1 #include "rs_types.rsh" 2 3 extern float2 __attribute__((overloadable)) convert_float2(int2 c); 4 extern float3 __attribute__((overloadable)) convert_float3(int3 c); 5 extern float4 __attribute__((overloadable)) convert_float4(int4 c); 6 7 extern int2 __attribute__((overloadable)) convert_int2(float2 c); 8 extern int3 __attribute__((overloadable)) convert_int3(float3 c); 9 extern int4 __attribute__((overloadable)) convert_int4(float4 c); 10 11 12 extern float __attribute__((overloadable)) fmin(float v, float v2); 13 extern float2 __attribute__((overloadable)) fmin(float2 v, float v2); 14 extern float3 __attribute__((overloadable)) fmin(float3 v, float v2); 15 extern float4 __attribute__((overloadable)) fmin(float4 v, float v2); 16 17 extern float __attribute__((overloadable)) fmax(float v, float v2); 18 extern float2 __attribute__((overloadable)) fmax(float2 v, float v2); 19 extern float3 __attribute__((overloadable)) fmax(float3 v, float v2); 20 extern float4 __attribute__((overloadable)) fmax(float4 v, float v2); 21 22 // Float ops, 6.11.2 23 24 #define FN_FUNC_FN(fnc) \ 25 extern float2 __attribute__((overloadable)) fnc(float2 v) { \ 26 float2 r; \ 27 r.x = fnc(v.x); \ 28 r.y = fnc(v.y); \ 29 return r; \ 30 } \ 31 extern float3 __attribute__((overloadable)) fnc(float3 v) { \ 32 float3 r; \ 33 r.x = fnc(v.x); \ 34 r.y = fnc(v.y); \ 35 r.z = fnc(v.z); \ 36 return r; \ 37 } \ 38 extern float4 __attribute__((overloadable)) fnc(float4 v) { \ 39 float4 r; \ 40 r.x = fnc(v.x); \ 41 r.y = fnc(v.y); \ 42 r.z = fnc(v.z); \ 43 r.w = fnc(v.w); \ 44 return r; \ 45 } 46 47 #define IN_FUNC_FN(fnc) \ 48 extern int2 __attribute__((overloadable)) fnc(float2 v) { \ 49 int2 r; \ 50 r.x = fnc(v.x); \ 51 r.y = fnc(v.y); \ 52 return r; \ 53 } \ 54 extern int3 __attribute__((overloadable)) fnc(float3 v) { \ 55 int3 r; \ 56 r.x = fnc(v.x); \ 57 r.y = fnc(v.y); \ 58 r.z = fnc(v.z); \ 59 return r; \ 60 } \ 61 extern int4 __attribute__((overloadable)) fnc(float4 v) { \ 62 int4 r; \ 63 r.x = fnc(v.x); \ 64 r.y = fnc(v.y); \ 65 r.z = fnc(v.z); \ 66 r.w = fnc(v.w); \ 67 return r; \ 68 } 69 70 #define FN_FUNC_FN_FN(fnc) \ 71 extern float2 __attribute__((overloadable)) fnc(float2 v1, float2 v2) { \ 72 float2 r; \ 73 r.x = fnc(v1.x, v2.x); \ 74 r.y = fnc(v1.y, v2.y); \ 75 return r; \ 76 } \ 77 extern float3 __attribute__((overloadable)) fnc(float3 v1, float3 v2) { \ 78 float3 r; \ 79 r.x = fnc(v1.x, v2.x); \ 80 r.y = fnc(v1.y, v2.y); \ 81 r.z = fnc(v1.z, v2.z); \ 82 return r; \ 83 } \ 84 extern float4 __attribute__((overloadable)) fnc(float4 v1, float4 v2) { \ 85 float4 r; \ 86 r.x = fnc(v1.x, v2.x); \ 87 r.y = fnc(v1.y, v2.y); \ 88 r.z = fnc(v1.z, v2.z); \ 89 r.w = fnc(v1.w, v2.w); \ 90 return r; \ 91 } 92 93 #define FN_FUNC_FN_F(fnc) \ 94 extern float2 __attribute__((overloadable)) fnc(float2 v1, float v2) { \ 95 float2 r; \ 96 r.x = fnc(v1.x, v2); \ 97 r.y = fnc(v1.y, v2); \ 98 return r; \ 99 } \ 100 extern float3 __attribute__((overloadable)) fnc(float3 v1, float v2) { \ 101 float3 r; \ 102 r.x = fnc(v1.x, v2); \ 103 r.y = fnc(v1.y, v2); \ 104 r.z = fnc(v1.z, v2); \ 105 return r; \ 106 } \ 107 extern float4 __attribute__((overloadable)) fnc(float4 v1, float v2) { \ 108 float4 r; \ 109 r.x = fnc(v1.x, v2); \ 110 r.y = fnc(v1.y, v2); \ 111 r.z = fnc(v1.z, v2); \ 112 r.w = fnc(v1.w, v2); \ 113 return r; \ 114 } 115 116 #define FN_FUNC_FN_IN(fnc) \ 117 extern float2 __attribute__((overloadable)) fnc(float2 v1, int2 v2) { \ 118 float2 r; \ 119 r.x = fnc(v1.x, v2.x); \ 120 r.y = fnc(v1.y, v2.y); \ 121 return r; \ 122 } \ 123 extern float3 __attribute__((overloadable)) fnc(float3 v1, int3 v2) { \ 124 float3 r; \ 125 r.x = fnc(v1.x, v2.x); \ 126 r.y = fnc(v1.y, v2.y); \ 127 r.z = fnc(v1.z, v2.z); \ 128 return r; \ 129 } \ 130 extern float4 __attribute__((overloadable)) fnc(float4 v1, int4 v2) { \ 131 float4 r; \ 132 r.x = fnc(v1.x, v2.x); \ 133 r.y = fnc(v1.y, v2.y); \ 134 r.z = fnc(v1.z, v2.z); \ 135 r.w = fnc(v1.w, v2.w); \ 136 return r; \ 137 } 138 139 #define FN_FUNC_FN_I(fnc) \ 140 extern float2 __attribute__((overloadable)) fnc(float2 v1, int v2) { \ 141 float2 r; \ 142 r.x = fnc(v1.x, v2); \ 143 r.y = fnc(v1.y, v2); \ 144 return r; \ 145 } \ 146 extern float3 __attribute__((overloadable)) fnc(float3 v1, int v2) { \ 147 float3 r; \ 148 r.x = fnc(v1.x, v2); \ 149 r.y = fnc(v1.y, v2); \ 150 r.z = fnc(v1.z, v2); \ 151 return r; \ 152 } \ 153 extern float4 __attribute__((overloadable)) fnc(float4 v1, int v2) { \ 154 float4 r; \ 155 r.x = fnc(v1.x, v2); \ 156 r.y = fnc(v1.y, v2); \ 157 r.z = fnc(v1.z, v2); \ 158 r.w = fnc(v1.w, v2); \ 159 return r; \ 160 } 161 162 #define FN_FUNC_FN_PFN(fnc) \ 163 extern float2 __attribute__((overloadable)) \ 164 fnc(float2 v1, float2 *v2) { \ 165 float2 r; \ 166 float t[2]; \ 167 r.x = fnc(v1.x, &t[0]); \ 168 r.y = fnc(v1.y, &t[1]); \ 169 v2->x = t[0]; \ 170 v2->y = t[1]; \ 171 return r; \ 172 } \ 173 extern float3 __attribute__((overloadable)) \ 174 fnc(float3 v1, float3 *v2) { \ 175 float3 r; \ 176 float t[3]; \ 177 r.x = fnc(v1.x, &t[0]); \ 178 r.y = fnc(v1.y, &t[1]); \ 179 r.z = fnc(v1.z, &t[2]); \ 180 v2->x = t[0]; \ 181 v2->y = t[1]; \ 182 v2->z = t[2]; \ 183 return r; \ 184 } \ 185 extern float4 __attribute__((overloadable)) \ 186 fnc(float4 v1, float4 *v2) { \ 187 float4 r; \ 188 float t[4]; \ 189 r.x = fnc(v1.x, &t[0]); \ 190 r.y = fnc(v1.y, &t[1]); \ 191 r.z = fnc(v1.z, &t[2]); \ 192 r.w = fnc(v1.w, &t[3]); \ 193 v2->x = t[0]; \ 194 v2->y = t[1]; \ 195 v2->z = t[2]; \ 196 v2->w = t[3]; \ 197 return r; \ 198 } 199 200 #define FN_FUNC_FN_PIN(fnc) \ 201 extern float2 __attribute__((overloadable)) fnc(float2 v1, int2 *v2) { \ 202 float2 r; \ 203 int t[2]; \ 204 r.x = fnc(v1.x, &t[0]); \ 205 r.y = fnc(v1.y, &t[1]); \ 206 v2->x = t[0]; \ 207 v2->y = t[1]; \ 208 return r; \ 209 } \ 210 extern float3 __attribute__((overloadable)) fnc(float3 v1, int3 *v2) { \ 211 float3 r; \ 212 int t[3]; \ 213 r.x = fnc(v1.x, &t[0]); \ 214 r.y = fnc(v1.y, &t[1]); \ 215 r.z = fnc(v1.z, &t[2]); \ 216 v2->x = t[0]; \ 217 v2->y = t[1]; \ 218 v2->z = t[2]; \ 219 return r; \ 220 } \ 221 extern float4 __attribute__((overloadable)) fnc(float4 v1, int4 *v2) { \ 222 float4 r; \ 223 int t[4]; \ 224 r.x = fnc(v1.x, &t[0]); \ 225 r.y = fnc(v1.y, &t[1]); \ 226 r.z = fnc(v1.z, &t[2]); \ 227 r.w = fnc(v1.w, &t[3]); \ 228 v2->x = t[0]; \ 229 v2->y = t[1]; \ 230 v2->z = t[2]; \ 231 v2->w = t[3]; \ 232 return r; \ 233 } 234 235 #define FN_FUNC_FN_FN_FN(fnc) \ 236 extern float2 __attribute__((overloadable)) \ 237 fnc(float2 v1, float2 v2, float2 v3) { \ 238 float2 r; \ 239 r.x = fnc(v1.x, v2.x, v3.x); \ 240 r.y = fnc(v1.y, v2.y, v3.y); \ 241 return r; \ 242 } \ 243 extern float3 __attribute__((overloadable)) \ 244 fnc(float3 v1, float3 v2, float3 v3) { \ 245 float3 r; \ 246 r.x = fnc(v1.x, v2.x, v3.x); \ 247 r.y = fnc(v1.y, v2.y, v3.y); \ 248 r.z = fnc(v1.z, v2.z, v3.z); \ 249 return r; \ 250 } \ 251 extern float4 __attribute__((overloadable)) \ 252 fnc(float4 v1, float4 v2, float4 v3) { \ 253 float4 r; \ 254 r.x = fnc(v1.x, v2.x, v3.x); \ 255 r.y = fnc(v1.y, v2.y, v3.y); \ 256 r.z = fnc(v1.z, v2.z, v3.z); \ 257 r.w = fnc(v1.w, v2.w, v3.w); \ 258 return r; \ 259 } 260 261 #define FN_FUNC_FN_FN_PIN(fnc) \ 262 extern float2 __attribute__((overloadable)) \ 263 fnc(float2 v1, float2 v2, int2 *v3) { \ 264 float2 r; \ 265 int t[2]; \ 266 r.x = fnc(v1.x, v2.x, &t[0]); \ 267 r.y = fnc(v1.y, v2.y, &t[1]); \ 268 v3->x = t[0]; \ 269 v3->y = t[1]; \ 270 return r; \ 271 } \ 272 extern float3 __attribute__((overloadable)) \ 273 fnc(float3 v1, float3 v2, int3 *v3) { \ 274 float3 r; \ 275 int t[3]; \ 276 r.x = fnc(v1.x, v2.x, &t[0]); \ 277 r.y = fnc(v1.y, v2.y, &t[1]); \ 278 r.z = fnc(v1.z, v2.z, &t[2]); \ 279 v3->x = t[0]; \ 280 v3->y = t[1]; \ 281 v3->z = t[2]; \ 282 return r; \ 283 } \ 284 extern float4 __attribute__((overloadable)) \ 285 fnc(float4 v1, float4 v2, int4 *v3) { \ 286 float4 r; \ 287 int t[4]; \ 288 r.x = fnc(v1.x, v2.x, &t[0]); \ 289 r.y = fnc(v1.y, v2.y, &t[1]); \ 290 r.z = fnc(v1.z, v2.z, &t[2]); \ 291 r.w = fnc(v1.w, v2.w, &t[3]); \ 292 v3->x = t[0]; \ 293 v3->y = t[1]; \ 294 v3->z = t[2]; \ 295 v3->w = t[3]; \ 296 return r; \ 297 } 298 299 static const int iposinf = 0x7f800000; 300 static const int ineginf = 0xff800000; 301 302 static const float posinf() { 303 float f = *((float*)&iposinf); 304 return f; 305 } 306 307 static const float neginf() { 308 float f = *((float*)&ineginf); 309 return f; 310 } 311 312 static bool isinf(float f) { 313 int i = *((int*)(void*)&f); 314 return (i == iposinf) || (i == ineginf); 315 } 316 317 static bool isnan(float f) { 318 int i = *((int*)(void*)&f); 319 return (((i & 0x7f800000) == 0x7f800000) && (i & 0x007fffff)); 320 } 321 322 static bool isposzero(float f) { 323 int i = *((int*)(void*)&f); 324 return (i == 0x00000000); 325 } 326 327 static bool isnegzero(float f) { 328 int i = *((int*)(void*)&f); 329 return (i == 0x80000000); 330 } 331 332 static bool iszero(float f) { 333 return isposzero(f) || isnegzero(f); 334 } 335 336 337 extern float __attribute__((overloadable)) acos(float); 338 FN_FUNC_FN(acos) 339 340 extern float __attribute__((overloadable)) acosh(float); 341 FN_FUNC_FN(acosh) 342 343 344 extern float __attribute__((overloadable)) acospi(float v) { 345 return acos(v) / M_PI; 346 } 347 FN_FUNC_FN(acospi) 348 349 extern float __attribute__((overloadable)) asin(float); 350 FN_FUNC_FN(asin) 351 352 extern float __attribute__((overloadable)) asinh(float); 353 FN_FUNC_FN(asinh) 354 355 extern float __attribute__((overloadable)) asinpi(float v) { 356 return asin(v) / M_PI; 357 } 358 FN_FUNC_FN(asinpi) 359 360 extern float __attribute__((overloadable)) atan(float); 361 FN_FUNC_FN(atan) 362 363 extern float __attribute__((overloadable)) atan2(float, float); 364 FN_FUNC_FN_FN(atan2) 365 366 extern float __attribute__((overloadable)) atanh(float); 367 FN_FUNC_FN(atanh) 368 369 extern float __attribute__((overloadable)) atanpi(float v) { 370 return atan(v) / M_PI; 371 } 372 FN_FUNC_FN(atanpi) 373 374 375 extern float __attribute__((overloadable)) atan2pi(float y, float x) { 376 return atan2(y, x) / M_PI; 377 } 378 FN_FUNC_FN_FN(atan2pi) 379 380 extern float __attribute__((overloadable)) cbrt(float); 381 FN_FUNC_FN(cbrt) 382 383 extern float __attribute__((overloadable)) ceil(float); 384 FN_FUNC_FN(ceil) 385 386 extern float __attribute__((overloadable)) copysign(float, float); 387 FN_FUNC_FN_FN(copysign) 388 389 extern float __attribute__((overloadable)) cos(float); 390 FN_FUNC_FN(cos) 391 392 extern float __attribute__((overloadable)) cosh(float); 393 FN_FUNC_FN(cosh) 394 395 extern float __attribute__((overloadable)) cospi(float v) { 396 return cos(v * M_PI); 397 } 398 FN_FUNC_FN(cospi) 399 400 extern float __attribute__((overloadable)) erfc(float); 401 FN_FUNC_FN(erfc) 402 403 extern float __attribute__((overloadable)) erf(float); 404 FN_FUNC_FN(erf) 405 406 extern float __attribute__((overloadable)) exp(float); 407 FN_FUNC_FN(exp) 408 409 extern float __attribute__((overloadable)) exp2(float); 410 FN_FUNC_FN(exp2) 411 412 extern float __attribute__((overloadable)) pow(float, float); 413 414 extern float __attribute__((overloadable)) exp10(float v) { 415 return exp2(v * 3.321928095f); 416 } 417 FN_FUNC_FN(exp10) 418 419 extern float __attribute__((overloadable)) expm1(float); 420 FN_FUNC_FN(expm1) 421 422 extern float __attribute__((overloadable)) fabs(float v) { 423 int i = *((int*)(void*)&v) & 0x7fffffff; 424 return *((float*)(void*)&i); 425 } 426 FN_FUNC_FN(fabs) 427 428 extern float __attribute__((overloadable)) fdim(float, float); 429 FN_FUNC_FN_FN(fdim) 430 431 extern float __attribute__((overloadable)) floor(float); 432 FN_FUNC_FN(floor) 433 434 extern float __attribute__((overloadable)) fma(float, float, float); 435 FN_FUNC_FN_FN_FN(fma) 436 437 extern float __attribute__((overloadable)) fmin(float, float); 438 439 extern float __attribute__((overloadable)) fmod(float, float); 440 FN_FUNC_FN_FN(fmod) 441 442 extern float __attribute__((overloadable)) fract(float v, float *iptr) { 443 int i = (int)floor(v); 444 if (iptr) { 445 iptr[0] = i; 446 } 447 return fmin(v - i, 0x1.fffffep-1f); 448 } 449 FN_FUNC_FN_PFN(fract) 450 451 extern float __attribute__((overloadable)) frexp(float, int *); 452 FN_FUNC_FN_PIN(frexp) 453 454 extern float __attribute__((overloadable)) hypot(float, float); 455 FN_FUNC_FN_FN(hypot) 456 457 extern int __attribute__((overloadable)) ilogb(float); 458 IN_FUNC_FN(ilogb) 459 460 extern float __attribute__((overloadable)) ldexp(float, int); 461 FN_FUNC_FN_IN(ldexp) 462 FN_FUNC_FN_I(ldexp) 463 464 extern float __attribute__((overloadable)) lgamma(float); 465 FN_FUNC_FN(lgamma) 466 extern float __attribute__((overloadable)) lgamma(float, int*); 467 FN_FUNC_FN_PIN(lgamma) 468 469 extern float __attribute__((overloadable)) log(float); 470 FN_FUNC_FN(log) 471 472 extern float __attribute__((overloadable)) log10(float); 473 FN_FUNC_FN(log10) 474 475 476 extern float __attribute__((overloadable)) log2(float v) { 477 return log10(v) * 3.321928095f; 478 } 479 FN_FUNC_FN(log2) 480 481 extern float __attribute__((overloadable)) log1p(float); 482 FN_FUNC_FN(log1p) 483 484 extern float __attribute__((overloadable)) logb(float); 485 FN_FUNC_FN(logb) 486 487 extern float __attribute__((overloadable)) mad(float a, float b, float c) { 488 return a * b + c; 489 } 490 extern float2 __attribute__((overloadable)) mad(float2 a, float2 b, float2 c) { 491 return a * b + c; 492 } 493 extern float3 __attribute__((overloadable)) mad(float3 a, float3 b, float3 c) { 494 return a * b + c; 495 } 496 extern float4 __attribute__((overloadable)) mad(float4 a, float4 b, float4 c) { 497 return a * b + c; 498 } 499 500 extern float __attribute__((overloadable)) modf(float, float *); 501 FN_FUNC_FN_PFN(modf); 502 503 extern float __attribute__((overloadable)) nan(uint v) { 504 float f[1]; 505 uint32_t *ip = (uint32_t *)f; 506 *ip = v | 0x7fc00000; 507 return f[0]; 508 } 509 510 extern float __attribute__((overloadable)) nextafter(float, float); 511 FN_FUNC_FN_FN(nextafter) 512 513 FN_FUNC_FN_FN(pow) 514 515 extern float __attribute__((overloadable)) pown(float v, int p) { 516 return pow(v, (float)p); 517 } 518 extern float2 __attribute__((overloadable)) pown(float2 v, int2 p) { 519 float2 f2 = convert_float2(p); 520 return pow(v, f2); 521 } 522 extern float3 __attribute__((overloadable)) pown(float3 v, int3 p) { 523 float3 f3 = convert_float3(p); 524 return pow(v, f3); 525 } 526 extern float4 __attribute__((overloadable)) pown(float4 v, int4 p) { 527 float4 f4 = convert_float4(p); 528 return pow(v, f4); 529 } 530 531 extern float __attribute__((overloadable)) powr(float v, float p) { 532 return pow(v, p); 533 } 534 extern float2 __attribute__((overloadable)) powr(float2 v, float2 p) { 535 return pow(v, p); 536 } 537 extern float3 __attribute__((overloadable)) powr(float3 v, float3 p) { 538 return pow(v, p); 539 } 540 extern float4 __attribute__((overloadable)) powr(float4 v, float4 p) { 541 return pow(v, p); 542 } 543 544 extern float __attribute__((overloadable)) remainder(float, float); 545 FN_FUNC_FN_FN(remainder) 546 547 extern float __attribute__((overloadable)) remquo(float, float, int *); 548 FN_FUNC_FN_FN_PIN(remquo) 549 550 extern float __attribute__((overloadable)) rint(float); 551 FN_FUNC_FN(rint) 552 553 extern float __attribute__((overloadable)) rootn(float v, int r) { 554 if (r == 0) { 555 return nan(0); 556 } 557 558 if (iszero(v)) { 559 if (r < 0) { 560 if (r & 1) { 561 return copysign(posinf(), v); 562 } else { 563 return posinf(); 564 } 565 } else { 566 if (r & 1) { 567 return copysign(0.f, v); 568 } else { 569 return 0.f; 570 } 571 } 572 } 573 574 if (!isinf(v) && !isnan(v) && (v < 0.f)) { 575 if (r & 1) { 576 return (-1.f * pow(-1.f * v, 1.f / r)); 577 } else { 578 return nan(0); 579 } 580 } 581 582 return pow(v, 1.f / r); 583 } 584 FN_FUNC_FN_IN(rootn); 585 586 extern float __attribute__((overloadable)) round(float); 587 FN_FUNC_FN(round) 588 589 590 extern float __attribute__((overloadable)) sqrt(float); 591 extern float __attribute__((overloadable)) rsqrt(float v) { 592 return 1.f / sqrt(v); 593 } 594 595 #if !defined(ARCH_X86_HAVE_SSE2) && !defined(ARCH_X86_HAVE_SSE3) 596 FN_FUNC_FN(sqrt) 597 #endif // !defined(ARCH_X86_HAVE_SSE2) && !defined(ARCH_X86_HAVE_SSE3) 598 599 FN_FUNC_FN(rsqrt) 600 601 extern float __attribute__((overloadable)) sin(float); 602 FN_FUNC_FN(sin) 603 604 extern float __attribute__((overloadable)) sincos(float v, float *cosptr) { 605 *cosptr = cos(v); 606 return sin(v); 607 } 608 extern float2 __attribute__((overloadable)) sincos(float2 v, float2 *cosptr) { 609 *cosptr = cos(v); 610 return sin(v); 611 } 612 extern float3 __attribute__((overloadable)) sincos(float3 v, float3 *cosptr) { 613 *cosptr = cos(v); 614 return sin(v); 615 } 616 extern float4 __attribute__((overloadable)) sincos(float4 v, float4 *cosptr) { 617 *cosptr = cos(v); 618 return sin(v); 619 } 620 621 extern float __attribute__((overloadable)) sinh(float); 622 FN_FUNC_FN(sinh) 623 624 extern float __attribute__((overloadable)) sinpi(float v) { 625 return sin(v * M_PI); 626 } 627 FN_FUNC_FN(sinpi) 628 629 extern float __attribute__((overloadable)) tan(float); 630 FN_FUNC_FN(tan) 631 632 extern float __attribute__((overloadable)) tanh(float); 633 FN_FUNC_FN(tanh) 634 635 extern float __attribute__((overloadable)) tanpi(float v) { 636 return tan(v * M_PI); 637 } 638 FN_FUNC_FN(tanpi) 639 640 641 extern float __attribute__((overloadable)) tgamma(float); 642 FN_FUNC_FN(tgamma) 643 644 extern float __attribute__((overloadable)) trunc(float); 645 FN_FUNC_FN(trunc) 646 647 // Int ops (partial), 6.11.3 648 649 #define XN_FUNC_YN(typeout, fnc, typein) \ 650 extern typeout __attribute__((overloadable)) fnc(typein); \ 651 extern typeout##2 __attribute__((overloadable)) fnc(typein##2 v) { \ 652 typeout##2 r; \ 653 r.x = fnc(v.x); \ 654 r.y = fnc(v.y); \ 655 return r; \ 656 } \ 657 extern typeout##3 __attribute__((overloadable)) fnc(typein##3 v) { \ 658 typeout##3 r; \ 659 r.x = fnc(v.x); \ 660 r.y = fnc(v.y); \ 661 r.z = fnc(v.z); \ 662 return r; \ 663 } \ 664 extern typeout##4 __attribute__((overloadable)) fnc(typein##4 v) { \ 665 typeout##4 r; \ 666 r.x = fnc(v.x); \ 667 r.y = fnc(v.y); \ 668 r.z = fnc(v.z); \ 669 r.w = fnc(v.w); \ 670 return r; \ 671 } 672 673 674 #define UIN_FUNC_IN(fnc) \ 675 XN_FUNC_YN(uchar, fnc, char) \ 676 XN_FUNC_YN(ushort, fnc, short) \ 677 XN_FUNC_YN(uint, fnc, int) 678 679 #define IN_FUNC_IN(fnc) \ 680 XN_FUNC_YN(uchar, fnc, uchar) \ 681 XN_FUNC_YN(char, fnc, char) \ 682 XN_FUNC_YN(ushort, fnc, ushort) \ 683 XN_FUNC_YN(short, fnc, short) \ 684 XN_FUNC_YN(uint, fnc, uint) \ 685 XN_FUNC_YN(int, fnc, int) 686 687 688 #define XN_FUNC_XN_XN_BODY(type, fnc, body) \ 689 extern type __attribute__((overloadable)) \ 690 fnc(type v1, type v2) { \ 691 return body; \ 692 } \ 693 extern type##2 __attribute__((overloadable)) \ 694 fnc(type##2 v1, type##2 v2) { \ 695 type##2 r; \ 696 r.x = fnc(v1.x, v2.x); \ 697 r.y = fnc(v1.y, v2.y); \ 698 return r; \ 699 } \ 700 extern type##3 __attribute__((overloadable)) \ 701 fnc(type##3 v1, type##3 v2) { \ 702 type##3 r; \ 703 r.x = fnc(v1.x, v2.x); \ 704 r.y = fnc(v1.y, v2.y); \ 705 r.z = fnc(v1.z, v2.z); \ 706 return r; \ 707 } \ 708 extern type##4 __attribute__((overloadable)) \ 709 fnc(type##4 v1, type##4 v2) { \ 710 type##4 r; \ 711 r.x = fnc(v1.x, v2.x); \ 712 r.y = fnc(v1.y, v2.y); \ 713 r.z = fnc(v1.z, v2.z); \ 714 r.w = fnc(v1.w, v2.w); \ 715 return r; \ 716 } 717 718 #define IN_FUNC_IN_IN_BODY(fnc, body) \ 719 XN_FUNC_XN_XN_BODY(uchar, fnc, body) \ 720 XN_FUNC_XN_XN_BODY(char, fnc, body) \ 721 XN_FUNC_XN_XN_BODY(ushort, fnc, body) \ 722 XN_FUNC_XN_XN_BODY(short, fnc, body) \ 723 XN_FUNC_XN_XN_BODY(uint, fnc, body) \ 724 XN_FUNC_XN_XN_BODY(int, fnc, body) \ 725 XN_FUNC_XN_XN_BODY(float, fnc, body) 726 727 728 /** 729 * abs 730 */ 731 extern uint32_t __attribute__((overloadable)) abs(int32_t v) { 732 if (v < 0) 733 return -v; 734 return v; 735 } 736 extern uint16_t __attribute__((overloadable)) abs(int16_t v) { 737 if (v < 0) 738 return -v; 739 return v; 740 } 741 extern uint8_t __attribute__((overloadable)) abs(int8_t v) { 742 if (v < 0) 743 return -v; 744 return v; 745 } 746 747 /** 748 * clz 749 */ 750 extern uint32_t __attribute__((overloadable)) clz(uint32_t v) { 751 return __builtin_clz(v); 752 } 753 extern uint16_t __attribute__((overloadable)) clz(uint16_t v) { 754 return (uint16_t)__builtin_clz(v); 755 } 756 extern uint8_t __attribute__((overloadable)) clz(uint8_t v) { 757 return (uint8_t)__builtin_clz(v); 758 } 759 extern int32_t __attribute__((overloadable)) clz(int32_t v) { 760 return (int32_t)__builtin_clz((uint32_t)v); 761 } 762 extern int16_t __attribute__((overloadable)) clz(int16_t v) { 763 return (int16_t)__builtin_clz(v); 764 } 765 extern int8_t __attribute__((overloadable)) clz(int8_t v) { 766 return (int8_t)__builtin_clz(v); 767 } 768 769 770 UIN_FUNC_IN(abs) 771 IN_FUNC_IN(clz) 772 773 774 // 6.11.4 775 776 777 extern float __attribute__((overloadable)) degrees(float radians) { 778 return radians * (180.f / M_PI); 779 } 780 extern float2 __attribute__((overloadable)) degrees(float2 radians) { 781 return radians * (180.f / M_PI); 782 } 783 extern float3 __attribute__((overloadable)) degrees(float3 radians) { 784 return radians * (180.f / M_PI); 785 } 786 extern float4 __attribute__((overloadable)) degrees(float4 radians) { 787 return radians * (180.f / M_PI); 788 } 789 790 extern float __attribute__((overloadable)) mix(float start, float stop, float amount) { 791 return start + (stop - start) * amount; 792 } 793 extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float2 amount) { 794 return start + (stop - start) * amount; 795 } 796 extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float3 amount) { 797 return start + (stop - start) * amount; 798 } 799 extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float4 amount) { 800 return start + (stop - start) * amount; 801 } 802 extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float amount) { 803 return start + (stop - start) * amount; 804 } 805 extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float amount) { 806 return start + (stop - start) * amount; 807 } 808 extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float amount) { 809 return start + (stop - start) * amount; 810 } 811 812 extern float __attribute__((overloadable)) radians(float degrees) { 813 return degrees * (M_PI / 180.f); 814 } 815 extern float2 __attribute__((overloadable)) radians(float2 degrees) { 816 return degrees * (M_PI / 180.f); 817 } 818 extern float3 __attribute__((overloadable)) radians(float3 degrees) { 819 return degrees * (M_PI / 180.f); 820 } 821 extern float4 __attribute__((overloadable)) radians(float4 degrees) { 822 return degrees * (M_PI / 180.f); 823 } 824 825 extern float __attribute__((overloadable)) step(float edge, float v) { 826 return (v < edge) ? 