Hi!
On 2020-12-09T12:06:21+0100, Tobias Burnus <tob...@codesourcery.com> wrote:
> On 08.12.20 13:30, Jakub Jelinek wrote:
>> On Tue, Dec 08, 2020 at 01:13:07PM +0100, Tobias Burnus wrote:
>>> + case EXEC_OMP_SCAN:
>>> + /* Flag is only used to checking, hence, it is unset afterwards. */
>>> + if (!code->ext.omp_clauses->if_present)
>> Isn't if_present used also for OpenACC? Then can't it with -fopenmp
>> -fopenacc allow
>> !$acc ... if_present...
>> !$omp scan inclusive(...)
>> !$add end ...
>> ?
Yeah, that re-purposing of 'if_present' made me raise an eyebrow, too.
> !$acc ends up in a different ST_OMP_/EXEC_OMP_; additionally, due to the
> tight restrictions imposed by 'inscan'/'omp scan' adding something
> inbetween is difficult. (It can be added in 'block ... end block' but it
> still does not make much sense for 'omp scan' and it still ends up in a
> different statement.)
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'"?
Regarding my comment "As long as '!$omp scan' inside '!$acc host_data'
(generally, all constructs using 'if_present') reliably results in a
compile-time error, [...]": do we need a more elaborate testcase for
that, like not directly nesting '!$omp scan' inside '!$acc host_data'?
.., or, well..., just implement this '!$omp scan' checking differently?
;-)
Grüße
Thomas
-----------------
Mentor Graphics (Deutschland) GmbH, Arnulfstraße 201, 80634 München / Germany
Registergericht München HRB 106955, Geschäftsführer: Thomas Heurung, Alexander
Walter
>From ceff1cdbd0fae3884efe9558704644add4ac9257 Mon Sep 17 00:00:00 2001
From: Thomas Schwinge <tho...@codesourcery.com>
Date: Wed, 9 Dec 2020 12:33:12 +0100
Subject: [PATCH] Add 'gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90'
... as a "Variant of '../gomp/scan-1.f90', checking
'code->ext.omp_clauses->if_present' implementation detail".
gcc/testsuite/
* gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90: New.
* gfortran.dg/gomp/scan-1.f90: Refer.
gcc/fortran/
* openmp.c (gfc_resolve_omp_do_blocks, gfc_resolve_omp_directive):
Refer 'gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90'.
---
gcc/fortran/openmp.c | 6 ++-
.../goacc-gomp/omp-scan-1-if_present.f90 | 38 +++++++++++++++++++
gcc/testsuite/gfortran.dg/gomp/scan-1.f90 | 2 +
3 files changed, 44 insertions(+), 2 deletions(-)
create mode 100644 gcc/testsuite/gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90
diff --git a/gcc/fortran/openmp.c b/gcc/fortran/openmp.c
index b1f009785e3..8010ce91478 100644
--- a/gcc/fortran/openmp.c
+++ b/gcc/fortran/openmp.c
@@ -6222,7 +6222,8 @@ gfc_resolve_omp_do_blocks (gfc_code *code, gfc_namespace *ns)
gfc_error ("With INSCAN at %L, expected loop body with !$OMP SCAN "
"between two structured-block-sequences", loc);
else
- /* Mark as checked; flag will be unset later. */
+ /* Mark as checked; flag will be unset later.
+ (See 'gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90'.) */
c->block->next->next->ext.omp_clauses->if_present = true;
}
}
@@ -7123,7 +7124,8 @@ gfc_resolve_omp_directive (gfc_code *code, gfc_namespace *ns)
"except when omp_sync_hint_none is used", &code->loc);
break;
case EXEC_OMP_SCAN:
- /* Flag is only used to checking, hence, it is unset afterwards. */
+ /* Flag is only used to checking, hence, it is unset afterwards.
+ See 'gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90'. */
if (!code->ext.omp_clauses->if_present)
gfc_error ("Unexpected !$OMP SCAN at %L outside loop construct with "
"%<inscan%> REDUCTION clause", &code->loc);
diff --git a/gcc/testsuite/gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90 b/gcc/testsuite/gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90
new file mode 100644
index 00000000000..08b21e0cbf5
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/goacc-gomp/omp-scan-1-if_present.f90
@@ -0,0 +1,38 @@
+! Variant of '../gomp/scan-1.f90', checking 'code->ext.omp_clauses->if_present' implementation detail.
+
+! As long as '!$omp scan' inside '!$acc host_data' (generally, all constructs
+! using 'if_present') reliably results in a compile-time error, we don't have
+! to worry about the case where the checking done for an '!$omp scan' might
+! inadvertently alter 'if_present' for an outer '!$acc host_data'.
+
+module m
+ integer a, b
+end module m
+
+subroutine f1
+ use m
+ !$acc host_data if_present
+ !$omp scan inclusive (a) ! { dg-error "Unexpected ..OMP SCAN at .1. outside loop construct with 'inscan' REDUCTION clause" }
+ ! { dg-error "The ..OMP SCAN directive cannot be specified within a ..ACC HOST_DATA region" "" { target *-*-* } .-1 }
+ !$omp scan exclusive (b) ! { dg-error "Unexpected ..OMP SCAN at .1. outside loop construct with 'inscan' REDUCTION clause" }
+ ! { dg-error "The ..OMP SCAN directive cannot be specified within a ..ACC HOST_DATA region" "" { target *-*-* } .-1 }
+ !$acc end host_data
+end
+
+subroutine f9
+ use m
+ implicit none
+ integer i
+ !$acc host_data if_present
+! The first error (exit) causes two follow-up errors:
+ !$omp simd reduction (inscan, +: a) ! { dg-error "With INSCAN at .1., expected loop body with ..OMP SCAN between two structured-block-sequences" }
+ ! { dg-error "The ..OMP SIMD directive cannot be specified within a ..ACC HOST_DATA region" "" { target *-*-* } .-1 }
+ do i = 1, 64
+ if (i == 23) &
+ exit ! { dg-error "EXIT statement at .1. terminating ..OMP DO loop" } */
+ !$omp scan exclusive (a) ! { dg-error "Unexpected ..OMP SCAN at .1. outside loop construct with 'inscan' REDUCTION clause" }
+ ! { dg-error "The ..OMP SCAN directive cannot be specified within a ..ACC HOST_DATA region" "" { target *-*-* } .-1 }
+ a = a + 1
+ end do
+ !$acc end host_data
+end
diff --git a/gcc/testsuite/gfortran.dg/gomp/scan-1.f90 b/gcc/testsuite/gfortran.dg/gomp/scan-1.f90
index 8c879fd98b9..91be4147db1 100644
--- a/gcc/testsuite/gfortran.dg/gomp/scan-1.f90
+++ b/gcc/testsuite/gfortran.dg/gomp/scan-1.f90
@@ -1,3 +1,5 @@
+! See also '../goacc-gomp/omp-scan-1-if_present.f90'.
+
module m
integer a, b
end module m
--
2.17.1
