rs_cl.c revision c117d8dad895ab9bae4ba6077365f0dfd33ece47
1#include "rs_types.rsh" 2 3extern float2 __attribute__((overloadable)) convert_float2(int2 c); 4extern float3 __attribute__((overloadable)) convert_float3(int3 c); 5extern float4 __attribute__((overloadable)) convert_float4(int4 c); 6 7extern int2 __attribute__((overloadable)) convert_int2(float2 c); 8extern int3 __attribute__((overloadable)) convert_int3(float3 c); 9extern int4 __attribute__((overloadable)) convert_int4(float4 c); 10 11 12extern float __attribute__((overloadable)) fmin(float v, float v2); 13extern float2 __attribute__((overloadable)) fmin(float2 v, float v2); 14extern float3 __attribute__((overloadable)) fmin(float3 v, float v2); 15extern float4 __attribute__((overloadable)) fmin(float4 v, float v2); 16 17extern float __attribute__((overloadable)) fmax(float v, float v2); 18extern float2 __attribute__((overloadable)) fmax(float2 v, float v2); 19extern float3 __attribute__((overloadable)) fmax(float3 v, float v2); 20extern float4 __attribute__((overloadable)) fmax(float4 v, float v2); 21 22// Float ops, 6.11.2 23 24#define FN_FUNC_FN(fnc) \ 25extern 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} \ 31extern 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} \ 38extern 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) \ 48extern 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} \ 54extern 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} \ 61extern 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) \ 71extern 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} \ 77extern 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} \ 84extern 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) \ 94extern 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} \ 100extern 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} \ 107extern 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) \ 117extern 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} \ 123extern 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} \ 130extern 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) \ 140extern 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} \ 146extern 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} \ 153extern 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) \ 163extern 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} \ 173extern 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} \ 185extern 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) \ 201extern 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} \ 210extern 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} \ 221extern 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) \ 236extern 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} \ 243extern 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} \ 251extern 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) \ 262extern 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} \ 272extern 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} \ 284extern 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 299static const int iposinf = 0x7f800000; 300static const int ineginf = 0xff800000; 301 302static const float posinf() { 303 float f = *((float*)&iposinf); 304 return f; 305} 306 307static const float neginf() { 308 float f = *((float*)&ineginf); 309 return f; 310} 311 312static bool isinf(float f) { 313 int i = *((int*)(void*)&f); 314 return (i == iposinf) || (i == ineginf); 315} 316 317static bool isnan(float f) { 318 int i = *((int*)(void*)&f); 319 return (((i & 0x7f800000) == 0x7f800000) && (i & 0x007fffff)); 320} 321 322static bool isposzero(float f) { 323 int i = *((int*)(void*)&f); 324 return (i == 0x00000000); 325} 326 327static bool isnegzero(float f) { 328 int i = *((int*)(void*)&f); 329 return (i == 0x80000000); 330} 331 332static bool iszero(float f) { 333 return isposzero(f) || isnegzero(f); 334} 335 336 337extern float __attribute__((overloadable)) acos(float); 338FN_FUNC_FN(acos) 339 340extern float __attribute__((overloadable)) acosh(float); 341FN_FUNC_FN(acosh) 342 343 344extern float __attribute__((overloadable)) acospi(float v) { 345 return acos(v) / M_PI; 346} 347FN_FUNC_FN(acospi) 348 349extern float __attribute__((overloadable)) asin(float); 350FN_FUNC_FN(asin) 351 352extern float __attribute__((overloadable)) asinh(float); 353FN_FUNC_FN(asinh) 354 355extern float __attribute__((overloadable)) asinpi(float v) { 356 return asin(v) / M_PI; 357} 358FN_FUNC_FN(asinpi) 359 360extern float __attribute__((overloadable)) atan(float); 361FN_FUNC_FN(atan) 362 363extern float __attribute__((overloadable)) atan2(float, float); 364FN_FUNC_FN_FN(atan2) 365 366extern float __attribute__((overloadable)) atanh(float); 367FN_FUNC_FN(atanh) 368 369extern float __attribute__((overloadable)) atanpi(float v) { 370 return atan(v) / M_PI; 371} 372FN_FUNC_FN(atanpi) 373 374 375extern float __attribute__((overloadable)) atan2pi(float y, float x) { 376 return atan2(y, x) / M_PI; 377} 378FN_FUNC_FN_FN(atan2pi) 379 380extern float __attribute__((overloadable)) cbrt(float); 381FN_FUNC_FN(cbrt) 382 383extern float __attribute__((overloadable)) ceil(float); 384FN_FUNC_FN(ceil) 385 386extern float __attribute__((overloadable)) copysign(float, float); 387FN_FUNC_FN_FN(copysign) 388 389extern float __attribute__((overloadable)) cos(float); 390FN_FUNC_FN(cos) 391 392extern float __attribute__((overloadable)) cosh(float); 393FN_FUNC_FN(cosh) 394 395extern float __attribute__((overloadable)) cospi(float v) { 396 return cos(v * M_PI); 397} 398FN_FUNC_FN(cospi) 399 400extern float __attribute__((overloadable)) erfc(float); 401FN_FUNC_FN(erfc) 402 403extern float __attribute__((overloadable)) erf(float); 404FN_FUNC_FN(erf) 405 406extern float __attribute__((overloadable)) exp(float); 407FN_FUNC_FN(exp) 408 409extern float __attribute__((overloadable)) exp2(float); 410FN_FUNC_FN(exp2) 411 412extern float __attribute__((overloadable)) pow(float, float); 413 414extern float __attribute__((overloadable)) exp10(float v) { 415 return exp2(v * 3.321928095f); 416} 417FN_FUNC_FN(exp10) 418 419extern float __attribute__((overloadable)) expm1(float); 420FN_FUNC_FN(expm1) 421 422extern float __attribute__((overloadable)) fabs(float v) { 423 int i = *((int*)(void*)&v) & 0x7fffffff; 424 return *((float*)(void*)&i); 425} 426FN_FUNC_FN(fabs) 427 428extern float __attribute__((overloadable)) fdim(float, float); 429FN_FUNC_FN_FN(fdim) 430 431extern float __attribute__((overloadable)) floor(float); 432FN_FUNC_FN(floor) 433 434extern float __attribute__((overloadable)) fma(float, float, float); 435FN_FUNC_FN_FN_FN(fma) 436 437extern float __attribute__((overloadable)) fmin(float, float); 438 439extern float __attribute__((overloadable)) fmod(float, float); 440FN_FUNC_FN_FN(fmod) 441 442extern 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} 449FN_FUNC_FN_PFN(fract) 450 451extern float __attribute__((overloadable)) frexp(float, int *); 452FN_FUNC_FN_PIN(frexp) 453 454extern float __attribute__((overloadable)) hypot(float, float); 455FN_FUNC_FN_FN(hypot) 456 457extern int __attribute__((overloadable)) ilogb(float); 458IN_FUNC_FN(ilogb) 459 460extern float __attribute__((overloadable)) ldexp(float, int); 461FN_FUNC_FN_IN(ldexp) 462FN_FUNC_FN_I(ldexp) 463 464extern float __attribute__((overloadable)) lgamma(float); 465FN_FUNC_FN(lgamma) 466extern float __attribute__((overloadable)) lgamma(float, int*); 