Hi!
The following patch adds support for lastprivate(conditional:) on for simd.
Bootstrapped/regtested on x86_64-linux and i686-linux, committed to trunk.
2019-06-04 Jakub Jelinek <ja...@redhat.com>
* gimplify.c (gimplify_scan_omp_clauses): Don't sorry_at on lastprivate
conditional on combined for simd.
* omp-low.c (struct omp_context): Add combined_into_simd_safelen0
member.
(lower_rec_input_clauses): For gimple_omp_for_combined_into_p max_vf 1
constructs, don't remove lastprivate_conditional_map, but instead set
ctx->combined_into_simd_safelen0 and adjust hash_map, so that it points
to parent construct temporaries.
(lower_lastprivate_clauses): Handle ctx->combined_into_simd_safelen0
like !ctx->lastprivate_conditional_map.
(lower_omp_1) <case GIMPLE_ASSIGN>: If up->combined_into_simd_safelen0,
use up->outer context instead of up.
* omp-expand.c (expand_omp_for_generic): Perform cond_var bump even if
gimple_omp_for_combined_p.
(expand_omp_for_static_nochunk): Likewise.
(expand_omp_for_static_chunk): Add forgotten cond_var bump that was
probably moved over into expand_omp_for_generic rather than being copied
there.
gcc/cp/
* cp-tree.h (CP_OMP_CLAUSE_INFO): Allow for any clauses up to _condvar_
instead of only up to linear.
gcc/testsuite/
* c-c++-common/gomp/lastprivate-conditional-2.c (foo): Don't expect
a sorry_at on any of the clauses.
libgomp/
* testsuite/libgomp.c-c++-common/lastprivate-conditional-7.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-8.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-9.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-10.c: New test.
--- gcc/gimplify.c.jj 2019-05-31 23:37:51.066938833 +0200
+++ gcc/gimplify.c 2019-06-03 10:22:30.272829191 +0200
@@ -8145,31 +8145,7 @@ gimplify_scan_omp_clauses (tree *list_p,
OMP_CLAUSE_LASTPRIVATE_CONDITIONAL (c) = 0;
}
if (OMP_CLAUSE_LASTPRIVATE_CONDITIONAL (c))
- {
- splay_tree_node n = NULL;
- if (code == OMP_SIMD
- && outer_ctx
- && outer_ctx->region_type == ORT_WORKSHARE)
- {
- n = splay_tree_lookup (outer_ctx->variables,
- (splay_tree_key) decl);
- if (n == NULL
- && outer_ctx->outer_context
- && (outer_ctx->outer_context->region_type
- == ORT_COMBINED_PARALLEL))
- n = splay_tree_lookup (outer_ctx->outer_context->variables,
- (splay_tree_key) decl);
- }
- if (n && (n->value & GOVD_LASTPRIVATE_CONDITIONAL) != 0)
- {
- sorry_at (OMP_CLAUSE_LOCATION (c),
- "%<conditional%> modifier on %<lastprivate%> "
- "clause not supported yet");
- OMP_CLAUSE_LASTPRIVATE_CONDITIONAL (c) = 0;
- }
- else
- flags |= GOVD_LASTPRIVATE_CONDITIONAL;
- }
+ flags |= GOVD_LASTPRIVATE_CONDITIONAL;
if (outer_ctx
&& (outer_ctx->region_type == ORT_COMBINED_PARALLEL
|| ((outer_ctx->region_type & ORT_COMBINED_TEAMS)
--- gcc/omp-low.c.jj 2019-05-31 23:37:51.069938786 +0200
+++ gcc/omp-low.c 2019-06-03 15:12:45.535613313 +0200
@@ -137,6 +137,10 @@ struct omp_context
/* True if this construct can be cancelled. */
bool cancellable;
+
+ /* True if lower_omp_1 should look up lastprivate conditional in parent
+ context. */
+ bool combined_into_simd_safelen0;
};
static splay_tree all_contexts;
@@ -4816,6 +4820,8 @@ lower_rec_input_clauses (tree clauses, g
void_node);
gimple_seq tseq = NULL;
