| 1 |
/* { dg-do compile { target { ! riscv_abi_e } } } */ |
| 2 |
/* { dg-add-options riscv_v } */ |
| 3 |
/* { dg-additional-options "-O3 -mrvv-max-lmul=dynamic" } */ |
| 4 |
/* { dg-final { check-function-bodies "**" "" } } */ |
| 5 |
|
| 6 |
#define MIN_VECTOR_BYTES (__riscv_v_min_vlen / 8) |
| 7 |
|
| 8 |
/* Tiny memsets should use scalar ops. |
| 9 |
** f1: |
| 10 |
** sb\s+a1,0\(a0\) |
| 11 |
** ret |
| 12 |
*/ |
| 13 |
void * |
| 14 |
f1 (void *a, int const b) |
| 15 |
{ |
| 16 |
return __builtin_memset (a, b, 1); |
| 17 |
} |
| 18 |
|
| 19 |
/* Tiny memsets should use scalar ops. |
| 20 |
** f2: |
| 21 |
** sb\s+a1,0\(a0\) |
| 22 |
** sb\s+a1,1\(a0\) |
| 23 |
** ret |
| 24 |
*/ |
| 25 |
void * |
| 26 |
f2 (void *a, int const b) |
| 27 |
{ |
| 28 |
return __builtin_memset (a, b, 2); |
| 29 |
} |
| 30 |
|
| 31 |
/* Tiny memsets should use scalar ops. |
| 32 |
** f3: |
| 33 |
** sb\s+a1,0\(a0\) |
| 34 |
** sb\s+a1,1\(a0\) |
| 35 |
** sb\s+a1,2\(a0\) |
| 36 |
** ret |
| 37 |
*/ |
| 38 |
void * |
| 39 |
f3 (void *a, int const b) |
| 40 |
{ |
| 41 |
return __builtin_memset (a, b, 3); |
| 42 |
} |
| 43 |
|
| 44 |
/* Vectorise+inline minimum vector register width with LMUL=1 |
| 45 |
** f4: |
| 46 |
** ( |
| 47 |
** vsetivli\s+zero,\d+,e8,m1,ta,ma |
| 48 |
** | |
| 49 |
** li\s+a\d+,\d+ |
| 50 |
** vsetvli\s+zero,a\d+,e8,m1,ta,ma |
| 51 |
** ) |
| 52 |
** vmv\.v\.x\s+v\d+,a1 |
| 53 |
** vse8\.v\s+v\d+,0\(a0\) |
| 54 |
** ret |
| 55 |
*/ |
| 56 |
void * |
| 57 |
f4 (void *a, int const b) |
| 58 |
{ |
| 59 |
return __builtin_memset (a, b, MIN_VECTOR_BYTES); |
| 60 |
} |
| 61 |
|
| 62 |
/* Vectorised code should use smallest lmul known to fit length |
| 63 |
** f5: |
| 64 |
** ( |
| 65 |
** vsetivli\s+zero,\d+,e8,m2,ta,ma |
| 66 |
** | |
| 67 |
** li\s+a\d+,\d+ |
| 68 |
** vsetvli\s+zero,a\d+,e8,m2,ta,ma |
| 69 |
** ) |
| 70 |
** vmv\.v\.x\s+v\d+,a1 |
| 71 |
** vse8\.v\s+v\d+,0\(a0\) |
| 72 |
** ret |
| 73 |
*/ |
| 74 |
void * |
| 75 |
f5 (void *a, int const b) |
| 76 |
{ |
| 77 |
return __builtin_memset (a, b, MIN_VECTOR_BYTES + 1); |
| 78 |
} |
| 79 |
|
| 80 |
/* Vectorise+inline up to LMUL=8 |
| 81 |
** f6: |
| 82 |
** li\s+a\d+,\d+ |
| 83 |
** vsetvli\s+zero,a\d+,e8,m8,ta,ma |
| 84 |
** vmv\.v\.x\s+v\d+,a1 |
| 85 |
** vse8\.v\s+v\d+,0\(a0\) |
| 86 |
** ret |
| 87 |
*/ |
| 88 |
void * |
| 89 |
f6 (void *a, int const b) |
| 90 |
{ |
| 91 |
return __builtin_memset (a, b, MIN_VECTOR_BYTES * 8); |
| 92 |
} |
| 93 |
|
| 94 |
/* Don't vectorise if the move is too large for one operation. |
| 95 |
** f7: |
| 96 |
** li\s+a2,\d+ |
| 97 |
** tail\s+memset(?:@plt)? |
| 98 |
*/ |
| 99 |
void * |
| 100 |
f7 (void *a, int const b) |
| 101 |
{ |
| 102 |
return __builtin_memset (a, b, MIN_VECTOR_BYTES * 8 + 1); |
| 103 |
} |