I've committed this to gomp4. It gets a few tile-related things out of the way.
1) we were asserting we never saw tile clauses in a few places. That'll change
soon, and the processing required of them is nothing, so just accept them. We
don't need to gimplify the operands, as they have to be INTEGER_CSTs
2) Broke out OACC_DIM_{SIZE,POS} internal function generation to a worker
function, as I need that soon.
3) Remvoe a stale comment and note OACC_DIM_POS's non-constness might be overly
conservative.
nathan
2016-10-03 Nathan Sidwell <nat...@codesourcery.com>
* gimplify.c (gimplify_scan_omp_clauses): No special handling for
OMP_CLAUSE_TILE.
* omp-low.c (scan_sharing_clauses): Allow OMP_CLAUSE_TILE.
(expand_oacc_for): Remove out of date note. Fix whitespace.
(oacc_dim_call): New.
(oacc_thread_numbers): Use it.
(oacc_loop_fixed_partitions): Dump partitioning.
* tree-nested.c (convert_nonlocal_omp_clauses): Allow OMP_CLAUSE_TILE.
* internal-fn.def (GOACC_DIM_POS): Comment may be overly conservative.
Index: gimplify.c
===================================================================
--- gimplify.c (revision 240724)
+++ gimplify.c (working copy)
@@ -7555,16 +7555,6 @@ gimplify_scan_omp_clauses (tree *list_p,
remove = true;
break;
- case OMP_CLAUSE_TILE:
- for (tree list = OMP_CLAUSE_TILE_LIST (c); !remove && list;
- list = TREE_CHAIN (list))
- {
- if (gimplify_expr (&TREE_VALUE (list), pre_p, NULL,
- is_gimple_val, fb_rvalue) == GS_ERROR)
- remove = true;
- }
- break;
-
case OMP_CLAUSE_DEVICE_RESIDENT:
remove = true;
break;
@@ -7573,6 +7563,7 @@ gimplify_scan_omp_clauses (tree *list_p,
case OMP_CLAUSE_ORDERED:
case OMP_CLAUSE_UNTIED:
case OMP_CLAUSE_COLLAPSE:
+ case OMP_CLAUSE_TILE:
case OMP_CLAUSE_AUTO:
case OMP_CLAUSE_SEQ:
case OMP_CLAUSE_INDEPENDENT:
Index: internal-fn.def
===================================================================
--- internal-fn.def (revision 240724)
+++ internal-fn.def (working copy)
@@ -175,7 +175,7 @@ DEF_INTERNAL_FN (UNIQUE, ECF_NOTHROW, NU
dimension. DIM_POS is pure (and not const) so that it isn't
thought to clobber memory and can be gcse'd within a single
parallel region, but not across FORK/JOIN boundaries. They take a
- single INTEGER_CST argument. */
+ single INTEGER_CST argument. This might be overly conservative. */
DEF_INTERNAL_FN (GOACC_DIM_SIZE, ECF_CONST | ECF_NOTHROW | ECF_LEAF, ".")
DEF_INTERNAL_FN (GOACC_DIM_POS, ECF_PURE | ECF_NOTHROW | ECF_LEAF, ".")
Index: omp-low.c
===================================================================
--- omp-low.c (revision 240724)
+++ omp-low.c (working copy)
@@ -2221,6 +2221,7 @@ scan_sharing_clauses (tree clauses, omp_
case OMP_CLAUSE_INDEPENDENT:
case OMP_CLAUSE_AUTO:
case OMP_CLAUSE_SEQ:
+ case OMP_CLAUSE_TILE:
case OMP_CLAUSE_DEVICE_TYPE:
break;
@@ -2234,7 +2235,6 @@ scan_sharing_clauses (tree clauses, omp_
case OMP_CLAUSE_BIND:
case OMP_CLAUSE_DEVICE_RESIDENT:
case OMP_CLAUSE_NOHOST:
- case OMP_CLAUSE_TILE:
case OMP_CLAUSE__CACHE_:
default:
gcc_unreachable ();
@@ -2395,6 +2395,7 @@ scan_sharing_clauses (tree clauses, omp_
case OMP_CLAUSE_INDEPENDENT:
case OMP_CLAUSE_AUTO:
case OMP_CLAUSE_SEQ:
+ case OMP_CLAUSE_TILE:
case OMP_CLAUSE__GRIDDIM_:
case OMP_CLAUSE_DEVICE_TYPE:
break;
@@ -2402,7 +2403,6 @@ scan_sharing_clauses (tree clauses, omp_
case OMP_CLAUSE_BIND:
case OMP_CLAUSE_DEVICE_RESIDENT:
case OMP_CLAUSE_NOHOST:
- case OMP_CLAUSE_TILE:
case OMP_CLAUSE__CACHE_:
default:
gcc_unreachable ();
@@ -11244,11 +11244,7 @@ expand_omp_taskloop_for_inner (struct om
<exit_bb> [incoming]
V = B + ((range -/+ 1) / S +/- 1) * S [*]
- [*] Needed if V live at end of loop
-
- Note: CHUNKING & GWV mask are specified explicitly here. This is a
- transition, and will be specified by a more general mechanism shortly.