gimplify_and_add (x, &tseq);
+ if (ctx->outer)
+ lower_omp (&tseq, ctx->outer);
gimple_seq_add_seq (&llist[1], tseq);
}
if (y)
@@ -5278,11 +5284,31 @@ lower_rec_input_clauses (tree clauses, g
sctx.is_simt = false;
if (ctx->lastprivate_conditional_map)
{
- /* When not vectorized, treat lastprivate(conditional:) like
- normal lastprivate, as there will be just one simd lane
- writing the privatized variable. */
- delete ctx->lastprivate_conditional_map;
- ctx->lastprivate_conditional_map = NULL;
+ if (gimple_omp_for_combined_into_p (ctx->stmt))
+ {
+ /* Signal to lower_omp_1 that it should use parent context. */
+ ctx->combined_into_simd_safelen0 = true;
+ for (c = clauses; c ; c = OMP_CLAUSE_CHAIN (c))
+ if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LASTPRIVATE
+ && OMP_CLAUSE_LASTPRIVATE_CONDITIONAL (c))
+ {
+ tree o = lookup_decl (OMP_CLAUSE_DECL (c), ctx);
+ tree *v
+ = ctx->lastprivate_conditional_map->get (o);
+ tree po = lookup_decl (OMP_CLAUSE_DECL (c), ctx->outer);
+ tree *pv
+ = ctx->outer->lastprivate_conditional_map->get (po);
+ *v = *pv;
+ }
+ }
+ else
+ {
+ /* When not vectorized, treat lastprivate(conditional:) like
+ normal lastprivate, as there will be just one simd lane
+ writing the privatized variable. */
+ delete ctx->lastprivate_conditional_map;
+ ctx->lastprivate_conditional_map = NULL;
+ }
}
}
@@ -5652,7 +5678,8 @@ lower_lastprivate_clauses (tree clauses,
if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_LASTPRIVATE
&& OMP_CLAUSE_LASTPRIVATE_CONDITIONAL (c)
- && ctx->lastprivate_conditional_map)
+ && ctx->lastprivate_conditional_map
+ && !ctx->combined_into_simd_safelen0)
{
gcc_assert (body_p);
if (simduid)
@@ -10812,6 +10839,8 @@ lower_omp_1 (gimple_stmt_iterator *gsi_p
if (tree *v = up->lastprivate_conditional_map->get (lhs))
{
tree clauses;
+ if (up->combined_into_simd_safelen0)
+ up = up->outer;
if (gimple_code (up->stmt) == GIMPLE_OMP_FOR)
clauses = gimple_omp_for_clauses (up->stmt);
else
--- gcc/omp-expand.c.jj 2019-05-31 23:37:51.071938754 +0200
+++ gcc/omp-expand.c 2019-06-03 12:04:20.342277395 +0200
@@ -3257,6 +3257,25 @@ expand_omp_for_generic (struct omp_regio
vmain = gimple_omp_continue_control_use (cont_stmt);
vback = gimple_omp_continue_control_def (cont_stmt);
+ if (cond_var)
+ {
+ tree itype = TREE_TYPE (cond_var);
+ tree t2;
+ if ((fd->ordered && fd->collapse == 1)
+ || bias
+ || POINTER_TYPE_P (type)
+ || TREE_CODE (fd->loop.n1) != INTEGER_CST
+ || fd->loop.cond_code != LT_EXPR)
+ t2 = build_int_cst (itype, 1);
+ else
+ t2 = fold_convert (itype, fd->loop.step);
+ t2 = fold_build2 (PLUS_EXPR, itype, cond_var, t2);
+ t2 = force_gimple_operand_gsi (&gsi, t2, false,
+ NULL_TREE, true, GSI_SAME_STMT);
+ assign_stmt = gimple_build_assign (cond_var, t2);
+ gsi_insert_before (&gsi, assign_stmt, GSI_SAME_STMT);
+ }
+
if (!gimple_omp_for_combined_p (fd->for_stmt))
{
if (POINTER_TYPE_P (type))
@@ -3270,25 +3289,6 @@ expand_omp_for_generic (struct omp_regio
assign_stmt = gimple_build_assign (vback, t);
gsi_insert_before (&gsi, assign_stmt, GSI_SAME_STMT);
- if (cond_var)
- {
- tree itype = TREE_TYPE (cond_var);
- tree t2;
- if ((fd->ordered && fd->collapse == 1)
- || bias
- || POINTER_TYPE_P (type)
- || TREE_CODE (fd->loop.n1) != INTEGER_CST
- || fd->loop.cond_code != LT_EXPR)
- t2 = build_int_cst (itype, 1);
- else
- t2 = fold_convert (itype, fd->loop.step);
- t2 = fold_build2 (PLUS_EXPR, itype, cond_var, t2);
- t2 = force_gimple_operand_gsi (&gsi, t2, false,
- NULL_TREE, true, GSI_SAME_STMT);
- assign_stmt = gimple_build_assign (cond_var, t2);
- gsi_insert_before (&gsi, assign_stmt, GSI_SAME_STMT);
- }
-
if (fd->ordered && counts[fd->collapse - 1] == NULL_TREE)
{
tree tem;
@@ -3962,6 +3962,23 @@ expand_omp_for_static_nochunk (struct om
vmain = gimple_omp_continue_control_use (cont_stmt);
vback = gimple_omp_continue_control_def (cont_stmt);
+ if (cond_var)
+ {
+ tree itype = TREE_TYPE (cond_var);
+ tree t2;
+ if (POINTER_TYPE_P (type)
+ || TREE_CODE (n1) != INTEGER_CST
+ || fd->loop.cond_code != LT_EXPR)
+ t2 = build_int_cst (itype, 1);
+ else
+ t2 = fold_convert (itype, step);
+ t2 = fold_build2 (PLUS_EXPR, itype, cond_var, t2);
+ t2 = force_gimple_operand_gsi (&gsi, t2, false,
+ NULL_TREE, true, GSI_SAME_STMT);
+ assign_stmt = gimple_build_assign (cond_var, t2);
+ gsi_insert_before (&gsi, assign_stmt, GSI_SAME_STMT);
+ }
+
if (!gimple_omp_for_combined_p (fd->for_stmt))
{
if (POINTER_TYPE_P (type))
@@ -3975,23 +3992,6 @@ expand_omp_for_static_nochunk (struct om
assign_stmt = gimple_build_assign (vback, t);
gsi_insert_before (&gsi, assign_stmt, GSI_SAME_STMT);
- if (cond_var)
- {
- tree itype = TREE_TYPE (cond_var);
- tree t2;
- if (POINTER_TYPE_P (type)
- || TREE_CODE (n1) != INTEGER_CST
- || fd->loop.cond_code != LT_EXPR)
- t2 = build_int_cst (itype, 1);
- else
- t2 = fold_convert (itype, step);
- t2 = fold_build2 (PLUS_EXPR, itype, cond_var, t2);
- t2 = force_gimple_operand_gsi (&gsi, t2, false,
- NULL_TREE, true, GSI_SAME_STMT);
- assign_stmt = gimple_build_assign (cond_var, t2);
- gsi_insert_before (&gsi, assign_stmt, GSI_SAME_STMT);
- }
-
t = build2 (fd->loop.cond_code, boolean_type_node,
DECL_P (vback) && TREE_ADDRESSABLE (vback)
? t : vback, e);
@@ -4607,6 +4607,23 @@ expand_omp_for_static_chunk (struct omp_
vmain = gimple_omp_continue_control_use (cont_stmt);
vback = gimple_omp_continue_control_def (cont_stmt);
+ if (cond_var)
+ {
+ tree itype = TREE_TYPE (cond_var);
+ tree t2;
+ if (POINTER_TYPE_P (type)
+ || TREE_CODE (n1) != INTEGER_CST
+ || fd->loop.cond_code != LT_EXPR)
+ t2 = build_int_cst (itype, 1);
+ else
+ t2 = fold_convert (itype, step);
+ t2 = fold_build2 (PLUS_EXPR, itype, cond_var, t2);
+ t2 = force_gimple_operand_gsi (&gsi, t2, false,
+ NULL_TREE, true, GSI_SAME_STMT);
+ assign_stmt = gimple_build_assign (cond_var, t2);
+ gsi_insert_before (&gsi, assign_stmt, GSI_SAME_STMT);
+ }
+
if (!gimple_omp_for_combined_p (fd->for_stmt))
{
if (POINTER_TYPE_P (type))
--- gcc/cp/cp-tree.h.jj 2019-05-31 23:35:38.461038702 +0200
+++ gcc/cp/cp-tree.h 2019-06-03 12:39:21.453395398 +0200
@@ -4924,7 +4924,7 @@ more_aggr_init_expr_args_p (const aggr_i
See semantics.c for details. */
#define CP_OMP_CLAUSE_INFO(NODE) \
TREE_TYPE (OMP_CLAUSE_RANGE_CHECK (NODE, OMP_CLAUSE_PRIVATE, \
- OMP_CLAUSE_LINEAR))
+ OMP_CLAUSE__CONDTEMP_))
/* Nonzero if this transaction expression's body contains statements. */
#define TRANSACTION_EXPR_IS_STMT(NODE) \
