| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-O2 -msse2 -mprefer-vector-width=512 -fdump-tree-optimized" } */ |
| 3 |
|
| 4 |
/* { dg-final { scan-tree-dump-times "\.REDUC_PLUS" 3 "optimized" } } */ |
| 5 |
/* { dg-final { scan-tree-dump-times "\.REDUC_MIN" 3 "optimized" } } */ |
| 6 |
/* { dg-final { scan-tree-dump-times "\.REDUC_MAX" 3 "optimized" } } */ |
| 7 |
|
| 8 |
_Float16 |
| 9 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 10 |
reduc_add_128 (_Float16* p) |
| 11 |
{ |
| 12 |
_Float16 sum = 0; |
| 13 |
for (int i = 0; i != 8; i++) |
| 14 |
sum += p[i]; |
| 15 |
return sum; |
| 16 |
} |
| 17 |
|
| 18 |
_Float16 |
| 19 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 20 |
reduc_add_256 (_Float16* p) |
| 21 |
{ |
| 22 |
_Float16 sum = 0; |
| 23 |
for (int i = 0; i != 16; i++) |
| 24 |
sum += p[i]; |
| 25 |
return sum; |
| 26 |
} |
| 27 |
|
| 28 |
_Float16 |
| 29 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 30 |
reduc_add_512 (_Float16* p) |
| 31 |
{ |
| 32 |
_Float16 sum = 0; |
| 33 |
for (int i = 0; i != 32; i++) |
| 34 |
sum += p[i]; |
| 35 |
return sum; |
| 36 |
} |
| 37 |
|
| 38 |
_Float16 |
| 39 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 40 |
reduc_min_128 (_Float16* p) |
| 41 |
{ |
| 42 |
_Float16 sum = p[0]; |
| 43 |
for (int i = 0; i != 8; i++) |
| 44 |
sum = sum > p[i] ? p[i] : sum; |
| 45 |
return sum; |
| 46 |
} |
| 47 |
|
| 48 |
_Float16 |
| 49 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 50 |
reduc_min_256 (_Float16* p) |
| 51 |
{ |
| 52 |
_Float16 sum = p[0]; |
| 53 |
for (int i = 0; i != 16; i++) |
| 54 |
sum = sum > p[i] ? p[i] : sum; |
| 55 |
return sum; |
| 56 |
} |
| 57 |
|
| 58 |
_Float16 |
| 59 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 60 |
reduc_min_512 (_Float16* p) |
| 61 |
{ |
| 62 |
_Float16 sum = p[0]; |
| 63 |
for (int i = 0; i != 32; i++) |
| 64 |
sum = sum > p[i] ? p[i] : sum; |
| 65 |
return sum; |
| 66 |
} |
| 67 |
|
| 68 |
_Float16 |
| 69 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 70 |
reduc_max_128 (_Float16* p) |
| 71 |
{ |
| 72 |
_Float16 sum = p[0]; |
| 73 |
for (int i = 0; i != 8; i++) |
| 74 |
sum = sum < p[i] ? p[i] : sum; |
| 75 |
return sum; |
| 76 |
} |
| 77 |
|
| 78 |
_Float16 |
| 79 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 80 |
reduc_max_256 (_Float16* p) |
| 81 |
{ |
| 82 |
_Float16 sum = p[0]; |
| 83 |
for (int i = 0; i != 16; i++) |
| 84 |
sum = sum < p[i] ? p[i] : sum; |
| 85 |
return sum; |
| 86 |
} |
| 87 |
|
| 88 |
_Float16 |
| 89 |
__attribute__((noipa, target("avx512fp16,avx512vl"), optimize("Ofast"))) |
| 90 |
reduc_max_512 (_Float16* p) |
| 91 |
{ |
| 92 |
_Float16 sum = p[0]; |
| 93 |
for (int i = 0; i != 32; i++) |
| 94 |
sum = sum < p[i] ? p[i] : sum; |
| 95 |
return sum; |
| 96 |
} |