Hi!
So, OpenMP 4.5 says that:
1) linear can be only specified for the loop iterator on
distribute {, parallel for} simd (because distribute can't
do firstprivate + lastprivate)
2) linear can't be specified at all on distribute parallel for
3) linear can't be specified on doacross loops (ordered(n) clause)
4) ordered can't be specified on distribute parallel for{, simd}
(as no synchronization exists between contention groups)
5) copyin can't be specified on target parallel{, for, for simd}
(as threadprivate is not supported in target regions)
This patch adds diagnostics for those restrictions and tweaks the testsuite.
2015-10-13 Jakub Jelinek <ja...@redhat.com>
gcc/
* gimplify.c (gimplify_scan_omp_clauses): Diagnose linear
clause on combined distribute {, parallel for} simd construct,
unless it is the loop iterator.
gcc/c/
* c-parser.c (c_parser_omp_for_loop): Disallow linear clause
if ordered(n) is present.
(c_parser_omp_for): Disallow ordered clause if combined with
distribute. Disallow linear clause if not combined with
simd and combined with distribute.
(c_parser_omp_parallel): Disallow copyin clause on
target parallel{, for, for simd}.
gcc/cp/
* parser.c (cp_parser_omp_for_loop): Disallow linear clause
if ordered(n) is present.
(cp_parser_omp_for): Disallow ordered clause if combined with
distribute. Disallow linear clause if not combined with
simd and combined with distribute.
(cp_parser_omp_parallel): Disallow copyin clause on
target parallel{, for, for simd}.
gcc/testsuite/
* c-c++-common/gomp/clauses-1.c (bar): Remove linear
and/or ordered clauses where they are no longer allowed.
* c-c++-common/gomp/clauses-4.c: New test.
* c-c++-common/gomp/pr61486-1.c (foo): Remove linear clause
on non-iterator.
* c-c++-common/gomp/pr61486-2.c (test, test2): Remove ordered
clause and ordered construct where no longer allowed.
libgomp/
* testsuite/libgomp.c/pr66199-2.c (f2): Adjust for linear clause
only allowed on the loop iterator.
* testsuite/libgomp.c/pr66199-4.c (f2): Adjust for linear clause
no longer allowed.
* testsuite/libgomp.c/linear-2.c: Remove.
* testsuite/libgomp.c++/linear-2.C: Remove.
--- gcc/gimplify.c.jj 2015-10-12 13:43:30.000000000 +0200
+++ gcc/gimplify.c 2015-10-12 19:19:34.563575391 +0200
@@ -6340,6 +6340,36 @@ gimplify_scan_omp_clauses (tree *list_p,
}
else
{
+ if (code == OMP_SIMD
+ && !OMP_CLAUSE_LINEAR_NO_COPYIN (c))
+ {
+ struct gimplify_omp_ctx *octx = outer_ctx;
+ if (octx
+ && octx->region_type == ORT_WORKSHARE
+ && octx->combined_loop
+ && !octx->distribute)
+ {
+ if (octx->outer_context
+ && (octx->outer_context->region_type
+ == ORT_COMBINED_PARALLEL))
+ octx = octx->outer_context->outer_context;
+ else
+ octx = octx->outer_context;
+ }
+ if (octx
+ && octx->region_type == ORT_WORKSHARE
+ && octx->combined_loop
+ && octx->distribute
+ && !lang_GNU_Fortran ())
+ {
+ error_at (OMP_CLAUSE_LOCATION (c),
+ "%<linear%> clause for variable other than "
+ "loop iterator specified on construct "
+ "combined with %<distribute%>");
+ remove = true;
+ break;
+ }
+ }
/* For combined #pragma omp parallel for simd, need to put
lastprivate and perhaps firstprivate too on the
parallel. Similarly for #pragma omp for simd. */
--- gcc/c/c-parser.c.jj 2015-10-12 13:26:52.000000000 +0200
+++ gcc/c/c-parser.c 2015-10-12 18:27:03.321992027 +0200
@@ -13662,6 +13662,19 @@ c_parser_omp_for_loop (location_t loc, c
= build_int_cst (NULL_TREE, collapse);
ordered = collapse;
}
+ if (ordered)
+ {
+ for (tree *pc = &clauses; *pc; )
+ if (OMP_CLAUSE_CODE (*pc) == OMP_CLAUSE_LINEAR)
+ {
+ error_at (OMP_CLAUSE_LOCATION (*pc),
+ "%<linear%> clause may not be specified together "
+ "with %<ordered%> clause with a parameter");
+ *pc = OMP_CLAUSE_CHAIN (*pc);
+ }
+ else
+ pc = &OMP_CLAUSE_CHAIN (*pc);
+ }
gcc_assert (collapse >= 1 && ordered >= 0);
count = ordered ? ordered : collapse;
@@ -14066,6 +14079,9 @@ c_parser_omp_for (location_t loc, c_pars
mask |= OMP_FOR_CLAUSE_MASK;
if (cclauses)
mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT);
+ /* Composite distribute parallel for{, simd} disallows ordered clause. */
+ if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) != 0)
+ mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ORDERED);
if (c_parser_next_token_is (parser, CPP_NAME))
{
@@ -14100,6 +14116,10 @@ c_parser_omp_for (location_t loc, c_pars
return NULL_TREE;
}
+ /* Composite distribute parallel for disallows linear clause. */
+ if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) != 0)
+ mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_LINEAR);
+
clauses = c_parser_omp_all_clauses (parser, mask, p_name, cclauses == NULL);
if (cclauses)
{
@@ -14340,6 +14360,10 @@ c_parser_omp_parallel (location_t loc, c
strcat (p_name, " parallel");
mask |= OMP_PARALLEL_CLAUSE_MASK;
+ /* #pragma omp target parallel{, for, for simd} disallow copyin clause. */
+ if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_MAP)) != 0
+ && (mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) == 0)
+ mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COPYIN);
if (c_parser_next_token_is_keyword (parser, RID_FOR))
{
--- gcc/cp/parser.c.jj 2015-10-12 13:27:43.000000000 +0200
+++ gcc/cp/parser.c 2015-10-12 18:44:31.150225238 +0200
@@ -32228,6 +32228,19 @@ cp_parser_omp_for_loop (cp_parser *parse
= build_int_cst (NULL_TREE, collapse);
ordered = collapse;
}
+ if (ordered)
+ {
+ for (tree *pc = &clauses; *pc; )
+ if (OMP_CLAUSE_CODE (*pc) == OMP_CLAUSE_LINEAR)
+ {
+ error_at (OMP_CLAUSE_LOCATION (*pc),
+ "%<linear%> clause may not be specified together "
+ "with %<ordered%> clause with a parameter");
+ *pc = OMP_CLAUSE_CHAIN (*pc);
+ }
+ else
+ pc = &OMP_CLAUSE_CHAIN (*pc);
+ }
gcc_assert (collapse >= 1 && ordered >= 0);
count = ordered ? ordered : collapse;
@@ -32587,6 +32600,9 @@ cp_parser_omp_for (cp_parser *parser, cp
mask |= OMP_FOR_CLAUSE_MASK;
if (cclauses)
mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_NOWAIT);
+ /* Composite distribute parallel for{, simd} disallows ordered clause. */
+ if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) != 0)
+ mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_ORDERED);
if (cp_lexer_next_token_is (parser->lexer, CPP_NAME))
{
@@ -32626,6 +32642,10 @@ cp_parser_omp_for (cp_parser *parser, cp
return NULL_TREE;
}
+ /* Composite distribute parallel for disallows linear clause. */
+ if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) != 0)
+ mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_LINEAR);
+
clauses = cp_parser_omp_all_clauses (parser, mask, p_name, pragma_tok,
cclauses == NULL);
if (cclauses)
@@ -32852,6 +32872,10 @@ cp_parser_omp_parallel (cp_parser *parse
strcat (p_name, " parallel");
mask |= OMP_PARALLEL_CLAUSE_MASK;
+ /* #pragma omp target parallel{, for, for simd} disallow copyin clause. */
+ if ((mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_MAP)) != 0
+ && (mask & (OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_DIST_SCHEDULE)) == 0)
+ mask &= ~(OMP_CLAUSE_MASK_1 << PRAGMA_OMP_CLAUSE_COPYIN);
if (cp_lexer_next_token_is_keyword (parser->lexer, RID_FOR))
{
--- gcc/testsuite/c-c++-common/gomp/clauses-1.c.jj 2015-07-16
18:09:25.000000000 +0200
+++ gcc/testsuite/c-c++-common/gomp/clauses-1.c 2015-10-12 17:47:00.355155717
+0200
@@ -15,19 +15,19 @@ foo (int d, int m, int i1, int i2, int p
#pragma omp distribute parallel for \
private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \
if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads
(nth) proc_bind(spread) \
- lastprivate (l) linear (ll:1) ordered schedule(static, 4)
+ lastprivate (l) schedule(static, 4)
for (int i = 0; i < 64; i++)
ll++;
#pragma omp distribute parallel for simd \
private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \
if (parallel: i2) default(shared) shared(s) reduction(+:r) num_threads
(nth) proc_bind(spread) \
- lastprivate (l) linear (ll:1) schedule(static, 4) \
+ lastprivate (l) schedule(static, 4) \
safelen(8) simdlen(4) aligned(q: 32)
for (int i = 0; i < 64; i++)
ll++;
#pragma omp distribute simd \
private (p) firstprivate (f) collapse(1) dist_schedule(static, 16) \
- safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32) reduction(+:r)
+ safelen(8) simdlen(4) aligned(q: 32) reduction(+:r)
for (int i = 0; i < 64; i++)
ll++;
}
@@ -94,7 +94,7 @@ bar (int d, int m, int i1, int i2, int p
shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \
collapse(1) dist_schedule(static, 16) \
if (parallel: i2) num_threads (nth) proc_bind(spread) \
- lastprivate (l) linear (ll:1) ordered schedule(static, 4)
+ lastprivate (l) schedule(static, 4)
for (int i = 0; i < 64; i++)
ll++;
#pragma omp target teams distribute parallel for simd \
@@ -102,7 +102,7 @@ bar (int d, int m, int i1, int i2, int p
shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \
collapse(1) dist_schedule(static, 16) \
if (parallel: i2) num_threads (nth) proc_bind(spread) \
- lastprivate (l) linear (ll:1) schedule(static, 4) \
+ lastprivate (l) schedule(static, 4) \
safelen(8) simdlen(4) aligned(q: 32)
for (int i = 0; i < 64; i++)
ll++;
@@ -110,7 +110,7 @@ bar (int d, int m, int i1, int i2, int p
device(d) map (tofrom: m) if (target: i1) private (p) firstprivate (f)
defaultmap(tofrom: scalar) is_device_ptr (idp) \
shared(s) default(shared) reduction(+:r) num_teams(nte) thread_limit(tl) \
collapse(1) dist_schedule(static, 16) \
- safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32)
+ safelen(8) simdlen(4) aligned(q: 32)
for (int i = 0; i < 64; i++)
ll++;
#pragma omp target simd \
@@ -140,7 +140,7 @@ bar (int d, int m, int i1, int i2, int p
private(p) firstprivate (f) shared(s) default(shared) reduction(+:r)
num_teams(nte) thread_limit(tl) \
collapse(1) dist_schedule(static, 16) \
if (parallel: i2) num_threads (nth) proc_bind(spread) \
- lastprivate (l) linear (ll:1) ordered schedule(static, 4)
+ lastprivate (l) schedule(static, 4)
for (int i = 0; i < 64; i++)
ll++;
#pragma omp target
@@ -148,7 +148,7 @@ bar (int d, int m, int i1, int i2, int p
private(p) firstprivate (f) shared(s) default(shared) reduction(+:r)
num_teams(nte) thread_limit(tl) \
collapse(1) dist_schedule(static, 16) \
if (parallel: i2) num_threads (nth) proc_bind(spread) \
- lastprivate (l) linear (ll:1) schedule(static, 4) \
+ lastprivate (l) schedule(static, 4) \
safelen(8) simdlen(4) aligned(q: 32)
for (int i = 0; i < 64; i++)
ll++;
@@ -156,7 +156,7 @@ bar (int d, int m, int i1, int i2, int p
#pragma omp teams distribute simd \
private(p) firstprivate (f) shared(s) default(shared) reduction(+:r)
num_teams(nte) thread_limit(tl) \
collapse(1) dist_schedule(static, 16) \
- safelen(8) simdlen(4) linear(ll: 1) aligned(q: 32)
+ safelen(8) simdlen(4) aligned(q: 32)
for (int i = 0; i < 64; i++)
ll++;
}
--- gcc/testsuite/c-c++-common/gomp/clauses-4.c.jj 2015-10-12
17:26:00.475096064 +0200
+++ gcc/testsuite/c-c++-common/gomp/clauses-4.c 2015-10-12 18:40:14.763083015
+0200
@@ -0,0 +1,96 @@
+int t;
+#pragma omp threadprivate (t)
+
+void
+foo (int y, short z)
+{
+ int x;
+ #pragma omp target teams map (from: x)
+ #pragma omp distribute simd linear (x : 2)
+ for (x = 0; x < 64; x += 2)
+ ;
+ #pragma omp target teams map (from: x)
+ #pragma omp distribute parallel for simd linear (x)
+ for (x = 0; x < 64; x++)
+ ;
+ #pragma omp target teams map (tofrom: y)
+ #pragma omp distribute simd linear (y : 2) /* { dg-error ".linear. clause
for variable other than loop iterator specified on construct combined with
.distribute." } */
+ for (x = 0; x < 64; x += 2)
+ y += 2;
+ #pragma omp target teams map (tofrom: z)
+ #pragma omp distribute parallel for simd linear (z) /* { dg-error ".linear.
clause for variable other than loop iterator specified on construct combined
with .distribute." } */
+ for (x = 0; x < 64; x++)
+ z++;
+ #pragma omp target teams map (tofrom: z)
+ #pragma omp distribute parallel for linear (z: 4) /* { dg-error ".linear.
is not valid for .#pragma omp distribute parallel for." } */
+ for (x = 0; x < 64; x++)
+ z += 4;
+ #pragma omp target map (from: x)
+ #pragma omp teams distribute simd linear (x : 2)
+ for (x = 0; x < 64; x += 2)
+ ;
+ #pragma omp target map (from: x)
+ #pragma omp teams distribute parallel for simd linear (x)
+ for (x = 0; x < 64; x++)
+ ;
+ #pragma omp target map (tofrom: y)
+ #pragma omp teams distribute simd linear (y : 2) /* { dg-error ".linear.
clause for variable other than loop iterator specified on construct combined
with .distribute." } */
+ for (x = 0; x < 64; x += 2)
+ y += 2;
+ #pragma omp target map (tofrom: z)
+ #pragma omp teams distribute parallel for simd linear (z) /* { dg-error
".linear. clause for variable other than loop iterator specified on construct
combined with .distribute." } */
+ for (x = 0; x < 64; x++)
+ z++;
+ #pragma omp target map (tofrom: z)
+ #pragma omp teams distribute parallel for linear (z: 4) /* { dg-error
".linear. is not valid for .#pragma omp teams distribute parallel for." } */
+ for (x = 0; x < 64; x++)
+ z += 4;
+ #pragma omp target parallel copyin (t) /* { dg-error ".copyin. is not
valid for .#pragma omp target parallel." } */
+ ;
+ #pragma omp target parallel for copyin (t) /* { dg-error ".copyin. is not
valid for .#pragma omp target parallel for." } */
+ for (x = 0; x < 64; x++)
+ ;
+ #pragma omp target parallel for simd copyin (t) /* { dg-error ".copyin.
