Discussion:
[PATCH, OpenACC] Properly handle wait clause with no arguments
Thomas Schwinge
2018-11-07 19:13:29 UTC
Permalink
Hi Chung-Lin!
Hi, this patch properly handles OpenACC 'wait' clauses without arguments, making it an equivalent of "wait all".
Thanks!
(current trunk basically discards and ignores such argument-less wait
clauses)
Bugs should be filed, for later reference. Now done:
<https://gcc.gnu.org/PR87924> "OpenACC wait clauses without
async-arguments". (I couldn't put you in CC because "***@gcc.gnu.org
did not match anything"?)
This adds additional handling in
the pack/unpack of the wait argument across the compiler/libgomp interface, but is done in a matter that
doesn't affect binary compatibility.
Hmm. See below. (Jakub, could you please review the last paragraph of
this email?)
This patch was part of the OpenACC async re-work that was done on the gomp4 branch (later merged to OG7/OG8), see [1].
I'm separating this part out and submitting it first because it's logically independent.
[1] https://gcc.gnu.org/ml/gcc-patches/2017-06/msg01842.html
Thanks for splitting it out!
Re-tested with offloading to ensure no regressions, is this okay for trunk?
A few comments.

No test cases included. I'm working on a few, will post/commit later.
gcc/c/
* c-parser.c (c_parser_oacc_clause_wait): Add representation of wait
clause without argument as 'wait (GOMP_ASYNC_NOVAL)', adjust comments.
gcc/cp/
* parser.c (cp_parser_oacc_clause_wait): Add representation of wait
clause without argument as 'wait (GOMP_ASYNC_NOVAL)', adjust comments.
gcc/fortran/
* trans-openmp.c (gfc_trans_omp_clauses_1): Add representation of wait
clause without argument as 'wait (GOMP_ASYNC_NOVAL)'.
gcc/
* omp-low.c (expand_omp_target): Add middle-end support for handling
OMP_CLAUSE_WAIT clause with a GOMP_ASYNC_NOVAL(-1) as the argument.
include/
* gomp-constants.h (GOMP_LAUNCH_OP_MASK): Define.
(GOMP_LAUNCH_PACK): Add bitwise-and of GOMP_LAUNCH_OP_MASK.
(GOMP_LAUNCH_OP): Likewise.
libgomp/
* oacc-parallel.c (GOACC_parallel_keyed): Interpret launch op as
signed 16-bit field, adjust num_waits handling.
(GOACC_enter_exit_data): Adjust num_waits handling.
(GOACC_update): Adjust num_waits handling.
--- gcc/c/c-parser.c (revision 263981)
+++ gcc/c/c-parser.c (working copy)
@@ -12719,7 +12719,7 @@ c_parser_oacc_clause_tile (c_parser *parser, tree
}
- wait ( int-expr-list ) */
+ wait [( int-expr-list )] */
static tree
c_parser_oacc_clause_wait (c_parser *parser, tree list)
@@ -12728,7 +12728,15 @@ c_parser_oacc_clause_wait (c_parser *parser, tree
if (c_parser_peek_token (parser)->type == CPP_OPEN_PAREN)
list = c_parser_oacc_wait_list (parser, clause_loc, list);
+ else
+ {
+ tree c = build_omp_clause (clause_loc, OMP_CLAUSE_WAIT);
+ OMP_CLAUSE_DECL (c) = build_int_cst (integer_type_node, GOMP_ASYNC_NOVAL);
+ OMP_CLAUSE_CHAIN (c) = list;
+ list = c;
+ }
+
return list;
}
ACK.
--- gcc/cp/parser.c (revision 263981)
+++ gcc/cp/parser.c (working copy)
@@ -32137,7 +32137,7 @@ cp_parser_oacc_wait_list (cp_parser *parser, locat
}
- wait ( int-expr-list ) */
+ wait [( int-expr-list )] */
static tree
cp_parser_oacc_clause_wait (cp_parser *parser, tree list)
@@ -32144,10 +32144,16 @@ cp_parser_oacc_clause_wait (cp_parser *parser, tre
{
location_t location = cp_lexer_peek_token (parser->lexer)->location;
- if (cp_lexer_peek_token (parser->lexer)->type != CPP_OPEN_PAREN)
- return list;
+ if (cp_lexer_peek_token (parser->lexer)->type == CPP_OPEN_PAREN)
+ list = cp_parser_oacc_wait_list (parser, location, list);
+ else
+ {
+ tree c = build_omp_clause (location, OMP_CLAUSE_WAIT);
- list = cp_parser_oacc_wait_list (parser, location, list);
+ OMP_CLAUSE_DECL (c) = build_int_cst (integer_type_node, GOMP_ASYNC_NOVAL);
+ OMP_CLAUSE_CHAIN (c) = list;
+ list = c;
+ }
return list;
}
ACK.
--- gcc/fortran/trans-openmp.c (revision 263981)
+++ gcc/fortran/trans-openmp.c (working copy)
@@ -2922,6 +2922,13 @@ gfc_trans_omp_clauses (stmtblock_t *block, gfc_omp
omp_clauses = c;
}
}
+ else if (clauses->wait)
+ {
+ c = build_omp_clause (where.lb->location, OMP_CLAUSE_WAIT);
+ OMP_CLAUSE_DECL (c) = build_int_cst (integer_type_node, GOMP_ASYNC_NOVAL);
+ OMP_CLAUSE_CHAIN (c) = omp_clauses;
+ omp_clauses = c;
+ }
if (clauses->num_gangs_expr)
{
tree num_gangs_var
NACK. Instead let's do the following, similar to C, C++, and also
similar to Fortran's OpenACC async clause handling without explicit
async-argument:

