[Bug fortran/27916] New: Problem with allocatable arrays inside OpenMP do loop

benjamin dot reveille at gmail dot com gcc-bugzilla@gcc.gnu.org
Tue Jun 6 12:01:00 GMT 2006


I've stumbled on a gfortran "Fortran runtime error" when using allocatable
arrays inside OpenMP PARALLEL DO LOOPS

Consider the folowing reduced testcase.

 >> cat allocate_in_loop.f90
program allocate_in_loop
  use omp_lib, only: omp_get_num_threads,omp_get_thread_num
  implicit none
  integer :: numthreads,n,it,i
  integer, dimension(:), allocatable :: int_array
  !
!$OMP PARALLEL DEFAULT(shared) NUM_THREADS(4)
   numthreads=omp_get_num_threads()
!$OMP DO PRIVATE(n,it,int_array,i) SCHEDULE(static)
  do n=1,numthreads
     it=omp_get_thread_num()
     allocate(int_array(3))
        do i=1,3
           int_array(i)=(it)*100+i
           write(*,'(a,i2,a,i1,a,i4)') 'thread :',it,' -->int_array(', &
                                       i,')=',int_array(i)
        end do
     deallocate(int_array)
  end do
!$OMP END DO
!
!$OMP END PARALLEL
end program allocate_in_loop


According to what I read in the OpenMP 2.5 Specification - in sections 2.8.3.3
page 75 - the usage of the allocatable array is correct since I allocate it and
deallocate inside the OMP loop and it is not allocated upon entering the OMP DO
loop.

Gfortran compile fine with
>> gfortran -fopenmp allocate_in_loop.f90
But fails on execution with: "Fortran runtime error: Attempting to allocate
already allocated array."


intel 9.1, pgi 6.1 and xl do fine, the output being for example...
 >> ./a.out
thread : 0 --> int_array(1)=   1
thread : 0 --> int_array(2)=   2
thread : 0 --> int_array(3)=   3
thread : 1 --> int_array(1)= 101
thread : 1 --> int_array(2)= 102
thread : 1 --> int_array(3)= 103
thread : 2 --> int_array(1)= 201
thread : 2 --> int_array(2)= 202
thread : 2 --> int_array(3)= 203
thread : 3 --> int_array(1)= 301
thread : 3 --> int_array(2)= 302
thread : 3 --> int_array(3)= 303

Good bug squashing

Benjamin


-- 
           Summary: Problem with allocatable arrays inside OpenMP do loop
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: benjamin dot reveille at gmail dot com
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27916



More information about the Gcc-bugs mailing list