Re: [Starpu-devel] Worker Binding Problem

[Samuel Thibault <samuel.thibault@inria.fr>]

Berenger Bramas, le Wed 09 Sep 2015 13:39:27 +0200, a écrit :
> So I still cannot figure it out, I was also thinking that it might come 
> from the binding or the stack but both look OK.

Did you check in top (by increasing the test duration to
have the time to see anything) that there is no other thread
eating CPU time concurrently?  Perhaps there is also some
speedstep/cpufreq/dyntick/whatever which happens to trigger in one case
but not the other for whatever reason?

Did you also check that the assembly code of the work function gets the
same in both cases?

You could also make sure to get rid of any memory issue by putting the
computation in registers, avoiding the optimization thanks to an empty
asm statement:

void work(){
    size_t Nb = 100000000;
    size_t idx = 0;
    while(idx != Nb){
        idx += 1;
        asm(""::"r" (idx));
    }
}

Samuel