aboutsummaryrefslogtreecommitdiff
path: root/src/bench/bench.cpp
diff options
context:
space:
mode:
authorWladimir J. van der Laan <laanwj@gmail.com>2016-11-22 10:22:08 +0100
committerWladimir J. van der Laan <laanwj@gmail.com>2016-11-22 12:20:57 +0100
commit35328187463a7078b4206e394c21d5515929c7de (patch)
treee310e5beca1e7ad006cb48423c64f72d6a4d00fd /src/bench/bench.cpp
parent55b2eddcc8fd407aa62fd280f4767e3034a7eb04 (diff)
downloadbitcoin-35328187463a7078b4206e394c21d5515929c7de.tar.xz
bench: Add support for measuring CPU cycles
This adds cycle min/max/avg to the statistics. Supported on x86 and x86_64 (natively through rdtsc), as well as Linux (perf syscall).
Diffstat (limited to 'src/bench/bench.cpp')
-rw-r--r--src/bench/bench.cpp22
1 files changed, 20 insertions, 2 deletions
diff --git a/src/bench/bench.cpp b/src/bench/bench.cpp
index 8942da8c74..af3d152c9a 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.
@@ -61,6 +74,8 @@ 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) {
@@ -72,6 +87,7 @@ bool State::KeepRunning()
}
}
lastTime = now;
+ lastCycles = nowCycles;
++count;
if (now - beginTime < maxElapsed) return true; // Keep going
@@ -80,7 +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;
}