[Patch, fortran] PR66549 [5/6 regression] ICE with OMP PARALLEL

[Mikael Morin <mikael dot morin at sfr dot fr>]

Hello,

I'm proposing here a fix for an OpenMP ICE regression introduced by me
at http://gcc.gnu.org/r221586

That revision changed the order in which procedures are resolved, making
it possible for a procedure to be resolved from within an OpenMP
construct body.  As the OpenMP constructs set some global state, the
procedure's do loop were registered as OpenMP-managed loops.

The attached patch clears the OpenMP state upon gfc_resolve entry
and restores it upon return.  I don't know the OpenMP code very well,
but it seems reasonable to me to do that.

Regression-tested on x86_64-linux.  OK for trunk/5 ?

Mikael

Attachment: pr66549_v1.1.CL
Description: Text document

diff --git a/gcc/fortran/resolve.c b/gcc/fortran/resolve.c
index e615cc6..23bee8f 100644
--- a/gcc/fortran/resolve.c
+++ b/gcc/fortran/resolve.c
@@ -2384,14 +2384,11 @@ resolve_global_procedure (gfc_symbol *sym, locus *where,
       if (!gsym->ns->resolved)
 	{
 	  gfc_dt_list *old_dt_list;
-	  struct gfc_omp_saved_state old_omp_state;
 
 	  /* Stash away derived types so that the backend_decls do not
 	     get mixed up.  */
 	  old_dt_list = gfc_derived_types;
 	  gfc_derived_types = NULL;
-	  /* And stash away openmp state.  */
-	  gfc_omp_save_and_clear_state (&old_omp_state);
 
 	  gfc_resolve (gsym->ns);
 
@@ -2401,8 +2398,6 @@ resolve_global_procedure (gfc_symbol *sym, locus *where,
 
 	  /* Restore the derived types of this namespace.  */
 	  gfc_derived_types = old_dt_list;
-	  /* And openmp state.  */
-	  gfc_omp_restore_state (&old_omp_state);
 	}
 
       /* Make sure that translation for the gsymbol occurs before
@@ -15071,6 +15066,7 @@ gfc_resolve (gfc_namespace *ns)
 {
   gfc_namespace *old_ns;
   code_stack *old_cs_base;
+  struct gfc_omp_saved_state old_omp_state;
 
   if (ns->resolved)
     return;
@@ -15079,6 +15075,11 @@ gfc_resolve (gfc_namespace *ns)
   old_ns = gfc_current_ns;
   old_cs_base = cs_base;
 
+  /* As gfc_resolve can be called during resolution of an OpenMP construct
+     body, we should clear any state associated to it, so that say NS's
+     DO loops are not interpreted as OpenMP loops.  */
+  gfc_omp_save_and_clear_state (&old_omp_state);
+
   resolve_types (ns);
   component_assignment_level = 0;
   resolve_codes (ns);
@@ -15088,4 +15089,6 @@ gfc_resolve (gfc_namespace *ns)
   ns->resolved = 1;
 
   gfc_run_passes (ns);
+
+  gfc_omp_restore_state (&old_omp_state);
 }

! { dg-do compile }
! { dg-additional-options "-fdump-tree-original" }
!
! PR fortran/66549
! The resolution of CVN in the middle CLWF's OpenMP construct was
! making the DO loop (wrongly) interpreted as an OpenMP-managed loop, leading
! to an ICE.
!
! Contributed by Andrew Benson <abensonca@gmail.com>.

module smfa
  type :: sgc
   contains
     procedure :: sla => sa
  end type sgc
  class(sgc), pointer :: sg_
  double precision, allocatable, dimension(:) :: vni 
contains
  double precision function sa(self,i)
    class(sgc), intent(in   ) :: self
  end function sa
  subroutine cvn(sg_,vn)
    class(sgc), intent(inout) :: sg_
    double precision, intent(  out), dimension(:) :: vn
    integer :: i
    do i=1,2
       vn(i)= sg_%sla(i)
    end do
  end subroutine cvn
  subroutine clwf()
    !$omp parallel
    call cvn(sg_,vni)
    !$omp end parallel
  end subroutine clwf
end module smfa

! { dg-final { scan-tree-dump-times "#pragma\\s+omp\\s+parallel\\n" 1 "original" } }