));
+    out[9] = u32::from_le_bytes(*array_ref!(bytes, 9 * 4, 4));
+    out[10] = u32::from_le_bytes(*array_ref!(bytes, 10 * 4, 4));
+    out[11] = u32::from_le_bytes(*array_ref!(bytes, 11 * 4, 4));
+    out[12] = u32::from_le_bytes(*array_ref!(bytes, 12 * 4, 4));
+    out[13] = u32::from_le_bytes(*array_ref!(bytes, 13 * 4, 4));
+    out[14] = u32::from_le_bytes(*array_ref!(bytes, 14 * 4, 4));
+    out[15] = u32::from_le_bytes(*array_ref!(bytes, 15 * 4, 4));
+    out
+}
+
+#[inline(always)]
+pub fn le_bytes_from_words_32(words: &[u32; 8]) -> [u8; 32] {
+    let mut out = [0; 32];
+    *array_mut_ref!(out, 0 * 4, 4) = words[0].to_le_bytes();
+    *array_mut_ref!(out, 1 * 4, 4) = words[1].to_le_bytes();
+    *array_mut_ref!(out, 2 * 4, 4) = words[2].to_le_bytes();
+    *array_mut_ref!(out, 3 * 4, 4) = words[3].to_le_bytes();
+    *array_mut_ref!(out, 4 * 4, 4) = words[4].to_le_bytes();
+    *array_mut_ref!(out, 5 * 4, 4) = words[5].to_le_bytes();
+    *array_mut_ref!(out, 6 * 4, 4) = words[6].to_le_bytes();
+    *array_mut_ref!(out, 7 * 4, 4) = words[7].to_le_bytes();
+    out
+}
+
+#[inline(always)]
+pub fn le_bytes_from_words_64(words: &[u32; 16]) -> [u8; 64] {
+    let mut out = [0; 64];
+    *array_mut_ref!(out, 0 * 4, 4) = words[0].to_le_bytes();
+    *array_mut_ref!(out, 1 * 4, 4) = words[1].to_le_bytes();
+    *array_mut_ref!(out, 2 * 4, 4) = words[2].to_le_bytes();
+    *array_mut_ref!(out, 3 * 4, 4) = words[3].to_le_bytes();
+    *array_mut_ref!(out, 4 * 4, 4) = words[4].to_le_bytes();
+    *array_mut_ref!(out, 5 * 4, 4) = words[5].to_le_bytes();
+    *array_mut_ref!(out, 6 * 4, 4) = words[6].to_le_bytes();
+    *array_mut_ref!(out, 7 * 4, 4) = words[7].to_le_bytes();
+    *array_mut_ref!(out, 8 * 4, 4) = words[8].to_le_bytes();
+    *array_mut_ref!(out, 9 * 4, 4) = words[9].to_le_bytes();
+    *array_mut_ref!(out, 10 * 4, 4) = words[10].to_le_bytes();
+    *array_mut_ref!(out, 11 * 4, 4) = words[11].to_le_bytes();
+    *array_mut_ref!(out, 12 * 4, 4) = words[12].to_le_bytes();
+    *array_mut_ref!(out, 13 * 4, 4) = words[13].to_le_bytes();
+    *array_mut_ref!(out, 14 * 4, 4) = words[14].to_le_bytes();
+    *array_mut_ref!(out, 15 * 4, 4) = words[15].to_le_bytes();
+    out
+}
diff --git a/thirdparty/blake3/src/portable.rs b/thirdparty/blake3/src/portable.rs
new file mode 100644
index 000000000..7af6828b0
--- /dev/null
+++ b/thirdparty/blake3/src/portable.rs
@@ -0,0 +1,198 @@
+use crate::{
+    counter_high, counter_low, CVBytes, CVWords, IncrementCounter, BLOCK_LEN, IV, MSG_SCHEDULE,
+    OUT_LEN,
+};
+use arrayref::{array_mut_ref, array_ref};
+
+#[inline(always)]
+fn g(state: &mut [u32; 16], a: usize, b: usize, c: usize, d: usize, x: u32, y: u32) {
+    state[a] = state[a].wrapping_add(state[b]).wrapping_add(x);
+    state[d] = (state[d] ^ state[a]).rotate_right(16);
+    state[c] = state[c].wrapping_add(state[d]);
+    state[b] = (state[b] ^ state[c]).rotate_right(12);
+    state[a] = state[a].wrapping_add(state[b]).wrapping_add(y);
+    state[d] = (state[d] ^ state[a]).rotate_right(8);
+    state[c] = state[c].wrapping_add(state[d]);
+    state[b] = (state[b] ^ state[c]).rotate_right(7);
+}
+
+#[inline(always)]
+fn round(state: &mut [u32; 16], msg: &[u32; 16], round: usize) {
+    // Select the message schedule based on the round.
+    let schedule = MSG_SCHEDULE[round];
+
+    // Mix the columns.
+    g(state, 0, 4, 8, 12, msg[schedule[0]], msg[schedule[1]]);
+    g(state, 1, 5, 9, 13, msg[schedule[2]], msg[schedule[3]]);
+    g(state, 2, 6, 10, 14, msg[schedule[4]], msg[schedule[5]]);
+    g(state, 3, 7, 11, 15, msg[schedule[6]], msg[schedule[7]]);
+
+    // Mix the diagonals.
+    g(state, 0, 5, 10, 15, msg[schedule[8]], msg[schedule[9]]);
+    g(state, 1, 6, 11, 12, msg[schedule[10]], msg[schedule[11]]);
+    g(state, 2, 7, 8, 13, msg[schedule[12]], msg[schedule[13]]);
+    g(state, 3, 4, 9, 14, msg[schedule[14]], msg[schedule[15]]);
+}
+
+#[inline(always)]
+fn compress_pre(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u32; 16] {
+    let block_words = crate::platform::words_from_le_bytes_64(block);
+
+    let mut state = [
+        cv[0],
+        cv[1],
+        cv[2],
+        cv[3],
+        cv[4],
+        cv[5],
+        cv[6],
+        cv[7],
+        IV[0],
+        IV[1],
+        IV[2],
+        IV[3],
+        counter_low(counter),
+        counter_high(counter),
+        block_len as u32,
+        flags as u32,
+    ];
+
+    round(&mut state, &block_words, 0);
+    round(&mut state, &block_words, 1);
+    round(&mut state, &block_words, 2);
+    round(&mut state, &block_words, 3);
+    round(&mut state, &block_words, 4);
+    round(&mut state, &block_words, 5);
+    round(&mut state, &block_words, 6);
+
+    state
+}
+
+pub fn compress_in_place(
+    cv: &mut CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    let state = compress_pre(cv, block, block_len, counter, flags);
+
+    cv[0] = state[0] ^ state[8];
+    cv[1] = state[1] ^ state[9];
+    cv[2] = state[2] ^ state[10];
+    cv[3] = state[3] ^ state[11];
+    cv[4] = state[4] ^ state[12];
+    cv[5] = state[5] ^ state[13];
+    cv[6] = state[6] ^ state[14];
+    cv[7] = state[7] ^ state[15];
+}
+
+pub fn compress_xof(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64] {
+    let mut state = compress_pre(cv, block, block_len, counter, flags);
+    state[0] ^= state[8];
+    state[1] ^= state[9];
+    state[2] ^= state[10];
+    state[3] ^= state[11];
+    state[4] ^= state[12];
+    state[5] ^= state[13];
+    state[6] ^= state[14];
+    state[7] ^= state[15];
+    state[8] ^= cv[0];
+    state[9] ^= cv[1];
+    state[10] ^= cv[2];
+    state[11] ^= cv[3];
+    state[12] ^= cv[4];
+    state[13] ^= cv[5];
+    state[14] ^= cv[6];
+    state[15] ^= cv[7];
+    crate::platform::le_bytes_from_words_64(&state)
+}
+
+pub fn hash1<const N: usize>(
+    input: &[u8; N],
+    key: &CVWords,
+    counter: u64,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut CVBytes,
+) {
+    debug_assert_eq!(N % BLOCK_LEN, 0, "uneven blocks");
+    let mut cv = *key;
+    let mut block_flags = flags | flags_start;
+    let mut slice = &input[..];
+    while slice.len() >= BLOCK_LEN {
+        if slice.len() == BLOCK_LEN {
+            block_flags |= flags_end;
+        }
+        compress_in_place(
+            &mut cv,
+            array_ref!(slice, 0, BLOCK_LEN),
+            BLOCK_LEN as u8,
+            counter,
+            block_flags,
+        );
+        block_flags = flags;
+        slice = &slice[BLOCK_LEN..];
+    }
+    *out = crate::platform::le_bytes_from_words_32(&cv);
+}
+
+pub fn hash_many<const N: usize>(
+    inputs: &[&[u8; N]],
+    key: &CVWords,
+    mut counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8],
+) {
+    debug_assert!(out.len() >= inputs.len() * OUT_LEN, "out too short");
+    for (&input, output) in inputs.iter().zip(out.chunks_exact_mut(OUT_LEN)) {
+        hash1(
+            input,
+            key,
+            counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(output, 0, OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += 1;
+        }
+    }
+}
+
+#[cfg(test)]
+pub mod test {
+    use super::*;
+
+    // This is basically testing the portable implementation against itself,
+    // but it also checks that compress_in_place and compress_xof are
+    // consistent. And there are tests against the reference implementation and
+    // against hardcoded test vectors elsewhere.
+    #[test]
+    fn test_compress() {
+        crate::test::test_compress_fn(compress_in_place, compress_xof);
+    }
+
+    // Ditto.
+    #[test]
+    fn test_hash_many() {
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/rust_avx2.rs b/thirdparty/blake3/src/rust_avx2.rs
new file mode 100644
index 000000000..a37a4caac
--- /dev/null
+++ b/thirdparty/blake3/src/rust_avx2.rs
@@ -0,0 +1,474 @@
+#[cfg(target_arch = "x86")]
+use core::arch::x86::*;
+#[cfg(target_arch = "x86_64")]
+use core::arch::x86_64::*;
+
+use crate::{
+    counter_high, counter_low, CVWords, IncrementCounter, BLOCK_LEN, IV, MSG_SCHEDULE, OUT_LEN,
+};
+use arrayref::{array_mut_ref, mut_array_refs};
+
+pub const DEGREE: usize = 8;
+
+#[inline(always)]
+unsafe fn loadu(src: *const u8) -> __m256i {
+    // This is an unaligned load, so the pointer cast is allowed.
+    _mm256_loadu_si256(src as *const __m256i)
+}
+
+#[inline(always)]
+unsafe fn storeu(src: __m256i, dest: *mut u8) {
+    // This is an unaligned store, so the pointer cast is allowed.
+    _mm256_storeu_si256(dest as *mut __m256i, src)
+}
+
+#[inline(always)]
+unsafe fn add(a: __m256i, b: __m256i) -> __m256i {
+    _mm256_add_epi32(a, b)
+}
+
+#[inline(always)]
+unsafe fn xor(a: __m256i, b: __m256i) -> __m256i {
+    _mm256_xor_si256(a, b)
+}
+
+#[inline(always)]
+unsafe fn set1(x: u32) -> __m256i {
+    _mm256_set1_epi32(x as i32)
+}
+
+#[inline(always)]
+unsafe fn set8(a: u32, b: u32, c: u32, d: u32, e: u32, f: u32, g: u32, h: u32) -> __m256i {
+    _mm256_setr_epi32(
+        a as i32, b as i32, c as i32, d as i32, e as i32, f as i32, g as i32, h as i32,
+    )
+}
+
+// These rotations are the "simple/shifts version". For the
+// "complicated/shuffles version", see
+// https://github.com/sneves/blake2-avx2/blob/b3723921f668df09ece52dcd225a36d4a4eea1d9/blake2s-common.h#L63-L66.
+// For a discussion of the tradeoffs, see
+// https://github.com/sneves/blake2-avx2/pull/5. Due to an LLVM bug
+// (https://bugs.llvm.org/show_bug.cgi?id=44379), this version performs better
+// on recent x86 chips.
+
+#[inline(always)]
+unsafe fn rot16(x: __m256i) -> __m256i {
+    _mm256_or_si256(_mm256_srli_epi32(x, 16), _mm256_slli_epi32(x, 32 - 16))
+}
+
+#[inline(always)]
+unsafe fn rot12(x: __m256i) -> __m256i {
+    _mm256_or_si256(_mm256_srli_epi32(x, 12), _mm256_slli_epi32(x, 32 - 12))
+}
+
+#[inline(always)]
+unsafe fn rot8(x: __m256i) -> __m256i {
+    _mm256_or_si256(_mm256_srli_epi32(x, 8), _mm256_slli_epi32(x, 32 - 8))
+}
+
+#[inline(always)]
+unsafe fn rot7(x: __m256i) -> __m256i {
+    _mm256_or_si256(_mm256_srli_epi32(x, 7), _mm256_slli_epi32(x, 32 - 7))
+}
+
+#[inline(always)]
+unsafe fn round(v: &mut [__m256i; 16], m: &[__m256i; 16], r: usize) {
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][0] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][2] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][4] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][6] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[15] = rot16(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot12(v[4]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][1] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][3] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][5] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][7] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[15] = rot8(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot7(v[4]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][8] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][10] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][12] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][14] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot16(v[15]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[4] = rot12(v[4]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][9] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][11] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][13] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][15] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot8(v[15]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+    v[4] = rot7(v[4]);
+}
+
+#[inline(always)]
+unsafe fn interleave128(a: __m256i, b: __m256i) -> (__m256i, __m256i) {
+    (
+        _mm256_permute2x128_si256(a, b, 0x20),
+        _mm256_permute2x128_si256(a, b, 0x31),
+    )
+}
+
+// There are several ways to do a transposition. We could do it naively, with 8 separate
+// _mm256_set_epi32 instructions, referencing each of the 32 words explicitly. Or we could copy
+// the vecs into contiguous storage and then use gather instructions. This third approach is to use
+// a series of unpack instructions to interleave the vectors. In my benchmarks, interleaving is the
+// fastest approach. To test this, run `cargo +nightly bench --bench libtest load_8` in the
+// https://github.com/oconnor663/bao_experiments repo.
+#[inline(always)]
+unsafe fn transpose_vecs(vecs: &mut [__m256i; DEGREE]) {
+    // Interleave 32-bit lanes. The low unpack is lanes 00/11/44/55, and the high is 22/33/66/77.
+    let ab_0145 = _mm256_unpacklo_epi32(vecs[0], vecs[1]);
+    let ab_2367 = _mm256_unpackhi_epi32(vecs[0], vecs[1]);
+    let cd_0145 = _mm256_unpacklo_epi32(vecs[2], vecs[3]);
+    let cd_2367 = _mm256_unpackhi_epi32(vecs[2], vecs[3]);
+    let ef_0145 = _mm256_unpacklo_epi32(vecs[4], vecs[5]);
+    let ef_2367 = _mm256_unpackhi_epi32(vecs[4], vecs[5]);
+    let gh_0145 = _mm256_unpacklo_epi32(vecs[6], vecs[7]);
+    let gh_2367 = _mm256_unpackhi_epi32(vecs[6], vecs[7]);
+
+    // Interleave 64-bit lanes. The low unpack is lanes 00/22 and the high is 11/33.
+    let abcd_04 = _mm256_unpacklo_epi64(ab_0145, cd_0145);
+    let abcd_15 = _mm256_unpackhi_epi64(ab_0145, cd_0145);
+    let abcd_26 = _mm256_unpacklo_epi64(ab_2367, cd_2367);
+    let abcd_37 = _mm256_unpackhi_epi64(ab_2367, cd_2367);
+    let efgh_04 = _mm256_unpacklo_epi64(ef_0145, gh_0145);
+    let efgh_15 = _mm256_unpackhi_epi64(ef_0145, gh_0145);
+    let efgh_26 = _mm256_unpacklo_epi64(ef_2367, gh_2367);
+    let efgh_37 = _mm256_unpackhi_epi64(ef_2367, gh_2367);
+
+    // Interleave 128-bit lanes.
+    let (abcdefgh_0, abcdefgh_4) = interleave128(abcd_04, efgh_04);
+    let (abcdefgh_1, abcdefgh_5) = interleave128(abcd_15, efgh_15);
+    let (abcdefgh_2, abcdefgh_6) = interleave128(abcd_26, efgh_26);
+    let (abcdefgh_3, abcdefgh_7) = interleave128(abcd_37, efgh_37);
+
+    vecs[0] = abcdefgh_0;
+    vecs[1] = abcdefgh_1;
+    vecs[2] = abcdefgh_2;
+    vecs[3] = abcdefgh_3;
+    vecs[4] = abcdefgh_4;
+    vecs[5] = abcdefgh_5;
+    vecs[6] = abcdefgh_6;
+    vecs[7] = abcdefgh_7;
+}
+
+#[inline(always)]
+unsafe fn transpose_msg_vecs(inputs: &[*const u8; DEGREE], block_offset: usize) -> [__m256i; 16] {
+    let mut vecs = [
+        loadu(inputs[0].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[4].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[5].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[6].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[7].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[4].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[5].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[6].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[7].add(block_offset + 1 * 4 * DEGREE)),
+    ];
+    for i in 0..DEGREE {
+        _mm_prefetch(inputs[i].add(block_offset + 256) as *const i8, _MM_HINT_T0);
+    }
+    let squares = mut_array_refs!(&mut vecs, DEGREE, DEGREE);
+    transpose_vecs(squares.0);
+    transpose_vecs(squares.1);
+    vecs
+}
+
+#[inline(always)]
+unsafe fn load_counters(counter: u64, increment_counter: IncrementCounter) -> (__m256i, __m256i) {
+    let mask = if increment_counter.yes() { !0 } else { 0 };
+    (
+        set8(
+            counter_low(counter + (mask & 0)),
+            counter_low(counter + (mask & 1)),
+            counter_low(counter + (mask & 2)),
+            counter_low(counter + (mask & 3)),
+            counter_low(counter + (mask & 4)),
+            counter_low(counter + (mask & 5)),
+            counter_low(counter + (mask & 6)),
+            counter_low(counter + (mask & 7)),
+        ),
+        set8(
+            counter_high(counter + (mask & 0)),
+            counter_high(counter + (mask & 1)),
+            counter_high(counter + (mask & 2)),
+            counter_high(counter + (mask & 3)),
+            counter_high(counter + (mask & 4)),
+            counter_high(counter + (mask & 5)),
+            counter_high(counter + (mask & 6)),
+            counter_high(counter + (mask & 7)),
+        ),
+    )
+}
+
+#[target_feature(enable = "avx2")]
+pub unsafe fn hash8(
+    inputs: &[*const u8; DEGREE],
+    blocks: usize,
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8; DEGREE * OUT_LEN],
+) {
+    let mut h_vecs = [
+        set1(key[0]),
+        set1(key[1]),
+        set1(key[2]),
+        set1(key[3]),
+        set1(key[4]),
+        set1(key[5]),
+        set1(key[6]),
+        set1(key[7]),
+    ];
+    let (counter_low_vec, counter_high_vec) = load_counters(counter, increment_counter);
+    let mut block_flags = flags | flags_start;
+
+    for block in 0..blocks {
+        if block + 1 == blocks {
+            block_flags |= flags_end;
+        }
+        let block_len_vec = set1(BLOCK_LEN as u32); // full blocks only
+        let block_flags_vec = set1(block_flags as u32);
+        let msg_vecs = transpose_msg_vecs(inputs, block * BLOCK_LEN);
+
+        // The transposed compression function. Note that inlining this
+        // manually here improves compile times by a lot, compared to factoring
+        // it out into its own function and making it #[inline(always)]. Just
+        // guessing, it might have something to do with loop unrolling.
+        let mut v = [
+            h_vecs[0],
+            h_vecs[1],
+            h_vecs[2],
+            h_vecs[3],
+            h_vecs[4],
+            h_vecs[5],
+            h_vecs[6],
+            h_vecs[7],
+            set1(IV[0]),
+            set1(IV[1]),
+            set1(IV[2]),
+            set1(IV[3]),
+            counter_low_vec,
+            counter_high_vec,
+            block_len_vec,
+            block_flags_vec,
+        ];
+        round(&mut v, &msg_vecs, 0);
+        round(&mut v, &msg_vecs, 1);
+        round(&mut v, &msg_vecs, 2);
+        round(&mut v, &msg_vecs, 3);
+        round(&mut v, &msg_vecs, 4);
+        round(&mut v, &msg_vecs, 5);
+        round(&mut v, &msg_vecs, 6);
+        h_vecs[0] = xor(v[0], v[8]);
+        h_vecs[1] = xor(v[1], v[9]);
+        h_vecs[2] = xor(v[2], v[10]);
+        h_vecs[3] = xor(v[3], v[11]);
+        h_vecs[4] = xor(v[4], v[12]);
+        h_vecs[5] = xor(v[5], v[13]);
+        h_vecs[6] = xor(v[6], v[14]);
+        h_vecs[7] = xor(v[7], v[15]);
+
+        block_flags = flags;
+    }
+
+    transpose_vecs(&mut h_vecs);
+    storeu(h_vecs[0], out.as_mut_ptr().add(0 * 4 * DEGREE));
+    storeu(h_vecs[1], out.as_mut_ptr().add(1 * 4 * DEGREE));
+    storeu(h_vecs[2], out.as_mut_ptr().add(2 * 4 * DEGREE));
+    storeu(h_vecs[3], out.as_mut_ptr().add(3 * 4 * DEGREE));
+    storeu(h_vecs[4], out.as_mut_ptr().add(4 * 4 * DEGREE));
+    storeu(h_vecs[5], out.as_mut_ptr().add(5 * 4 * DEGREE));
+    storeu(h_vecs[6], out.as_mut_ptr().add(6 * 4 * DEGREE));
+    storeu(h_vecs[7], out.as_mut_ptr().add(7 * 4 * DEGREE));
+}
+
+#[target_feature(enable = "avx2")]
+pub unsafe fn hash_many<const N: usize>(
+    mut inputs: &[&[u8; N]],
+    key: &CVWords,
+    mut counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    mut out: &mut [u8],
+) {
+    debug_assert!(out.len() >= inputs.len() * OUT_LEN, "out too short");
+    while inputs.len() >= DEGREE && out.len() >= DEGREE * OUT_LEN {
+        // Safe because the layout of arrays is guaranteed, and because the
+        // `blocks` count is determined statically from the argument type.
+        let input_ptrs: &[*const u8; DEGREE] = &*(inputs.as_ptr() as *const [*const u8; DEGREE]);
+        let blocks = N / BLOCK_LEN;
+        hash8(
+            input_ptrs,
+            blocks,
+            key,
+            counter,
+            increment_counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(out, 0, DEGREE * OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += DEGREE as u64;
+        }
+        inputs = &inputs[DEGREE..];
+        out = &mut out[DEGREE * OUT_LEN..];
+    }
+    crate::sse41::hash_many(
+        inputs,
+        key,
+        counter,
+        increment_counter,
+        flags,
+        flags_start,
+        flags_end,
+        out,
+    );
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_transpose() {
+        if !crate::platform::avx2_detected() {
+            return;
+        }
+
+        #[target_feature(enable = "avx2")]
+        unsafe fn transpose_wrapper(vecs: &mut [__m256i; DEGREE]) {
+            transpose_vecs(vecs);
+        }
+
+        let mut matrix = [[0 as u32; DEGREE]; DEGREE];
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                matrix[i][j] = (i * DEGREE + j) as u32;
+            }
+        }
+
+        unsafe {
+            let mut vecs: [__m256i; DEGREE] = core::mem::transmute(matrix);
+            transpose_wrapper(&mut vecs);
+            matrix = core::mem::transmute(vecs);
+        }
+
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                // Reversed indexes from above.
+                assert_eq!(matrix[j][i], (i * DEGREE + j) as u32);
+            }
+        }
+    }
+
+    #[test]
+    fn test_hash_many() {
+        if !crate::platform::avx2_detected() {
+            return;
+        }
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/rust_sse2.rs b/thirdparty/blake3/src/rust_sse2.rs
new file mode 100644
index 000000000..bd2be69f6
--- /dev/null
+++ b/thirdparty/blake3/src/rust_sse2.rs
@@ -0,0 +1,775 @@
+#[cfg(target_arch = "x86")]
+use core::arch::x86::*;
+#[cfg(target_arch = "x86_64")]
+use core::arch::x86_64::*;
+
+use crate::{
+    counter_high, counter_low, CVBytes, CVWords, IncrementCounter, BLOCK_LEN, IV, MSG_SCHEDULE,
+    OUT_LEN,
+};
+use arrayref::{array_mut_ref, array_ref, mut_array_refs};
+
+pub const DEGREE: usize = 4;
+
+#[inline(always)]
+unsafe fn loadu(src: *const u8) -> __m128i {
+    // This is an unaligned load, so the pointer cast is allowed.
+    _mm_loadu_si128(src as *const __m128i)
+}
+
+#[inline(always)]
+unsafe fn storeu(src: __m128i, dest: *mut u8) {
+    // This is an unaligned store, so the pointer cast is allowed.
+    _mm_storeu_si128(dest as *mut __m128i, src)
+}
+
+#[inline(always)]
+unsafe fn add(a: __m128i, b: __m128i) -> __m128i {
+    _mm_add_epi32(a, b)
+}
+
+#[inline(always)]
+unsafe fn xor(a: __m128i, b: __m128i) -> __m128i {
+    _mm_xor_si128(a, b)
+}
+
+#[inline(always)]
+unsafe fn set1(x: u32) -> __m128i {
+    _mm_set1_epi32(x as i32)
+}
+
+#[inline(always)]
+unsafe fn set4(a: u32, b: u32, c: u32, d: u32) -> __m128i {
+    _mm_setr_epi32(a as i32, b as i32, c as i32, d as i32)
+}
+
+// These rotations are the "simple/shifts version". For the
+// "complicated/shuffles version", see
+// https://github.com/sneves/blake2-avx2/blob/b3723921f668df09ece52dcd225a36d4a4eea1d9/blake2s-common.h#L63-L66.