467FN_FUNC_FN_PIN(lgamma) 468 469extern float __attribute__((overloadable)) log(float); 470FN_FUNC_FN(log) 471 472extern float __attribute__((overloadable)) log10(float); 473FN_FUNC_FN(log10) 474 475 476extern float __attribute__((overloadable)) log2(float v) { 477 return log10(v) * 3.321928095f; 478} 479FN_FUNC_FN(log2) 480 481extern float __attribute__((overloadable)) log1p(float); 482FN_FUNC_FN(log1p) 483 484extern float __attribute__((overloadable)) logb(float); 485FN_FUNC_FN(logb) 486 487extern float __attribute__((overloadable)) mad(float a, float b, float c) { 488 return a * b + c; 489} 490extern float2 __attribute__((overloadable)) mad(float2 a, float2 b, float2 c) { 491 return a * b + c; 492} 493extern float3 __attribute__((overloadable)) mad(float3 a, float3 b, float3 c) { 494 return a * b + c; 495} 496extern float4 __attribute__((overloadable)) mad(float4 a, float4 b, float4 c) { 497 return a * b + c; 498} 499 500extern float __attribute__((overloadable)) modf(float, float *); 501FN_FUNC_FN_PFN(modf); 502 503extern 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 510extern float __attribute__((overloadable)) nextafter(float, float); 511FN_FUNC_FN_FN(nextafter) 512 513FN_FUNC_FN_FN(pow) 514 515extern float __attribute__((overloadable)) pown(float v, int p) { 516 return pow(v, (float)p); 517} 518extern float2 __attribute__((overloadable)) pown(float2 v, int2 p) { 519 float2 f2 = convert_float2(p); 520 return pow(v, f2); 521} 522extern float3 __attribute__((overloadable)) pown(float3 v, int3 p) { 523 float3 f3 = convert_float3(p); 524 return pow(v, f3); 525} 526extern float4 __attribute__((overloadable)) pown(float4 v, int4 p) { 527 float4 f4 = convert_float4(p); 528 return pow(v, f4); 529} 530 531extern float __attribute__((overloadable)) powr(float v, float p) { 532 return pow(v, p); 533} 534extern float2 __attribute__((overloadable)) powr(float2 v, float2 p) { 535 return pow(v, p); 536} 537extern float3 __attribute__((overloadable)) powr(float3 v, float3 p) { 538 return pow(v, p); 539} 540extern float4 __attribute__((overloadable)) powr(float4 v, float4 p) { 541 return pow(v, p); 542} 543 544extern float __attribute__((overloadable)) remainder(float, float); 545FN_FUNC_FN_FN(remainder) 546 547extern float __attribute__((overloadable)) remquo(float, float, int *); 548FN_FUNC_FN_FN_PIN(remquo) 549 550extern float __attribute__((overloadable)) rint(float); 551FN_FUNC_FN(rint) 552 553extern 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} 584FN_FUNC_FN_IN(rootn); 585 586extern float __attribute__((overloadable)) round(float); 587FN_FUNC_FN(round) 588 589 590extern float __attribute__((overloadable)) sqrt(float); 591extern float __attribute__((overloadable)) rsqrt(float v) { 592 return 1.f / sqrt(v); 593} 594 595#if !defined(__i386__) 596FN_FUNC_FN(sqrt) 597#endif // !defined(__i386__) 598 599FN_FUNC_FN(rsqrt) 600 601extern float __attribute__((overloadable)) sin(float); 602FN_FUNC_FN(sin) 603 604extern float __attribute__((overloadable)) sincos(float v, float *cosptr) { 605 *cosptr = cos(v); 606 return sin(v); 607} 608extern float2 __attribute__((overloadable)) sincos(float2 v, float2 *cosptr) { 609 *cosptr = cos(v); 610 return sin(v); 611} 612extern float3 __attribute__((overloadable)) sincos(float3 v, float3 *cosptr) { 613 *cosptr = cos(v); 614 return sin(v); 615} 616extern float4 __attribute__((overloadable)) sincos(float4 v, float4 *cosptr) { 617 *cosptr = cos(v); 618 return sin(v); 619} 620 621extern float __attribute__((overloadable)) sinh(float); 622FN_FUNC_FN(sinh) 623 624extern float __attribute__((overloadable)) sinpi(float v) { 625 return sin(v * M_PI); 626} 627FN_FUNC_FN(sinpi) 628 629extern float __attribute__((overloadable)) tan(float); 630FN_FUNC_FN(tan) 631 632extern float __attribute__((overloadable)) tanh(float); 633FN_FUNC_FN(tanh) 634 635extern float __attribute__((overloadable)) tanpi(float v) { 636 return tan(v * M_PI); 637} 638FN_FUNC_FN(tanpi) 639 640 641extern float __attribute__((overloadable)) tgamma(float); 642FN_FUNC_FN(tgamma) 643 644extern float __attribute__((overloadable)) trunc(float); 645FN_FUNC_FN(trunc) 646 647// Int ops (partial), 6.11.3 648 649#define XN_FUNC_YN(typeout, fnc, typein) \ 650extern typeout __attribute__((overloadable)) fnc(typein); \ 651extern 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} \ 657extern 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} \ 664extern 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) \ 675XN_FUNC_YN(uchar, fnc, char) \ 676XN_FUNC_YN(ushort, fnc, short) \ 677XN_FUNC_YN(uint, fnc, int) 678 679#define IN_FUNC_IN(fnc) \ 680XN_FUNC_YN(uchar, fnc, uchar) \ 681XN_FUNC_YN(char, fnc, char) \ 682XN_FUNC_YN(ushort, fnc, ushort) \ 683XN_FUNC_YN(short, fnc, short) \ 684XN_FUNC_YN(uint, fnc, uint) \ 685XN_FUNC_YN(int, fnc, int) 686 687 688#define XN_FUNC_XN_XN_BODY(type, fnc, body) \ 689extern type __attribute__((overloadable)) \ 690 fnc(type v1, type v2) { \ 691 return body; \ 692} \ 693extern 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} \ 700extern 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} \ 708extern 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) \ 719XN_FUNC_XN_XN_BODY(uchar, fnc, body) \ 720XN_FUNC_XN_XN_BODY(char, fnc, body) \ 721XN_FUNC_XN_XN_BODY(ushort, fnc, body) \ 722XN_FUNC_XN_XN_BODY(short, fnc, body) \ 723XN_FUNC_XN_XN_BODY(uint, fnc, body) \ 724XN_FUNC_XN_XN_BODY(int, fnc, body) \ 725XN_FUNC_XN_XN_BODY(float, fnc, body) 726 727 728/** 729 * abs 730 */ 731extern uint32_t __attribute__((overloadable)) abs(int32_t v) { 732 if (v < 0) 733 return -v; 734 return v; 735} 736extern uint16_t __attribute__((overloadable)) abs(int16_t v) { 737 if (v < 0) 738 return -v; 739 return v; 740} 741extern uint8_t __attribute__((overloadable)) abs(int8_t v) { 742 if (v < 0) 743 return -v; 744 return v; 745} 746 747/** 748 * clz 749 * __builtin_clz only accepts a 32-bit unsigned int, so every input will be 750 * expanded to 32 bits. For our smaller data types, we need to subtract off 751 * these unused top bits (that will be always be composed of zeros). 752 */ 753extern uint32_t __attribute__((overloadable)) clz(uint32_t v) { 754 return __builtin_clz(v); 755} 756extern uint16_t __attribute__((overloadable)) clz(uint16_t v) { 757 return __builtin_clz(v) - 16; 758} 759extern uint8_t __attribute__((overloadable)) clz(uint8_t v) { 760 return __builtin_clz(v) - 24; 761} 762extern int32_t __attribute__((overloadable)) clz(int32_t v) { 763 return __builtin_clz(v); 764} 765extern int16_t __attribute__((overloadable)) clz(int16_t v) { 766 return __builtin_clz(((uint32_t)v) & 0x0000ffff) - 16; 767} 768extern int8_t __attribute__((overloadable)) clz(int8_t v) { 769 return __builtin_clz(((uint32_t)v) & 0x000000ff) - 24; 770} 771 772 773UIN_FUNC_IN(abs) 774IN_FUNC_IN(clz) 775 776 777// 6.11.4 778 779 780extern float __attribute__((overloadable)) degrees(float radians) { 781 return radians * (180.f / M_PI); 782} 783extern float2 __attribute__((overloadable)) degrees(float2 radians) { 784 return radians * (180.f / M_PI); 785} 786extern float3 __attribute__((overloadable)) degrees(float3 radians) { 787 return radians * (180.f / M_PI); 788} 789extern float4 __attribute__((overloadable)) degrees(float4 radians) { 790 return radians * (180.f / M_PI); 791} 792 793extern float __attribute__((overloadable)) mix(float start, float stop, float amount) { 794 return start + (stop - start) * amount; 795} 796extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float2 