| 1 |
/* { dg-do run { target avx512fp16 } } */ |
| 2 |
/* { dg-options "-O2 -mavx512fp16 -mavx512vl -mprefer-vector-width=512" } */ |
| 3 |
|
| 4 |
static void condmov_test (void); |
| 5 |
#define DO_TEST condmov_test |
| 6 |
#define AVX512FP16 |
| 7 |
#define AVX512VL |
| 8 |
#include "avx512f-check.h" |
| 9 |
#include "avx512fp16-vcondmn-loop-1.c" |
| 10 |
|
| 11 |
_Float16 a[32], b[32], c[32], fexp[32], fref[32]; |
| 12 |
s16 sa[32], sb[32], sc[32], sexp[32], sref[32]; |
| 13 |
u16 ua[32], ub[32], uc[32], uexp[32], uref[32]; |
| 14 |
|
| 15 |
#define EMULATE_CONDMOV_LOOP(size, type, ptype, op, name) \ |
| 16 |
void \ |
| 17 |
__attribute__ ((noinline, noclone)) \ |
| 18 |
scalar_cond_##size##ptype##type##name ( \ |
| 19 |
ptype * restrict a, ptype * restrict b, \ |
| 20 |
type * restrict c, type * restrict d) \ |
| 21 |
{ \ |
| 22 |
int i; \ |
| 23 |
for (i = 0; i < size; i++) \ |
| 24 |
{ \ |
| 25 |
if (a[i] op b[i]) \ |
| 26 |
d[i] = c[i]; \ |
| 27 |
} \ |
| 28 |
} |
| 29 |
|
| 30 |
EMULATE_CONDMOV_LOOP (32, _Float16, _Float16, <, lt) |
| 31 |
EMULATE_CONDMOV_LOOP (32, _Float16, _Float16, >, gt) |
| 32 |
EMULATE_CONDMOV_LOOP (32, _Float16, _Float16, ==, eq) |
| 33 |
EMULATE_CONDMOV_LOOP (16, _Float16, _Float16, <, lt) |
| 34 |
EMULATE_CONDMOV_LOOP (16, _Float16, _Float16, >, gt) |
| 35 |
EMULATE_CONDMOV_LOOP (16, _Float16, _Float16, ==, eq) |
| 36 |
EMULATE_CONDMOV_LOOP (8, _Float16, _Float16, <, lt) |
| 37 |
EMULATE_CONDMOV_LOOP (8, _Float16, _Float16, >, gt) |
| 38 |
EMULATE_CONDMOV_LOOP (8, _Float16, _Float16, ==, eq) |
| 39 |
EMULATE_CONDMOV_LOOP (32, _Float16, s16, <, lt) |
| 40 |
EMULATE_CONDMOV_LOOP (32, _Float16, s16, >, gt) |
| 41 |
EMULATE_CONDMOV_LOOP (32, _Float16, s16, ==, eq) |
| 42 |
EMULATE_CONDMOV_LOOP (16, _Float16, s16, <, lt) |
| 43 |
EMULATE_CONDMOV_LOOP (16, _Float16, s16, >, gt) |
| 44 |
EMULATE_CONDMOV_LOOP (16, _Float16, s16, ==, eq) |
| 45 |
EMULATE_CONDMOV_LOOP (8, _Float16, s16, <, lt) |
| 46 |
EMULATE_CONDMOV_LOOP (8, _Float16, s16, >, gt) |
| 47 |
EMULATE_CONDMOV_LOOP (8, _Float16, s16, ==, eq) |
| 48 |
EMULATE_CONDMOV_LOOP (32, s16, _Float16, <, lt) |
| 49 |
EMULATE_CONDMOV_LOOP (32, s16, _Float16, >, gt) |
| 50 |
EMULATE_CONDMOV_LOOP (32, s16, _Float16, ==, eq) |
| 51 |
EMULATE_CONDMOV_LOOP (16, s16, _Float16, <, lt) |
| 52 |
EMULATE_CONDMOV_LOOP (16, s16, _Float16, >, gt) |
| 53 |
EMULATE_CONDMOV_LOOP (16, s16, _Float16, ==, eq) |
| 54 |
EMULATE_CONDMOV_LOOP (8, s16, _Float16, <, lt) |
| 55 |
EMULATE_CONDMOV_LOOP (8, s16, _Float16, >, gt) |
| 56 |
EMULATE_CONDMOV_LOOP (8, s16, _Float16, ==, eq) |
| 57 |
EMULATE_CONDMOV_LOOP (32, _Float16, u16, <, lt) |
| 58 |
EMULATE_CONDMOV_LOOP (32, _Float16, u16, >, gt) |
| 59 |
EMULATE_CONDMOV_LOOP (32, _Float16, u16, ==, eq) |
| 60 |
EMULATE_CONDMOV_LOOP (16, _Float16, u16, <, lt) |
| 61 |
EMULATE_CONDMOV_LOOP (16, _Float16, u16, >, gt) |
| 62 |
EMULATE_CONDMOV_LOOP (16, _Float16, u16, ==, eq) |
| 63 |
EMULATE_CONDMOV_LOOP (8, _Float16, u16, <, lt) |
