[maz]

Hi,

Here is an OpenMP implementation of a Laplace solver (a parallel Fortran code). I can get a good speedup on up to 4 processors, but it does not improve when I use more processors.
Could you please let me know how I can get a better speedup by using more processors?

program lpomp
integer imax,jmax,im1,im2,jm1,jm2,it,itmax
parameter (imax=4001,jmax=4001)
parameter (im1=imax-1,im2=imax-2,jm1=jmax-1,jm2=jmax-2)
parameter (itmax=200)
real*8 u(imax,jmax),du(imax,jmax),umax,dumax,tol
parameter (umax=10.0,tol=1.0e-6)
call omp_set_num_threads(2)
!$OMP PARALLEL DEFAULT(SHARED) PRIVATE(i,j)
!$OMP DO
do j=1,jmax
do i=1,imax-1
u(i,j)=0.0
du(i,j)=0.0
enddo
u(imax,j)=umax
enddo
!$OMP END DO
!$OMP END PARALLEL
it=1
dumax=1.0
!$OMP PARALLEL DEFAULT(SHARED) PRIVATE(i,j)
! Main computation loop
do it=1,itmax
dumax=0.0
!$OMP DO REDUCTION (max:dumax)
do j=2,jm1
do i=2,im1
du(i,j)=0.25*(u(i-1,j)+u(i+1,j)+u(i,j-1)+u(i,j+1))-u(i,j)
dumax=max(dumax,abs(du(i,j)))
enddo
enddo
!$OMP END DO
!$OMP DO
do j=2,jm1
do i=2,im1
u(i,j)=u(i,j)+du(i,j)
enddo
enddo
!$OMP END DO
enddo
!$OMP END PARALLEL
stop
end

Compile: ifort -openmp -parallel Laplace.for

Thanks!

[DoHander]

Comment (or erase) this line:

call omp_set_num_threads(2)

recompile your code and let me know if you see any improvement.
You could also try to enable all optimizations by using for compilation:

ifort -O3 -openmp -parallel Laplace.for

Do

[maz]

Hi,
I commented the line, but did not see any improvement.
Thanks!

[G.Y.Liang]

I can get a good speedup on up to 4 processors, but it does not improve when I use more processors.

Usually I use ifort -fast /Qopenmp filename.f90. I don't know why ifort -fast -openmp cann't be used on my computer(Windows serve OS）

For fixed grids, of course, the performance will be slow down if you use more processors. You can fix you processors first, then increase grids to test the speedup.

[shirazbj]

I　ｃａｎ　ｃｏｍｐｉｌｅ　ｉｔ　ｗｉｔｈ:

ｇｆｏｒｔｒａｎ　-fopenmp laplace.f90

but when I run it, there is error. Don't kow why?

Thx

[Ravindra Shende]

Why are you using du(i,j) while you can directly do u(i,j) = 0.25*(u(i-,j)+u(i+1,j)+u(i,j-1)+u(i,j+1)). This will increase the speed of serial code itself.