| 1 |
/* { dg-do assemble { target aarch64_asm_sve_ok } } */ |
| 2 |
/* { dg-options "-O3 -msve-vector-bits=256 --save-temps" } */ |
| 3 |
|
| 4 |
#include <stdint.h> |
| 5 |
|
| 6 |
typedef int8_t vnx16qi __attribute__((vector_size(32))); |
| 7 |
typedef int16_t vnx8hi __attribute__((vector_size(32))); |
| 8 |
typedef int32_t vnx4si __attribute__((vector_size(32))); |
| 9 |
typedef int64_t vnx2di __attribute__((vector_size(32))); |
| 10 |
|
| 11 |
#define DO_OP(TYPE) \ |
| 12 |
void vmla_##TYPE (TYPE *x, TYPE y, TYPE z) \ |
| 13 |
{ \ |
| 14 |
register TYPE dst asm("z0"); \ |
| 15 |
register TYPE src1 asm("z2"); \ |
| 16 |
register TYPE src2 asm("z4"); \ |
| 17 |
dst = *x; \ |
| 18 |
src1 = y; \ |
| 19 |
src2 = z; \ |
| 20 |
asm volatile ("" :: "w" (dst), "w" (src1), "w" (src2)); \ |
| 21 |
dst = (src1 * src2) + dst; \ |
| 22 |
asm volatile ("" :: "w" (dst)); \ |
| 23 |
*x = dst; \ |
| 24 |
} |
| 25 |
|
| 26 |
DO_OP (vnx16qi) |
| 27 |
DO_OP (vnx8hi) |
| 28 |
DO_OP (vnx4si) |
| 29 |
DO_OP (vnx2di) |
| 30 |
|
| 31 |
/* { dg-final { scan-assembler-times {\tmla\tz0\.b, p[0-7]/m, z2\.b, z4\.b\n} 1 } } */ |
| 32 |
/* { dg-final { scan-assembler-times {\tmla\tz0\.h, p[0-7]/m, z2\.h, z4\.h\n} 1 } } */ |
| 33 |
/* { dg-final { scan-assembler-times {\tmla\tz0\.s, p[0-7]/m, z2\.s, z4\.s\n} 1 } } */ |
| 34 |
/* { dg-final { scan-assembler-times {\tmla\tz0\.d, p[0-7]/m, z2\.d, z4\.d\n} 1 } } */ |