| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-march=x86-64 -mavx512fp16 -mavx512vl -O3" } */ |
| 3 |
/* { dg-final { scan-assembler-times "vcvtpd2ps" 3 } } */ |
| 4 |
/* { dg-final { scan-assembler-times "vcvtpd2ph" 3 } } */ |
| 5 |
/* { dg-final { scan-assembler-times "vcvtps2ph" 3 } } */ |
| 6 |
|
| 7 |
#include <x86intrin.h> |
| 8 |
|
| 9 |
typedef _Float16 __v2hf __attribute__ ((__vector_size__ (4))); |
| 10 |
typedef _Float16 __v4hf __attribute__ ((__vector_size__ (8))); |
| 11 |
|
| 12 |
__v2sf mm_cvtpd_ps_builtin_convertvector(__v2df a) |
| 13 |
{ |
| 14 |
return __builtin_convertvector(a, __v2sf); |
| 15 |
} |
| 16 |
|
| 17 |
__v4sf mm256_cvtpd_ps_builtin_convertvector(__v4df a) |
| 18 |
{ |
| 19 |
return __builtin_convertvector(a, __v4sf); |
| 20 |
} |
| 21 |
|
| 22 |
__v8sf mm512_cvtpd_ps_builtin_convertvector(__v8df a) |
| 23 |
{ |
| 24 |
return __builtin_convertvector(a, __v8sf); |
| 25 |
} |
| 26 |
|
| 27 |
__v2hf mm_cvtpd_ph_builtin_convertvector(__v2df a) |
| 28 |
{ |
| 29 |
return __builtin_convertvector(a, __v2hf); |
| 30 |
} |
| 31 |
|
| 32 |
__v4hf mm256_cvtpd_ph_builtin_convertvector(__v4df a) |
| 33 |
{ |
| 34 |
return __builtin_convertvector(a, __v4hf); |
| 35 |
} |
| 36 |
|
| 37 |
__v8hf mm512_cvtpd_ph_builtin_convertvector(__v8df a) |
| 38 |
{ |
| 39 |
return __builtin_convertvector(a, __v8hf); |
| 40 |
} |
| 41 |
|
| 42 |
__v4hf mm_cvtps_ph_builtin_convertvector(__v4sf a) |
| 43 |
{ |
| 44 |
return __builtin_convertvector(a, __v4hf); |
| 45 |
} |
| 46 |
|
| 47 |
__v8hf mm256_cvtps_ph_builtin_convertvector(__v8sf a) |
| 48 |
{ |
| 49 |
return __builtin_convertvector(a, __v8hf); |
| 50 |
} |
| 51 |
|
| 52 |
__v16hf mm512_cvtps_ph_builtin_convertvector(__v16sf a) |
| 53 |
{ |
| 54 |
return __builtin_convertvector(a, __v16hf); |
| 55 |
} |