CPU usage of nanosleep on Win32 when called repeatedly

[Michael Goffioul]

Hi,

In the context of octave, I need to have a loop that sleeps for about
100ms then perform some action. Most of the time, the action to
perform is empty so the loop is mainly sleeping. I don't expect such
loop to be CPU-consuming. To build this loop in octave, I'm using
octave_usleep, which calls gnulib nanosleep. I've noticed that such
loop on Win32 is consuming a significant amount of CPU, about 5-6%,
which is suprizing.

I've built a small example (see below) to compare the CPU usage of
gnulib nanosleep and Win32 Sleep (the delay being 100ms, Sleep is good
enough). When running it, I see a steady CPU usage of 5-6% when using
nanosleep, and 0% when using Win32 Sleep.

Is this something that can be fixed in gnulib? Or is this something I
have to live with?
Note that the main octave developer is not really willing to have a
special case with #ifdef for Win32 in octave's code as this is the
specific reason for octave moving to gnulib.

Thanks,
Michael.


#include <config.h>
#include <time.h>
#define WIN32_LEAN_AND_MEAN
#include <windows.h>
#include <stdio.h>

int main (int argc, char **argv)
{
  struct timespec t1, rem;
  int i;

  t1.tv_sec = 0;
  t1.tv_nsec = 100000000;

  printf ("testing gnulib::nanosleep...\n");
  for (i = 0; i < 100; i++)
    nanosleep (&t1, &rem);

  printf ("testing Win32 Sleep...\n");
  for (i = 0; i < 100; i++)
    Sleep (100);

  return 0;
}