Question

When using the following example code, the tv_nsec value, I presume is looping round because it’s only an long;

#include <iostream>
using namespace std;
#include <time.h>

int main(int argc, char *argv[]) {
  timespec time1, time2;
  timespec timens = {0};
  timens.tv_sec = 1;

  clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &time1);
  nanosleep(&timens, (struct timespec *)NULL);
  clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &time2);

  cout << time1.tv_nsec << " : " << time2.tv_nsec << endl;
  cout << (time2.tv_nsec-time1.tv_nsec) << endl;
  return 0;
}

$ ./microtime 
2754095 : 2766801
12706

The difference between the start and end time is 12706ns (12.7us), it didn’t take 12us to sleep for 1 second, I’m reasonably sure of that! So what has happened here, has tv_nsec looped round? Assuming it’s an singed long to give a smaller range of positive values (0 to 2147483647) instead of an unsigned long (0 to 4294967295), 2147483647 nanoseconds is still just over 2 full seconds!

If I change the code to the following I get a seemingly invalid output still;

timens.tv_sec = 0;
timens.tv_nsec = 100000;

$ ./microtime
2743552 : 2754327
10775

The sleep duration was set for 100,000ns but according to that output the program has slept for 10,000ns. I know that is a small amount of time but I’m trying to show here that whether it is a long time like 1 second or a short time like 100us, there seems to be no accuracy here, or anywhere in between these values. Am I doing something wrong, have I implemented this incorrectly somehow? Why can’t I get the actual duration of the event in between the calls to clock_gettime()?

Many thanks for your time.

Update

@Oli;
I can test your suggestion with the following code but it doesn’t seem to work:

int main(int argc, char *argv[]) {
  timespec time1, time2;
  timespec timens = {0};
  timens.tv_sec = 1;
  timens.tv_nsec = 0;

  clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &time1);
  nanosleep(&timens, (struct timespec *)NULL);
  clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &time2);

  cout <<time1.tv_sec<<"s:"<<time1.tv_nsec<<"ns : "<<
         time2.tv_sec<<"s:"<<time2.tv_nsec<<"ns"<<endl;
  cout <<(time2.tv_nsec-time1.tv_nsec)<<endl;
  return 0;
}

$ ./microtime
0s:2801478ns : 0s:2813732ns
12254

Answer 1

You also need to take the tv_sec field into account when calculating the time interval.

UPDATE

Also, CLOCK_PROCESS_CPUTIME_ID measures CPU time used for this process. If you yield with a nanosleep, this timer won’t increment. Try e.g. CLOCK_MONOTONIC instead.