| 64 |
EMULATE_CONDMOV_LOOP (8, _Float16, u16, >, gt) |
| 65 |
EMULATE_CONDMOV_LOOP (8, _Float16, u16, ==, eq) |
| 66 |
EMULATE_CONDMOV_LOOP (32, u16, _Float16, <, lt) |
| 67 |
EMULATE_CONDMOV_LOOP (32, u16, _Float16, >, gt) |
| 68 |
EMULATE_CONDMOV_LOOP (32, u16, _Float16, ==, eq) |
| 69 |
EMULATE_CONDMOV_LOOP (16, u16, _Float16, <, lt) |
| 70 |
EMULATE_CONDMOV_LOOP (16, u16, _Float16, >, gt) |
| 71 |
EMULATE_CONDMOV_LOOP (16, u16, _Float16, ==, eq) |
| 72 |
EMULATE_CONDMOV_LOOP (8, u16, _Float16, <, lt) |
| 73 |
EMULATE_CONDMOV_LOOP (8, u16, _Float16, >, gt) |
| 74 |
EMULATE_CONDMOV_LOOP (8, u16, _Float16, ==, eq) |
| 75 |
|
| 76 |
void init() |
| 77 |
{ |
| 78 |
int i; |
| 79 |
for (i = 0; i < 32; i++) |
| 80 |
{ |
| 81 |
ua[i] = sa[i] = a[i] = i; |
| 82 |
ub[i] = sb[i] = b[i] = i; |
| 83 |
uc[i] = sc[i] = c[i] = (32 - i) * 2; |
| 84 |
uexp[i] = sexp[i] = fexp[i] = -1; |
| 85 |
uref[i] = sref[i] = fref[i] = -1; |
| 86 |
} |
| 87 |
} |
| 88 |
|
| 89 |
int check_cond(void *a, void *b, int size) |
| 90 |
{ |
| 91 |
int i; |
| 92 |
u16 *pa = (u16 *)a, *pb = (u16 *)b; |
| 93 |
for (i = 0; i < size; i++) |
| 94 |
if (pa[i] != pb[i]) |
| 95 |
return 0; |
| 96 |
return 1; |
| 97 |
} |
| 98 |
|
| 99 |
#define TEST_CONDMOV_LOOP(size, name) \ |
| 100 |
{ \ |
| 101 |
init (); \ |
| 102 |
scalar_cond_##size##_Float16_Float16##name (a, b, c, fexp); \ |
| 103 |
loop_cond_##size##_Float16_Float16##name (a, b, c, fref); \ |
| 104 |
if (!check_cond ((void *)fexp, (void *)fref, size)) \ |
| 105 |
abort(); \ |
| 106 |
\ |
| 107 |
init (); \ |
| 108 |
scalar_cond_##size##_Float16s16##name (a, b, sc, sexp); \ |
| 109 |
loop_cond_##size##_Float16s16##name (a, b, sc, sref); \ |
| 110 |
if (!check_cond ((void *)sexp, (void *)sref, size)) \ |
| 111 |
abort(); \ |
| 112 |
\ |
| 113 |
init (); \ |
| 114 |
scalar_cond_##size##s16_Float16##name (sa, sb, c, fexp); \ |
| 115 |
loop_cond_##size##s16_Float16##name (sa, sb, c, fref); \ |
| 116 |
if (!check_cond ((void *)fexp, (void *)fref, size)) \ |
| 117 |
abort(); \ |
| 118 |
\ |
| 119 |
init (); \ |
| 120 |
scalar_cond_##size##_Float16u16##name (a, b, uc, uexp); \ |
| 121 |
loop_cond_##size##_Float16u16##name (a, b, uc, uref); \ |
| 122 |
if (!check_cond ((void *)uexp, (void *)uref, size)) \ |
| 123 |
abort(); \ |
| 124 |
\ |
| 125 |
init (); \ |
| 126 |
scalar_cond_##size##u16_Float16##name (ua, ub, c, fexp); \ |
| 127 |
loop_cond_##size##u16_Float16##name (ua, ub, c, fref); \ |
| 128 |
if (!check_cond ((void *)fexp, (void *)fref, size)) \ |
| 129 |
abort(); \ |
| 130 |
} |
| 131 |
|
| 132 |
static void condmov_test() |
| 133 |
{ |
| 134 |
TEST_CONDMOV_LOOP (32, lt) |
| 135 |
TEST_CONDMOV_LOOP (32, gt) |
| 136 |
TEST_CONDMOV_LOOP (32, eq) |
| 137 |
TEST_CONDMOV_LOOP (16, lt) |
| 138 |
TEST_CONDMOV_LOOP (16, gt) |
| 139 |
TEST_CONDMOV_LOOP (16, eq) |
| 140 |
TEST_CONDMOV_LOOP (8, lt) |
| 141 |
TEST_CONDMOV_LOOP (8, gt) |
| 142 |
TEST_CONDMOV_LOOP (8, eq) |
| 143 |
} |