--- gcc/testsuite/c-c++-common/gomp/lastprivate-conditional-2.c.jj
2019-05-31 23:37:51.071938754 +0200
+++ gcc/testsuite/c-c++-common/gomp/lastprivate-conditional-2.c 2019-06-03
12:25:31.233599215 +0200
@@ -13,7 +13,7 @@ foo (int *p)
if (p[i])
b = i;
#pragma omp parallel
- #pragma omp for simd lastprivate (conditional: c) /* { dg-message "not
supported yet" } */
+ #pragma omp for simd lastprivate (conditional: c)
for (i = 0; i < 32; i++)
if (p[i])
c = i;
@@ -21,7 +21,7 @@ foo (int *p)
for (i = 0; i < 32; i++)
if (p[i])
d = i;
- #pragma omp parallel for simd lastprivate (conditional: e) /* { dg-message
"not supported yet" } */
+ #pragma omp parallel for simd lastprivate (conditional: e)
for (i = 0; i < 32; i++)
if (p[i])
e = i;
--- libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-7.c.jj
2019-06-03 12:30:01.087307199 +0200
+++ libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-7.c
2019-06-03 12:57:18.597254143 +0200
@@ -0,0 +1,63 @@
+/* { dg-do run } */
+/* { dg-additional-options "-O2 -fdump-tree-vect-details" } */
+/* { dg-additional-options "-mavx" { target avx_runtime } } */
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 3 "vect" { target
avx_runtime } } } */
+
+int v, x;
+
+__attribute__((noipa)) void
+foo (int *a)
+{
+ #pragma omp for simd lastprivate (conditional: x) schedule(simd: static)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i];
+}
+
+__attribute__((noipa)) void
+bar (int *a, int *b)
+{
+ #pragma omp for simd lastprivate (conditional: x, v) schedule(static, 16)
+ for (int i = 16; i < 128; ++i)
+ {
+ if (a[i])
+ x = a[i];
+ if (b[i])
+ v = b[i] + 10;
+ }
+}
+
+__attribute__((noipa)) void
+baz (int *a)
+{
+ #pragma omp for simd lastprivate (conditional: x) schedule(simd: dynamic, 16)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i] + 5;
+}
+
+int
+main ()
+{
+ int a[128], b[128], i;
+ for (i = 0; i < 128; i++)
+ {
+ a[i] = ((i % 11) == 2) ? i + 10 : 0;
+ asm volatile ("" : "+g" (i));
+ b[i] = ((i % 13) == 5) ? i * 2 : 0;
+ }
+ #pragma omp parallel
+ foo (a);
+ if (x != 133)
+ __builtin_abort ();
+ x = -3;
+ #pragma omp parallel
+ bar (b, a);
+ if (x != 244 || v != 143)
+ __builtin_abort ();
+ #pragma omp parallel
+ baz (b);
+ if (x != 249)
+ __builtin_abort ();
+ return 0;
+}
--- libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-8.c.jj
2019-06-03 13:01:22.102373429 +0200
+++ libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-8.c
2019-06-03 13:02:38.295159149 +0200
@@ -0,0 +1,60 @@
+/* { dg-do run } */
+/* { dg-additional-options "-O2 -fdump-tree-vect-details" } */
+/* { dg-additional-options "-mavx" { target avx_runtime } } */
+/* { dg-final { scan-tree-dump-times "vectorized 1 loops" 3 "vect" { target
avx_runtime } } } */
+
+int v, x;
+
+__attribute__((noipa)) int
+foo (int *a)
+{
+ #pragma omp parallel for simd lastprivate (conditional: x) schedule(simd:
static)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i];
+ return x;
+}
+
+__attribute__((noipa)) int
+bar (int *a, int *b)
+{
+ #pragma omp parallel
+ #pragma omp for simd lastprivate (conditional: x, v) schedule(static, 16)
+ for (int i = 16; i < 128; ++i)
+ {
+ if (a[i])
+ x = a[i];
+ if (b[i])
+ v = b[i] + 10;
+ }
+ return x;
+}
+
+__attribute__((noipa)) int
