gimplify.c (gimplify_scan_omp_clauses): No special handling for OMP_CLAUSE_TILE.

2017-02-09  Nathan Sidwell  <nathan@codesourcery.com>
	    Cesar Philippidis  <cesar@codesourcery.com>
	    Joseph Myers  <joseph@codesourcery.com>
	    Chung-Lin Tang  <cltang@codesourcery.com>

	gcc/
	* gimplify.c (gimplify_scan_omp_clauses): No special handling for
	OMP_CLAUSE_TILE.
	(gimplify_adjust_omp_clauses): Don't delete TILE.
	(gimplify_omp_for): Deal with TILE.
	* internal-fn.c (expand_GOACC_TILE): New function.
	* internal-fn.def (GOACC_DIM_POS): Comment may be overly conservative.
	(GOACC_TILE): New.
	* omp-expand.c (struct oacc_collapse): Add tile and outer fields.
	(expand_oacc_collapse_init): Add LOC paramter.  Initialize tile
	element fields.
	(expand_oacc_collapse_vars): Add INNER parm, adjust for tiling,
	avoid DIV for outermost collapse var.
	(expand_oacc_for): Insert tile element loop as needed.  Adjust.
	Remove out of date comments, fix whitespace.
	* omp-general.c (omp_extract_for_data): Deal with tiling.
	* omp-general.h (enum oacc_loop_flags): Add OLF_TILE flag,
	adjust OLF_DIM_BASE value.
	(struct omp_for_data): Add tiling field.
	* omp-low.c (scan_sharing_clauses): Allow OMP_CLAUSE_TILE.
	(lower_oacc_head_mark): Add OLF_TILE as appropriate.  Ensure 2 levels
	for auto loops.  Remove default auto determining, moved to
	oacc_loop_fixed_partitions.
	* omp-offload.c (struct oacc_loop): Change 'ifns' to vector of call
	stmts, add e_mask field.
	(oacc_dim_call): New function, abstracted out from oacc_thread_numbers.
	(oacc_thread_numbers): Use oacc_dim_call.
	(oacc_xform_tile): New.
	(new_oacc_loop_raw): Initialize e_mask, adjust for ifns vector.
	(finish_oacc_loop): Adjust for ifns vector.
	(oacc_loop_discover_walk): Append loop abstraction sites to list,
	add case for GOACC_TILE fns.
	(oacc_loop_xform_loop): Delete.
	(oacc_loop_process): Iterate over call list directly, and add
	handling for GOACC_TILE fns.
	(oacc_loop_fixed_partitions): Determine default auto, deal with TILE,
	dump partitioning.
	(oacc_loop_auto_partitions): Add outer_assign parm. Assign all but
	vector partitioning to outer loops.  Assign 2 partitions to loops
	when available. Add TILE handling.
	(oacc_loop_partition): Adjust oacc_loop_auto_partitions call.
	(execite_oacc_device_lower): Process GOACC_TILE fns, ignore unknown specs.
	* tree-nested.c (convert_nonlocal_omp_clauses): Allow OMP_CLAUSE_TILE.
	* tree.c (omp_clause_num_ops): Adjust TILE ops.
	* tree.h (OMP_CLAUSE_TILE_ITERVAR, OMP_CLAUSE_TILE_COUNT): New.

	gcc/c/
	* c-parser.c (c_parser_omp_clause_collapse): Disallow tile.
	(c_parser_oacc_clause_tile): Disallow collapse. Fix parsing and
	semantic checking.
	* c-parser.c (c_parser_omp_for_loop): Accept tiling constructs.

	gcc/cp/
	* parser.c (cp_parser_oacc_clause_tile): Disallow collapse.  Fix
	parsing.  Parse constant expression. Remove semantic checking.
	(cp_parser_omp_clause_collapse): Disallow tile.
	(cp_parser_omp_for_loop): Deal with tile clause.  Don't emit a parse
	error about missing for after already emitting one.  Use more
	conventional for idiom for unbounded loop.
	* pt.c (tsubst_omp_clauses): Handle OMP_CLAUSE_TILE.
	* semantics.c (finish_omp_clauses): Correct TILE semantic check.
	(finish_omp_for): Deal with tile clause.

	gcc/fortran/
	* openmp.c (resolve_omp_clauses): Error on directives
	containing both tile and collapse clauses.
	(resolve_oacc_loop_blocks): Represent '*' tile arguments as zero.
	* trans-openmp.c (gfc_trans_omp_do): Lower tiled loops like
	collapsed loops.