is not valid for .#pragma omp target parallel for simd." } */
+ for (x = 0; x < 64; x++)
+ ;
+ #pragma omp target teams
+ #pragma omp distribute parallel for ordered /* { dg-error
".ordered. is not valid for .#pragma omp distribute parallel for." } */
+ for (x = 0; x < 64; x++)
+ {
+ #pragma omp ordered /* { dg-error "ordered region must be closely
nested inside a loop region with an ordered clause" } */
+ ;
+ }
+ #pragma omp target teams
+ #pragma omp distribute parallel for simd ordered /* { dg-error
".ordered. is not valid for .#pragma omp distribute parallel for simd." } */
+ for (x = 0; x < 64; x++)
+ {
+ #pragma omp ordered simd, threads /* { dg-error "OpenMP
constructs other than .#pragma omp ordered simd. may not be nested inside simd
region" } */
+ ;
+ }
+ #pragma omp target
+ #pragma omp teams distribute parallel for ordered /* { dg-error
".ordered. is not valid for .#pragma omp teams distribute parallel for." } */
+ for (x = 0; x < 64; x++)
+ {
+ #pragma omp ordered /* { dg-error "ordered region must be closely
nested inside a loop region with an ordered clause" } */
+ ;
+ }
+ #pragma omp target
+ #pragma omp teams distribute parallel for simd ordered /* { dg-error
".ordered. is not valid for .#pragma omp teams distribute parallel for simd." }
*/
+ for (x = 0; x < 64; x++)
+ {
+ #pragma omp ordered simd, threads /* { dg-error "OpenMP
constructs other than .#pragma omp ordered simd. may not be nested inside simd
region" } */
+ ;
+ }
+ #pragma omp target teams distribute parallel for ordered /* {
dg-error ".ordered. is not valid for .#pragma omp target teams distribute
parallel for." } */
+ for (x = 0; x < 64; x++)
+ {
+ #pragma omp ordered /* { dg-error "ordered region must be closely
nested inside a loop region with an ordered clause" } */
+ ;
+ }
+ #pragma omp target teams distribute parallel for simd ordered /* {
dg-error ".ordered. is not valid for .#pragma omp target teams distribute
parallel for simd." } */
+ for (x = 0; x < 64; x++)
+ {
+ #pragma omp ordered simd, threads /* { dg-error "OpenMP
constructs other than .#pragma omp ordered simd. may not be nested inside simd
region" } */
+ ;
+ }
+}
--- gcc/testsuite/c-c++-common/gomp/pr61486-1.c.jj 2015-04-24
12:32:01.000000000 +0200
+++ gcc/testsuite/c-c++-common/gomp/pr61486-1.c 2015-10-12 17:49:42.041721972
+0200
@@ -6,8 +6,8 @@ int
foo (int *a)
{
int i, j = 0;
- #pragma omp target teams distribute simd linear(i, j) map(a[:10])
+ #pragma omp target teams distribute simd linear(i) map(a[:10])
for (i = 0; i < 10; i++)
- a[i] = j++;
- return i + j;
+ a[i] = j;
+ return i;
}
--- gcc/testsuite/c-c++-common/gomp/pr61486-2.c.jj 2015-09-03
16:37:51.000000000 +0200
+++ gcc/testsuite/c-c++-common/gomp/pr61486-2.c 2015-10-12 17:55:47.205225446
+0200
@@ -50,22 +50,21 @@ test (int n, int o, int p, int q, int r,
private (p) firstprivate (q) shared (n) reduction (+: r) \
thread_limit (n * 2) dist_schedule (static, 4) collapse (2) \
num_threads (n + 4) proc_bind (spread) lastprivate (s) \
- ordered schedule (static, 8)
+ schedule (static, 8)
for (i = 0; i < 10; i++)
for (j = 0; j < 10; j++)
{
r = r + 1;
p = q;
dosomething (a, n, p + q);
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10 + j;
}
#pragma omp target teams distribute parallel for device (n + 1) num_teams
(n + 4) \
if (n != 6)map (from: n) map (alloc: a[2:o-2]) default(shared) \
private (p) firstprivate (q) shared (n) reduction (+: r) \
thread_limit (n * 2) dist_schedule (static, 4) num_threads (n + 4) \
- proc_bind (master) lastprivate (s) ordered schedule (static, 8)
+ proc_bind (master) lastprivate (s) schedule (static, 8)
for (i = 0; i < 10; i++)
{
for (j = 0; j < 10; j++)
@@ -74,8 +73,7 @@ test (int n, int o, int p, int q, int r,
p = q;
dosomething (a, n, p + q);
}
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10;
}
#pragma omp target teams distribute parallel for simd device (n + 1) \
@@ -165,22 +163,21 @@ test (int n, int o, int p, int q, int r,
default(shared) private (p) firstprivate (q) shared (n) reduction (+:
r) \
thread_limit (n * 2) dist_schedule (static, 4) collapse (2) \
num_threads (n + 4) proc_bind (spread) lastprivate (s) \
- ordered schedule (static, 8)
+ schedule (static, 8)
for (i = 0; i < 10; i++)
for (j = 0; j < 10; j++)
{
r = r + 1;
p = q;
dosomething (a, n, p + q);
