| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-O2 -ftree-vectorize -msve-vector-bits=scalable" } */ |
| 3 |
|
| 4 |
#include <stdint.h> |
| 5 |
|
| 6 |
#define VEC_PERM(TYPE) \ |
| 7 |
TYPE __attribute__ ((noinline, noclone)) \ |
| 8 |
vec_slp_##TYPE (TYPE *restrict a, int n) \ |
| 9 |
{ \ |
| 10 |
for (int i = 0; i < n; ++i) \ |
| 11 |
{ \ |
| 12 |
a[i * 8] += 99; \ |
| 13 |
a[i * 8 + 1] += 11; \ |
| 14 |
a[i * 8 + 2] += 17; \ |
| 15 |
a[i * 8 + 3] += 80; \ |
| 16 |
a[i * 8 + 4] += 63; \ |
| 17 |
a[i * 8 + 5] += 37; \ |
| 18 |
a[i * 8 + 6] += 24; \ |
| 19 |
a[i * 8 + 7] += 81; \ |
| 20 |
} \ |
| 21 |
} |
| 22 |
|
| 23 |
#define TEST_ALL(T) \ |
| 24 |
T (int8_t) \ |
| 25 |
T (uint8_t) \ |
| 26 |
T (int16_t) \ |
| 27 |
T (uint16_t) \ |
| 28 |
T (int32_t) \ |
| 29 |
T (uint32_t) \ |
| 30 |
T (int64_t) \ |
| 31 |
T (uint64_t) \ |
| 32 |
T (_Float16) \ |
| 33 |
T (float) \ |
| 34 |
T (double) |
| 35 |
|
| 36 |
TEST_ALL (VEC_PERM) |
| 37 |
|
| 38 |
/* 1 for each 8-bit type */ |
| 39 |
/* { dg-final { scan-assembler-times {\tld1rd\tz[0-9]+\.d, } 2 } } */ |
| 40 |
/* 1 for each 16-bit type. */ |
| 41 |
/* { dg-final { scan-assembler-times {\tld1rqh\tz[0-9]+\.h, } 3 } } */ |
| 42 |
/* { dg-final { scan-assembler-times {\tld1rqd\tz[0-9]+\.d, } 18 } } */ |
| 43 |
/* The 32-bit types need: |
| 44 |
|
| 45 |
ZIP1 ZIP2 |
| 46 |
|
| 47 |
and the 64-bit types need: |
| 48 |
|
| 49 |
ZIP1 ZIP2 ZIP1 ZIP2 |
| 50 |
ZIP1 ZIP2 ZIP1 ZIP2. */ |
| 51 |
/* { dg-final { scan-assembler-times {\tzip1\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 15 } } */ |
| 52 |
/* { dg-final { scan-assembler-times {\tzip2\tz[0-9]+\.d, z[0-9]+\.d, z[0-9]+\.d\n} 15 } } */ |
| 53 |
|
| 54 |
/* The loop should be fully-masked. The 32-bit types need two loads |
| 55 |
and stores each and the 64-bit types need four. */ |
| 56 |
/* { dg-final { scan-assembler-times {\tld1b\t} 2 } } */ |
| 57 |
/* { dg-final { scan-assembler-times {\tst1b\t} 2 } } */ |
| 58 |
/* { dg-final { scan-assembler-times {\tld1h\t} 3 } } */ |
| 59 |
/* { dg-final { scan-assembler-times {\tst1h\t} 3 } } */ |
| 60 |
/* { dg-final { scan-assembler-times {\tld1w\t} 6 } } */ |
| 61 |
/* { dg-final { scan-assembler-times {\tst1w\t} 6 } } */ |
| 62 |
/* { dg-final { scan-assembler-times {\tld1d\t} 12 } } */ |
| 63 |
/* { dg-final { scan-assembler-times {\tst1d\t} 12 } } */ |
| 64 |
/* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.b} 4 } } */ |
| 65 |
/* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.h} 6 } } */ |
| 66 |
/* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.s} 12 } } */ |
| 67 |
/* { dg-final { scan-assembler-times {\twhilelo\tp[0-7]\.d} 24 } } */ |
| 68 |
/* { dg-final { scan-assembler-not {\tldr} } } */ |
| 69 |
/* { dg-final { scan-assembler-not {\tstr} } } */ |
| 70 |
|
| 71 |
/* { dg-final { scan-assembler-not {\tuqdec[bh]\t} } } */ |
| 72 |
/* We use UQDECW instead of UQDECD ..., MUL #2. */ |
| 73 |
/* { dg-final { scan-assembler-times {\tuqdecw\t} 6 } } */ |
| 74 |
/* { dg-final { scan-assembler-times {\tuqdecd\t} 6 } } */ |