OpenMP

This allows to write valid OpenMP instructions and are handled in two steps:

• in the grammar, in order to parse the omp pragams
• as a Pyccel header. Therefor, you can import and call OpenMP functions as you would do it in Fortran or C.

Examples

See script.

Now, run the command:

pyccel tests/scripts/openmp/core/ex1.py --openmp
export OMP_NUM_THREADS=4

Executing the associated binary gives:

> threads number :            1
> maximum available threads :            4
> thread  id :            0
> thread  id :            3
> thread  id :            1
> thread  id :            2

See more OpenMP examples.

Matrix multiplication

Let’s take a look at the file tests/examples/openmp/matrix_product.py, listed below

from numpy import zeros

n = 500
m = 700
p = 500

a = zeros((n,m), double)
b = zeros((m,p), double)
c = zeros((n,p), double)

#$ omp parallel
#$ omp do schedule(runtime)
for i in range(0, n):
    for j in range(0, m):
        a[i,j] = i-j
#$ omp end do nowait

#$ omp do schedule(runtime)
for i in range(0, m):
    for j in range(0, p):
        b[i,j] = i+j
#$ omp end do nowait

#$ omp do schedule(runtime)
for i in range(0, n):
    for j in range(0, p):
        for k in range(0, p):
            c[i,j] = c[i,j] + a[i,k]*b[k,j]
#$ omp end do
#$ omp end parallel

Now, run the command:

pyccel tests/examples/openmp/matrix_product.py --compiler="gfortran" --openmp

This will parse the Python file, generate the corresponding Fortran file and compile it.

Note

Openmp is activated using the flag –openmp in the command line.

The generated Fortran code is

program main
use omp_lib
implicit none
real(kind=8), allocatable :: a (:, :)
real(kind=8), allocatable :: c (:, :)
real(kind=8), allocatable :: b (:, :)
integer :: i
integer :: k
integer :: j
integer :: m
integer :: n
integer :: p

!
n = 500
m = 700
p = 500
allocate(a(0:n-1, 0:m-1)) ; a = 0
allocate(b(0:m-1, 0:p-1)) ; b = 0
allocate(c(0:n-1, 0:p-1)) ; c = 0
!$omp parallel
!$omp do schedule(runtime)
do i = 0, n - 1, 1
  do j = 0, m - 1, 1
    a(i, j) = i - j
  end do
end do
!$omp end do  nowait
!$omp do schedule(runtime)
do i = 0, m - 1, 1
  do j = 0, p - 1, 1
    b(i, j) = i + j
  end do
end do
!$omp end do  nowait
!$omp do schedule(runtime)
do i = 0, n - 1, 1
  do j = 0, p - 1, 1
    do k = 0, p - 1, 1
      c(i, j) = a(i, k)*b(k, j) + c(i, j)
    end do
  end do
end do
!$omp end do
!$omp end parallel

end