public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r12-4035] openmp: Differentiate between order(concurrent) and order(reproducible:concurrent)
@ 2021-10-01 8:55 Jakub Jelinek
0 siblings, 0 replies; only message in thread
From: Jakub Jelinek @ 2021-10-01 8:55 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:e705b8533aa0a00a65734eb5fd6344295723dccc
commit r12-4035-ge705b8533aa0a00a65734eb5fd6344295723dccc
Author: Jakub Jelinek <jakub@redhat.com>
Date: Fri Oct 1 10:45:48 2021 +0200
openmp: Differentiate between order(concurrent) and order(reproducible:concurrent)
While OpenMP 5.1 implies order(concurrent) is the same thing as
order(reproducible:concurrent), this is going to change in OpenMP 5.2, where
essentially order(concurrent) means nothing is stated on whether it is
reproducible or unconstrained (and is determined by other means, e.g. for/do
with schedule static or runtime with static being selected is implicitly
reproducible, distribute with dist_schedule static is implicitly reproducible,
loop is implicitly reproducible) and when the modifier is specified explicitly,
it overrides the implicit behavior either way.
And, when order(reproducible:concurrent) is used with e.g. schedule(dynamic)
or some other schedule that is by definition not reproducible, it is
implementation's duty to ensure it is reproducible, either by remembering how
it scheduled some loop and then replaying the same schedule when seeing loops
with the same directive/schedule/number of iterations, or by overriding the
schedule to some reproducible one.
This patch doesn't implement the 5.2 wording just yet, but in the FEs
differentiates between the 3 states - no explicit modifier, explicit reproducible
or explicit unconstrainted, so that the middle-end can easily switch any time.
Instead it follows the 5.1 wording where both order(concurrent) (implicit or
explicit) or order(reproducible:concurrent) imply reproducibility.
And, it implements the easier method, when for/do should be reproducible, it
just chooses static schedule. order(concurrent) implies no OpenMP APIs in the
loop body nor threadprivate vars, so the exact scheduling isn't (easily at least)
observable.
2021-10-01 Jakub Jelinek <jakub@redhat.com>
gcc/
* tree.h (OMP_CLAUSE_ORDER_REPRODUCIBLE): Define.
* tree-pretty-print.c (dump_omp_clause) <case OMP_CLAUSE_ORDER>: Print
reproducible: for OMP_CLAUSE_ORDER_REPRODUCIBLE.
* omp-general.c (omp_extract_for_data): If OMP_CLAUSE_ORDER is seen
without OMP_CLAUSE_ORDER_UNCONSTRAINED, overwrite sched_kind to
OMP_CLAUSE_SCHEDULE_STATIC.
gcc/c-family/
* c-omp.c (c_omp_split_clauses): Also copy
OMP_CLAUSE_ORDER_REPRODUCIBLE.
gcc/c/
* c-parser.c (c_parser_omp_clause_order): Set
OMP_CLAUSE_ORDER_REPRODUCIBLE for explicit reproducible: modifier.
gcc/cp/
* parser.c (cp_parser_omp_clause_order): Set
OMP_CLAUSE_ORDER_REPRODUCIBLE for explicit reproducible: modifier.
gcc/fortran/
* gfortran.h (gfc_omp_clauses): Add order_reproducible bitfield.
* dump-parse-tree.c (show_omp_clauses): Print REPRODUCIBLE: for it.
* openmp.c (gfc_match_omp_clauses): Set order_reproducible for
explicit reproducible: modifier.
* trans-openmp.c (gfc_trans_omp_clauses): Set
OMP_CLAUSE_ORDER_REPRODUCIBLE for order_reproducible.
(gfc_split_omp_clauses): Also copy order_reproducible.
gcc/testsuite/
* gfortran.dg/gomp/order-5.f90: Adjust scan-tree-dump-times regexps.
libgomp/
* testsuite/libgomp.c-c++-common/order-reproducible-1.c: New test.
* testsuite/libgomp.c-c++-common/order-reproducible-2.c: New test.