+// For a discussion of the tradeoffs, see
+// https://github.com/sneves/blake2-avx2/pull/5. Due to an LLVM bug
+// (https://bugs.llvm.org/show_bug.cgi?id=44379), this version performs better
+// on recent x86 chips.
+
+#[inline(always)]
+unsafe fn rot16(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 16), _mm_slli_epi32(a, 32 - 16))
+}
+
+#[inline(always)]
+unsafe fn rot12(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 12), _mm_slli_epi32(a, 32 - 12))
+}
+
+#[inline(always)]
+unsafe fn rot8(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 8), _mm_slli_epi32(a, 32 - 8))
+}
+
+#[inline(always)]
+unsafe fn rot7(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 7), _mm_slli_epi32(a, 32 - 7))
+}
+
+#[inline(always)]
+unsafe fn g1(
+    row0: &mut __m128i,
+    row1: &mut __m128i,
+    row2: &mut __m128i,
+    row3: &mut __m128i,
+    m: __m128i,
+) {
+    *row0 = add(add(*row0, m), *row1);
+    *row3 = xor(*row3, *row0);
+    *row3 = rot16(*row3);
+    *row2 = add(*row2, *row3);
+    *row1 = xor(*row1, *row2);
+    *row1 = rot12(*row1);
+}
+
+#[inline(always)]
+unsafe fn g2(
+    row0: &mut __m128i,
+    row1: &mut __m128i,
+    row2: &mut __m128i,
+    row3: &mut __m128i,
+    m: __m128i,
+) {
+    *row0 = add(add(*row0, m), *row1);
+    *row3 = xor(*row3, *row0);
+    *row3 = rot8(*row3);
+    *row2 = add(*row2, *row3);
+    *row1 = xor(*row1, *row2);
+    *row1 = rot7(*row1);
+}
+
+// Adapted from https://github.com/rust-lang-nursery/stdsimd/pull/479.
+macro_rules! _MM_SHUFFLE {
+    ($z:expr, $y:expr, $x:expr, $w:expr) => {
+        ($z << 6) | ($y << 4) | ($x << 2) | $w
+    };
+}
+
+macro_rules! shuffle2 {
+    ($a:expr, $b:expr, $c:expr) => {
+        _mm_castps_si128(_mm_shuffle_ps(
+            _mm_castsi128_ps($a),
+            _mm_castsi128_ps($b),
+            $c,
+        ))
+    };
+}
+
+// Note the optimization here of leaving row1 as the unrotated row, rather than
+// row0. All the message loads below are adjusted to compensate for this. See
+// discussion at https://github.com/sneves/blake2-avx2/pull/4
+#[inline(always)]
+unsafe fn diagonalize(row0: &mut __m128i, row2: &mut __m128i, row3: &mut __m128i) {
+    *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE!(2, 1, 0, 3));
+    *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE!(1, 0, 3, 2));
+    *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE!(0, 3, 2, 1));
+}
+
+#[inline(always)]
+unsafe fn undiagonalize(row0: &mut __m128i, row2: &mut __m128i, row3: &mut __m128i) {
+    *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE!(0, 3, 2, 1));
+    *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE!(1, 0, 3, 2));
+    *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE!(2, 1, 0, 3));
+}
+
+#[inline(always)]
+unsafe fn blend_epi16(a: __m128i, b: __m128i, imm8: i32) -> __m128i {
+    let bits = _mm_set_epi16(0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01);
+    let mut mask = _mm_set1_epi16(imm8 as i16);
+    mask = _mm_and_si128(mask, bits);
+    mask = _mm_cmpeq_epi16(mask, bits);
+    _mm_or_si128(_mm_and_si128(mask, b), _mm_andnot_si128(mask, a))
+}
+
+#[inline(always)]
+unsafe fn compress_pre(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [__m128i; 4] {
+    let row0 = &mut loadu(cv.as_ptr().add(0) as *const u8);
+    let row1 = &mut loadu(cv.as_ptr().add(4) as *const u8);
+    let row2 = &mut set4(IV[0], IV[1], IV[2], IV[3]);
+    let row3 = &mut set4(
+        counter_low(counter),
+        counter_high(counter),
+        block_len as u32,
+        flags as u32,
+    );
+
+    let mut m0 = loadu(block.as_ptr().add(0 * 4 * DEGREE));
+    let mut m1 = loadu(block.as_ptr().add(1 * 4 * DEGREE));
+    let mut m2 = loadu(block.as_ptr().add(2 * 4 * DEGREE));
+    let mut m3 = loadu(block.as_ptr().add(3 * 4 * DEGREE));
+
+    let mut t0;
+    let mut t1;
+    let mut t2;
+    let mut t3;
+    let mut tt;
+
+    // Round 1. The first round permutes the message words from the original
+    // input order, into the groups that get mixed in parallel.
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(2, 0, 2, 0)); //  6  4  2  0
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 3, 1)); //  7  5  3  1
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = shuffle2!(m2, m3, _MM_SHUFFLE!(2, 0, 2, 0)); // 14 12 10  8
+    t2 = _mm_shuffle_epi32(t2, _MM_SHUFFLE!(2, 1, 0, 3)); // 12 10  8 14
+    g1(row0, row1, row2, row3, t2);
+    t3 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 1, 3, 1)); // 15 13 11  9
+    t3 = _mm_shuffle_epi32(t3, _MM_SHUFFLE!(2, 1, 0, 3)); // 13 11  9 15
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 2. This round and all following rounds apply a fixed permutation
+    // to the message words from the round before.
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 3
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 4
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 5
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 6
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 7
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+
+    [*row0, *row1, *row2, *row3]
+}
+
+#[target_feature(enable = "sse2")]
+pub unsafe fn compress_in_place(
+    cv: &mut CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    let [row0, row1, row2, row3] = compress_pre(cv, block, block_len, counter, flags);
+    storeu(xor(row0, row2), cv.as_mut_ptr().add(0) as *mut u8);
+    storeu(xor(row1, row3), cv.as_mut_ptr().add(4) as *mut u8);
+}
+
+#[target_feature(enable = "sse2")]
+pub unsafe fn compress_xof(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64] {
+    let [mut row0, mut row1, mut row2, mut row3] =
+        compress_pre(cv, block, block_len, counter, flags);
+    row0 = xor(row0, row2);
+    row1 = xor(row1, row3);
+    row2 = xor(row2, loadu(cv.as_ptr().add(0) as *const u8));
+    row3 = xor(row3, loadu(cv.as_ptr().add(4) as *const u8));
+    core::mem::transmute([row0, row1, row2, row3])
+}
+
+#[inline(always)]
+unsafe fn round(v: &mut [__m128i; 16], m: &[__m128i; 16], r: usize) {
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][0] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][2] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][4] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][6] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[15] = rot16(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot12(v[4]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][1] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][3] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][5] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][7] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[15] = rot8(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot7(v[4]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][8] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][10] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][12] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][14] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot16(v[15]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[4] = rot12(v[4]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][9] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][11] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][13] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][15] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot8(v[15]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+    v[4] = rot7(v[4]);
+}
+
+#[inline(always)]
+unsafe fn transpose_vecs(vecs: &mut [__m128i; DEGREE]) {
+    // Interleave 32-bit lanes. The low unpack is lanes 00/11 and the high is
+    // 22/33. Note that this doesn't split the vector into two lanes, as the
+    // AVX2 counterparts do.
+    let ab_01 = _mm_unpacklo_epi32(vecs[0], vecs[1]);
+    let ab_23 = _mm_unpackhi_epi32(vecs[0], vecs[1]);
+    let cd_01 = _mm_unpacklo_epi32(vecs[2], vecs[3]);
+    let cd_23 = _mm_unpackhi_epi32(vecs[2], vecs[3]);
+
+    // Interleave 64-bit lanes.
+    let abcd_0 = _mm_unpacklo_epi64(ab_01, cd_01);
+    let abcd_1 = _mm_unpackhi_epi64(ab_01, cd_01);
+    let abcd_2 = _mm_unpacklo_epi64(ab_23, cd_23);
+    let abcd_3 = _mm_unpackhi_epi64(ab_23, cd_23);
+
+    vecs[0] = abcd_0;
+    vecs[1] = abcd_1;
+    vecs[2] = abcd_2;
+    vecs[3] = abcd_3;
+}
+
+#[inline(always)]
+unsafe fn transpose_msg_vecs(inputs: &[*const u8; DEGREE], block_offset: usize) -> [__m128i; 16] {
+    let mut vecs = [
+        loadu(inputs[0].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 3 * 4 * DEGREE)),
+    ];
+    for i in 0..DEGREE {
+        _mm_prefetch(inputs[i].add(block_offset + 256) as *const i8, _MM_HINT_T0);
+    }
+    let squares = mut_array_refs!(&mut vecs, DEGREE, DEGREE, DEGREE, DEGREE);
+    transpose_vecs(squares.0);
+    transpose_vecs(squares.1);
+    transpose_vecs(squares.2);
+    transpose_vecs(squares.3);
+    vecs
+}
+
+#[inline(always)]
+unsafe fn load_counters(counter: u64, increment_counter: IncrementCounter) -> (__m128i, __m128i) {
+    let mask = if increment_counter.yes() { !0 } else { 0 };
+    (
+        set4(
+            counter_low(counter + (mask & 0)),
+            counter_low(counter + (mask & 1)),
+            counter_low(counter + (mask & 2)),
+            counter_low(counter + (mask & 3)),
+        ),
+        set4(
+            counter_high(counter + (mask & 0)),
+            counter_high(counter + (mask & 1)),
+            counter_high(counter + (mask & 2)),
+            counter_high(counter + (mask & 3)),
+        ),
+    )
+}
+
+#[target_feature(enable = "sse2")]
+pub unsafe fn hash4(
+    inputs: &[*const u8; DEGREE],
+    blocks: usize,
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8; DEGREE * OUT_LEN],
+) {
+    let mut h_vecs = [
+        set1(key[0]),
+        set1(key[1]),
+        set1(key[2]),
+        set1(key[3]),
+        set1(key[4]),
+        set1(key[5]),
+        set1(key[6]),
+        set1(key[7]),
+    ];
+    let (counter_low_vec, counter_high_vec) = load_counters(counter, increment_counter);
+    let mut block_flags = flags | flags_start;
+
+    for block in 0..blocks {
+        if block + 1 == blocks {
+            block_flags |= flags_end;
+        }
+        let block_len_vec = set1(BLOCK_LEN as u32); // full blocks only
+        let block_flags_vec = set1(block_flags as u32);
+        let msg_vecs = transpose_msg_vecs(inputs, block * BLOCK_LEN);
+
+        // The transposed compression function. Note that inlining this
+        // manually here improves compile times by a lot, compared to factoring
+        // it out into its own function and making it #[inline(always)]. Just
+        // guessing, it might have something to do with loop unrolling.
+        let mut v = [
+            h_vecs[0],
+            h_vecs[1],
+            h_vecs[2],
+            h_vecs[3],
+            h_vecs[4],
+            h_vecs[5],
+            h_vecs[6],
+            h_vecs[7],
+            set1(IV[0]),
+            set1(IV[1]),
+            set1(IV[2]),
+            set1(IV[3]),
+            counter_low_vec,
+            counter_high_vec,
+            block_len_vec,
+            block_flags_vec,
+        ];
+        round(&mut v, &msg_vecs, 0);
+        round(&mut v, &msg_vecs, 1);
+        round(&mut v, &msg_vecs, 2);
+        round(&mut v, &msg_vecs, 3);
+        round(&mut v, &msg_vecs, 4);
+        round(&mut v, &msg_vecs, 5);
+        round(&mut v, &msg_vecs, 6);
+        h_vecs[0] = xor(v[0], v[8]);
+        h_vecs[1] = xor(v[1], v[9]);
+        h_vecs[2] = xor(v[2], v[10]);
+        h_vecs[3] = xor(v[3], v[11]);
+        h_vecs[4] = xor(v[4], v[12]);
+        h_vecs[5] = xor(v[5], v[13]);
+        h_vecs[6] = xor(v[6], v[14]);
+        h_vecs[7] = xor(v[7], v[15]);
+
+        block_flags = flags;
+    }
+
+    let squares = mut_array_refs!(&mut h_vecs, DEGREE, DEGREE);
+    transpose_vecs(squares.0);
+    transpose_vecs(squares.1);
+    // The first four vecs now contain the first half of each output, and the
+    // second four vecs contain the second half of each output.
+    storeu(h_vecs[0], out.as_mut_ptr().add(0 * 4 * DEGREE));
+    storeu(h_vecs[4], out.as_mut_ptr().add(1 * 4 * DEGREE));
+    storeu(h_vecs[1], out.as_mut_ptr().add(2 * 4 * DEGREE));
+    storeu(h_vecs[5], out.as_mut_ptr().add(3 * 4 * DEGREE));
+    storeu(h_vecs[2], out.as_mut_ptr().add(4 * 4 * DEGREE));
+    storeu(h_vecs[6], out.as_mut_ptr().add(5 * 4 * DEGREE));
+    storeu(h_vecs[3], out.as_mut_ptr().add(6 * 4 * DEGREE));
+    storeu(h_vecs[7], out.as_mut_ptr().add(7 * 4 * DEGREE));
+}
+
+#[target_feature(enable = "sse2")]
+unsafe fn hash1<const N: usize>(
+    input: &[u8; N],
+    key: &CVWords,
+    counter: u64,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut CVBytes,
+) {
+    debug_assert_eq!(N % BLOCK_LEN, 0, "uneven blocks");
+    let mut cv = *key;
+    let mut block_flags = flags | flags_start;
+    let mut slice = &input[..];
+    while slice.len() >= BLOCK_LEN {
+        if slice.len() == BLOCK_LEN {
+            block_flags |= flags_end;
+        }
+        compress_in_place(
+            &mut cv,
+            array_ref!(slice, 0, BLOCK_LEN),
+            BLOCK_LEN as u8,
+            counter,
+            block_flags,
+        );
+        block_flags = flags;
+        slice = &slice[BLOCK_LEN..];
+    }
+    *out = core::mem::transmute(cv); // x86 is little-endian
+}
+
+#[target_feature(enable = "sse2")]
+pub unsafe fn hash_many<const N: usize>(
+    mut inputs: &[&[u8; N]],
+    key: &CVWords,
+    mut counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    mut out: &mut [u8],
+) {
+    debug_assert!(out.len() >= inputs.len() * OUT_LEN, "out too short");
+    while inputs.len() >= DEGREE && out.len() >= DEGREE * OUT_LEN {
+        // Safe because the layout of arrays is guaranteed, and because the
+        // `blocks` count is determined statically from the argument type.
+        let input_ptrs: &[*const u8; DEGREE] = &*(inputs.as_ptr() as *const [*const u8; DEGREE]);
+        let blocks = N / BLOCK_LEN;
+        hash4(
+            input_ptrs,
+            blocks,
+            key,
+            counter,
+            increment_counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(out, 0, DEGREE * OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += DEGREE as u64;
+        }
+        inputs = &inputs[DEGREE..];
+        out = &mut out[DEGREE * OUT_LEN..];
+    }
+    for (&input, output) in inputs.iter().zip(out.chunks_exact_mut(OUT_LEN)) {
+        hash1(
+            input,
+            key,
+            counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(output, 0, OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += 1;
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_transpose() {
+        if !crate::platform::sse2_detected() {
+            return;
+        }
+
+        #[target_feature(enable = "sse2")]
+        unsafe fn transpose_wrapper(vecs: &mut [__m128i; DEGREE]) {
+            transpose_vecs(vecs);
+        }
+
+        let mut matrix = [[0 as u32; DEGREE]; DEGREE];
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                matrix[i][j] = (i * DEGREE + j) as u32;
+            }
+        }
+
+        unsafe {
+            let mut vecs: [__m128i; DEGREE] = core::mem::transmute(matrix);
+            transpose_wrapper(&mut vecs);
+            matrix = core::mem::transmute(vecs);
+        }
+
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                // Reversed indexes from above.
+                assert_eq!(matrix[j][i], (i * DEGREE + j) as u32);
+            }
+        }
+    }
+
+    #[test]
+    fn test_compress() {
+        if !crate::platform::sse2_detected() {
+            return;
+        }
+        crate::test::test_compress_fn(compress_in_place, compress_xof);
+    }
+
+    #[test]
+    fn test_hash_many() {
+        if !crate::platform::sse2_detected() {
+            return;
+        }
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/rust_sse41.rs b/thirdparty/blake3/src/rust_sse41.rs
new file mode 100644
index 000000000..1ebadc482
--- /dev/null
+++ b/thirdparty/blake3/src/rust_sse41.rs
@@ -0,0 +1,766 @@
+#[cfg(target_arch = "x86")]
+use core::arch::x86::*;
+#[cfg(target_arch = "x86_64")]
+use core::arch::x86_64::*;
+
+use crate::{
+    counter_high, counter_low, CVBytes, CVWords, IncrementCounter, BLOCK_LEN, IV, MSG_SCHEDULE,
+    OUT_LEN,
+};
+use arrayref::{array_mut_ref, array_ref, mut_array_refs};
+
+pub const DEGREE: usize = 4;
+
+#[inline(always)]
+unsafe fn loadu(src: *const u8) -> __m128i {
+    // This is an unaligned load, so the pointer cast is allowed.
+    _mm_loadu_si128(src as *const __m128i)
+}
+
+#[inline(always)]
+unsafe fn storeu(src: __m128i, dest: *mut u8) {
+    // This is an unaligned store, so the pointer cast is allowed.
+    _mm_storeu_si128(dest as *mut __m128i, src)
+}
+
+#[inline(always)]
+unsafe fn add(a: __m128i, b: __m128i) -> __m128i {
+    _mm_add_epi32(a, b)
+}
+
+#[inline(always)]
+unsafe fn xor(a: __m128i, b: __m128i) -> __m128i {
+    _mm_xor_si128(a, b)
+}
+
+#[inline(always)]
+unsafe fn set1(x: u32) -> __m128i {
+    _mm_set1_epi32(x as i32)
+}
+
+#[inline(always)]
+unsafe fn set4(a: u32, b: u32, c: u32, d: u32) -> __m128i {
+    _mm_setr_epi32(a as i32, b as i32, c as i32, d as i32)
+}
+
+// These rotations are the "simple/shifts version". For the
+// "complicated/shuffles version", see
+// https://github.com/sneves/blake2-avx2/blob/b3723921f668df09ece52dcd225a36d4a4eea1d9/blake2s-common.h#L63-L66.
+// For a discussion of the tradeoffs, see
+// https://github.com/sneves/blake2-avx2/pull/5. Due to an LLVM bug
+// (https://bugs.llvm.org/show_bug.cgi?id=44379), this version performs better
+// on recent x86 chips.
+
+#[inline(always)]
+unsafe fn rot16(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 16), _mm_slli_epi32(a, 32 - 16))
+}
+
+#[inline(always)]
+unsafe fn rot12(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 12), _mm_slli_epi32(a, 32 - 12))
+}
+
+#[inline(always)]
+unsafe fn rot8(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 8), _mm_slli_epi32(a, 32 - 8))
+}
+
+#[inline(always)]
+unsafe fn rot7(a: __m128i) -> __m128i {
+    _mm_or_si128(_mm_srli_epi32(a, 7), _mm_slli_epi32(a, 32 - 7))
+}
+
+#[inline(always)]
+unsafe fn g1(
+    row0: &mut __m128i,
+    row1: &mut __m128i,
+    row2: &mut __m128i,
+    row3: &mut __m128i,
+    m: __m128i,
+) {
+    *row0 = add(add(*row0, m), *row1);
+    *row3 = xor(*row3, *row0);
+    *row3 = rot16(*row3);
+    *row2 = add(*row2, *row3);
+    *row1 = xor(*row1, *row2);
+    *row1 = rot12(*row1);
+}
+
+#[inline(always)]
+unsafe fn g2(
+    row0: &mut __m128i,
+    row1: &mut __m128i,
+    row2: &mut __m128i,
+    row3: &mut __m128i,
+    m: __m128i,
+) {
+    *row0 = add(add(*row0, m), *row1);
+    *row3 = xor(*row3, *row0);
+    *row3 = rot8(*row3);
+    *row2 = add(*row2, *row3);
+    *row1 = xor(*row1, *row2);
+    *row1 = rot7(*row1);
+}
+
+// Adapted from https://github.com/rust-lang-nursery/stdsimd/pull/479.
+macro_rules! _MM_SHUFFLE {
+    ($z:expr, $y:expr, $x:expr, $w:expr) => {
+        ($z << 6) | ($y << 4) | ($x << 2) | $w
+    };
+}
+
+macro_rules! shuffle2 {
+    ($a:expr, $b:expr, $c:expr) => {
+        _mm_castps_si128(_mm_shuffle_ps(
+            _mm_castsi128_ps($a),
+            _mm_castsi128_ps($b),
+            $c,
+        ))
+    };
+}
+
+// Note the optimization here of leaving row1 as the unrotated row, rather than
+// row0. All the message loads below are adjusted to compensate for this. See
+// discussion at https://github.com/sneves/blake2-avx2/pull/4
+#[inline(always)]
+unsafe fn diagonalize(row0: &mut __m128i, row2: &mut __m128i, row3: &mut __m128i) {
+    *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE!(2, 1, 0, 3));
+    *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE!(1, 0, 3, 2));
+    *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE!(0, 3, 2, 1));
+}
+
+#[inline(always)]
+unsafe fn undiagonalize(row0: &mut __m128i, row2: &mut __m128i, row3: &mut __m128i) {
+    *row0 = _mm_shuffle_epi32(*row0, _MM_SHUFFLE!(0, 3, 2, 1));
+    *row3 = _mm_shuffle_epi32(*row3, _MM_SHUFFLE!(1, 0, 3, 2));
+    *row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE!(2, 1, 0, 3));
+}
+
+#[inline(always)]
+unsafe fn compress_pre(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [__m128i; 4] {
+    let row0 = &mut loadu(cv.as_ptr().add(0) as *const u8);
+    let row1 = &mut loadu(cv.as_ptr().add(4) as *const u8);
+    let row2 = &mut set4(IV[0], IV[1], IV[2], IV[3]);
+    let row3 = &mut set4(
+        counter_low(counter),
+        counter_high(counter),
+        block_len as u32,
+        flags as u32,
+    );
+
+    let mut m0 = loadu(block.as_ptr().add(0 * 4 * DEGREE));
+    let mut m1 = loadu(block.as_ptr().add(1 * 4 * DEGREE));
+    let mut m2 = loadu(block.as_ptr().add(2 * 4 * DEGREE));
+    let mut m3 = loadu(block.as_ptr().add(3 * 4 * DEGREE));
+
+    let mut t0;
+    let mut t1;
+    let mut t2;
+    let mut t3;
+    let mut tt;
+
+    // Round 1. The first round permutes the message words from the original
+    // input order, into the groups that get mixed in parallel.
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(2, 0, 2, 0)); //  6  4  2  0
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 3, 1)); //  7  5  3  1
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = shuffle2!(m2, m3, _MM_SHUFFLE!(2, 0, 2, 0)); // 14 12 10  8
+    t2 = _mm_shuffle_epi32(t2, _MM_SHUFFLE!(2, 1, 0, 3)); // 12 10  8 14
+    g1(row0, row1, row2, row3, t2);
+    t3 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 1, 3, 1)); // 15 13 11  9
+    t3 = _mm_shuffle_epi32(t3, _MM_SHUFFLE!(2, 1, 0, 3)); // 13 11  9 15
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 2. This round and all following rounds apply a fixed permutation
+    // to the message words from the round before.
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = _mm_blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = _mm_blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 3
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = _mm_blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = _mm_blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 4
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = _mm_blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = _mm_blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 5
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = _mm_blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = _mm_blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 6
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = _mm_blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = _mm_blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 7
+    t0 = shuffle2!(m0, m1, _MM_SHUFFLE!(3, 1, 1, 2));
+    t0 = _mm_shuffle_epi32(t0, _MM_SHUFFLE!(0, 3, 2, 1));
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
+    tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
+    t1 = _mm_blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = _mm_unpacklo_epi64(m3, m1);
+    tt = _mm_blend_epi16(t2, m2, 0xC0);
+    t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
+    g1(row0, row1, row2, row3, t2);
+    t3 = _mm_unpackhi_epi32(m1, m3);
+    tt = _mm_unpacklo_epi32(m2, t3);
+    t3 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(0, 1, 3, 2));
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+
+    [*row0, *row1, *row2, *row3]
+}
+
+#[target_feature(enable = "sse4.1")]
+pub unsafe fn compress_in_place(
+    cv: &mut CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    let [row0, row1, row2, row3] = compress_pre(cv, block, block_len, counter, flags);
+    storeu(xor(row0, row2), cv.as_mut_ptr().add(0) as *mut u8);
+    storeu(xor(row1, row3), cv.as_mut_ptr().add(4) as *mut u8);
+}
+
+#[target_feature(enable = "sse4.1")]
+pub unsafe fn compress_xof(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64] {
+    let [mut row0, mut row1, mut row2, mut row3] =
+        compress_pre(cv, block, block_len, counter, flags);
+    row0 = xor(row0, row2);
+    row1 = xor(row1, row3);
+    row2 = xor(row2, loadu(cv.as_ptr().add(0) as *const u8));
+    row3 = xor(row3, loadu(cv.as_ptr().add(4) as *const u8));
+    core::mem::transmute([row0, row1, row2, row3])
+}
+
+#[inline(always)]
+unsafe fn round(v: &mut [__m128i; 16], m: &[__m128i; 16], r: usize) {
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][0] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][2] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][4] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][6] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[15] = rot16(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot12(v[4]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][1] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][3] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][5] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][7] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[15] = rot8(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot7(v[4]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][8] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][10] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][12] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][14] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot16(v[15]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[4] = rot12(v[4]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][9] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][11] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][13] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][15] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot8(v[15]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+    v[4] = rot7(v[4]);
+}
+
+#[inline(always)]
+unsafe fn transpose_vecs(vecs: &mut [__m128i; DEGREE]) {
+    // Interleave 32-bit lanes. The low unpack is lanes 00/11 and the high is
+    // 22/33. Note that this doesn't split the vector into two lanes, as the
+    // AVX2 counterparts do.
