我正在尝试将GPU与CPU性能进行比较。对于NVIDIA GPU，我一直在使用cudaEvent_t类型来获得非常精确的计时。
对于CPU，我一直在使用以下代码：
// Timers
clock_t start, stop;
float elapsedTime = 0;
// Capture the start time
start = clock();
// Do something here
.......
// Capture the stop time
stop = clock();
// Retrieve time elapsed in milliseconds
elapsedTime = (float)(stop - start) / (float)CLOCKS_PER_SEC * 1000.0f;
显然，只要你在几秒钟内计算，这段代码就是好的。而且，结果有时候很奇怪。
有没有人知道在Linux中创建高分辨率计时器的某种方法？
答案 0 :(得分：42)
查看clock_gettime，它是高分辨率计时器的POSIX接口。
如果在阅读了联机帮助页后，您仍然想知道CLOCK_REALTIME和CLOCK_MONOTONIC之间的区别，请参阅Difference between CLOCK_REALTIME and CLOCK_MONOTONIC?
有关完整示例，请参阅以下页面：http://www.guyrutenberg.com/2007/09/22/profiling-code-using-clock_gettime/
#include <iostream>
#include <time.h>
using namespace std;
timespec diff(timespec start, timespec end);
int main()
{
timespec time1, time2;
int temp;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &time1);
for (int i = 0; i< 242000000; i++)
temp+=temp;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &time2);
cout<<diff(time1,time2).tv_sec<<":"<<diff(time1,time2).tv_nsec<<endl;
return 0;
}
timespec diff(timespec start, timespec end)
{
timespec temp;
if ((end.tv_nsec-start.tv_nsec)<0) {
temp.tv_sec = end.tv_sec-start.tv_sec-1;
temp.tv_nsec = 1000000000+end.tv_nsec-start.tv_nsec;
} else {
temp.tv_sec = end.tv_sec-start.tv_sec;
temp.tv_nsec = end.tv_nsec-start.tv_nsec;
}
return temp;
}
答案 1 :(得分：18)
总结目前提供的信息，这些是典型应用所需的两个功能。
#include <time.h>
// call this function to start a nanosecond-resolution timer
struct timespec timer_start(){
struct timespec start_time;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &start_time);
return start_time;
}
// call this function to end a timer, returning nanoseconds elapsed as a long
long timer_end(struct timespec start_time){
struct timespec end_time;
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &end_time);
long diffInNanos = (end_time.tv_sec - start_time.tv_sec) * (long)1e9 + (end_time.tv_nsec - start_time.tv_nsec);
return diffInNanos;
}
以下是如何使用它们计算计算输入列表方差所需时间的示例。
struct timespec vartime = timer_start(); // begin a timer called 'vartime'
double variance = var(input, MAXLEN); // perform the task we want to time
long time_elapsed_nanos = timer_end(vartime);
printf("Variance = %f, Time taken (nanoseconds): %ld\n", variance, time_elapsed_nanos);
答案 2 :(得分：1)
struct timespec t;
clock_gettime(CLOCK_REALTIME, &t);
还有CLOCK_REALTIME_HR，但我不确定它是否有任何区别..