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10 + j;
}
#pragma omp target device (n + 1) if (n != 6)map(from:n)
map(alloc:a[2:o-2])
#pragma omp teams distribute parallel for num_teams (n + 4) if (n != 6) \
default(shared) private (p) firstprivate (q) shared (n) reduction (+:
r) \
thread_limit (n * 2) dist_schedule (static, 4) num_threads (n + 4) \
- proc_bind (master) lastprivate (s) ordered schedule (static, 8)
+ proc_bind (master) lastprivate (s) schedule (static, 8)
for (i = 0; i < 10; i++)
{
for (j = 0; j < 10; j++)
@@ -189,8 +186,7 @@ test (int n, int o, int p, int q, int r,
p = q;
dosomething (a, n, p + q);
}
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10;
}
#pragma omp target device (n + 1) if (n != 6)map(from:n)
map(alloc:a[2:o-2])
@@ -273,15 +269,14 @@ test (int n, int o, int p, int q, int r,
default(shared) private (p) firstprivate (q) shared (n) reduction (+:
r) \
collapse (2) dist_schedule (static, 4) \
num_threads (n + 4) proc_bind (spread) lastprivate (s) \
- ordered schedule (static, 8)
+ schedule (static, 8)
for (i = 0; i < 10; i++)
for (j = 0; j < 10; j++)
{
r = r + 1;
p = q;
dosomething (a, n, p + q);
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10 + j;
}
#pragma omp target teams device (n + 1) if (n != 6)map(from:n)
map(alloc:a[2:o-2]) \
@@ -289,7 +284,7 @@ test (int n, int o, int p, int q, int r,
#pragma omp distribute parallel for if (n != 6) \
default(shared) private (p) firstprivate (q) shared (n) reduction (+:
r) \
num_threads (n + 4) dist_schedule (static, 4) \
- proc_bind (master) lastprivate (s) ordered schedule (static, 8)
+ proc_bind (master) lastprivate (s) schedule (static, 8)
for (i = 0; i < 10; i++)
{
for (j = 0; j < 10; j++)
@@ -298,8 +293,7 @@ test (int n, int o, int p, int q, int r,
p = q;
dosomething (a, n, p + q);
}
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10;
}
#pragma omp target teams device (n + 1) if (n != 6)map(from:n)
map(alloc:a[2:o-2]) \
@@ -385,21 +379,20 @@ test2 (int n, int o, int p, int r, int s
default(shared) private (p) firstprivate (q) shared (n) reduction (+:
r) \
collapse (2) dist_schedule (static, 4) \
num_threads (n + 4) proc_bind (spread) lastprivate (s) \
- ordered schedule (static, 8)
+ schedule (static, 8)
for (i = 0; i < 10; i++)
for (j = 0; j < 10; j++)
{
r = r + 1;
p = q;
dosomething (a, n, p + q);
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10 + j;
}
#pragma omp distribute parallel for if (n != 6) \
default(shared) private (p) firstprivate (q) shared (n) reduction (+:
r) \
num_threads (n + 4) dist_schedule (static, 4) \
- proc_bind (master) lastprivate (s) ordered schedule (static, 8)
+ proc_bind (master) lastprivate (s) schedule (static, 8)
for (i = 0; i < 10; i++)
{
for (j = 0; j < 10; j++)
@@ -408,8 +401,7 @@ test2 (int n, int o, int p, int r, int s
p = q;
dosomething (a, n, p + q);
}
- #pragma omp ordered
- p = q;
+ p = q;
s = i * 10;
}
#pragma omp distribute parallel for simd if (n != 6)default(shared) \
--- libgomp/testsuite/libgomp.c/pr66199-2.c.jj 2015-06-17 21:00:37.000000000
+0200
+++ libgomp/testsuite/libgomp.c/pr66199-2.c 2015-10-12 19:16:16.276558954
+0200
@@ -18,12 +18,11 @@ __attribute__((noinline, noclone)) void
f2 (long a, long b, long c)
{
long d, e;
- #pragma omp target teams distribute parallel for simd default(none)
firstprivate (a, b) shared(u, v, w) linear(d) linear(c:5) lastprivate(e)
+ #pragma omp target teams distribute parallel for simd default(none)
firstprivate (a, b, c) shared(u, v, w) linear(d) lastprivate(e)
for (d = a; d < b; d++)
{
u[d] = v[d] + w[d];
- c += 5;
- e = c;
+ e = c + d * 5;
}
}
--- libgomp/testsuite/libgomp.c/pr66199-4.c.jj 2015-06-18 15:16:18.000000000
+0200
+++ libgomp/testsuite/libgomp.c/pr66199-4.c 2015-10-12 19:16:40.353196680
+0200
@@ -19,12 +19,11 @@ __attribute__((noinline, noclone)) void
f2 (long a, long b, long c)
{
long d, e;
- #pragma omp target teams distribute parallel for default(none) firstprivate
(a, b) shared(u, v, w) linear(d) linear(c:5) lastprivate(e)
+ #pragma omp target teams distribute parallel for default(none) firstprivate
(a, b, c) shared(u, v, w) lastprivate(d, e)
for (d = a; d < b; d++)
{
u[d] = v[d] + w[d];
- c += 5;
- e = c;
+ e = c + d * 5;
}
}
--- libgomp/testsuite/libgomp.c/linear-2.c.jj 2015-07-17 15:23:45.000000000
+0200
