Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-06-10 Thread Thomas Schwinge 
Hi!

I'd still have preferred if that patch described (adding source code
comments) why that special handling is required, but oh well...

On Fri, 3 Jun 2016 15:13:40 +0800, Chung-Lin Tang  
wrote:
> Tested again with no regressions.

Your gomp-4_0-branch commit of this patch, r237210, forgot to clean up
XFAILs; now committed to gomp-4_0-branch in r237298:

commit f26e600bdf37b2d68bd479c1e7d6a33866da5acd
Author: tschwinge 
Date:   Fri Jun 10 10:12:09 2016 +

Clean up libgomp.oacc-c++/template-reduction.C XFAIL

libgomp/
* testsuite/libgomp.oacc-c++/template-reduction.C: Remove XFAIL.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gomp-4_0-branch@237298 
138bc75d-0d04-0410-961f-82ee72b054a4
---
 libgomp/ChangeLog.gomp  | 4 
 libgomp/testsuite/libgomp.oacc-c++/template-reduction.C | 4 
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git libgomp/ChangeLog.gomp libgomp/ChangeLog.gomp
index fc1decf..ccee4d1c 100644
--- libgomp/ChangeLog.gomp
+++ libgomp/ChangeLog.gomp
@@ -1,3 +1,7 @@
+2016-06-10  Thomas Schwinge  
+
+   * testsuite/libgomp.oacc-c++/template-reduction.C: Remove XFAIL.
+
 2016-06-08  Chung-Lin Tang  
 
Backport from trunk r237070:
diff --git libgomp/testsuite/libgomp.oacc-c++/template-reduction.C 
libgomp/testsuite/libgomp.oacc-c++/template-reduction.C
index e2a0ca0..6c85fba 100644
--- libgomp/testsuite/libgomp.oacc-c++/template-reduction.C
+++ libgomp/testsuite/libgomp.oacc-c++/template-reduction.C
@@ -1,7 +1,3 @@
-// TODO: async_sum is currently failing on nvptx.
-
-// { dg-xfail-run-if "TODO" { openacc_nvidia_accel_selected } }
-
 const int n = 100;
 
 // Check explicit template copy map


Grüße
 Thomas


signature.asc
Description: PGP signature

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-06-03 Thread Jakub Jelinek 
On Fri, Jun 03, 2016 at 03:13:40PM +0800, Chung-Lin Tang wrote:
> On 2016/6/2 10:00 PM, Jakub Jelinek wrote:
> > Wouldn't it be better to pass either a bool openacc_async flag, or
> > whole clauses, down to gfc_trans_omp_reduction_list and handle it there
> > instead of walking the list after the fact?
> 
> You mean this style? (patch attached)
> Tested again with no regressions.
> 
> Thanks,
> Chung-Lin
> 
> * trans-openmp.c (gfc_trans_omp_reduction_list): Add mark_addressable
> bool parameter, set reduction clause DECLs as addressable when true.
> (gfc_trans_omp_clauses): Pass clauses->async to
> gfc_trans_omp_reduction_list, add comment describing OpenACC 
> situation.

Yep, thanks (and the C/C++ patch is ok too).