- */
+ [*] Needed if V live at end of loop. */
static void
expand_oacc_for (struct omp_region *region, struct omp_for_data *fd)
@@ -11357,7 +11353,6 @@ expand_oacc_for (struct omp_region *regi
ass = gimple_build_assign (fd->loop.n2, total);
gsi_insert_before (&gsi, ass, GSI_SAME_STMT);
}
-
}
tree b = fd->loop.n1;
@@ -18906,6 +18901,23 @@ omp_finish_file (void)
}
}
+/* Call dim_pos (POS == true) or dim_size (POS == false) builtins for
+ axis DIM. Return a tmp var holding the result. */
+
+static tree
+oacc_dim_call (bool pos, int dim, gimple_seq *seq)
+{
+ tree arg = build_int_cst (unsigned_type_node, dim);
+ tree size = create_tmp_var (integer_type_node);
+ enum internal_fn fn = pos ? IFN_GOACC_DIM_POS : IFN_GOACC_DIM_SIZE;
+ gimple *call = gimple_build_call_internal (fn, 1, arg);
+
+ gimple_call_set_lhs (call, size);
+ gimple_seq_add_stmt (seq, call);
+
+ return size;
+}
+
/* Find the number of threads (POS = false), or thread number (POS =
true) for an OpenACC region partitioned as MASK. Setup code
required for the calculation is added to SEQ. */
@@ -18920,29 +18932,17 @@ oacc_thread_numbers (bool pos, int mask,
for (ix = GOMP_DIM_GANG; ix != GOMP_DIM_MAX; ix++)
if (GOMP_DIM_MASK (ix) & mask)
{
- tree arg = build_int_cst (unsigned_type_node, ix);
-
if (res)
{
/* We had an outer index, so scale that by the size of
this dimension. */
- tree n = create_tmp_var (integer_type_node);
- gimple *call
- = gimple_build_call_internal (IFN_GOACC_DIM_SIZE, 1, arg);
-
- gimple_call_set_lhs (call, n);
- gimple_seq_add_stmt (seq, call);
+ tree n = oacc_dim_call (false, ix, seq);
res = fold_build2 (MULT_EXPR, integer_type_node, res, n);
}
if (pos)
{
/* Determine index in this dimension. */
- tree id = create_tmp_var (integer_type_node);
- gimple *call = gimple_build_call_internal
- (IFN_GOACC_DIM_POS, 1, arg);
-
- gimple_call_set_lhs (call, id);
- gimple_seq_add_stmt (seq, call);
+ tree id = oacc_dim_call (true, ix, seq);
if (res)
res = fold_build2 (PLUS_EXPR, integer_type_node, res, id);
else
@@ -19829,9 +19829,14 @@ oacc_loop_fixed_partitions (oacc_loop *l
}
}
- loop->mask = this_mask;
mask_all |= this_mask;
+ loop->mask = this_mask;
+ if (dump_file)
+ fprintf (dump_file, "Loop %s:%d user specified %d\n",
+ LOCATION_FILE (loop->loc), LOCATION_LINE (loop->loc),
+ loop->mask);
+
if (loop->child)
{
loop->inner = oacc_loop_fixed_partitions (loop->child,
Index: tree-nested.c
===================================================================
--- tree-nested.c (revision 240724)
+++ tree-nested.c (working copy)
@@ -1191,6 +1191,7 @@ convert_nonlocal_omp_clauses (tree *pcla
case OMP_CLAUSE_DEFAULT:
case OMP_CLAUSE_COPYIN:
case OMP_CLAUSE_COLLAPSE:
+ case OMP_CLAUSE_TILE:
case OMP_CLAUSE_UNTIED:
case OMP_CLAUSE_MERGEABLE:
case OMP_CLAUSE_PROC_BIND:
@@ -1207,11 +1208,6 @@ convert_nonlocal_omp_clauses (tree *pcla
/* TODO. */
gcc_unreachable ();
- case OMP_CLAUSE_TILE:
- /* OpenACC tile clauses are discarded during gimplification, so we
- don't expect to see anything here. */
- gcc_unreachable ();
-
case OMP_CLAUSE__CACHE_:
/* These clauses belong to the OpenACC cache directive, which is
discarded during gimplification, so we don't expect to see
@@ -1894,6 +1890,7 @@ convert_local_omp_clauses (tree *pclause
case OMP_CLAUSE_DEFAULT:
case OMP_CLAUSE_COPYIN:
case OMP_CLAUSE_COLLAPSE:
+ case OMP_CLAUSE_TILE:
case OMP_CLAUSE_UNTIED:
case OMP_CLAUSE_MERGEABLE:
case OMP_CLAUSE_PROC_BIND:
@@ -1910,11 +1907,6 @@ convert_local_omp_clauses (tree *pclause
/* TODO. */
gcc_unreachable ();
- case OMP_CLAUSE_TILE:
- /* OpenACC tile clauses are discarded during gimplification, so we
- don't expect to see anything here. */
- gcc_unreachable ();
-
case OMP_CLAUSE__CACHE_:
/* These clauses belong to the OpenACC cache directive, which is
discarded during gimplification, so we don't expect to see
