Bringing /src/ up to speed with 0.8.6.2

1 files changed, 42 insertions, 36 deletions

diff --git a/src/scrypt.cpp b/src/scrypt.cpp
index 1342ea782..efce1f4d2 100644
--- a/src/scrypt.cpp
+++ b/src/scrypt.cpp
@@ -27,7 +27,7 @@
  * online backup system.
  */
 
-// Copyright (c) 2013-2014 Dogecoin Developers
+ // Copyright (c) 2013-2014 Dogecoin Developers
 
 #include "scrypt.h"
 #include "util.h"
@@ -36,6 +36,16 @@
 #include <string.h>
 #include <openssl/sha.h>
 
+#if defined(USE_SSE2) && !defined(USE_SSE2_ALWAYS)
+#ifdef _MSC_VER
+// MSVC 64bit is unable to use inline asm
+#include <intrin.h>
+#else
+// GCC Linux or i686-w64-mingw32
+#include <cpuid.h>
+#endif
+#endif
+
 static inline uint32_t be32dec(const void *pp)
 {
 	const uint8_t *p = (uint8_t const *)pp;
@@ -204,26 +214,26 @@ static inline void xor_salsa8(uint32_t B[16], const uint32_t Bx[16])
 		/* Operate on columns. */
 		x04 ^= ROTL(x00 + x12,  7);  x09 ^= ROTL(x05 + x01,  7);
 		x14 ^= ROTL(x10 + x06,  7);  x03 ^= ROTL(x15 + x11,  7);
-		
+
 		x08 ^= ROTL(x04 + x00,  9);  x13 ^= ROTL(x09 + x05,  9);
 		x02 ^= ROTL(x14 + x10,  9);  x07 ^= ROTL(x03 + x15,  9);
-		
+
 		x12 ^= ROTL(x08 + x04, 13);  x01 ^= ROTL(x13 + x09, 13);
 		x06 ^= ROTL(x02 + x14, 13);  x11 ^= ROTL(x07 + x03, 13);
-		
+
 		x00 ^= ROTL(x12 + x08, 18);  x05 ^= ROTL(x01 + x13, 18);
 		x10 ^= ROTL(x06 + x02, 18);  x15 ^= ROTL(x11 + x07, 18);
-		
+
 		/* Operate on rows. */
 		x01 ^= ROTL(x00 + x03,  7);  x06 ^= ROTL(x05 + x04,  7);
 		x11 ^= ROTL(x10 + x09,  7);  x12 ^= ROTL(x15 + x14,  7);
-		
+
 		x02 ^= ROTL(x01 + x00,  9);  x07 ^= ROTL(x06 + x05,  9);
 		x08 ^= ROTL(x11 + x10,  9);  x13 ^= ROTL(x12 + x15,  9);
-		
+
 		x03 ^= ROTL(x02 + x01, 13);  x04 ^= ROTL(x07 + x06, 13);
 		x09 ^= ROTL(x08 + x11, 13);  x14 ^= ROTL(x13 + x12, 13);
-		
+
 		x00 ^= ROTL(x03 + x02, 18);  x05 ^= ROTL(x04 + x07, 18);
 		x10 ^= ROTL(x09 + x08, 18);  x15 ^= ROTL(x14 + x13, 18);
 	}
@@ -253,7 +263,7 @@ void scrypt_1024_1_1_256_sp_generic(const char *input, char *output, char *scrat
 	uint32_t i, j, k;
 
 	V = (uint32_t *)(((uintptr_t)(scratchpad) + 63) & ~ (uintptr_t)(63));
-	
+
 	PBKDF2_SHA256((const uint8_t *)input, 80, (const uint8_t *)input, 80, 1, B, 128);
 
 	for (k = 0; k < 32; k++)
@@ -279,47 +289,43 @@ void scrypt_1024_1_1_256_sp_generic(const char *input, char *output, char *scrat
 }
 
 #if defined(USE_SSE2)
-#if defined(_M_X64) || defined(__x86_64__) || defined(_M_AMD64) || (defined(MAC_OSX) && defined(__i386__))
-/* Always SSE2 */
-void scrypt_detect_sse2(unsigned int cpuid_edx)
+// By default, set to generic scrypt function. This will prevent crash in case when scrypt_detect_sse2() wasn't called
+void (*scrypt_1024_1_1_256_sp_detected)(const char *input, char *output, char *scratchpad) = &scrypt_1024_1_1_256_sp_generic;
+
+void scrypt_detect_sse2()
 {
+#if defined(USE_SSE2_ALWAYS)
     printf("scrypt: using scrypt-sse2 as built.\n");
-}
-#else
-/* Detect SSE2 */
-void (*scrypt_1024_1_1_256_sp)(const char *input, char *output, char *scratchpad);
+#else // USE_SSE2_ALWAYS
+    // 32bit x86 Linux or Windows, detect cpuid features
+    unsigned int cpuid_edx=0;
+#if defined(_MSC_VER)
+    // MSVC
+    int x86cpuid[4];
+    __cpuid(x86cpuid, 1);
+    cpuid_edx = (unsigned int)buffer[3];
+#else // _MSC_VER
+    // Linux or i686-w64-mingw32 (gcc-4.6.3)
+    unsigned int eax, ebx, ecx;
+    __get_cpuid(1, &eax, &ebx, &ecx, &cpuid_edx);
+#endif // _MSC_VER
 
-void scrypt_detect_sse2(unsigned int cpuid_edx)
-{
     if (cpuid_edx & 1<<26)
     {
-        scrypt_1024_1_1_256_sp = &scrypt_1024_1_1_256_sp_sse2;
+        scrypt_1024_1_1_256_sp_detected = &scrypt_1024_1_1_256_sp_sse2;
         printf("scrypt: using scrypt-sse2 as detected.\n");
     }
     else
     {
-        scrypt_1024_1_1_256_sp = &scrypt_1024_1_1_256_sp_generic;
+        scrypt_1024_1_1_256_sp_detected = &scrypt_1024_1_1_256_sp_generic;
         printf("scrypt: using scrypt-generic, SSE2 unavailable.\n");
     }
+#endif // USE_SSE2_ALWAYS
 }
 #endif
-#endif
 
 void scrypt_1024_1_1_256(const char *input, char *output)
 {
 	char scratchpad[SCRYPT_SCRATCHPAD_SIZE];
-#if defined(USE_SSE2)
-        // Detection would work, but in cases where we KNOW it always has SSE2,
-        // it is faster to use directly than to use a function pointer or conditional.
-#if defined(_M_X64) || defined(__x86_64__) || defined(_M_AMD64) || (defined(MAC_OSX) && defined(__i386__))
-        // Always SSE2: x86_64 or Intel MacOS X
-        scrypt_1024_1_1_256_sp_sse2(input, output, scratchpad);
-#else
-        // Detect SSE2: 32bit x86 Linux or Windows
-        scrypt_1024_1_1_256_sp(input, output, scratchpad);
-#endif
-#else
-        // Generic scrypt
-        scrypt_1024_1_1_256_sp_generic(input, output, scratchpad);
-#endif
-}
+    scrypt_1024_1_1_256_sp(input, output, scratchpad);
+}
\ No newline at end of file