aboutsummaryrefslogtreecommitdiff
path: root/src/bench
diff options
context:
space:
mode:
Diffstat (limited to 'src/bench')
-rw-r--r--src/bench/bench.cpp41
-rw-r--r--src/bench/bench.h25
-rw-r--r--src/bench/rollingbloom.cpp7
3 files changed, 40 insertions, 33 deletions
diff --git a/src/bench/bench.cpp b/src/bench/bench.cpp
index 849d924af2..4c5a036773 100644
--- a/src/bench/bench.cpp
+++ b/src/bench/bench.cpp
@@ -8,29 +8,25 @@
#include <assert.h>
#include <iostream>
#include <iomanip>
-#include <sys/time.h>
benchmark::BenchRunner::BenchmarkMap &benchmark::BenchRunner::benchmarks() {
static std::map<std::string, benchmark::BenchFunction> benchmarks_map;
return benchmarks_map;
}
-static double gettimedouble(void) {
- struct timeval tv;
- gettimeofday(&tv, nullptr);
- return tv.tv_usec * 0.000001 + tv.tv_sec;
-}
-
benchmark::BenchRunner::BenchRunner(std::string name, benchmark::BenchFunction func)
{
benchmarks().insert(std::make_pair(name, func));
}
void
-benchmark::BenchRunner::RunAll(double elapsedTimeForOne)
+benchmark::BenchRunner::RunAll(benchmark::duration elapsedTimeForOne)
{
perf_init();
- std::cout << "#Benchmark" << "," << "count" << "," << "min" << "," << "max" << "," << "average" << ","
+ if (std::ratio_less_equal<benchmark::clock::period, std::micro>::value) {
+ std::cerr << "WARNING: Clock precision is worse than microsecond - benchmarks may be less accurate!\n";
+ }
+ std::cout << "#Benchmark" << "," << "count" << "," << "min(ns)" << "," << "max(ns)" << "," << "average(ns)" << ","
<< "min_cycles" << "," << "max_cycles" << "," << "average_cycles" << "\n";
for (const auto &p: benchmarks()) {
@@ -46,22 +42,23 @@ bool benchmark::State::KeepRunning()
++count;
return true;
}
- double now;
+ time_point now;
+
uint64_t nowCycles;
if (count == 0) {
- lastTime = beginTime = now = gettimedouble();
+ lastTime = beginTime = now = clock::now();
lastCycles = beginCycles = nowCycles = perf_cpucycles();
}
else {
- now = gettimedouble();
- double elapsed = now - lastTime;
- double elapsedOne = elapsed * countMaskInv;
+ now = clock::now();
+ auto elapsed = now - lastTime;
+ auto elapsedOne = elapsed / (countMask + 1);
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;
+ uint64_t elapsedOneCycles = (nowCycles - lastCycles) / (countMask + 1);
if (elapsedOneCycles < minCycles) minCycles = elapsedOneCycles;
if (elapsedOneCycles > maxCycles) maxCycles = elapsedOneCycles;
@@ -69,10 +66,9 @@ bool benchmark::State::KeepRunning()
// 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.
countMask = ((countMask<<3)|7) & ((1LL<<60)-1);
- countMaskInv = 1./(countMask+1);
count = 0;
- minTime = std::numeric_limits<double>::max();
- maxTime = std::numeric_limits<double>::min();
+ minTime = duration::max();
+ maxTime = duration::zero();
minCycles = std::numeric_limits<uint64_t>::max();
maxCycles = std::numeric_limits<uint64_t>::min();
return true;
@@ -81,7 +77,6 @@ bool benchmark::State::KeepRunning()
uint64_t newCountMask = ((countMask<<1)|1) & ((1LL<<60)-1);
if ((count & newCountMask)==0) {
countMask = newCountMask;
- countMaskInv = 1./(countMask+1);
}
}
}
@@ -96,9 +91,13 @@ bool benchmark::State::KeepRunning()
assert(count != 0 && "count == 0 => (now == 0 && beginTime == 0) => return above");
// Output results
- double average = (now-beginTime)/count;
+ // Duration casts are only necessary here because hardware with sub-nanosecond clocks
+ // will lose precision.
