| 1 |
/* { dg-do run { target aarch64_sve256_hw } } */ |
| 2 |
/* { dg-options "-msve-vector-bits=256 -fopenmp -O2" } */ |
| 3 |
|
| 4 |
#pragma GCC target "+sve" |
| 5 |
|
| 6 |
#include <arm_sve.h> |
| 7 |
#include <omp.h> |
| 8 |
|
| 9 |
static svint32_t __attribute__ ((noipa)) |
| 10 |
foo (svint32_t *vb, svint32_t *vc, int tn) |
| 11 |
{ |
| 12 |
svint32_t temp = svindex_s32 (tn, 0); |
| 13 |
temp = svadd_s32_z (svptrue_b32 (), temp, *vb); |
| 14 |
return svadd_s32_z (svptrue_b32 (), temp, *vc); |
| 15 |
} |
| 16 |
|
| 17 |
void __attribute__ ((noipa)) |
| 18 |
lastprivate_sections () |
| 19 |
{ |
| 20 |
int a[8], b[8], c[8]; |
| 21 |
svint32_t va, vb, vc; |
| 22 |
int i; |
| 23 |
|
| 24 |
#pragma omp parallel for |
| 25 |
for (i = 0; i < 8; i++) |
| 26 |
{ |
| 27 |
b[i] = i; |
| 28 |
c[i] = i + 1; |
| 29 |
} |
| 30 |
|
| 31 |
#pragma omp parallel sections lastprivate (vb, vc) num_threads (2) |
| 32 |
{ |
| 33 |
#pragma omp section |
| 34 |
vb = svld1_s32 (svptrue_b32 (), b); |
| 35 |
#pragma omp section |
| 36 |
vb = svld1_s32 (svptrue_b32 (), b); |
| 37 |
vc = svld1_s32 (svptrue_b32 (), c); |
| 38 |
} |
| 39 |
|
| 40 |
va = svadd_s32_z (svptrue_b32 (), vb, vc); |
| 41 |
svst1_s32 (svptrue_b32 (), a, va); |
| 42 |
|
| 43 |
for (i = 0; i < 8; i++) |
| 44 |
if (a[i] != b[i] + c[i]) |
| 45 |
__builtin_abort (); |
| 46 |
} |
| 47 |
|
| 48 |
void __attribute__ ((noipa)) |
| 49 |
lastprivate_for () |
| 50 |
{ |
| 51 |
int a[32], b[32], c[32]; |
| 52 |
int aa[8], bb[8], cc[8]; |
| 53 |
svint32_t va, vb, vc; |
| 54 |
int i, tn; |
| 55 |
|
| 56 |
#pragma omp parallel for |
| 57 |
for (i = 0; i < 32; i++) |
| 58 |
{ |
| 59 |
b[i] = i; |
| 60 |
c[i] = i + 1; |
| 61 |
} |
| 62 |
|
| 63 |
#pragma omp parallel for lastprivate (va, vb, vc, tn) |
| 64 |
for (i = 0; i < 4; i++) |
| 65 |
{ |
| 66 |
vb = svld1_s32 (svptrue_b32 (), b + i * 8); |
| 67 |
vc = svld1_s32 (svptrue_b32 (), c + i * 8); |
| 68 |
tn = i; |
| 69 |
va = foo (&vb, &vc, tn); |
| 70 |
svst1_s32 (svptrue_b32 (), a + i * 8, va); |
| 71 |
} |
| 72 |
|
| 73 |
svst1_s32 (svptrue_b32 (), aa, va); |
| 74 |
svst1_s32 (svptrue_b32 (), bb, vb); |
| 75 |
svst1_s32 (svptrue_b32 (), cc, vc); |
| 76 |
|
| 77 |
for (i = 0; i < 8; i++) |
| 78 |
if (aa[i] != bb[i] + cc[i] + tn) |
| 79 |
__builtin_abort (); |
| 80 |
|
| 81 |
for (i = 0; i < 32; i++) |
| 82 |
if (a[i] != b[i] + c[i] + i / 8) |
| 83 |
__builtin_abort (); |
| 84 |
} |