+    let ab_01 = _mm_unpacklo_epi32(vecs[0], vecs[1]);
+    let ab_23 = _mm_unpackhi_epi32(vecs[0], vecs[1]);
+    let cd_01 = _mm_unpacklo_epi32(vecs[2], vecs[3]);
+    let cd_23 = _mm_unpackhi_epi32(vecs[2], vecs[3]);
+
+    // Interleave 64-bit lanes.
+    let abcd_0 = _mm_unpacklo_epi64(ab_01, cd_01);
+    let abcd_1 = _mm_unpackhi_epi64(ab_01, cd_01);
+    let abcd_2 = _mm_unpacklo_epi64(ab_23, cd_23);
+    let abcd_3 = _mm_unpackhi_epi64(ab_23, cd_23);
+
+    vecs[0] = abcd_0;
+    vecs[1] = abcd_1;
+    vecs[2] = abcd_2;
+    vecs[3] = abcd_3;
+}
+
+#[inline(always)]
+unsafe fn transpose_msg_vecs(inputs: &[*const u8; DEGREE], block_offset: usize) -> [__m128i; 16] {
+    let mut vecs = [
+        loadu(inputs[0].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 3 * 4 * DEGREE)),
+    ];
+    for i in 0..DEGREE {
+        _mm_prefetch(inputs[i].add(block_offset + 256) as *const i8, _MM_HINT_T0);
+    }
+    let squares = mut_array_refs!(&mut vecs, DEGREE, DEGREE, DEGREE, DEGREE);
+    transpose_vecs(squares.0);
+    transpose_vecs(squares.1);
+    transpose_vecs(squares.2);
+    transpose_vecs(squares.3);
+    vecs
+}
+
+#[inline(always)]
+unsafe fn load_counters(counter: u64, increment_counter: IncrementCounter) -> (__m128i, __m128i) {
+    let mask = if increment_counter.yes() { !0 } else { 0 };
+    (
+        set4(
+            counter_low(counter + (mask & 0)),
+            counter_low(counter + (mask & 1)),
+            counter_low(counter + (mask & 2)),
+            counter_low(counter + (mask & 3)),
+        ),
+        set4(
+            counter_high(counter + (mask & 0)),
+            counter_high(counter + (mask & 1)),
+            counter_high(counter + (mask & 2)),
+            counter_high(counter + (mask & 3)),
+        ),
+    )
+}
+
+#[target_feature(enable = "sse4.1")]
+pub unsafe fn hash4(
+    inputs: &[*const u8; DEGREE],
+    blocks: usize,
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8; DEGREE * OUT_LEN],
+) {
+    let mut h_vecs = [
+        set1(key[0]),
+        set1(key[1]),
+        set1(key[2]),
+        set1(key[3]),
+        set1(key[4]),
+        set1(key[5]),
+        set1(key[6]),
+        set1(key[7]),
+    ];
+    let (counter_low_vec, counter_high_vec) = load_counters(counter, increment_counter);
+    let mut block_flags = flags | flags_start;
+
+    for block in 0..blocks {
+        if block + 1 == blocks {
+            block_flags |= flags_end;
+        }
+        let block_len_vec = set1(BLOCK_LEN as u32); // full blocks only
+        let block_flags_vec = set1(block_flags as u32);
+        let msg_vecs = transpose_msg_vecs(inputs, block * BLOCK_LEN);
+
+        // The transposed compression function. Note that inlining this
+        // manually here improves compile times by a lot, compared to factoring
+        // it out into its own function and making it #[inline(always)]. Just
+        // guessing, it might have something to do with loop unrolling.
+        let mut v = [
+            h_vecs[0],
+            h_vecs[1],
+            h_vecs[2],
+            h_vecs[3],
+            h_vecs[4],
+            h_vecs[5],
+            h_vecs[6],
+            h_vecs[7],
+            set1(IV[0]),
+            set1(IV[1]),
+            set1(IV[2]),
+            set1(IV[3]),
+            counter_low_vec,
+            counter_high_vec,
+            block_len_vec,
+            block_flags_vec,
+        ];
+        round(&mut v, &msg_vecs, 0);
+        round(&mut v, &msg_vecs, 1);
+        round(&mut v, &msg_vecs, 2);
+        round(&mut v, &msg_vecs, 3);
+        round(&mut v, &msg_vecs, 4);
+        round(&mut v, &msg_vecs, 5);
+        round(&mut v, &msg_vecs, 6);
+        h_vecs[0] = xor(v[0], v[8]);
+        h_vecs[1] = xor(v[1], v[9]);
+        h_vecs[2] = xor(v[2], v[10]);
+        h_vecs[3] = xor(v[3], v[11]);
+        h_vecs[4] = xor(v[4], v[12]);
+        h_vecs[5] = xor(v[5], v[13]);
+        h_vecs[6] = xor(v[6], v[14]);
+        h_vecs[7] = xor(v[7], v[15]);
+
+        block_flags = flags;
+    }
+
+    let squares = mut_array_refs!(&mut h_vecs, DEGREE, DEGREE);
+    transpose_vecs(squares.0);
+    transpose_vecs(squares.1);
+    // The first four vecs now contain the first half of each output, and the
+    // second four vecs contain the second half of each output.
+    storeu(h_vecs[0], out.as_mut_ptr().add(0 * 4 * DEGREE));
+    storeu(h_vecs[4], out.as_mut_ptr().add(1 * 4 * DEGREE));
+    storeu(h_vecs[1], out.as_mut_ptr().add(2 * 4 * DEGREE));
+    storeu(h_vecs[5], out.as_mut_ptr().add(3 * 4 * DEGREE));
+    storeu(h_vecs[2], out.as_mut_ptr().add(4 * 4 * DEGREE));
+    storeu(h_vecs[6], out.as_mut_ptr().add(5 * 4 * DEGREE));
+    storeu(h_vecs[3], out.as_mut_ptr().add(6 * 4 * DEGREE));
+    storeu(h_vecs[7], out.as_mut_ptr().add(7 * 4 * DEGREE));
+}
+
+#[target_feature(enable = "sse4.1")]
+unsafe fn hash1<const N: usize>(
+    input: &[u8; N],
+    key: &CVWords,
+    counter: u64,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut CVBytes,
+) {
+    debug_assert_eq!(N % BLOCK_LEN, 0, "uneven blocks");
+    let mut cv = *key;
+    let mut block_flags = flags | flags_start;
+    let mut slice = &input[..];
+    while slice.len() >= BLOCK_LEN {
+        if slice.len() == BLOCK_LEN {
+            block_flags |= flags_end;
+        }
+        compress_in_place(
+            &mut cv,
+            array_ref!(slice, 0, BLOCK_LEN),
+            BLOCK_LEN as u8,
+            counter,
+            block_flags,
+        );
+        block_flags = flags;
+        slice = &slice[BLOCK_LEN..];
+    }
+    *out = core::mem::transmute(cv); // x86 is little-endian
+}
+
+#[target_feature(enable = "sse4.1")]
+pub unsafe fn hash_many<const N: usize>(
+    mut inputs: &[&[u8; N]],
+    key: &CVWords,
+    mut counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    mut out: &mut [u8],
+) {
+    debug_assert!(out.len() >= inputs.len() * OUT_LEN, "out too short");
+    while inputs.len() >= DEGREE && out.len() >= DEGREE * OUT_LEN {
+        // Safe because the layout of arrays is guaranteed, and because the
+        // `blocks` count is determined statically from the argument type.
+        let input_ptrs: &[*const u8; DEGREE] = &*(inputs.as_ptr() as *const [*const u8; DEGREE]);
+        let blocks = N / BLOCK_LEN;
+        hash4(
+            input_ptrs,
+            blocks,
+            key,
+            counter,
+            increment_counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(out, 0, DEGREE * OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += DEGREE as u64;
+        }
+        inputs = &inputs[DEGREE..];
+        out = &mut out[DEGREE * OUT_LEN..];
+    }
+    for (&input, output) in inputs.iter().zip(out.chunks_exact_mut(OUT_LEN)) {
+        hash1(
+            input,
+            key,
+            counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(output, 0, OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += 1;
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_transpose() {
+        if !crate::platform::sse41_detected() {
+            return;
+        }
+
+        #[target_feature(enable = "sse4.1")]
+        unsafe fn transpose_wrapper(vecs: &mut [__m128i; DEGREE]) {
+            transpose_vecs(vecs);
+        }
+
+        let mut matrix = [[0 as u32; DEGREE]; DEGREE];
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                matrix[i][j] = (i * DEGREE + j) as u32;
+            }
+        }
+
+        unsafe {
+            let mut vecs: [__m128i; DEGREE] = core::mem::transmute(matrix);
+            transpose_wrapper(&mut vecs);
+            matrix = core::mem::transmute(vecs);
+        }
+
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                // Reversed indexes from above.
+                assert_eq!(matrix[j][i], (i * DEGREE + j) as u32);
+            }
+        }
+    }
+
+    #[test]
+    fn test_compress() {
+        if !crate::platform::sse41_detected() {
+            return;
+        }
+        crate::test::test_compress_fn(compress_in_place, compress_xof);
+    }
+
+    #[test]
+    fn test_hash_many() {
+        if !crate::platform::sse41_detected() {
+            return;
+        }
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/src/test.rs b/thirdparty/blake3/src/test.rs
new file mode 100644
index 000000000..2d21856a0
--- /dev/null
+++ b/thirdparty/blake3/src/test.rs
@@ -0,0 +1,1049 @@
+use crate::{CVBytes, CVWords, IncrementCounter, BLOCK_LEN, CHUNK_LEN, OUT_LEN};
+use arrayref::array_ref;
+use arrayvec::ArrayVec;
+use core::usize;
+use rand::prelude::*;
+
+// Interesting input lengths to run tests on.
+pub const TEST_CASES: &[usize] = &[
+    0,
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    BLOCK_LEN - 1,
+    BLOCK_LEN,
+    BLOCK_LEN + 1,
+    2 * BLOCK_LEN - 1,
+    2 * BLOCK_LEN,
+    2 * BLOCK_LEN + 1,
+    CHUNK_LEN - 1,
+    CHUNK_LEN,
+    CHUNK_LEN + 1,
+    2 * CHUNK_LEN,
+    2 * CHUNK_LEN + 1,
+    3 * CHUNK_LEN,
+    3 * CHUNK_LEN + 1,
+    4 * CHUNK_LEN,
+    4 * CHUNK_LEN + 1,
+    5 * CHUNK_LEN,
+    5 * CHUNK_LEN + 1,
+    6 * CHUNK_LEN,
+    6 * CHUNK_LEN + 1,
+    7 * CHUNK_LEN,
+    7 * CHUNK_LEN + 1,
+    8 * CHUNK_LEN,
+    8 * CHUNK_LEN + 1,
+    16 * CHUNK_LEN - 1,
+    16 * CHUNK_LEN, // AVX512's bandwidth
+    16 * CHUNK_LEN + 1,
+    31 * CHUNK_LEN - 1,
+    31 * CHUNK_LEN, // 16 + 8 + 4 + 2 + 1
+    31 * CHUNK_LEN + 1,
+    100 * CHUNK_LEN, // subtrees larger than MAX_SIMD_DEGREE chunks
+];
+
+pub const TEST_CASES_MAX: usize = 100 * CHUNK_LEN;
+
+// There's a test to make sure these two are equal below.
+pub const TEST_KEY: CVBytes = *b"whats the Elvish word for friend";
+pub const TEST_KEY_WORDS: CVWords = [
+    1952540791, 1752440947, 1816469605, 1752394102, 1919907616, 1868963940, 1919295602, 1684956521,
+];
+
+// Paint the input with a repeating byte pattern. We use a cycle length of 251,
+// because that's the largest prime number less than 256. This makes it
+// unlikely to swapping any two adjacent input blocks or chunks will give the
+// same answer.
+pub fn paint_test_input(buf: &mut [u8]) {
+    for (i, b) in buf.iter_mut().enumerate() {
+        *b = (i % 251) as u8;
+    }
+}
+
+type CompressInPlaceFn =
+    unsafe fn(cv: &mut CVWords, block: &[u8; BLOCK_LEN], block_len: u8, counter: u64, flags: u8);
+
+type CompressXofFn = unsafe fn(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64];
+
+// A shared helper function for platform-specific tests.
+pub fn test_compress_fn(compress_in_place_fn: CompressInPlaceFn, compress_xof_fn: CompressXofFn) {
+    let initial_state = TEST_KEY_WORDS;
+    let block_len: u8 = 61;
+    let mut block = [0; BLOCK_LEN];
+    paint_test_input(&mut block[..block_len as usize]);
+    // Use a counter with set bits in both 32-bit words.
+    let counter = (5u64 << 32) + 6;
+    let flags = crate::CHUNK_END | crate::ROOT | crate::KEYED_HASH;
+
+    let portable_out =
+        crate::portable::compress_xof(&initial_state, &block, block_len, counter as u64, flags);
+
+    let mut test_state = initial_state;
+    unsafe { compress_in_place_fn(&mut test_state, &block, block_len, counter as u64, flags) };
+    let test_state_bytes = crate::platform::le_bytes_from_words_32(&test_state);
+    let test_xof =
+        unsafe { compress_xof_fn(&initial_state, &block, block_len, counter as u64, flags) };
+
+    assert_eq!(&portable_out[..32], &test_state_bytes[..]);
+    assert_eq!(&portable_out[..], &test_xof[..]);
+}
+
+type HashManyFn<A> = unsafe fn(
+    inputs: &[&A],
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8],
+);
+
+// A shared helper function for platform-specific tests.
+pub fn test_hash_many_fn(
+    hash_many_chunks_fn: HashManyFn<[u8; CHUNK_LEN]>,
+    hash_many_parents_fn: HashManyFn<[u8; 2 * OUT_LEN]>,
+) {
+    // Test a few different initial counter values.
+    // - 0: The base case.
+    // - u32::MAX: The low word of the counter overflows for all inputs except the first.
+    // - i32::MAX: *No* overflow. But carry bugs in tricky SIMD code can screw this up, if you XOR
+    //   when you're supposed to ANDNOT...
+    let initial_counters = [0, u32::MAX as u64, i32::MAX as u64];
+    for counter in initial_counters {
+        #[cfg(feature = "std")]
+        dbg!(counter);
+
+        // 31 (16 + 8 + 4 + 2 + 1) inputs
+        const NUM_INPUTS: usize = 31;
+        let mut input_buf = [0; CHUNK_LEN * NUM_INPUTS];
+        crate::test::paint_test_input(&mut input_buf);
+
+        // First hash chunks.
+        let mut chunks = ArrayVec::<&[u8; CHUNK_LEN], NUM_INPUTS>::new();
+        for i in 0..NUM_INPUTS {
+            chunks.push(array_ref!(input_buf, i * CHUNK_LEN, CHUNK_LEN));
+        }
+        let mut portable_chunks_out = [0; NUM_INPUTS * OUT_LEN];
+        crate::portable::hash_many(
+            &chunks,
+            &TEST_KEY_WORDS,
+            counter,
+            IncrementCounter::Yes,
+            crate::KEYED_HASH,
+            crate::CHUNK_START,
+            crate::CHUNK_END,
+            &mut portable_chunks_out,
+        );
+
+        let mut test_chunks_out = [0; NUM_INPUTS * OUT_LEN];
+        unsafe {
+            hash_many_chunks_fn(
+                &chunks[..],
+                &TEST_KEY_WORDS,
+                counter,
+                IncrementCounter::Yes,
+                crate::KEYED_HASH,
+                crate::CHUNK_START,
+                crate::CHUNK_END,
+                &mut test_chunks_out,
+            );
+        }
+        for n in 0..NUM_INPUTS {
+            #[cfg(feature = "std")]
+            dbg!(n);
+            assert_eq!(
+                &portable_chunks_out[n * OUT_LEN..][..OUT_LEN],
+                &test_chunks_out[n * OUT_LEN..][..OUT_LEN]
+            );
+        }
+
+        // Then hash parents.
+        let mut parents = ArrayVec::<&[u8; 2 * OUT_LEN], NUM_INPUTS>::new();
+        for i in 0..NUM_INPUTS {
+            parents.push(array_ref!(input_buf, i * 2 * OUT_LEN, 2 * OUT_LEN));
+        }
+        let mut portable_parents_out = [0; NUM_INPUTS * OUT_LEN];
+        crate::portable::hash_many(
+            &parents,
+            &TEST_KEY_WORDS,
+            counter,
+            IncrementCounter::No,
+            crate::KEYED_HASH | crate::PARENT,
+            0,
+            0,
+            &mut portable_parents_out,
+        );
+
+        let mut test_parents_out = [0; NUM_INPUTS * OUT_LEN];
+        unsafe {
+            hash_many_parents_fn(
+                &parents[..],
+                &TEST_KEY_WORDS,
+                counter,
+                IncrementCounter::No,
+                crate::KEYED_HASH | crate::PARENT,
+                0,
+                0,
+                &mut test_parents_out,
+            );
+        }
+        for n in 0..NUM_INPUTS {
+            #[cfg(feature = "std")]
+            dbg!(n);
+            assert_eq!(
+                &portable_parents_out[n * OUT_LEN..][..OUT_LEN],
+                &test_parents_out[n * OUT_LEN..][..OUT_LEN]
+            );
+        }
+    }
+}
+
+#[allow(unused)]
+type XofManyFunction = unsafe fn(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+    out: &mut [u8],
+);
+
+// A shared helper function for platform-specific tests.
+#[allow(unused)]
+pub fn test_xof_many_fn(xof_many_function: XofManyFunction) {
+    let mut block = [0; BLOCK_LEN];
+    let block_len = 42;
+    crate::test::paint_test_input(&mut block[..block_len]);
+    let cv = [40, 41, 42, 43, 44, 45, 46, 47];
+    let flags = crate::KEYED_HASH;
+
+    // Test a few different initial counter values.
+    // - 0: The base case.
+    // - u32::MAX: The low word of the counter overflows for all inputs except the first.
+    // - i32::MAX: *No* overflow. But carry bugs in tricky SIMD code can screw this up, if you XOR
+    //   when you're supposed to ANDNOT...
+    let initial_counters = [0, u32::MAX as u64, i32::MAX as u64];
+    for counter in initial_counters {
+        #[cfg(feature = "std")]
+        dbg!(counter);
+
+        // 31 (16 + 8 + 4 + 2 + 1) outputs
+        const OUTPUT_SIZE: usize = 31 * BLOCK_LEN;
+
+        let mut portable_out = [0u8; OUTPUT_SIZE];
+        for (i, out_block) in portable_out.chunks_exact_mut(64).enumerate() {
+            out_block.copy_from_slice(&crate::portable::compress_xof(
+                &cv,
+                &block,
+                block_len as u8,
+                counter + i as u64,
+                flags,
+            ));
+        }
+
+        let mut test_out = [0u8; OUTPUT_SIZE];
+        unsafe {
+            xof_many_function(&cv, &block, block_len as u8, counter, flags, &mut test_out);
+        }
+
+        assert_eq!(portable_out, test_out);
+    }
+
+    // Test that xof_many doesn't write more blocks than requested. Note that the current assembly
+    // implementation always outputs at least one block, so we don't test the zero case.
+    for block_count in 1..=32 {
+        let mut array = [0; BLOCK_LEN * 33];
+        let output_start = 17;
+        let output_len = block_count * BLOCK_LEN;
+        let output_end = output_start + output_len;
+        let output = &mut array[output_start..output_end];
+        unsafe {
+            xof_many_function(&cv, &block, block_len as u8, 0, flags, output);
+        }
+        for i in 0..array.len() {
+            if i < output_start || output_end <= i {
+                assert_eq!(0, array[i], "index {i}");
+            }
+        }
+    }
+}
+
+#[test]
+fn test_key_bytes_equal_key_words() {
+    assert_eq!(
+        TEST_KEY_WORDS,
+        crate::platform::words_from_le_bytes_32(&TEST_KEY),
+    );
+}
+
+#[test]
+fn test_reference_impl_size() {
+    // Because the Rust compiler optimizes struct layout, it's possible that
+    // some future version of the compiler will produce a different size. If
+    // that happens, we can either disable this test, or test for multiple
+    // expected values. For now, the purpose of this test is to make sure we
+    // notice if that happens.
+    assert_eq!(1880, core::mem::size_of::<reference_impl::Hasher>());
+}
+
+#[test]
+fn test_counter_words() {
+    let counter: u64 = (1 << 32) + 2;
+    assert_eq!(crate::counter_low(counter), 2);
+    assert_eq!(crate::counter_high(counter), 1);
+}
+
+#[test]
+fn test_largest_power_of_two_leq() {
+    let input_output = &[
+        // The zero case is nonsensical, but it does work.
+        (0, 1),
+        (1, 1),
+        (2, 2),
+        (3, 2),
+        (4, 4),
+        (5, 4),
+        (6, 4),
+        (7, 4),
+        (8, 8),
+        // the largest possible usize
+        (usize::MAX, (usize::MAX >> 1) + 1),
+    ];
+    for &(input, output) in input_output {
+        assert_eq!(
+            output,
+            crate::largest_power_of_two_leq(input),
+            "wrong output for n={}",
+            input
+        );
+    }
+}
+
+#[test]
+fn test_compare_reference_impl() {
+    const OUT: usize = 303; // more than 64, not a multiple of 4
+    let mut input_buf = [0; TEST_CASES_MAX];
+    paint_test_input(&mut input_buf);
+    for &case in TEST_CASES {
+        let input = &input_buf[..case];
+        #[cfg(feature = "std")]
+        dbg!(case);
+
+        // regular
+        {
+            let mut reference_hasher = reference_impl::Hasher::new();
+            reference_hasher.update(input);
+            let mut expected_out = [0; OUT];
+            reference_hasher.finalize(&mut expected_out);
+
+            // all at once
+            let test_out = crate::hash(input);
+            assert_eq!(test_out, *array_ref!(expected_out, 0, 32));
+            // incremental
+            let mut hasher = crate::Hasher::new();
+            hasher.update(input);
+            assert_eq!(hasher.finalize(), *array_ref!(expected_out, 0, 32));
+            assert_eq!(hasher.finalize(), test_out);
+            // incremental (rayon)
+            #[cfg(feature = "rayon")]
+            {
+                let mut hasher = crate::Hasher::new();
+                hasher.update_rayon(input);
+                assert_eq!(hasher.finalize(), *array_ref!(expected_out, 0, 32));
+                assert_eq!(hasher.finalize(), test_out);
+            }
+            // xof
+            let mut extended = [0; OUT];
+            hasher.finalize_xof().fill(&mut extended);
+            assert_eq!(extended, expected_out);
+        }
+
+        // keyed
+        {
+            let mut reference_hasher = reference_impl::Hasher::new_keyed(&TEST_KEY);
+            reference_hasher.update(input);
+            let mut expected_out = [0; OUT];
+            reference_hasher.finalize(&mut expected_out);
+
+            // all at once
+            let test_out = crate::keyed_hash(&TEST_KEY, input);
+            assert_eq!(test_out, *array_ref!(expected_out, 0, 32));
+            // incremental
+            let mut hasher = crate::Hasher::new_keyed(&TEST_KEY);
+            hasher.update(input);
+            assert_eq!(hasher.finalize(), *array_ref!(expected_out, 0, 32));
+            assert_eq!(hasher.finalize(), test_out);
+            // incremental (rayon)
+            #[cfg(feature = "rayon")]
+            {
+                let mut hasher = crate::Hasher::new_keyed(&TEST_KEY);
+                hasher.update_rayon(input);
+                assert_eq!(hasher.finalize(), *array_ref!(expected_out, 0, 32));
+                assert_eq!(hasher.finalize(), test_out);
+            }
+            // xof
+            let mut extended = [0; OUT];
+            hasher.finalize_xof().fill(&mut extended);
+            assert_eq!(extended, expected_out);
+        }
+
+        // derive_key
+        {
+            let context = "BLAKE3 2019-12-27 16:13:59 example context (not the test vector one)";
+            let mut reference_hasher = reference_impl::Hasher::new_derive_key(context);
+            reference_hasher.update(input);
+            let mut expected_out = [0; OUT];
+            reference_hasher.finalize(&mut expected_out);
+
+            // all at once
+            let test_out = crate::derive_key(context, input);
+            assert_eq!(test_out, expected_out[..32]);
+            // incremental
+            let mut hasher = crate::Hasher::new_derive_key(context);
+            hasher.update(input);
+            assert_eq!(hasher.finalize(), *array_ref!(expected_out, 0, 32));
+            assert_eq!(hasher.finalize(), *array_ref!(test_out, 0, 32));
+            // incremental (rayon)
+            #[cfg(feature = "rayon")]
+            {
+                let mut hasher = crate::Hasher::new_derive_key(context);
+                hasher.update_rayon(input);
+                assert_eq!(hasher.finalize(), *array_ref!(expected_out, 0, 32));
+                assert_eq!(hasher.finalize(), *array_ref!(test_out, 0, 32));
+            }
+            // xof
+            let mut extended = [0; OUT];
+            hasher.finalize_xof().fill(&mut extended);
+            assert_eq!(extended, expected_out);
+        }
+    }
+}
+
+#[test]
+fn test_compare_reference_impl_long_xof() {
+    let mut reference_output = [0u8; 32 * BLOCK_LEN - 1];
+    let mut reference_hasher = reference_impl::Hasher::new_keyed(&TEST_KEY);
+    reference_hasher.update(b"hello world");
+    reference_hasher.finalize(&mut reference_output);
+
+    let mut test_output = [0u8; 32 * BLOCK_LEN - 1];
+    let mut test_hasher = crate::Hasher::new_keyed(&TEST_KEY);
+    test_hasher.update(b"hello world");
+    test_hasher.finalize_xof().fill(&mut test_output);
+
+    assert_eq!(reference_output, test_output);
+}
+
+#[test]
+fn test_xof_partial_blocks() {
+    const OUT_LEN: usize = 6 * BLOCK_LEN;
+    let mut reference_out = [0u8; OUT_LEN];
+    reference_impl::Hasher::new().finalize(&mut reference_out);
+
+    let mut all_at_once_out = [0u8; OUT_LEN];
+    crate::Hasher::new()
+        .finalize_xof()
+        .fill(&mut all_at_once_out);
+    assert_eq!(reference_out, all_at_once_out);
+
+    let mut partial_out = [0u8; OUT_LEN];
+    let partial_start = 32;
+    let partial_end = OUT_LEN - 32;
+    let mut xof = crate::Hasher::new().finalize_xof();
+    xof.fill(&mut partial_out[..partial_start]);
+    xof.fill(&mut partial_out[partial_start..partial_end]);
+    xof.fill(&mut partial_out[partial_end..]);
+    assert_eq!(reference_out, partial_out);
+}
+
+fn reference_hash(input: &[u8]) -> crate::Hash {
+    let mut hasher = reference_impl::Hasher::new();
+    hasher.update(input);
+    let mut bytes = [0; 32];
+    hasher.finalize(&mut bytes);
+    bytes.into()
+}
+
+#[test]
+fn test_compare_update_multiple() {
+    // Don't use all the long test cases here, since that's unnecessarily slow
+    // in debug mode.
