aboutsummaryrefslogtreecommitdiff
path: root/src/bench
diff options
context:
space:
mode:
Diffstat (limited to 'src/bench')
-rw-r--r--src/bench/Examples.cpp2
-rw-r--r--src/bench/base58.cpp2
-rw-r--r--src/bench/bench.cpp29
-rw-r--r--src/bench/bench.h10
-rw-r--r--src/bench/bench_bitcoin.cpp2
-rw-r--r--src/bench/checkblock.cpp3
-rw-r--r--src/bench/coin_selection.cpp2
-rw-r--r--src/bench/perf.cpp53
-rw-r--r--src/bench/perf.h37
9 files changed, 129 insertions, 11 deletions
diff --git a/src/bench/Examples.cpp b/src/bench/Examples.cpp
index b6b020a97..9f35a1ea0 100644
--- a/src/bench/Examples.cpp
+++ b/src/bench/Examples.cpp
@@ -3,7 +3,7 @@
// file COPYING or http://www.opensource.org/licenses/mit-license.php.
#include "bench.h"
-#include "main.h"
+#include "validation.h"
#include "utiltime.h"
// Sanity test: this should loop ten times, and
diff --git a/src/bench/base58.cpp b/src/bench/base58.cpp
index a791b5b7f..3319c179b 100644
--- a/src/bench/base58.cpp
+++ b/src/bench/base58.cpp
@@ -4,7 +4,7 @@
#include "bench.h"
-#include "main.h"
+#include "validation.h"
#include "base58.h"
#include <vector>
diff --git a/src/bench/bench.cpp b/src/bench/bench.cpp
index 227546a7a..af3d152c9 100644
--- a/src/bench/bench.cpp
+++ b/src/bench/bench.cpp
@@ -3,6 +3,7 @@
// file COPYING or http://www.opensource.org/licenses/mit-license.php.
#include "bench.h"
+#include "perf.h"
#include <iostream>
#include <iomanip>
@@ -26,7 +27,9 @@ BenchRunner::BenchRunner(std::string name, BenchFunction func)
void
BenchRunner::RunAll(double elapsedTimeForOne)
{
- std::cout << "#Benchmark" << "," << "count" << "," << "min" << "," << "max" << "," << "average" << "\n";
+ perf_init();
+ std::cout << "#Benchmark" << "," << "count" << "," << "min" << "," << "max" << "," << "average" << ","
+ << "min_cycles" << "," << "max_cycles" << "," << "average_cycles" << "\n";
for (std::map<std::string,BenchFunction>::iterator it = benchmarks.begin();
it != benchmarks.end(); ++it) {
@@ -35,6 +38,7 @@ BenchRunner::RunAll(double elapsedTimeForOne)
BenchFunction& func = it->second;
func(state);
}
+ perf_fini();
}
bool State::KeepRunning()
@@ -44,8 +48,10 @@ bool State::KeepRunning()
return true;
}
double now;
+ uint64_t nowCycles;
if (count == 0) {
lastTime = beginTime = now = gettimedouble();
+ lastCycles = beginCycles = nowCycles = perf_cpucycles();
}
else {
now = gettimedouble();
@@ -53,6 +59,13 @@ bool State::KeepRunning()
double elapsedOne = elapsed * countMaskInv;
if (elapsedOne < minTime) minTime = elapsedOne;
if (elapsedOne > maxTime) maxTime = elapsedOne;
+
+ // We only use relative values, so don't have to handle 64-bit wrap-around specially
+ nowCycles = perf_cpucycles();
+ uint64_t elapsedOneCycles = (nowCycles - lastCycles) * countMaskInv;
+ if (elapsedOneCycles < minCycles) minCycles = elapsedOneCycles;
+ if (elapsedOneCycles > maxCycles) maxCycles = elapsedOneCycles;
+
if (elapsed*128 < maxElapsed) {
// If the execution was much too fast (1/128th of maxElapsed), increase the count mask by 8x and restart timing.
// The restart avoids including the overhead of this code in the measurement.
