Loop Collapse

Loop Collapse in OpenMP 3.0

The COLLAPSE clause can appear on a do/for loop directive. It is used to parallelize perfectly nested loops without using nested parallelism. The COLLAPSE clause takes a constant positive integer expression as parameter. The parameter specifies how many loops should be collapsed by the compiler into one loop before parallelizing the resulting loop.

The COLLAPSE clause is described in Section 2.5.1 (p. 39) of the OpenMP specification.

Example (Fortran):

       program test
      use omp_lib

      integer, parameter :: N = 10
      integer i, j, x

      call omp_set_dynamic (.false.)
      call omp_set_num_threads (4)

      x = 10

!$OMP PARALLEL SHARED(x) PRIVATE(i, j)
!$OMP DO COLLAPSE(2)
      do i = 1, N
         do j = 1, N
!$OMP CRITICAL
            x = x + 1
!$OMP END CRITICAL
         end do
      end do
!$OMP END PARALLEL

      print *, "x = ", x
      end


% f90 -xopenmp -xO3 test.f
% a.out
 x =  110

Example (C/C++):

#include <stdio.h>
#include <omp.h>

#define N 10

int main(void)
{
  int i, j, x;

  omp_set_dynamic(0);
  omp_set_num_threads(4);

  x = 10;

  #pragma omp parallel shared(x) private(i, j)
  {
     #pragma omp for collapse(2)
     for (i = 0; i < N; i = i + 2)
     {
        for (j = 0; j < N; j = j + 2)
        {
           #pragma omp atomic
           x++;
        }
     }
  }

  printf ("x = %d\n", x);
}


% cc -xopenmp -xO3 test.c
% a.out
x = 35

OpenMP 3.0 Features in Express 7.08

Loop Collapse

Loop Collapse in OpenMP 3.0

Labels

Comments (1)

crow34 says: