1 // Copyright 2016 The SwiftShader Authors. All Rights Reserved. 2 // 3 // Licensed under the Apache License, Version 2.0 (the "License"); 4 // you may not use this file except in compliance with the License. 5 // You may obtain a copy of the License at 6 // 7 // http://www.apache.org/licenses/LICENSE-2.0 8 // 9 // Unless required by applicable law or agreed to in writing, software 10 // distributed under the License is distributed on an "AS IS" BASIS, 11 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 // See the License for the specific language governing permissions and 13 // limitations under the License. 14 15 #include "ShaderCore.hpp" 16 17 #include "Renderer/Renderer.hpp" 18 #include "Common/Debug.hpp" 19 20 #include <limits.h> 21 22 namespace sw 23 { 24 extern TranscendentalPrecision logPrecision; 25 extern TranscendentalPrecision expPrecision; 26 extern TranscendentalPrecision rcpPrecision; 27 extern TranscendentalPrecision rsqPrecision; 28 29 Vector4s::Vector4s() 30 { 31 } 32 33 Vector4s::Vector4s(unsigned short x, unsigned short y, unsigned short z, unsigned short w) 34 { 35 this->x = Short4(x); 36 this->y = Short4(y); 37 this->z = Short4(z); 38 this->w = Short4(w); 39 } 40 41 Vector4s::Vector4s(const Vector4s &rhs) 42 { 43 x = rhs.x; 44 y = rhs.y; 45 z = rhs.z; 46 w = rhs.w; 47 } 48 49 Vector4s &Vector4s::operator=(const Vector4s &rhs) 50 { 51 x = rhs.x; 52 y = rhs.y; 53 z = rhs.z; 54 w = rhs.w; 55 56 return *this; 57 } 58 59 Short4 &Vector4s::operator[](int i) 60 { 61 switch(i) 62 { 63 case 0: return x; 64 case 1: return y; 65 case 2: return z; 66 case 3: return w; 67 } 68 69 return x; 70 } 71 72 Vector4f::Vector4f() 73 { 74 } 75 76 Vector4f::Vector4f(float x, float y, float z, float w) 77 { 78 this->x = Float4(x); 79 this->y = Float4(y); 80 this->z = Float4(z); 81 this->w = Float4(w); 82 } 83 84 Vector4f::Vector4f(const Vector4f &rhs) 85 { 86 x = rhs.x; 87 y = rhs.y; 88 z = rhs.z; 89 w = rhs.w; 90 } 91 92 Vector4f &Vector4f::operator=(const Vector4f &rhs) 93 { 94 x = rhs.x; 95 y = rhs.y; 96 z = rhs.z; 97 w = rhs.w; 98 99 return *this; 100 } 101 102 Float4 &Vector4f::operator[](int i) 103 { 104 switch(i) 105 { 106 case 0: return x; 107 case 1: return y; 108 case 2: return z; 109 case 3: return w; 110 } 111 112 return x; 113 } 114 115 Float4 exponential2(RValue<Float4> x, bool pp) 116 { 117 // This implementation is based on 2^(i + f) = 2^i * 2^f, 118 // where i is the integer part of x and f is the fraction. 119 120 // For 2^i we can put the integer part directly in the exponent of 121 // the IEEE-754 floating-point number. Clamp to prevent overflow 122 // past the representation of infinity. 123 Float4 x0 = x; 124 x0 = Min(x0, As<Float4>(Int4(0x43010000))); // 129.00000e+0f 125 x0 = Max(x0, As<Float4>(Int4(0xC2FDFFFF))); // -126.99999e+0f 126 127 Int4 i = RoundInt(x0 - Float4(0.5f)); 128 Float4 ii = As<Float4>((i + Int4(127)) << 23); // Add single-precision bias, and shift into exponent. 129 130 // For the fractional part use a polynomial 131 // which approximates 2^f in the 0 to 1 range. 132 Float4 f = x0 - Float4(i); 133 Float4 ff = As<Float4>(Int4(0x3AF61905)); // 1.8775767e-3f 134 ff = ff * f + As<Float4>(Int4(0x3C134806)); // 8.9893397e-3f 135 ff = ff * f + As<Float4>(Int4(0x3D64AA23)); // 5.5826318e-2f 136 ff = ff * f + As<Float4>(Int4(0x3E75EAD4)); // 2.4015361e-1f 137 ff = ff * f + As<Float4>(Int4(0x3F31727B)); // 6.9315308e-1f 138 ff = ff * f + Float4(1.0f); 139 140 return ii * ff; 141 } 142 143 Float4 logarithm2(RValue<Float4> x, bool absolute, bool pp) 144 { 145 Float4 x0; 146 Float4 x1; 147 Float4 x2; 148 Float4 x3; 149 150 x0 = x; 151 152 x1 = As<Float4>(As<Int4>(x0) & Int4(0x7F800000)); 153 x1 = As<Float4>(As<UInt4>(x1) >> 8); 154 x1 = As<Float4>(As<Int4>(x1) | As<Int4>(Float4(1.0f))); 155 x1 = (x1 - Float4(1.4960938f)) * Float4(256.0f); // FIXME: (x1 - 1.4960938f) * 256.0f; 156 x0 = As<Float4>((As<Int4>(x0) & Int4(0x007FFFFF)) | As<Int4>(Float4(1.0f))); 157 158 x2 = (Float4(9.5428179e-2f) * x0 + Float4(4.7779095e-1f)) * x0 + Float4(1.9782813e-1f); 159 x3 = ((Float4(1.6618466e-2f) * x0 + Float4(2.0350508e-1f)) * x0 + Float4(2.7382900e-1f)) * x0 + Float4(4.0496687e-2f); 160 x2 /= x3; 161 162 x1 += (x0 - Float4(1.0f)) * x2; 163 164 Int4 pos_inf_x = CmpEQ(As<Int4>(x), Int4(0x7F800000)); 165 return As<Float4>((pos_inf_x & As<Int4>(x)) | (~pos_inf_x & As<Int4>(x1))); 166 } 167 168 Float4 exponential(RValue<Float4> x, bool pp) 169 { 170 // FIXME: Propagate the constant 171 return exponential2(Float4(1.44269504f) * x, pp); // 1/ln(2) 172 } 173 174 Float4 logarithm(RValue<Float4> x, bool absolute, bool pp) 175 { 176 // FIXME: Propagate the constant 177 return Float4(6.93147181e-1f) * logarithm2(x, absolute, pp); // ln(2) 178 } 179 180 Float4 power(RValue<Float4> x, RValue<Float4> y, bool pp) 181 { 182 Float4 log = logarithm2(x, true, pp); 183 log *= y; 184 return exponential2(log, pp); 185 } 186 187 Float4 reciprocal(RValue<Float4> x, bool pp, bool finite, bool exactAtPow2) 188 { 189 Float4 rcp; 190 191 if(!pp && rcpPrecision >= WHQL) 192 { 193 rcp = Float4(1.0f) / x; 194 } 195 else 196 { 197 rcp = Rcp_pp(x, exactAtPow2); 198 199 if(!pp) 200 { 201 rcp = (rcp + rcp) - (x * rcp * rcp); 202 } 203 } 204 205 if(finite) 206 { 207 int big = 0x7F7FFFFF; 208 rcp = Min(rcp, Float4((float&)big)); 209 } 210 211 return rcp; 212 } 213 214 Float4 reciprocalSquareRoot(RValue<Float4> x, bool absolute, bool pp) 215 { 216 Float4 abs = x; 217 218 if(absolute) 219 { 220 abs = Abs(abs); 221 } 222 223 Float4 rsq; 224 225 if(!pp) 226 { 227 rsq = Float4(1.0f) / Sqrt(abs); 228 } 229 else 230 { 231 rsq = RcpSqrt_pp(abs); 232 233 if(!pp) 234 { 235 rsq = rsq * (Float4(3.0f) - rsq * rsq * abs) * Float4(0.5f); 236 } 237 238 rsq = As<Float4>(CmpNEQ(As<Int4>(abs), Int4(0x7F800000)) & As<Int4>(rsq)); 239 } 240 241 return rsq; 242 } 243 244 Float4 modulo(RValue<Float4> x, RValue<Float4> y) 245 { 246 return x - y * Floor(x / y); 247 } 248 249 Float4 sine_pi(RValue<Float4> x, bool pp) 250 { 251 const Float4 A = Float4(-4.05284734e-1f); // -4/pi^2 252 const Float4 B = Float4(1.27323954e+0f); // 4/pi 253 const Float4 C = Float4(7.75160950e-1f); 254 const Float4 D = Float4(2.24839049e-1f); 255 256 // Parabola approximating sine 257 Float4 sin = x * (Abs(x) * A + B); 258 259 // Improve precision from 0.06 to 0.001 260 if(true) 261 { 262 sin = sin * (Abs(sin) * D + C); 263 } 264 265 return sin; 266 } 267 268 Float4 cosine_pi(RValue<Float4> x, bool pp) 269 { 270 // cos(x) = sin(x + pi/2) 271 Float4 y = x + Float4(1.57079632e+0f); 272 273 // Wrap around 274 y -= As<Float4>(CmpNLT(y, Float4(3.14159265e+0f)) & As<Int4>(Float4(6.28318530e+0f))); 275 276 return sine_pi(y, pp); 277 } 278 279 Float4 sine(RValue<Float4> x, bool pp) 280 { 281 // Reduce to [-0.5, 0.5] range 282 Float4 y = x * Float4(1.59154943e-1f); // 1/2pi 283 y = y - Round(y); 284 285 if(!pp) 286 { 287 // From the paper: "A Fast, Vectorizable Algorithm for Producing Single-Precision Sine-Cosine Pairs" 288 // This implementation passes OpenGL ES 3.0 precision requirements, at the cost of more operations: 289 // !pp : 17 mul, 7 add, 1 sub, 1 reciprocal 290 // pp : 4 mul, 2 add, 2 abs 291 292 Float4 y2 = y * y; 293 Float4 c1 = y2 * (y2 * (y2 * Float4(-0.0204391631f) + Float4(0.2536086171f)) + Float4(-1.2336977925f)) + Float4(1.0f); 294 Float4 s1 = y * (y2 * (y2 * (y2 * Float4(-0.0046075748f) + Float4(0.0796819754f)) + Float4(-0.645963615f)) + Float4(1.5707963235f)); 295 Float4 c2 = (c1 * c1) - (s1 * s1); 296 Float4 s2 = Float4(2.0f) * s1 * c1; 297 return Float4(2.0f) * s2 * c2 * reciprocal(s2 * s2 + c2 * c2, pp, true); 298 } 299 300 const Float4 A = Float4(-16.0f); 301 const Float4 B = Float4(8.0f); 302 const Float4 C = Float4(7.75160950e-1f); 303 const Float4 D = Float4(2.24839049e-1f); 304 305 // Parabola approximating sine 306 Float4 sin = y * (Abs(y) * A + B); 307 308 // Improve precision from 0.06 to 0.001 309 if(true) 310 { 311 sin = sin * (Abs(sin) * D + C); 312 } 313 314 return sin; 315 } 316 317 Float4 cosine(RValue<Float4> x, bool pp) 318 { 319 // cos(x) = sin(x + pi/2) 320 Float4 y = x + Float4(1.57079632e+0f); 321 return sine(y, pp); 322 } 323 324 Float4 tangent(RValue<Float4> x, bool pp) 325 { 326 return sine(x, pp) / cosine(x, pp); 327 } 328 329 Float4 arccos(RValue<Float4> x, bool pp) 330 { 331 // pi/2 - arcsin(x) 332 return Float4(1.57079632e+0f) - arcsin(x); 333 } 334 335 Float4 arcsin(RValue<Float4> x, bool pp) 336 { 337 if(false) // Simpler implementation fails even lowp precision tests 338 { 339 // x*(pi/2-sqrt(1-x*x)*pi/5) 340 return x * (Float4(1.57079632e+0f) - Sqrt(Float4(1.0f) - x*x) * Float4(6.28318531e-1f)); 341 } 342 else 343 { 344 // From 4.4.45, page 81 of the Handbook of Mathematical Functions, by Milton Abramowitz and Irene Stegun 345 const Float4 half_pi(1.57079632f); 346 const Float4 a0(1.5707288f); 347 const Float4 a1(-0.2121144f); 348 const Float4 a2(0.0742610f); 349 const Float4 a3(-0.0187293f); 350 Float4 absx = Abs(x); 351 return As<Float4>(As<Int4>(half_pi - Sqrt(Float4(1.0f) - absx) * (a0 + absx * (a1 + absx * (a2 + absx * a3)))) ^ 352 (As<Int4>(x) & Int4(0x80000000))); 353 } 354 } 355 356 // Approximation of atan in [0..1] 357 Float4 arctan_01(Float4 x, bool pp) 358 { 359 if(pp) 360 { 361 return x * (Float4(-0.27f) * x + Float4(1.05539816f)); 362 } 363 else 364 { 365 // From 4.4.49, page 81 of the Handbook of Mathematical Functions, by Milton Abramowitz and Irene Stegun 366 const Float4 a2(-0.3333314528f); 367 const Float4 a4(0.1999355085f); 368 const Float4 a6(-0.1420889944f); 369 const Float4 a8(0.1065626393f); 370 const Float4 a10(-0.0752896400f); 371 const Float4 a12(0.0429096138f); 372 const Float4 a14(-0.0161657367f); 373 const Float4 a16(0.0028662257f); 374 Float4 x2 = x * x; 375 return (x + x * (x2 * (a2 + x2 * (a4 + x2 * (a6 + x2 * (a8 + x2 * (a10 + x2 * (a12 + x2 * (a14 + x2 * a16))))))))); 376 } 377 } 378 379 Float4 arctan(RValue<Float4> x, bool pp) 380 { 381 Float4 absx = Abs(x); 382 Int4 O = CmpNLT(absx, Float4(1.0f)); 383 Float4 y = As<Float4>((O & As<Int4>(Float4(1.0f) / absx)) | (~O & As<Int4>(absx))); // FIXME: Vector select 384 385 const Float4 half_pi(1.57079632f); 386 Float4 theta = arctan_01(y, pp); 387 return As<Float4>(((O & As<Int4>(half_pi - theta)) | (~O & As<Int4>(theta))) ^ // FIXME: Vector select 388 (As<Int4>(x) & Int4(0x80000000))); 389 } 390 391 Float4 arctan(RValue<Float4> y, RValue<Float4> x, bool pp) 392 { 393 const Float4 pi(3.14159265f); // pi 394 const Float4 minus_pi(-3.14159265f); // -pi 395 const Float4 half_pi(1.57079632f); // pi/2 396 const Float4 