Diff:
---
gcc/c-family/c-omp.c | 4 ++
gcc/c/c-parser.c | 6 ++-
gcc/cp/parser.c | 6 ++-
gcc/fortran/dump-parse-tree.c | 2 +
gcc/fortran/gfortran.h | 4 +-
gcc/fortran/openmp.c | 5 +-
gcc/fortran/trans-openmp.c | 7 +++
gcc/omp-general.c | 16 ++++++
gcc/testsuite/gfortran.dg/gomp/order-5.f90 | 8 +--
gcc/tree-pretty-print.c | 2 +
gcc/tree.h | 3 ++
.../libgomp.c-c++-common/order-reproducible-1.c | 63 ++++++++++++++++++++++
.../libgomp.c-c++-common/order-reproducible-2.c | 28 ++++++++++
13 files changed, 144 insertions(+), 10 deletions(-)
diff --git a/gcc/c-family/c-omp.c b/gcc/c-family/c-omp.c
index 1f07a0a454b..2849fdabc3d 100644
--- a/gcc/c-family/c-omp.c
+++ b/gcc/c-family/c-omp.c
@@ -2128,6 +2128,8 @@ c_omp_split_clauses (location_t loc, enum tree_code code,
OMP_CLAUSE_ORDER);
OMP_CLAUSE_ORDER_UNCONSTRAINED (c)
= OMP_CLAUSE_ORDER_UNCONSTRAINED (clauses);
+ OMP_CLAUSE_ORDER_REPRODUCIBLE (c)
+ = OMP_CLAUSE_ORDER_REPRODUCIBLE (clauses);
OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_DISTRIBUTE];
cclauses[C_OMP_CLAUSE_SPLIT_DISTRIBUTE] = c;
}
@@ -2139,6 +2141,8 @@ c_omp_split_clauses (location_t loc, enum tree_code code,
OMP_CLAUSE_ORDER);
OMP_CLAUSE_ORDER_UNCONSTRAINED (c)
= OMP_CLAUSE_ORDER_UNCONSTRAINED (clauses);
+ OMP_CLAUSE_ORDER_REPRODUCIBLE (c)
+ = OMP_CLAUSE_ORDER_REPRODUCIBLE (clauses);
OMP_CLAUSE_CHAIN (c) = cclauses[C_OMP_CLAUSE_SPLIT_FOR];
cclauses[C_OMP_CLAUSE_SPLIT_FOR] = c;
s = C_OMP_CLAUSE_SPLIT_SIMD;
diff --git a/gcc/c/c-parser.c b/gcc/c/c-parser.c
index 21ff25d3b2c..a66f43f6dc2 100644
--- a/gcc/c/c-parser.c
+++ b/gcc/c/c-parser.c
@@ -14626,6 +14626,7 @@ c_parser_omp_clause_order (c_parser *parser, tree list)
tree c;
const char *p;
bool unconstrained = false;
+ bool reproducible = false;
matching_parens parens;
if (!parens.require_open (parser))
@@ -14636,7 +14637,9 @@ c_parser_omp_clause_order (c_parser *parser, tree list)
p = IDENTIFIER_POINTER (c_parser_peek_token (parser)->value);
if (strcmp (p, "unconstrained") == 0)
unconstrained = true;
- else if (strcmp (p, "reproducible") != 0)
+ else if (strcmp (p, "reproducible") == 0)
+ reproducible = true;
+ else
{
c_parser_error (parser, "expected %<reproducible%> or "
"%<unconstrained%>");
@@ -14661,6 +14664,7 @@ c_parser_omp_clause_order (c_parser *parser, tree list)
check_no_duplicate_clause (list, OMP_CLAUSE_ORDER, "order");
c = build_omp_clause (loc, OMP_CLAUSE_ORDER);
OMP_CLAUSE_ORDER_UNCONSTRAINED (c) = unconstrained;
+ OMP_CLAUSE_ORDER_REPRODUCIBLE (c) = reproducible;
OMP_CLAUSE_CHAIN (c) = list;
return c;
diff --git a/gcc/cp/parser.c b/gcc/cp/parser.c
index 04f5a24cc03..c6f1a9796c5 100644
--- a/gcc/cp/parser.c
+++ b/gcc/cp/parser.c
@@ -37735,6 +37735,7 @@ cp_parser_omp_clause_order (cp_parser *parser, tree list, location_t location)
tree c, id;
const char *p;
bool unconstrained = false;
+ bool reproducible = false;
matching_parens parens;
if (!parens.require_open (parser))
@@ -37747,7 +37748,9 @@ cp_parser_omp_clause_order (cp_parser *parser, tree list, location_t location)
p = IDENTIFIER_POINTER (id);
if (strcmp (p, "unconstrained") == 0)
unconstrained = true;
- else if (strcmp (p, "reproducible") != 0)
+ else if (strcmp (p, "reproducible") == 0)
+ reproducible = true;
+ else
{
cp_parser_error (parser, "expected %<reproducible%> or "
"%<unconstrained%>");
@@ -37778,6 +37781,7 @@ cp_parser_omp_clause_order (cp_parser *parser, tree list, location_t location)
check_no_duplicate_clause (list, OMP_CLAUSE_ORDER, "order", location);
c = build_omp_clause (location, OMP_CLAUSE_ORDER);
OMP_CLAUSE_ORDER_UNCONSTRAINED (c) = unconstrained;
+ OMP_CLAUSE_ORDER_REPRODUCIBLE (c) = reproducible;
OMP_CLAUSE_CHAIN (c) = list;
return c;
diff --git a/gcc/fortran/dump-parse-tree.c b/gcc/fortran/dump-parse-tree.c
index 28eb09e261d..64e04c043f6 100644
--- a/gcc/fortran/dump-parse-tree.c
+++ b/gcc/fortran/dump-parse-tree.c
@@ -1634,6 +1634,8 @@ show_omp_clauses (gfc_omp_clauses *omp_clauses)
fputs (" ORDER(", dumpfile);
if (omp_clauses->order_unconstrained)
fputs ("UNCONSTRAINED:", dumpfile);
+ else if (omp_clauses->order_reproducible)
+ fputs ("REPRODUCIBLE:", dumpfile);
fputs ("CONCURRENT)", dumpfile);
}
if (omp_clauses->ordered)
diff --git a/gcc/fortran/gfortran.h b/gcc/fortran/gfortran.h
index 7ef835b211a..c25d1cca3a8 100644
--- a/gcc/fortran/gfortran.h
+++ b/gcc/fortran/gfortran.h
@@ -1491,8 +1491,8 @@ typedef struct gfc_omp_clauses
unsigned inbranch:1, notinbranch:1, nogroup:1;
unsigned sched_simd:1, sched_monotonic:1, sched_nonmonotonic:1;
unsigned simd:1, threads:1, depend_source:1, destroy:1, order_concurrent:1;
- unsigned order_unconstrained:1, capture:1, grainsize_strict:1;
- unsigned num_tasks_strict:1;
+ unsigned order_unconstrained:1, order_reproducible:1, capture:1;
+ unsigned grainsize_strict:1, num_tasks_strict:1;
ENUM_BITFIELD (gfc_omp_sched_kind) sched_kind:3;
ENUM_BITFIELD (gfc_omp_device_type) device_type:2;
ENUM_BITFIELD (gfc_omp_memorder) memorder:3;
diff --git a/gcc/fortran/openmp.c b/gcc/fortran/openmp.c
index 9ee52d6b0ea..6a4ca2868f8 100644
--- a/gcc/fortran/openmp.c
+++ b/gcc/fortran/openmp.c
@@ -2374,8 +2374,9 @@ gfc_match_omp_clauses (gfc_omp_clauses **cp, const omp_mask mask,
{
if (m == MATCH_ERROR)
goto error;
- if (gfc_match (" reproducible : concurrent )") == MATCH_YES
- || gfc_match (" concurrent )") == MATCH_YES)
+ if (gfc_match (" reproducible : concurrent )") == MATCH_YES)
+ c->order_reproducible = true;
+ else if (gfc_match (" concurrent )") == MATCH_YES)
;
else if (gfc_match (" unconstrained : concurrent )") == MATCH_YES)
c->order_unconstrained = true;
diff --git a/gcc/fortran/trans-openmp.c b/gcc/fortran/trans-openmp.c
index 4ca2c3f9e7f..d234d1b070f 100644
--- a/gcc/fortran/trans-openmp.c
+++ b/gcc/fortran/trans-openmp.c
@@ -3804,6 +3804,7 @@ gfc_trans_omp_clauses (stmtblock_t *block, gfc_omp_clauses *clauses,
{
c = build_omp_clause (gfc_get_location (&where), OMP_CLAUSE_ORDER);
OMP_CLAUSE_ORDER_UNCONSTRAINED (c) = clauses->order_unconstrained;
+ OMP_CLAUSE_ORDER_REPRODUCIBLE (c) = clauses->order_reproducible;
omp_clauses = gfc_trans_add_clause (c, omp_clauses);
}
@@ -5895,6 +5896,8 @@ gfc_split_omp_clauses (gfc_code *code,
= code->ext.omp_clauses->order_concurrent;
clausesa[GFC_OMP_SPLIT_DISTRIBUTE].order_unconstrained
= code->ext.omp_clauses->order_unconstrained;
+ clausesa[GFC_OMP_SPLIT_DISTRIBUTE].order_reproducible
+ = code->ext.omp_clauses->order_reproducible;
}
if (mask & GFC_OMP_MASK_PARALLEL)
{
@@ -5951,6 +5954,8 @@ gfc_split_omp_clauses (gfc_code *code,
= code->ext.omp_clauses->order_concurrent;
clausesa[GFC_OMP_SPLIT_DO].order_unconstrained
= code->ext.omp_clauses->order_unconstrained;
+ clausesa[GFC_OMP_SPLIT_DO].order_reproducible
+ = code->ext.omp_clauses->order_reproducible;
}
if (mask & GFC_OMP_MASK_SIMD)
{
@@ -5969,6 +5974,8 @@ gfc_split_omp_clauses (gfc_code *code,
= code->ext.omp_clauses->order_concurrent;
clausesa[GFC_OMP_SPLIT_SIMD].order_unconstrained
= code->ext.omp_clauses->order_unconstrained;
+ clausesa[GFC_OMP_SPLIT_SIMD].order_reproducible
+ = code->ext.omp_clauses->order_reproducible;
/* And this is copied to all. */
clausesa[GFC_OMP_SPLIT_SIMD].if_expr
= code->ext.omp_clauses->if_expr;
diff --git a/gcc/omp-general.c b/gcc/omp-general.c
index cc6aecb1d66..1e4c0b25531 100644
--- a/gcc/omp-general.c
+++ b/gcc/omp-general.c
@@ -193,6 +193,7 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
== GF_OMP_FOR_KIND_DISTRIBUTE;
bool taskloop = gimple_omp_for_kind (for_stmt)
== GF_OMP_FOR_KIND_TASKLOOP;
+ bool order_reproducible = false;
tree iterv, countv;
fd->for_stmt = for_stmt;
@@ -277,10 +278,25 @@ omp_extract_for_data (gomp_for *for_stmt, struct omp_for_data *fd,
&& !OMP_CLAUSE__SCANTEMP__CONTROL (t))
fd->have_nonctrl_scantemp = true;
break;
+ case OMP_CLAUSE_ORDER:
+ /* FIXME: For OpenMP 5.2 this should change to
+ if (OMP_CLAUSE_ORDER_REPRODUCIBLE (t))
+ (with the exception of loop construct but that lowers to
+ no schedule/dist_schedule clauses currently). */
+ if (!OMP_CLAUSE_ORDER_UNCONSTRAINED (t))
+ order_reproducible = true;
default:
break;
}
+ /* For order(reproducible:concurrent) schedule ({dynamic,guided,runtime})
+ we have either the option to expensively remember at runtime how we've
+ distributed work from first loop and reuse that in following loops with
+ the same number of iterations and schedule, or just force static schedule.
