| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-O" } */ |
| 3 |
/* { dg-final { check-function-bodies "**" "" "" { target lp64 } } } */ |
| 4 |
|
| 5 |
#include <arm_neon.h> |
| 6 |
|
| 7 |
void ext(); |
| 8 |
|
| 9 |
/* |
| 10 |
** s32_1: |
| 11 |
** fmov s0, w0 |
| 12 |
** ins v0\.s\[1\], w1 |
| 13 |
** ret |
| 14 |
*/ |
| 15 |
int32x2_t s32_1(int32_t a0, int32_t a1) { |
| 16 |
if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 17 |
return (int32x2_t) { a1, a0 }; |
| 18 |
else |
| 19 |
return (int32x2_t) { a0, a1 }; |
| 20 |
} |
| 21 |
/* |
| 22 |
** s32_2: |
| 23 |
** fmov s0, w0 |
| 24 |
** ld1 {v0\.s}\[1\], \[x1\] |
| 25 |
** ret |
| 26 |
*/ |
| 27 |
int32x2_t s32_2(int32_t a0, int32_t *ptr) { |
| 28 |
if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 29 |
return (int32x2_t) { ptr[0], a0 }; |
| 30 |
else |
| 31 |
return (int32x2_t) { a0, ptr[0] }; |
| 32 |
} |
| 33 |
/* |
| 34 |
** s32_3: |
| 35 |
** ldr s0, \[x0\] |
| 36 |
** ins v0\.s\[1\], w1 |
| 37 |
** ret |
| 38 |
*/ |
| 39 |
int32x2_t s32_3(int32_t *ptr, int32_t a1) { |
| 40 |
if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 41 |
return (int32x2_t) { a1, ptr[0] }; |
| 42 |
else |
| 43 |
return (int32x2_t) { ptr[0], a1 }; |
| 44 |
} |
| 45 |
/* |
| 46 |
** s32_4: |
| 47 |
** stp w1, w2, \[x0\] |
| 48 |
** ret |
| 49 |
*/ |
| 50 |
void s32_4(int32x2_t *res, int32_t a0, int32_t a1) { |
| 51 |
res[0] = (int32x2_t) { a0, a1 }; |
| 52 |
} |
| 53 |
/* |
| 54 |
** s32_5: |
| 55 |
** stp w1, w2, \[x0, #?4\] |
| 56 |
** ret |
| 57 |
*/ |
| 58 |
void s32_5(uintptr_t res, int32_t a0, int32_t a1) { |
| 59 |
*(int32x2_t *)(res + 4) = (int32x2_t) { a0, a1 }; |
| 60 |
} |
| 61 |
/* Currently uses d8 to hold res across the call. */ |
| 62 |
int32x2_t s32_6(int32_t a0, int32_t a1) { |
| 63 |
int32x2_t res = { a0, a1 }; |
| 64 |
ext (); |
| 65 |
return res; |
| 66 |
} |
| 67 |
|
| 68 |
/* |
| 69 |
** f32_1: |
| 70 |
** uzp1 v0\.2s, v0\.2s, v1\.2s |
| 71 |
** ret |
| 72 |
*/ |
| 73 |
float32x2_t f32_1(float32_t a0, float32_t a1) { |
| 74 |
if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 75 |
return (float32x2_t) { a1, a0 }; |
| 76 |
else |
| 77 |
return (float32x2_t) { a0, a1 }; |
| 78 |
} |
| 79 |
/* |
| 80 |
** f32_2: |
| 81 |
** ld1 {v0\.s}\[1\], \[x0\] |
| 82 |
** ret |
| 83 |
*/ |
| 84 |
float32x2_t f32_2(float32_t a0, float32_t *ptr) { |
| 85 |
if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 86 |
return (float32x2_t) { ptr[0], a0 }; |
| 87 |
else |
| 88 |
return (float32x2_t) { a0, ptr[0] }; |
| 89 |
} |
| 90 |
/* |
| 91 |
** f32_3: |
| 92 |
** ldr s0, \[x0\] |
| 93 |
** uzp1 v0\.2s, v0\.2s, v1\.2s |
| 94 |
** ret |
| 95 |
*/ |
| 96 |
float32x2_t f32_3(float32_t a0, float32_t a1, float32_t *ptr) { |
| 97 |
if (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) |
| 98 |
return (float32x2_t) { a1, ptr[0] }; |
| 99 |
else |
| 100 |
return (float32x2_t) { ptr[0], a1 }; |
| 101 |
} |
| 102 |
/* |
| 103 |
** f32_4: |
| 104 |
** stp s0, s1, \[x0\] |
| 105 |
** ret |
| 106 |
*/ |
| 107 |
void f32_4(float32x2_t *res, float32_t a0, float32_t a1) { |
| 108 |
res[0] = (float32x2_t) { a0, a1 }; |
| 109 |
} |
| 110 |
/* |
| 111 |
** f32_5: |
| 112 |
** stp s0, s1, \[x0, #?4\] |
| 113 |
** ret |
| 114 |
*/ |
| 115 |
void f32_5(uintptr_t res, float32_t a0, float32_t a1) { |
| 116 |
*(float32x2_t *)(res + 4) = (float32x2_t) { a0, a1 }; |
| 117 |
} |
| 118 |
/* Currently uses d8 to hold res across the call. */ |
| 119 |
float32x2_t f32_6(float32_t a0, float32_t a1) { |
| 120 |
float32x2_t res = { a0, a1 }; |
| 121 |
ext (); |
| 122 |
return res; |
| 123 |
} |