| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-O3 -fno-vect-cost-model" } */ |
| 3 |
|
| 4 |
/* Check that we vectorize to a full 128-bit vector for _Float16 and __fp16 |
| 5 |
types. */ |
| 6 |
|
| 7 |
/* Enable ARMv8.2-A+fp16 so we have access to the vector instructions. */ |
| 8 |
#pragma GCC target ("arch=armv8.2-a+fp16") |
| 9 |
|
| 10 |
_Float16 |
| 11 |
sum_Float16 (_Float16 *__restrict__ __attribute__ ((__aligned__ (16))) a, |
| 12 |
_Float16 *__restrict__ __attribute__ ((__aligned__ (16))) b, |
| 13 |
_Float16 *__restrict__ __attribute__ ((__aligned__ (16))) c) |
| 14 |
{ |
| 15 |
for (int i = 0; i < 256; i++) |
| 16 |
a[i] = b[i] + c[i]; |
| 17 |
} |
| 18 |
|
| 19 |
_Float16 |
| 20 |
sum_fp16 (__fp16 *__restrict__ __attribute__ ((__aligned__ (16))) a, |
| 21 |
__fp16 *__restrict__ __attribute__ ((__aligned__ (16))) b, |
| 22 |
__fp16 *__restrict__ __attribute__ ((__aligned__ (16))) c) |
| 23 |
{ |
| 24 |
for (int i = 0; i < 256; i++) |
| 25 |
a[i] = b[i] + c[i]; |
| 26 |
} |
| 27 |
|
| 28 |
/* Two FADD operations on "8h" data widths, one from sum_Float16, one from |
| 29 |
sum_fp16. */ |
| 30 |
/* { dg-final { scan-assembler-times "fadd\tv\[0-9\]\+.8h" 2 } } */ |