0.f : 1.f; 827 } 828 extern float2 __attribute__((overloadable)) step(float2 edge, float2 v) { 829 float2 r; 830 r.x = (v.x < edge.x) ? 0.f : 1.f; 831 r.y = (v.y < edge.y) ? 0.f : 1.f; 832 return r; 833 } 834 extern float3 __attribute__((overloadable)) step(float3 edge, float3 v) { 835 float3 r; 836 r.x = (v.x < edge.x) ? 0.f : 1.f; 837 r.y = (v.y < edge.y) ? 0.f : 1.f; 838 r.z = (v.z < edge.z) ? 0.f : 1.f; 839 return r; 840 } 841 extern float4 __attribute__((overloadable)) step(float4 edge, float4 v) { 842 float4 r; 843 r.x = (v.x < edge.x) ? 0.f : 1.f; 844 r.y = (v.y < edge.y) ? 0.f : 1.f; 845 r.z = (v.z < edge.z) ? 0.f : 1.f; 846 r.w = (v.w < edge.w) ? 0.f : 1.f; 847 return r; 848 } 849 extern float2 __attribute__((overloadable)) step(float2 edge, float v) { 850 float2 r; 851 r.x = (v < edge.x) ? 0.f : 1.f; 852 r.y = (v < edge.y) ? 0.f : 1.f; 853 return r; 854 } 855 extern float3 __attribute__((overloadable)) step(float3 edge, float v) { 856 float3 r; 857 r.x = (v < edge.x) ? 0.f : 1.f; 858 r.y = (v < edge.y) ? 0.f : 1.f; 859 r.z = (v < edge.z) ? 0.f : 1.f; 860 return r; 861 } 862 extern float4 __attribute__((overloadable)) step(float4 edge, float v) { 863 float4 r; 864 r.x = (v < edge.x) ? 0.f : 1.f; 865 r.y = (v < edge.y) ? 0.f : 1.f; 866 r.z = (v < edge.z) ? 0.f : 1.f; 867 r.w = (v < edge.w) ? 0.f : 1.f; 868 return r; 869 } 870 871 extern float __attribute__((overloadable)) smoothstep(float, float, float); 872 extern float2 __attribute__((overloadable)) smoothstep(float2, float2, float2); 873 extern float3 __attribute__((overloadable)) smoothstep(float3, float3, float3); 874 extern float4 __attribute__((overloadable)) smoothstep(float4, float4, float4); 875 extern float2 __attribute__((overloadable)) smoothstep(float, float, float2); 876 extern float3 __attribute__((overloadable)) smoothstep(float, float, float3); 877 extern float4 __attribute__((overloadable)) smoothstep(float, float, float4); 878 879 extern float __attribute__((overloadable)) sign(float v) { 880 if (v > 0) return 1.f; 881 if (v < 0) return -1.f; 882 return v; 883 } 884 FN_FUNC_FN(sign) 885 886 887 // 6.11.5 888 extern float3 __attribute__((overloadable)) cross(float3 lhs, float3 rhs) { 889 float3 r; 890 r.x = lhs.y * rhs.z - lhs.z * rhs.y; 891 r.y = lhs.z * rhs.x - lhs.x * rhs.z; 892 r.z = lhs.x * rhs.y - lhs.y * rhs.x; 893 return r; 894 } 895 896 extern float4 __attribute__((overloadable)) cross(float4 lhs, float4 rhs) { 897 float4 r; 898 r.x = lhs.y * rhs.z - lhs.z * rhs.y; 899 r.y = lhs.z * rhs.x - lhs.x * rhs.z; 900 r.z = lhs.x * rhs.y - lhs.y * rhs.x; 901 r.w = 0.f; 902 return r; 903 } 904 905 #if !defined(ARCH_X86_HAVE_SSE3) 906 907 extern float __attribute__((overloadable)) dot(float lhs, float rhs) { 908 return lhs * rhs; 909 } 910 extern float __attribute__((overloadable)) dot(float2 lhs, float2 rhs) { 911 return lhs.x*rhs.x + lhs.y*rhs.y; 912 } 913 extern float __attribute__((overloadable)) dot(float3 lhs, float3 rhs) { 914 return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z; 915 } 916 extern float __attribute__((overloadable)) dot(float4 lhs, float4 rhs) { 917 return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z + lhs.w*rhs.w; 918 } 919 920 extern float __attribute__((overloadable)) length(float v) { 921 return fabs(v); 922 } 923 extern float __attribute__((overloadable)) length(float2 v) { 924 return sqrt(v.x*v.x + v.y*v.y); 925 } 926 extern float __attribute__((overloadable)) length(float3 v) { 927 return sqrt(v.x*v.x + v.y*v.y + v.z*v.z); 928 } 929 extern float __attribute__((overloadable)) length(float4 v) { 930 return sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w); 931 } 932 933 #else 934 935 extern float __attribute__((overloadable)) length(float v); 936 extern float __attribute__((overloadable)) length(float2 v); 937 extern float __attribute__((overloadable)) length(float3 v); 938 extern float __attribute__((overloadable)) length(float4 v); 939 940 #endif 941 942 extern float __attribute__((overloadable)) distance(float lhs, float rhs) { 943 return length(lhs - rhs); 944 } 945 extern float __attribute__((overloadable)) distance(float2 lhs, float2 rhs) { 946 return length(lhs - rhs); 947 } 948 extern float __attribute__((overloadable)) distance(float3 lhs, float3 rhs) { 949 return length(lhs - rhs); 950 } 951 extern float __attribute__((overloadable)) distance(float4 lhs, float4 rhs) { 952 return length(lhs - rhs); 953 } 954 955 extern float __attribute__((overloadable)) normalize(float v) { 956 return 1.f; 957 } 958 extern float2 __attribute__((overloadable)) normalize(float2 v) { 959 return v / length(v); 960 } 961 extern float3 __attribute__((overloadable)) normalize(float3 v) { 962 return v / length(v); 963 } 964 extern float4 __attribute__((overloadable)) normalize(float4 v) { 965 return v / length(v); 966 } 967 968 extern float __attribute__((overloadable)) half_sqrt(float); 969 970 extern float __attribute__((overloadable)) fast_length(float v) { 971 return fabs(v); 972 } 973 extern float __attribute__((overloadable)) fast_length(float2 v) { 974 return half_sqrt(v.x*v.x + v.y*v.y); 975 } 976 extern float __attribute__((overloadable)) fast_length(float3 v) { 977 return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z); 978 } 979 extern float __attribute__((overloadable)) fast_length(float4 v) { 980 return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w); 981 } 982 983 extern float __attribute__((overloadable)) fast_distance(float lhs, float rhs) { 984 return fast_length(lhs - rhs); 985 } 986 extern float __attribute__((overloadable)) fast_distance(float2 lhs, float2 rhs) { 987 return fast_length(lhs - rhs); 988 } 989 extern float __attribute__((overloadable)) fast_distance(float3 lhs, float3 rhs) { 990 return fast_length(lhs - rhs); 991 } 992 extern float __attribute__((overloadable)) fast_distance(float4 lhs, float4 rhs) { 993 return fast_length(lhs - rhs); 994 } 995 996 extern float __attribute__((overloadable)) half_rsqrt(float); 997 998 extern float __attribute__((overloadable)) fast_normalize(float v) { 999 return 1.f; 1000 } 1001 extern float2 __attribute__((overloadable)) fast_normalize(float2 v) { 1002 return v * half_rsqrt(v.x*v.x + v.y*v.y); 1003 } 1004 extern float3 __attribute__((overloadable)) fast_normalize(float3 v) { 1005 return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z); 1006 } 1007 extern float4 __attribute__((overloadable)) fast_normalize(float4 v) { 1008 return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w); 1009 } 1010 1011 extern float __attribute__((overloadable)) half_recip(float); 1012 1013 /* 1014 extern float __attribute__((overloadable)) approx_atan(float x) { 1015 if (x == 0.f) 1016 return 0.f; 1017 if (x < 0.f) 1018 return -1.f * approx_atan(-1.f * x); 1019 if (x > 1.f) 1020 return M_PI_2 - approx_atan(approx_recip(x)); 1021 return x * approx_recip(1.f + 0.28f * x*x); 1022 } 1023 FN_FUNC_FN(approx_atan) 1024 */ 1025 1026 typedef union 1027 { 1028 float fv; 1029 int32_t iv; 1030 } ieee_float_shape_type; 1031 1032 /* Get a 32 bit int from a float. */ 1033 1034 #define GET_FLOAT_WORD(i,d) \ 1035 do { \ 1036 ieee_float_shape_type gf_u; \ 1037 gf_u.fv = (d); \ 1038 (i) = gf_u.iv; \ 1039 } while (0) 1040 1041 /* Set a float from a 32 bit int. */ 1042 1043 #define SET_FLOAT_WORD(d,i) \ 1044 do { \ 1045 ieee_float_shape_type sf_u; \ 1046 sf_u.iv = (i); \ 1047 (d) = sf_u.fv; \ 1048 } while (0) 1049 1050 1051 1052 // Valid -125 to 125 1053 extern float __attribute__((overloadable)) native_exp2(float v) { 1054 int32_t iv = (int)v; 1055 int32_t x = iv + (iv >> 31); // ~floor(v) 1056 float r = (v - x); 1057 1058 float fo; 1059 SET_FLOAT_WORD(fo, (x + 127) << 23); 1060 1061 r *= 0.694f; // ~ log(e) / log(2) 1062 float r2 = r*r; 1063 float adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f); 1064 return fo * adj; 1065 } 1066 1067 extern float2 __attribute__((overloadable)) native_exp2(float2 v) { 1068 int2 iv = convert_int2(v); 1069 int2 x = iv + (iv >> (int2)31);//floor(v); 1070 float2 r = (v - convert_float2(x)); 1071 1072 x += 127; 1073 1074 float2 fo = (float2)(x << (int2)23); 1075 1076 r *= 0.694f; // ~ log(e) / log(2) 1077 float2 r2 = r*r; 1078 float2 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f); 1079 return fo * adj; 1080 } 1081 1082 extern float4 __attribute__((overloadable)) native_exp2(float4 v) { 1083 int4 iv = convert_int4(v); 1084 int4 x = iv + (iv >> (int4)31);//floor(v); 1085 float4 r = (v - convert_float4(x)); 1086 1087 x += 127; 1088 1089 float4 fo = (float4)(x << (int4)23); 1090 1091 r *= 0.694f; // ~ log(e) / log(2) 1092 float4 r2 = r*r; 1093 float4 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f); 1094 