amount) { 797 return start + (stop - start) * amount; 798} 799extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float3 amount) { 800 return start + (stop - start) * amount; 801} 802extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float4 amount) { 803 return start + (stop - start) * amount; 804} 805extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float amount) { 806 return start + (stop - start) * amount; 807} 808extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float amount) { 809 return start + (stop - start) * amount; 810} 811extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float amount) { 812 return start + (stop - start) * amount; 813} 814 815extern float __attribute__((overloadable)) radians(float degrees) { 816 return degrees * (M_PI / 180.f); 817} 818extern float2 __attribute__((overloadable)) radians(float2 degrees) { 819 return degrees * (M_PI / 180.f); 820} 821extern float3 __attribute__((overloadable)) radians(float3 degrees) { 822 return degrees * (M_PI / 180.f); 823} 824extern float4 __attribute__((overloadable)) radians(float4 degrees) { 825 return degrees * (M_PI / 180.f); 826} 827 828extern float __attribute__((overloadable)) step(float edge, float v) { 829 return (v < edge) ? 0.f : 1.f; 830} 831extern float2 __attribute__((overloadable)) step(float2 edge, float2 v) { 832 float2 r; 833 r.x = (v.x < edge.x) ? 0.f : 1.f; 834 r.y = (v.y < edge.y) ? 0.f : 1.f; 835 return r; 836} 837extern float3 __attribute__((overloadable)) step(float3 edge, float3 v) { 838 float3 r; 839 r.x = (v.x < edge.x) ? 0.f : 1.f; 840 r.y = (v.y < edge.y) ? 0.f : 1.f; 841 r.z = (v.z < edge.z) ? 0.f : 1.f; 842 return r; 843} 844extern float4 __attribute__((overloadable)) step(float4 edge, float4 v) { 845 float4 r; 846 r.x = (v.x < edge.x) ? 0.f : 1.f; 847 r.y = (v.y < edge.y) ? 0.f : 1.f; 848 r.z = (v.z < edge.z) ? 0.f : 1.f; 849 r.w = (v.w < edge.w) ? 0.f : 1.f; 850 return r; 851} 852extern float2 __attribute__((overloadable)) step(float2 edge, float v) { 853 float2 r; 854 r.x = (v < edge.x) ? 0.f : 1.f; 855 r.y = (v < edge.y) ? 0.f : 1.f; 856 return r; 857} 858extern float3 __attribute__((overloadable)) step(float3 edge, float v) { 859 float3 r; 860 r.x = (v < edge.x) ? 0.f : 1.f; 861 r.y = (v < edge.y) ? 0.f : 1.f; 862 r.z = (v < edge.z) ? 0.f : 1.f; 863 return r; 864} 865extern float4 __attribute__((overloadable)) step(float4 edge, float v) { 866 float4 r; 867 r.x = (v < edge.x) ? 0.f : 1.f; 868 r.y = (v < edge.y) ? 0.f : 1.f; 869 r.z = (v < edge.z) ? 0.f : 1.f; 870 r.w = (v < edge.w) ? 0.f : 1.f; 871 return r; 872} 873extern float2 __attribute__((overloadable)) step(float edge, float2 v) { 874 float2 r; 875 r.x = (v.x < edge) ? 0.f : 1.f; 876 r.y = (v.y < edge) ? 0.f : 1.f; 877 return r; 878} 879extern float3 __attribute__((overloadable)) step(float edge, float3 v) { 880 float3 r; 881 r.x = (v.x < edge) ? 0.f : 1.f; 882 r.y = (v.y < edge) ? 0.f : 1.f; 883 r.z = (v.z < edge) ? 0.f : 1.f; 884 return r; 885} 886extern float4 __attribute__((overloadable)) step(float edge, float4 v) { 887 float4 r; 888 r.x = (v.x < edge) ? 0.f : 1.f; 889 r.y = (v.y < edge) ? 0.f : 1.f; 890 r.z = (v.z < edge) ? 0.f : 1.f; 891 r.w = (v.w < edge) ? 0.f : 1.f; 892 return r; 893} 894 895extern float __attribute__((overloadable)) smoothstep(float, float, float); 896extern float2 __attribute__((overloadable)) smoothstep(float2, float2, float2); 897extern float3 __attribute__((overloadable)) smoothstep(float3, float3, float3); 898extern float4 __attribute__((overloadable)) smoothstep(float4, float4, float4); 899extern float2 __attribute__((overloadable)) smoothstep(float, float, float2); 900extern float3 __attribute__((overloadable)) smoothstep(float, float, float3); 901extern float4 __attribute__((overloadable)) smoothstep(float, float, float4); 902 903extern float __attribute__((overloadable)) sign(float v) { 904 if (v > 0) return 1.f; 905 if (v < 0) return -1.f; 906 return v; 907} 908FN_FUNC_FN(sign) 909 910 911// 6.11.5 912extern float3 __attribute__((overloadable)) cross(float3 lhs, float3 rhs) { 913 float3 r; 914 r.x = lhs.y * rhs.z - lhs.z * rhs.y; 915 r.y = lhs.z * rhs.x - lhs.x * rhs.z; 916 r.z = lhs.x * rhs.y - lhs.y * rhs.x; 917 return r; 918} 919 920extern float4 __attribute__((overloadable)) cross(float4 lhs, float4 rhs) { 921 float4 r; 922 r.x = lhs.y * rhs.z - lhs.z * rhs.y; 923 r.y = lhs.z * rhs.x - lhs.x * rhs.z; 924 r.z = lhs.x * rhs.y - lhs.y * rhs.x; 925 r.w = 0.f; 926 return r; 927} 928 929#if !defined(__i386__) 930 931extern float __attribute__((overloadable)) dot(float lhs, float rhs) { 932 return lhs * rhs; 933} 934extern float __attribute__((overloadable)) dot(float2 lhs, float2 rhs) { 935 return lhs.x*rhs.x + lhs.y*rhs.y; 936} 937extern float __attribute__((overloadable)) dot(float3 lhs, float3 rhs) { 938 return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z; 939} 940extern float __attribute__((overloadable)) dot(float4 lhs, float4 rhs) { 941 return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z + lhs.w*rhs.w; 942} 943 944extern float __attribute__((overloadable)) length(float v) { 945 return fabs(v); 946} 947extern float __attribute__((overloadable)) length(float2 v) { 948 return sqrt(v.x*v.x + v.y*v.y); 949} 950extern float __attribute__((overloadable)) length(float3 v) { 951 return sqrt(v.x*v.x + v.y*v.y + v.z*v.z); 952} 953extern float __attribute__((overloadable)) length(float4 v) { 954 return sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w); 955} 956 957#else 958 959extern float __attribute__((overloadable)) length(float v); 960extern float __attribute__((overloadable)) length(float2 v); 961extern float __attribute__((overloadable)) length(float3 v); 962extern float __attribute__((overloadable)) length(float4 v); 963 964#endif 965 966extern float __attribute__((overloadable)) distance(float lhs, float rhs) { 967 return length(lhs - rhs); 968} 969extern float __attribute__((overloadable)) distance(float2 lhs, float2 rhs) { 970 return length(lhs - rhs); 971} 972extern float __attribute__((overloadable)) distance(float3 lhs, float3 rhs) { 973 return length(lhs - rhs); 974} 975extern float __attribute__((overloadable)) distance(float4 lhs, float4 rhs) { 976 return length(lhs - rhs); 977} 978 979extern float __attribute__((overloadable)) normalize(float v) { 980 return 1.f; 981} 982extern float2 __attribute__((overloadable)) normalize(float2 v) { 983 return v / length(v); 984} 985extern float3 __attribute__((overloadable)) normalize(float3 v) { 986 return v / length(v); 987} 988extern float4 __attribute__((overloadable)) normalize(float4 v) { 989 return v / length(v); 990} 991 992extern float __attribute__((overloadable)) half_sqrt(float); 993 994extern float __attribute__((overloadable)) fast_length(float v) { 995 return fabs(v); 996} 997extern float __attribute__((overloadable)) fast_length(float2 v) { 998 return half_sqrt(v.x*v.x + v.y*v.y); 999} 1000extern float __attribute__((overloadable)) fast_length(float3 v) { 1001 return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z); 1002} 1003extern float __attribute__((overloadable)) fast_length(float4 v) { 1004 return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w); 1005} 1006 1007extern float __attribute__((overloadable)) fast_distance(float lhs, float rhs) { 1008 return fast_length(lhs - rhs); 1009} 1010extern float __attribute__((overloadable)) fast_distance(float2 lhs, float2 rhs) { 1011 return