+++ libgomp/testsuite/libgomp.c/linear-2.c 2015-10-12 19:10:36.912665256
+0200
@@ -1,244 +0,0 @@
-#pragma omp declare target
-int a[256];
-#pragma omp end declare target
-
-__attribute__((noinline, noclone)) void
-f1 (int i)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
- for (int j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f2 (short int i, char k)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
- for (long j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f3 (long long int i, long long int k)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
- for (short j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f4 (int i)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(static, 3)
- for (int j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f5 (short int i, char k)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(static, 5)
- for (long j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f6 (long long int i, long long int k)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(static, 7)
- for (short j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f7 (int i)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(dynamic, 3)
- for (int j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f8 (short int i, char k)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(dynamic, 5)
- for (long j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f9 (long long int i, long long int k)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(dynamic, 7)
- for (short j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f10 (int i, char start, long step)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
- for (int j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f11 (short int i, char k, long start, char step)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
- for (long j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f12 (long long int i, long long int k, long long int start, int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
- for (short j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f13 (int i, int start, long long int step)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(static, 3)
- for (int j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f14 (short int i, char k, unsigned long long int start, int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(static, 5)
- for (long j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f15 (long long int i, long long int k, char start, long int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(static, 7)
- for (short j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f16 (int i, int start, long long int step)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(dynamic, 3)
- for (int j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f17 (short int i, char k, long start, int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(dynamic, 5)
- for (long j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f18 (long long int i, long long int k, short start, long int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(dynamic, 7)
- for (short j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-void
-verify (void)
-{
- int err;
- #pragma omp target map(from:err)
- {
- err = 0;
- for (int i = 0; i < 256; i++)
- if (a[i] != (((i & 3) == 0 && i >= 8
- && i < 8 + 48 * 4)
- ? ((i - 8) / 4) + 16 : 0))
- err = 1;
- __builtin_memset (a, 0, sizeof (a));
- }
- if (err)
- __builtin_abort ();
-}
-
-int
-main ()
-{
-#define TEST(x) x; verify ()
- TEST (f1 (8));
- TEST (f2 (8, 3));
- TEST (f3 (8LL, 4LL));
- TEST (f4 (8));
- TEST (f5 (8, 3));
- TEST (f6 (8LL, 4LL));
- TEST (f7 (8));
- TEST (f8 (8, 3));
- TEST (f9 (8LL, 4LL));
- TEST (f10 (8, 16, 2));
- TEST (f11 (8, 3, 16, 2));
- TEST (f12 (8LL, 4LL, 16, 2));
- TEST (f13 (8, 16, 2));
- TEST (f14 (8, 3, 16, 2));
- TEST (f15 (8LL, 4LL, 16, 2));
- TEST (f16 (8, 16, 2));
- TEST (f17 (8, 3, 16, 2));
- TEST (f18 (8LL, 4LL, 16, 2));
- return 0;
-}
--- libgomp/testsuite/libgomp.c++/linear-2.C.jj 2015-07-17 15:26:20.000000000
+0200
+++ libgomp/testsuite/libgomp.c++/linear-2.C 2015-10-12 19:16:53.194003469
+0200
@@ -1,261 +0,0 @@
-#pragma omp declare target
-int a[256];
-#pragma omp end declare target
-
-template <typename T>
-__attribute__((noinline, noclone)) void
