/ds_benchmark.h
Last active Apr 29, 2020
C macros for performance testing
| /******************************************************************************************** | |
| * ds_benchmark.h: Macros for simple benchmarking of C code. | |
| * | |
| * See instructions for usage below. | |
| * Software originally developed by Douglas Stebila. | |
| * Most recent version at https://gist.github.com/dstebila/6980008ec98209ef6075 | |
| * | |
| * This is free and unencumbered software released into the public domain. | |
| * | |
| * Anyone is free to copy, modify, publish, use, compile, sell, or | |
| * distribute this software, either in source code form or as a compiled | |
| * binary, for any purpose, commercial or non-commercial, and by any | |
| * means. | |
| * | |
| * In jurisdictions that recognize copyright laws, the author or authors | |
| * of this software dedicate any and all copyright interest in the | |
| * software to the public domain. We make this dedication for the benefit | |
| * of the public at large and to the detriment of our heirs and | |
| * successors. We intend this dedication to be an overt act of | |
| * relinquishment in perpetuity of all present and future rights to this | |
| * software under copyright law. | |
| * | |
| * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
| * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
| * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. | |
| * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR | |
| * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, | |
| * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR | |
| * OTHER DEALINGS IN THE SOFTWARE. | |
| * | |
| * For more information, please refer to <http://unlicense.org> | |
| ********************************************************************************************/ | |
| /** \file ds_benchmark.h | |
| * Macros for simple benchmarking of C code. | |
| */ | |
| #if 0 | |
| /* example code: timing two operations */ | |
| #include "ds_benchmark.h" | |
| ... | |
| DEFINE_TIMER_VARIABLES | |
| INITIALIZE_TIMER | |
| START_TIMER | |
| // your operation here | |
| STOP_TIMER | |
| START_TIMER | |
| // another operation here | |
| STOP_TIMER | |
| FINALIZE_TIMER | |
| PRINT_TIMER_HEADER | |
| PRINT_TIMER_AVG("my operation") | |
| PRINT_TIMER_FOOTER | |
| /* example code: average multiple runs, run for e.g. 30 seconds */ | |
| #include "ds_benchmark.h" | |
| ... | |
| PRINT_TIMER_HEADER | |
| TIME_OPERATION_SECONDS(MyFunction(myarg1, myarg2, ...), "my operation", 30) | |
| TIME_OPERATION_SECONDS(MyOtherFunction(myarg3), "my other operation", 30) | |
| PRINT_TIMER_FOOTER | |
| /* example code: average multiple runs, run for e.g. 100 iterations */ | |
| #include "ds_benchmark.h" | |
| ... | |
| PRINT_TIMER_HEADER | |
| TIME_OPERATION_ITERATIONS(MyFunction(myarg1, myarg2, ...), "my operation", 1000) | |
| TIME_OPERATION_ITERATIONS(MyOtherFunction(myarg3), "my other operation", 100) | |
| PRINT_TIMER_FOOTER | |
| /* For most accurate results: | |
| * - disable hyperthreading a.k.a. hardware multithreading | |
| * (Linux instructions: http://bench.cr.yp.to/supercop.html) | |
| * (Mac OS X instructions: Instruments -> Preferences -> CPUs -> uncheck "Hardware Multi-Threading" | |
| * http://forums.macrumors.com/showthread.php?t=1484684) | |
| * - disable TurboBoost | |
| * (Linux instructions: http://bench.cr.yp.to/supercop.html) | |
| * (Max OS X: use http://www.rugarciap.com/turbo-boost-switcher-for-os-x/) | |
| * - run when the computer is idle (e.g., shut down all other applications, disable network access if possible, ...) | |
| */ | |
| /* On Raspberry Pi, you need to additionally define the macro _RASPBERRY_PI since | |
| * the high-precision cycle count register is not available to user-space programs. | |
| */ | |
| #endif | |
