f38b20d68f
OpenMP 5.2 adds support for firstprivate and allocate clauses on the scope construct and this patch adds that support to GCC. 5.2 unfortunately (IMNSHO mistakenly) marked scope construct as worksharing, which implies that it isn't possible to nest inside of it other scope, worksharing loop, sections, explicit barriers, single etc. which would make scope far less useful. I'm not implementing that part, keeping the 5.1 behavior here, and will file an issue to revert that for OpenMP 6.0. But, for firstprivate it keeps the restriction that is now implied from worksharing construct that listed var can't be private in outer context, where for reduction 5.1 had similar restriction explicit even for scope and 5.2 has it implicitly through worksharing construct. 2022-05-31 Jakub Jelinek <jakub@redhat.com> gcc/ * omp-low.cc (build_outer_var_ref): For code == OMP_CLAUSE_ALLOCATE allow var to be private in the outer context. (lower_private_allocate): Pass OMP_CLAUSE_ALLOCATE as last argument to build_outer_var_ref. gcc/c/ * c-parser.cc (OMP_SCOPE_CLAUSE_MASK): Add firstprivate and allocate clauses. gcc/cp/ * parser.cc (OMP_SCOPE_CLAUSE_MASK): Add firstprivate and allocate clauses. gcc/testsuite/ * c-c++-common/gomp/scope-5.c: New test. * c-c++-common/gomp/scope-6.c: New test. * g++.dg/gomp/attrs-1.C (bar): Add firstprivate and allocate clauses to scope construct. * g++.dg/gomp/attrs-2.C (bar): Likewise. libgomp/ * testsuite/libgomp.c-c++-common/allocate-1.c (foo): Add testcase for scope construct with allocate clause. * testsuite/libgomp.c-c++-common/allocate-3.c (foo): Likewise. * testsuite/libgomp.c-c++-common/scope-2.c: New test.
438 lines
12 KiB
C
438 lines
12 KiB
C
#include <omp.h>
|
|
#include <stdlib.h>
|
|
#include <stdint.h>
|
|
|
|
struct S { int a, b; };
|
|
|
|
void
|
|
foo (int x, int *p, int *q, int px, omp_allocator_handle_t h, int fl)
|
|
{
|
|
int y = 0, r = 0, i, i1, l, l2[4], l3, n = 8;
|
|
int i2, j2, n2 = 9, l4;
|
|
int i3, j3, n3 = 10, l5;
|
|
int i4, j4, n4 = 11, l6;
|
|
int i5, n5;
|
|
int v[x], w[x];
|
|
int r2[4] = { 0, 0, 0, 0 };
|
|
int xo = x;
|
|
struct S s = { 27, 29 };
|
|
for (i = 0; i < 4; i++)
|
|
p[i] = 0;
|
|
for (i = 0; i < 3; i++)
|
|
q[i] = 0;
|
|
for (i = 0; i < x; i++)
|
|
w[i] = i;
|
|
#pragma omp parallel private (y, v) firstprivate (x) allocate (align (32) : x) allocate (align (128) : y) allocate (v)
|
|
{
|
|
int *volatile p1 = &x;
|
|
int *volatile p2 = &y;
|
|
if (x != 42)
|
|
abort ();
|
|
#pragma omp barrier
|
|
*p2 = 1;
|
|
p1[0]++;
|
|
v[0] = 7;
|
|
v[41] = 8;
|
|
#pragma omp barrier
|
|
if (x != 43 || y != 1)
|
|
abort ();
|
|
if (v[0] != 7 || v[41] != 8)
|
|
abort ();
|
|
if ((fl & 2) && (((uintptr_t) p1 | (uintptr_t) p2
|
|
| (uintptr_t) &v[0]) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p1) & 31) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p2) & 127) != 0)