--- gcc/fortran/openmp.c
+++ gcc/fortran/openmp.c
@@ -1885,7 +1885,19 @@ gfc_match_omp_clauses (gfc_omp_clauses **cp, const omp_mask mask,
break;
}
else if (m == MATCH_NO)
- needs_space = true;
+ {
+ gfc_expr *expr
+ = gfc_get_constant_expr (BT_INTEGER,
+ gfc_default_integer_kind,
+ &gfc_current_locus);
+ mpz_set_si (expr->value.integer, GOMP_ASYNC_NOVAL);
+ gfc_expr_list **expr_list = &c->wait_list;
+ while (*expr_list)
+ expr_list = &(*expr_list)->next;
+ *expr_list = gfc_get_expr_list ();
+ (*expr_list)->expr = expr;
+ needs_space = true;
+ }
continue;
}
if ((mask & OMP_CLAUSE_WORKER)

Now, why do we need the following changes, in this rather "convoluted"
--- gcc/omp-expand.c (revision 263981)
+++ gcc/omp-expand.c (working copy)
@@ -7381,16 +7381,32 @@ expand_omp_target (struct omp_region *region)
/* ... push a placeholder. */
args.safe_push (integer_zero_node);
+ bool noval_seen = false;
+ tree noval = build_int_cst (integer_type_node, GOMP_ASYNC_NOVAL);
+
for (; c; c = OMP_CLAUSE_CHAIN (c))
if (OMP_CLAUSE_CODE (c) == OMP_CLAUSE_WAIT)
{
+ tree wait_expr = OMP_CLAUSE_WAIT_EXPR (c);
+
+ if (TREE_CODE (wait_expr) == INTEGER_CST
+ && tree_int_cst_compare (wait_expr, noval) == 0)
+ {
+ noval_seen = true;
+ continue;
+ }
+
args.safe_push (fold_convert_loc (OMP_CLAUSE_LOCATION (c),
- integer_type_node,
- OMP_CLAUSE_WAIT_EXPR (c)));
+ integer_type_node, wait_expr));
num_waits++;
}
- if (!tagging || num_waits)
+ if (noval_seen && num_waits == 0)
+ args[t_wait_idx] =
+ (tagging
+ ? oacc_launch_pack (GOMP_LAUNCH_WAIT, NULL_TREE, GOMP_ASYNC_NOVAL)
+ : noval);
+ else if (!tagging || num_waits)
{
tree len;
--- include/gomp-constants.h (revision 263981)
+++ include/gomp-constants.h (working copy)
@@ -221,13 +221,14 @@ enum gomp_map_kind
#define GOMP_LAUNCH_CODE_SHIFT 28
#define GOMP_LAUNCH_DEVICE_SHIFT 16
#define GOMP_LAUNCH_OP_SHIFT 0
+#define GOMP_LAUNCH_OP_MASK 0xffff
#define GOMP_LAUNCH_PACK(CODE,DEVICE,OP) \
(((CODE) << GOMP_LAUNCH_CODE_SHIFT) \
| ((DEVICE) << GOMP_LAUNCH_DEVICE_SHIFT) \
- | ((OP) << GOMP_LAUNCH_OP_SHIFT))
+ | (((OP) & GOMP_LAUNCH_OP_MASK) << GOMP_LAUNCH_OP_SHIFT))