+    let mut short_test_cases = TEST_CASES;
+    while *short_test_cases.last().unwrap() > 4 * CHUNK_LEN {
+        short_test_cases = &short_test_cases[..short_test_cases.len() - 1];
+    }
+    assert_eq!(*short_test_cases.last().unwrap(), 4 * CHUNK_LEN);
+
+    let mut input_buf = [0; 2 * TEST_CASES_MAX];
+    paint_test_input(&mut input_buf);
+
+    for &first_update in short_test_cases {
+        #[cfg(feature = "std")]
+        dbg!(first_update);
+        let first_input = &input_buf[..first_update];
+        let mut test_hasher = crate::Hasher::new();
+        test_hasher.update(first_input);
+
+        for &second_update in short_test_cases {
+            #[cfg(feature = "std")]
+            dbg!(second_update);
+            let second_input = &input_buf[first_update..][..second_update];
+            let total_input = &input_buf[..first_update + second_update];
+
+            // Clone the hasher with first_update bytes already written, so
+            // that the next iteration can reuse it.
+            let mut test_hasher = test_hasher.clone();
+            test_hasher.update(second_input);
+            let expected = reference_hash(total_input);
+            assert_eq!(expected, test_hasher.finalize());
+        }
+    }
+}
+
+#[test]
+fn test_fuzz_hasher() {
+    const INPUT_MAX: usize = 4 * CHUNK_LEN;
+    let mut input_buf = [0; 3 * INPUT_MAX];
+    paint_test_input(&mut input_buf);
+
+    // Don't do too many iterations in debug mode, to keep the tests under a
+    // second or so. CI should run tests in release mode also. Provide an
+    // environment variable for specifying a larger number of fuzz iterations.
+    let num_tests = if cfg!(debug_assertions) { 100 } else { 10_000 };
+
+    // Use a fixed RNG seed for reproducibility.
+    let mut rng = rand_chacha::ChaCha8Rng::from_seed([1; 32]);
+    for _num_test in 0..num_tests {
+        #[cfg(feature = "std")]
+        dbg!(_num_test);
+        let mut hasher = crate::Hasher::new();
+        let mut total_input = 0;
+        // For each test, write 3 inputs of random length.
+        for _ in 0..3 {
+            let input_len = rng.random_range(0..(INPUT_MAX + 1));
+            #[cfg(feature = "std")]
+            dbg!(input_len);
+            let input = &input_buf[total_input..][..input_len];
+            hasher.update(input);
+            total_input += input_len;
+        }
+        let expected = reference_hash(&input_buf[..total_input]);
+        assert_eq!(expected, hasher.finalize());
+    }
+}
+
+#[test]
+fn test_fuzz_xof() {
+    let mut input_buf = [0u8; 3 * BLOCK_LEN];
+    paint_test_input(&mut input_buf);
+
+    // Don't do too many iterations in debug mode, to keep the tests under a
+    // second or so. CI should run tests in release mode also. Provide an
+    // environment variable for specifying a larger number of fuzz iterations.
+    let num_tests = if cfg!(debug_assertions) { 100 } else { 2500 };
+
+    // Use a fixed RNG seed for reproducibility.
+    let mut rng = rand_chacha::ChaCha8Rng::from_seed([1; 32]);
+    for _num_test in 0..num_tests {
+        #[cfg(feature = "std")]
+        dbg!(_num_test);
+        // 31 (16 + 8 + 4 + 2 + 1) outputs
+        let mut output_buf = [0; 31 * CHUNK_LEN];
+        let input_len = rng.random_range(0..input_buf.len());
+        let mut xof = crate::Hasher::new()
+            .update(&input_buf[..input_len])
+            .finalize_xof();
+        let partial_start = rng.random_range(0..output_buf.len());
+        let partial_end = rng.random_range(partial_start..output_buf.len());
+        xof.fill(&mut output_buf[..partial_start]);
+        xof.fill(&mut output_buf[partial_start..partial_end]);
+        xof.fill(&mut output_buf[partial_end..]);
+
+        let mut reference_buf = [0; 31 * CHUNK_LEN];
+        let mut reference_hasher = reference_impl::Hasher::new();
+        reference_hasher.update(&input_buf[..input_len]);
+        reference_hasher.finalize(&mut reference_buf);
+
+        assert_eq!(reference_buf, output_buf);
+    }
+}
+
+#[test]
+fn test_xof_seek() {
+    let mut out = [0; 533];
+    let mut hasher = crate::Hasher::new();
+    hasher.update(b"foo");
+    hasher.finalize_xof().fill(&mut out);
+    assert_eq!(hasher.finalize().as_bytes(), &out[0..32]);
+
+    let mut reader = hasher.finalize_xof();
+    reader.set_position(303);
+    let mut out2 = [0; 102];
+    reader.fill(&mut out2);
+    assert_eq!(&out[303..][..102], &out2[..]);
+
+    #[cfg(feature = "std")]
+    {
+        use std::io::prelude::*;
+        let mut reader = hasher.finalize_xof();
+        reader.seek(std::io::SeekFrom::Start(303)).unwrap();
+        let mut out3 = Vec::new();
+        reader.by_ref().take(102).read_to_end(&mut out3).unwrap();
+        assert_eq!(&out[303..][..102], &out3[..]);
+
+        assert_eq!(
+            reader.seek(std::io::SeekFrom::Current(0)).unwrap(),
+            303 + 102
+        );
+        reader.seek(std::io::SeekFrom::Current(-5)).unwrap();
+        assert_eq!(
+            reader.seek(std::io::SeekFrom::Current(0)).unwrap(),
+            303 + 102 - 5
+        );
+        let mut out4 = [0; 17];
+        assert_eq!(reader.read(&mut out4).unwrap(), 17);
+        assert_eq!(&out[303 + 102 - 5..][..17], &out4[..]);
+        assert_eq!(
+            reader.seek(std::io::SeekFrom::Current(0)).unwrap(),
+            303 + 102 - 5 + 17
+        );
+        assert!(reader.seek(std::io::SeekFrom::End(0)).is_err());
+        assert!(reader.seek(std::io::SeekFrom::Current(-1000)).is_err());
+    }
+}
+
+#[test]
+fn test_msg_schedule_permutation() {
+    let permutation = [2, 6, 3, 10, 7, 0, 4, 13, 1, 11, 12, 5, 9, 14, 15, 8];
+
+    let mut generated = [[0; 16]; 7];
+    generated[0] = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15];
+
+    for round in 1..7 {
+        for i in 0..16 {
+            generated[round][i] = generated[round - 1][permutation[i]];
+        }
+    }
+
+    assert_eq!(generated, crate::MSG_SCHEDULE);
+}
+
+#[test]
+fn test_reset() {
+    let mut hasher = crate::Hasher::new();
+    hasher.update(&[42; 3 * CHUNK_LEN + 7]);
+    hasher.reset();
+    hasher.update(&[42; CHUNK_LEN + 3]);
+    assert_eq!(hasher.finalize(), crate::hash(&[42; CHUNK_LEN + 3]));
+
+    let key = &[99; crate::KEY_LEN];
+    let mut keyed_hasher = crate::Hasher::new_keyed(key);
+    keyed_hasher.update(&[42; 3 * CHUNK_LEN + 7]);
+    keyed_hasher.reset();
+    keyed_hasher.update(&[42; CHUNK_LEN + 3]);
+    assert_eq!(
+        keyed_hasher.finalize(),
+        crate::keyed_hash(key, &[42; CHUNK_LEN + 3]),
+    );
+
+    let context = "BLAKE3 2020-02-12 10:20:58 reset test";
+    let mut kdf = crate::Hasher::new_derive_key(context);
+    kdf.update(&[42; 3 * CHUNK_LEN + 7]);
+    kdf.reset();
+    kdf.update(&[42; CHUNK_LEN + 3]);
+    let expected = crate::derive_key(context, &[42; CHUNK_LEN + 3]);
+    assert_eq!(kdf.finalize(), expected);
+}
+
+#[test]
+fn test_hex_encoding_decoding() {
+    let digest_str = "04e0bb39f30b1a3feb89f536c93be15055482df748674b00d26e5a75777702e9";
+    let mut hasher = crate::Hasher::new();
+    hasher.update(b"foo");
+    let digest = hasher.finalize();
+    assert_eq!(digest.to_hex().as_str(), digest_str);
+    #[cfg(feature = "std")]
+    assert_eq!(digest.to_string(), digest_str);
+
+    // Test round trip
+    let digest = crate::Hash::from_hex(digest_str).unwrap();
+    assert_eq!(digest.to_hex().as_str(), digest_str);
+
+    // Test uppercase
+    let digest = crate::Hash::from_hex(digest_str.to_uppercase()).unwrap();
+    assert_eq!(digest.to_hex().as_str(), digest_str);
+
+    // Test string parsing via FromStr
+    let digest: crate::Hash = digest_str.parse().unwrap();
+    assert_eq!(digest.to_hex().as_str(), digest_str);
+
+    // Test errors
+    let bad_len = "04e0bb39f30b1";
+    let _result = crate::Hash::from_hex(bad_len).unwrap_err();
+    #[cfg(feature = "std")]
+    assert_eq!(_result.to_string(), "expected 64 hex bytes, received 13");
+
+    let bad_char = "Z4e0bb39f30b1a3feb89f536c93be15055482df748674b00d26e5a75777702e9";
+    let _result = crate::Hash::from_hex(bad_char).unwrap_err();
+    #[cfg(feature = "std")]
+    assert_eq!(_result.to_string(), "invalid hex character: 'Z'");
+
+    let _result = crate::Hash::from_hex([128; 64]).unwrap_err();
+    #[cfg(feature = "std")]
+    assert_eq!(_result.to_string(), "invalid hex character: 0x80");
+}
+
+// This test is a mimized failure case for the Windows SSE2 bug described in
+// https://github.com/BLAKE3-team/BLAKE3/issues/206.
+//
+// Before that issue was fixed, this test would fail on Windows in the following configuration:
+//
+//     cargo test --features=no_avx512,no_avx2,no_sse41 --release
+//
+// Bugs like this one (stomping on a caller's register) are very sensitive to the details of
+// surrounding code, so it's not especially likely that this test will catch another bug (or even
+// the same bug) in the future. Still, there's no harm in keeping it.
+#[test]
+fn test_issue_206_windows_sse2() {
+    // This stupid loop has to be here to trigger the bug. I don't know why.
+    for _ in &[0] {
+        // The length 65 (two blocks) is significant. It doesn't repro with 64 (one block). It also
+        // doesn't repro with an all-zero input.
+        let input = &[0xff; 65];
+        let expected_hash = [
+            183, 235, 50, 217, 156, 24, 190, 219, 2, 216, 176, 255, 224, 53, 28, 95, 57, 148, 179,
+            245, 162, 90, 37, 121, 0, 142, 219, 62, 234, 204, 225, 161,
+        ];
+
+        // This throwaway call has to be here to trigger the bug.
+        crate::Hasher::new().update(input);
+
+        // This assert fails when the bug is triggered.
+        assert_eq!(crate::Hasher::new().update(input).finalize(), expected_hash);
+    }
+}
+
+#[test]
+fn test_hash_conversions() {
+    let bytes1 = [42; 32];
+    let hash1: crate::Hash = bytes1.into();
+    let bytes2: [u8; 32] = hash1.into();
+    assert_eq!(bytes1, bytes2);
+
+    let bytes3 = *hash1.as_bytes();
+    assert_eq!(bytes1, bytes3);
+
+    let hash2 = crate::Hash::from_bytes(bytes1);
+    assert_eq!(hash1, hash2);
+
+    let hex = hash1.to_hex();
+    let hash3 = crate::Hash::from_hex(hex.as_bytes()).unwrap();
+    assert_eq!(hash1, hash3);
+
+    let slice1: &[u8] = bytes1.as_slice();
+    let hash4 = crate::Hash::from_slice(slice1).expect("correct length");
+    assert_eq!(hash1, hash4);
+
+    assert!(crate::Hash::from_slice(&[]).is_err());
+    assert!(crate::Hash::from_slice(&[42]).is_err());
+    assert!(crate::Hash::from_slice([42; 31].as_slice()).is_err());
+    assert!(crate::Hash::from_slice([42; 33].as_slice()).is_err());
+    assert!(crate::Hash::from_slice([42; 100].as_slice()).is_err());
+}
+
+#[test]
+const fn test_hash_const_conversions() {
+    let bytes = [42; 32];
+    let hash = crate::Hash::from_bytes(bytes);
+    _ = hash.as_bytes();
+}
+
+#[cfg(feature = "zeroize")]
+#[test]
+fn test_zeroize() {
+    use zeroize::Zeroize;
+
+    let mut hash = crate::Hash([42; 32]);
+    hash.zeroize();
+    assert_eq!(hash.0, [0u8; 32]);
+
+    let mut hasher = crate::Hasher {
+        chunk_state: crate::ChunkState {
+            cv: [42; 8],
+            chunk_counter: 42,
+            buf: [42; 64],
+            buf_len: 42,
+            blocks_compressed: 42,
+            flags: 42,
+            platform: crate::Platform::Portable,
+        },
+        initial_chunk_counter: 42,
+        key: [42; 8],
+        cv_stack: [[42; 32]; { crate::MAX_DEPTH + 1 }].into(),
+    };
+    hasher.zeroize();
+    assert_eq!(hasher.chunk_state.cv, [0; 8]);
+    assert_eq!(hasher.chunk_state.chunk_counter, 0);
+    assert_eq!(hasher.chunk_state.buf, [0; 64]);
+    assert_eq!(hasher.chunk_state.buf_len, 0);
+    assert_eq!(hasher.chunk_state.blocks_compressed, 0);
+    assert_eq!(hasher.chunk_state.flags, 0);
+    assert!(matches!(
+        hasher.chunk_state.platform,
+        crate::Platform::Portable
+    ));
+    assert_eq!(hasher.initial_chunk_counter, 0);
+    assert_eq!(hasher.key, [0; 8]);
+    assert_eq!(&*hasher.cv_stack, &[[0u8; 32]; 0]);
+
+    let mut output_reader = crate::OutputReader {
+        inner: crate::Output {
+            input_chaining_value: [42; 8],
+            block: [42; 64],
+            counter: 42,
+            block_len: 42,
+            flags: 42,
+            platform: crate::Platform::Portable,
+        },
+        position_within_block: 42,
+    };
+
+    output_reader.zeroize();
+    assert_eq!(output_reader.inner.input_chaining_value, [0; 8]);
+    assert_eq!(output_reader.inner.block, [0; 64]);
+    assert_eq!(output_reader.inner.counter, 0);
+    assert_eq!(output_reader.inner.block_len, 0);
+    assert_eq!(output_reader.inner.flags, 0);
+    assert!(matches!(
+        output_reader.inner.platform,
+        crate::Platform::Portable
+    ));
+    assert_eq!(output_reader.position_within_block, 0);
+}
+
+#[test]
+#[cfg(feature = "std")]
+fn test_update_reader() -> Result<(), std::io::Error> {
+    // This is a brief test, since update_reader() is mostly a wrapper around update(), which already
+    // has substantial testing.
+    let mut input = vec![0; 1_000_000];
+    paint_test_input(&mut input);
+    assert_eq!(
+        crate::Hasher::new().update_reader(&input[..])?.finalize(),
+        crate::hash(&input),
+    );
+    Ok(())
+}
+
+#[test]
+#[cfg(feature = "std")]
+fn test_update_reader_interrupted() -> std::io::Result<()> {
+    use std::io;
+    struct InterruptingReader<'a> {
+        already_interrupted: bool,
+        slice: &'a [u8],
+    }
+    impl<'a> InterruptingReader<'a> {
+        fn new(slice: &'a [u8]) -> Self {
+            Self {
+                already_interrupted: false,
+                slice,
+            }
+        }
+    }
+    impl<'a> io::Read for InterruptingReader<'a> {
+        fn read(&mut self, buf: &mut [u8]) -> io::Result<usize> {
+            if !self.already_interrupted {
+                self.already_interrupted = true;
+                return Err(io::Error::from(io::ErrorKind::Interrupted));
+            }
+            let take = std::cmp::min(self.slice.len(), buf.len());
+            buf[..take].copy_from_slice(&self.slice[..take]);
+            self.slice = &self.slice[take..];
+            Ok(take)
+        }
+    }
+
+    let input = b"hello world";
+    let mut reader = InterruptingReader::new(input);
+    let mut hasher = crate::Hasher::new();
+    hasher.update_reader(&mut reader)?;
+    assert_eq!(hasher.finalize(), crate::hash(input));
+    Ok(())
+}
+
+#[test]
+#[cfg(feature = "mmap")]
+// NamedTempFile isn't Miri-compatible
+#[cfg(not(miri))]
+fn test_mmap() -> Result<(), std::io::Error> {
+    // This is a brief test, since update_mmap() is mostly a wrapper around update(), which already
+    // has substantial testing.
+    use std::io::prelude::*;
+    let mut input = vec![0; 1_000_000];
+    paint_test_input(&mut input);
+    let mut tempfile = tempfile::NamedTempFile::new()?;
+    tempfile.write_all(&input)?;
+    tempfile.flush()?;
+    assert_eq!(
+        crate::Hasher::new()
+            .update_mmap(tempfile.path())?
+            .finalize(),
+        crate::hash(&input),
+    );
+    Ok(())
+}
+
+#[test]
+#[cfg(feature = "mmap")]
+#[cfg(target_os = "linux")]
+fn test_mmap_virtual_file() -> Result<(), std::io::Error> {
+    // Virtual files like /proc/version can't be mmapped, because their contents don't actually
+    // exist anywhere in memory. Make sure we fall back to regular file IO in these cases.
+    // Currently this is handled with a length check, where the assumption is that virtual files
+    // will always report length 0. If that assumption ever breaks, hopefully this test will catch
+    // it.
+    let virtual_filepath = "/proc/version";
+    let mut mmap_hasher = crate::Hasher::new();
+    // We'll fail right here if the fallback doesn't work.
+    mmap_hasher.update_mmap(virtual_filepath)?;
+    let mut read_hasher = crate::Hasher::new();
+    read_hasher.update_reader(std::fs::File::open(virtual_filepath)?)?;
+    assert_eq!(mmap_hasher.finalize(), read_hasher.finalize());
+    Ok(())
+}
+
+#[test]
+#[cfg(feature = "mmap")]
+#[cfg(feature = "rayon")]
+// NamedTempFile isn't Miri-compatible
+#[cfg(not(miri))]
+fn test_mmap_rayon() -> Result<(), std::io::Error> {
+    // This is a brief test, since update_mmap_rayon() is mostly a wrapper around update_rayon(),
+    // which already has substantial testing.
+    use std::io::prelude::*;
+    let mut input = vec![0; 1_000_000];
+    paint_test_input(&mut input);
+    let mut tempfile = tempfile::NamedTempFile::new()?;
+    tempfile.write_all(&input)?;
+    tempfile.flush()?;
+    assert_eq!(
+        crate::Hasher::new()
+            .update_mmap_rayon(tempfile.path())?
+            .finalize(),
+        crate::hash(&input),
+    );
+    Ok(())
+}
+
+#[test]
+#[cfg(feature = "std")]
+#[cfg(feature = "serde")]
+fn test_serde() {
+    // Henrik suggested that we use 0xfe / 254 for byte test data instead of 0xff / 255, due to the
+    // fact that 0xfe is not a well formed CBOR item.
+    let hash: crate::Hash = [0xfe; 32].into();
+
+    let json = serde_json::to_string(&hash).unwrap();
+    assert_eq!(
+        json,
+        "[254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254,254]",
+    );
+    let hash2: crate::Hash = serde_json::from_str(&json).unwrap();
+    assert_eq!(hash, hash2);
+
+    let mut cbor = Vec::<u8>::new();
+    ciborium::into_writer(&hash, &mut cbor).unwrap();
+    assert_eq!(
+        cbor,
+        [
+            0x98, 0x20, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe,
+            0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe,
+            0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe,
+            0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe,
+            0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe, 0x18, 0xfe,
+        ]
+    );
+    let hash_from_cbor: crate::Hash = ciborium::from_reader(&cbor[..]).unwrap();
+    assert_eq!(hash_from_cbor, hash);
+
+    // Version 1.5.2 of this crate changed the default serialization format to a bytestring
+    // (instead of an array/list) to save bytes on the wire. That was a backwards compatibility
+    // mistake for non-self-describing formats, and it's been reverted. Since some small number of
+    // serialized bytestrings will probably exist forever in the wild, we shold test that we can
+    // still deserialize these from self-describing formats.
+    let bytestring_cbor: &[u8] = &[
+        0x58, 0x20, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe,
+        0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe, 0xfe,
+        0xfe, 0xfe, 0xfe, 0xfe,
+    ];
+    let hash_from_bytestring_cbor: crate::Hash = ciborium::from_reader(bytestring_cbor).unwrap();
+    assert_eq!(hash_from_bytestring_cbor, hash);
+}
+
+// `cargo +nightly miri test` currently works, but it takes forever, because some of our test
+// inputs are quite large. Most of our unsafe code is platform specific and incompatible with Miri
+// anyway, but we'd like it to be possible for callers to run their own tests under Miri, assuming
+// they don't use incompatible features like Rayon or mmap. This test should get reasonable
+// coverage of our public API without using any large inputs, so we can run it in CI and catch
+// obvious breaks. (For example, constant_time_eq is not compatible with Miri.)
+#[test]
+fn test_miri_smoketest() {
+    let mut hasher = crate::Hasher::new_derive_key("Miri smoketest");
+    hasher.update(b"foo");
+    #[cfg(feature = "std")]
+    hasher.update_reader(&b"bar"[..]).unwrap();
+    assert_eq!(hasher.finalize(), hasher.finalize());
+    let mut reader = hasher.finalize_xof();
+    reader.set_position(999999);
+    reader.fill(&mut [0]);
+}
+
+// I had to move these tests out of the deprecated guts module, because leaving them there causes
+// an un-silenceable warning: https://github.com/rust-lang/rust/issues/47238
+#[cfg(test)]
+#[allow(deprecated)]
+mod guts_tests {
+    use crate::guts::*;
+
+    #[test]
+    fn test_chunk() {
+        assert_eq!(
+            crate::hash(b"foo"),
+            ChunkState::new(0).update(b"foo").finalize(true)
+        );
+    }
+
+    #[test]
+    fn test_parents() {
+        let mut hasher = crate::Hasher::new();
+        let mut buf = [0; crate::CHUNK_LEN];
+
+        buf[0] = 'a' as u8;
+        hasher.update(&buf);
+        let chunk0_cv = ChunkState::new(0).update(&buf).finalize(false);
+
+        buf[0] = 'b' as u8;
+        hasher.update(&buf);
+        let chunk1_cv = ChunkState::new(1).update(&buf).finalize(false);
+
+        hasher.update(b"c");
+        let chunk2_cv = ChunkState::new(2).update(b"c").finalize(false);
+
+        let parent = parent_cv(&chunk0_cv, &chunk1_cv, false);
+        let root = parent_cv(&parent, &chunk2_cv, true);
+        assert_eq!(hasher.finalize(), root);
+    }
+}
diff --git a/thirdparty/blake3/src/traits.rs b/thirdparty/blake3/src/traits.rs
new file mode 100644
index 000000000..70b1c0687
--- /dev/null
+++ b/thirdparty/blake3/src/traits.rs
@@ -0,0 +1,227 @@
+//! Implementations of commonly used traits like `Digest` and `Mac` from the
+//! [`digest`](https://crates.io/crates/digest) crate.
