| 1 |
// { dg-options "-O2 -fno-schedule-insns -fno-schedule-insns2" } |
| 2 |
// { dg-final { check-function-bodies "**" "" } } |
| 3 |
|
| 4 |
#include <arm_sme.h> |
| 5 |
|
| 6 |
#pragma GCC target "+sme2" |
| 7 |
|
| 8 |
// This file deliberately contains nonsense code. |
| 9 |
|
| 10 |
/* |
| 11 |
** test1: |
| 12 |
** ptrue [^\n]+ |
| 13 |
** ld1w [^\n]+ |
| 14 |
** ld1w [^\n]+ |
| 15 |
** ld1w [^\n]+ |
| 16 |
** ld1w [^\n]+ |
| 17 |
** st1w [^\n]+ |
| 18 |
** st1w [^\n]+ |
| 19 |
** st1w [^\n]+ |
| 20 |
** st1w [^\n]+ |
| 21 |
** ret |
| 22 |
*/ |
| 23 |
void test1(int32_t *dest, int32_t *src) __arm_streaming |
| 24 |
{ |
| 25 |
svcount_t pg = svptrue_c32(); |
| 26 |
svint32x4_t l0 = svld1_vnum_x4(pg, src, 0); |
| 27 |
svint32x4_t l1 = svld1_vnum_x4(pg, src, 4); |
| 28 |
svint32x4_t l2 = svld1_vnum_x4(pg, src, 8); |
| 29 |
svint32x4_t l3 = svld1_vnum_x4(pg, src, 12); |
| 30 |
svst1_vnum(pg, dest, 0, |
| 31 |
svcreate4(svget4(l0, 0), svget4(l1, 0), |
| 32 |
svget4(l2, 0), svget4(l3, 0))); |
| 33 |
svst1_vnum(pg, dest, 4, |
| 34 |
svcreate4(svget4(l0, 1), svget4(l1, 1), |
| 35 |
svget4(l2, 1), svget4(l3, 1))); |
| 36 |
svst1_vnum(pg, dest, 8, |
| 37 |
svcreate4(svget4(l0, 2), svget4(l1, 2), |
| 38 |
svget4(l2, 2), svget4(l3, 2))); |
| 39 |
svst1_vnum(pg, dest, 12, |
| 40 |
svcreate4(svget4(l0, 3), svget4(l1, 3), |
| 41 |
svget4(l2, 3), svget4(l3, 3))); |
| 42 |
} |
| 43 |
|
| 44 |
/* |
| 45 |
** test2: |
| 46 |
** ptrue [^\n]+ |
| 47 |
** ld1w [^\n]+ |
| 48 |
** ld1w [^\n]+ |
| 49 |
** ld1w [^\n]+ |
| 50 |
** ld1w [^\n]+ |
| 51 |
** st1w [^\n]+ |
| 52 |
** st1w [^\n]+ |
| 53 |
** st1w [^\n]+ |
| 54 |
** st1w [^\n]+ |
| 55 |
** st1w [^\n]+ |
| 56 |
** st1w [^\n]+ |
| 57 |
** st1w [^\n]+ |
| 58 |
** st1w [^\n]+ |
| 59 |
** ret |
| 60 |
*/ |
| 61 |
void test2(int32_t *dest, int32_t *src) __arm_streaming |
| 62 |
{ |
| 63 |
svcount_t pg = svptrue_c32(); |
| 64 |
svint32x4_t l0 = svld1_vnum_x4(pg, src, 0); |
| 65 |
svint32x4_t l1 = svld1_vnum_x4(pg, src, 4); |
| 66 |
svint32x4_t l2 = svld1_vnum_x4(pg, src, 8); |
| 67 |
svint32x4_t l3 = svld1_vnum_x4(pg, src, 12); |
| 68 |
svst1_vnum(pg, dest, 0, |
| 69 |
svcreate4(svget4(l0, 0), svget4(l1, 0), |
| 70 |
svget4(l2, 0), svget4(l3, 0))); |
| 71 |
svst1_vnum(pg, dest, 4, |
| 72 |
svcreate4(svget4(l0, 1), svget4(l1, 1), |
| 73 |
svget4(l2, 1), svget4(l3, 1))); |
| 74 |
svst1_vnum(pg, dest, 8, |
| 75 |
svcreate4(svget4(l0, 2), svget4(l1, 2), |
| 76 |
svget4(l2, 2), svget4(l3, 2))); |
| 77 |
svst1_vnum(pg, dest, 12, |
| 78 |