|
|
abort ();
|
|
}
|
|
x = xo;
|
|
#pragma omp teams
|
|
#pragma omp parallel private (y) firstprivate (x, w) allocate (allocator (h), align (32) : x) allocate (align (128), allocator (h):y) allocate(allocator(h):w)
|
|
{
|
|
int *volatile p1 = &x;
|
|
int *volatile p2 = &y;
|
|
if (x != 42 || w[17] != 17 || w[41] != 41)
|
|
abort ();
|
|
#pragma omp barrier
|
|
*p2 = 1;
|
|
p1[0]++;
|
|
w[19]++;
|
|
#pragma omp barrier
|
|
if (x != 43 || y != 1 || w[19] != 20)
|
|
abort ();
|
|
if ((fl & 1) && (((uintptr_t) p1 | (uintptr_t) p2
|
|
| (uintptr_t) &w[0]) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p1) & 31) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p2) & 127) != 0)
|
|
abort ();
|
|
}
|
|
x = xo;
|
|
#pragma omp parallel for private (y) firstprivate (x) allocate (allocator (h), align (32) : x, y, r) allocate (align (128), allocator (h) : l, n) reduction(+: r) lastprivate (l) linear (n: 16)
|
|
for (i = 0; i < 64; i++)
|
|
{
|
|
if (x != 42)
|
|
abort ();
|
|
y = 1;
|
|
l = i;
|
|
n += y + 15;
|
|
r += i;
|
|
if ((fl & 1) && (((uintptr_t) &x | (uintptr_t) &y | (uintptr_t) &r
|
|
| (uintptr_t) &l | (uintptr_t) &n) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &x | (uintptr_t) &y | (uintptr_t) &r) & 31) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &l | (uintptr_t) &n) & 127) != 0)
|
|
abort ();
|
|
}
|
|
x = xo;
|
|
#pragma omp parallel
|
|
{
|
|
#pragma omp for lastprivate (l2) private (i1) allocate (allocator (h): l2, l3) allocate (allocator (h),align(16): i1) lastprivate (conditional: l3)
|
|
for (i1 = 0; i1 < 64; i1++)
|
|
{
|
|
l2[0] = i1;
|
|
l2[1] = i1 + 1;
|
|
l2[2] = i1 + 2;
|
|
l2[3] = i1 + 3;
|
|
if (i1 < 37)
|
|
l3 = i1;
|
|
if ((fl & 1) && (((uintptr_t) &l2[0] | (uintptr_t) &l3 | (uintptr_t) &i1) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &i1) & 15) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp for collapse(2) lastprivate(l4, i2, j2) linear (n2:17) allocate (allocator (h), align (8): n2, l4, i2, j2)
|
|
for (i2 = 3; i2 < 5; i2++)
|
|
for (j2 = 17; j2 < 22; j2 += 2)
|
|
{
|
|
n2 += 17;
|
|
l4 = i2 * 31 + j2;
|
|
if ((fl & 1) && (((uintptr_t) &l4 | (uintptr_t) &n2
|
|
| (uintptr_t) &i2 | (uintptr_t) &j2) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &l4 | (uintptr_t) &n2 | (uintptr_t) &i2 | (uintptr_t) &j2) & 7) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp for collapse(2) lastprivate(l5, i3, j3) linear (n3:17) schedule (static, 3) allocate (align (16): n3, l5, i3, j3)
|
|
for (i3 = 3; i3 < 5; i3++)
|
|
for (j3 = 17; j3 < 23; j3 += 2)
|
|
{
|
|
n3 += 17;
|
|
l5 = i3 * 31 + j3;
|
|
if ((fl & 2) && (((uintptr_t) &l5 | (uintptr_t) &n3
|
|
| (uintptr_t) &i3 | (uintptr_t) &j3) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &l5 | (uintptr_t) &n3 | (uintptr_t) &i3 | (uintptr_t) &j3) & 15) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp for collapse(2) lastprivate(l6, i4, j4) linear (n4:17) schedule (dynamic) allocate (align (16), allocator (h): n4, l6, i4, j4)