fast_length(lhs - rhs); 1012} 1013extern float __attribute__((overloadable)) fast_distance(float3 lhs, float3 rhs) { 1014 return fast_length(lhs - rhs); 1015} 1016extern float __attribute__((overloadable)) fast_distance(float4 lhs, float4 rhs) { 1017 return fast_length(lhs - rhs); 1018} 1019 1020extern float __attribute__((overloadable)) half_rsqrt(float); 1021 1022extern float __attribute__((overloadable)) fast_normalize(float v) { 1023 return 1.f; 1024} 1025extern float2 __attribute__((overloadable)) fast_normalize(float2 v) { 1026 return v * half_rsqrt(v.x*v.x + v.y*v.y); 1027} 1028extern float3 __attribute__((overloadable)) fast_normalize(float3 v) { 1029 return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z); 1030} 1031extern float4 __attribute__((overloadable)) fast_normalize(float4 v) { 1032 return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w); 1033} 1034 1035extern float __attribute__((overloadable)) half_recip(float); 1036 1037/* 1038extern float __attribute__((overloadable)) approx_atan(float x) { 1039 if (x == 0.f) 1040 return 0.f; 1041 if (x < 0.f) 1042 return -1.f * approx_atan(-1.f * x); 1043 if (x > 1.f) 1044 return M_PI_2 - approx_atan(approx_recip(x)); 1045 return x * approx_recip(1.f + 0.28f * x*x); 1046} 1047FN_FUNC_FN(approx_atan) 1048*/ 1049 1050typedef union 1051{ 1052 float fv; 1053 int32_t iv; 1054} ieee_float_shape_type; 1055 1056/* Get a 32 bit int from a float. */ 1057 1058#define GET_FLOAT_WORD(i,d) \ 1059do { \ 1060 ieee_float_shape_type gf_u; \ 1061 gf_u.fv = (d); \ 1062 (i) = gf_u.iv; \ 1063} while (0) 1064 1065/* Set a float from a 32 bit int. */ 1066 1067#define SET_FLOAT_WORD(d,i) \ 1068do { \ 1069 ieee_float_shape_type sf_u; \ 1070 sf_u.iv = (i); \ 1071 (d) = sf_u.fv; \ 1072} while (0) 1073 1074 1075 1076// Valid -125 to 125 1077extern float __attribute__((overloadable)) native_exp2(float v) { 1078 int32_t iv = (int)v; 1079 int32_t x = iv + (iv >> 31); // ~floor(v) 1080 float r = (v - x); 1081 1082 float fo; 1083 SET_FLOAT_WORD(fo, (x + 127) << 23); 1084 1085 r *= 0.694f; // ~ log(e) / log(2) 1086 float r2 = r*r; 1087 float adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f); 1088 return fo * adj; 1089} 1090 1091extern float2 __attribute__((overloadable)) native_exp2(float2 v) { 1092 int2 iv = convert_int2(v); 1093 int2 x = iv + (iv >> (int2)31);//floor(v); 1094 float2 r = (v - convert_float2(x)); 1095 1096 x += 127; 1097 1098 float2 fo = (float2)(x << (int2)23); 1099 1100 r *= 0.694f; // ~ log(e) / log(2) 1101 float2 r2 = r*r; 1102 float2 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f); 1103 return fo * adj; 1104} 1105 1106extern float4 __attribute__((overloadable)) native_exp2(float4 v) { 1107 int4 iv = convert_int4(v); 1108 int4 x = iv + (iv >> (int4)31);//floor(v); 1109 float4 r = (v - convert_float4(x)); 1110 1111 x += 127; 1112 1113 float4 fo = (float4)(x << (int4)23); 1114 1115 r *= 0.694f; // ~ log(e) / log(2) 1116 float4 r2 = r*r; 1117 float4 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f); 1118 return fo * adj; 1119} 1120 1121extern float3 __attribute__((overloadable)) native_exp2(float3 v) { 1122 float4 t = 1.f; 1123 t.xyz = v; 1124 return native_exp2(t).xyz; 1125} 1126 1127 1128extern float __attribute__((overloadable)) native_exp(float v) { 1129 return native_exp2(v * 1.442695041f); 1130} 1131extern float2 __attribute__((overloadable)) native_exp(float2 v) { 1132 return native_exp2(v * 1.442695041f); 1133} 1134extern float3 __attribute__((overloadable)) native_exp(float3 v) { 1135 return native_exp2(v * 1.442695041f); 1136} 1137extern float4 __attribute__((overloadable)) native_exp(float4 v) { 1138 return native_exp2(v * 1.442695041f); 1139} 1140 1141extern float __attribute__((overloadable)) native_exp10(float v) { 1142 return native_exp2(v * 3.321928095f); 1143} 1144extern float2 __attribute__((overloadable)) native_exp10(float2 v) { 1145 return native_exp2(v * 3.321928095f); 1146} 1147extern float3 __attribute__((overloadable)) native_exp10(float3 v) { 1148 return native_exp2(v * 3.321928095f); 1149} 1150extern float4 __attribute__((overloadable)) native_exp10(float4 v) { 1151 return native_exp2(v * 3.321928095f); 1152} 1153 1154extern float __attribute__((overloadable)) native_log2(float v) { 1155 int32_t ibits; 1156 GET_FLOAT_WORD(ibits, v); 1157 1158 int32_t e = (ibits >> 23) & 0xff; 1159 1160 ibits &= 0x7fffff; 1161 ibits |= 127 << 23; 1162 1163 float ir; 1164 SET_FLOAT_WORD(ir, ibits); 1165 1166 ir -= 1.5f; 1167 float ir2 = ir*ir; 1168 float adj2 = 0.405465108f + // -0.00009f + 1169 (0.666666667f * ir) - 1170 (0.222222222f * ir2) + 1171 (0.098765432f * ir*ir2) - 1172 (0.049382716f * ir2*ir2) + 1173 (0.026337449f * ir*ir2*ir2) - 1174 (0.014631916f * ir2*ir2*ir2); 1175 adj2 *= (1.f / 0.693147181f); 1176 1177 return (float)(e - 127) + adj2; 1178} 1179extern float2 __attribute__((overloadable)) native_log2(float2 v) { 1180 float2 v2 = {native_log2(v.x), native_log2(v.y)}; 1181 return v2; 1182} 1183extern float3 __attribute__((overloadable)) native_log2(float3 v) { 1184 float3 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z)}; 1185 return v2; 1186} 1187extern float4 __attribute__((overloadable)) native_log2(float4 v) { 1188 float4 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z), native_log2(v.w)}; 1189 return v2; 1190} 1191 1192extern float __attribute__((overloadable)) native_log(float v) { 1193 return native_log2(v) * (1.f / 1.442695041f); 1194} 1195extern float2 __attribute__((overloadable)) native_log(float2 v) { 1196 return native_log2(v) * (1.f / 1.442695041f); 1197} 1198extern float3 __attribute__((overloadable)) native_log(float3 v) { 1199 return native_log2(v) * (1.f / 1.442695041f); 1200} 1201extern float4 __attribute__((overloadable)) native_log(float4 v) { 1202 return native_log2(v) * (1.f / 1.442695041f); 1203} 1204 1205extern float __attribute__((overloadable)) native_log10(float v) { 1206 return native_log2(v) * (1.f / 3.321928095f); 1207} 1208extern float2 __attribute__((overloadable)) native_log10(float2 v) { 1209 return native_log2(v) * (1.f / 3.321928095f); 1210} 1211extern float3 __attribute__((overloadable)) native_log10(float3 v) { 1212 return native_log2(v) * (1.f / 3.321928095f); 1213} 1214extern float4 __attribute__((overloadable)) native_log10(float4 v) { 1215 return native_log2(v) * (1.f / 3.321928095f); 1216} 1217 1218 1219extern float __attribute__((overloadable)) native_powr(float v, float y) { 1220 float v2 = native_log2(v); 1221 v2 = fmax(v2, -125.f); 1222 return native_exp2(v2 * y); 1223} 1224extern float2 __attribute__((overloadable)) native_powr(float2 v, float2 y) { 1225 float2 v2 = native_log2(v); 1226 v2 = fmax(v2, -125.f); 1227 return native_exp2(v2 * y); 1228} 1229extern float3 __attribute__((overloadable)) native_powr(float3 v, float3 y) { 1230 float3 v2 = native_log2(v); 1231 v2 = fmax(v2, -125.f); 1232 return native_exp2(v2 * y); 1233} 1234extern float4 __attribute__((overloadable)) native_powr(float4 v, float4 y) { 1235 float4 v2 = native_log2(v); 1236 v2 = fmax(v2, -125.f); 1237 return native_exp2(v2 * y); 1238} 1239 1240 1241#undef FN_FUNC_FN 1242#undef IN_FUNC_FN 1243#undef FN_FUNC_FN_FN 1244#undef FN_FUNC_FN_F 1245#undef FN_FUNC_FN_IN 1246#undef FN_FUNC_FN_I 1247#undef FN_FUNC_FN_PFN 1248#undef FN_FUNC_FN_PIN 1249#undef FN_FUNC_FN_FN_FN 1250#undef FN_FUNC_FN_FN_PIN 1251#undef XN_FUNC_YN 1252#undef UIN_FUNC_IN 1253#undef IN_FUNC_IN 1254#undef XN_FUNC_XN_XN_BODY 1255#undef IN_FUNC_IN_IN_BODY 1256