-f1 (T &i)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
- for (int j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f2 (short int i, char k)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
- for (long j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f3 (long long int &i, long long int k)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
- for (short j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f4 (int i)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(static, 3)
- for (int j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f5 (short int i, char k)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(static, 5)
- for (long j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-template <int N>
-__attribute__((noinline, noclone)) void
-f6 (long long int &i, long long int k)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(static, 7)
- for (short j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f7 (int i)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(dynamic, 3)
- for (int j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f8 (short int &i, char &k)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(dynamic, 5)
- for (long j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f9 (long long int i, long long int k)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(dynamic, 7)
- for (short j = 16; j < 64; j++)
- {
- a[i] = j;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f10 (int i, char start, long step)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
- for (int j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f11 (short int &i, char &k, long &start, char step)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
- for (long j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f12 (long long int i, long long int k, long long int start, int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
- for (short j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f13 (int i, int start, long long int step)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(static, 3)
- for (int j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f14 (short int i, char k, unsigned long long int start, int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(static, 5)
- for (long j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-template <typename T>
-__attribute__((noinline, noclone)) void
-f15 (T &i, T k, char &start, long int &step)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(static, 7)
- for (short j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-template <typename T>
-__attribute__((noinline, noclone)) void
-f16 (T i, T start, long long int step)
-{
- #pragma omp target teams distribute parallel for linear (i: 4)
schedule(dynamic, 3)
- for (int j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f17 (short int i, char k, long start, int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k + 1)
schedule(dynamic, 5)
- for (long j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-__attribute__((noinline, noclone)) void
-f18 (long long int i, long long int k, short start, long int step)
-{
- #pragma omp target teams distribute parallel for linear (i: k)
schedule(dynamic, 7)
- for (short j = start; j < 112; j += step)
- {
- a[i] = j / 2 + 8;
- i += 4;
- }
-}
-
-void
-verify (void)
-{
- int err;
- #pragma omp target map(from:err)
- {
- err = 0;
- for (int i = 0; i < 256; i++)
- if (a[i] != (((i & 3) == 0 && i >= 8
- && i < 8 + 48 * 4)
- ? ((i - 8) / 4) + 16 : 0))
- err = 1;
- __builtin_memset (a, 0, sizeof (a));
- }
- if (err)
- __builtin_abort ();
-}
-
-int
-main ()
-{
-#define TEST(x) x; verify ()
- int vi = 8;
- TEST (f1 (vi));
- TEST (f2 (8, 3));
- long long int vll = 8LL;
- TEST (f3 (vll, 4LL));
- TEST (f4 (8));
- TEST (f5 (8, 3));
- vll = 8LL;
- TEST (f6<9> (vll, 4LL));
- TEST (f7 (8));
- short int vs = 8;
- char vk = 3;
- TEST (f8 (vs, vk));
- TEST (f9 (8LL, 4LL));
- TEST (f10 (8, 16, 2));
- vs = 8;
- vk = 3;
- long int vl = 16;
- TEST (f11 (vs, vk, vl, 2));
- TEST (f12 (8LL, 4LL, 16, 2));
- TEST (f13 (8, 16, 2));
- TEST (f14 (8, 3, 16, 2));
- vll = 8LL;
- vk = 16;
- vl = 2;
- TEST (f15 (vll, 4LL, vk, vl));
- vi = 8;
- int vi2 = 16;
- TEST (f16<int &> (vi, vi2, 2));
- TEST (f17 (8, 3, 16, 2));
- TEST (f18 (8LL, 4LL, 16, 2));
- return 0;
-}
Jakub