quarter_pi(7.85398163e-1f); // pi/4 397 398 // Rotate to upper semicircle when in lower semicircle 399 Int4 S = CmpLT(y, Float4(0.0f)); 400 Float4 theta = As<Float4>(S & As<Int4>(minus_pi)); 401 Float4 x0 = As<Float4>((As<Int4>(y) & Int4(0x80000000)) ^ As<Int4>(x)); 402 Float4 y0 = Abs(y); 403 404 // Rotate to right quadrant when in left quadrant 405 Int4 non_zero_y = CmpNEQ(y0, Float4(0.0f)); 406 Int4 Q = CmpLT(x0, Float4(0.0f)) & non_zero_y; 407 theta += As<Float4>(Q & As<Int4>(half_pi)); 408 Float4 x1 = As<Float4>((Q & As<Int4>(y0)) | (~Q & As<Int4>(x0))); // FIXME: Vector select 409 Float4 y1 = As<Float4>((Q & As<Int4>(-x0)) | (~Q & As<Int4>(y0))); // FIXME: Vector select 410 411 // Mirror to first octant when in second octant 412 Int4 O = CmpNLT(y1, x1) & non_zero_y; 413 Float4 x2 = As<Float4>((O & As<Int4>(y1)) | (~O & As<Int4>(x1))); // FIXME: Vector select 414 Float4 y2 = As<Float4>((O & As<Int4>(x1)) | (~O & As<Int4>(y1))); // FIXME: Vector select 415 416 // Approximation of atan in [0..1] 417 Int4 zero_x = CmpEQ(x2, Float4(0.0f)); 418 Int4 inf_y = IsInf(y2); // Since x2 >= y2, this means x2 == y2 == inf, so we use 45 degrees or pi/4 419 Float4 atan2_theta = arctan_01(y2 / x2, pp); 420 theta += As<Float4>((~zero_x & ~inf_y & non_zero_y & ((O & As<Int4>(half_pi - atan2_theta)) | (~O & (As<Int4>(atan2_theta))))) | // FIXME: Vector select 421 (inf_y & As<Int4>(quarter_pi))); 422 423 // Recover loss of precision for tiny theta angles 424 Int4 precision_loss = S & Q & O & ~inf_y; // This combination results in (-pi + half_pi + half_pi - atan2_theta) which is equivalent to -atan2_theta 425 return As<Float4>((precision_loss & As<Int4>(-atan2_theta)) | (~precision_loss & As<Int4>(theta))); // FIXME: Vector select 426 } 427 428 Float4 sineh(RValue<Float4> x, bool pp) 429 { 430 return (exponential(x, pp) - exponential(-x, pp)) * Float4(0.5f); 431 } 432 433 Float4 cosineh(RValue<Float4> x, bool pp) 434 { 435 return (exponential(x, pp) + exponential(-x, pp)) * Float4(0.5f); 436 } 437 438 Float4 tangenth(RValue<Float4> x, bool pp) 439 { 440 Float4 e_x = exponential(x, pp); 441 Float4 e_minus_x = exponential(-x, pp); 442 return (e_x - e_minus_x) / (e_x + e_minus_x); 443 } 444 445 Float4 arccosh(RValue<Float4> x, bool pp) 446 { 447 return logarithm(x + Sqrt(x + Float4(1.0f)) * Sqrt(x - Float4(1.0f)), pp); 448 } 449 450 Float4 arcsinh(RValue<Float4> x, bool pp) 451 { 452 return logarithm(x + Sqrt(x * x + Float4(1.0f)), pp); 453 } 454 455 Float4 arctanh(RValue<Float4> x, bool pp) 456 { 457 return logarithm((Float4(1.0f) + x) / (Float4(1.0f) - x), pp) * Float4(0.5f); 458 } 459 460 Float4 dot2(const Vector4f &v0, const