+ OpenMP API calls etc. aren't allowed in order(concurrent) bodies so
+ users can't observe it easily anyway. */
+ if (order_reproducible)
+ fd->sched_kind = OMP_CLAUSE_SCHEDULE_STATIC;
if (fd->collapse > 1 || fd->tiling)
fd->loops = loops;
else
diff --git a/gcc/testsuite/gfortran.dg/gomp/order-5.f90 b/gcc/testsuite/gfortran.dg/gomp/order-5.f90
index 4d9e33642af..0dddb968cb4 100644
--- a/gcc/testsuite/gfortran.dg/gomp/order-5.f90
+++ b/gcc/testsuite/gfortran.dg/gomp/order-5.f90
@@ -116,14 +116,14 @@ subroutine f4 (a)
end do
end
-! { dg-final { scan-tree-dump-times "#pragma omp distribute order\\(concurrent\\)" 6 "original"} }
+! { dg-final { scan-tree-dump-times "#pragma omp distribute order\\(reproducible:concurrent\\)" 6 "original"} }
! { dg-final { scan-tree-dump-times "#pragma omp distribute order\\(unconstrained:concurrent\\)" 6 "original"} }
-! { dg-final { scan-tree-dump-times "#pragma omp for nowait order\\(concurrent\\)" 6 "original"} }
+! { dg-final { scan-tree-dump-times "#pragma omp for nowait order\\(reproducible:concurrent\\)" 6 "original"} }
! { dg-final { scan-tree-dump-times "#pragma omp for nowait order\\(unconstrained:concurrent\\)" 6 "original"} }
-! { dg-final { scan-tree-dump-times "#pragma omp for order\\(concurrent\\)" 2 "original"} }
+! { dg-final { scan-tree-dump-times "#pragma omp for order\\(reproducible:concurrent\\)" 2 "original"} }
! { dg-final { scan-tree-dump-times "#pragma omp for order\\(unconstrained:concurrent\\)" 2 "original"} }
! { dg-final { scan-tree-dump-times "#pragma omp parallel" 12 "original"} }
-! { dg-final { scan-tree-dump-times "#pragma omp simd linear\\(i:1\\) order\\(concurrent\\)" 6 "original"} }
+! { dg-final { scan-tree-dump-times "#pragma omp simd linear\\(i:1\\) order\\(reproducible:concurrent\\)" 6 "original"} }
! { dg-final { scan-tree-dump-times "#pragma omp simd linear\\(i:1\\) order\\(unconstrained:concurrent\\)" 6 "original"} }
! { dg-final { scan-tree-dump-times "#pragma omp taskloop" 2 "original"} }
! { dg-final { scan-tree-dump-times "#pragma omp teams" 8 "original"} }
diff --git a/gcc/tree-pretty-print.c b/gcc/tree-pretty-print.c
index 35e567cbfaf..0b5bdd78f06 100644
--- a/gcc/tree-pretty-print.c
+++ b/gcc/tree-pretty-print.c
@@ -1165,6 +1165,8 @@ dump_omp_clause (pretty_printer *pp, tree clause, int spc, dump_flags_t flags)
pp_string (pp, "order(");
if (OMP_CLAUSE_ORDER_UNCONSTRAINED (clause))
pp_string (pp, "unconstrained:");
+ else if (OMP_CLAUSE_ORDER_REPRODUCIBLE (clause))
+ pp_string (pp, "reproducible:");
pp_string (pp, "concurrent)");
break;
diff --git a/gcc/tree.h b/gcc/tree.h
index 7d1257b7a75..7542d97ce12 100644
--- a/gcc/tree.h
+++ b/gcc/tree.h
@@ -1718,6 +1718,9 @@ class auto_suppress_location_wrappers
/* True for unconstrained modifier on order(concurrent) clause. */
#define OMP_CLAUSE_ORDER_UNCONSTRAINED(NODE) \
(OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_ORDER)->base.public_flag)