答案 3 :(得分：1)
您是否对壁挂时间（实际经过的时间）或周期数（周期数）感兴趣？在第一种情况下，您应该使用类似gettimeofday的内容。
最高分辨率计时器使用RDTSC x86汇编指令。但是，这会测量时钟周期，因此您应该确保禁用省电模式。
TSC的维基页面提供了一些示例：http://en.wikipedia.org/wiki/Time_Stamp_Counter
答案 4 :(得分：0)
epoll实现： https://github.com/ielife/simple-timer-for-c-language
像这样使用：
timer_server_handle_t *timer_handle = timer_server_init(1024);
if (NULL == timer_handle) {
fprintf(stderr, "timer_server_init failed\n");
return -1;
}
ctimer timer1;
timer1.count_ = 3;
timer1.timer_internal_ = 0.5;
timer1.timer_cb_ = timer_cb1;
int *user_data1 = (int *)malloc(sizeof(int));
*user_data1 = 100;
timer1.user_data_ = user_data1;
timer_server_addtimer(timer_handle, &timer1);
ctimer timer2;
timer2.count_ = -1;
timer2.timer_internal_ = 0.5;
timer2.timer_cb_ = timer_cb2;
int *user_data2 = (int *)malloc(sizeof(int));
*user_data2 = 10;
timer2.user_data_ = user_data2;
timer_server_addtimer(timer_handle, &timer2);
sleep(10);
timer_server_deltimer(timer_handle, timer1.fd);
timer_server_deltimer(timer_handle, timer2.fd);
timer_server_uninit(timer_handle);
答案 5 :(得分：0)
阅读此线程后，我开始针对c ++ 11的chrono测试clock_gettime的代码，它们似乎不匹配。
它们之间有很大的距离！
std :: chrono :: seconds（1）似乎等于 clock_gettime
的〜30,000#include <ctime>
#include <cstdlib>
#include <cstring>
#include <iostream>
#include <thread>
#include <chrono>
#include <iomanip>
#include <vector>
timespec diff(timespec start, timespec end);
timespec get_cpu_now_time();
std::vector<timespec> get_start_end_pairs();
void output_deltas(const std::vector<timespec> &start_end_pairs);
//=============================================================
int main()
{
std::cout << "Hello waiter" << std::endl; // flush is intentional
std::vector<timespec> start_end_pairs = get_start_end_pairs();
output_deltas(start_end_pairs);
return EXIT_SUCCESS;
}
//=============================================================
std::vector<timespec> get_start_end_pairs()
{
std::vector<timespec> start_end_pairs;
for (int i = 0; i < 20; ++i)
{
start_end_pairs.push_back(get_cpu_now_time());
std::this_thread::sleep_for(std::chrono::seconds(1));
start_end_pairs.push_back(get_cpu_now_time());
}
return start_end_pairs;
}
//=============================================================
void output_deltas(const std::vector<timespec> &start_end_pairs)
{
for (auto it_start = start_end_pairs.begin(); it_start != start_end_pairs.end(); it_start += 2)
{
auto it_end = it_start + 1;
auto delta = diff(*it_start, *it_end);
std::cout
<< "Waited ("
<< delta.tv_sec
<< "\ts\t"
<< std::setw(9)
<< std::setfill('0')
<< delta.tv_nsec
<< "\tns)"
<< std::endl;
}
}
//=============================================================
timespec diff(timespec start, timespec end)
{
timespec temp;
temp.tv_sec = end.tv_sec-start.tv_sec;
temp.tv_nsec = end.tv_nsec-start.tv_nsec;
if (temp.tv_nsec < 0) {
++temp.tv_sec;
temp.tv_nsec += 1000000000;
}
return temp;
}
//=============================================================
timespec get_cpu_now_time()
{
timespec now_time;
memset(&now_time, 0, sizeof(timespec));
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &now_time);
return now_time;
}
输出：
Waited (0 s 000064802 ns)
Waited (0 s 000028512 ns)
Waited (0 s 000030664 ns)
Waited (0 s 000041233 ns)
Waited (0 s 000013458 ns)
Waited (0 s 000024068 ns)
Waited (0 s 000027591 ns)
Waited (0 s 000028148 ns)
Waited (0 s 000033783 ns)
Waited (0 s 000022382 ns)
Waited (0 s 000027866 ns)
Waited (0 s 000028085 ns)
Waited (0 s 000028012 ns)
Waited (0 s 000028172 ns)
Waited (0 s 000022121 ns)
Waited (0 s 000052940 ns)
Waited (0 s 000032138 ns)
Waited (0 s 000028082 ns)
Waited (0 s 000034486 ns)
Waited (0 s 000018875 ns)
答案 6 :(得分：-1)