Vector4f &v1) 461 { 462 return v0.x * v1.x + v0.y * v1.y; 463 } 464 465 Float4 dot3(const Vector4f &v0, const Vector4f &v1) 466 { 467 return v0.x * v1.x + v0.y * v1.y + v0.z * v1.z; 468 } 469 470 Float4 dot4(const Vector4f &v0, const Vector4f &v1) 471 { 472 return v0.x * v1.x + v0.y * v1.y + v0.z * v1.z + v0.w * v1.w; 473 } 474 475 void transpose4x4(Short4 &row0, Short4 &row1, Short4 &row2, Short4 &row3) 476 { 477 Int2 tmp0 = UnpackHigh(row0, row1); 478 Int2 tmp1 = UnpackHigh(row2, row3); 479 Int2 tmp2 = UnpackLow(row0, row1); 480 Int2 tmp3 = UnpackLow(row2, row3); 481 482 row0 = UnpackLow(tmp2, tmp3); 483 row1 = UnpackHigh(tmp2, tmp3); 484 row2 = UnpackLow(tmp0, tmp1); 485 row3 = UnpackHigh(tmp0, tmp1); 486 } 487 488 void transpose4x3(Short4 &row0, Short4 &row1, Short4 &row2, Short4 &row3) 489 { 490 Int2 tmp0 = UnpackHigh(row0, row1); 491 Int2 tmp1 = UnpackHigh(row2, row3); 492 Int2 tmp2 = UnpackLow(row0, row1); 493 Int2 tmp3 = UnpackLow(row2, row3); 494 495 row0 = UnpackLow(tmp2, tmp3); 496 row1 = UnpackHigh(tmp2, tmp3); 497 row2 = UnpackLow(tmp0, tmp1); 498 } 499 500 void transpose4x4(Float4 &row0, Float4 &row1, Float4 &row2, Float4 &row3) 501 { 502 Float4 tmp0 = UnpackLow(row0, row1); 503 Float4 tmp1 = UnpackLow(row2, row3); 504 Float4 tmp2 = UnpackHigh(row0, row1); 505 Float4 tmp3 = UnpackHigh(row2, row3); 506 507 row0 = Float4(tmp0.xy, tmp1.xy); 508 row1 = Float4(tmp0.zw, tmp1.zw); 509 row2 = Float4(tmp2.xy, tmp3.xy); 510 row3 = Float4(tmp2.zw, tmp3.zw); 511 } 512 513 void transpose4x3(Float4 &row0, Float4 &row1, Float4 &row2, Float4 &row3) 514 { 515 Float4 tmp0 = UnpackLow(row0, row1); 516 Float4 tmp1 = UnpackLow(row2, row3); 517 Float4 tmp2 = UnpackHigh(row0, row1); 518 Float4 tmp3 = UnpackHigh(row2, row3); 519 520 row0 = Float4(tmp0.xy, tmp1.xy); 521 row1 = Float4(tmp0.zw, tmp1.zw); 522 row2 = Float4(tmp2.xy, tmp3.xy); 523 } 524 525 void transpose4x2(Float4 &row0, Float4 &row1, Float4 &row2, Float4 &row3) 526 { 527 Float4 tmp0 = UnpackLow(row0, row1); 528 Float4 tmp1 = UnpackLow(row2, row3); 529 530 row0 = Float4(tmp0.xy, tmp1.xy); 531 row1 = Float4(tmp0.zw, tmp1.zw); 532 } 533 534 void transpose4x1(Float4 &row0, Float4 &row1, Float4 &row2, Float4 &row3) 535 { 536 Float4 tmp0 = UnpackLow(row0, row1); 537 Float4 tmp1 = UnpackLow(row2, row3); 538 539 row0 = Float4(tmp0.xy, tmp1.xy); 540 } 541 542 void transpose2x4(Float4 &row0, Float4 &row1, Float4 &row2, Float4 &row3) 543 { 544 Float4 tmp01 = UnpackLow(row0, row1); 545 Float4 tmp23 = UnpackHigh(row0, row1); 546 547 row0 = tmp01; 548 row1 = Float4(tmp01.zw, row1.zw); 549 row2 = tmp23; 550 row3 = Float4(tmp23.zw, row3.zw); 551 } 552 553 void