PGI User Forum

[Ansa]

Hi.
I don't know if it's normal, I wrote a little program to test OpenMp in
a Linux machine (RH 7.1 2.96-98, Intel Xeon 2.8GHz, two CPUs).

program testeOpenMP
implicit none

integer,dimension(1000,2) :: board
integer :: i,j,k
integer::omp_get_thread_num

call omp_set_dynamic(.FALSE.)
call omp_set_num_threads(2)
!$OMP PARALLEL DO PRIVATE(j)
do i=1,1000
j = omp_get_thread_num()+ 1
board(i,j)=i+j
enddo

read(*,*) k

END

Then I made de executable.

pgf90 -O2 -mp -o testeOpenMP testeOpenMP.f90

Well, when I use top command (program is waiting the keyboard
input (read(*,*))), its shows me 3 processes, and one eating CPU without doing anything useful, at least for me. In a IRIX machine this
don't occur.


CPU0 states: 0.1% user, 0.3% system, 0.0% nice, 99.1% idle
CPU1 states: 100.0% user, 0.0% system, 0.0% nice, 0.0% idle
Mem: 2061776K av, 1979060K used, 82716K free, 0K shrd, 129864K
buff
Swap: 8193104K av, 1381520K used, 6811584K free 1445708K
cached

PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND
14608 s099 16 0 692 692 496 R 99.9 0.0 49:41 testeOpenMP
14606 s099 9 0 692 692 496 S 0.0 0.0 0:00 testeOpenMP
14607 s099 9 0 692 692 496 R 0.0 0.0 0:00 testeOpenMP



Is it normal? Am I doing anything wrong?
Thanks, bye.
Anselmo

[mkcolg]

Hi Anselmo,

This is normal behavior. Our OpenMP runtime only creates each thread once and then attached and detaches to the threads as needed. When not in use, the threads are in a 'sched_yield' state, meaning that while they may be using CPU time, if another process needs that CPU, the thread will yield.

- Mat

[Ansa]

Hi Mat.
First of all I would like to thank you for your attention. I read in another forum (Programming and Compiling-OpenMP high system usage) a similar problem and, in the last message, you suggested to set the number of threads to one after de parallel region. I did this and the top command shows me now only two threads sleeeping.

PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME CMD
27248 s099 9 0 736 732 528 S 0.0 0.0 0:10 testeOpenMP
27249 s099 8 0 736 732 528 S 0.0 0.0 0:00 testeOpenMP


As my application will have few parallel regions, I think that thil will solve my concern.
Thank you again.
Bye
Anselmo