|
|
for (i4 = 3; i4 < 5; i4++)
|
|
for (j4 = 17; j4 < 22; j4 += 2)
|
|
{
|
|
n4 += 17;
|
|
l6 = i4 * 31 + j4;
|
|
if ((fl & 1) && (((uintptr_t) &l6 | (uintptr_t) &n4
|
|
| (uintptr_t) &i4 | (uintptr_t) &j4) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &l6 | (uintptr_t) &n4 | (uintptr_t) &i4 | (uintptr_t) &j4) & 15) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp for lastprivate (i5) allocate (align (32): i5)
|
|
for (i5 = 1; i5 < 17; i5 += 3)
|
|
{
|
|
if ((fl & 2) && (((uintptr_t) &i5) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &i5) & 31) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp for reduction(+:p[2:px], q[:3], r2) allocate(align (16), allocator (h): p, q, r2)
|
|
for (i = 0; i < 32; i++)
|
|
{
|
|
p[2] += i;
|
|
p[3] += 2 * i;
|
|
q[0] += 3 * i;
|
|
q[2] += 4 * i;
|
|
r2[0] += 5 * i;
|
|
r2[3] += 6 * i;
|
|
/* Can't really rely on alignment of &p[0], the implementation could
|
|
allocate the whole array or do what GCC does and allocate only part
|
|
of it. */
|
|
if ((fl & 1) && (((uintptr_t) &q[0] | (uintptr_t) &r2[0]) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) &q[0] | (uintptr_t) &r2[0]) & 15) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp task private(y) firstprivate(x) allocate(align (8) : x, y)
|
|
{
|
|
int *volatile p1 = &x;
|
|
int *volatile p2 = &y;
|
|
if (x != 42)
|
|
abort ();
|
|
p1[0]++;
|
|
p2[0] = 21;
|
|
if (x != 43 || y != 21)
|
|
abort ();
|
|
if ((fl & 2) && (((uintptr_t) p1 | (uintptr_t) p2) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p1 | (uintptr_t) p2) & 7) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp task private(y) firstprivate(x) allocate(allocator (h),align(32): x, y)
|
|
{
|
|
int *volatile p1 = &x;
|
|
int *volatile p2 = &y;
|
|
if (x != 42)
|
|
abort ();
|
|
p1[0]++;
|
|
p2[0] = 21;
|
|
if (x != 43 || y != 21)
|
|
abort ();
|
|
if ((fl & 1) && (((uintptr_t) p1 | (uintptr_t) p2) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p1 | (uintptr_t) p2) & 31) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp task private(y) firstprivate(s) allocate(align(16): s, y)
|
|
{
|
|
int *volatile p1 = &s.a;
|
|
int *volatile p2 = &s.b;
|
|
int *volatile p3 = &y;
|
|
if (s.a != 27 || s.b != 29)
|
|
abort ();
|
|
p1[0]++;
|
|
p2[0]++;
|
|
p3[0] = 21;
|
|
if (s.a != 28 || s.b != 30 || y != 21)
|
|
abort ();
|
|
if ((fl & 2) && (((uintptr_t) p1 | (uintptr_t) p3) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p1 | (uintptr_t) p3) & 15) != 0)
|
|
abort ();
|
|
}
|
|
#pragma omp task private(y) firstprivate(s) allocate(allocator (h): s, y)
|
|
{
|
|
int *volatile p1 = &s.a;
|
|
int *volatile p2 = &s.b;
|
|
int *volatile p3 = &y;
|
|
if (s.a != 27 || s.b != 29)
|
|
abort ();
|
|
p1[0]++;
|
|
p2[0]++;
|
|
p3[0] = 21;
|
|