return fo * adj; 1095 } 1096 1097 extern float3 __attribute__((overloadable)) native_exp2(float3 v) { 1098 float4 t = 1.f; 1099 t.xyz = v; 1100 return native_exp2(t).xyz; 1101 } 1102 1103 1104 extern float __attribute__((overloadable)) native_exp(float v) { 1105 return native_exp2(v * 1.442695041f); 1106 } 1107 extern float2 __attribute__((overloadable)) native_exp(float2 v) { 1108 return native_exp2(v * 1.442695041f); 1109 } 1110 extern float3 __attribute__((overloadable)) native_exp(float3 v) { 1111 return native_exp2(v * 1.442695041f); 1112 } 1113 extern float4 __attribute__((overloadable)) native_exp(float4 v) { 1114 return native_exp2(v * 1.442695041f); 1115 } 1116 1117 extern float __attribute__((overloadable)) native_exp10(float v) { 1118 return native_exp2(v * 3.321928095f); 1119 } 1120 extern float2 __attribute__((overloadable)) native_exp10(float2 v) { 1121 return native_exp2(v * 3.321928095f); 1122 } 1123 extern float3 __attribute__((overloadable)) native_exp10(float3 v) { 1124 return native_exp2(v * 3.321928095f); 1125 } 1126 extern float4 __attribute__((overloadable)) native_exp10(float4 v) { 1127 return native_exp2(v * 3.321928095f); 1128 } 1129 1130 extern float __attribute__((overloadable)) native_log2(float v) { 1131 int32_t ibits; 1132 GET_FLOAT_WORD(ibits, v); 1133 1134 int32_t e = (ibits >> 23) & 0xff; 1135 1136 ibits &= 0x7fffff; 1137 ibits |= 127 << 23; 1138 1139 float ir; 1140 SET_FLOAT_WORD(ir, ibits); 1141 1142 ir -= 1.5f; 1143 float ir2 = ir*ir; 1144 float adj2 = 0.405465108f + // -0.00009f + 1145 (0.666666667f * ir) - 1146 (0.222222222f * ir2) + 1147 (0.098765432f * ir*ir2) - 1148 (0.049382716f * ir2*ir2) + 1149 (0.026337449f * ir*ir2*ir2) - 1150 (0.014631916f * ir2*ir2*ir2); 1151 adj2 *= (1.f / 0.693147181f); 1152 1153 return (float)(e - 127) + adj2; 1154 } 1155 extern float2 __attribute__((overloadable)) native_log2(float2 v) { 1156 float2 v2 = {native_log2(v.x), native_log2(v.y)}; 1157 return v2; 1158 } 1159 extern float3 __attribute__((overloadable)) native_log2(float3 v) { 1160 float3 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z)}; 1161 return v2; 1162 } 1163 extern float4 __attribute__((overloadable)) native_log2(float4 v) { 1164 float4 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z), native_log2(v.w)}; 1165 return v2; 1166 } 1167 1168 extern float __attribute__((overloadable)) native_log(float v) { 1169 return native_log2(v) * (1.f / 1.442695041f); 1170 } 1171 extern float2 __attribute__((overloadable)) native_log(float2 v) { 1172 return native_log2(v) * (1.f / 1.442695041f); 1173 } 1174 extern float3 __attribute__((overloadable)) native_log(float3 v) { 1175 return native_log2(v) * (1.f / 1.442695041f); 1176 } 1177 extern float4 __attribute__((overloadable)) native_log(float4 v) { 1178 return native_log2(v) * (1.f / 1.442695041f); 1179 } 1180 1181 extern float __attribute__((overloadable)) native_log10(float v) { 1182 return native_log2(v) * (1.f / 3.321928095f); 1183 } 1184 extern float2 __attribute__((overloadable)) native_log10(float2 v) { 1185 return native_log2(v) * (1.f / 3.321928095f); 1186 } 1187 extern float3 __attribute__((overloadable)) native_log10(float3 v) { 1188 return native_log2(v) * (1.f / 3.321928095f); 1189 } 1190 extern float4 __attribute__((overloadable)) native_log10(float4 v) { 1191 return native_log2(v) * (1.f / 3.321928095f); 1192 } 1193 1194 1195 extern float __attribute__((overloadable)) native_powr(float v, float y) { 1196 float v2 = native_log2(v); 1197 v2 = fmax(v2, -125.f); 1198 return native_exp2(v2 * y); 1199 } 1200 extern float2 __attribute__((overloadable)) native_powr(float2 v, float2 y) { 1201 float2 v2 = native_log2(v); 1202 v2 = fmax(v2, -125.f); 1203 return native_exp2(v2 * y); 1204 } 1205 extern float3 __attribute__((overloadable)) native_powr(float3 v, float3 y) { 1206 float3 v2 = native_log2(v); 1207 v2 = fmax(v2, -125.f); 1208 return native_exp2(v2 * y); 1209 } 1210 extern float4 __attribute__((overloadable)) native_powr(float4 v, float4 y) { 1211 float4 v2 = native_log2(v); 1212 v2 = fmax(v2, -125.f); 1213 return native_exp2(v2 * y); 1214 } 1215 1216 1217 #undef FN_FUNC_FN 1218 #undef IN_FUNC_FN 1219 #undef FN_FUNC_FN_FN 1220 #undef FN_FUNC_FN_F 1221 #undef FN_FUNC_FN_IN 1222 #undef FN_FUNC_FN_I 1223 #undef FN_FUNC_FN_PFN 1224 #undef FN_FUNC_FN_PIN 1225 #undef FN_FUNC_FN_FN_FN 1226 #undef FN_FUNC_FN_FN_PIN 1227 #undef XN_FUNC_YN 1228 #undef UIN_FUNC_IN 1229 #undef IN_FUNC_IN 1230 #undef XN_FUNC_XN_XN_BODY 1231 #undef IN_FUNC_IN_IN_BODY 1232