> Index: trans-openmp.c
> ===
> --- trans-openmp.c(revision 236845)
> +++ trans-openmp.c(working copy)
> @@ -1646,7 +1646,7 @@ gfc_trans_omp_array_reduction_or_udr (tree c, gfc_
>  
>  static tree
>  gfc_trans_omp_reduction_list (gfc_omp_namelist *namelist, tree list,
> -   locus where)
> +   locus where, bool mark_addressable)
>  {
>for (; namelist != NULL; namelist = namelist->next)
>  if (namelist->sym->attr.referenced)
> @@ -1657,6 +1657,8 @@ gfc_trans_omp_reduction_list (gfc_omp_namelist *na
>   tree node = build_omp_clause (where.lb->location,
> OMP_CLAUSE_REDUCTION);
>   OMP_CLAUSE_DECL (node) = t;
> + if (mark_addressable)
> +   TREE_ADDRESSABLE (t) = 1;
>   switch (namelist->u.reduction_op)
> {
> case OMP_REDUCTION_PLUS:
> @@ -1747,7 +1749,10 @@ gfc_trans_omp_clauses (stmtblock_t *block, gfc_omp
>switch (list)
>   {
>   case OMP_LIST_REDUCTION:
> -   omp_clauses = gfc_trans_omp_reduction_list (n, omp_clauses, where);
> +   /* An OpenACC async clause indicates the need to set reduction
> +  arguments addressable, to allow asynchronous copy-out.  */
> +   omp_clauses = gfc_trans_omp_reduction_list (n, omp_clauses, where,
> +   clauses->async);
> break;
>   case OMP_LIST_PRIVATE:
> clause_code = OMP_CLAUSE_PRIVATE;


Jakub

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-06-03 Thread Chung-Lin Tang 
On 2016/6/2 10:00 PM, Jakub Jelinek wrote:
> Wouldn't it be better to pass either a bool openacc_async flag, or
> whole clauses, down to gfc_trans_omp_reduction_list and handle it there
> instead of walking the list after the fact?

You mean this style? (patch attached)
Tested again with no regressions.

Thanks,
Chung-Lin

* trans-openmp.c (gfc_trans_omp_reduction_list): Add mark_addressable
bool parameter, set reduction clause DECLs as addressable when true.
(gfc_trans_omp_clauses): Pass clauses->async to
gfc_trans_omp_reduction_list, add comment describing OpenACC situation.

Index: trans-openmp.c
===
--- trans-openmp.c  (revision 236845)
+++ trans-openmp.c  (working copy)
@@ -1646,7 +1646,7 @@ gfc_trans_omp_array_reduction_or_udr (tree c, gfc_
 
 static tree
 gfc_trans_omp_reduction_list (gfc_omp_namelist *namelist, tree list,
- locus where)
+ locus where, bool mark_addressable)
 {
   for (; namelist != NULL; namelist = namelist->next)
 if (namelist->sym->attr.referenced)
@@ -1657,6 +1657,8 @@ gfc_trans_omp_reduction_list (gfc_omp_namelist *na
tree node = build_omp_clause (where.lb->location,
  OMP_CLAUSE_REDUCTION);
OMP_CLAUSE_DECL (node) = t;
+   if (mark_addressable)
+ TREE_ADDRESSABLE (t) = 1;
switch (namelist->u.reduction_op)
  {
  case OMP_REDUCTION_PLUS:
@@ -1747,7 +1749,10 @@ gfc_trans_omp_clauses (stmtblock_t *block, gfc_omp
   switch (list)
{
case OMP_LIST_REDUCTION:
- omp_clauses = gfc_trans_omp_reduction_list (n, omp_clauses, where);
+ /* An OpenACC async clause indicates the need to set reduction
+arguments addressable, to allow asynchronous copy-out.  */
+ omp_clauses = gfc_trans_omp_reduction_list (n, omp_clauses, where,
+ clauses->async);
  break;
case OMP_LIST_PRIVATE:
  clause_code = OMP_CLAUSE_PRIVATE;

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-06-02 Thread Jakub Jelinek 
On Thu, Jun 02, 2016 at 09:55:05PM +0800, Chung-Lin Tang wrote:
>   fortran/
>   * trans-openmp.c (gfc_trans_omp_clauses): Mark OpenACC reduction
>   arguments as addressable when async clause exists.

Wouldn't it be better to pass either a bool openacc_async flag, or
whole clauses, down to gfc_trans_omp_reduction_list and handle it there
instead of walking the list after the fact?