| #ifndef _DS_BENCHMARK_H | |
| #define _DS_BENCHMARK_H | |
| #include <inttypes.h> | |
| #include <stdio.h> | |
| #include <stdlib.h> | |
| #if !defined(_WIN32) | |
| #include <sys/time.h> | |
| #endif | |
| #include <math.h> | |
| #include <time.h> | |
| #if defined(_WIN32) | |
| #include <Windows.h> | |
| int gettimeofday(struct timeval *tp, struct timezone *tzp) { | |
| // Note: some broken versions only have 8 trailing zero's, the correct epoch has 9 trailing zero's | |
| static const uint64_t EPOCH = ((uint64_t) 116444736000000000ULL); | |
| SYSTEMTIME system_time; | |
| FILETIME file_time; | |
| uint64_t time; | |
| GetSystemTime(&system_time); | |
| SystemTimeToFileTime(&system_time, &file_time); | |
| time = ((uint64_t) file_time.dwLowDateTime); | |
| time += ((uint64_t) file_time.dwHighDateTime) << 32; | |
| tp->tv_sec = (long) ((time - EPOCH) / 10000000L); | |
| tp->tv_usec = (long) (system_time.wMilliseconds * 1000); | |
| return 0; | |
| } | |
| #endif | |
| static uint64_t _bench_rdtsc(void) { | |
| #if defined(_WIN32) || defined(_WIN64) | |
| return __rdtsc(); | |
| #elif defined(__i586__) || defined(__amd64__) | |
| uint64_t x; | |
| __asm__ volatile(".byte 0x0f, 0x31" | |
| : "=A"(x)); | |
| return x; | |
| #elif defined(__arm__) && !defined(_RASPBERRY_PI) | |
| /* Use the ARM performance counters. */ | |
| unsigned int value; | |
| /* Read CCNT Register */ | |
| asm volatile("mrc p15, 0, %0, c9, c13, 0\t\n" | |
| : "=r"(value)); | |
| return value; | |
| #else | |
| #define USING_TIME_RATHER_THAN_CYCLES | |
| struct timespec time; | |
| clock_gettime(CLOCK_REALTIME, &time); | |
| return (int64_t)(time.tv_sec * 1e9 + time.tv_nsec); | |
| #endif | |
| } | |
| #if defined(__arm__) && !defined(_RASPBERRY_PI) | |
| static void _bench_init_perfcounters(int32_t do_reset, int32_t enable_divider) { | |
| /* In general enable all counters (including cycle counter) */ | |
| int32_t value = 1; | |
| /* Peform reset */ | |
| if (do_reset) { | |
| value |= 2; /* reset all counters to zero */ | |
| value |= 4; /* reset cycle counter to zero */ | |
| } | |
| if (enable_divider) | |
| value |= 8; /* enable "by 64" divider for CCNT */ | |
| value |= 16; | |
| /* Program the performance-counter control-register */ | |
| asm volatile("mcr p15, 0, %0, c9, c12, 0\t\n" ::"r"(value)); | |
| /* Enable all counters */ | |
| asm volatile("mcr p15, 0, %0, c9, c12, 1\t\n" ::"r"(0x8000000f)); | |
| /* Clear overflows */ | |
| asm volatile("mcr p15, 0, %0, c9, c12, 3\t\n" ::"r"(0x8000000f)); | |
| } | |
| #endif | |
| #define DEFINE_TIMER_VARIABLES \ | |
| volatile uint64_t _bench_cycles_start, _bench_cycles_end; \ | |
| uint64_t _bench_cycles_cumulative = 0; \ | |
| int64_t _bench_cycles_diff; \ | |
| struct timeval _bench_timeval_start, _bench_timeval_end; \ | |
| uint64_t _bench_iterations, _bench_time_cumulative; \ | |
| double _bench_cycles_x, _bench_cycles_mean, _bench_cycles_delta, _bench_cycles_M2, _bench_cycles_stdev; \ | |
| double _bench_time_x, _bench_time_mean, _bench_time_delta, _bench_time_M2, _bench_time_stdev; | |
| #if defined(__arm__) && !defined(_RASPBERRY_PI) | |
| #define INITIALIZE_TIMER \ | |
| _bench_init_perfcounters(1, 0); \ | |
| _bench_iterations = 0; \ | |
| _bench_cycles_mean = 0.0; \ | |
| _bench_cycles_M2 = 0.0; \ | |
| _bench_time_cumulative = 0; \ | |
| _bench_time_mean = 0.0; \ | |
| _bench_time_M2 = 0.0; | |
| #else | |
| #define INITIALIZE_TIMER \ | |
| _bench_iterations = 0; \ | |
| _bench_cycles_mean = 0.0; \ | |
| _bench_cycles_M2 = 0.0; \ | |
| _bench_time_cumulative = 0; \ | |
| _bench_time_mean = 0.0; \ | |
| _bench_time_M2 = 0.0; | |
| #endif | |
| #define START_TIMER \ | |
| gettimeofday(&_bench_timeval_start, NULL); \ | |
| _bench_cycles_start = _bench_rdtsc(); | |
| // Mean and population standard deviation are calculated in an online way using the algorithm in | |