+ int64_t min_elapsed = std::chrono::duration_cast<std::chrono::nanoseconds>(minTime).count();
+ int64_t max_elapsed = std::chrono::duration_cast<std::chrono::nanoseconds>(maxTime).count();
+ int64_t avg_elapsed = std::chrono::duration_cast<std::chrono::nanoseconds>((now-beginTime)/count).count();
int64_t averageCycles = (nowCycles-beginCycles)/count;
- std::cout << std::fixed << std::setprecision(15) << name << "," << count << "," << minTime << "," << maxTime << "," << average << ","
+ std::cout << std::fixed << std::setprecision(15) << name << "," << count << "," << min_elapsed << "," << max_elapsed << "," << avg_elapsed << ","
<< minCycles << "," << maxCycles << "," << averageCycles << "\n";
std::cout.copyfmt(std::ios(nullptr));
diff --git a/src/bench/bench.h b/src/bench/bench.h
index 1f36f2a4bc..ab5c3d5604 100644
--- a/src/bench/bench.h
+++ b/src/bench/bench.h
@@ -9,6 +9,7 @@
#include <limits>
#include <map>
#include <string>
+#include <chrono>
#include <boost/preprocessor/cat.hpp>
#include <boost/preprocessor/stringize.hpp>
@@ -36,12 +37,21 @@ BENCHMARK(CODE_TO_TIME);
*/
namespace benchmark {
+ // In case high_resolution_clock is steady, prefer that, otherwise use steady_clock.
+ struct best_clock {
+ using hi_res_clock = std::chrono::high_resolution_clock;
+ using steady_clock = std::chrono::steady_clock;
+ using type = std::conditional<hi_res_clock::is_steady, hi_res_clock, steady_clock>::type;
+ };
+ using clock = best_clock::type;
+ using time_point = clock::time_point;
+ using duration = clock::duration;
class State {
std::string name;
- double maxElapsed;
- double beginTime;
- double lastTime, minTime, maxTime, countMaskInv;
+ duration maxElapsed;
+ time_point beginTime, lastTime;
+ duration minTime, maxTime;
uint64_t count;
uint64_t countMask;
uint64_t beginCycles;
@@ -49,13 +59,12 @@ namespace benchmark {
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();
+ State(std::string _name, duration _maxElapsed) : name(_name), maxElapsed(_maxElapsed), count(0) {
+ minTime = duration::max();
+ maxTime = duration::zero();
minCycles = std::numeric_limits<uint64_t>::max();
maxCycles = std::numeric_limits<uint64_t>::min();
countMask = 1;
- countMaskInv = 1./(countMask + 1);
}
bool KeepRunning();
};
@@ -70,7 +79,7 @@ namespace benchmark {
public:
BenchRunner(std::string name, BenchFunction func);
- static void RunAll(double elapsedTimeForOne=1.0);
+ static void RunAll(duration elapsedTimeForOne = std::chrono::seconds(1));
};
}
diff --git a/src/bench/rollingbloom.cpp b/src/bench/rollingbloom.cpp
index 73c02cf718..a93d0fb0a5 100644
--- a/src/bench/rollingbloom.cpp
+++ b/src/bench/rollingbloom.cpp
@@ -6,7 +6,6 @@
#include "bench.h"
#include "bloom.h"
-#include "utiltime.h"
static void RollingBloom(benchmark::State& state)
{
@@ -23,10 +22,10 @@ static void RollingBloom(benchmark::State& state)
data[2] = count >> 16;
data[3] = count >> 24;
if (countnow == nEntriesPerGeneration) {
- int64_t b = GetTimeMicros();
+ auto b = benchmark::clock::now();
filter.insert(data);
- int64_t e = GetTimeMicros();
- std::cout << "RollingBloom-refresh,1," << (e-b)*0.000001 << "," << (e-b)*0.000001 << "," << (e-b)*0.000001 << "\n";
+ auto total = std::chrono::duration_cast<std::chrono::nanoseconds>(benchmark::clock::now() - b).count();
+ std::cout << "RollingBloom-refresh,1," << total << "," << total << "," << total << "\n";
countnow = 0;
} else {
filter.insert(data);
generated by cgit v1.2.3 (git 2.26.2) at 2024-08-24 22:04:53 +0000