| 1 |
/* { dg-do assemble { target aarch64_asm_sve_ok } } */ |
| 2 |
/* { dg-options "-O2 -msve-vector-bits=256 --save-temps" } */ |
| 3 |
|
| 4 |
#include <stdint.h> |
| 5 |
|
| 6 |
typedef int32_t vnx4si __attribute__((vector_size (32))); |
| 7 |
typedef int64_t vnx2di __attribute__((vector_size (32))); |
| 8 |
|
| 9 |
typedef float vnx4sf __attribute__((vector_size (32))); |
| 10 |
typedef double vnx2df __attribute__((vector_size (32))); |
| 11 |
|
| 12 |
/* Predicate vector: 1 0 0 0 ... */ |
| 13 |
|
| 14 |
#define MASK_32 { 0, 33, 34, 35, 4, 37, 38, 39, 8, 41, 42, 43, 12, \ |
| 15 |
45, 46, 47, 16, 49, 50, 51, 20, 53, 54, 55, 24, \ |
| 16 |
57, 58, 59, 28, 61, 62, 63 } |
| 17 |
|
| 18 |
#define MASK_16 {0, 17, 18, 19, 4, 21, 22, 23, 8, 25, 26, 27, 12, 29, 30, 31} |
| 19 |
#define MASK_8 {0, 9, 10, 11, 4, 13, 14, 15} |
| 20 |
#define MASK_4 {0, 5, 6, 7} |
| 21 |
|
| 22 |
#define INDEX_8 vnx4si |
| 23 |
#define INDEX_4 vnx2di |
| 24 |
|
| 25 |
#define PERMUTE(type, nunits) \ |
| 26 |
type permute_##type (type x, type y) \ |
| 27 |
{ \ |
| 28 |
return __builtin_shuffle (x, y, (INDEX_##nunits) MASK_##nunits); \ |
| 29 |
} |
| 30 |
|
| 31 |
PERMUTE(vnx4si, 8) |
| 32 |
PERMUTE(vnx2di, 4) |
| 33 |
|
| 34 |
PERMUTE(vnx4sf, 8) |
| 35 |
PERMUTE(vnx2df, 4) |
| 36 |
|
| 37 |
/* { dg-final { scan-assembler-not {\ttbl\t} } } */ |
| 38 |
|
| 39 |
/* { dg-final { scan-assembler-times {\tsel\tz[0-9]+\.s, p[0-9]+, z[0-9]+\.s, z[0-9]+\.s\n} 2 } } */ |
| 40 |
/* { dg-final { scan-assembler-times {\tsel\tz[0-9]+\.d, p[0-9]+, z[0-9]+\.d, z[0-9]+\.d\n} 2 } } */ |
| 41 |
|
| 42 |
/* { dg-final { scan-assembler-times {\tptrue\tp[0-9]+\.d, vl4\n} 2 } } */ |