+
+pub use digest;
+
+use crate::{Hasher, OutputReader};
+use digest::crypto_common;
+use digest::generic_array::{typenum::U32, typenum::U64, GenericArray};
+
+impl digest::HashMarker for Hasher {}
+
+impl digest::Update for Hasher {
+    #[inline]
+    fn update(&mut self, data: &[u8]) {
+        self.update(data);
+    }
+}
+
+impl digest::Reset for Hasher {
+    #[inline]
+    fn reset(&mut self) {
+        self.reset(); // the inherent method
+    }
+}
+
+impl digest::OutputSizeUser for Hasher {
+    type OutputSize = U32;
+}
+
+impl digest::FixedOutput for Hasher {
+    #[inline]
+    fn finalize_into(self, out: &mut GenericArray<u8, Self::OutputSize>) {
+        out.copy_from_slice(self.finalize().as_bytes());
+    }
+}
+
+impl digest::FixedOutputReset for Hasher {
+    #[inline]
+    fn finalize_into_reset(&mut self, out: &mut GenericArray<u8, Self::OutputSize>) {
+        out.copy_from_slice(self.finalize().as_bytes());
+        self.reset();
+    }
+}
+
+impl digest::ExtendableOutput for Hasher {
+    type Reader = OutputReader;
+
+    #[inline]
+    fn finalize_xof(self) -> Self::Reader {
+        Hasher::finalize_xof(&self)
+    }
+}
+
+impl digest::ExtendableOutputReset for Hasher {
+    #[inline]
+    fn finalize_xof_reset(&mut self) -> Self::Reader {
+        let reader = Hasher::finalize_xof(self);
+        self.reset();
+        reader
+    }
+}
+
+impl digest::XofReader for OutputReader {
+    #[inline]
+    fn read(&mut self, buffer: &mut [u8]) {
+        self.fill(buffer);
+    }
+}
+
+impl crypto_common::KeySizeUser for Hasher {
+    type KeySize = U32;
+}
+
+impl crypto_common::BlockSizeUser for Hasher {
+    type BlockSize = U64;
+}
+
+impl digest::MacMarker for Hasher {}
+
+impl digest::KeyInit for Hasher {
+    #[inline]
+    fn new(key: &digest::Key<Self>) -> Self {
+        let key_bytes: [u8; 32] = (*key).into();
+        Hasher::new_keyed(&key_bytes)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_digest_traits() {
+        // Inherent methods.
+        let mut hasher1 = crate::Hasher::new();
+        hasher1.update(b"foo");
+        hasher1.update(b"bar");
+        hasher1.update(b"baz");
+        let out1 = hasher1.finalize();
+        let mut xof1 = [0; 301];
+        hasher1.finalize_xof().fill(&mut xof1);
+        assert_eq!(out1.as_bytes(), &xof1[..32]);
+
+        // Trait implementations.
+        let mut hasher2: crate::Hasher = digest::Digest::new();
+        digest::Digest::update(&mut hasher2, b"xxx");
+        digest::Digest::reset(&mut hasher2);
+        digest::Digest::update(&mut hasher2, b"foo");
+        digest::Digest::update(&mut hasher2, b"bar");
+        digest::Digest::update(&mut hasher2, b"baz");
+        let out2 = digest::Digest::finalize(hasher2.clone());
+        let mut xof2 = [0; 301];
+        digest::XofReader::read(
+            &mut digest::ExtendableOutput::finalize_xof(hasher2.clone()),
+            &mut xof2,
+        );
+        assert_eq!(out1.as_bytes(), &out2[..]);
+        assert_eq!(xof1[..], xof2[..]);
+
+        // Again with the resetting variants.
+        let mut hasher3: crate::Hasher = digest::Digest::new();
+        digest::Digest::update(&mut hasher3, b"foobarbaz");
+        let mut out3 = [0; 32];
+        digest::FixedOutputReset::finalize_into_reset(
+            &mut hasher3,
+            GenericArray::from_mut_slice(&mut out3),
+        );
+        digest::Digest::update(&mut hasher3, b"foobarbaz");
+        let mut out4 = [0; 32];
+        digest::FixedOutputReset::finalize_into_reset(
+            &mut hasher3,
+            GenericArray::from_mut_slice(&mut out4),
+        );
+        digest::Digest::update(&mut hasher3, b"foobarbaz");
+        let mut xof3 = [0; 301];
+        digest::XofReader::read(
+            &mut digest::ExtendableOutputReset::finalize_xof_reset(&mut hasher3),
+            &mut xof3,
+        );
+        digest::Digest::update(&mut hasher3, b"foobarbaz");
+        let mut xof4 = [0; 301];
+        digest::XofReader::read(
+            &mut digest::ExtendableOutputReset::finalize_xof_reset(&mut hasher3),
+            &mut xof4,
+        );
+        assert_eq!(out1.as_bytes(), &out3[..]);
+        assert_eq!(out1.as_bytes(), &out4[..]);
+        assert_eq!(xof1[..], xof3[..]);
+        assert_eq!(xof1[..], xof4[..]);
+    }
+
+    #[test]
+    fn test_mac_trait() {
+        // Inherent methods.
+        let key = b"some super secret key bytes fooo";
+        let mut hasher1 = crate::Hasher::new_keyed(key);
+        hasher1.update(b"foo");
+        hasher1.update(b"bar");
+        hasher1.update(b"baz");
+        let out1 = hasher1.finalize();
+
+        // Trait implementation.
+        let generic_key = (*key).into();
+        let mut hasher2: crate::Hasher = digest::Mac::new(&generic_key);
+        digest::Mac::update(&mut hasher2, b"xxx");
+        digest::Mac::reset(&mut hasher2);
+        digest::Mac::update(&mut hasher2, b"foo");
+        digest::Mac::update(&mut hasher2, b"bar");
+        digest::Mac::update(&mut hasher2, b"baz");
+        let out2 = digest::Mac::finalize(hasher2);
+        assert_eq!(out1.as_bytes(), out2.into_bytes().as_slice());
+    }
+
+    fn expected_hmac_blake3(key: &[u8], input: &[u8]) -> [u8; 32] {
+        // See https://en.wikipedia.org/wiki/HMAC.
+        let key_hash;
+        let key_prime = if key.len() <= 64 {
+            key
+        } else {
+            key_hash = *crate::hash(key).as_bytes();
+            &key_hash
+        };
+        let mut ipad = [0x36; 64];
+        let mut opad = [0x5c; 64];
+        for i in 0..key_prime.len() {
+            ipad[i] ^= key_prime[i];
+            opad[i] ^= key_prime[i];
+        }
+        let mut inner_state = crate::Hasher::new();
+        inner_state.update(&ipad);
+        inner_state.update(input);
+        let mut outer_state = crate::Hasher::new();
+        outer_state.update(&opad);
+        outer_state.update(inner_state.finalize().as_bytes());
+        outer_state.finalize().into()
+    }
+
+    #[test]
+    fn test_hmac_compatibility() {
+        use hmac::{Mac, SimpleHmac};
+
+        // Test a short key.
+        let mut x = SimpleHmac::<Hasher>::new_from_slice(b"key").unwrap();
+        hmac::digest::Update::update(&mut x, b"data");
+        let output = x.finalize().into_bytes();
+        assert_ne!(output.len(), 0);
+        let expected = expected_hmac_blake3(b"key", b"data");
+        assert_eq!(expected, output.as_ref());
+
+        // Test a range of key and data lengths, particularly to exercise the long-key logic.
+        let mut input_bytes = [0; crate::test::TEST_CASES_MAX];
+        crate::test::paint_test_input(&mut input_bytes);
+        for &input_len in crate::test::TEST_CASES {
+            #[cfg(feature = "std")]
+            dbg!(input_len);
+            let input = &input_bytes[..input_len];
+
+            let mut x = SimpleHmac::<Hasher>::new_from_slice(input).unwrap();
+            hmac::digest::Update::update(&mut x, input);
+            let output = x.finalize().into_bytes();
+            assert_ne!(output.len(), 0);
+
+            let expected = expected_hmac_blake3(input, input);
+            assert_eq!(expected, output.as_ref());
+        }
+    }
+}
diff --git a/thirdparty/blake3/src/wasm32_simd.rs b/thirdparty/blake3/src/wasm32_simd.rs
new file mode 100644
index 000000000..135249158
--- /dev/null
+++ b/thirdparty/blake3/src/wasm32_simd.rs
@@ -0,0 +1,794 @@
+/*
+ * This code is based on rust_sse2.rs of the same distribution, and is subject to further improvements.
+ * Some comments are left intact even if their applicability is questioned.
+ *
+ * Performance measurements with a primitive benchmark with ~16Kb of data:
+ *
+ * | M1 native     | 11,610 ns |
+ * | M1 Wasm SIMD  | 13,355 ns |
+ * | M1 Wasm       | 22,037 ns |
+ * | x64 native    |  6,713 ns |
+ * | x64 Wasm SIMD | 11,985 ns |
+ * | x64 Wasm      | 25,978 ns |
+ *
+ * wasmtime v12.0.1 was used on both platforms.
+ */
+
+use core::arch::wasm32::*;
+
+use crate::{
+    counter_high, counter_low, CVBytes, CVWords, IncrementCounter, BLOCK_LEN, IV, MSG_SCHEDULE,
+    OUT_LEN,
+};
+use arrayref::{array_mut_ref, array_ref, mut_array_refs};
+
+pub const DEGREE: usize = 4;
+
+#[inline(always)]
+unsafe fn loadu(src: *const u8) -> v128 {
+    // This is an unaligned load, so the pointer cast is allowed.
+    v128_load(src as *const v128)
+}
+
+#[inline(always)]
+unsafe fn storeu(src: v128, dest: *mut u8) {
+    // This is an unaligned store, so the pointer cast is allowed.
+    v128_store(dest as *mut v128, src)
+}
+
+#[inline(always)]
+fn add(a: v128, b: v128) -> v128 {
+    i32x4_add(a, b)
+}
+
+#[inline(always)]
+fn xor(a: v128, b: v128) -> v128 {
+    v128_xor(a, b)
+}
+
+#[inline(always)]
+fn set1(x: u32) -> v128 {
+    i32x4_splat(x as i32)
+}
+
+#[inline(always)]
+fn set4(a: u32, b: u32, c: u32, d: u32) -> v128 {
+    i32x4(a as i32, b as i32, c as i32, d as i32)
+}
+
+// These rotations are the "simple/shifts version". For the
+// "complicated/shuffles version", see
+// https://github.com/sneves/blake2-avx2/blob/b3723921f668df09ece52dcd225a36d4a4eea1d9/blake2s-common.h#L63-L66.
+// For a discussion of the tradeoffs, see
+// https://github.com/sneves/blake2-avx2/pull/5. Due to an LLVM bug
+// (https://bugs.llvm.org/show_bug.cgi?id=44379), this version performs better
+// on recent x86 chips.
+#[inline(always)]
+fn rot16(a: v128) -> v128 {
+    v128_or(u32x4_shr(a, 16), u32x4_shl(a, 32 - 16))
+}
+
+#[inline(always)]
+fn rot12(a: v128) -> v128 {
+    v128_or(u32x4_shr(a, 12), u32x4_shl(a, 32 - 12))
+}
+
+#[inline(always)]
+fn rot8(a: v128) -> v128 {
+    v128_or(u32x4_shr(a, 8), u32x4_shl(a, 32 - 8))
+}
+
+#[inline(always)]
+fn rot7(a: v128) -> v128 {
+    v128_or(u32x4_shr(a, 7), u32x4_shl(a, 32 - 7))
+}
+
+#[inline(always)]
+fn g1(row0: &mut v128, row1: &mut v128, row2: &mut v128, row3: &mut v128, m: v128) {
+    *row0 = add(add(*row0, m), *row1);
+    *row3 = xor(*row3, *row0);
+    *row3 = rot16(*row3);
+    *row2 = add(*row2, *row3);
+    *row1 = xor(*row1, *row2);
+    *row1 = rot12(*row1);
+}
+
+#[inline(always)]
+fn g2(row0: &mut v128, row1: &mut v128, row2: &mut v128, row3: &mut v128, m: v128) {
+    *row0 = add(add(*row0, m), *row1);
+    *row3 = xor(*row3, *row0);
+    *row3 = rot8(*row3);
+    *row2 = add(*row2, *row3);
+    *row1 = xor(*row1, *row2);
+    *row1 = rot7(*row1);
+}
+
+// It could be a function, but artimetics in const generics is too limited yet.
+macro_rules! shuffle {
+    ($a: expr, $b: expr, $z:expr, $y:expr, $x:expr, $w:expr) => {
+        i32x4_shuffle::<{ $w }, { $x }, { $y + 4 }, { $z + 4 }>($a, $b)
+    };
+}
+
+#[inline(always)]
+fn unpacklo_epi64(a: v128, b: v128) -> v128 {
+    i64x2_shuffle::<0, 2>(a, b)
+}
+
+#[inline(always)]
+fn unpackhi_epi64(a: v128, b: v128) -> v128 {
+    i64x2_shuffle::<1, 3>(a, b)
+}
+
+#[inline(always)]
+fn unpacklo_epi32(a: v128, b: v128) -> v128 {
+    i32x4_shuffle::<0, 4, 1, 5>(a, b)
+}
+
+#[inline(always)]
+fn unpackhi_epi32(a: v128, b: v128) -> v128 {
+    i32x4_shuffle::<2, 6, 3, 7>(a, b)
+}
+
+#[inline(always)]
+fn shuffle_epi32<const I3: usize, const I2: usize, const I1: usize, const I0: usize>(
+    a: v128,
+) -> v128 {
+    // Please note that generic arguments in delcaration and imlementation are in
+    // different order.
+    // second arg is actually ignored.
+    i32x4_shuffle::<I0, I1, I2, I3>(a, a)
+}
+
+#[inline(always)]
+fn blend_epi16(a: v128, b: v128, imm8: i32) -> v128 {
+    // imm8 is always constant; it allows to implement this function with
+    // i16x8_shuffle.  However, it is marginally slower on x64.
+    let bits = i16x8(0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80);
+    let mut mask = i16x8_splat(imm8 as i16);
+    mask = v128_and(mask, bits);
+    mask = i16x8_eq(mask, bits);
+    // The swapped argument order is equivalent to mask negation.
+    v128_bitselect(b, a, mask)
+}
+
+// Note the optimization here of leaving row1 as the unrotated row, rather than
+// row0. All the message loads below are adjusted to compensate for this. See
+// discussion at https://github.com/sneves/blake2-avx2/pull/4
+#[inline(always)]
+fn diagonalize(row0: &mut v128, row2: &mut v128, row3: &mut v128) {
+    *row0 = shuffle_epi32::<2, 1, 0, 3>(*row0);
+    *row3 = shuffle_epi32::<1, 0, 3, 2>(*row3);
+    *row2 = shuffle_epi32::<0, 3, 2, 1>(*row2);
+}
+
+#[inline(always)]
+fn undiagonalize(row0: &mut v128, row2: &mut v128, row3: &mut v128) {
+    *row0 = shuffle_epi32::<0, 3, 2, 1>(*row0);
+    *row3 = shuffle_epi32::<1, 0, 3, 2>(*row3);
+    *row2 = shuffle_epi32::<2, 1, 0, 3>(*row2);
+}
+
+#[inline(always)]
+fn compress_pre(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [v128; 4] {
+    // safe because CVWords is [u32; 8]
+    let row0 = &mut unsafe { loadu(cv.as_ptr().add(0) as *const u8) };
+    let row1 = &mut unsafe { loadu(cv.as_ptr().add(4) as *const u8) };
+    let row2 = &mut set4(IV[0], IV[1], IV[2], IV[3]);
+    let row3 = &mut set4(
+        counter_low(counter),
+        counter_high(counter),
+        block_len as u32,
+        flags as u32,
+    );
+
+    // safe because block is &[u8; 64]
+    let mut m0 = unsafe { loadu(block.as_ptr().add(0 * 4 * DEGREE)) };
+    let mut m1 = unsafe { loadu(block.as_ptr().add(1 * 4 * DEGREE)) };
+    let mut m2 = unsafe { loadu(block.as_ptr().add(2 * 4 * DEGREE)) };
+    let mut m3 = unsafe { loadu(block.as_ptr().add(3 * 4 * DEGREE)) };
+
+    let mut t0;
+    let mut t1;
+    let mut t2;
+    let mut t3;
+    let mut tt;
+
+    // Round 1. The first round permutes the message words from the original
+    // input order, into the groups that get mixed in parallel.
+    t0 = shuffle!(m0, m1, 2, 0, 2, 0); //  6  4  2  0
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle!(m0, m1, 3, 1, 3, 1); //  7  5  3  1
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = shuffle!(m2, m3, 2, 0, 2, 0); // 14 12 10  8
+    t2 = shuffle_epi32::<2, 1, 0, 3>(t2); // 12 10  8 14
+    g1(row0, row1, row2, row3, t2);
+    t3 = shuffle!(m2, m3, 3, 1, 3, 1); // 15 13 11  9
+    t3 = shuffle_epi32::<2, 1, 0, 3>(t3); // 13 11  9 15
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 2. This round and all following rounds apply a fixed permutation
+    // to the message words from the round before.
+    t0 = shuffle!(m0, m1, 3, 1, 1, 2);
+    t0 = shuffle_epi32::<0, 3, 2, 1>(t0);
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle!(m2, m3, 3, 3, 2, 2);
+    tt = shuffle_epi32::<0, 0, 3, 3>(m0);
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = shuffle_epi32::<1, 3, 2, 0>(tt);
+    g1(row0, row1, row2, row3, t2);
+    t3 = unpackhi_epi32(m1, m3);
+    tt = unpacklo_epi32(m2, t3);
+    t3 = shuffle_epi32::<0, 1, 3, 2>(tt);
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 3
+    t0 = shuffle!(m0, m1, 3, 1, 1, 2);
+    t0 = shuffle_epi32::<0, 3, 2, 1>(t0);
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle!(m2, m3, 3, 3, 2, 2);
+    tt = shuffle_epi32::<0, 0, 3, 3>(m0);
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = shuffle_epi32::<1, 3, 2, 0>(tt);
+    g1(row0, row1, row2, row3, t2);
+    t3 = unpackhi_epi32(m1, m3);
+    tt = unpacklo_epi32(m2, t3);
+    t3 = shuffle_epi32::<0, 1, 3, 2>(tt);
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 4
+    t0 = shuffle!(m0, m1, 3, 1, 1, 2);
+    t0 = shuffle_epi32::<0, 3, 2, 1>(t0);
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle!(m2, m3, 3, 3, 2, 2);
+    tt = shuffle_epi32::<0, 0, 3, 3>(m0);
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = shuffle_epi32::<1, 3, 2, 0>(tt);
+    g1(row0, row1, row2, row3, t2);
+    t3 = unpackhi_epi32(m1, m3);
+    tt = unpacklo_epi32(m2, t3);
+    t3 = shuffle_epi32::<0, 1, 3, 2>(tt);
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 5
+    t0 = shuffle!(m0, m1, 3, 1, 1, 2);
+    t0 = shuffle_epi32::<0, 3, 2, 1>(t0);
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle!(m2, m3, 3, 3, 2, 2);
+    tt = shuffle_epi32::<0, 0, 3, 3>(m0);
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = shuffle_epi32::<1, 3, 2, 0>(tt);
+    g1(row0, row1, row2, row3, t2);
+    t3 = unpackhi_epi32(m1, m3);
+    tt = unpacklo_epi32(m2, t3);
+    t3 = shuffle_epi32::<0, 1, 3, 2>(tt);
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 6
+    t0 = shuffle!(m0, m1, 3, 1, 1, 2);
+    t0 = shuffle_epi32::<0, 3, 2, 1>(t0);
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle!(m2, m3, 3, 3, 2, 2);
+    tt = shuffle_epi32::<0, 0, 3, 3>(m0);
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = shuffle_epi32::<1, 3, 2, 0>(tt);
+    g1(row0, row1, row2, row3, t2);
+    t3 = unpackhi_epi32(m1, m3);
+    tt = unpacklo_epi32(m2, t3);
+    t3 = shuffle_epi32::<0, 1, 3, 2>(tt);
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+    m0 = t0;
+    m1 = t1;
+    m2 = t2;
+    m3 = t3;
+
+    // Round 7
+    t0 = shuffle!(m0, m1, 3, 1, 1, 2);
+    t0 = shuffle_epi32::<0, 3, 2, 1>(t0);
+    g1(row0, row1, row2, row3, t0);
+    t1 = shuffle!(m2, m3, 3, 3, 2, 2);
+    tt = shuffle_epi32::<0, 0, 3, 3>(m0);
+    t1 = blend_epi16(tt, t1, 0xCC);
+    g2(row0, row1, row2, row3, t1);
+    diagonalize(row0, row2, row3);
+    t2 = unpacklo_epi64(m3, m1);
+    tt = blend_epi16(t2, m2, 0xC0);
+    t2 = shuffle_epi32::<1, 3, 2, 0>(tt);
+    g1(row0, row1, row2, row3, t2);
+    t3 = unpackhi_epi32(m1, m3);
+    tt = unpacklo_epi32(m2, t3);
+    t3 = shuffle_epi32::<0, 1, 3, 2>(tt);
+    g2(row0, row1, row2, row3, t3);
+    undiagonalize(row0, row2, row3);
+
+    [*row0, *row1, *row2, *row3]
+}
+
+#[target_feature(enable = "simd128")]
+pub fn compress_in_place(
+    cv: &mut CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) {
+    let [row0, row1, row2, row3] = compress_pre(cv, block, block_len, counter, flags);
+    // it stores in reversed order...
+    // safe because CVWords is [u32; 8]
+    unsafe {
+        storeu(xor(row0, row2), cv.as_mut_ptr().add(0) as *mut u8);
+        storeu(xor(row1, row3), cv.as_mut_ptr().add(4) as *mut u8);
+    }
+}
+
+#[target_feature(enable = "simd128")]
+pub fn compress_xof(
+    cv: &CVWords,
+    block: &[u8; BLOCK_LEN],
+    block_len: u8,
+    counter: u64,
+    flags: u8,
+) -> [u8; 64] {
+    let [mut row0, mut row1, mut row2, mut row3] =
+        compress_pre(cv, block, block_len, counter, flags);
+    row0 = xor(row0, row2);
+    row1 = xor(row1, row3);
+    // safe because CVWords is [u32; 8]
+    row2 = xor(row2, unsafe { loadu(cv.as_ptr().add(0) as *const u8) });
+    row3 = xor(row3, unsafe { loadu(cv.as_ptr().add(4) as *const u8) });
+    // It seems to be architecture dependent, but works.
+    // safe because sizes match, and every state of u8 is valid.
+    unsafe { core::mem::transmute([row0, row1, row2, row3]) }
+}
+
+#[inline(always)]
+fn round(v: &mut [v128; 16], m: &[v128; 16], r: usize) {
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][0] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][2] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][4] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][6] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[15] = rot16(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot12(v[4]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][1] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][3] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][5] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][7] as usize]);
+    v[0] = add(v[0], v[4]);
+    v[1] = add(v[1], v[5]);
+    v[2] = add(v[2], v[6]);
+    v[3] = add(v[3], v[7]);
+    v[12] = xor(v[12], v[0]);
+    v[13] = xor(v[13], v[1]);
+    v[14] = xor(v[14], v[2]);
+    v[15] = xor(v[15], v[3]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[15] = rot8(v[15]);
+    v[8] = add(v[8], v[12]);
+    v[9] = add(v[9], v[13]);
+    v[10] = add(v[10], v[14]);
+    v[11] = add(v[11], v[15]);
+    v[4] = xor(v[4], v[8]);
+    v[5] = xor(v[5], v[9]);
+    v[6] = xor(v[6], v[10]);
+    v[7] = xor(v[7], v[11]);
+    v[4] = rot7(v[4]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][8] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][10] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][12] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][14] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot16(v[15]);
+    v[12] = rot16(v[12]);
+    v[13] = rot16(v[13]);
+    v[14] = rot16(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot12(v[5]);
+    v[6] = rot12(v[6]);
+    v[7] = rot12(v[7]);
+    v[4] = rot12(v[4]);
+    v[0] = add(v[0], m[MSG_SCHEDULE[r][9] as usize]);
+    v[1] = add(v[1], m[MSG_SCHEDULE[r][11] as usize]);
+    v[2] = add(v[2], m[MSG_SCHEDULE[r][13] as usize]);
+    v[3] = add(v[3], m[MSG_SCHEDULE[r][15] as usize]);
+    v[0] = add(v[0], v[5]);
+    v[1] = add(v[1], v[6]);
+    v[2] = add(v[2], v[7]);
+    v[3] = add(v[3], v[4]);
+    v[15] = xor(v[15], v[0]);
+    v[12] = xor(v[12], v[1]);
+    v[13] = xor(v[13], v[2]);
+    v[14] = xor(v[14], v[3]);
+    v[15] = rot8(v[15]);
+    v[12] = rot8(v[12]);
+    v[13] = rot8(v[13]);
+    v[14] = rot8(v[14]);
+    v[10] = add(v[10], v[15]);
+    v[11] = add(v[11], v[12]);
+    v[8] = add(v[8], v[13]);
+    v[9] = add(v[9], v[14]);
+    v[5] = xor(v[5], v[10]);
+    v[6] = xor(v[6], v[11]);
+    v[7] = xor(v[7], v[8]);
+    v[4] = xor(v[4], v[9]);
+    v[5] = rot7(v[5]);
+    v[6] = rot7(v[6]);
+    v[7] = rot7(v[7]);
+    v[4] = rot7(v[4]);
+}
+
+#[inline(always)]
+fn transpose_vecs(vecs: &mut [v128; DEGREE]) {
+    // Interleave 32-bit lanes. The low unpack is lanes 00/11 and the high is
+    // 22/33. Note that this doesn't split the vector into two lanes, as the
+    // AVX2 counterparts do.
+    let ab_01 = unpacklo_epi32(vecs[0], vecs[1]);
+    let ab_23 = unpackhi_epi32(vecs[0], vecs[1]);
+    let cd_01 = unpacklo_epi32(vecs[2], vecs[3]);
+    let cd_23 = unpackhi_epi32(vecs[2], vecs[3]);
+
+    // Interleave 64-bit lanes.
