| 1 |
/* { dg-do assemble { target aarch64_asm_sve2p1_ok } } */ |
| 2 |
/* { dg-do compile { target { ! aarch64_asm_sve2p1_ok } } } */ |
| 3 |
/* { dg-skip-if "" { *-*-* } { "-DSTREAMING_COMPATIBLE" } { "" } } */ |
| 4 |
/* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */ |
| 5 |
|
| 6 |
#include "test_sve_acle.h" |
| 7 |
|
| 8 |
#pragma GCC target "+sve2p1" |
| 9 |
|
| 10 |
/* |
| 11 |
** ld1uwq_f32_base: |
| 12 |
** ld1w {z0\.q}, p0/z, \[x0\] |
| 13 |
** ret |
| 14 |
*/ |
| 15 |
TEST_LOAD (ld1uwq_f32_base, svfloat32_t, float32_t, |
| 16 |
z0 = svld1uwq_f32 (p0, x0), |
| 17 |
z0 = svld1uwq (p0, x0)) |
| 18 |
|
| 19 |
/* |
| 20 |
** ld1uwq_f32_index: |
| 21 |
** ld1w {z0\.q}, p0/z, \[x0, x1, lsl 2\] |
| 22 |
** ret |
| 23 |
*/ |
| 24 |
TEST_LOAD (ld1uwq_f32_index, svfloat32_t, float32_t, |
| 25 |
z0 = svld1uwq_f32 (p0, x0 + x1), |
| 26 |
z0 = svld1uwq (p0, x0 + x1)) |
| 27 |
|
| 28 |
/* |
| 29 |
** ld1uwq_f32_1: |
| 30 |
** ld1w {z0\.q}, p0/z, \[x0, #1, mul vl\] |
| 31 |
** ret |
| 32 |
*/ |
| 33 |
TEST_LOAD (ld1uwq_f32_1, svfloat32_t, float32_t, |
| 34 |
z0 = svld1uwq_f32 (p0, x0 + svcntw () / 4), |
| 35 |
z0 = svld1uwq (p0, x0 + svcntw () / 4)) |
| 36 |
|
| 37 |
/* |
| 38 |
** ld1uwq_f32_7: |
| 39 |
** ld1w {z0\.q}, p0/z, \[x0, #7, mul vl\] |
| 40 |
** ret |
| 41 |
*/ |
| 42 |
TEST_LOAD (ld1uwq_f32_7, svfloat32_t, float32_t, |
| 43 |
z0 = svld1uwq_f32 (p0, x0 + svcntw () * 7 / 4), |
| 44 |
z0 = svld1uwq (p0, x0 + svcntw () * 7 / 4)) |
| 45 |
|
| 46 |
/* Moving the constant into a register would also be OK. */ |
| 47 |
/* |
| 48 |
** ld1uwq_f32_8: |
| 49 |
** incb x0, all, mul #2 |
| 50 |
** ld1w {z0\.q}, p0/z, \[x0\] |
| 51 |
** ret |
| 52 |
*/ |
| 53 |
TEST_LOAD (ld1uwq_f32_8, svfloat32_t, float32_t, |
| 54 |
z0 = svld1uwq_f32 (p0, x0 + svcntw () * 2), |
| 55 |
z0 = svld1uwq (p0, x0 + svcntw () * 2)) |
| 56 |
|
| 57 |
/* |
| 58 |
** ld1uwq_f32_m1: |
| 59 |
** ld1w {z0\.q}, p0/z, \[x0, #-1, mul vl\] |
| 60 |
** ret |
| 61 |
*/ |
| 62 |
TEST_LOAD (ld1uwq_f32_m1, svfloat32_t, float32_t, |
| 63 |
z0 = svld1uwq_f32 (p0, x0 - svcntw () / 4), |
| 64 |
z0 = svld1uwq (p0, x0 - svcntw () / 4)) |
| 65 |
|
| 66 |
/* |
| 67 |
** ld1uwq_f32_m8: |
| 68 |
** ld1w {z0\.q}, p0/z, \[x0, #-8, mul vl\] |
| 69 |
** ret |
| 70 |
*/ |
| 71 |
TEST_LOAD (ld1uwq_f32_m8, svfloat32_t, float32_t, |
| 72 |
z0 = svld1uwq_f32 (p0, x0 - svcntw () * 2), |
| 73 |
z0 = svld1uwq (p0, x0 - svcntw () * 2)) |
| 74 |
|
| 75 |
/* Moving the constant into a register would also be OK. */ |
| 76 |
/* |
| 77 |
** ld1uwq_f32_m9: |
| 78 |
** decw x0, all, mul #9 |
| 79 |
** ld1w {z0\.q}, p0/z, \[x0\] |
| 80 |
** ret |
| 81 |
*/ |
| 82 |
TEST_LOAD (ld1uwq_f32_m9, svfloat32_t, float32_t, |
| 83 |
