Hi!
On Wed, 12 Nov 2014 16:58:54 +0100, I wrote:
> Hi Casar!
>
> Can you please help me trying to understand the following change, that
> you've done a long time ago:
>
> On Tue, 8 Jul 2014 07:28:24 -0700, Cesar Philippidis
> <cesar_philippi...@mentor.com> wrote:
> > On 07/07/2014 02:55 AM, Thomas Schwinge wrote:
> >
> > > On Sun, 6 Jul 2014 16:10:56 -0700, Cesar Philippidis
> > > <cesar_philippi...@mentor.com> wrote:
> > >> This patch is the first step to enabling parallel reductions in openacc.
>
> > --- a/gcc/omp-low.c
> > +++ b/gcc/omp-low.c
>
> > @@ -1750,7 +1808,16 @@ scan_sharing_clauses (tree clauses, omp_context *ctx)
> > && TREE_CODE (TREE_TYPE (decl)) == ARRAY_TYPE)
> > install_var_field (decl, true, 7, ctx);
> > else
> > - install_var_field (decl, true, 3, ctx);
> > + {
> > + if (!is_gimple_omp_oacc_specifically (ctx->stmt))
> > + install_var_field (decl, true, 3, ctx);
> > + else
> > + {
> > + /* decl goes heres. */
> > + omp_context *c = (ctx->field_map ? ctx : ctx->outer);
> > + install_var_field (decl, true, 3, c);
> > + }
> > + }
> > if (is_gimple_omp_offloaded (ctx->stmt))
> > install_var_local (decl, ctx);
> > }
>
> Writing your change differently, easier to read:
>
> if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_MAP
> && OMP_CLAUSE_MAP_KIND (c) == OMP_CLAUSE_MAP_POINTER
> && !OMP_CLAUSE_MAP_ZERO_BIAS_ARRAY_SECTION (c)
> && TREE_CODE (TREE_TYPE (decl)) == ARRAY_TYPE)
> install_var_field (decl, true, 7, ctx);
> + else if (is_gimple_omp_oacc_specifically (ctx->stmt))
> + {
> + /* decl goes heres. */
> + omp_context *c = (ctx->field_map ? ctx :
> ctx->outer);
> + install_var_field (decl, true, 3, c);
> + }
> else
> install_var_field (decl, true, 3, ctx);
>
> It is a generic OpenACC (that is, not only OpenACC reductions) code path
> that you're changing here. Can you still come up with a rationale for
> that change, or should this possibly be restricted to OpenACC reductions
> processing only?
Cesar could neither remember, nor make up a new rationale on the spot;
trusting in regression testing, and that showing no issues, I have now
reverted this in r217462 on gomp-4_0-branch:
commit 389a8054007e619a917c16234cc697cae711de02
Author: tschwinge <tschwinge@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Thu Nov 13 08:13:33 2014 +0000
Middle end: Revert earlier change.
This change in OMP_CLAUSE_MAP handling had originally been applied for
reductions support, but is now obsolete.
gcc/
* omp-low.c (scan_sharing_clauses) <OMP_CLAUSE_MAP>: Revert
earlier change.
git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gomp-4_0-branch@217462
138bc75d-0d04-0410-961f-82ee72b054a4
---
gcc/ChangeLog.gomp | 5 +++++
gcc/omp-low.c | 11 +----------
2 files changed, 6 insertions(+), 10 deletions(-)
diff --git gcc/ChangeLog.gomp gcc/ChangeLog.gomp
index 7501679..174235d 100644
--- gcc/ChangeLog.gomp
+++ gcc/ChangeLog.gomp
@@ -1,3 +1,8 @@
+2014-11-13 Thomas Schwinge <tho...@codesourcery.com>
+
+ * omp-low.c (scan_sharing_clauses) <OMP_CLAUSE_MAP>: Revert
+ earlier change.
+
2014-11-12 Cesar Philippidis <ce...@codesourcery.com>
* tree.c (omp_clause_code_name): Add missing comma
diff --git gcc/omp-low.c gcc/omp-low.c
index 44e14b4..e511846 100644
--- gcc/omp-low.c
+++ gcc/omp-low.c
@@ -1902,16 +1902,7 @@ scan_sharing_clauses (tree clauses, omp_context *ctx)
&& TREE_CODE (TREE_TYPE (decl)) == ARRAY_TYPE)
install_var_field (decl, true, 7, ctx);
else
- {
- if (!is_gimple_omp_oacc_specifically (ctx->stmt))
- install_var_field (decl, true, 3, ctx);
- else
- {
- /* decl goes heres. */
- omp_context *c = (ctx->field_map ? ctx : ctx->outer);
- install_var_field (decl, true, 3, c);
- }
- }
+ install_var_field (decl, true, 3, ctx);
if (is_gimple_omp_offloaded (ctx->stmt))
install_var_local (decl, ctx);
}
Grüße,
Thomas
pgpWw2Hx5SpKp.pgp
Description: PGP signature