+    let abcd_0 = unpacklo_epi64(ab_01, cd_01);
+    let abcd_1 = unpackhi_epi64(ab_01, cd_01);
+    let abcd_2 = unpacklo_epi64(ab_23, cd_23);
+    let abcd_3 = unpackhi_epi64(ab_23, cd_23);
+
+    vecs[0] = abcd_0;
+    vecs[1] = abcd_1;
+    vecs[2] = abcd_2;
+    vecs[3] = abcd_3;
+}
+
+#[inline(always)]
+unsafe fn transpose_msg_vecs(inputs: &[*const u8; DEGREE], block_offset: usize) -> [v128; 16] {
+    let mut vecs = [
+        loadu(inputs[0].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 0 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 1 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 2 * 4 * DEGREE)),
+        loadu(inputs[0].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[1].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[2].add(block_offset + 3 * 4 * DEGREE)),
+        loadu(inputs[3].add(block_offset + 3 * 4 * DEGREE)),
+    ];
+    let squares = mut_array_refs!(&mut vecs, DEGREE, DEGREE, DEGREE, DEGREE);
+    transpose_vecs(squares.0);
+    transpose_vecs(squares.1);
+    transpose_vecs(squares.2);
+    transpose_vecs(squares.3);
+    vecs
+}
+
+#[inline(always)]
+fn load_counters(counter: u64, increment_counter: IncrementCounter) -> (v128, v128) {
+    let mask = if increment_counter.yes() { !0 } else { 0 };
+    (
+        set4(
+            counter_low(counter + (mask & 0)),
+            counter_low(counter + (mask & 1)),
+            counter_low(counter + (mask & 2)),
+            counter_low(counter + (mask & 3)),
+        ),
+        set4(
+            counter_high(counter + (mask & 0)),
+            counter_high(counter + (mask & 1)),
+            counter_high(counter + (mask & 2)),
+            counter_high(counter + (mask & 3)),
+        ),
+    )
+}
+
+#[target_feature(enable = "simd128")]
+pub unsafe fn hash4(
+    inputs: &[*const u8; DEGREE],
+    blocks: usize,
+    key: &CVWords,
+    counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut [u8; DEGREE * OUT_LEN],
+) {
+    let mut h_vecs = [
+        set1(key[0]),
+        set1(key[1]),
+        set1(key[2]),
+        set1(key[3]),
+        set1(key[4]),
+        set1(key[5]),
+        set1(key[6]),
+        set1(key[7]),
+    ];
+    let (counter_low_vec, counter_high_vec) = load_counters(counter, increment_counter);
+    let mut block_flags = flags | flags_start;
+
+    for block in 0..blocks {
+        if block + 1 == blocks {
+            block_flags |= flags_end;
+        }
+        let block_len_vec = set1(BLOCK_LEN as u32); // full blocks only
+        let block_flags_vec = set1(block_flags as u32);
+        let msg_vecs = transpose_msg_vecs(inputs, block * BLOCK_LEN);
+
+        // The transposed compression function. Note that inlining this
+        // manually here improves compile times by a lot, compared to factoring
+        // it out into its own function and making it #[inline(always)]. Just
+        // guessing, it might have something to do with loop unrolling.
+        let mut v = [
+            h_vecs[0],
+            h_vecs[1],
+            h_vecs[2],
+            h_vecs[3],
+            h_vecs[4],
+            h_vecs[5],
+            h_vecs[6],
+            h_vecs[7],
+            set1(IV[0]),
+            set1(IV[1]),
+            set1(IV[2]),
+            set1(IV[3]),
+            counter_low_vec,
+            counter_high_vec,
+            block_len_vec,
+            block_flags_vec,
+        ];
+        round(&mut v, &msg_vecs, 0);
+        round(&mut v, &msg_vecs, 1);
+        round(&mut v, &msg_vecs, 2);
+        round(&mut v, &msg_vecs, 3);
+        round(&mut v, &msg_vecs, 4);
+        round(&mut v, &msg_vecs, 5);
+        round(&mut v, &msg_vecs, 6);
+        h_vecs[0] = xor(v[0], v[8]);
+        h_vecs[1] = xor(v[1], v[9]);
+        h_vecs[2] = xor(v[2], v[10]);
+        h_vecs[3] = xor(v[3], v[11]);
+        h_vecs[4] = xor(v[4], v[12]);
+        h_vecs[5] = xor(v[5], v[13]);
+        h_vecs[6] = xor(v[6], v[14]);
+        h_vecs[7] = xor(v[7], v[15]);
+
+        block_flags = flags;
+    }
+
+    let squares = mut_array_refs!(&mut h_vecs, DEGREE, DEGREE);
+    transpose_vecs(squares.0);
+    transpose_vecs(squares.1);
+    // The first four vecs now contain the first half of each output, and the
+    // second four vecs contain the second half of each output.
+    storeu(h_vecs[0], out.as_mut_ptr().add(0 * 4 * DEGREE));
+    storeu(h_vecs[4], out.as_mut_ptr().add(1 * 4 * DEGREE));
+    storeu(h_vecs[1], out.as_mut_ptr().add(2 * 4 * DEGREE));
+    storeu(h_vecs[5], out.as_mut_ptr().add(3 * 4 * DEGREE));
+    storeu(h_vecs[2], out.as_mut_ptr().add(4 * 4 * DEGREE));
+    storeu(h_vecs[6], out.as_mut_ptr().add(5 * 4 * DEGREE));
+    storeu(h_vecs[3], out.as_mut_ptr().add(6 * 4 * DEGREE));
+    storeu(h_vecs[7], out.as_mut_ptr().add(7 * 4 * DEGREE));
+}
+
+#[target_feature(enable = "simd128")]
+unsafe fn hash1<const N: usize>(
+    input: &[u8; N],
+    key: &CVWords,
+    counter: u64,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    out: &mut CVBytes,
+) {
+    debug_assert_eq!(N % BLOCK_LEN, 0, "uneven blocks");
+    let mut cv = *key;
+    let mut block_flags = flags | flags_start;
+    let mut slice = &input[..];
+    while slice.len() >= BLOCK_LEN {
+        if slice.len() == BLOCK_LEN {
+            block_flags |= flags_end;
+        }
+        compress_in_place(
+            &mut cv,
+            array_ref!(slice, 0, BLOCK_LEN),
+            BLOCK_LEN as u8,
+            counter,
+            block_flags,
+        );
+        block_flags = flags;
+        slice = &slice[BLOCK_LEN..];
+    }
+    *out = core::mem::transmute(cv);
+}
+
+#[target_feature(enable = "simd128")]
+pub unsafe fn hash_many<const N: usize>(
+    mut inputs: &[&[u8; N]],
+    key: &CVWords,
+    mut counter: u64,
+    increment_counter: IncrementCounter,
+    flags: u8,
+    flags_start: u8,
+    flags_end: u8,
+    mut out: &mut [u8],
+) {
+    debug_assert!(out.len() >= inputs.len() * OUT_LEN, "out too short");
+    while inputs.len() >= DEGREE && out.len() >= DEGREE * OUT_LEN {
+        // Safe because the layout of arrays is guaranteed, and because the
+        // `blocks` count is determined statically from the argument type.
+        let input_ptrs: &[*const u8; DEGREE] = &*(inputs.as_ptr() as *const [*const u8; DEGREE]);
+        let blocks = N / BLOCK_LEN;
+        hash4(
+            input_ptrs,
+            blocks,
+            key,
+            counter,
+            increment_counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(out, 0, DEGREE * OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += DEGREE as u64;
+        }
+        inputs = &inputs[DEGREE..];
+        out = &mut out[DEGREE * OUT_LEN..];
+    }
+    for (&input, output) in inputs.iter().zip(out.chunks_exact_mut(OUT_LEN)) {
+        hash1(
+            input,
+            key,
+            counter,
+            flags,
+            flags_start,
+            flags_end,
+            array_mut_ref!(output, 0, OUT_LEN),
+        );
+        if increment_counter.yes() {
+            counter += 1;
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use super::*;
+
+    #[test]
+    fn test_transpose() {
+        #[target_feature(enable = "simd128")]
+        fn transpose_wrapper(vecs: &mut [v128; DEGREE]) {
+            transpose_vecs(vecs);
+        }
+
+        let mut matrix = [[0 as u32; DEGREE]; DEGREE];
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                matrix[i][j] = (i * DEGREE + j) as u32;
+            }
+        }
+
+        unsafe {
+            let mut vecs: [v128; DEGREE] = core::mem::transmute(matrix);
+            transpose_wrapper(&mut vecs);
+            matrix = core::mem::transmute(vecs);
+        }
+
+        for i in 0..DEGREE {
+            for j in 0..DEGREE {
+                // Reversed indexes from above.
+                assert_eq!(matrix[j][i], (i * DEGREE + j) as u32);
+            }
+        }
+    }
+
+    #[test]
+    fn test_compress() {
+        crate::test::test_compress_fn(compress_in_place, compress_xof);
+    }
+
+    #[test]
+    fn test_hash_many() {
+        crate::test::test_hash_many_fn(hash_many, hash_many);
+    }
+}
diff --git a/thirdparty/blake3/test_vectors/Cargo.toml b/thirdparty/blake3/test_vectors/Cargo.toml
new file mode 100644
index 000000000..46ba58cf9
--- /dev/null
+++ b/thirdparty/blake3/test_vectors/Cargo.toml
@@ -0,0 +1,19 @@
+[package]
+name = "test_vectors"
+version = "0.0.0"
+edition = "2021"
+
+[features]
+neon = ["blake3/neon"]
+prefer_intrinsics = ["blake3/prefer_intrinsics"]
+pure = ["blake3/pure"]
+wasm32_simd = ["blake3/wasm32_simd"]
+
+[dependencies]
+# If you ever change these path dependencies, you'll probably need to update
+# cross_test.sh, or CI will break. I'm sorry >.<
+blake3 = { path = "../" }
+hex = "0.4.0"
+reference_impl = { path = "../reference_impl" }
+serde = { version = "1.0", features = ["derive"] }
+serde_json = "1.0"
diff --git a/thirdparty/blake3/test_vectors/cross_test.sh b/thirdparty/blake3/test_vectors/cross_test.sh
new file mode 100755
index 000000000..c4d280c9d
--- /dev/null
+++ b/thirdparty/blake3/test_vectors/cross_test.sh
@@ -0,0 +1,25 @@
+#! /usr/bin/env bash
+
+# This hacky script works around the fact that `cross test` does not support
+# path dependencies. (It uses a docker shared folder to let the guest access
+# project files, so parent directories aren't available.) Solve this problem by
+# copying the entire project to a temp dir and rearranging paths to put
+# "blake3" and "reference_impl" underneath "test_vectors", so that everything
+# is accessible. Hopefully this will just run on CI forever and no one will
+# ever read this and discover my deep shame.
+
+set -e -u -o pipefail
+
+project_root="$(realpath "$(dirname "$BASH_SOURCE")/..")"
+tmpdir="$(mktemp -d)"
+echo "Running cross tests in $tmpdir"
+cd "$tmpdir"
+git clone "$project_root" blake3
+mv blake3/test_vectors .
+mv blake3/reference_impl test_vectors
+mv blake3 test_vectors
+cd test_vectors
+sed -i 's|blake3 = { path = "../" }|blake3 = { path = "./blake3" }|' Cargo.toml
+sed -i 's|reference_impl = { path = "../reference_impl" }|reference_impl = { path = "reference_impl" }|' Cargo.toml
+
+cross test "$@"
diff --git a/thirdparty/blake3/test_vectors/src/lib.rs b/thirdparty/blake3/test_vectors/src/lib.rs
new file mode 100644
index 000000000..8df28cd65
--- /dev/null
+++ b/thirdparty/blake3/test_vectors/src/lib.rs
@@ -0,0 +1,350 @@
+use blake3::{BLOCK_LEN, CHUNK_LEN};
+use serde::{Deserialize, Serialize};
+
+// Reading files at runtime requires special configuration under WASM/WASI, so including this at
+// compile time is simpler.
+const TEST_VECTORS_JSON: &str = include_str!("../test_vectors.json");
+
+// A non-multiple of 4 is important, since one possible bug is to fail to emit
+// partial words.
+pub const OUTPUT_LEN: usize = 2 * BLOCK_LEN + 3;
+
+pub const TEST_CASES: &[usize] = &[
+    0,
+    1,
+    2,
+    3,
+    4,
+    5,
+    6,
+    7,
+    8,
+    BLOCK_LEN - 1,
+    BLOCK_LEN,
+    BLOCK_LEN + 1,
+    2 * BLOCK_LEN - 1,
+    2 * BLOCK_LEN,
+    2 * BLOCK_LEN + 1,
+    CHUNK_LEN - 1,
+    CHUNK_LEN,
+    CHUNK_LEN + 1,
+    2 * CHUNK_LEN,
+    2 * CHUNK_LEN + 1,
+    3 * CHUNK_LEN,
+    3 * CHUNK_LEN + 1,
+    4 * CHUNK_LEN,
+    4 * CHUNK_LEN + 1,
+    5 * CHUNK_LEN,
+    5 * CHUNK_LEN + 1,
+    6 * CHUNK_LEN,
+    6 * CHUNK_LEN + 1,
+    7 * CHUNK_LEN,
+    7 * CHUNK_LEN + 1,
+    8 * CHUNK_LEN,
+    8 * CHUNK_LEN + 1,
+    16 * CHUNK_LEN,  // AVX512's bandwidth
+    31 * CHUNK_LEN,  // 16 + 8 + 4 + 2 + 1
+    100 * CHUNK_LEN, // subtrees larger than MAX_SIMD_DEGREE chunks
+];
+
+pub const TEST_KEY: &[u8; blake3::KEY_LEN] = b"whats the Elvish word for friend";
+pub const TEST_CONTEXT: &str = "BLAKE3 2019-12-27 16:29:52 test vectors context";
+
+const COMMENT: &str = r#"
+Each test is an input length and three outputs, one for each of the hash,
+keyed_hash, and derive_key modes. The input in each case is filled with a
+repeating sequence of 251 bytes: 0, 1, 2, ..., 249, 250, 0, 1, ..., and so on.
+The key used with keyed_hash is the 32-byte ASCII string "whats the Elvish word
+for friend", also given in the `key` field below. The context string used with
+derive_key is the ASCII string "BLAKE3 2019-12-27 16:29:52 test vectors
+context", also given in the `context_string` field below. Outputs are encoded
+as hexadecimal. Each case is an extended output, and implementations should
+also check that the first 32 bytes match their default-length output.
+"#;
+
+// Paint the input with a repeating byte pattern. We use a cycle length of 251,
+// because that's the largest prime number less than 256. This makes it
+// unlikely to swapping any two adjacent input blocks or chunks will give the
+// same answer.
+pub fn paint_test_input(buf: &mut [u8]) {
+    for (i, b) in buf.iter_mut().enumerate() {
+        *b = (i % 251) as u8;
+    }
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+pub struct Cases {
+    pub _comment: String,
+    pub key: String,
+    pub context_string: String,
+    pub cases: Vec<Case>,
+}
+
+#[derive(Debug, Serialize, Deserialize)]
+pub struct Case {
+    pub input_len: usize,
+    pub hash: String,
+    pub keyed_hash: String,
+    pub derive_key: String,
+}
+
+pub fn generate_json() -> String {
+    let mut cases = Vec::new();
+    for &input_len in TEST_CASES {
+        let mut input = vec![0; input_len];
+        paint_test_input(&mut input);
+
+        let mut hash_out = [0; OUTPUT_LEN];
+        blake3::Hasher::new()
+            .update(&input)
+            .finalize_xof()
+            .fill(&mut hash_out);
+
+        let mut keyed_hash_out = [0; OUTPUT_LEN];
+        blake3::Hasher::new_keyed(TEST_KEY)
+            .update(&input)
+            .finalize_xof()
+            .fill(&mut keyed_hash_out);
+
+        let mut derive_key_out = [0; OUTPUT_LEN];
+        blake3::Hasher::new_derive_key(TEST_CONTEXT)
+            .update(&input)
+            .finalize_xof()
+            .fill(&mut derive_key_out);
+
+        cases.push(Case {
+            input_len,
+            hash: hex::encode(&hash_out[..]),
+            keyed_hash: hex::encode(&keyed_hash_out[..]),
+            derive_key: hex::encode(&derive_key_out[..]),
+        });
+    }
+
+    let mut json = serde_json::to_string_pretty(&Cases {
+        _comment: COMMENT.trim().replace("\n", " "),
+        key: std::str::from_utf8(TEST_KEY).unwrap().to_string(),
+        context_string: TEST_CONTEXT.to_string(),
+        cases,
+    })
+    .unwrap();
+
+    // Add a trailing newline.
+    json.push('\n');
+    json
+}
+
+pub fn parse_test_cases() -> Cases {
+    serde_json::from_str(TEST_VECTORS_JSON).expect("failed to parse test_vectors.json")
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn test_reference_impl_all_at_once(
+        key: &[u8; blake3::KEY_LEN],
+        input: &[u8],
+        expected_hash: &[u8],
+        expected_keyed_hash: &[u8],
+        expected_derive_key: &[u8],
+    ) {
+        let mut out = vec![0; expected_hash.len()];
+        let mut hasher = reference_impl::Hasher::new();
+        hasher.update(input);
+        hasher.finalize(&mut out);
+        assert_eq!(expected_hash, &out[..]);
+
+        let mut out = vec![0; expected_keyed_hash.len()];
+        let mut hasher = reference_impl::Hasher::new_keyed(key);
+        hasher.update(input);
+        hasher.finalize(&mut out);
+        assert_eq!(expected_keyed_hash, &out[..]);
+
+        let mut out = vec![0; expected_derive_key.len()];
+        let mut hasher = reference_impl::Hasher::new_derive_key(TEST_CONTEXT);
+        hasher.update(input);
+        hasher.finalize(&mut out);
+        assert_eq!(expected_derive_key, &out[..]);
+    }
+
+    fn test_reference_impl_one_at_a_time(
+        key: &[u8; blake3::KEY_LEN],
+        input: &[u8],
+        expected_hash: &[u8],
+        expected_keyed_hash: &[u8],
+        expected_derive_key: &[u8],
+    ) {
+        let mut out = vec![0; expected_hash.len()];
+        let mut hasher = reference_impl::Hasher::new();
+        for &b in input {
+            hasher.update(&[b]);
+        }
+        hasher.finalize(&mut out);
+        assert_eq!(expected_hash, &out[..]);
+
+        let mut out = vec![0; expected_keyed_hash.len()];
+        let mut hasher = reference_impl::Hasher::new_keyed(key);
+        for &b in input {
+            hasher.update(&[b]);
+        }
+        hasher.finalize(&mut out);
+        assert_eq!(expected_keyed_hash, &out[..]);
+
+        let mut out = vec![0; expected_derive_key.len()];
+        let mut hasher = reference_impl::Hasher::new_derive_key(TEST_CONTEXT);
+        for &b in input {
+            hasher.update(&[b]);
+        }
+        hasher.finalize(&mut out);
+        assert_eq!(expected_derive_key, &out[..]);
+    }
+
+    fn test_incremental_all_at_once(
+        key: &[u8; blake3::KEY_LEN],
+        input: &[u8],
+        expected_hash: &[u8],
+        expected_keyed_hash: &[u8],
+        expected_derive_key: &[u8],
+    ) {
+        let mut out = vec![0; expected_hash.len()];
+        let mut hasher = blake3::Hasher::new();
+        hasher.update(input);
+        hasher.finalize_xof().fill(&mut out);
+        assert_eq!(expected_hash, &out[..]);
+        assert_eq!(&expected_hash[..32], hasher.finalize().as_bytes());
+
+        let mut out = vec![0; expected_keyed_hash.len()];
+        let mut hasher = blake3::Hasher::new_keyed(key);
+        hasher.update(input);
+        hasher.finalize_xof().fill(&mut out);
+        assert_eq!(expected_keyed_hash, &out[..]);
+        assert_eq!(&expected_keyed_hash[..32], hasher.finalize().as_bytes());
+
+        let mut out = vec![0; expected_derive_key.len()];
+        let mut hasher = blake3::Hasher::new_derive_key(TEST_CONTEXT);
+        hasher.update(input);
+        hasher.finalize_xof().fill(&mut out);
+        assert_eq!(expected_derive_key, &out[..]);
+        assert_eq!(&expected_derive_key[..32], hasher.finalize().as_bytes());
+    }
+
+    fn test_incremental_one_at_a_time(
+        key: &[u8; blake3::KEY_LEN],
+        input: &[u8],
+        expected_hash: &[u8],
+        expected_keyed_hash: &[u8],
+        expected_derive_key: &[u8],
+    ) {
+        let mut out = vec![0; expected_hash.len()];
+        let mut hasher = blake3::Hasher::new();
+        for i in 0..input.len() {
+            hasher.update(&[input[i]]);
+            assert_eq!(i as u64 + 1, hasher.count());
+        }
+        hasher.finalize_xof().fill(&mut out);
+        assert_eq!(expected_hash, &out[..]);
+        assert_eq!(&expected_hash[..32], hasher.finalize().as_bytes());
+
+        let mut out = vec![0; expected_keyed_hash.len()];
+        let mut hasher = blake3::Hasher::new_keyed(key);
+        for i in 0..input.len() {
+            hasher.update(&[input[i]]);
+            assert_eq!(i as u64 + 1, hasher.count());
+        }
+        hasher.finalize_xof().fill(&mut out);
+        assert_eq!(expected_keyed_hash, &out[..]);
+        assert_eq!(&expected_keyed_hash[..32], hasher.finalize().as_bytes());
+
+        let mut out = vec![0; expected_derive_key.len()];
+        let mut hasher = blake3::Hasher::new_derive_key(TEST_CONTEXT);
+        for i in 0..input.len() {
+            hasher.update(&[input[i]]);
+            assert_eq!(i as u64 + 1, hasher.count());
+        }
+        hasher.finalize_xof().fill(&mut out);
+        assert_eq!(expected_derive_key, &out[..]);
+        assert_eq!(&expected_derive_key[..32], hasher.finalize().as_bytes());
+    }
+
+    fn test_recursive(
+        key: &[u8; blake3::KEY_LEN],
+        input: &[u8],
+        expected_hash: &[u8],
+        expected_keyed_hash: &[u8],
+        expected_derive_key: &[u8],
+    ) {
+        assert_eq!(&expected_hash[..32], blake3::hash(input).as_bytes());
+        assert_eq!(
+            &expected_keyed_hash[..32],
+            blake3::keyed_hash(key, input).as_bytes(),
+        );
+        assert_eq!(
+            expected_derive_key[..32],
+            blake3::derive_key(TEST_CONTEXT, input)
+        );
+    }
+
+    #[test]
+    fn run_test_vectors() {
+        let cases = parse_test_cases();
+        let key: &[u8; blake3::KEY_LEN] = cases.key.as_bytes().try_into().unwrap();
+        for case in &cases.cases {
+            dbg!(case.input_len);
+            let mut input = vec![0; case.input_len];
+            paint_test_input(&mut input);
+            let expected_hash = hex::decode(&case.hash).unwrap();
+            let expected_keyed_hash = hex::decode(&case.keyed_hash).unwrap();
+            let expected_derive_key = hex::decode(&case.derive_key).unwrap();
+
+            test_reference_impl_all_at_once(
+                key,
+                &input,
+                &expected_hash,
+                &expected_keyed_hash,
+                &expected_derive_key,
+            );
+
+            test_reference_impl_one_at_a_time(
+                key,
+                &input,
+                &expected_hash,
+                &expected_keyed_hash,
+                &expected_derive_key,
+            );
+
+            test_incremental_all_at_once(
+                key,
+                &input,
+                &expected_hash,
+                &expected_keyed_hash,
+                &expected_derive_key,
+            );
+
+            test_incremental_one_at_a_time(
+                key,
+                &input,
+                &expected_hash,
+                &expected_keyed_hash,
+                &expected_derive_key,
+            );
+
+            test_recursive(
+                key,
+                &input,
+                &expected_hash,
+                &expected_keyed_hash,
+                &expected_derive_key,
+            );
+        }
+    }
+
+    #[test]
+    fn test_checked_in_vectors_up_to_date() {
+        // Replace Windows newlines, in case Git is configured to alter
+        // newlines when files are checked out.
