Re: openacc reference reductions

[Cesar Philippidis <cesar@codesourcery.com>]

[-- Attachment #1: Type: text/plain, Size: 1667 bytes --]

On 02/22/2016 07:34 AM, Cesar Philippidis wrote:
> Ping. This patch still needs a review.

Ping. I've attached a rebased version of this patch. The omp-low.c bits
haven't changed, but the test cases have since Thomas has been merging
some of them from trunk. I separated the omp-low.c changes because the
test cases are relatively large.

This patch fixes PR70533 and PR70535. Basically it teaches
lower_oacc_reductions how to cope with reference-type variables
(PR70533) and not to remap the reduction variables on parallel
constructs (PR70535).

Is it OK for trunk?

Cesar

> On 02/09/2016 08:17 AM, Cesar Philippidis wrote:
>> On 02/09/2016 07:33 AM, Nathan Sidwell wrote:
>>> While I've not looked at the rest of the patch, this bit stood out:
>>>
>>>> +static bool
>>>> +is_oacc_parallel_reduction (tree var, omp_context *ctx)
>>>> +{
>>>> +  if (!is_oacc_parallel (ctx))
>>>> +    return false;
>>>> +
>>>> +  tree clauses = gimple_omp_target_clauses (ctx->stmt);
>>>> +
>>>> +  /* Don't install a local copy of the decl if it used
>>>> +     inside a acc parallel reduction.  */
>>>
>>> ^^ comment is misleading -- this routine's not installing anything
>>>
>>>> +  if (is_oacc_parallel (ctx))
>>>
>>> ^^ already checked above.
>>>
>>>> +    for (tree c = clauses; c; c = OMP_CLAUSE_CHAIN (c))
>>>> +      if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_REDUCTION
>>>> +      && OMP_CLAUSE_DECL (c) == var)
>>>> +    return true;
>>>> +
>>>> +  return false;
>>>> +}
>>>> +
>>
>> Thanks for catching that. Those are artifacts from when this code used
>> to be located exclusively in scan_sharing_clauses. I've updated the
>> patch with those changes.
>>
>> Cesar
>>
> 


[-- Attachment #2: pr70533-omplow.diff --]
[-- Type: text/x-patch, Size: 9744 bytes --]

2016-04-05  Cesar Philippidis  <cesar@codesourcery.com>

	gcc/
	* omp-low.c (is_oacc_parallel_reduction): New function.
	(scan_sharing_clauses): Use it to prevent installing local variables
	for those used in acc parallel reductions.
	(lower_rec_input_clauses): Remove dead code.
	(lower_oacc_reductions): Add support for reference reductions.
	(lower_reduction_clauses): Remove dead code.
	(lower_omp_target): Don't remap variables appearing in acc parallel
	reductions.

	gcc/testsuite/
	* gfortran.dg/goacc/reduction-promotions.f90: Add more coverage.

	libgomp/
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-gang-np-1.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-gv-np-1.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-gw-np-1.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-1.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-2.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-3.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-gwv-np-4.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-vector-p-1.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-vector-p-2.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-worker-p-1.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-wv-p-1.c: New test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-wv-p-2.c: New test.
	* testsuite/libgomp.oacc-c-c++-common/loop-reduction-wv-p-3.c: New test.
	* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-1.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-2.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-3.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-4.c: New
	test.
	* testsuite/libgomp.oacc-c-c++-common/par-reduction-1.c: Adjust test.
	* testsuite/libgomp.oacc-c-c++-common/par-reduction-2.c: Likewise.
	* testsuite/libgomp.oacc-c-c++-common/parallel-dims.c: Likewise.
	* testsuite/libgomp.oacc-c-c++-common/parallel-reduction.c: New test.
	* testsuite/libgomp.oacc-c-c++-common/reduction-1.c: Adjust test.
	* testsuite/libgomp.oacc-c-c++-common/reduction-2.c: Likewise.
	* testsuite/libgomp.oacc-c-c++-common/reduction-3.c: Likewise.
	* testsuite/libgomp.oacc-c-c++-common/reduction-4.c: Likewise.
	* testsuite/libgomp.oacc-c-c++-common/reduction-5.c: Likewise.
	* testsuite/libgomp.oacc-c-c++-common/reduction-6.c: New test.
	* testsuite/libgomp.oacc-c-c++-common/reduction.h: New test.
	* testsuite/libgomp.oacc-fortran/parallel-reduction.f90: New test.
	* testsuite/libgomp.oacc-fortran/reduction-1.f90: Adjust test.
	* testsuite/libgomp.oacc-fortran/reduction-2.f90: Likewise.
	* testsuite/libgomp.oacc-fortran/reduction-3.f90: Likewise.
	* testsuite/libgomp.oacc-fortran/reduction-4.f90: Likewise.
	* testsuite/libgomp.oacc-fortran/reduction-5.f90: Likewise.
	* testsuite/libgomp.oacc-fortran/reduction-6.f90: Likewise.
	* testsuite/libgomp.oacc-fortran/reduction-7.f90: New test.


