Automatic parallelization in SCA forward subroutine

[clementval]

Description

Currently, the parallelisation is done automatically in the root SCA subroutine/function. When using sea forward along the call graph, some parallization must be done as well. But user must do it explicitly at the moment. This should be done by CLAW.

Original code

!$claw parallelize forward
CALL column%compute_column(nz, q, t)

q = q + sum(t)

Transformed code

  CALL column % compute_column ( nz , q , t , nproma = nproma )
  q = q + sum ( t )

Expected code

CALL column % compute_column ( nz , q , t , nproma = nproma )
!$acc parallel
!$acc loop gang vector collapse(2)
do iter1 = lbound(q, 1), ubound(q, 1)
  do iter2 = lbound(q, 2), ubound(q, 2)
    q(iter1, iter2) = q(iter1, iter2) + sum ( t )
  end do
end do
!$acc end parallel

Execution of clawfc

clawfc --target=gpu --directive=acc -o transformed_code.f90 original_code.f90

Related issues (if any)

• None