why openmp is slower?

[pt663359]

Hi there,

I am trying some parallelisation using openMP but to my surprize all I get is slower execution times.
See the example below: with openmp is about 4 time slower.

Any ideea what can be helpfull here? Thanks.

!$OMP parallel shared(V,V1,V2) , private(k)
!$OMP do schedule (static)

do k=1,10000000
do i =1,Nmax
! j=OMP_GET_THREAD_NUM()
! print*,j,'i=',i
V (i)=V1(i)*V2(i)
enddo
enddo
!$OMP end do
!$OMP END parallel

!master:~/debug$ ifc -O3 -tpp7 -xW -openmp t1_omp.f90
!time a.out
!real 5m8.800s
!user 19m8.230s
!sys 0m3.000s !with openmp
!
!master:~/debug$ ifc -O3 -tpp7 -xW t1_omp.f90
!time a.out
!
!real 1m59.402s
!user 1m59.160s
!sys 0m0.000s !without openmp

[TimP]

If your program is so simple, more than embarrassingly parallel, such that the compiler can shortcut your loops, parallelizing it may force it to perform several times as much work.

[pt663359]

OK, thanks for replay, I think your explanation make sense, actually this is what I suspected: In the case of loops like:
do i=1,N
do j=1,M
do few things
do one more thing :)
endo
endo

ifc alone is already doing a very good job and overloading openmp stuff may actually slow down things.