if (s.a != 28 || s.b != 30 || y != 21)
|
|
abort ();
|
|
if ((fl & 1) && (((uintptr_t) p1 | (uintptr_t) p3) & 63) != 0)
|
|
abort ();
|
|
}
|
|
}
|
|
if (r != 64 * 63 / 2 || l != 63 || n != 8 + 16 * 64)
|
|
abort ();
|
|
if (l2[0] != 63 || l2[1] != 63 + 1 || l2[2] != 63 + 2 || l2[3] != 63 + 3 || l3 != 36)
|
|
abort ();
|
|
if (i2 != 5 || j2 != 23 || n2 != 9 + 6 * 17 || l4 != 4 * 31 + 21)
|
|
abort ();
|
|
if (i3 != 5 || j3 != 23 || n3 != 10 + 6 * 17 || l5 != 4 * 31 + 21)
|
|
abort ();
|
|
if (i4 != 5 || j4 != 23 || n4 != 11 + 6 * 17 || l6 != 4 * 31 + 21)
|
|
abort ();
|
|
if (i5 != 19)
|
|
abort ();
|
|
if (p[2] != (32 * 31) / 2 || p[3] != 2 * (32 * 31) / 2
|
|
|| q[0] != 3 * (32 * 31) / 2 || q[2] != 4 * (32 * 31) / 2
|
|
|| r2[0] != 5 * (32 * 31) / 2 || r2[3] != 6 * (32 * 31) / 2)
|
|
abort ();
|
|
r = 0;
|
|
x = xo;
|
|
#pragma omp parallel shared (x, y, r, n5) firstprivate (h)
|
|
{
|
|
#pragma omp masked
|
|
n5 = omp_get_num_threads ();
|
|
#pragma omp scope private (y) firstprivate (x) reduction(+:r) \
|
|
allocate (allocator (h), align (32): x) \
|
|
allocate (align (128), allocator (h): y) \
|
|
allocate (align (32), allocator (h): r)
|
|
{
|
|
int *volatile p1 = &x;
|
|
int *volatile p2 = &y;
|
|
if (x != 42)
|
|
abort ();
|
|
#pragma omp barrier
|
|
*p2 = 1;
|
|
p1[0]++;
|
|
r++;
|
|
#pragma omp barrier
|
|
if (x != 43 || y != 1 || r != 1)
|
|
abort ();
|
|
if ((fl & 1) && (((uintptr_t) p1 | (uintptr_t) p2
|
|
| (uintptr_t) &r) & 63) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p1 | (uintptr_t) &r) & 31) != 0)
|
|
abort ();
|
|
if ((((uintptr_t) p2) & 127) != 0)
|
|
abort ();
|
|
}
|
|
}
|
|
if (x != 42 || r != n5)
|
|
abort ();
|
|
}
|
|
|
|
void
|
|
bar (int x, omp_allocator_handle_t h)
|
|
{
|
|
int y = 0, r = 0, i, i1, l, l2[4], l3, n = 8;
|
|
int i2, j2, n2 = 9, l4;
|
|
int i3, j3, n3 = 10, l5;
|
|
int i4, j4, n4 = 11, l6;
|
|
int i5;
|
|
struct S s = { 27, 29 };
|
|
int xo = x;
|
|
#pragma omp parallel private (y) firstprivate (x) allocate (x, y)
|
|
{
|
|
if (x != 42)
|
|
abort ();
|
|
#pragma omp barrier
|
|
y = 1;
|
|
x++;
|
|
#pragma omp barrier
|
|
if (x != 43 || y != 1)
|
|
abort ();
|
|
}
|
|
x = xo;
|
|
#pragma omp teams
|
|
#pragma omp parallel private (y) firstprivate (x) allocate (allocator (h): x, y)
|
|
{
|
|
if (x != 42)
|
|
abort ();
|
|
#pragma omp barrier
|
|
y = 1;
|
|
x++;
|
|
#pragma omp barrier
|
|
if (x != 43 || y != 1)
|
|
abort ();
|
|
}
|
|
x = xo;
|
|
#pragma omp parallel for private (y) firstprivate (x) allocate (allocator (h): x, y, r, l, n) reduction(+: r) lastprivate (l) linear (n: 16)
|
|
for (i = 0; i < 64; i++)
|
|
{
|
|
if (x != 42)
|
|
abort ();
|
|
y = 1;
|
|
l = i;
|
|
n += y + 15;
|
|
r += i;
|
|
}
|
|
x = xo;
|
|
#pragma omp parallel
|
|
{
|
|
#pragma omp for lastprivate (l2) private (i1) allocate (allocator (h): l2, l3, i1) lastprivate (conditional: l3)
|
|
for (i1 = 0; i1 < 64; i1++)