z0 = svld1uwq_f32 (p0, x0 - svcntw () * 9 / 4), |
| 84 |
z0 = svld1uwq (p0, x0 - svcntw () * 9 / 4)) |
| 85 |
|
| 86 |
/* |
| 87 |
** ld1uwq_vnum_f32_0: |
| 88 |
** ld1w {z0\.q}, p0/z, \[x0\] |
| 89 |
** ret |
| 90 |
*/ |
| 91 |
TEST_LOAD (ld1uwq_vnum_f32_0, svfloat32_t, float32_t, |
| 92 |
z0 = svld1uwq_vnum_f32 (p0, x0, 0), |
| 93 |
z0 = svld1uwq_vnum (p0, x0, 0)) |
| 94 |
|
| 95 |
/* |
| 96 |
** ld1uwq_vnum_f32_1: |
| 97 |
** ld1w {z0\.q}, p0/z, \[x0, #1, mul vl\] |
| 98 |
** ret |
| 99 |
*/ |
| 100 |
TEST_LOAD (ld1uwq_vnum_f32_1, svfloat32_t, float32_t, |
| 101 |
z0 = svld1uwq_vnum_f32 (p0, x0, 1), |
| 102 |
z0 = svld1uwq_vnum (p0, x0, 1)) |
| 103 |
|
| 104 |
/* |
| 105 |
** ld1uwq_vnum_f32_7: |
| 106 |
** ld1w {z0\.q}, p0/z, \[x0, #7, mul vl\] |
| 107 |
** ret |
| 108 |
*/ |
| 109 |
TEST_LOAD (ld1uwq_vnum_f32_7, svfloat32_t, float32_t, |
| 110 |
z0 = svld1uwq_vnum_f32 (p0, x0, 7), |
| 111 |
z0 = svld1uwq_vnum (p0, x0, 7)) |
| 112 |
|
| 113 |
/* Moving the constant into a register would also be OK. */ |
| 114 |
/* |
| 115 |
** ld1uwq_vnum_f32_8: |
| 116 |
** incb x0, all, mul #2 |
| 117 |
** ld1w {z0\.q}, p0/z, \[x0\] |
| 118 |
** ret |
| 119 |
*/ |
| 120 |
TEST_LOAD (ld1uwq_vnum_f32_8, svfloat32_t, float32_t, |
| 121 |
z0 = svld1uwq_vnum_f32 (p0, x0, 8), |
| 122 |
z0 = svld1uwq_vnum (p0, x0, 8)) |
| 123 |
|
| 124 |
/* |
| 125 |
** ld1uwq_vnum_f32_m1: |
| 126 |
** ld1w {z0\.q}, p0/z, \[x0, #-1, mul vl\] |
| 127 |
** ret |
| 128 |
*/ |
| 129 |
TEST_LOAD (ld1uwq_vnum_f32_m1, svfloat32_t, float32_t, |
| 130 |
z0 = svld1uwq_vnum_f32 (p0, x0, -1), |
| 131 |
z0 = svld1uwq_vnum (p0, x0, -1)) |
| 132 |
|
| 133 |
/* |
| 134 |
** ld1uwq_vnum_f32_m8: |
| 135 |
** ld1w {z0\.q}, p0/z, \[x0, #-8, mul vl\] |
| 136 |
** ret |
| 137 |
*/ |
| 138 |
TEST_LOAD (ld1uwq_vnum_f32_m8, svfloat32_t, float32_t, |
| 139 |
z0 = svld1uwq_vnum_f32 (p0, x0, -8), |
| 140 |
z0 = svld1uwq_vnum (p0, x0, -8)) |
| 141 |
|
| 142 |
/* Moving the constant into a register would also be OK. */ |
| 143 |
/* |
| 144 |
** ld1uwq_vnum_f32_m9: |
| 145 |
** decw x0, all, mul #9 |
| 146 |
** ld1w {z0\.q}, p0/z, \[x0\] |
| 147 |
** ret |
| 148 |
*/ |
| 149 |
TEST_LOAD (ld1uwq_vnum_f32_m9, svfloat32_t, float32_t, |
| 150 |
z0 = svld1uwq_vnum_f32 (p0, x0, -9), |
| 151 |
z0 = svld1uwq_vnum (p0, x0, -9)) |
| 152 |
|
| 153 |
/* Using MUL to calculate an index would also be OK. */ |
| 154 |
/* |
| 155 |
** ld1uwq_vnum_f32_x1: |
| 156 |
** cntw (x[0-9]+) |
| 157 |
** madd (x[0-9]+), (x1, \1|\1, x1), x0 |
| 158 |
** ld1w {z0\.q}, p0/z, \[\2\] |
| 159 |
** ret |
| 160 |
*/ |
| 161 |
TEST_LOAD (ld1uwq_vnum_f32_x1, svfloat32_t, float32_t, |
| 162 |
z0 = svld1uwq_vnum_f32 (p0, x0, x1), |
| 163 |
z0 = svld1uwq_vnum (p0, x0, x1)) |