svcreate4(svget4(l0, 3), svget4(l1, 3), |
| 79 |
svget4(l2, 3), svget4(l3, 3))); |
| 80 |
svst1_vnum(pg, dest, 16, |
| 81 |
svcreate4(svget4(l0, 0), svget4(l1, 0), |
| 82 |
svget4(l2, 0), svget4(l3, 0))); |
| 83 |
svst1_vnum(pg, dest, 20, |
| 84 |
svcreate4(svget4(l0, 1), svget4(l1, 1), |
| 85 |
svget4(l2, 1), svget4(l3, 1))); |
| 86 |
svst1_vnum(pg, dest, 24, |
| 87 |
svcreate4(svget4(l0, 2), svget4(l1, 2), |
| 88 |
svget4(l2, 2), svget4(l3, 2))); |
| 89 |
svst1_vnum(pg, dest, 28, |
| 90 |
svcreate4(svget4(l0, 3), svget4(l1, 3), |
| 91 |
svget4(l2, 3), svget4(l3, 3))); |
| 92 |
} |
| 93 |
|
| 94 |
/* |
| 95 |
** test3: |
| 96 |
** ptrue ([^\n]+)\.s |
| 97 |
** ld1w {z16\.s - z19\.s}, \1/z, \[x1\] |
| 98 |
** ld1w {z20\.s - z23\.s}, \1/z, \[x1, #4, mul vl\] |
| 99 |
** ld1w {z24\.s - z27\.s}, \1/z, \[x1, #8, mul vl\] |
| 100 |
** ld1w {z28\.s - z31\.s}, \1/z, \[x1, #12, mul vl\] |
| 101 |
** sclamp [^\n]+ |
| 102 |
** st1w {z19\.s, z23\.s, z27\.s, z31\.s}, \1, \[x0, #12, mul vl\] |
| 103 |
** st1w {z16\.s, z20\.s, z24\.s, z28\.s}, \1, \[x0\] |
| 104 |
** st1w {z17\.s, z21\.s, z25\.s, z29\.s}, \1, \[x0, #4, mul vl\] |
| 105 |
** st1w {z18\.s, z22\.s, z26\.s, z30\.s}, \1, \[x0, #8, mul vl\] |
| 106 |
** ret |
| 107 |
*/ |
| 108 |
void test3(int32_t *dest, int32_t *src) __arm_streaming |
| 109 |
{ |
| 110 |
svcount_t pg = svptrue_c32(); |
| 111 |
svint32x4_t al0 = svld1_vnum_x4(pg, src, 0); |
| 112 |
svint32x4_t l1 = svld1_vnum_x4(pg, src, 4); |
| 113 |
svint32x4_t l2 = svld1_vnum_x4(pg, src, 8); |
| 114 |
svint32x4_t l3 = svld1_vnum_x4(pg, src, 12); |
| 115 |
svint32x4_t l0 = svclamp(al0, svget4(l3, 0), svget4(l3, 1)); |
| 116 |
svst1_vnum(pg, dest, 12, |
| 117 |
svcreate4(svget4(l0, 3), svget4(l1, 3), |
| 118 |
svget4(l2, 3), svget4(l3, 3))); |
| 119 |
svst1_vnum(pg, dest, 0, |
| 120 |
svcreate4(svget4(l0, 0), svget4(l1, 0), |
| 121 |
svget4(l2, 0), svget4(l3, 0))); |
| 122 |
svst1_vnum(pg, dest, 4, |
| 123 |
svcreate4(svget4(l0, 1), svget4(l1, 1), |
| 124 |
svget4(l2, 1), svget4(l3, 1))); |
| 125 |
svst1_vnum(pg, dest, 8, |
| 126 |
svcreate4(svget4(l0, 2), svget4(l1, 2), |
| 127 |
svget4(l2, 2), svget4(l3, 2))); |
| 128 |
} |
| 129 |
|
| 130 |
/* |
| 131 |
** test4: |
| 132 |
** ptrue ([^\n]+)\.s |
| 133 |
** ld1w {z16\.s - z19\.s}, \1/z, \[x1\] |
| 134 |
** ld1w {z20\.s - z23\.s}, \1/z, \[x1, #4, mul vl\] |
| 135 |
** ld1w {z24\.s - z27\.s}, \1/z, \[x1, #8, mul vl\] |
| 136 |
** ld1w {z28\.s - z31\.s}, \1/z, \[x1, #12, mul vl\] |
| 137 |
** sclamp [^\n]+ |
| 138 |
** st1w {z16\.s, z20\.s, z24\.s, z28\.s}, \1, \[x0\] |
| 139 |
** st1w {z17\.s, z21\.s, z25\.s, z29\.s}, \1, \[x0, #4, mul vl\] |
| 140 |
** st1w {z18\.s, z22\.s, z26\.s, z30\.s}, \1, \[x0, #8, mul vl\] |
| 141 |
** st1w {z19\.s, z23\.s, z27\.s, z31\.s}, \1, \[x0, #12, mul vl\] |
| 142 |
** st1w {z16\.s, z20\.s, z24\.s, z28\.s}, \1, \[x0, #16, mul vl\] |
| 143 |
** st1w {z17\.s, z21\.s, z25\.s, z29\.s}, \1, \[x0, #20, mul vl\] |
| 144 |
** st1w {z18\.s, z22\.s, z26\.s, z30\.s}, \1, \[x0, #24, mul vl\] |
| 145 |
** st1w {z19\.s, z23\.s, z27\.s, z31\.s}, \1, \[x0, #28, mul vl\] |
| 146 |
** ... |
| 147 |
** ret |
| 148 |
*/ |
| 149 |
void test4(int32_t *dest, int32_t *src) __arm_streaming |
| 150 |
{ |
| 151 |
svcount_t pg = svptrue_c32(); |
| 152 |
svint32x4_t l0 = svld1_vnum_x4(pg, src, 0); |
| 153 |
svint32x4_t l1 = svld1_vnum_x4(pg, src, 4); |
| 154 |
svint32x4_t l2 = svld1_vnum_x4(pg, src, 8); |
| 155 |
svint32x4_t l3 = svld1_vnum_x4(pg, src, 12); |
| 156 |
l0 = svclamp(l0, svget4(l3, 0), svget4(l3, 1)); |
| 157 |
svst1_vnum(pg, dest, 0, |
| 158 |
svcreate4(svget4(l0, 0), svget4(l1, 0), |
| 159 |
svget4(l2, 0), svget4(l3, 0))); |
| 160 |
svst1_vnum(pg, dest, 4, |
| 161 |
svcreate4(svget4(l0, 1), svget4(l1, 1), |
| 162 |
svget4(l2, 1), svget4(l3, 1))); |
| 163 |
svst1_vnum(pg, dest, 8, |
| 164 |
svcreate4(svget4(l0, 2), svget4(l1, 2), |
| 165 |
svget4(l2, 2), svget4(l3, 2))); |
| 166 |
svst1_vnum(pg, dest, 12, |
| 167 |
svcreate4(svget4(l0, 3), svget4(l1, 3), |
| 168 |
svget4(l2, 3), svget4(l3, 3))); |
| 169 |
svst1_vnum(pg, dest, 16, |
| 170 |
svcreate4(svget4(l0, 0), svget4(l1, 0), |
| 171 |
svget4(l2, 0), svget4(l3, 0))); |
| 172 |
svst1_vnum(pg, dest, 20, |
| 173 |
svcreate4(svget4(l0, 1), svget4(l1, 1), |
| 174 |
svget4(l2, 1), svget4(l3, 1))); |
| 175 |
svst1_vnum(pg, dest, 24, |
| 176 |
svcreate4(svget4(l0, 2), svget4(l1, 2), |
| 177 |
svget4(l2, 2), svget4(l3, 2))); |
| 178 |
svst1_vnum(pg, dest, 28, |
| 179 |
svcreate4(svget4(l0, 3), svget4(l1, 3), |
| 180 |
svget4(l2, 3), svget4(l3, 3))); |
| 181 |
} |
| 182 |
|
| 183 |
/* |
| 184 |
** test6: |
| 185 |
** ptrue [^\n]+ |
| 186 |
** ld1h [^\n]+ |
| 187 |
** sclamp [^\n]+ |
| 188 |
** st1h [^\n]+ |
| 189 |
** ret |
| 190 |
*/ |
| 191 |
void test6(int16_t *ptr) |
| 192 |
__arm_streaming __arm_preserves("za") __arm_inout("zt0") |
| 193 |
{ |
| 194 |
svcount_t pg = svptrue_c16(); |
| 195 |
svint16x4_t x0 = svld1_x4(pg, ptr); |
| 196 |
x0 = svclamp(x0, svget4(x0, 0), svget4(x0, 3)); |
| 197 |
svst1(pg, ptr, x0); |
| 198 |
} |