> Index: trans-openmp.c
> ===
> --- trans-openmp.c(revision 236845)
> +++ trans-openmp.c(working copy)
> @@ -1748,6 +1748,12 @@ gfc_trans_omp_clauses (stmtblock_t *block, gfc_omp
>   {
>   case OMP_LIST_REDUCTION:
> omp_clauses = gfc_trans_omp_reduction_list (n, omp_clauses, where);
> +   /* An OpenACC async clause indicates the need to set reduction
> +  arguments addressable, to allow asynchronous copy-out.  */
> +   if (clauses->async)
> + for (c = omp_clauses; c; c = OMP_CLAUSE_CHAIN (c))
> +   if (DECL_P (OMP_CLAUSE_DECL (c)))
> + TREE_ADDRESSABLE (OMP_CLAUSE_DECL (c)) = 1;
> break;
>   case OMP_LIST_PRIVATE:
> clause_code = OMP_CLAUSE_PRIVATE;


Jakub

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-06-02 Thread Chung-Lin Tang 
On 2016/6/1 09:38 PM, Jakub Jelinek wrote:
>>  construct_clauses.lists[OMP_LIST_REDUCTION] = NULL;
>> >oacc_clauses = gfc_trans_omp_clauses (&block, &construct_clauses,
>> >code->loc);
>> > +  for (tree c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
>> > +  if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ASYNC)
>> > +{
>> > +  for (c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
>> > +if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_REDUCTION
>> > +&& DECL_P (OMP_CLAUSE_DECL (c)))
>> > +  TREE_ADDRESSABLE (OMP_CLAUSE_DECL (c)) = 1;
>> > +  break;
>> > +}
>> >  }
> These 2 look wrong to me.  1) you really don't need to walk all the clauses
> to find if there is OMP_CLAUSE_ASYNC, you can just test the
> async field of struct gfc_omp_clauses.  And, 2) is there any reason why you
> can't just do this in gfc_trans_omp_clauses instead, when crating
> OMP_CLAUSE_REDUCTION if clauses->async is set?  Or are there some cases
> where on OpenACC constructs you don't want to do this?

Thanks for reminding, I didn't notice there was such a gfc_omp_clauses->async 
field.

Here's a much more succinct patch that does it inside gfc_trans_omp_clauses.
Again re-tested gfortran and libgomp without regressions.
Is this and the C/C++ patches (and the new testsuite cases) okay for trunk?

Thanks,
Chung-Lin

fortran/
* trans-openmp.c (gfc_trans_omp_clauses): Mark OpenACC reduction
arguments as addressable when async clause exists.

Index: trans-openmp.c
===
--- trans-openmp.c  (revision 236845)
+++ trans-openmp.c  (working copy)
@@ -1748,6 +1748,12 @@ gfc_trans_omp_clauses (stmtblock_t *block, gfc_omp
{
case OMP_LIST_REDUCTION:
  omp_clauses = gfc_trans_omp_reduction_list (n, omp_clauses, where);
+ /* An OpenACC async clause indicates the need to set reduction
+arguments addressable, to allow asynchronous copy-out.  */
+ if (clauses->async)
+   for (c = omp_clauses; c; c = OMP_CLAUSE_CHAIN (c))
+ if (DECL_P (OMP_CLAUSE_DECL (c)))
+   TREE_ADDRESSABLE (OMP_CLAUSE_DECL (c)) = 1;
  break;
case OMP_LIST_PRIVATE:
  clause_code = OMP_CLAUSE_PRIVATE;