diff --git a/gcc/omp-low.c b/gcc/omp-low.c
index 3fd6eb3..fa2d318 100644
--- a/gcc/omp-low.c
+++ b/gcc/omp-low.c
@@ -309,6 +309,25 @@ is_oacc_kernels (omp_context *ctx)
 	      == GF_OMP_TARGET_KIND_OACC_KERNELS));
 }
 
+/* Return true if CTX corresponds to an oacc parallel region and if
+   VAR is used in a reduction.  */
+
+static bool
+is_oacc_parallel_reduction (tree var, omp_context *ctx)
+{
+  if (!is_oacc_parallel (ctx))
+    return false;
+
+  tree clauses = gimple_omp_target_clauses (ctx->stmt);
+
+  for (tree c = clauses; c; c = OMP_CLAUSE_CHAIN (c))
+    if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_REDUCTION
+	&& OMP_CLAUSE_DECL (c) == var)
+      return true;
+
+  return false;
+}
+
 /* If DECL is the artificial dummy VAR_DECL created for non-static
    data member privatization, return the underlying "this" parameter,
    otherwise return NULL.  */
@@ -2122,7 +2141,8 @@ scan_sharing_clauses (tree clauses, omp_context *ctx,
 		  else
 		    install_var_field (decl, true, 3, ctx,
 				       base_pointers_restrict);
-		  if (is_gimple_omp_offloaded (ctx->stmt))
+		  if (is_gimple_omp_offloaded (ctx->stmt)
+		      && !is_oacc_parallel_reduction (decl, ctx))
 		    install_var_local (decl, ctx);
 		}
 	    }
@@ -4837,7 +4857,7 @@ lower_rec_input_clauses (tree clauses, gimple_seq *ilist, gimple_seq *dlist,
 		  gimplify_assign (ptr, x, ilist);
 		}
 	    }
-	  else if (is_reference (var) && !is_oacc_parallel (ctx))
+	  else if (is_reference (var))
 	    {
 	      /* For references that are being privatized for Fortran,
 		 allocate new backing storage for the new pointer
@@ -5573,7 +5593,8 @@ lower_oacc_reductions (location_t loc, tree clauses, tree level, bool inner,
 	tree orig = OMP_CLAUSE_DECL (c);
 	tree var = maybe_lookup_decl (orig, ctx);
 	tree ref_to_res = NULL_TREE;
-	tree incoming, outgoing;
+	tree incoming, outgoing, v1, v2, v3;
+	bool is_private = false;
 
 	enum tree_code rcode = OMP_CLAUSE_REDUCTION_CODE (c);
 	if (rcode == MINUS_EXPR)
@@ -5586,7 +5607,6 @@ lower_oacc_reductions (location_t loc, tree clauses, tree level, bool inner,
 
 	if (!var)
 	  var = orig;
-	gcc_assert (!is_reference (var));
 
 	incoming = outgoing = var;
 	
@@ -5622,22 +5642,38 @@ lower_oacc_reductions (location_t loc, tree clauses, tree level, bool inner,
 		for (; cls;  cls = OMP_CLAUSE_CHAIN (cls))
 		  if (OMP_CLAUSE_CODE (cls) == OMP_CLAUSE_REDUCTION
 		      && orig == OMP_CLAUSE_DECL (cls))
-		    goto has_outer_reduction;
+		    {
+		      incoming = outgoing = lookup_decl (orig, probe);
+		      goto has_outer_reduction;
+		    }
+		  else if ((OMP_CLAUSE_CODE (cls) == OMP_CLAUSE_FIRSTPRIVATE
+			    || OMP_CLAUSE_CODE (cls) == OMP_CLAUSE_PRIVATE)
+			   && orig == OMP_CLAUSE_DECL (cls))
+		    {
+		      is_private = true;
+		      goto do_lookup;
+		    }
 	      }
 
 	  do_lookup:
 	    /* This is the outermost construct with this reduction,
 	       see if there's a mapping for it.  */
 	    if (gimple_code (outer->stmt) == GIMPLE_OMP_TARGET
-		&& maybe_lookup_field (orig, outer))
+		&& maybe_lookup_field (orig, outer) && !is_private)
 	      {
 		ref_to_res = build_receiver_ref (orig, false, outer);
 		if (is_reference (orig))
 		  ref_to_res = build_simple_mem_ref (ref_to_res);
 