#define GOMP_LAUNCH_CODE(X) (((X) >> GOMP_LAUNCH_CODE_SHIFT) & 0xf)
#define GOMP_LAUNCH_DEVICE(X) (((X) >> GOMP_LAUNCH_DEVICE_SHIFT) & 0xfff)
-#define GOMP_LAUNCH_OP(X) (((X) >> GOMP_LAUNCH_OP_SHIFT) & 0xffff)
+#define GOMP_LAUNCH_OP(X) (((X) >> GOMP_LAUNCH_OP_SHIFT) & GOMP_LAUNCH_OP_MASK)
#define GOMP_LAUNCH_OP_MAX 0xffff
/* Bitmask to apply in order to find out the intended device of a target
--- libgomp/oacc-parallel.c (revision 263981)
+++ libgomp/oacc-parallel.c (working copy)
@@ -194,10 +194,14 @@ GOACC_parallel_keyed (int device, void (*fn) (void
{
- unsigned num_waits = GOMP_LAUNCH_OP (tag);
+ /* Be careful to cast the op field as a signed 16-bit, and
+ sign-extend to full integer. */
+ int num_waits = ((signed short) GOMP_LAUNCH_OP (tag));
- if (num_waits)
+ if (num_waits > 0)
goacc_wait (async, num_waits, &ap);
+ else if (num_waits == acc_async_noval)
+ acc_wait_all_async (async);
break;
}
@@ -351,7 +355,7 @@ GOACC_enter_exit_data (int device, size_t mapnum,
|| host_fallback)
return;
- if (num_waits)
+ if (num_waits > 0)
{
va_list ap;
@@ -359,6 +363,8 @@ GOACC_enter_exit_data (int device, size_t mapnum,
goacc_wait (async, num_waits, &ap);
va_end (ap);
}
+ else if (num_waits == acc_async_noval)
+ acc_wait_all_async (async);
/* Determine whether "finalize" semantics apply to all mappings of this
OpenACC directive. */
@@ -542,7 +548,7 @@ GOACC_update (int device, size_t mapnum,
|| host_fallback)
return;
- if (num_waits)
+ if (num_waits > 0)
{
va_list ap;
@@ -550,6 +556,8 @@ GOACC_update (int device, size_t mapnum,
goacc_wait (async, num_waits, &ap);
va_end (ap);
}
+ else if (num_waits == acc_async_noval)
+ acc_wait_all_async (async);
acc_dev->openacc.async_set_async_func (async);
Why can't we just pass "GOMP_ASYNC_NOVAL" through like any other
async-argument (that is, map a single "wait" clause to "num_waits == 1,
*ap == GOMP_ASYNC_NOVAL"), and then handle that case in "goacc_wait",
avoiding all these interface changes and special casing in different
functions?

Or am I not understanding correctly what the purpose of this is?


