| 1 |
/* { dg-do run } */ |
| 2 |
/* { dg-options "-O2 -mavx512f" } */ |
| 3 |
/* { dg-require-effective-target avx512f } */ |
| 4 |
|
| 5 |
#define AVX512F |
| 6 |
|
| 7 |
#include "avx512f-helper.h" |
| 8 |
|
| 9 |
__attribute__((noinline, noclone)) int |
| 10 |
test_reduce_add_epi32 (__m512i a) |
| 11 |
{ |
| 12 |
return _mm512_reduce_add_epi32 (a); |
| 13 |
} |
| 14 |
|
| 15 |
__attribute__((noinline, noclone)) int |
| 16 |
test_reduce_mul_epi32 (__m512i a) |
| 17 |
{ |
| 18 |
return _mm512_reduce_mul_epi32 (a); |
| 19 |
} |
| 20 |
|
| 21 |
__attribute__((noinline, noclone)) int |
| 22 |
test_reduce_and_epi32 (__m512i a) |
| 23 |
{ |
| 24 |
return _mm512_reduce_and_epi32 (a); |
| 25 |
} |
| 26 |
|
| 27 |
__attribute__((noinline, noclone)) int |
| 28 |
test_reduce_or_epi32 (__m512i a) |
| 29 |
{ |
| 30 |
return _mm512_reduce_or_epi32 (a); |
| 31 |
} |
| 32 |
|
| 33 |
__attribute__((noinline, noclone)) int |
| 34 |
test_mask_reduce_add_epi32 (__mmask16 u, __m512i a) |
| 35 |
{ |
| 36 |
return _mm512_mask_reduce_add_epi32 (u, a); |
| 37 |
} |
| 38 |
|
| 39 |
__attribute__((noinline, noclone)) int |
| 40 |
test_mask_reduce_mul_epi32 (__mmask16 u, __m512i a) |
| 41 |
{ |
| 42 |
return _mm512_mask_reduce_mul_epi32 (u, a); |
| 43 |
} |
| 44 |
|
| 45 |
__attribute__((noinline, noclone)) int |
| 46 |
test_mask_reduce_and_epi32 (__mmask16 u, __m512i a) |
| 47 |
{ |
| 48 |
return _mm512_mask_reduce_and_epi32 (u, a); |
| 49 |
} |
| 50 |
|
| 51 |
__attribute__((noinline, noclone)) int |
| 52 |
test_mask_reduce_or_epi32 (__mmask16 u, __m512i a) |
| 53 |
{ |
| 54 |
return _mm512_mask_reduce_or_epi32 (u, a); |
| 55 |
} |
| 56 |
|
| 57 |
__attribute__((noinline, noclone)) int |
| 58 |
test_reduce_min_epi32 (__m512i a) |
| 59 |
{ |
| 60 |
return _mm512_reduce_min_epi32 (a); |
| 61 |
} |
| 62 |
|
| 63 |
__attribute__((noinline, noclone)) int |
| 64 |
test_reduce_max_epi32 (__m512i a) |
| 65 |
{ |
| 66 |
return _mm512_reduce_max_epi32 (a); |
| 67 |
} |
| 68 |
|
| 69 |
__attribute__((noinline, noclone)) unsigned int |
| 70 |
test_reduce_min_epu32 (__m512i a) |
| 71 |
{ |
| 72 |
return _mm512_reduce_min_epu32 (a); |
| 73 |
} |
| 74 |
|
| 75 |
__attribute__((noinline, noclone)) unsigned int |
| 76 |
test_reduce_max_epu32 (__m512i a) |
| 77 |
{ |
| 78 |
return _mm512_reduce_max_epu32 (a); |
| 79 |
} |
| 80 |
|
| 81 |
__attribute__((noinline, noclone)) int |
| 82 |
test_mask_reduce_min_epi32 (__mmask16 u, __m512i a) |
| 83 |
{ |
| 84 |
return _mm512_mask_reduce_min_epi32 (u, a); |
| 85 |
} |
| 86 |
|
| 87 |
__attribute__((noinline, noclone)) int |
| 88 |
test_mask_reduce_max_epi32 (__mmask16 u, __m512i a) |
| 89 |
{ |
| 90 |
return _mm512_mask_reduce_max_epi32 (u, a); |
| 91 |
} |
| 92 |
|
| 93 |
__attribute__((noinline, noclone)) unsigned int |
| 94 |
test_mask_reduce_min_epu32 (__mmask16 u, __m512i a) |
| 95 |
{ |
| 96 |
return _mm512_mask_reduce_min_epu32 (u, a); |
| 97 |
} |
| 98 |
|
| 99 |
__attribute__((noinline, noclone)) unsigned int |
| 100 |
test_mask_reduce_max_epu32 (__mmask16 u, __m512i a) |
| 101 |
{ |
| 102 |
return _mm512_mask_reduce_max_epu32 (u, a); |
| 103 |
} |
| 104 |
|
| 105 |
__attribute__((noinline, noclone)) float |
| 106 |
test_reduce_add_ps (__m512 a) |
| 107 |
{ |
| 108 |
return _mm512_reduce_add_ps (a); |
| 109 |
} |
| 110 |
|
| 111 |
__attribute__((noinline, noclone)) float |
| 112 |
test_reduce_mul_ps (__m512 a) |
| 113 |
{ |
| 114 |
return _mm512_reduce_mul_ps (a); |
| 115 |
} |
| 116 |
|
| 117 |
__attribute__((noinline, noclone)) float |
| 118 |
test_mask_reduce_add_ps (__mmask16 u, __m512 a) |
| 119 |
{ |
| 120 |
return _mm512_mask_reduce_add_ps (u, a); |
| 121 |
} |
| 122 |
|
| 123 |
__attribute__((noinline, noclone)) float |
| 124 |
test_mask_reduce_mul_ps (__mmask16 u, __m512 a) |
| 125 |
{ |
| 126 |
return _mm512_mask_reduce_mul_ps (u, a); |
| 127 |
} |
| 128 |
|
| 129 |
__attribute__((noinline, noclone)) float |
| 130 |
test_reduce_min_ps (__m512 a) |
| 131 |
{ |
| 132 |
return _mm512_reduce_min_ps (a); |
| 133 |
} |
| 134 |
|
| 135 |
__attribute__((noinline, noclone)) float |
| 136 |
test_reduce_max_ps (__m512 a) |
| 137 |
{ |
| 138 |
return _mm512_reduce_max_ps (a); |
| 139 |
} |
| 140 |
|
| 141 |
__attribute__((noinline, noclone)) float |
| 142 |
test_mask_reduce_min_ps (__mmask16 u, __m512 a) |
| 143 |
{ |
| 144 |
return _mm512_mask_reduce_min_ps (u, a); |
| 145 |
} |
| 146 |
|
| 147 |
__attribute__((noinline, noclone)) float |
| 148 |
test_mask_reduce_max_ps (__mmask16 u, __m512 a) |
| 149 |
{ |
| 150 |
return _mm512_mask_reduce_max_ps (u, a); |
| 151 |
} |
| 152 |
|
| 153 |
__attribute__((noinline, noclone)) long long |
| 154 |
test_reduce_add_epi64 (__m512i a) |
| 155 |
{ |
| 156 |
return _mm512_reduce_add_epi64 (a); |
| 157 |
} |
| 158 |
|
| 159 |
__attribute__((noinline, noclone)) long long |
| 160 |
test_reduce_mul_epi64 (__m512i a) |
| 161 |
{ |
| 162 |
return _mm512_reduce_mul_epi64 (a); |
| 163 |
} |
| 164 |
|
| 165 |
__attribute__((noinline, noclone)) long long |
| 166 |
test_reduce_and_epi64 (__m512i a) |
| 167 |
{ |
| 168 |
return _mm512_reduce_and_epi64 (a); |
| 169 |
} |
| 170 |
|
| 171 |
__attribute__((noinline, noclone)) long long |
| 172 |
test_reduce_or_epi64 (__m512i a) |
| 173 |
{ |
| 174 |
return _mm512_reduce_or_epi64 (a); |
| 175 |
} |
| 176 |
|
| 177 |
__attribute__((noinline, noclone)) long long |
| 178 |
test_mask_reduce_add_epi64 (__mmask8 u, __m512i a) |
| 179 |
{ |
| 180 |
return _mm512_mask_reduce_add_epi64 (u, a); |
| 181 |
} |
| 182 |
|
| 183 |
__attribute__((noinline, noclone)) long long |
| 184 |
test_mask_reduce_mul_epi64 (__mmask8 u, __m512i a) |
| 185 |
{ |
| 186 |
return _mm512_mask_reduce_mul_epi64 (u, a); |
| 187 |
} |
| 188 |
|
| 189 |
__attribute__((noinline, noclone)) long long |
| 190 |
test_mask_reduce_and_epi64 (__mmask8 u, __m512i a) |
| 191 |
{ |
| 192 |
return _mm512_mask_reduce_and_epi64 (u, a); |
| 193 |
} |
| 194 |
|
| 195 |
__attribute__((noinline, noclone)) long long |
| 196 |
test_mask_reduce_or_epi64 (__mmask8 u, __m512i a) |
| 197 |
{ |
| 198 |
return _mm512_mask_reduce_or_epi64 (u, a); |
| 199 |
} |
| 200 |
|
| 201 |
__attribute__((noinline, noclone)) long long |
| 202 |
test_reduce_min_epi64 (__m512i a) |
| 203 |
{ |
| 204 |
return _mm512_reduce_min_epi64 (a); |
| 205 |
} |
| 206 |
|
| 207 |
__attribute__((noinline, noclone)) long long |
| 208 |
test_reduce_max_epi64 (__m512i a) |
| 209 |
{ |
| 210 |
return _mm512_reduce_max_epi64 (a); |
| 211 |
} |
| 212 |
|
| 213 |
__attribute__((noinline, noclone)) unsigned long long |
| 214 |
test_reduce_min_epu64 (__m512i a) |
| 215 |
{ |
| 216 |
return _mm512_reduce_min_epu64 (a); |
| 217 |
} |
| 218 |
|
| 219 |
__attribute__((noinline, noclone)) unsigned long long |
| 220 |
test_reduce_max_epu64 (__m512i a) |
| 221 |
{ |
| 222 |
return _mm512_reduce_max_epu64 (a); |
| 223 |
} |
| 224 |
|
| 225 |
__attribute__((noinline, noclone)) long long |
| 226 |
test_mask_reduce_min_epi64 (__mmask8 u, __m512i a) |
| 227 |
{ |
| 228 |
return _mm512_mask_reduce_min_epi64 (u, a); |
| 229 |
} |
| 230 |
|
| 231 |
__attribute__((noinline, noclone)) long long |
| 232 |
test_mask_reduce_max_epi64 (__mmask8 u, __m512i a) |
| 233 |
{ |
| 234 |
return _mm512_mask_reduce_max_epi64 (u, a); |
| 235 |
} |
| 236 |
|
| 237 |
__attribute__((noinline, noclone)) unsigned long long |
| 238 |
test_mask_reduce_min_epu64 (__mmask8 u, __m512i a) |
| 239 |
{ |
| 240 |
return _mm512_mask_reduce_min_epu64 (u, a); |
| 241 |
} |
| 242 |
|
| 243 |
__attribute__((noinline, noclone)) unsigned long long |
| 244 |
test_mask_reduce_max_epu64 (__mmask8 u, __m512i a) |
| 245 |
{ |
| 246 |
return _mm512_mask_reduce_max_epu64 (u, a); |
| 247 |
} |
| 248 |
|
| 249 |
__attribute__((noinline, noclone)) double |
| 250 |
test_reduce_add_pd (__m512d a) |
| 251 |
{ |
| 252 |
return _mm512_reduce_add_pd (a); |
| 253 |
} |
| 254 |
|
| 255 |
__attribute__((noinline, noclone)) double |
| 256 |
test_reduce_mul_pd (__m512d a) |
| 257 |
{ |
| 258 |
return _mm512_reduce_mul_pd (a); |
| 259 |
} |
| 260 |
|
| 261 |
__attribute__((noinline, noclone)) double |
| 262 |
test_mask_reduce_add_pd (__mmask8 u, __m512d a) |
| 263 |
{ |
| 264 |
return _mm512_mask_reduce_add_pd (u, a); |
| 265 |
} |
| 266 |
|
| 267 |
__attribute__((noinline, noclone)) double |
| 268 |
test_mask_reduce_mul_pd (__mmask8 u, __m512d a) |
| 269 |
{ |
| 270 |
return _mm512_mask_reduce_mul_pd (u, a); |
| 271 |
} |
| 272 |
|
| 273 |
__attribute__((noinline, noclone)) double |
| 274 |
test_reduce_min_pd (__m512d a) |
| 275 |
{ |
| 276 |
return _mm512_reduce_min_pd (a); |
| 277 |
} |
| 278 |
|
| 279 |
__attribute__((noinline, noclone)) double |
| 280 |
test_reduce_max_pd (__m512d a) |
| 281 |
{ |
| 282 |
return _mm512_reduce_max_pd (a); |
| 283 |
} |
| 284 |
|
| 285 |
__attribute__((noinline, noclone)) double |
| 286 |
test_mask_reduce_min_pd (__mmask8 u, __m512d a) |
| 287 |
{ |
| 288 |
return _mm512_mask_reduce_min_pd (u, a); |
| 289 |
} |
| 290 |
|
| 291 |
__attribute__((noinline, noclone)) double |
| 292 |
test_mask_reduce_max_pd (__mmask8 u, __m512d a) |
| 293 |
{ |
| 294 |
return _mm512_mask_reduce_max_pd (u, a); |
| 295 |
} |
| 296 |
|
| 297 |
#define TESTOP(opname, op, type, suffix, neutral) \ |
| 298 |
do { \ |
| 299 |
type r1 = _mm512_reduce_##opname##_##suffix (v.x); \ |
| 300 |
type r2 = test_reduce_##opname##_##suffix (v.x); \ |
| 301 |
type r3 = neutral; \ |
| 302 |
if (r1 != r2) \ |
| 303 |
__builtin_abort (); \ |
| 304 |
for (int i = 0; i < SIZE; i++) \ |
| 305 |
r3 = r3 op v.a[i]; \ |
| 306 |
if (r1 != r3) \ |
| 307 |
__builtin_abort (); \ |
| 308 |
type r4 = _mm512_mask_reduce_##opname##_##suffix (MASK_VALUE, v.x); \ |
| 309 |
type r5 = test_mask_reduce_##opname##_##suffix (MASK_VALUE, v.x); \ |
| 310 |
if (r4 != r5) \ |
| 311 |
__builtin_abort (); \ |
| 312 |
r3 = neutral; \ |
| 313 |
for (int i = 0; i < SIZE; i++) \ |
| 314 |
if (MASK_VALUE & (1 << i)) \ |
| 315 |
r3 = r3 op v.a[i]; \ |
| 316 |
if (r4 != r3) \ |
| 317 |
__builtin_abort (); \ |
| 318 |
type r6 = _mm512_mask_reduce_##opname##_##suffix (0, v.x); \ |
| 319 |
type r7 = test_mask_reduce_##opname##_##suffix (0, v.x); \ |
| 320 |
if (r6 != r7 || r6 != neutral) \ |
| 321 |
__builtin_abort (); \ |
| 322 |
} while (0) |
| 323 |
|
| 324 |
#define SIZE (AVX512F_LEN / 32) |
| 325 |
#include "avx512f-mask-type.h" |
| 326 |
|
| 327 |
#define TEST_EPI32(c1, c2, c3, c4, c5, c6, c7, c8, \ |
| 328 |
c9, c10, c11, c12, c13, c14, c15, c16) \ |
| 329 |
do { \ |
| 330 |
UNION_TYPE (AVX512F_LEN, i_d) v; \ |
| 331 |
v.x = _mm512_set_epi32 (c1, c2, c3, c4, c5, c6, c7, c8, \ |
| 332 |
c9, c10, c11, c12, c13, c14, c15, c16); \ |
| 333 |
TESTOP (add, +, int, epi32, 0); \ |
| 334 |
TESTOP (mul, *, int, epi32, 1); \ |
| 335 |
TESTOP (and, &, int, epi32, ~0); \ |
| 336 |
TESTOP (or, |, int, epi32, 0); \ |
| 337 |
TESTOP (min, < v.a[i] ? r3 :, int, epi32, __INT_MAX__); \ |
| 338 |
TESTOP (max, > v.a[i] ? r3 :, int, epi32, -__INT_MAX__ - 1); \ |
| 339 |
TESTOP (min, < (unsigned) v.a[i] ? r3 :, unsigned, epu32, ~0U); \ |
| 340 |
TESTOP (max, > (unsigned) v.a[i] ? r3 :, unsigned, epu32, 0); \ |
| 341 |
} while (0) |
| 342 |
|
| 343 |