|
|
{
|
|
l2[0] = i1;
|
|
l2[1] = i1 + 1;
|
|
l2[2] = i1 + 2;
|
|
l2[3] = i1 + 3;
|
|
if (i1 < 37)
|
|
l3 = i1;
|
|
}
|
|
#pragma omp for collapse(2) lastprivate(l4, i2, j2) linear (n2:17) allocate (allocator (h): n2, l4, i2, j2)
|
|
for (i2 = 3; i2 < 5; i2++)
|
|
for (j2 = 17; j2 < 22; j2 += 2)
|
|
{
|
|
n2 += 17;
|
|
l4 = i2 * 31 + j2;
|
|
}
|
|
#pragma omp for collapse(2) lastprivate(l5, i3, j3) linear (n3:17) schedule (static, 3) allocate (n3, l5, i3, j3)
|
|
for (i3 = 3; i3 < 5; i3++)
|
|
for (j3 = 17; j3 < 23; j3 += 2)
|
|
{
|
|
n3 += 17;
|
|
l5 = i3 * 31 + j3;
|
|
}
|
|
#pragma omp for collapse(2) lastprivate(l6, i4, j4) linear (n4:17) schedule (dynamic) allocate (allocator (h): n4, l6, i4, j4)
|
|
for (i4 = 3; i4 < 5; i4++)
|
|
for (j4 = 17; j4 < 22; j4 += 2)
|
|
{
|
|
n4 += 17;
|
|
l6 = i4 * 31 + j4;
|
|
}
|
|
#pragma omp for lastprivate (i5) allocate (i5)
|
|
for (i5 = 1; i5 < 17; i5 += 3)
|
|
;
|
|
#pragma omp task private(y) firstprivate(x) allocate(x, y)
|
|
{
|
|
if (x != 42)
|
|
abort ();
|
|
x++;
|
|
y = 21;
|
|
if (x != 43 || y != 21)
|
|
abort ();
|
|
}
|
|
#pragma omp task private(y) firstprivate(x) allocate(allocator (h): x, y)
|
|
{
|
|
if (x != 42)
|
|
abort ();
|
|
x++;
|
|
y = 21;
|
|
if (x != 43 || y != 21)
|
|
abort ();
|
|
}
|
|
#pragma omp task private(y) firstprivate(s) allocate(s, y)
|
|
{
|
|
if (s.a != 27 || s.b != 29)
|
|
abort ();
|
|
s.a++;
|
|
s.b++;
|
|
y = 21;
|
|
if (s.a != 28 || s.b != 30 || y != 21)
|
|
abort ();
|
|
}
|
|
#pragma omp task private(y) firstprivate(s) allocate(allocator (h), align (16): s, y)
|
|
{
|
|
if (s.a != 27 || s.b != 29)
|
|
abort ();
|
|
s.a++;
|
|
s.b++;
|
|
y = 21;
|
|
if (s.a != 28 || s.b != 30 || y != 21)
|
|
abort ();
|
|
}
|
|
}
|
|
if (r != 64 * 63 / 2 || l != 63 || n != 8 + 16 * 64)
|
|
abort ();
|
|
if (l2[0] != 63 || l2[1] != 63 + 1 || l2[2] != 63 + 2 || l2[3] != 63 + 3 || l3 != 36)
|
|
abort ();
|
|
if (i2 != 5 || j2 != 23 || n2 != 9 + 6 * 17 || l4 != 4 * 31 + 21)
|
|
abort ();
|
|
if (i3 != 5 || j3 != 23 || n3 != 10 + 6 * 17 || l5 != 4 * 31 + 21)
|
|
abort ();
|
|
if (i4 != 5 || j4 != 23 || n4 != 11 + 6 * 17 || l6 != 4 * 31 + 21)
|
|
abort ();
|
|
if (i5 != 19)
|
|
abort ();
|
|
}
|
|
|
|
int
|
|
main ()
|
|
{
|
|
omp_alloctrait_t traits[3]
|
|
= { { omp_atk_alignment, 64 },
|
|
{ omp_atk_fallback, omp_atv_null_fb } };
|
|
omp_allocator_handle_t a
|
|
= omp_init_allocator (omp_default_mem_space, 2, traits);
|
|
int p[4], q[3];
|
|
if (a == omp_null_allocator)
|
|
abort ();
|
|
omp_set_default_allocator (omp_default_mem_alloc);
|
|
foo (42, p, q, 2, omp_null_allocator, 0);
|
|
foo (42, p, q, 2, omp_default_mem_alloc, 0);
|
|
foo (42, p, q, 2, a, 1);
|
|
omp_set_default_allocator (a);
|
|
foo (42, p, q, 2, omp_null_allocator, 3);
|
|
foo (42, p, q, 2, omp_default_mem_alloc, 2);
|
|
bar (42, a);
|
|
omp_destroy_allocator (a);
|
|
return 0;
|
|
}
|