@@ -61,14 +74,20 @@ bool State::KeepRunning()
count = 0;
minTime = std::numeric_limits<double>::max();
maxTime = std::numeric_limits<double>::min();
+ minCycles = std::numeric_limits<uint64_t>::max();
+ maxCycles = std::numeric_limits<uint64_t>::min();
return true;
}
if (elapsed*16 < maxElapsed) {
- countMask = ((countMask<<1)|1) & ((1LL<<60)-1);
- countMaskInv = 1./(countMask+1);
+ uint64_t newCountMask = ((countMask<<1)|1) & ((1LL<<60)-1);
+ if ((count & newCountMask)==0) {
+ countMask = newCountMask;
+ countMaskInv = 1./(countMask+1);
+ }
}
}
lastTime = now;
+ lastCycles = nowCycles;
++count;
if (now - beginTime < maxElapsed) return true; // Keep going
@@ -77,7 +96,9 @@ bool State::KeepRunning()
// Output results
double average = (now-beginTime)/count;
- std::cout << std::fixed << std::setprecision(15) << name << "," << count << "," << minTime << "," << maxTime << "," << average << "\n";
+ int64_t averageCycles = (nowCycles-beginCycles)/count;
+ std::cout << std::fixed << std::setprecision(15) << name << "," << count << "," << minTime << "," << maxTime << "," << average << ","
+ << minCycles << "," << maxCycles << "," << averageCycles << "\n";
return false;
}
diff --git a/src/bench/bench.h b/src/bench/bench.h
index f13b145aa..caf73e949 100644
--- a/src/bench/bench.h
+++ b/src/bench/bench.h
@@ -41,12 +41,18 @@ namespace benchmark {
double maxElapsed;
double beginTime;
double lastTime, minTime, maxTime, countMaskInv;
- int64_t count;
- int64_t countMask;
+ uint64_t count;
+ uint64_t countMask;
+ uint64_t beginCycles;
+ uint64_t lastCycles;
+ uint64_t minCycles;
+ uint64_t maxCycles;
public:
State(std::string _name, double _maxElapsed) : name(_name), maxElapsed(_maxElapsed), count(0) {
minTime = std::numeric_limits<double>::max();
maxTime = std::numeric_limits<double>::min();
+ minCycles = std::numeric_limits<uint64_t>::max();
+ maxCycles = std::numeric_limits<uint64_t>::min();
countMask = 1;
countMaskInv = 1./(countMask + 1);
}
diff --git a/src/bench/bench_bitcoin.cpp b/src/bench/bench_bitcoin.cpp
index db1402216..bd768180c 100644
--- a/src/bench/bench_bitcoin.cpp
+++ b/src/bench/bench_bitcoin.cpp
@@ -5,7 +5,7 @@
#include "bench.h"
#include "key.h"
-#include "main.h"
+#include "validation.h"
#include "util.h"
int
diff --git a/src/bench/checkblock.cpp b/src/bench/checkblock.cpp
index 4a564d3fc..230e4ca77 100644
--- a/src/bench/checkblock.cpp
+++ b/src/bench/checkblock.cpp
@@ -5,7 +5,8 @@
#include "bench.h"
#include "chainparams.h"
-#include "main.h"
+#include "validation.h"
+#include "streams.h"
#include "consensus/validation.h"
namespace block_bench {
diff --git a/src/bench/coin_selection.cpp b/src/bench/coin_selection.cpp
index 7091ee3e1..32690fe48 100644
--- a/src/bench/coin_selection.cpp
+++ b/src/bench/coin_selection.cpp
@@ -19,7 +19,7 @@ static void addCoin(const CAmount& nValue, const CWallet& wallet, vector<COutput
tx.nLockTime = nextLockTime++; // so all transactions get different hashes
tx.vout.resize(nInput + 1);
tx.vout[nInput].nValue = nValue;
- CWalletTx* wtx = new CWalletTx(&wallet, tx);
+ CWalletTx* wtx = new CWalletTx(&wallet, MakeTransactionRef(std::move(tx)));
int nAge = 6 * 24;
COutput output(wtx, nInput, nAge, true, true);
diff --git a/src/bench/perf.cpp b/src/bench/perf.cpp
new file mode 100644
index 000000000..1f43e5d3a
--- /dev/null
+++ b/src/bench/perf.cpp
@@ -0,0 +1,53 @@
+// Copyright (c) 2016 The Bitcoin Core developers
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+
+#include "perf.h"
+
+#if defined(__i386__) || defined(__x86_64__)
+
+/* These architectures support quering the cycle counter
+ * from user space, no need for any syscall overhead.
+ */
+void perf_init(void) { }
+void perf_fini(void) { }
+
+#elif defined(__linux__)
+
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <linux/perf_event.h>
+
+static int fd = -1;
+static struct perf_event_attr attr;
+
+void perf_init(void)
+{
+ attr.type = PERF_TYPE_HARDWARE;
+ attr.config = PERF_COUNT_HW_CPU_CYCLES;
+ fd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, 0);
+}
+
+void perf_fini(void)
+{
+ if (fd != -1) {
+ close(fd);
+ }
+}
+
+uint64_t perf_cpucycles(void)
+{
+ uint64_t result = 0;
+ if (fd == -1 || read(fd, &result, sizeof(result)) < (ssize_t)sizeof(result)) {
+ return 0;
+ }
+ return result;
+}
+
+#else /* Unhandled platform */
+
+void perf_init(void) { }
+void perf_fini(void) { }
+uint64_t perf_cpucycles(void) { return 0; }
+
+#endif
diff --git a/src/bench/perf.h b/src/bench/perf.h
new file mode 100644
index 000000000..681bd0c8a
--- /dev/null
+++ b/src/bench/perf.h
@@ -0,0 +1,37 @@
+// Copyright (c) 2016 The Bitcoin Core developers
+// Distributed under the MIT software license, see the accompanying
+// file COPYING or http://www.opensource.org/licenses/mit-license.php.
+
+/** Functions for measurement of CPU cycles */
+#ifndef H_PERF
+#define H_PERF
+
+#include <stdint.h>
+
+#if defined(__i386__)
+
+static inline uint64_t perf_cpucycles(void)
+{
+ uint64_t x;
+ __asm__ volatile (".byte 0x0f, 0x31" : "=A" (x));
+ return x;
+}
+
+#elif defined(__x86_64__)
+
+static inline uint64_t perf_cpucycles(void)
+{
+ uint32_t hi, lo;
+ __asm__ __volatile__ ("rdtsc" : "=a"(lo), "=d"(hi));
+ return ((uint64_t)lo)|(((uint64_t)hi)<<32);
+}
+#else
+
+uint64_t perf_cpucycles(void);
+
+#endif
+
+void perf_init(void);
+void perf_fini(void);
+
+#endif // H_PERF