| // http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance#Online_algorithm | |
| #define STOP_TIMER \ | |
| _bench_cycles_end = _bench_rdtsc(); \ | |
| gettimeofday(&_bench_timeval_end, NULL); \ | |
| _bench_iterations += 1; \ | |
| if (_bench_cycles_end < _bench_cycles_start) { \ | |
| _bench_cycles_end += (uint64_t) 1 << 32; \ | |
| } \ | |
| _bench_cycles_diff = _bench_cycles_end; \ | |
| _bench_cycles_diff -= _bench_cycles_start; \ | |
| _bench_cycles_cumulative += _bench_cycles_diff; \ | |
| _bench_cycles_x = (double) (_bench_cycles_diff); \ | |
| _bench_cycles_delta = _bench_cycles_x - _bench_cycles_mean; \ | |
| _bench_cycles_mean += _bench_cycles_delta / (double) _bench_iterations; \ | |
| _bench_cycles_M2 += _bench_cycles_delta * (_bench_cycles_x - _bench_cycles_mean); \ | |
| _bench_time_x = (double) ((_bench_timeval_end.tv_sec * 1000000 + _bench_timeval_end.tv_usec) - (_bench_timeval_start.tv_sec * 1000000 + _bench_timeval_start.tv_usec)); \ | |
| _bench_time_delta = _bench_time_x - _bench_time_mean; \ | |
| _bench_time_mean += _bench_time_delta / (double) _bench_iterations; \ | |
| _bench_time_M2 += _bench_time_delta * (_bench_time_x - _bench_time_mean); \ | |
| _bench_time_cumulative += _bench_time_x; | |
| #define FINALIZE_TIMER \ | |
| if (_bench_iterations == 2) { \ | |
| _bench_cycles_stdev = 0.0; \ | |
| } else { \ | |
| _bench_cycles_stdev = sqrt(_bench_cycles_M2 / (double) _bench_iterations); \ | |
| } \ | |
| if (_bench_iterations == 2) { \ | |
| _bench_time_stdev = 0.0; \ | |
| } else { \ | |
| _bench_time_stdev = sqrt(_bench_time_M2 / (double) _bench_iterations); \ | |
| } | |
| #define PRINT_CURRENT_TIME \ | |
| { \ | |
| char _bench_time_buff[20]; \ | |
| time_t _bench_time_now = time(0); \ | |
| strftime(_bench_time_buff, 20, "%Y-%m-%d %H:%M:%S", localtime(&_bench_time_now)); \ | |
| printf("%s", _bench_time_buff); \ | |
| } | |
| #ifdef USING_TIME_RATHER_THAN_CYCLES | |
| #define HIGH_PREC_HEADER "High-prec time (ns): mean" | |
| #else | |
| #define HIGH_PREC_HEADER "CPU cycles: mean " | |
| #endif | |
| #define PRINT_TIMER_HEADER \ | |
| printf("Started at "); \ | |
| PRINT_CURRENT_TIME \ | |
| printf("\n"); \ | |
| printf("%-30s | %10s | %14s | %15s | %10s | %25s | %10s\n", "Operation ", "Iterations", "Total time (s)", "Time (us): mean", "pop. stdev", HIGH_PREC_HEADER, "pop. stdev"); \ | |
| printf("%-30s | %10s:| %14s:| %15s:| %10s:| %25s:| %10s:\n", "------------------------------", "----------", "--------------", "---------------", "----------", "-------------------------", "----------"); | |
| /* colons are used in above to right-align cell contents in Markdown */ | |
| #define PRINT_TIMER_FOOTER \ | |
| printf("Ended at "); \ | |
| PRINT_CURRENT_TIME \ | |
| printf("\n"); | |
| #define PRINT_TIMER_AVG(op_name) \ | |
| printf("%-30s | %10" PRIu64 " | %14.3f | %15.3f | %10.3f | %25.0f | %10.0f\n", (op_name), _bench_iterations, ((double) _bench_time_cumulative) / 1000000.0, _bench_time_mean, _bench_time_stdev, ((double) _bench_cycles_cumulative) / (double) _bench_iterations, _bench_cycles_stdev); | |
| #define TIME_OPERATION_ITERATIONS(op, op_name, it) \ | |
| { \ | |
| DEFINE_TIMER_VARIABLES \ | |
| INITIALIZE_TIMER \ | |
| for (int i = 0; i < (it); i++) { \ | |
| START_TIMER { op; } \ | |
| STOP_TIMER \ | |
| } \ | |
| FINALIZE_TIMER \ | |
| PRINT_TIMER_AVG(op_name) \ | |
| } | |
| #define TIME_OPERATION_SECONDS(op, op_name, secs) \ | |
| { \ | |
| DEFINE_TIMER_VARIABLES \ | |
| INITIALIZE_TIMER \ | |
| uint64_t _bench_time_goal_usecs = 1000000 * secs; \ | |
| while (_bench_time_cumulative < _bench_time_goal_usecs) { \ | |
| START_TIMER { op; } \ | |
| STOP_TIMER \ | |
| } \ | |
| FINALIZE_TIMER \ | |
| PRINT_TIMER_AVG(op_name) \ | |
| } | |
| #endif |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment