| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-O -mfp16-format=ieee" } */ |
| 3 |
/* { dg-require-effective-target arm_v8_1m_mve_ok } */ |
| 4 |
/* { dg-add-options arm_v8_1m_mve } */ |
| 5 |
/* { dg-additional-options "-mfloat-abi=hard" } */ |
| 6 |
/* { dg-final { check-function-bodies "**" "" } } */ |
| 7 |
|
| 8 |
/* |
| 9 |
** r_w: |
| 10 |
** vmov.f16 r0, s0 @ __fp16 |
| 11 |
** bx lr |
| 12 |
*/ |
| 13 |
void |
| 14 |
r_w (_Float16 s0) |
| 15 |
{ |
| 16 |
register _Float16 r0 asm ("r0"); |
| 17 |
r0 = s0; |
| 18 |
asm volatile ("" :: "r" (r0)); |
| 19 |
} |
| 20 |
|
| 21 |
/* |
| 22 |
** w_r: |
| 23 |
** vmov.f16 s0, r0 @ __fp16 |
| 24 |
** bx lr |
| 25 |
*/ |
| 26 |
_Float16 |
| 27 |
w_r () |
| 28 |
{ |
| 29 |
register _Float16 r0 asm ("r0"); |
| 30 |
asm volatile ("" : "=r" (r0)); |
| 31 |
return r0; |
| 32 |
} |
| 33 |
|
| 34 |
/* |
| 35 |
** w_w: |
| 36 |
** vmov s1, s0 @ __fp16 |
| 37 |
** bx lr |
| 38 |
*/ |
| 39 |
void |
| 40 |
w_w (_Float16 s0) |
| 41 |
{ |
| 42 |
register _Float16 s1 asm ("s1"); |
| 43 |
s1 = s0; |
| 44 |
asm volatile ("" :: "w" (s1)); |
| 45 |
} |
| 46 |
|
| 47 |
/* |
| 48 |
** r_m_m128: |
| 49 |
** sub (r[0-9]+), r0, #256 |
| 50 |
** ldrh r1, \[\1\] @ __fp16 |
| 51 |
** bx lr |
| 52 |
*/ |
| 53 |
void |
| 54 |
r_m_m128 (_Float16 *r0) |
| 55 |
{ |
| 56 |
register _Float16 r1 asm ("r1"); |
| 57 |
r1 = r0[-128]; |
| 58 |
asm volatile ("" :: "r" (r1)); |
| 59 |
} |
| 60 |
|
| 61 |
/* |
| 62 |
** r_m_m127: |
| 63 |
** ldrh r1, \[r0, #-254\] @ __fp16 |
| 64 |
** bx lr |
| 65 |
*/ |
| 66 |
void |
| 67 |
r_m_m127 (_Float16 *r0) |
| 68 |
{ |
| 69 |
register _Float16 r1 asm ("r1"); |
| 70 |
r1 = r0[-127]; |
| 71 |
asm volatile ("" :: "r" (r1)); |
| 72 |
} |
| 73 |
|
| 74 |
/* |
| 75 |
** r_m_m1: |
| 76 |
** ldrh r1, \[r0, #-2\] @ __fp16 |
| 77 |
** bx lr |
| 78 |
*/ |
| 79 |
void |
| 80 |
r_m_m1 (_Float16 *r0) |
| 81 |
{ |
| 82 |
register _Float16 r1 asm ("r1"); |
| 83 |
r1 = r0[-1]; |
| 84 |
asm volatile ("" :: "r" (r1)); |
| 85 |
} |
| 86 |
|
| 87 |
/* |
| 88 |
** r_m_0: |
| 89 |
** ldrh r1, \[r0\] @ __fp16 |
| 90 |
** bx lr |
| 91 |
*/ |
| 92 |
void |
| 93 |
r_m_0 (_Float16 *r0) |
| 94 |
{ |
| 95 |
register _Float16 r1 asm ("r1"); |
| 96 |
r1 = r0[0]; |
| 97 |
asm volatile ("" :: "r" (r1)); |
| 98 |
} |
| 99 |
|
| 100 |
/* |
| 101 |
** r_m_1: |
| 102 |
** ldrh r1, \[r0, #2\] @ __fp16 |
| 103 |
** bx lr |
| 104 |
*/ |
| 105 |
void |
| 106 |
r_m_1 (_Float16 *r0) |
| 107 |
{ |
| 108 |
register _Float16 r1 asm ("r1"); |
| 109 |
r1 = r0[1]; |
| 110 |
asm volatile ("" :: "r" (r1)); |
| 111 |
} |
| 112 |
|
| 113 |
/* |
| 114 |
** r_m_255: |
| 115 |
** ldrh r1, \[r0, #510\] @ __fp16 |
| 116 |
** bx lr |
| 117 |
*/ |
| 118 |
void |
| 119 |
r_m_255 (_Float16 *r0) |
| 120 |
{ |
| 121 |
register _Float16 r1 asm ("r1"); |
| 122 |
r1 = r0[255]; |
| 123 |
asm volatile ("" :: "r" (r1)); |
| 124 |
} |
| 125 |
|
| 126 |
/* |
| 127 |
** r_m_256: |
| 128 |
** ldrh r1, \[r0, #512\] @ __fp16 |
| 129 |
** bx lr |
| 130 |
*/ |
| 131 |
void |
| 132 |
r_m_256 (_Float16 *r0) |
| 133 |
{ |
| 134 |
register _Float16 r1 asm ("r1"); |
| 135 |
r1 = r0[256]; |
| 136 |
asm volatile ("" :: "r" (r1)); |
| 137 |
} |
| 138 |
|
| 139 |
/* ??? This could be done in one instruction, but without mve.fp, |
| 140 |
it makes more sense for memory_operand to enforce the GPR range. */ |
| 141 |
/* |
| 142 |
** w_m_m128: |
| 143 |
** sub (r[0-9]+), r0, #256 |
| 144 |
** vldr.16 s0, \[\1\] |
| 145 |
** bx lr |
| 146 |
*/ |
| 147 |
void |
| 148 |
w_m_m128 (_Float16 *r0) |
| 149 |
{ |
| 150 |
register _Float16 s0 asm ("s0"); |
| 151 |
s0 = r0[-128]; |
| 152 |
asm volatile ("" :: "w" (s0)); |
| 153 |
} |
| 154 |
|
| 155 |
/* |
| 156 |
** w_m_m127: |
| 157 |
** vldr.16 s0, \[r0, #-254\] |
| 158 |
** bx lr |
| 159 |
*/ |
| 160 |
void |
| 161 |
w_m_m127 (_Float16 *r0) |
| 162 |
{ |
| 163 |
register _Float16 s0 asm ("s0"); |
| 164 |
s0 = r0[-127]; |
| 165 |
asm volatile ("" :: "w" (s0)); |
| 166 |
} |
| 167 |
|
| 168 |
/* |
| 169 |
** w_m_m1: |
| 170 |
** vldr.16 s0, \[r0, #-2\] |
| 171 |
** bx lr |
| 172 |
*/ |
| 173 |
void |
| 174 |
w_m_m1 (_Float16 *r0) |
| 175 |
{ |
| 176 |
register _Float16 s0 asm ("s0"); |
| 177 |
s0 = r0[-1]; |
| 178 |
asm volatile ("" :: "w" (s0)); |
| 179 |
} |
| 180 |
|
| 181 |
/* |
| 182 |
** w_m_0: |
| 183 |
** vldr.16 s0, \[r0\] |
| 184 |
** bx lr |
| 185 |
*/ |
| 186 |
void |
| 187 |
w_m_0 (_Float16 *r0) |
| 188 |
{ |
| 189 |
register _Float16 s0 asm ("s0"); |
| 190 |
s0 = r0[0]; |
| 191 |
asm volatile ("" :: "w" (s0)); |
| 192 |
} |
| 193 |
|
| 194 |
/* |
| 195 |
** w_m_1: |
| 196 |
** vldr.16 s0, \[r0, #2\] |
| 197 |
** bx lr |
| 198 |
*/ |
| 199 |
void |
| 200 |
w_m_1 (_Float16 *r0) |
| 201 |
{ |
| 202 |
register _Float16 s0 asm ("s0"); |
| 203 |
s0 = r0[1]; |
| 204 |
asm volatile ("" :: "w" (s0)); |
| 205 |
} |
| 206 |
|
| 207 |
/* |
| 208 |
** w_m_255: |
| 209 |
** vldr.16 s0, \[r0, #510\] |
| 210 |
** bx lr |
| 211 |
*/ |
| 212 |
void |
| 213 |
w_m_255 (_Float16 *r0) |
| 214 |
{ |
| 215 |
register _Float16 s0 asm ("s0"); |
| 216 |
s0 = r0[255]; |
| 217 |
asm volatile ("" :: "w" (s0)); |
| 218 |
} |
| 219 |
|
| 220 |
/* |
| 221 |
** w_m_256: |
| 222 |
** add (r[0-9]+), r0, #512 |
| 223 |
** vldr.16 s0, \[\1\] |
| 224 |
** bx lr |
| 225 |
*/ |
| 226 |
void |
| 227 |
w_m_256 (_Float16 *r0) |
| 228 |
{ |
| 229 |
register _Float16 s0 asm ("s0"); |
| 230 |
s0 = r0[256]; |
| 231 |
asm volatile ("" :: "w" (s0)); |
| 232 |
} |
| 233 |
|
| 234 |
/* |
| 235 |
** m_m128_r: |
| 236 |
** sub (r[0-9]+), r0, #256 |
| 237 |
** strh r1, \[\1\] @ __fp16 |
| 238 |
** bx lr |
| 239 |
*/ |
| 240 |
void |
| 241 |
m_m128_r (_Float16 *r0) |
| 242 |
{ |
| 243 |
register _Float16 r1 asm ("r1"); |
| 244 |
asm volatile ("" : "=r" (r1)); |
| 245 |
r0[-128] = r1; |
| 246 |
} |
| 247 |
|
| 248 |
/* |
| 249 |
** m_m127_r: |
| 250 |
** strh r1, \[r0, #-254\] @ __fp16 |
| 251 |
** bx lr |
| 252 |
*/ |
| 253 |
void |
| 254 |
m_m127_r (_Float16 *r0) |
| 255 |
{ |
| 256 |
register _Float16 r1 asm ("r1"); |
| 257 |
asm volatile ("" : "=r" (r1)); |
| 258 |
r0[-127] = r1; |
| 259 |
} |
| 260 |
|
| 261 |
/* |
| 262 |
** m_m1_r: |
| 263 |
** strh r1, \[r0, #-2\] @ __fp16 |
| 264 |
** bx lr |
| 265 |
*/ |
| 266 |
void |
| 267 |
m_m1_r (_Float16 *r0) |
| 268 |
{ |
| 269 |
register _Float16 r1 asm ("r1"); |
| 270 |
asm volatile ("" : "=r" (r1)); |
| 271 |
r0[-1] = r1; |
| 272 |
} |
| 273 |
|
| 274 |
/* |
| 275 |
** m_0_r: |
| 276 |
** strh r1, \[r0\] @ __fp16 |
| 277 |
** bx lr |
| 278 |
*/ |
| 279 |
void |
| 280 |
m_0_r (_Float16 *r0) |
| 281 |
{ |
| 282 |
register _Float16 r1 asm ("r1"); |
| 283 |
asm volatile ("" : "=r" (r1)); |
| 284 |
r0[0] = r1; |
| 285 |
} |
| 286 |
|
| 287 |
/* |
| 288 |
** m_1_r: |
| 289 |
** strh r1, \[r0, #2\] @ __fp16 |
| 290 |
** bx lr |
| 291 |
*/ |
| 292 |
void |
| 293 |
m_1_r (_Float16 *r0) |
| 294 |
{ |
| 295 |
register _Float16 r1 asm ("r1"); |
| 296 |
asm volatile ("" : "=r" (r1)); |
| 297 |
r0[1] = r1; |
| 298 |
} |
| 299 |
|
| 300 |
/* |
| 301 |
** m_255_r: |
| 302 |
** strh r1, \[r0, #510\] @ __fp16 |
| 303 |
** bx lr |
| 304 |
*/ |
| 305 |
void |
| 306 |
m_255_r (_Float16 *r0) |
| 307 |
{ |
| 308 |
register _Float16 r1 asm ("r1"); |
| 309 |
asm volatile ("" : "=r" (r1)); |
| 310 |
r0[255] = r1; |
| 311 |
} |
| 312 |
|
| 313 |
/* |
| 314 |
** m_256_r: |
| 315 |
** strh r1, \[r0, #512\] @ __fp16 |
| 316 |
** bx lr |
| 317 |
*/ |
| 318 |
void |
| 319 |
m_256_r (_Float16 *r0) |
| 320 |
{ |
| 321 |
register _Float16 r1 asm ("r1"); |
| 322 |
asm volatile ("" : "=r" (r1)); |
| 323 |
r0[256] = r1; |
| 324 |
} |
| 325 |
|
| 326 |
/* ??? This could be done in one instruction, but without mve.fp, |
| 327 |
it makes more sense for memory_operand to enforce the GPR range. */ |
| 328 |
/* |
| 329 |
** m_m128_w: |
| 330 |
** sub (r[0-9]+), r0, #256 |
| 331 |
** vstr.16 s0, \[\1\] |
| 332 |
** bx lr |
| 333 |
*/ |
| 334 |
void |
| 335 |
m_m128_w (_Float16 *r0) |
| 336 |
{ |
| 337 |
register _Float16 s0 asm ("s0"); |
| 338 |
asm volatile ("" : "=w" (s0)); |
| 339 |
r0[-128] = s0; |
| 340 |
} |
| 341 |
|
| 342 |
/* |
| 343 |
** m_m127_w: |
| 344 |
** vstr.16 s0, \[r0, #-254\] |
| 345 |
** bx lr |
| 346 |
*/ |
| 347 |
void |
| 348 |
m_m127_w (_Float16 *r0) |
| 349 |
{ |
| 350 |
register _Float16 s0 asm ("s0"); |
| 351 |
asm volatile ("" : "=w" (s0)); |
| 352 |
r0[-127] = s0; |
| 353 |
} |
| 354 |
|
| 355 |
/* |
| 356 |
** m_m1_w: |
| 357 |
** vstr.16 s0, \[r0, #-2\] |
| 358 |
** bx lr |
| 359 |
*/ |
| 360 |
void |
| 361 |
m_m1_w (_Float16 *r0) |
| 362 |
{ |
| 363 |
register _Float16 s0 asm ("s0"); |
| 364 |
asm volatile ("" : "=w" (s0)); |
| 365 |
r0[-1] = s0; |
| 366 |
} |
| 367 |
|
| 368 |
/* |
| 369 |
** m_0_w: |
| 370 |
** vstr.16 s0, \[r0\] |
| 371 |
** bx lr |
| 372 |
*/ |
| 373 |
void |
| 374 |
m_0_w (_Float16 *r0) |
| 375 |
{ |
| 376 |
register _Float16 s0 asm ("s0"); |
| 377 |
asm volatile ("" : "=w" (s0)); |
| 378 |
r0[0] = s0; |
| 379 |
} |
| 380 |
|
| 381 |
/* |
| 382 |
** m_1_w: |
| 383 |
** vstr.16 s0, \[r0, #2\] |
| 384 |
** bx lr |
| 385 |
*/ |
| 386 |
void |
| 387 |
m_1_w (_Float16 *r0) |
| 388 |
{ |
| 389 |
register _Float16 s0 asm ("s0"); |
| 390 |
asm volatile ("" : "=w" (s0)); |
| 391 |
r0[1] = s0; |
| 392 |
} |
| 393 |
|
| 394 |
/* |
| 395 |
** m_255_w: |
| 396 |
** vstr.16 s0, \[r0, #510\] |
| 397 |
** bx lr |
| 398 |
*/ |
| 399 |
void |
| 400 |
m_255_w (_Float16 *r0) |
| 401 |
{ |
| 402 |
register _Float16 s0 asm ("s0"); |
| 403 |
asm volatile ("" : "=w" (s0)); |
| 404 |
r0[255] = s0; |
| 405 |
} |
| 406 |
|
| 407 |
/* |
| 408 |
** m_256_w: |
| 409 |
** add (r[0-9]+), r0, #512 |
| 410 |
** vstr.16 s0, \[\1\] |
| 411 |
** bx lr |
| 412 |
*/ |
| 413 |
void |
| 414 |
m_256_w (_Float16 *r0) |
| 415 |
{ |
| 416 |
register _Float16 s0 asm ("s0"); |
| 417 |
asm volatile ("" : "=w" (s0)); |
| 418 |
r0[256] = s0; |
| 419 |
} |