#define TEST_PS(c1, c2, c3, c4, c5, c6, c7, c8, \ |
| 344 |
c9, c10, c11, c12, c13, c14, c15, c16) \ |
| 345 |
do { \ |
| 346 |
UNION_TYPE (AVX512F_LEN, ) v; \ |
| 347 |
v.x = _mm512_set_ps (c1, c2, c3, c4, c5, c6, c7, c8, \ |
| 348 |
c9, c10, c11, c12, c13, c14, c15, c16); \ |
| 349 |
TESTOP (add, +, float, ps, 0.0f); \ |
| 350 |
TESTOP (mul, *, float, ps, 1.0f); \ |
| 351 |
TESTOP (min, < v.a[i] ? r3 :, float, ps, __builtin_inff ()); \ |
| 352 |
TESTOP (max, > v.a[i] ? r3 :, float, ps, -__builtin_inff ()); \ |
| 353 |
} while (0) |
| 354 |
|
| 355 |
static void |
| 356 |
test_epi32_ps (void) |
| 357 |
{ |
| 358 |
TEST_EPI32 (1, 2, 3, 4, 5, 6, 6, 5, 4, 3, 2, 1, 7, 6, 5, 4); |
| 359 |
TEST_EPI32 (-1, 15, -1, 7, -1, 7, -1, -1, 6, 6, -1, -1, -1, -1, 7, 6); |
| 360 |
TEST_PS (1, 2, 3, 4, 5, 6, 6, 5, 4, 3, 2, 1, 7, 6, 5, 4); |
| 361 |
TEST_PS (1.25f, 2.25f, -0.25f, 4.0f, -2.0f, 4.0f, -3.0f, 2.0f, |
| 362 |
-0.5f, -1.0f, 1.0f, -1.0f, 1.0f, 1.0f, 2.0f, 4.0f); |
| 363 |
} |
| 364 |
|
| 365 |
#undef SIZE |
| 366 |
#define SIZE (AVX512F_LEN / 64) |
| 367 |
#include "avx512f-mask-type.h" |
| 368 |
|
| 369 |
#define TEST_EPI64(c1, c2, c3, c4, c5, c6, c7, c8) \ |
| 370 |
do { \ |
| 371 |
UNION_TYPE (AVX512F_LEN, i_q) v; \ |
| 372 |
v.x = _mm512_set_epi64 (c1, c2, c3, c4, c5, c6, c7, c8); \ |
| 373 |
TESTOP (add, +, long long, epi64, 0); \ |
| 374 |
TESTOP (mul, *, long long, epi64, 1); \ |
| 375 |
TESTOP (and, &, long long, epi64, ~0LL); \ |
| 376 |
TESTOP (or, |, long long, epi64, 0); \ |
| 377 |
TESTOP (min, < v.a[i] ? r3 :, long long, epi64, __LONG_LONG_MAX__); \ |
| 378 |
TESTOP (max, > v.a[i] ? r3 :, long long, epi64, \ |
| 379 |
-__LONG_LONG_MAX__ - 1); \ |
| 380 |
TESTOP (min, < (unsigned long long) v.a[i] ? r3 :, \ |
| 381 |
unsigned long long, epu64, ~0ULL); \ |
| 382 |
TESTOP (max, > (unsigned long long) v.a[i] ? r3 :, \ |
| 383 |
unsigned long long, epu64, 0); \ |
| 384 |
} while (0) |
| 385 |
|
| 386 |
#define TEST_PD(c1, c2, c3, c4, c5, c6, c7, c8) \ |
| 387 |
do { \ |
| 388 |
UNION_TYPE (AVX512F_LEN, d) v; \ |
| 389 |
v.x = _mm512_set_pd (c1, c2, c3, c4, c5, c6, c7, c8); \ |
| 390 |
TESTOP (add, +, double, pd, 0.0); \ |
| 391 |
TESTOP (mul, *, double, pd, 1.0); \ |
| 392 |
TESTOP (min, < v.a[i] ? r3 :, double, pd, __builtin_inf ()); \ |
| 393 |
TESTOP (max, > v.a[i] ? r3 :, double, pd, -__builtin_inf ()); \ |
| 394 |
} while (0) |
| 395 |
|
| 396 |
static void |
| 397 |
test_epi64_pd (void) |
| 398 |
{ |
| 399 |
TEST_EPI64 (1, 2, 3, 4, 5, 6, 6, 5); |
| 400 |
TEST_EPI64 (-1, 15, -1, 7, -1, 7, -1, -1); |
| 401 |
TEST_PD (1, 2, 3, 4, 5, 6, 6, 5); |
| 402 |
TEST_PD (1.25f, 2.25f, -0.25f, 4.0f, -2.0f, 4.0f, -3.0f, 2.0f); |
| 403 |
} |
| 404 |
|
| 405 |
void |
| 406 |
test_512 (void) |
| 407 |
{ |
| 408 |
test_epi32_ps (); |
| 409 |
test_epi64_pd (); |
| 410 |
} |