+        let json = TEST_VECTORS_JSON.replace("\r\n", "\n");
+        if generate_json() != json {
+            panic!("Checked-in test_vectors.json is not up to date. Regenerate with `cargo run --bin generate > ./test_vectors.json`.");
+        }
+    }
+}
diff --git a/thirdparty/blake3/test_vectors/test_vectors.json b/thirdparty/blake3/test_vectors/test_vectors.json
new file mode 100644
index 000000000..f6da91792
--- /dev/null
+++ b/thirdparty/blake3/test_vectors/test_vectors.json
@@ -0,0 +1,217 @@
+{
+  "_comment": "Each test is an input length and three outputs, one for each of the hash, keyed_hash, and derive_key modes. The input in each case is filled with a repeating sequence of 251 bytes: 0, 1, 2, ..., 249, 250, 0, 1, ..., and so on. The key used with keyed_hash is the 32-byte ASCII string \"whats the Elvish word for friend\", also given in the `key` field below. The context string used with derive_key is the ASCII string \"BLAKE3 2019-12-27 16:29:52 test vectors context\", also given in the `context_string` field below. Outputs are encoded as hexadecimal. Each case is an extended output, and implementations should also check that the first 32 bytes match their default-length output.",
+  "key": "whats the Elvish word for friend",
+  "context_string": "BLAKE3 2019-12-27 16:29:52 test vectors context",
+  "cases": [
+    {
+      "input_len": 0,
+      "hash": "af1349b9f5f9a1a6a0404dea36dcc9499bcb25c9adc112b7cc9a93cae41f3262e00f03e7b69af26b7faaf09fcd333050338ddfe085b8cc869ca98b206c08243a26f5487789e8f660afe6c99ef9e0c52b92e7393024a80459cf91f476f9ffdbda7001c22e159b402631f277ca96f2defdf1078282314e763699a31c5363165421cce14d",
+      "keyed_hash": "92b2b75604ed3c761f9d6f62392c8a9227ad0ea3f09573e783f1498a4ed60d26b18171a2f22a4b94822c701f107153dba24918c4bae4d2945c20ece13387627d3b73cbf97b797d5e59948c7ef788f54372df45e45e4293c7dc18c1d41144a9758be58960856be1eabbe22c2653190de560ca3b2ac4aa692a9210694254c371e851bc8f",
+      "derive_key": "2cc39783c223154fea8dfb7c1b1660f2ac2dcbd1c1de8277b0b0dd39b7e50d7d905630c8be290dfcf3e6842f13bddd573c098c3f17361f1f206b8cad9d088aa4a3f746752c6b0ce6a83b0da81d59649257cdf8eb3e9f7d4998e41021fac119deefb896224ac99f860011f73609e6e0e4540f93b273e56547dfd3aa1a035ba6689d89a0"
+    },
+    {
+      "input_len": 1,
+      "hash": "2d3adedff11b61f14c886e35afa036736dcd87a74d27b5c1510225d0f592e213c3a6cb8bf623e20cdb535f8d1a5ffb86342d9c0b64aca3bce1d31f60adfa137b358ad4d79f97b47c3d5e79f179df87a3b9776ef8325f8329886ba42f07fb138bb502f4081cbcec3195c5871e6c23e2cc97d3c69a613eba131e5f1351f3f1da786545e5",
+      "keyed_hash": "6d7878dfff2f485635d39013278ae14f1454b8c0a3a2d34bc1ab38228a80c95b6568c0490609413006fbd428eb3fd14e7756d90f73a4725fad147f7bf70fd61c4e0cf7074885e92b0e3f125978b4154986d4fb202a3f331a3fb6cf349a3a70e49990f98fe4289761c8602c4e6ab1138d31d3b62218078b2f3ba9a88e1d08d0dd4cea11",
+      "derive_key": "b3e2e340a117a499c6cf2398a19ee0d29cca2bb7404c73063382693bf66cb06c5827b91bf889b6b97c5477f535361caefca0b5d8c4746441c57617111933158950670f9aa8a05d791daae10ac683cbef8faf897c84e6114a59d2173c3f417023a35d6983f2c7dfa57e7fc559ad751dbfb9ffab39c2ef8c4aafebc9ae973a64f0c76551"
+    },
+    {
+      "input_len": 2,
+      "hash": "7b7015bb92cf0b318037702a6cdd81dee41224f734684c2c122cd6359cb1ee63d8386b22e2ddc05836b7c1bb693d92af006deb5ffbc4c70fb44d0195d0c6f252faac61659ef86523aa16517f87cb5f1340e723756ab65efb2f91964e14391de2a432263a6faf1d146937b35a33621c12d00be8223a7f1919cec0acd12097ff3ab00ab1",
+      "keyed_hash": "5392ddae0e0a69d5f40160462cbd9bd889375082ff224ac9c758802b7a6fd20a9ffbf7efd13e989a6c246f96d3a96b9d279f2c4e63fb0bdff633957acf50ee1a5f658be144bab0f6f16500dee4aa5967fc2c586d85a04caddec90fffb7633f46a60786024353b9e5cebe277fcd9514217fee2267dcda8f7b31697b7c54fab6a939bf8f",
+      "derive_key": "1f166565a7df0098ee65922d7fea425fb18b9943f19d6161e2d17939356168e6daa59cae19892b2d54f6fc9f475d26031fd1c22ae0a3e8ef7bdb23f452a15e0027629d2e867b1bb1e6ab21c71297377750826c404dfccc2406bd57a83775f89e0b075e59a7732326715ef912078e213944f490ad68037557518b79c0086de6d6f6cdd2"
+    },
+    {
+      "input_len": 3,
+      "hash": "e1be4d7a8ab5560aa4199eea339849ba8e293d55ca0a81006726d184519e647f5b49b82f805a538c68915c1ae8035c900fd1d4b13902920fd05e1450822f36de9454b7e9996de4900c8e723512883f93f4345f8a58bfe64ee38d3ad71ab027765d25cdd0e448328a8e7a683b9a6af8b0af94fa09010d9186890b096a08471e4230a134",
+      "keyed_hash": "39e67b76b5a007d4921969779fe666da67b5213b096084ab674742f0d5ec62b9b9142d0fab08e1b161efdbb28d18afc64d8f72160c958e53a950cdecf91c1a1bbab1a9c0f01def762a77e2e8545d4dec241e98a89b6db2e9a5b070fc110caae2622690bd7b76c02ab60750a3ea75426a6bb8803c370ffe465f07fb57def95df772c39f",
+      "derive_key": "440aba35cb006b61fc17c0529255de438efc06a8c9ebf3f2ddac3b5a86705797f27e2e914574f4d87ec04c379e12789eccbfbc15892626042707802dbe4e97c3ff59dca80c1e54246b6d055154f7348a39b7d098b2b4824ebe90e104e763b2a447512132cede16243484a55a4e40a85790038bb0dcf762e8c053cabae41bbe22a5bff7"
+    },
+    {
+      "input_len": 4,
+      "hash": "f30f5ab28fe047904037f77b6da4fea1e27241c5d132638d8bedce9d40494f328f603ba4564453e06cdcee6cbe728a4519bbe6f0d41e8a14b5b225174a566dbfa61b56afb1e452dc08c804f8c3143c9e2cc4a31bb738bf8c1917b55830c6e65797211701dc0b98daa1faeaa6ee9e56ab606ce03a1a881e8f14e87a4acf4646272cfd12",
+      "keyed_hash": "7671dde590c95d5ac9616651ff5aa0a27bee5913a348e053b8aa9108917fe070116c0acff3f0d1fa97ab38d813fd46506089118147d83393019b068a55d646251ecf81105f798d76a10ae413f3d925787d6216a7eb444e510fd56916f1d753a5544ecf0072134a146b2615b42f50c179f56b8fae0788008e3e27c67482349e249cb86a",
+      "derive_key": "f46085c8190d69022369ce1a18880e9b369c135eb93f3c63550d3e7630e91060fbd7d8f4258bec9da4e05044f88b91944f7cab317a2f0c18279629a3867fad0662c9ad4d42c6f27e5b124da17c8c4f3a94a025ba5d1b623686c6099d202a7317a82e3d95dae46a87de0555d727a5df55de44dab799a20dffe239594d6e99ed17950910"
+    },
+    {
+      "input_len": 5,
+      "hash": "b40b44dfd97e7a84a996a91af8b85188c66c126940ba7aad2e7ae6b385402aa2ebcfdac6c5d32c31209e1f81a454751280db64942ce395104e1e4eaca62607de1c2ca748251754ea5bbe8c20150e7f47efd57012c63b3c6a6632dc1c7cd15f3e1c999904037d60fac2eb9397f2adbe458d7f264e64f1e73aa927b30988e2aed2f03620",
+      "keyed_hash": "73ac69eecf286894d8102018a6fc729f4b1f4247d3703f69bdc6a5fe3e0c84616ab199d1f2f3e53bffb17f0a2209fe8b4f7d4c7bae59c2bc7d01f1ff94c67588cc6b38fa6024886f2c078bfe09b5d9e6584cd6c521c3bb52f4de7687b37117a2dbbec0d59e92fa9a8cc3240d4432f91757aabcae03e87431dac003e7d73574bfdd8218",
+      "derive_key": "1f24eda69dbcb752847ec3ebb5dd42836d86e58500c7c98d906ecd82ed9ae47f6f48a3f67e4e43329c9a89b1ca526b9b35cbf7d25c1e353baffb590fd79be58ddb6c711f1a6b60e98620b851c688670412fcb0435657ba6b638d21f0f2a04f2f6b0bd8834837b10e438d5f4c7c2c71299cf7586ea9144ed09253d51f8f54dd6bff719d"
+    },
+    {
+      "input_len": 6,
+      "hash": "06c4e8ffb6872fad96f9aaca5eee1553eb62aed0ad7198cef42e87f6a616c844611a30c4e4f37fe2fe23c0883cde5cf7059d88b657c7ed2087e3d210925ede716435d6d5d82597a1e52b9553919e804f5656278bd739880692c94bff2824d8e0b48cac1d24682699e4883389dc4f2faa2eb3b4db6e39debd5061ff3609916f3e07529a",
+      "keyed_hash": "82d3199d0013035682cc7f2a399d4c212544376a839aa863a0f4c91220ca7a6dc2ffb3aa05f2631f0fa9ac19b6e97eb7e6669e5ec254799350c8b8d189e8807800842a5383c4d907c932f34490aaf00064de8cdb157357bde37c1504d2960034930887603abc5ccb9f5247f79224baff6120a3c622a46d7b1bcaee02c5025460941256",
+      "derive_key": "be96b30b37919fe4379dfbe752ae77b4f7e2ab92f7ff27435f76f2f065f6a5f435ae01a1d14bd5a6b3b69d8cbd35f0b01ef2173ff6f9b640ca0bd4748efa398bf9a9c0acd6a66d9332fdc9b47ffe28ba7ab6090c26747b85f4fab22f936b71eb3f64613d8bd9dfabe9bb68da19de78321b481e5297df9e40ec8a3d662f3e1479c65de0"
+    },
+    {
+      "input_len": 7,
+      "hash": "3f8770f387faad08faa9d8414e9f449ac68e6ff0417f673f602a646a891419fe66036ef6e6d1a8f54baa9fed1fc11c77cfb9cff65bae915045027046ebe0c01bf5a941f3bb0f73791d3fc0b84370f9f30af0cd5b0fc334dd61f70feb60dad785f070fef1f343ed933b49a5ca0d16a503f599a365a4296739248b28d1a20b0e2cc8975c",
+      "keyed_hash": "af0a7ec382aedc0cfd626e49e7628bc7a353a4cb108855541a5651bf64fbb28a7c5035ba0f48a9c73dabb2be0533d02e8fd5d0d5639a18b2803ba6bf527e1d145d5fd6406c437b79bcaad6c7bdf1cf4bd56a893c3eb9510335a7a798548c6753f74617bede88bef924ba4b334f8852476d90b26c5dc4c3668a2519266a562c6c8034a6",
+      "derive_key": "dc3b6485f9d94935329442916b0d059685ba815a1fa2a14107217453a7fc9f0e66266db2ea7c96843f9d8208e600a73f7f45b2f55b9e6d6a7ccf05daae63a3fdd10b25ac0bd2e224ce8291f88c05976d575df998477db86fb2cfbbf91725d62cb57acfeb3c2d973b89b503c2b60dde85a7802b69dc1ac2007d5623cbea8cbfb6b181f5"
+    },
+    {
+      "input_len": 8,
+      "hash": "2351207d04fc16ade43ccab08600939c7c1fa70a5c0aaca76063d04c3228eaeb725d6d46ceed8f785ab9f2f9b06acfe398c6699c6129da084cb531177445a682894f9685eaf836999221d17c9a64a3a057000524cd2823986db378b074290a1a9b93a22e135ed2c14c7e20c6d045cd00b903400374126676ea78874d79f2dd7883cf5c",
+      "keyed_hash": "be2f5495c61cba1bb348a34948c004045e3bd4dae8f0fe82bf44d0da245a060048eb5e68ce6dea1eb0229e144f578b3aa7e9f4f85febd135df8525e6fe40c6f0340d13dd09b255ccd5112a94238f2be3c0b5b7ecde06580426a93e0708555a265305abf86d874e34b4995b788e37a823491f25127a502fe0704baa6bfdf04e76c13276",
+      "derive_key": "2b166978cef14d9d438046c720519d8b1cad707e199746f1562d0c87fbd32940f0e2545a96693a66654225ebbaac76d093bfa9cd8f525a53acb92a861a98c42e7d1c4ae82e68ab691d510012edd2a728f98cd4794ef757e94d6546961b4f280a51aac339cc95b64a92b83cc3f26d8af8dfb4c091c240acdb4d47728d23e7148720ef04"
+    },
+    {
+      "input_len": 63,
+      "hash": "e9bc37a594daad83be9470df7f7b3798297c3d834ce80ba85d6e207627b7db7b1197012b1e7d9af4d7cb7bdd1f3bb49a90a9b5dec3ea2bbc6eaebce77f4e470cbf4687093b5352f04e4a4570fba233164e6acc36900e35d185886a827f7ea9bdc1e5c3ce88b095a200e62c10c043b3e9bc6cb9b6ac4dfa51794b02ace9f98779040755",
+      "keyed_hash": "bb1eb5d4afa793c1ebdd9fb08def6c36d10096986ae0cfe148cd101170ce37aea05a63d74a840aecd514f654f080e51ac50fd617d22610d91780fe6b07a26b0847abb38291058c97474ef6ddd190d30fc318185c09ca1589d2024f0a6f16d45f11678377483fa5c005b2a107cb9943e5da634e7046855eaa888663de55d6471371d55d",
+      "derive_key": "b6451e30b953c206e34644c6803724e9d2725e0893039cfc49584f991f451af3b89e8ff572d3da4f4022199b9563b9d70ebb616efff0763e9abec71b550f1371e233319c4c4e74da936ba8e5bbb29a598e007a0bbfa929c99738ca2cc098d59134d11ff300c39f82e2fce9f7f0fa266459503f64ab9913befc65fddc474f6dc1c67669"
+    },
+    {
+      "input_len": 64,
+      "hash": "4eed7141ea4a5cd4b788606bd23f46e212af9cacebacdc7d1f4c6dc7f2511b98fc9cc56cb831ffe33ea8e7e1d1df09b26efd2767670066aa82d023b1dfe8ab1b2b7fbb5b97592d46ffe3e05a6a9b592e2949c74160e4674301bc3f97e04903f8c6cf95b863174c33228924cdef7ae47559b10b294acd660666c4538833582b43f82d74",
+      "keyed_hash": "ba8ced36f327700d213f120b1a207a3b8c04330528586f414d09f2f7d9ccb7e68244c26010afc3f762615bbac552a1ca909e67c83e2fd5478cf46b9e811efccc93f77a21b17a152ebaca1695733fdb086e23cd0eb48c41c034d52523fc21236e5d8c9255306e48d52ba40b4dac24256460d56573d1312319afcf3ed39d72d0bfc69acb",
+      "derive_key": "a5c4a7053fa86b64746d4bb688d06ad1f02a18fce9afd3e818fefaa7126bf73e9b9493a9befebe0bf0c9509fb3105cfa0e262cde141aa8e3f2c2f77890bb64a4cca96922a21ead111f6338ad5244f2c15c44cb595443ac2ac294231e31be4a4307d0a91e874d36fc9852aeb1265c09b6e0cda7c37ef686fbbcab97e8ff66718be048bb"
+    },
+    {
+      "input_len": 65,
+      "hash": "de1e5fa0be70df6d2be8fffd0e99ceaa8eb6e8c93a63f2d8d1c30ecb6b263dee0e16e0a4749d6811dd1d6d1265c29729b1b75a9ac346cf93f0e1d7296dfcfd4313b3a227faaaaf7757cc95b4e87a49be3b8a270a12020233509b1c3632b3485eef309d0abc4a4a696c9decc6e90454b53b000f456a3f10079072baaf7a981653221f2c",
+      "keyed_hash": "c0a4edefa2d2accb9277c371ac12fcdbb52988a86edc54f0716e1591b4326e72d5e795f46a596b02d3d4bfb43abad1e5d19211152722ec1f20fef2cd413e3c22f2fc5da3d73041275be6ede3517b3b9f0fc67ade5956a672b8b75d96cb43294b9041497de92637ed3f2439225e683910cb3ae923374449ca788fb0f9bea92731bc26ad",
+      "derive_key": "51fd05c3c1cfbc8ed67d139ad76f5cf8236cd2acd26627a30c104dfd9d3ff8a82b02e8bd36d8498a75ad8c8e9b15eb386970283d6dd42c8ae7911cc592887fdbe26a0a5f0bf821cd92986c60b2502c9be3f98a9c133a7e8045ea867e0828c7252e739321f7c2d65daee4468eb4429efae469a42763f1f94977435d10dccae3e3dce88d"
+    },
+    {
+      "input_len": 127,
+      "hash": "d81293fda863f008c09e92fc382a81f5a0b4a1251cba1634016a0f86a6bd640de3137d477156d1fde56b0cf36f8ef18b44b2d79897bece12227539ac9ae0a5119da47644d934d26e74dc316145dcb8bb69ac3f2e05c242dd6ee06484fcb0e956dc44355b452c5e2bbb5e2b66e99f5dd443d0cbcaaafd4beebaed24ae2f8bb672bcef78",
+      "keyed_hash": "c64200ae7dfaf35577ac5a9521c47863fb71514a3bcad18819218b818de85818ee7a317aaccc1458f78d6f65f3427ec97d9c0adb0d6dacd4471374b621b7b5f35cd54663c64dbe0b9e2d95632f84c611313ea5bd90b71ce97b3cf645776f3adc11e27d135cbadb9875c2bf8d3ae6b02f8a0206aba0c35bfe42574011931c9a255ce6dc",
+      "derive_key": "c91c090ceee3a3ac81902da31838012625bbcd73fcb92e7d7e56f78deba4f0c3feeb3974306966ccb3e3c69c337ef8a45660ad02526306fd685c88542ad00f759af6dd1adc2e50c2b8aac9f0c5221ff481565cf6455b772515a69463223202e5c371743e35210bbbbabd89651684107fd9fe493c937be16e39cfa7084a36207c99bea3"
+    },
+    {
+      "input_len": 128,
+      "hash": "f17e570564b26578c33bb7f44643f539624b05df1a76c81f30acd548c44b45efa69faba091427f9c5c4caa873aa07828651f19c55bad85c47d1368b11c6fd99e47ecba5820a0325984d74fe3e4058494ca12e3f1d3293d0010a9722f7dee64f71246f75e9361f44cc8e214a100650db1313ff76a9f93ec6e84edb7add1cb4a95019b0c",
+      "keyed_hash": "b04fe15577457267ff3b6f3c947d93be581e7e3a4b018679125eaf86f6a628ecd86bbe0001f10bda47e6077b735016fca8119da11348d93ca302bbd125bde0db2b50edbe728a620bb9d3e6f706286aedea973425c0b9eedf8a38873544cf91badf49ad92a635a93f71ddfcee1eae536c25d1b270956be16588ef1cfef2f1d15f650bd5",
+      "derive_key": "81720f34452f58a0120a58b6b4608384b5c51d11f39ce97161a0c0e442ca022550e7cd651e312f0b4c6afb3c348ae5dd17d2b29fab3b894d9a0034c7b04fd9190cbd90043ff65d1657bbc05bfdecf2897dd894c7a1b54656d59a50b51190a9da44db426266ad6ce7c173a8c0bbe091b75e734b4dadb59b2861cd2518b4e7591e4b83c9"
+    },
+    {
+      "input_len": 129,
+      "hash": "683aaae9f3c5ba37eaaf072aed0f9e30bac0865137bae68b1fde4ca2aebdcb12f96ffa7b36dd78ba321be7e842d364a62a42e3746681c8bace18a4a8a79649285c7127bf8febf125be9de39586d251f0d41da20980b70d35e3dac0eee59e468a894fa7e6a07129aaad09855f6ad4801512a116ba2b7841e6cfc99ad77594a8f2d181a7",
+      "keyed_hash": "d4a64dae6cdccbac1e5287f54f17c5f985105457c1a2ec1878ebd4b57e20d38f1c9db018541eec241b748f87725665b7b1ace3e0065b29c3bcb232c90e37897fa5aaee7e1e8a2ecfcd9b51463e42238cfdd7fee1aecb3267fa7f2128079176132a412cd8aaf0791276f6b98ff67359bd8652ef3a203976d5ff1cd41885573487bcd683",
+      "derive_key": "938d2d4435be30eafdbb2b7031f7857c98b04881227391dc40db3c7b21f41fc18d72d0f9c1de5760e1941aebf3100b51d64644cb459eb5d20258e233892805eb98b07570ef2a1787cd48e117c8d6a63a68fd8fc8e59e79dbe63129e88352865721c8d5f0cf183f85e0609860472b0d6087cefdd186d984b21542c1c780684ed6832d8d"
+    },
+    {
+      "input_len": 1023,
+      "hash": "10108970eeda3eb932baac1428c7a2163b0e924c9a9e25b35bba72b28f70bd11a182d27a591b05592b15607500e1e8dd56bc6c7fc063715b7a1d737df5bad3339c56778957d870eb9717b57ea3d9fb68d1b55127bba6a906a4a24bbd5acb2d123a37b28f9e9a81bbaae360d58f85e5fc9d75f7c370a0cc09b6522d9c8d822f2f28f485",
+      "keyed_hash": "c951ecdf03288d0fcc96ee3413563d8a6d3589547f2c2fb36d9786470f1b9d6e890316d2e6d8b8c25b0a5b2180f94fb1a158ef508c3cde45e2966bd796a696d3e13efd86259d756387d9becf5c8bf1ce2192b87025152907b6d8cc33d17826d8b7b9bc97e38c3c85108ef09f013e01c229c20a83d9e8efac5b37470da28575fd755a10",
+      "derive_key": "74a16c1c3d44368a86e1ca6df64be6a2f64cce8f09220787450722d85725dea59c413264404661e9e4d955409dfe4ad3aa487871bcd454ed12abfe2c2b1eb7757588cf6cb18d2eccad49e018c0d0fec323bec82bf1644c6325717d13ea712e6840d3e6e730d35553f59eff5377a9c350bcc1556694b924b858f329c44ee64b884ef00d"
+    },
+    {
+      "input_len": 1024,
+      "hash": "42214739f095a406f3fc83deb889744ac00df831c10daa55189b5d121c855af71cf8107265ecdaf8505b95d8fcec83a98a6a96ea5109d2c179c47a387ffbb404756f6eeae7883b446b70ebb144527c2075ab8ab204c0086bb22b7c93d465efc57f8d917f0b385c6df265e77003b85102967486ed57db5c5ca170ba441427ed9afa684e",
+      "keyed_hash": "75c46f6f3d9eb4f55ecaaee480db732e6c2105546f1e675003687c31719c7ba4a78bc838c72852d4f49c864acb7adafe2478e824afe51c8919d06168414c265f298a8094b1ad813a9b8614acabac321f24ce61c5a5346eb519520d38ecc43e89b5000236df0597243e4d2493fd626730e2ba17ac4d8824d09d1a4a8f57b8227778e2de",
+      "derive_key": "7356cd7720d5b66b6d0697eb3177d9f8d73a4a5c5e968896eb6a6896843027066c23b601d3ddfb391e90d5c8eccdef4ae2a264bce9e612ba15e2bc9d654af1481b2e75dbabe615974f1070bba84d56853265a34330b4766f8e75edd1f4a1650476c10802f22b64bd3919d246ba20a17558bc51c199efdec67e80a227251808d8ce5bad"
+    },
+    {
+      "input_len": 1025,
+      "hash": "d00278ae47eb27b34faecf67b4fe263f82d5412916c1ffd97c8cb7fb814b8444f4c4a22b4b399155358a994e52bf255de60035742ec71bd08ac275a1b51cc6bfe332b0ef84b409108cda080e6269ed4b3e2c3f7d722aa4cdc98d16deb554e5627be8f955c98e1d5f9565a9194cad0c4285f93700062d9595adb992ae68ff12800ab67a",
+      "keyed_hash": "357dc55de0c7e382c900fd6e320acc04146be01db6a8ce7210b7189bd664ea69362396b77fdc0d2634a552970843722066c3c15902ae5097e00ff53f1e116f1cd5352720113a837ab2452cafbde4d54085d9cf5d21ca613071551b25d52e69d6c81123872b6f19cd3bc1333edf0c52b94de23ba772cf82636cff4542540a7738d5b930",
+      "derive_key": "effaa245f065fbf82ac186839a249707c3bddf6d3fdda22d1b95a3c970379bcb5d31013a167509e9066273ab6e2123bc835b408b067d88f96addb550d96b6852dad38e320b9d940f86db74d398c770f462118b35d2724efa13da97194491d96dd37c3c09cbef665953f2ee85ec83d88b88d11547a6f911c8217cca46defa2751e7f3ad"
+    },
+    {
+      "input_len": 2048,
+      "hash": "e776b6028c7cd22a4d0ba182a8bf62205d2ef576467e838ed6f2529b85fba24a9a60bf80001410ec9eea6698cd537939fad4749edd484cb541aced55cd9bf54764d063f23f6f1e32e12958ba5cfeb1bf618ad094266d4fc3c968c2088f677454c288c67ba0dba337b9d91c7e1ba586dc9a5bc2d5e90c14f53a8863ac75655461cea8f9",
+      "keyed_hash": "879cf1fa2ea0e79126cb1063617a05b6ad9d0b696d0d757cf053439f60a99dd10173b961cd574288194b23ece278c330fbb8585485e74967f31352a8183aa782b2b22f26cdcadb61eed1a5bc144b8198fbb0c13abbf8e3192c145d0a5c21633b0ef86054f42809df823389ee40811a5910dcbd1018af31c3b43aa55201ed4edaac74fe",
+      "derive_key": "7b2945cb4fef70885cc5d78a87bf6f6207dd901ff239201351ffac04e1088a23e2c11a1ebffcea4d80447867b61badb1383d842d4e79645d48dd82ccba290769caa7af8eaa1bd78a2a5e6e94fbdab78d9c7b74e894879f6a515257ccf6f95056f4e25390f24f6b35ffbb74b766202569b1d797f2d4bd9d17524c720107f985f4ddc583"