My understanding is that before, GCC never generates "negative
async-arguments" (now used for "GOMP_ASYNC_NOVAL"), but only non-negative
ones (real "async-arguments"), which we continue to handle, as before.

Isn't that sufficient for the ABI compatibility that we promise, which is
(unless I'm confused now?) that old (existing) executables continue to
run correctly when dynamically linking against a new libgomp. Or do we
also have to care about the case that an executable built with a new
version of GCC has to work when dynamically linked against an old
libgomp?


Grüße
Thomas
Jakub Jelinek
2018-11-07 19:19:35 UTC
Permalink
Post by Thomas Schwinge
Isn't that sufficient for the ABI compatibility that we promise, which is
(unless I'm confused now?) that old (existing) executables continue to
run correctly when dynamically linking against a new libgomp. Or do we
also have to care about the case that an executable built with a new
version of GCC has to work when dynamically linked against an old
libgomp?
Only old executables/libraries need to continue running correctly when
linking against new libgomp. New programs against old libgomp might work,
or might not.

Jakub
Chung-Lin Tang
2018-11-27 14:41:54 UTC
Permalink
Post by Thomas Schwinge
NACK. Instead let's do the following, similar to C, C++, and also
similar to Fortran's OpenACC async clause handling without explicit
--- gcc/fortran/openmp.c
+++ gcc/fortran/openmp.c
@@ -1885,7 +1885,19 @@ gfc_match_omp_clauses (gfc_omp_clauses **cp, const omp_mask mask,
break;
}
else if (m == MATCH_NO)
- needs_space = true;
+ {
+ gfc_expr *expr
+ = gfc_get_constant_expr (BT_INTEGER,
+ gfc_default_integer_kind,
+ &gfc_current_locus);
+ mpz_set_si (expr->value.integer, GOMP_ASYNC_NOVAL);
+ gfc_expr_list **expr_list = &c->wait_list;
+ while (*expr_list)
+ expr_list = &(*expr_list)->next;
+ *expr_list = gfc_get_expr_list ();
+ (*expr_list)->expr = expr;
+ needs_space = true;
+ }
continue;
}
if ((mask & OMP_CLAUSE_WORKER)
Okay, I see what you mean.
Post by Thomas Schwinge
Now, why do we need the following changes, in this rather "convoluted"
+ tree wait_expr = OMP_CLAUSE_WAIT_EXPR (c);
+
+ if (TREE_CODE (wait_expr) == INTEGER_CST
+ && tree_int_cst_compare (wait_expr, noval) == 0)
+ {
+ noval_seen = true;
+ continue;
+ }
+
args.safe_push (fold_convert_loc (OMP_CLAUSE_LOCATION (c),
- integer_type_node,
- OMP_CLAUSE_WAIT_EXPR (c)));
+ integer_type_node, wait_expr));
num_waits++;
}
- if (!tagging || num_waits)
+ if (noval_seen && num_waits == 0)
+ args[t_wait_idx] =
+ (tagging
+ ? oacc_launch_pack (GOMP_LAUNCH_WAIT, NULL_TREE, GOMP_ASYNC_NOVAL)
+ : noval);
+ else if (!tagging || num_waits)
{
tree len;
{
- unsigned num_waits = GOMP_LAUNCH_OP (tag);
+ /* Be careful to cast the op field as a signed 16-bit, and
+ sign-extend to full integer. */
+ int num_waits = ((signed short) GOMP_LAUNCH_OP (tag));
- if (num_waits)
+ if (num_waits > 0)
goacc_wait (async, num_waits, &ap);
+ else if (num_waits == acc_async_noval)
+ acc_wait_all_async (async);
Why can't we just pass "GOMP_ASYNC_NOVAL" through like any other
async-argument (that is, map a single "wait" clause to "num_waits == 1,
*ap == GOMP_ASYNC_NOVAL"), and then handle that case in "goacc_wait",
avoiding all these interface changes and special casing in different
functions?
Or am I not understanding correctly what the purpose of this is?
I think the original intention was that wait(acc_async_noval) should
correspond to "wait all" semantics, hence we should be able to ignore
and discard other wait(<arg>) clauses if they exist.