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-06-01 Thread Chung-Lin Tang 
On 2016/5/31 05:51 PM, Chung-Lin Tang wrote:
> On 2016/5/31 3:28 PM, Thomas Schwinge wrote:
>> > Hi!
>> > 
>> > On Mon, 30 May 2016 18:53:41 +0200, Jakub Jelinek  wrote:
>>> >> On Mon, May 30, 2016 at 10:38:59PM +0800, Chung-Lin Tang wrote:
 >>> Hi, a previous patch of Cesar's has made the middle-end omp-lowering
 >>> automatically create and insert a tofrom (i.e. present_or_copy) map for
 >>> parallel reductions.  This allowed the user to not need explicit
 >>> clauses to copy out the reduction result, but because reduction 
 >>> arguments
 >>> are not marked addressable, async does not work as expected,
 >>> i.e. the asynchronous copy-out results are not used in the compiler 
 >>> generated code.
>>> >>
>>> >> If you need it only for async parallel/kernels? regions, can't you do 
>>> >> that
>>> >> only for those and not for others?
> That is achievable, but not in line with how we currently treat all other
> data clause OMP_CLAUSE_MAPs, which are all marked addressable. Is this special
> case handling really better here?
> 

Hi Jakub, here's a version of the patch with the addressable marking restricted
to when there's an async clause. Tests re-ran to ensure no regressions.  Please 
inform
if this way seems better.  Also attached are some new testcases.

Thanks,
Chung-Lin

2016-06-01  Chung-Lin Tang  

c/
* c-typeck.c (c_finish_omp_clauses): Mark OpenACC reduction
arguments as addressable when async clause exists.

cp/
* semantics.c (finish_omp_clauses): Mark OpenACC reduction
arguments as addressable when async clause exists.

fortran/
* trans-openmp.c (gfc_trans_oacc_construct): Mark OpenACC reduction
arguments as addressable. when async clause exists.
(gfc_trans_oacc_combined_directive): Likewise.

libgomp/testsuite/
* libgomp/testsuite/libgomp.oacc-c-c++-common/reduction-8.c: New test.
* libgomp/testsuite/libgomp.oacc-fortran/reduction-8.f90: New test.

Index: c/c-typeck.c
===
--- c/c-typeck.c	(revision 236845)
+++ c/c-typeck.c	(working copy)
@@ -12529,6 +12529,7 @@ c_finish_omp_clauses (tree clauses, enum c_omp_reg
   tree *nowait_clause = NULL;
   bool ordered_seen = false;
   tree schedule_clause = NULL_TREE;
+  bool oacc_async = false;
 
   bitmap_obstack_initialize (NULL);
   bitmap_initialize (&generic_head, &bitmap_default_obstack);
@@ -12539,6 +12540,14 @@ c_finish_omp_clauses (tree clauses, enum c_omp_reg
   bitmap_initialize (&map_field_head, &bitmap_default_obstack);
   bitmap_initialize (&oacc_reduction_head, &bitmap_default_obstack);
 
+  if (ort & C_ORT_ACC)
+for (c = clauses; c; c = OMP_CLAUSE_CHAIN (c))
+  if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ASYNC)
+	{
+	  oacc_async = true;
+	  break;
+	}
+	  
   for (pc = &clauses, c = clauses; c ; c = *pc)
 {
   bool remove = false;
@@ -12575,6 +12584,8 @@ c_finish_omp_clauses (tree clauses, enum c_omp_reg
 	  remove = true;
 	  break;
 	}
+	  if (oacc_async)
+	c_mark_addressable (t);
 	  type = TREE_TYPE (t);
 	  if (TREE_CODE (t) == MEM_REF)
 	type = TREE_TYPE (type);
Index: cp/semantics.c
===
--- cp/semantics.c	(revision 236845)
+++ cp/semantics.c	(working copy)
@@ -5774,6 +5774,7 @@ finish_omp_clauses (tree clauses, enum c_omp_regio
   bool branch_seen = false;
   bool copyprivate_seen = false;
   bool ordered_seen = false;
+  bool oacc_async = false;
 
   bitmap_obstack_initialize (NULL);
   bitmap_initialize (&generic_head, &bitmap_default_obstack);
@@ -5784,6 +5785,14 @@ finish_omp_clauses (tree clauses, enum c_omp_regio
   bitmap_initialize (&map_field_head, &bitmap_default_obstack);
   bitmap_initialize (&oacc_reduction_head, &bitmap_default_obstack);
 
