| 1 |
/* { dg-do run { target { ! ia32 } } } */ |
| 2 |
/* { dg-options "-O1 -msse4.1" } */ |
| 3 |
/* { dg-require-effective-target sse4 } */ |
| 4 |
|
| 5 |
#include "sse4_1-check.h" |
| 6 |
|
| 7 |
typedef _Float16 v4hf __attribute__((vector_size(8))); |
| 8 |
|
| 9 |
v4hf |
| 10 |
__attribute__((noipa)) |
| 11 |
vector_init_dupv4hf (_Float16 a) |
| 12 |
{ |
| 13 |
return __extension__(v4hf){a, a, a, a}; |
| 14 |
} |
| 15 |
|
| 16 |
v4hf |
| 17 |
__attribute__((noipa)) |
| 18 |
vector_init_allzero (_Float16 a) |
| 19 |
{ |
| 20 |
return __extension__(v4hf){0, 0, 0, 0}; |
| 21 |
} |
| 22 |
|
| 23 |
v4hf |
| 24 |
__attribute__((noipa)) |
| 25 |
vector_init_one_nonzero (_Float16 a) |
| 26 |
{ |
| 27 |
return __extension__(v4hf){0, 0, a, 0}; |
| 28 |
} |
| 29 |
|
| 30 |
v4hf |
| 31 |
__attribute__((noipa)) |
| 32 |
vector_init_one_var (_Float16 a) |
| 33 |
{ |
| 34 |
return __extension__(v4hf){1, 2, a, 4}; |
| 35 |
} |
| 36 |
|
| 37 |
v4hf |
| 38 |
__attribute__((noipa)) |
| 39 |
vector_init_general (_Float16 a, _Float16 a1, _Float16 a2, _Float16 a3) |
| 40 |
{ |
| 41 |
return __extension__(v4hf){a3, a2, a1, a}; |
| 42 |
} |
| 43 |
|
| 44 |
v4hf |
| 45 |
__attribute__((noipa)) |
| 46 |
vec_set (_Float16 a, v4hf b) |
| 47 |
{ |
| 48 |
b[1] = a; |
| 49 |
return b; |
| 50 |
} |
| 51 |
|
| 52 |
v4hf |
| 53 |
__attribute__((noipa)) |
| 54 |
vec_set_var (_Float16 a, v4hf b, int c) |
| 55 |
{ |
| 56 |
b[c] = a; |
| 57 |
return b; |
| 58 |
} |
| 59 |
|
| 60 |
_Float16 |
| 61 |
__attribute__((noipa)) |
| 62 |
vec_extract (v4hf b) |
| 63 |
{ |
| 64 |
return b[2]; |
| 65 |
} |
| 66 |
|
| 67 |
static void |
| 68 |
sse4_1_test () |
| 69 |
{ |
| 70 |
typedef union { |
| 71 |
_Float16 a[4]; |
| 72 |
v4hf x;}union64hf; |
| 73 |
union64hf res, exp, src; |
| 74 |
|
| 75 |
res.x = vector_init_dupv4hf (1.0f16); |
| 76 |
for (int i = 0; i != 4; i++) |
| 77 |
exp.a[i] = 1.0f16; |
| 78 |
if (__builtin_memcmp (&res.a[0], &exp.a[0], 8) != 0) |
| 79 |
__builtin_abort (); |
| 80 |
|
| 81 |
res.x = vector_init_allzero (1.0f16); |
| 82 |
for (int i = 0; i != 4; i++) |
| 83 |
exp.a[i] = 0.0f16; |
| 84 |
if (__builtin_memcmp (&res.a[0], &exp.a[0], 8) != 0) |
| 85 |
__builtin_abort (); |
| 86 |
|
| 87 |
res.x = vector_init_one_nonzero (1.0f16); |
| 88 |
for (int i = 0; i != 4; i++) |
| 89 |
exp.a[i] = 0.0f16; |
| 90 |
exp.a[2] = 1.0f16; |
| 91 |
if (__builtin_memcmp (&res.a[0], &exp.a[0], 8) != 0) |
| 92 |
__builtin_abort (); |
| 93 |
|
| 94 |
res.x = vector_init_one_var (3.0f16); |
| 95 |
for (int i = 0; i != 4; i++) |
| 96 |
exp.a[i] = i + 1; |
| 97 |
if (__builtin_memcmp (&res.a[0], &exp.a[0], 8) != 0) |
| 98 |
__builtin_abort (); |
| 99 |
|
| 100 |
res.x = vector_init_general (4.0, 3.0f, 2.0f, 1.0); |
| 101 |
for (int i = 0; i != 4; i++) |
| 102 |
exp.a[i] = 1 + i; |
| 103 |
if (__builtin_memcmp (&res.a[0], &exp.a[0], 8) != 0) |
| 104 |
__builtin_abort (); |
| 105 |
|
| 106 |
for (int i = 0; i != 4; i++) |
| 107 |
{ |
| 108 |
src.a[i] = i; |
| 109 |
exp.a[i] = i; |
| 110 |
} |
| 111 |
res.x = vec_set (3.0f, src.x); |
| 112 |
exp.a[1] = 3.0f; |
| 113 |
if (__builtin_memcmp (&res.a[0], &exp.a[0], 8) != 0) |
| 114 |
__builtin_abort (); |
| 115 |
|
| 116 |
for (int i = 0; i != 4; i++) |
| 117 |
{ |
| 118 |
src.a[i] = i; |
| 119 |
exp.a[i] = i; |
| 120 |
} |
| 121 |
res.x = vec_set_var (3.0f, src.x, 1); |
| 122 |
exp.a[1] = 3.0f; |
| 123 |
if (__builtin_memcmp (&res.a[0], &exp.a[0], 8) != 0) |
| 124 |
__builtin_abort (); |
| 125 |
|
| 126 |
for (int i = 0; i != 4; i++) |
| 127 |
{ |
| 128 |
src.a[i] = i; |
| 129 |
exp.a[i] = i; |
| 130 |
} |
| 131 |
_Float16 res_scalar = vec_extract (src.x); |
| 132 |
if (res_scalar != 2.0f) |
| 133 |
__builtin_abort (); |
| 134 |
return ; |
| 135 |
} |