Having that said, I think there is some incorrect code in my patch wrt
this intended behavior, which I'll revise.

(The assumption of an argument-less wait clause to mean "wait all" is
derived from the closely documented OpenACC wait *directive* specification.
Frankly speaking, the prior section on the wait *clause* is not explicitly
clear on this, though 'wait all' is a reasonable assumption. It would still
be helpful if we asked the OpenACC SC to clarify)

As for the idea on stuffing more code into goacc_wait(), I think that's
a pretty good suggestion, since all uses of it in oacc-parallel.c are
actually quite similar; re-factoring this part should make things more elegant.
Post by Thomas Schwinge
My understanding is that before, GCC never generates "negative
async-arguments" (now used for "GOMP_ASYNC_NOVAL"), but only non-negative
ones (real "async-arguments"), which we continue to handle, as before.
Isn't that sufficient for the ABI compatibility that we promise, which is
(unless I'm confused now?) that old (existing) executables continue to
run correctly when dynamically linking against a new libgomp. Or do we
also have to care about the case that an executable built with a new
version of GCC has to work when dynamically linked against an old
libgomp?
I think either way, encoding GOMP_ASYNC_NOVAL in num_waits or as an argument
should be okay for backward compatibility, i.e. old binaries should still
work with new libgomp with this modification.

As for new binaries vs old libgomp, I believe with the original libgomp
oacc-parallel.c code, it's not quite possible to achieve the intended wait all
behavior by playing with num_waits or arguments.

I'll revise the patch and re-submit later.

Thanks,
Chung-Lin
Thomas Schwinge
2018-11-30 21:01:05 UTC
Permalink
Hi Chung-Lin!
Post by Thomas Schwinge
Hi, this patch properly handles OpenACC 'wait' clauses without arguments, making it an equivalent of "wait all".
(current trunk basically discards and ignores such argument-less wait
clauses)
<https://gcc.gnu.org/PR87924> "OpenACC wait clauses without
did not match anything"?)
This will, by the way, need to be fixed on all active release branches.
Post by Thomas Schwinge
No test cases included. I'm working on a few, will post/commit later.
I thought I had also written a libgomp execution test case, during
travel/attending the SuperComputing 2018 conference, but I can't find it
right now... ;-|

Anyway, with XFAILs (which you then please remove as part of your patch),
at least the following compile-time test cases now committed to trunk in
r266686:

commit 1d89613e77d7db420b13ce3ad8b98f07aaf474e8
Author: tschwinge <***@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Fri Nov 30 20:39:30 2018 +0000

[PR87924] Add (XFAILed) test cases for OpenACC wait clauses without async-arguments

gcc/testsuite/
PR c/87924
* c-c++-common/goacc/asyncwait-5.c: Update.
* gfortran.dg/goacc/asyncwait-5.f: Likewise.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/***@266686 138bc75d-0d04-0410-961f-82ee72b054a4
---
gcc/testsuite/ChangeLog | 4 ++++
gcc/testsuite/c-c++-common/goacc/asyncwait-5.c | 21 +++++++++++++++++++++
gcc/testsuite/gfortran.dg/goacc/asyncwait-5.f | 20 ++++++++++++++++++++
3 files changed, 45 insertions(+)

diff --git gcc/testsuite/ChangeLog gcc/testsuite/ChangeLog
index 75ca70b4af28..68186d8ab837 100644
--- gcc/testsuite/ChangeLog
+++ gcc/testsuite/ChangeLog
@@ -1,5 +1,9 @@
2018-11-30 Thomas Schwinge <***@codesourcery.com>

+ PR c/87924
+ * c-c++-common/goacc/asyncwait-5.c: Update.
+ * gfortran.dg/goacc/asyncwait-5.f: Likewise.
+
* c-c++-common/goacc/asyncwait-5.c: New file.
* gfortran.dg/goacc/asyncwait-5.f: Likewise.