+  if (ort & C_ORT_ACC)
+for (c = clauses; c; c = OMP_CLAUSE_CHAIN (c))
+  if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ASYNC)
+	{
+	  oacc_async = true;
+	  break;
+	}
+
   for (pc = &clauses, c = clauses; c ; c = *pc)
 {
   bool remove = false;
@@ -5827,6 +5836,8 @@ finish_omp_clauses (tree clauses, enum c_omp_regio
 		t = n;
 	  goto check_dup_generic_t;
 	}
+	  if (oacc_async)
+	cxx_mark_addressable (t);
 	  goto check_dup_generic;
 	case OMP_CLAUSE_COPYPRIVATE:
 	  copyprivate_seen = true;
Index: fortran/trans-openmp.c
===
--- fortran/trans-openmp.c	(revision 236845)
+++ fortran/trans-openmp.c	(working copy)
@@ -2704,6 +2704,15 @@ gfc_trans_oacc_construct (gfc_code *code)
   gfc_start_block (&block);
   oacc_clauses = gfc_trans_omp_clauses (&block, code->ext.omp_clauses,
 	code->loc);
+  for (tree c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
+if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ASYNC)
+  {
+	for (c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
+	  if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_REDUC

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-06-01 Thread Jakub Jelinek 
On Wed, Jun 01, 2016 at 09:32:26PM +0800, Chung-Lin Tang wrote:
> 2016-06-01  Chung-Lin Tang  
> 
>   c/
>   * c-typeck.c (c_finish_omp_clauses): Mark OpenACC reduction
>   arguments as addressable when async clause exists.
> 
>   cp/
>   * semantics.c (finish_omp_clauses): Mark OpenACC reduction
>   arguments as addressable when async clause exists.

This LGTM.
> 
>   fortran/
>   * trans-openmp.c (gfc_trans_oacc_construct): Mark OpenACC reduction
>   arguments as addressable. when async clause exists.
>   (gfc_trans_oacc_combined_directive): Likewise.

> --- fortran/trans-openmp.c(revision 236845)
> +++ fortran/trans-openmp.c(working copy)
> @@ -2704,6 +2704,15 @@ gfc_trans_oacc_construct (gfc_code *code)
>gfc_start_block (&block);
>oacc_clauses = gfc_trans_omp_clauses (&block, code->ext.omp_clauses,
>   code->loc);
> +  for (tree c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
> +if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ASYNC)
> +  {
> + for (c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
> +   if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_REDUCTION
> +   && DECL_P (OMP_CLAUSE_DECL (c)))
> + TREE_ADDRESSABLE (OMP_CLAUSE_DECL (c)) = 1;
> + break;
> +  }
>stmt = gfc_trans_omp_code (code->block->next, true);
>stmt = build2_loc (input_location, construct_code, void_type_node, stmt,
>oacc_clauses);
> @@ -3501,6 +3510,15 @@ gfc_trans_oacc_combined_directive (gfc_code *code)
>   construct_clauses.lists[OMP_LIST_REDUCTION] = NULL;
>oacc_clauses = gfc_trans_omp_clauses (&block, &construct_clauses,
>   code->loc);
> +  for (tree c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
> + if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_ASYNC)
> +   {
> + for (c = oacc_clauses; c; c = OMP_CLAUSE_CHAIN (c))
> +   if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_REDUCTION
> +   && DECL_P (OMP_CLAUSE_DECL (c)))
> + TREE_ADDRESSABLE (OMP_CLAUSE_DECL (c)) = 1;
> + break;
> +   }
>  }

These 2 look wrong to me.  1) you really don't need to walk all the clauses
to find if there is OMP_CLAUSE_ASYNC, you can just test the
async field of struct gfc_omp_clauses.  And, 2) is there any reason why you
can't just do this in gfc_trans_omp_clauses instead, when crating
OMP_CLAUSE_REDUCTION if clauses->async is set?  Or are there some cases
where on OpenACC constructs you don't want to do this?

