how to calculate machine cycle counts for a process

QNX DEVELOPERS Porting
1

hi all

i have a codec ported on QNX IDE. i need to calculate the machine cycles consumed by the codec. i got some info in the application profiler once i implemented post moterm profiling. But i just want something that i can obtain after each frame processed, apart from the entire test case/vector run. Normally incase of some dsp applications, i used something like

                       TIMER_START();
                   
                        ENCODER_PROCESS();

                        cycles = TIMER_STOP();

to obtain the cycle count information. I need similar kind of thing in QNX IDE

please help

regards
mukuntan

2

#include <sys/neutrino.h>
#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/syspage.h>

int main( void )
{
uint64_t cps, cycle1, cycle2, ncycles;
float sec;

/* snap the time */
cycle1=ClockCycles( );

/* do something */
printf("poo\n");

/* snap the time again */
cycle2=ClockCycles( );
ncycles=cycle2-cycle1;
printf("%lld cycles elapsed \n", ncycles);

/* find out how many cycles per second */
cps = SYSPAGE_ENTRY(qtime)->cycles_per_sec;
printf( "This system has %lld cycles/sec.\n",cps );
sec=(float)ncycles/cps;
printf("The cycles in seconds is %f \n",sec);

return EXIT_SUCCESS;

}

3

Hello Mukuntan,
Consider using ClockCycles()

qnx.com/developers/docs/6.3. … ycles.html

Regards,
Yuriy

4

hi all

thank for ur reply. actually i tried using clockcycles() earlier. but i was getting some inconsistent results. the values obtained are not similar everytime(for example im getting different value while debugging and altogether a different value while running ). is there anything to do with the target processor. because my target is dual core processor. or i need to introduce few more inbuilt APIs before while using clockcycles().

5

ClockCycles gives the current time, hence if the process is performaing at different speed for whatever reason you will get different value. Dual core can be nasty because ClockCycles gets a value from a counter that could be different in each core. If the process/thread migrate from core to core the result could be bogus.

Maybe clock() is what you are looking for?

6

since mine is a dual core processor, i got to know that i need to create a thread and it should be set with high priority and i need to put my API there and then i need to use clockcycles(). since i was new to RTOS and QNX. could u please suggest how to proceed

7

If you wish to use ClockCycles() on an SMP machine, you must use the following call to “lock” the thread to a single CPU:
ThreadCtl(_NTO_TCTL_RUNMASK, …)

See
qnx.com/developers/docs/6.3. … ycles.html

8

that’s not 100% true, many intel processor have their counter synchronised. Even better on the later model I believe the counter frequency won’t be affected by power saving throttling of the processor frequency.

9

Even if it locked to a single cpu, the thread still can be pre-empted. so will the value return by clockcycles() make sense?

10

If you want to measure CPU time used by a thread you can do it manually with pidin ti. If you want to do it programatically get the source of pidin and check out how it does it.

11

I am not quite sure why ClockCycles() is required here.

A possible solution:

gimme_clock(void)
{
uint64_t time_internal;
ClockTime(CLOCK_MONOTONIC, NULL, &time_internal);
return time_internal;
}

…………
now= gimme_clock();