+    },
+    {
+      "input_len": 2049,
+      "hash": "5f4d72f40d7a5f82b15ca2b2e44b1de3c2ef86c426c95c1af0b687952256303096de31d71d74103403822a2e0bc1eb193e7aecc9643a76b7bbc0c9f9c52e8783aae98764ca468962b5c2ec92f0c74eb5448d519713e09413719431c802f948dd5d90425a4ecdadece9eb178d80f26efccae630734dff63340285adec2aed3b51073ad3",
+      "keyed_hash": "9f29700902f7c86e514ddc4df1e3049f258b2472b6dd5267f61bf13983b78dd5f9a88abfefdfa1e00b418971f2b39c64ca621e8eb37fceac57fd0c8fc8e117d43b81447be22d5d8186f8f5919ba6bcc6846bd7d50726c06d245672c2ad4f61702c646499ee1173daa061ffe15bf45a631e2946d616a4c345822f1151284712f76b2b0e",
+      "derive_key": "2ea477c5515cc3dd606512ee72bb3e0e758cfae7232826f35fb98ca1bcbdf27316d8e9e79081a80b046b60f6a263616f33ca464bd78d79fa18200d06c7fc9bffd808cc4755277a7d5e09da0f29ed150f6537ea9bed946227ff184cc66a72a5f8c1e4bd8b04e81cf40fe6dc4427ad5678311a61f4ffc39d195589bdbc670f63ae70f4b6"
+    },
+    {
+      "input_len": 3072,
+      "hash": "b98cb0ff3623be03326b373de6b9095218513e64f1ee2edd2525c7ad1e5cffd29a3f6b0b978d6608335c09dc94ccf682f9951cdfc501bfe47b9c9189a6fc7b404d120258506341a6d802857322fbd20d3e5dae05b95c88793fa83db1cb08e7d8008d1599b6209d78336e24839724c191b2a52a80448306e0daa84a3fdb566661a37e11",
+      "keyed_hash": "044a0e7b172a312dc02a4c9a818c036ffa2776368d7f528268d2e6b5df19177022f302d0529e4174cc507c463671217975e81dab02b8fdeb0d7ccc7568dd22574c783a76be215441b32e91b9a904be8ea81f7a0afd14bad8ee7c8efc305ace5d3dd61b996febe8da4f56ca0919359a7533216e2999fc87ff7d8f176fbecb3d6f34278b",
+      "derive_key": "050df97f8c2ead654d9bb3ab8c9178edcd902a32f8495949feadcc1e0480c46b3604131bbd6e3ba573b6dd682fa0a63e5b165d39fc43a625d00207607a2bfeb65ff1d29292152e26b298868e3b87be95d6458f6f2ce6118437b632415abe6ad522874bcd79e4030a5e7bad2efa90a7a7c67e93f0a18fb28369d0a9329ab5c24134ccb0"
+    },
+    {
+      "input_len": 3073,
+      "hash": "7124b49501012f81cc7f11ca069ec9226cecb8a2c850cfe644e327d22d3e1cd39a27ae3b79d68d89da9bf25bc27139ae65a324918a5f9b7828181e52cf373c84f35b639b7fccbb985b6f2fa56aea0c18f531203497b8bbd3a07ceb5926f1cab74d14bd66486d9a91eba99059a98bd1cd25876b2af5a76c3e9eed554ed72ea952b603bf",
+      "keyed_hash": "68dede9bef00ba89e43f31a6825f4cf433389fedae75c04ee9f0cf16a427c95a96d6da3fe985054d3478865be9a092250839a697bbda74e279e8a9e69f0025e4cfddd6cfb434b1cd9543aaf97c635d1b451a4386041e4bb100f5e45407cbbc24fa53ea2de3536ccb329e4eb9466ec37093a42cf62b82903c696a93a50b702c80f3c3c5",
+      "derive_key": "72613c9ec9ff7e40f8f5c173784c532ad852e827dba2bf85b2ab4b76f7079081576288e552647a9d86481c2cae75c2dd4e7c5195fb9ada1ef50e9c5098c249d743929191441301c69e1f48505a4305ec1778450ee48b8e69dc23a25960fe33070ea549119599760a8a2d28aeca06b8c5e9ba58bc19e11fe57b6ee98aa44b2a8e6b14a5"
+    },
+    {
+      "input_len": 4096,
+      "hash": "015094013f57a5277b59d8475c0501042c0b642e531b0a1c8f58d2163229e9690289e9409ddb1b99768eafe1623da896faf7e1114bebeadc1be30829b6f8af707d85c298f4f0ff4d9438aef948335612ae921e76d411c3a9111df62d27eaf871959ae0062b5492a0feb98ef3ed4af277f5395172dbe5c311918ea0074ce0036454f620",
+      "keyed_hash": "befc660aea2f1718884cd8deb9902811d332f4fc4a38cf7c7300d597a081bfc0bbb64a36edb564e01e4b4aaf3b060092a6b838bea44afebd2deb8298fa562b7b597c757b9df4c911c3ca462e2ac89e9a787357aaf74c3b56d5c07bc93ce899568a3eb17d9250c20f6c5f6c1e792ec9a2dcb715398d5a6ec6d5c54f586a00403a1af1de",
+      "derive_key": "1e0d7f3db8c414c97c6307cbda6cd27ac3b030949da8e23be1a1a924ad2f25b9d78038f7b198596c6cc4a9ccf93223c08722d684f240ff6569075ed81591fd93f9fff1110b3a75bc67e426012e5588959cc5a4c192173a03c00731cf84544f65a2fb9378989f72e9694a6a394a8a30997c2e67f95a504e631cd2c5f55246024761b245"
+    },
+    {
+      "input_len": 4097,
+      "hash": "9b4052b38f1c5fc8b1f9ff7ac7b27cd242487b3d890d15c96a1c25b8aa0fb99505f91b0b5600a11251652eacfa9497b31cd3c409ce2e45cfe6c0a016967316c426bd26f619eab5d70af9a418b845c608840390f361630bd497b1ab44019316357c61dbe091ce72fc16dc340ac3d6e009e050b3adac4b5b2c92e722cffdc46501531956",
+      "keyed_hash": "00df940cd36bb9fa7cbbc3556744e0dbc8191401afe70520ba292ee3ca80abbc606db4976cfdd266ae0abf667d9481831ff12e0caa268e7d3e57260c0824115a54ce595ccc897786d9dcbf495599cfd90157186a46ec800a6763f1c59e36197e9939e900809f7077c102f888caaf864b253bc41eea812656d46742e4ea42769f89b83f",
+      "derive_key": "aca51029626b55fda7117b42a7c211f8c6e9ba4fe5b7a8ca922f34299500ead8a897f66a400fed9198fd61dd2d58d382458e64e100128075fc54b860934e8de2e84170734b06e1d212a117100820dbc48292d148afa50567b8b84b1ec336ae10d40c8c975a624996e12de31abbe135d9d159375739c333798a80c64ae895e51e22f3ad"
+    },
+    {
+      "input_len": 5120,
+      "hash": "9cadc15fed8b5d854562b26a9536d9707cadeda9b143978f319ab34230535833acc61c8fdc114a2010ce8038c853e121e1544985133fccdd0a2d507e8e615e611e9a0ba4f47915f49e53d721816a9198e8b30f12d20ec3689989175f1bf7a300eee0d9321fad8da232ece6efb8e9fd81b42ad161f6b9550a069e66b11b40487a5f5059",
+      "keyed_hash": "2c493e48e9b9bf31e0553a22b23503c0a3388f035cece68eb438d22fa1943e209b4dc9209cd80ce7c1f7c9a744658e7e288465717ae6e56d5463d4f80cdb2ef56495f6a4f5487f69749af0c34c2cdfa857f3056bf8d807336a14d7b89bf62bef2fb54f9af6a546f818dc1e98b9e07f8a5834da50fa28fb5874af91bf06020d1bf0120e",
+      "derive_key": "7a7acac8a02adcf3038d74cdd1d34527de8a0fcc0ee3399d1262397ce5817f6055d0cefd84d9d57fe792d65a278fd20384ac6c30fdb340092f1a74a92ace99c482b28f0fc0ef3b923e56ade20c6dba47e49227166251337d80a037e987ad3a7f728b5ab6dfafd6e2ab1bd583a95d9c895ba9c2422c24ea0f62961f0dca45cad47bfa0d"
+    },
+    {
+      "input_len": 5121,
+      "hash": "628bd2cb2004694adaab7bbd778a25df25c47b9d4155a55f8fbd79f2fe154cff96adaab0613a6146cdaabe498c3a94e529d3fc1da2bd08edf54ed64d40dcd6777647eac51d8277d70219a9694334a68bc8f0f23e20b0ff70ada6f844542dfa32cd4204ca1846ef76d811cdb296f65e260227f477aa7aa008bac878f72257484f2b6c95",
+      "keyed_hash": "6ccf1c34753e7a044db80798ecd0782a8f76f33563accaddbfbb2e0ea4b2d0240d07e63f13667a8d1490e5e04f13eb617aea16a8c8a5aaed1ef6fbde1b0515e3c81050b361af6ead126032998290b563e3caddeaebfab592e155f2e161fb7cba939092133f23f9e65245e58ec23457b78a2e8a125588aad6e07d7f11a85b88d375b72d",
+      "derive_key": "b07f01e518e702f7ccb44a267e9e112d403a7b3f4883a47ffbed4b48339b3c341a0add0ac032ab5aaea1e4e5b004707ec5681ae0fcbe3796974c0b1cf31a194740c14519273eedaabec832e8a784b6e7cfc2c5952677e6c3f2c3914454082d7eb1ce1766ac7d75a4d3001fc89544dd46b5147382240d689bbbaefc359fb6ae30263165"
+    },
+    {
+      "input_len": 6144,
+      "hash": "3e2e5b74e048f3add6d21faab3f83aa44d3b2278afb83b80b3c35164ebeca2054d742022da6fdda444ebc384b04a54c3ac5839b49da7d39f6d8a9db03deab32aade156c1c0311e9b3435cde0ddba0dce7b26a376cad121294b689193508dd63151603c6ddb866ad16c2ee41585d1633a2cea093bea714f4c5d6b903522045b20395c83",
+      "keyed_hash": "3d6b6d21281d0ade5b2b016ae4034c5dec10ca7e475f90f76eac7138e9bc8f1dc35754060091dc5caf3efabe0603c60f45e415bb3407db67e6beb3d11cf8e4f7907561f05dace0c15807f4b5f389c841eb114d81a82c02a00b57206b1d11fa6e803486b048a5ce87105a686dee041207e095323dfe172df73deb8c9532066d88f9da7e",
+      "derive_key": "2a95beae63ddce523762355cf4b9c1d8f131465780a391286a5d01abb5683a1597099e3c6488aab6c48f3c15dbe1942d21dbcdc12115d19a8b8465fb54e9053323a9178e4275647f1a9927f6439e52b7031a0b465c861a3fc531527f7758b2b888cf2f20582e9e2c593709c0a44f9c6e0f8b963994882ea4168827823eef1f64169fef"
+    },
+    {
+      "input_len": 6145,
+      "hash": "f1323a8631446cc50536a9f705ee5cb619424d46887f3c376c695b70e0f0507f18a2cfdd73c6e39dd75ce7c1c6e3ef238fd54465f053b25d21044ccb2093beb015015532b108313b5829c3621ce324b8e14229091b7c93f32db2e4e63126a377d2a63a3597997d4f1cba59309cb4af240ba70cebff9a23d5e3ff0cdae2cfd54e070022",
+      "keyed_hash": "9ac301e9e39e45e3250a7e3b3df701aa0fb6889fbd80eeecf28dbc6300fbc539f3c184ca2f59780e27a576c1d1fb9772e99fd17881d02ac7dfd39675aca918453283ed8c3169085ef4a466b91c1649cc341dfdee60e32231fc34c9c4e0b9a2ba87ca8f372589c744c15fd6f985eec15e98136f25beeb4b13c4e43dc84abcc79cd4646c",
+      "derive_key": "379bcc61d0051dd489f686c13de00d5b14c505245103dc040d9e4dd1facab8e5114493d029bdbd295aaa744a59e31f35c7f52dba9c3642f773dd0b4262a9980a2aef811697e1305d37ba9d8b6d850ef07fe41108993180cf779aeece363704c76483458603bbeeb693cffbbe5588d1f3535dcad888893e53d977424bb707201569a8d2"
+    },
+    {
+      "input_len": 7168,
+      "hash": "61da957ec2499a95d6b8023e2b0e604ec7f6b50e80a9678b89d2628e99ada77a5707c321c83361793b9af62a40f43b523df1c8633cecb4cd14d00bdc79c78fca5165b863893f6d38b02ff7236c5a9a8ad2dba87d24c547cab046c29fc5bc1ed142e1de4763613bb162a5a538e6ef05ed05199d751f9eb58d332791b8d73fb74e4fce95",
+      "keyed_hash": "b42835e40e9d4a7f42ad8cc04f85a963a76e18198377ed84adddeaecacc6f3fca2f01d5277d69bb681c70fa8d36094f73ec06e452c80d2ff2257ed82e7ba348400989a65ee8daa7094ae0933e3d2210ac6395c4af24f91c2b590ef87d7788d7066ea3eaebca4c08a4f14b9a27644f99084c3543711b64a070b94f2c9d1d8a90d035d52",
+      "derive_key": "11c37a112765370c94a51415d0d651190c288566e295d505defdad895dae223730d5a5175a38841693020669c7638f40b9bc1f9f39cf98bda7a5b54ae24218a800a2116b34665aa95d846d97ea988bfcb53dd9c055d588fa21ba78996776ea6c40bc428b53c62b5f3ccf200f647a5aae8067f0ea1976391fcc72af1945100e2a6dcb88"
+    },
+    {
+      "input_len": 7169,
+      "hash": "a003fc7a51754a9b3c7fae0367ab3d782dccf28855a03d435f8cfe74605e781798a8b20534be1ca9eb2ae2df3fae2ea60e48c6fb0b850b1385b5de0fe460dbe9d9f9b0d8db4435da75c601156df9d047f4ede008732eb17adc05d96180f8a73548522840779e6062d643b79478a6e8dbce68927f36ebf676ffa7d72d5f68f050b119c8",
+      "keyed_hash": "ed9b1a922c046fdb3d423ae34e143b05ca1bf28b710432857bf738bcedbfa5113c9e28d72fcbfc020814ce3f5d4fc867f01c8f5b6caf305b3ea8a8ba2da3ab69fabcb438f19ff11f5378ad4484d75c478de425fb8e6ee809b54eec9bdb184315dc856617c09f5340451bf42fd3270a7b0b6566169f242e533777604c118a6358250f54",
+      "derive_key": "554b0a5efea9ef183f2f9b931b7497995d9eb26f5c5c6dad2b97d62fc5ac31d99b20652c016d88ba2a611bbd761668d5eda3e568e940faae24b0d9991c3bd25a65f770b89fdcadabcb3d1a9c1cb63e69721cacf1ae69fefdcef1e3ef41bc5312ccc17222199e47a26552c6adc460cf47a72319cb5039369d0060eaea59d6c65130f1dd"
+    },
+    {
+      "input_len": 8192,
+      "hash": "aae792484c8efe4f19e2ca7d371d8c467ffb10748d8a5a1ae579948f718a2a635fe51a27db045a567c1ad51be5aa34c01c6651c4d9b5b5ac5d0fd58cf18dd61a47778566b797a8c67df7b1d60b97b19288d2d877bb2df417ace009dcb0241ca1257d62712b6a4043b4ff33f690d849da91ea3bf711ed583cb7b7a7da2839ba71309bbf",
+      "keyed_hash": "dc9637c8845a770b4cbf76b8daec0eebf7dc2eac11498517f08d44c8fc00d58a4834464159dcbc12a0ba0c6d6eb41bac0ed6585cabfe0aca36a375e6c5480c22afdc40785c170f5a6b8a1107dbee282318d00d915ac9ed1143ad40765ec120042ee121cd2baa36250c618adaf9e27260fda2f94dea8fb6f08c04f8f10c78292aa46102",
+      "derive_key": "ad01d7ae4ad059b0d33baa3c01319dcf8088094d0359e5fd45d6aeaa8b2d0c3d4c9e58958553513b67f84f8eac653aeeb02ae1d5672dcecf91cd9985a0e67f4501910ecba25555395427ccc7241d70dc21c190e2aadee875e5aae6bf1912837e53411dabf7a56cbf8e4fb780432b0d7fe6cec45024a0788cf5874616407757e9e6bef7"
+    },
+    {
+      "input_len": 8193,
+      "hash": "bab6c09cb8ce8cf459261398d2e7aef35700bf488116ceb94a36d0f5f1b7bc3bb2282aa69be089359ea1154b9a9286c4a56af4de975a9aa4a5c497654914d279bea60bb6d2cf7225a2fa0ff5ef56bbe4b149f3ed15860f78b4e2ad04e158e375c1e0c0b551cd7dfc82f1b155c11b6b3ed51ec9edb30d133653bb5709d1dbd55f4e1ff6",
+      "keyed_hash": "954a2a75420c8d6547e3ba5b98d963e6fa6491addc8c023189cc519821b4a1f5f03228648fd983aef045c2fa8290934b0866b615f585149587dda2299039965328835a2b18f1d63b7e300fc76ff260b571839fe44876a4eae66cbac8c67694411ed7e09df51068a22c6e67d6d3dd2cca8ff12e3275384006c80f4db68023f24eebba57",
+      "derive_key": "af1e0346e389b17c23200270a64aa4e1ead98c61695d917de7d5b00491c9b0f12f20a01d6d622edf3de026a4db4e4526225debb93c1237934d71c7340bb5916158cbdafe9ac3225476b6ab57a12357db3abbad7a26c6e66290e44034fb08a20a8d0ec264f309994d2810c49cfba6989d7abb095897459f5425adb48aba07c5fb3c83c0"
+    },
+    {
+      "input_len": 16384,
+      "hash": "f875d6646de28985646f34ee13be9a576fd515f76b5b0a26bb324735041ddde49d764c270176e53e97bdffa58d549073f2c660be0e81293767ed4e4929f9ad34bbb39a529334c57c4a381ffd2a6d4bfdbf1482651b172aa883cc13408fa67758a3e47503f93f87720a3177325f7823251b85275f64636a8f1d599c2e49722f42e93893",
+      "keyed_hash": "9e9fc4eb7cf081ea7c47d1807790ed211bfec56aa25bb7037784c13c4b707b0df9e601b101e4cf63a404dfe50f2e1865bb12edc8fca166579ce0c70dba5a5c0fc960ad6f3772183416a00bd29d4c6e651ea7620bb100c9449858bf14e1ddc9ecd35725581ca5b9160de04060045993d972571c3e8f71e9d0496bfa744656861b169d65",
+      "derive_key": "160e18b5878cd0df1c3af85eb25a0db5344d43a6fbd7a8ef4ed98d0714c3f7e160dc0b1f09caa35f2f417b9ef309dfe5ebd67f4c9507995a531374d099cf8ae317542e885ec6f589378864d3ea98716b3bbb65ef4ab5e0ab5bb298a501f19a41ec19af84a5e6b428ecd813b1a47ed91c9657c3fba11c406bc316768b58f6802c9e9b57"
+    },
+    {
+      "input_len": 31744,
+      "hash": "62b6960e1a44bcc1eb1a611a8d6235b6b4b78f32e7abc4fb4c6cdcce94895c47860cc51f2b0c28a7b77304bd55fe73af663c02d3f52ea053ba43431ca5bab7bfea2f5e9d7121770d88f70ae9649ea713087d1914f7f312147e247f87eb2d4ffef0ac978bf7b6579d57d533355aa20b8b77b13fd09748728a5cc327a8ec470f4013226f",
+      "keyed_hash": "efa53b389ab67c593dba624d898d0f7353ab99e4ac9d42302ee64cbf9939a4193a7258db2d9cd32a7a3ecfce46144114b15c2fcb68a618a976bd74515d47be08b628be420b5e830fade7c080e351a076fbc38641ad80c736c8a18fe3c66ce12f95c61c2462a9770d60d0f77115bbcd3782b593016a4e728d4c06cee4505cb0c08a42ec",
+      "derive_key": "39772aef80e0ebe60596361e45b061e8f417429d529171b6764468c22928e28e9759adeb797a3fbf771b1bcea30150a020e317982bf0d6e7d14dd9f064bc11025c25f31e81bd78a921db0174f03dd481d30e93fd8e90f8b2fee209f849f2d2a52f31719a490fb0ba7aea1e09814ee912eba111a9fde9d5c274185f7bae8ba85d300a2b"
+    },
+    {
+      "input_len": 102400,
+      "hash": "bc3e3d41a1146b069abffad3c0d44860cf664390afce4d9661f7902e7943e085e01c59dab908c04c3342b816941a26d69c2605ebee5ec5291cc55e15b76146e6745f0601156c3596cb75065a9c57f35585a52e1ac70f69131c23d611ce11ee4ab1ec2c009012d236648e77be9295dd0426f29b764d65de58eb7d01dd42248204f45f8e",
+      "keyed_hash": "1c35d1a5811083fd7119f5d5d1ba027b4d01c0c6c49fb6ff2cf75393ea5db4a7f9dbdd3e1d81dcbca3ba241bb18760f207710b751846faaeb9dff8262710999a59b2aa1aca298a032d94eacfadf1aa192418eb54808db23b56e34213266aa08499a16b354f018fc4967d05f8b9d2ad87a7278337be9693fc638a3bfdbe314574ee6fc4",
+      "derive_key": "4652cff7a3f385a6103b5c260fc1593e13c778dbe608efb092fe7ee69df6e9c6d83a3e041bc3a48df2879f4a0a3ed40e7c961c73eff740f3117a0504c2dff4786d44fb17f1549eb0ba585e40ec29bf7732f0b7e286ff8acddc4cb1e23b87ff5d824a986458dcc6a04ac83969b80637562953df51ed1a7e90a7926924d2763778be8560"
+    }
+  ]
+}
diff --git a/thirdparty/blake3/tools/compiler_version/Cargo.toml b/thirdparty/blake3/tools/compiler_version/Cargo.toml
new file mode 100644
index 000000000..862509406
--- /dev/null
+++ b/thirdparty/blake3/tools/compiler_version/Cargo.toml
@@ -0,0 +1,7 @@
+[package]
+name = "compiler_version"
+version = "0.0.0"
+edition = "2021"
+
+[build-dependencies]
+cc = "1.0.50"
diff --git a/thirdparty/blake3/tools/compiler_version/build.rs b/thirdparty/blake3/tools/compiler_version/build.rs
new file mode 100644
index 000000000..3e14ebe67
--- /dev/null
+++ b/thirdparty/blake3/tools/compiler_version/build.rs
@@ -0,0 +1,6 @@
+fn main() {
+    let build = cc::Build::new();
+    let compiler = build.get_compiler();
+    let compiler_path = compiler.path().to_string_lossy();
+    println!("cargo:rustc-env=COMPILER_PATH={}", compiler_path);
+}
diff --git a/thirdparty/blake3/tools/compiler_version/src/main.rs b/thirdparty/blake3/tools/compiler_version/src/main.rs
new file mode 100644
index 000000000..767cb31bd
--- /dev/null
+++ b/thirdparty/blake3/tools/compiler_version/src/main.rs
@@ -0,0 +1,27 @@
+use std::process::Command;
+
+fn main() {
+    // Print the rustc version.
+    Command::new(env!("CARGO"))
+        .args(&["rustc", "--quiet", "--", "--version"])
+        .status()
+        .unwrap();
+    println!();
+
+    // Print the Cargo version.
+    Command::new(env!("CARGO"))
+        .args(&["--version"])
+        .status()
+        .unwrap();
+    println!();
+
+    // Print the C compiler version. This relies on C compiler detection done
+    // in build.rs, which sets the COMPILER_PATH variable.
+    let compiler_path = env!("COMPILER_PATH");
+    let mut compiler_command = Command::new(compiler_path);
+    // Use the --version flag on everything other than MSVC.
+    if !cfg!(target_env = "msvc") {
+        compiler_command.arg("--version");
+    }
+    let _ = compiler_command.status().unwrap();
+}
diff --git a/thirdparty/blake3/tools/instruction_set_support/Cargo.toml b/thirdparty/blake3/tools/instruction_set_support/Cargo.toml
new file mode 100644
index 000000000..eef8a80d2
--- /dev/null
+++ b/thirdparty/blake3/tools/instruction_set_support/Cargo.toml
@@ -0,0 +1,6 @@
+[package]
+name = "instruction_set_support"
+version = "0.0.0"
+edition = "2021"
+
+[dependencies]
diff --git a/thirdparty/blake3/tools/instruction_set_support/src/main.rs b/thirdparty/blake3/tools/instruction_set_support/src/main.rs
new file mode 100644
index 000000000..6b509b053
--- /dev/null
+++ b/thirdparty/blake3/tools/instruction_set_support/src/main.rs
@@ -0,0 +1,10 @@
+fn main() {
+    #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
+    {
+        dbg!(is_x86_feature_detected!("sse2"));
+        dbg!(is_x86_feature_detected!("sse4.1"));
+        dbg!(is_x86_feature_detected!("avx2"));
+        dbg!(is_x86_feature_detected!("avx512f"));
+        dbg!(is_x86_feature_detected!("avx512vl"));
+    }
+}
diff --git a/thirdparty/blake3/tools/release.md b/thirdparty/blake3/tools/release.md
new file mode 100644
index 000000000..26c492723
--- /dev/null
+++ b/thirdparty/blake3/tools/release.md
@@ -0,0 +1,16 @@
+# Release checklist
+
+- Make sure `cargo outdated -R` is clean in the root and in b3sum/.
+- Bump the version in the root Cargo.toml.
+- Bump the version in b3sum/Cargo.toml.
+    - Bump the dependency version too, if new features are used.
+- Delete b3sum/Cargo.lock and recreate it with `cargo build` or similar.
+- Update the `-h` output in b3sum/README.md if it's changed.
+- Bump `BLAKE3_VERSION_STRING` in c/blake3.h.
+- Bump `VERSION` in c/CMakeLists.txt.
+- Make a version bump commit with change notes.
+- `git push` and make sure CI is green.
+- `git tag` the version bump commit with the new version number.
+- `git push --tags`
+- `cargo publish` in the root.
+- `cargo publish` in b3sum/.