+		tree type = TREE_TYPE (var);
+		if (POINTER_TYPE_P (type))
+		  type = TREE_TYPE (type);
+
 		outgoing = var;
-		incoming = omp_reduction_init_op (loc, rcode, TREE_TYPE (var));
+		incoming = omp_reduction_init_op (loc, rcode, type);
 	      }
+	    else if (ctx->outer)
+	      incoming = outgoing = lookup_decl (orig, ctx->outer);
 	    else
 	      incoming = outgoing = orig;
 	      
@@ -5647,6 +5683,37 @@ lower_oacc_reductions (location_t loc, tree clauses, tree level, bool inner,
 	if (!ref_to_res)
 	  ref_to_res = integer_zero_node;
 
+        if (is_reference (orig))
+	  {
+	    tree type = TREE_TYPE (var);
+	    const char *id = IDENTIFIER_POINTER (DECL_NAME (var));
+
+	    if (!inner)
+	      {
+		tree x = create_tmp_var (TREE_TYPE (type), id);
+		gimplify_assign (var, build_fold_addr_expr (x), fork_seq);
+	      }
+
+	    v1 = create_tmp_var (type, id);
+	    v2 = create_tmp_var (type, id);
+	    v3 = create_tmp_var (type, id);
+
+	    gimplify_assign (v1, var, fork_seq);
+	    gimplify_assign (v2, var, fork_seq);
+	    gimplify_assign (v3, var, fork_seq);
+
+	    var = build_simple_mem_ref (var);
+	    v1 = build_simple_mem_ref (v1);
+	    v2 = build_simple_mem_ref (v2);
+	    v3 = build_simple_mem_ref (v3);
+	    outgoing = build_simple_mem_ref (outgoing);
+
+	    if (TREE_CODE (incoming) != INTEGER_CST)
+	      incoming = build_simple_mem_ref (incoming);
+	  }
+	else
+	  v1 = v2 = v3 = var;
+
 	/* Determine position in reduction buffer, which may be used
 	   by target.  */
 	enum machine_mode mode = TYPE_MODE (TREE_TYPE (var));
@@ -5676,20 +5743,20 @@ lower_oacc_reductions (location_t loc, tree clauses, tree level, bool inner,
 	  = build_call_expr_internal_loc (loc, IFN_GOACC_REDUCTION,
 					  TREE_TYPE (var), 6, init_code,
 					  unshare_expr (ref_to_res),
-					  var, level, op, off);
+					  v1, level, op, off);
 	tree fini_call
 	  = build_call_expr_internal_loc (loc, IFN_GOACC_REDUCTION,
 					  TREE_TYPE (var), 6, fini_code,
 					  unshare_expr (ref_to_res),
-					  var, level, op, off);
+					  v2, level, op, off);
 	tree teardown_call
 	  = build_call_expr_internal_loc (loc, IFN_GOACC_REDUCTION,
 					  TREE_TYPE (var), 6, teardown_code,
-					  ref_to_res, var, level, op, off);
+					  ref_to_res, v3, level, op, off);
 
-	gimplify_assign (var, setup_call, &before_fork);
-	gimplify_assign (var, init_call, &after_fork);
-	gimplify_assign (var, fini_call, &before_join);
+	gimplify_assign (v1, setup_call, &before_fork);
+	gimplify_assign (v2, init_call, &after_fork);
+	gimplify_assign (v3, fini_call, &before_join);
 	gimplify_assign (outgoing, teardown_call, &after_join);
       }
 
@@ -5931,9 +5998,6 @@ lower_reduction_clauses (tree clauses, gimple_seq *stmt_seqp, omp_context *ctx)
 	}
     }
 
-  if (is_gimple_omp_oacc (ctx->stmt))
-    return;
-
   stmt = gimple_build_call (builtin_decl_explicit (BUILT_IN_GOMP_ATOMIC_START),
 			    0);
   gimple_seq_add_stmt (stmt_seqp, stmt);
@@ -15820,7 +15884,9 @@ lower_omp_target (gimple_stmt_iterator *gsi_p, omp_context *ctx)
 	if (!maybe_lookup_field (var, ctx))
 	  continue;
 
-	if (offloaded)
+	/* Don't remap oacc parallel reduction variables, because the
+	   intermediate result must be local to each gang.  */
+	if (offloaded && !is_oacc_parallel_reduction (var, ctx))
 	  {
 	    x = build_receiver_ref (var, true, ctx);
 	    tree new_var = lookup_decl (var, ctx);

[-- Attachment #3: pr70533-tests.diff.gz --]
[-- Type: application/gzip, Size: 9838 bytes --]