	gcc/testsuite/
	* c-c++-common/goacc/combined-directives.c: Remove xfail.
	* c-c++-common/goacc/loop-auto-1.c: Adjust and add additional case.
	* c-c++-common/goacc/loop-auto-2.c: New.
	* c-c++-common/goacc/tile.c: Include stdbool, fix expected errors.
	* c-c++-common/goacc/tile-2.c: New.
	* g++.dg/goacc/template.C: Test tile subst.  Adjust erroneous uses.
	* g++.dg/goacc/tile-1.C: New, check tile subst.
	* gcc.dg/goacc/loop-processing-1.c: Adjust dg-final pattern.
	* gfortran.dg/goacc/combined-directives.f90: Remove xfail.
	* gfortran.dg/goacc/tile-1.f90: New test.
	* gfortran.dg/goacc/tile-2.f90: New test.
	* gfortran.dg/goacc/tile-lowering.f95: New test.

	libgomp/
	* testsuite/libgomp.oacc-c-c++-common/tile-1.c: New.
	* testsuite/libgomp.oacc-c-c++-common/loop-auto-1.c: Adjust and
	add additional case.
	* testsuite/libgomp.oacc-c-c++-common/vprop.c: XFAIL under
	"openacc_nvidia_accel_selected".
	* libgomp.oacc-fortran/nested-function-1.f90 (test2):
	Add num_workers(8) clause.

From-SVN: r245300

1 files changed, 21 insertions, 9 deletions

diff --git a/gcc/omp-general.c b/gcc/omp-general.c
index 649dbf3..3f9aec8 100644
--- a/gcc/omp-general.c
+++ b/gcc/omp-general.c
@@ -133,13 +133,9 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
 
   fd->for_stmt = for_stmt;
   fd->pre = NULL;
-  if (gimple_omp_for_collapse (for_stmt) > 1)
-    fd->loops = loops;
-  else
-    fd->loops = &fd->loop;
-
   fd->have_nowait = distribute || simd;
   fd->have_ordered = false;
+  fd->tiling = NULL_TREE;
   fd->collapse = 1;
   fd->ordered = 0;
   fd->sched_kind = OMP_CLAUSE_SCHEDULE_STATIC;
@@ -184,9 +180,22 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
 	    collapse_count = &OMP_CLAUSE_COLLAPSE_COUNT (t);
 	  }
 	break;
+      case OMP_CLAUSE_TILE:
+	fd->tiling = OMP_CLAUSE_TILE_LIST (t);
+	fd->collapse = list_length (fd->tiling);
+	gcc_assert (fd->collapse);
+	collapse_iter = &OMP_CLAUSE_TILE_ITERVAR (t);
+	collapse_count = &OMP_CLAUSE_TILE_COUNT (t);
+	break;
       default:
 	break;
       }
+
+  if (fd->collapse > 1 || fd->tiling)
+    fd->loops = loops;
+  else
+    fd->loops = &fd->loop;
+
   if (fd->ordered && fd->collapse == 1 && loops != NULL)
     {
       fd->loops = loops;
@@ -205,7 +214,7 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
       fd->sched_kind = OMP_CLAUSE_SCHEDULE_STATIC;
       gcc_assert (fd->chunk_size == NULL);
     }
-  gcc_assert (fd->collapse == 1 || collapse_iter != NULL);
+  gcc_assert ((fd->collapse == 1 && !fd->tiling) || collapse_iter != NULL);
   if (taskloop)
     fd->sched_kind = OMP_CLAUSE_SCHEDULE_RUNTIME;
   if (fd->sched_kind == OMP_CLAUSE_SCHEDULE_RUNTIME)
@@ -223,7 +232,10 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
   int cnt = fd->ordered ? fd->ordered : fd->collapse;
   for (i = 0; i < cnt; i++)
     {
-      if (i == 0 && fd->collapse == 1 && (fd->ordered == 0 || loops == NULL))
+      if (i == 0
+	  && fd->collapse == 1
+	  && !fd->tiling
+	  && (fd->ordered == 0 || loops == NULL))
 	loop = &fd->loop;
       else if (loops != NULL)
 	loop = loops + i;
@@ -252,7 +264,7 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
 	  || (fd->sched_kind == OMP_CLAUSE_SCHEDULE_STATIC
 	      && !fd->have_ordered))
 	{
-	  if (fd->collapse == 1)
+	  if (fd->collapse == 1 && !fd->tiling)
 	    iter_type = TREE_TYPE (loop->v);
 	  else if (i == 0
 		   || TYPE_PRECISION (iter_type)
@@ -383,7 +395,7 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
 	*collapse_count = create_tmp_var (iter_type, ".count");
     }
 
-  if (fd->collapse > 1 || (fd->ordered && loops))
+  if (fd->collapse > 1 || fd->tiling || (fd->ordered && loops))
     {
       fd->loop.v = *collapse_iter;
       fd->loop.n1 = build_int_cst (TREE_TYPE (fd->loop.v), 0);