| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-mavx512bw -mavx512vl -mno-avx512bf16 -mno-avxneconvert -O2 -ffast-math" } */ |
| 3 |
/* { dg-final { scan-assembler-times {(?n)(?:vpermw|vpshufb)} 6 } } */ |
| 4 |
|
| 5 |
typedef float v4sf __attribute__((vector_size(16))); |
| 6 |
typedef float v8sf __attribute__((vector_size(32))); |
| 7 |
typedef float v16sf __attribute__((vector_size(64))); |
| 8 |
typedef __bf16 v4bf __attribute__((vector_size(8))); |
| 9 |
typedef __bf16 v8bf __attribute__((vector_size(16))); |
| 10 |
typedef __bf16 v16bf __attribute__((vector_size(32))); |
| 11 |
|
| 12 |
v4bf |
| 13 |
foo (v4sf b, v4sf a) |
| 14 |
{ |
| 15 |
return __builtin_convertvector (a, v4bf); |
| 16 |
} |
| 17 |
|
| 18 |
v8bf |
| 19 |
foo2 (v8sf b, v8sf a) |
| 20 |
{ |
| 21 |
return __builtin_convertvector (a, v8bf); |
| 22 |
} |
| 23 |
|
| 24 |
v16bf |
| 25 |
foo3 (v16sf b, v16sf a) |
| 26 |
{ |
| 27 |
return __builtin_convertvector (a, v16bf); |
| 28 |
} |
| 29 |
|
| 30 |
v4bf |
| 31 |
foo_mem (v4sf* a) |
| 32 |
{ |
| 33 |
return __builtin_convertvector (*a, v4bf); |
| 34 |
} |
| 35 |
|
| 36 |
v8bf |
| 37 |
foo2_mem (v8sf* a) |
| 38 |
{ |
| 39 |
return __builtin_convertvector (*a, v8bf); |
| 40 |
} |
| 41 |
|
| 42 |
v16bf |
| 43 |
foo3_mem (v16sf* a) |
| 44 |
{ |
| 45 |
return __builtin_convertvector (*a, v16bf); |
| 46 |
} |