Hi Everyone!!! I've start parallelizing my CFD code. It spends a lot of work in a loop for a finite difference computation. I decided to use openmp but I just got poor performance. I can't understand what's the matter with my implementation. I've posted my prototype code below.
program my_first_openmp
use omp_lib
integer,parameter:: nx = 1000,ny=1000
integer,parameter: dp = kind(1.0d0)
real(dp),allocatable,dimension(:,:):: u,dudx,dudy
real:: t1,t2
call OMP_SET_NUM_THREADS(4)
allocate(u(nx,ny))
allocate(dudx(nx,ny))
allocate(dudy(nx,ny))
u = 3.141592654_dp
dudx = 0.0_dp
dudy = 0.0_dp
call cpu_time(t1)
call calc_dif(u,dudx,dudy)
call cpu_time(t2)
print*,' t2-t1 = ',t2-t1
contains
subroutine(u,dudx,dudy)
implicit none
integer:: i,j
real(dp),dimension(nx,ny),intent(in):: u
real(dp),dimension(nx,ny),intent(out):: dudx,dudy
!$omp parallel do
do j = 1,ny
do i = 1,nx
if(i==1) then
dudx(i,j) = u(i+1,j)-u(i,j)
elseif(i==nx) then
dudx(i,j) = u(i,j)-u(i-1,j)
else
dudx(i,j) = u(i+1,j)-u(i-1,j)
endif
if(j==1) then
dudy(i,j) = u(i,j+1)-u(i,j)
elseif(j==ny) then
dudy(i,j) = u(i,j)-u(i,j-1)
else
dudy(i,j) = u(i,j+1)-u(i,j-1)
endif
enddo
enddo
return
end subroutine calc_diff
end program my_first_openmp
