[Patch] Fortran: Add 'omp scan' support of OpenMP 5.0

[Tobias Burnus]

On 09.12.20 12:36, Thomas Schwinge wrote:
> I'm confirming that it seems to work (that is, doesn't seem to cause any
> obvious interference); OK to verify/document that as in the attached
> "Add 'gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90'"?

I don't think the testcase is useful, but I wouldn't veto it.

However, I think the comment change is completely misleading.
And additionally the testcase misses the point in terms what
happens internally.

Namely:
   !$acc update ... if_present
gets translated into an
   gfc_code->op == EXEC_OACC_UPDATE
   gfc_code->code->ext.omp_clauses->if_present = true

While
   !$omp scan ...
gets translated into
   gfc_code->op == EXEC_OMP_SCAN
which for a short time sets:
   gfc_code->code->ext.omp_clauses->if_present = true

(And those are different gfc_code variables.)

If we worry about this, we should also add a testcase that for
  !$acc update host(a)
  !$acc update self(b) if_present
checking that the first 'acc' does not have if_present set.


That it is also almost impossible to generate a compilable
testcase – due to the restrictions of 'omp scan' but also
because '!$acc update' cannot appear in an 'omp do' loop
— comes on top of this but focusing on the testcase is
really a red herring.

Tobias

PS: Unsetting 'if_present' for OMP_SCAN makes sense as otherwise
trans-openmp.c creates an OMP_CLAUSE_IF_PRESENT clause,
which may cause problems in the middle end. But that's
completely independent of -fopenacc and !$acc.

-----------------
Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München / Germany
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Alexander Walter