+/* True for reproducible modifier on order(concurrent) clause. */
+#define OMP_CLAUSE_ORDER_REPRODUCIBLE(NODE) \
+ TREE_PROTECTED (OMP_CLAUSE_SUBCODE_CHECK (NODE, OMP_CLAUSE_ORDER))
#define OMP_CLAUSE_REDUCTION_CODE(NODE) \
(OMP_CLAUSE_RANGE_CHECK (NODE, OMP_CLAUSE_REDUCTION, \
diff --git a/libgomp/testsuite/libgomp.c-c++-common/order-reproducible-1.c b/libgomp/testsuite/libgomp.c-c++-common/order-reproducible-1.c
new file mode 100644
index 00000000000..7e6968fe980
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c-c++-common/order-reproducible-1.c
@@ -0,0 +1,63 @@
+#include <unistd.h>
+#include <stdlib.h>
+
+int
+main ()
+{
+ int a[128];
+ #pragma omp teams num_teams(5)
+ {
+ #pragma omp loop bind(teams)
+ for (int i = 0; i < 128; i++)
+ {
+ a[i] = i;
+ if (i == 0)
+ usleep (20);
+ else if (i == 17)
+ usleep (40);
+ }
+ #pragma omp loop bind(teams)
+ for (int i = 0; i < 128; i++)
+ a[i] += i;
+ }
+ for (int i = 0; i < 128; i++)
+ if (a[i] != 2 * i)
+ abort ();
+ #pragma omp teams num_teams(5)
+ {
+ #pragma omp loop bind(teams) order(concurrent)
+ for (int i = 0; i < 128; i++)
+ {
+ a[i] *= 2;
+ if (i == 1)
+ usleep (20);
+ else if (i == 13)
+ usleep (40);
+ }
+ #pragma omp loop bind(teams) order(concurrent)
+ for (int i = 0; i < 128; i++)
+ a[i] += i;
+ }
+ for (int i = 0; i < 128; i++)
+ if (a[i] != 5 * i)
+ abort ();
+ #pragma omp teams num_teams(5)
+ {
+ #pragma omp loop bind(teams) order(reproducible:concurrent)
+ for (int i = 0; i < 128; i++)
+ {
+ a[i] *= 2;
+ if (i == 2)
+ usleep (20);
+ else if (i == 105)
+ usleep (40);
+ }
+ #pragma omp loop bind(teams) order(reproducible:concurrent)
+ for (int i = 0; i < 128; i++)
+ a[i] += i;
+ }
+ for (int i = 0; i < 128; i++)
+ if (a[i] != 11 * i)
+ abort ();
+ return 0;
+}
diff --git a/libgomp/testsuite/libgomp.c-c++-common/order-reproducible-2.c b/libgomp/testsuite/libgomp.c-c++-common/order-reproducible-2.c
new file mode 100644
index 00000000000..c8ba658cf5e
--- /dev/null
+++ b/libgomp/testsuite/libgomp.c-c++-common/order-reproducible-2.c
@@ -0,0 +1,28 @@
+#include <unistd.h>
+#include <stdlib.h>
+
+int
+main ()
+{
+ int a[128];
+ #pragma omp parallel num_threads(8)
+ {
+ #pragma omp barrier
+ #pragma omp for nowait schedule (dynamic, 2) order(reproducible:concurrent)
+ for (int i = 0; i < 128; i++)
+ {
+ a[i] = i;
+ if (i == 0)
+ usleep (20);
+ else if (i == 17)
+ usleep (40);
+ }
+ #pragma omp for nowait schedule (dynamic, 2) order(reproducible:concurrent)
+ for (int i = 0; i < 128; i++)
+ a[i] += i;
+ }
+ for (int i = 0; i < 128; i++)
+ if (a[i] != 2 * i)
+ abort ();
+ return 0;
+}
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2021-10-01 8:55 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-10-01 8:55 [gcc r12-4035] openmp: Differentiate between order(concurrent) and order(reproducible:concurrent) Jakub Jelinek
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).