+baz (int *a)
+{
+ #pragma omp parallel for simd lastprivate (conditional: x) schedule(simd:
dynamic, 16)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i] + 5;
+ return x;
+}
+
+int
+main ()
+{
+ int a[128], b[128], i;
+ for (i = 0; i < 128; i++)
+ {
+ a[i] = ((i % 11) == 2) ? i + 10 : 0;
+ asm volatile ("" : "+g" (i));
+ b[i] = ((i % 13) == 5) ? i * 2 : 0;
+ }
+ if (foo (a) != 133)
+ __builtin_abort ();
+ if (bar (b, a) != 244 || v != 143)
+ __builtin_abort ();
+ if (baz (b) != 249)
+ __builtin_abort ();
+ return 0;
+}
--- libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-9.c.jj
2019-06-03 13:10:44.860421306 +0200
+++ libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-9.c
2019-06-03 13:11:32.053675094 +0200
@@ -0,0 +1,60 @@
+/* { dg-do run } */
+
+int v, x;
+
+__attribute__((noipa)) void
+foo (int *a)
+{
+ #pragma omp for simd lastprivate (conditional: x) schedule(simd: static) if
(simd: 0)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i];
+}
+
+__attribute__((noipa)) void
+bar (int *a, int *b)
+{
+ #pragma omp for simd lastprivate (conditional: x, v) schedule(static, 16)
simdlen(1)
+ for (int i = 16; i < 128; ++i)
+ {
+ if (a[i])
+ x = a[i];
+ if (b[i])
+ v = b[i] + 10;
+ }
+}
+
+__attribute__((noipa)) void
+baz (int *a)
+{
+ #pragma omp for simd lastprivate (conditional: x) schedule(simd: dynamic,
16) if (0)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i] + 5;
+}
+
+int
+main ()
+{
+ int a[128], b[128], i;
+ for (i = 0; i < 128; i++)
+ {
+ a[i] = ((i % 11) == 2) ? i + 10 : 0;
+ asm volatile ("" : "+g" (i));
+ b[i] = ((i % 13) == 5) ? i * 2 : 0;
+ }
+ #pragma omp parallel
+ foo (a);
+ if (x != 133)
+ __builtin_abort ();
+ x = -3;
+ #pragma omp parallel
+ bar (b, a);
+ if (x != 244 || v != 143)
+ __builtin_abort ();
+ #pragma omp parallel
+ baz (b);
+ if (x != 249)
+ __builtin_abort ();
+ return 0;
+}
--- libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-10.c.jj
2019-06-03 15:17:33.368060408 +0200
+++ libgomp/testsuite/libgomp.c-c++-common/lastprivate-conditional-10.c
2019-06-03 15:19:11.717504728 +0200
@@ -0,0 +1,57 @@
+/* { dg-do run } */
+
+int v, x;
+
+__attribute__((noipa)) int
+foo (int *a)
+{
+ #pragma omp parallel for simd lastprivate (conditional: x) schedule(simd:
static) if (simd: 0)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i];
+ return x;
+}
+
+__attribute__((noipa)) int
+bar (int *a, int *b)
+{
+ #pragma omp parallel
+ #pragma omp for simd lastprivate (conditional: x, v) schedule(static, 16)
simdlen (1)
+ for (int i = 16; i < 128; ++i)
+ {
+ if (a[i])
+ x = a[i];
+ if (b[i])
+ v = b[i] + 10;
+ }
+ return x;
+}
+
+__attribute__((noipa)) int
+baz (int *a)
+{
+ #pragma omp parallel for simd if (simd: 0) lastprivate (conditional: x)
schedule(simd: dynamic, 16)
+ for (int i = 0; i < 128; i++)
+ if (a[i])
+ x = a[i] + 5;
+ return x;
+}
+
+int
+main ()
+{
+ int a[128], b[128], i;
+ for (i = 0; i < 128; i++)
+ {
+ a[i] = ((i % 11) == 2) ? i + 10 : 0;
+ asm volatile ("" : "+g" (i));
+ b[i] = ((i % 13) == 5) ? i * 2 : 0;
+ }
+ if (foo (a) != 133)
+ __builtin_abort ();
+ if (bar (b, a) != 244 || v != 143)
+ __builtin_abort ();
+ if (baz (b) != 249)
+ __builtin_abort ();
+ return 0;
+}
Jakub
