C/C++ Millisecond timers?

[Demirug]

If you use RDTSC make sure you use SetProcessAffinityMask too. If you don't do this on a multiprocessor system you will run in big trouble.

 

[Rolf N]

If you use RDTSC make sure you use SetProcessAffinityMask too. If you don't do this on a multiprocessor system you will run in big trouble.

You mean Win32 can actually move a running thread to another processor at any unforeseen time?

 

[Demirug]

If you use RDTSC make sure you use SetProcessAffinityMask too. If you don't do this on a multiprocessor system you will run in big trouble.

You mean Win32 can actually move a running thread to another processor at any unforeseen time?

Not at any time but after each thread switch your thread can run on another processor. If you have a thread switch between your start and end time the RDTSC method can give you very strange results like a negative execution time. I have see this.

 

[psurge]

On UNIX systems you can also do the following :

1.) Fairly portable, fall back to gettimeofday() if it's unavailable :

#include <time.h>
...
struct timespec ts;
clock_gettime(CLOCK_REALTIME, &ts);

2.) Solaris also supports :

#include <time.h>
#include <sys/time.h>
...
/* number of nanoseconds since some arbitrary point in the past,
    unaffected by changes to the system time */
struct timespec ts;
clock_gettime(CLOCK_HIGHRES, &ts);
hrtime_t t;
t = gethrtime();

2.) More recent versions of Linux support the following clocks as well:

/*
       CLOCK_MONOTONIC
              Clock  that  cannot  be  set and represents monotonic time since
              some unspecified starting point.

       CLOCK_PROCESS_CPUTIME_ID
              High-resolution per-process timer from the CPU.

       CLOCK_THREAD_CPUTIME_ID
              Thread-specific CPU-time clock.
*/
clock_gettime(CLOCK_MONOTONIC, &ts);
clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts);
...