| 85 |
|
| 86 |
void __attribute__ ((noipa)) |
| 87 |
lastprivate_simd () |
| 88 |
{ |
| 89 |
|
| 90 |
int a[64], b[64], c[64]; |
| 91 |
int aa[8], bb[8], cc[8]; |
| 92 |
svint32_t va, vb, vc; |
| 93 |
int i; |
| 94 |
|
| 95 |
#pragma omp parallel for |
| 96 |
for (i = 0; i < 64; i++) |
| 97 |
{ |
| 98 |
b[i] = i; |
| 99 |
c[i] = i + 1; |
| 100 |
} |
| 101 |
|
| 102 |
#pragma omp simd lastprivate (va, vb, vc) |
| 103 |
for (i = 0; i < 8; i++) |
| 104 |
{ |
| 105 |
vb = svld1_s32 (svptrue_b32 (), b + i * 8); |
| 106 |
vc = svld1_s32 (svptrue_b32 (), c + i * 8); |
| 107 |
va = svadd_s32_z (svptrue_b32 (), vb, vc); |
| 108 |
svst1_s32 (svptrue_b32 (), a + i * 8, va); |
| 109 |
} |
| 110 |
|
| 111 |
svst1_s32 (svptrue_b32 (), aa, va); |
| 112 |
svst1_s32 (svptrue_b32 (), bb, vb); |
| 113 |
svst1_s32 (svptrue_b32 (), cc, vc); |
| 114 |
|
| 115 |
for (i = 0; i < 8; i++) |
| 116 |
if (aa[i] != bb[i] + cc[i]) |
| 117 |
__builtin_abort (); |
| 118 |
|
| 119 |
for (i = 0; i < 64; i++) |
| 120 |
if (a[i] != b[i] + c[i]) |
| 121 |
__builtin_abort (); |
| 122 |
} |
| 123 |
|
| 124 |
void __attribute__ ((noipa)) |
| 125 |
lastprivate_distribute () |
| 126 |
{ |
| 127 |
|
| 128 |
int a[32], b[32], c[32]; |
| 129 |
int aa[8], bb[8], cc[8]; |
| 130 |
svint32_t va, vb, vc; |
| 131 |
int i, tn; |
| 132 |
|
| 133 |
#pragma omp parallel for |
| 134 |
for (i = 0; i < 32; i++) |
| 135 |
{ |
| 136 |
b[i] = i; |
| 137 |
c[i] = i + 1; |
| 138 |
} |
| 139 |
|
| 140 |
#pragma omp teams |
| 141 |
#pragma omp distribute lastprivate (va, vb, vc, tn) |
| 142 |
for (i = 0; i < 4; i++) |
| 143 |
{ |
| 144 |
vb = svld1_s32 (svptrue_b32 (), b + i * 8); |
| 145 |
vc = svld1_s32 (svptrue_b32 (), c + i * 8); |
| 146 |
tn = i; |
| 147 |
va = foo (&vb, &vc, tn); |
| 148 |
svst1_s32 (svptrue_b32 (), a + i * 8, va); |
| 149 |
} |
| 150 |
|
| 151 |
svst1_s32 (svptrue_b32 (), aa, va); |
| 152 |
svst1_s32 (svptrue_b32 (), bb, vb); |
| 153 |
svst1_s32 (svptrue_b32 (), cc, vc); |
| 154 |
|
| 155 |
for (i = 0; i < 8; i++) |
| 156 |
if (aa[i] != bb[i] + cc[i] + tn) |
| 157 |
__builtin_abort (); |
| 158 |
|
| 159 |
for (i = 0; i < 32; i++) |
| 160 |
if (a[i] != b[i] + c[i] + i / 8) |
| 161 |
__builtin_abort (); |
| 162 |
} |
| 163 |
|
| 164 |
int |
| 165 |
main () |
| 166 |
{ |
| 167 |
lastprivate_for (); |
| 168 |
lastprivate_sections (); |
| 169 |
lastprivate_simd (); |
| 170 |
lastprivate_distribute (); |
| 171 |
} |