Jakub

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-05-31 Thread Chung-Lin Tang 
On 2016/5/31 3:28 PM, Thomas Schwinge wrote:
> Hi!
> 
> On Mon, 30 May 2016 18:53:41 +0200, Jakub Jelinek  wrote:
>> On Mon, May 30, 2016 at 10:38:59PM +0800, Chung-Lin Tang wrote:
>>> Hi, a previous patch of Cesar's has made the middle-end omp-lowering
>>> automatically create and insert a tofrom (i.e. present_or_copy) map for
>>> parallel reductions.  This allowed the user to not need explicit
>>> clauses to copy out the reduction result, but because reduction arguments
>>> are not marked addressable, async does not work as expected,
>>> i.e. the asynchronous copy-out results are not used in the compiler 
>>> generated code.
>>
>> If you need it only for async parallel/kernels? regions, can't you do that
>> only for those and not for others?

That is achievable, but not in line with how we currently treat all other
data clause OMP_CLAUSE_MAPs, which are all marked addressable. Is this special
case handling really better here?

> Also, please add comments to the source code to document the need for
> such special handling.
> 
>>> This patch fixes this in the front-ends, I've tested this patch without
>>> new regressions, and fixes some C++ OpenACC tests that regressed after
>>> my last OpenACC async patch.  Is this okay for trunk?
>>
>> Testcases in the testsuite or others?  If the latter, we should add them.
> 
> The r236772 commit "[PATCH, libgomp] Rewire OpenACC async",
> 
> regressed (or, triggered/exposed the existing wrong behavior?)
> libgomp.oacc-c++/template-reduction.C execution testing for nvptx
> offloading.  (Please always send email about such known regressions, and
> XFAIL them with your commit -- that would have saved me an hour
> yesterday, when I bisected recent changes to figure out why that test
> suddenly fails.)

Sorry, Thomas. I was going to quickly send this follow-up patch, so glossed 
over XFAILing.

> For reference, here is a test case, a reduced C version of
> libgomp/testsuite/libgomp.oacc-c++/template-reduction.C.  This test case
> works (without Chung-Lin's "[PATCH, OpenACC] Make reduction arguments
> addressable" patch) if I enable "POCs", which surprises me a bit, because
> I thought after Cesar's recent changes, the gimplifier is doing the same
> thing of adding a data clause next to the reduction clause.  Probably
> it's not doing the exactly same thing, though.  Should it?  Cesar, do you
> have any comments on this?  For example (just guessing), should
> TREE_ADDRESSABLE be set where the gimplifier does its work, instead of in
> the three front ends?

There's really nothing wrong about Cesar's patch. The marking addressable needs
to be done earlier, or it may be too late during gimplification. I already
tried to fix this in gimplify.c before, but didn't completely work.

I'll add more testcases for this before I commit any final patches.

Thanks,
Chung-Lin

> 
> // Reduced C version of 
> libgomp/testsuite/libgomp.oacc-c++/template-reduction.C.
> 
> const int n = 100;
> 
> // Check present and async and an explicit firstprivate
> 
> int
> async_sum (int c)
> {
>   int s = 0;
> 
> #define POCs //present_or_copy(s)
> #pragma acc parallel loop num_gangs (10) gang reduction (+:s) POCs 
> firstprivate (c) async
>   for (int i = 0; i < n; i++)
> s += i+c;
> 
> #pragma acc wait
> 
>   return s;
> }
> 
> int
> main()
> {
>   int result = 0;
> 
>   for (int i = 0; i < n; i++)
> {
>   result += i+1;
> }
> 
>   if (async_sum (1) != result)
> __builtin_abort ();
> 
>   return 0;
> }
> 
> 
> Grüße
>  Thomas
>

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-05-31 Thread Thomas Schwinge 
Hi!