diff --git gcc/testsuite/c-c++-common/goacc/asyncwait-5.c gcc/testsuite/c-c++-common/goacc/asyncwait-5.c
index fe6f8a0cf2da..80d4a8477b93 100644
--- gcc/testsuite/c-c++-common/goacc/asyncwait-5.c
+++ gcc/testsuite/c-c++-common/goacc/asyncwait-5.c
@@ -11,4 +11,25 @@ void f()
#pragma acc parallel async (2) wait (11, 12) wait (13)
;
/* { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel wait\\(13\\) wait\\(12\\) wait\\(11\\) async\\(2\\)\$" 1 "original" } } */
+
+
+#pragma acc parallel async (3) wait
+ ;
+ /* { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel wait\\(-1\\) async\\(3\\)$" 1 "original" { xfail *-*-* } } } */
+
+#pragma acc parallel async (4) wait (100) wait
+ ;
+ /* { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel wait\\(-1\\) wait\\(100\\) async\\(4\\)$" 1 "original" { xfail *-*-* } } } */
+
+#pragma acc parallel async (5) wait wait (101)
+ ;
+ /* { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel wait\\(101\\) wait\\(-1\\) async\\(5\\)$" 1 "original" { xfail *-*-* } } } */
+
+#pragma acc parallel async (6) wait wait (102, 103) wait wait
+ ;
+ /* { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel wait\\(-1\\) wait\\(-1\\) wait\\(103\\) wait\\(102\\) wait\\(-1\\) async\\(6\\)$" 1 "original" { xfail *-*-* } } } */
+
+#pragma acc parallel async (7) wait (104) wait wait (105, 106)
+ ;
+ /* { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel wait\\(106\\) wait\\(105\\) wait\\(-1\\) wait\\(104\\) async\\(7\\)$" 1 "original" { xfail *-*-* } } } */
}
diff --git gcc/testsuite/gfortran.dg/goacc/asyncwait-5.f gcc/testsuite/gfortran.dg/goacc/asyncwait-5.f
index 59b886343af6..7ad5813b8a03 100644
--- gcc/testsuite/gfortran.dg/goacc/asyncwait-5.f
+++ gcc/testsuite/gfortran.dg/goacc/asyncwait-5.f
@@ -10,4 +10,24 @@
!$ACC END PARALLEL
! { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel async\\(2\\) wait\\(11\\) wait\\(12\\) wait\\(13\\)$" 1 "original" } }

+!$ACC PARALLEL ASYNC (3) WAIT
+!$ACC END PARALLEL
+! { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel async\\(3\\) wait\\(-1\\)$" 1 "original" { xfail *-*-* } } }
+
+!$ACC PARALLEL ASYNC (4) WAIT (100) WAIT
+!$ACC END PARALLEL
+! { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel async\\(4\\) wait\\(100\\) wait\\(-1\\)$" 1 "original" { xfail *-*-* } } }
+
+!$ACC PARALLEL ASYNC (5) WAIT WAIT (101)
+!$ACC END PARALLEL
+! { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel async\\(5\\) wait\\(-1\\) wait\\(101\\)$" 1 "original" { xfail *-*-* } } }
+
+!$ACC PARALLEL ASYNC (6) WAIT WAIT (102, 103) WAIT WAIT
+!$ACC END PARALLEL
+! { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel async\\(6\\) wait\\(-1\\) wait\\(102\\) wait\\(103\\) wait\\(-1\\) wait\\(-1\\)$" 1 "original" { xfail *-*-* } } }
+
+!$ACC PARALLEL ASYNC (7) WAIT (104) WAIT WAIT (105, 106)
+!$ACC END PARALLEL
+! { dg-final { scan-tree-dump-times "(?n)#pragma acc parallel async\\(7\\) wait\\(104\\) wait\\(-1\\) wait\\(105\\) wait\\(106\\)$" 1 "original" { xfail *-*-* } } }
+
END


Grüße
Thomas

Loading...