On Mon, 30 May 2016 18:53:41 +0200, Jakub Jelinek  wrote:
> On Mon, May 30, 2016 at 10:38:59PM +0800, Chung-Lin Tang wrote:
> > Hi, a previous patch of Cesar's has made the middle-end omp-lowering
> > automatically create and insert a tofrom (i.e. present_or_copy) map for
> > parallel reductions.  This allowed the user to not need explicit
> > clauses to copy out the reduction result, but because reduction arguments
> > are not marked addressable, async does not work as expected,
> > i.e. the asynchronous copy-out results are not used in the compiler 
> > generated code.
> 
> If you need it only for async parallel/kernels? regions, can't you do that
> only for those and not for others?

Also, please add comments to the source code to document the need for
such special handling.

> > This patch fixes this in the front-ends, I've tested this patch without
> > new regressions, and fixes some C++ OpenACC tests that regressed after
> > my last OpenACC async patch.  Is this okay for trunk?
> 
> Testcases in the testsuite or others?  If the latter, we should add them.

The r236772 commit "[PATCH, libgomp] Rewire OpenACC async",

regressed (or, triggered/exposed the existing wrong behavior?)
libgomp.oacc-c++/template-reduction.C execution testing for nvptx
offloading.  (Please always send email about such known regressions, and
XFAIL them with your commit -- that would have saved me an hour
yesterday, when I bisected recent changes to figure out why that test
suddenly fails.)

For reference, here is a test case, a reduced C version of
libgomp/testsuite/libgomp.oacc-c++/template-reduction.C.  This test case
works (without Chung-Lin's "[PATCH, OpenACC] Make reduction arguments
addressable" patch) if I enable "POCs", which surprises me a bit, because
I thought after Cesar's recent changes, the gimplifier is doing the same
thing of adding a data clause next to the reduction clause.  Probably
it's not doing the exactly same thing, though.  Should it?  Cesar, do you
have any comments on this?  For example (just guessing), should
TREE_ADDRESSABLE be set where the gimplifier does its work, instead of in
the three front ends?

// Reduced C version of 
libgomp/testsuite/libgomp.oacc-c++/template-reduction.C.

const int n = 100;

// Check present and async and an explicit firstprivate

int
async_sum (int c)
{
  int s = 0;

#define POCs //present_or_copy(s)
#pragma acc parallel loop num_gangs (10) gang reduction (+:s) POCs 
firstprivate (c) async
  for (int i = 0; i < n; i++)
s += i+c;

#pragma acc wait

  return s;
}

int
main()
{
  int result = 0;

  for (int i = 0; i < n; i++)
{
  result += i+1;
}

  if (async_sum (1) != result)
__builtin_abort ();

  return 0;
}


Grüße
 Thomas


signature.asc
Description: PGP signature

Re: [PATCH, OpenACC] Make reduction arguments addressable

2016-05-30 Thread Jakub Jelinek 
On Mon, May 30, 2016 at 10:38:59PM +0800, Chung-Lin Tang wrote:
> Hi, a previous patch of Cesar's has made the middle-end omp-lowering
> automatically create and insert a tofrom (i.e. present_or_copy) map for
> parallel reductions.  This allowed the user to not need explicit
> clauses to copy out the reduction result, but because reduction arguments
> are not marked addressable, async does not work as expected,
> i.e. the asynchronous copy-out results are not used in the compiler generated 
> code.

If you need it only for async parallel/kernels? regions, can't you do that
only for those and not for others?

> This patch fixes this in the front-ends, I've tested this patch without
> new regressions, and fixes some C++ OpenACC tests that regressed after
> my last OpenACC async patch.  Is this okay for trunk?

Testcases in the testsuite or others?  If the latter, we should add them.

Jakub