transpose4xN(Float4 &row0, Float4 &row1, Float4 &row2, Float4 &row3, int N) 554 { 555 switch(N) 556 { 557 case 1: transpose4x1(row0, row1, row2, row3); break; 558 case 2: transpose4x2(row0, row1, row2, row3); break; 559 case 3: transpose4x3(row0, row1, row2, row3); break; 560 case 4: transpose4x4(row0, row1, row2, row3); break; 561 } 562 } 563 564 void ShaderCore::mov(Vector4f &dst, const Vector4f &src, bool integerDestination) 565 { 566 if(integerDestination) 567 { 568 dst.x = As<Float4>(RoundInt(src.x)); 569 dst.y = As<Float4>(RoundInt(src.y)); 570 dst.z = As<Float4>(RoundInt(src.z)); 571 dst.w = As<Float4>(RoundInt(src.w)); 572 } 573 else 574 { 575 dst = src; 576 } 577 } 578 579 void ShaderCore::neg(Vector4f &dst, const Vector4f &src) 580 { 581 dst.x = -src.x; 582 dst.y = -src.y; 583 dst.z = -src.z; 584 dst.w = -src.w; 585 } 586 587 void ShaderCore::ineg(Vector4f &dst, const Vector4f &src) 588 { 589 dst.x = As<Float4>(-As<Int4>(src.x)); 590 dst.y = As<Float4>(-As<Int4>(src.y)); 591 dst.z = As<Float4>(-As<Int4>(src.z)); 592 dst.w = As<Float4>(-As<Int4>(src.w)); 593 } 594 595 void ShaderCore::f2b(Vector4f &dst, const Vector4f &src) 596 { 597 dst.x = As<Float4>(CmpNEQ(src.x, Float4(0.0f))); 598 dst.y = As<Float4>(CmpNEQ(src.y, Float4(0.0f))); 599 dst.z = As<Float4>(CmpNEQ(src.z, Float4(0.0f))); 600 dst.w = As<Float4>(CmpNEQ(src.w, Float4(0.0f))); 601 } 602 603 void ShaderCore::b2f(Vector4f &dst, const Vector4f &src) 604 { 605 dst.x = As<Float4>(As<Int4>(src.x) & As<Int4>(Float4(1.0f))); 606 dst.y = As<Float4>(As<Int4>(src.y) & As<Int4>(Float4(1.0f))); 607 dst.z = As<Float4>(As<Int4>(src.z) & As<Int4>(Float4(1.0f))); 608 dst.w = As<Float4>(As<Int4>(src.w) & As<Int4>(Float4(1.0f))); 609 } 610 611 void ShaderCore::f2i(Vector4f &dst, const Vector4f &src) 612 { 613 dst.x = As<Float4>(Int4(src.x)); 614 dst.y = As<Float4>(Int4(src.y)); 615 dst.z = As<Float4>(Int4(src.z)); 616 dst.w = As<Float4>(Int4(src.w)); 617 } 618 619 void ShaderCore::i2f(Vector4f &dst, const Vector4f &src) 620 { 621 dst.x = Float4(As<Int4>(src.x)); 622 dst.y = Float4(As<Int4>(src.y)); 623 dst.z = Float4(As<Int4>(src.z)); 624 dst.w = Float4(As<Int4>(src.w)); 625 } 626 627 void ShaderCore::f2u(Vector4f &dst, const Vector4f &src) 628 { 629 dst.x = As<Float4>(UInt4(src.x)); 630 dst.y = As<Float4>(UInt4(src.y)); 631 dst.z = As<Float4>(UInt4(src.z)); 632 dst.w = As<Float4>(UInt4(src.w)); 633 } 634 635 void ShaderCore::u2f(Vector4f &dst, const Vector4f &src) 636 { 637 dst.x = Float4(As<UInt4>(src.x)); 638 dst.y = Float4(As<UInt4>(src.y)); 639 dst.z = Float4(As<UInt4>(src.z)); 640 dst.w = Float4(As<UInt4>(src.w)); 641 } 642 643 void ShaderCore::i2b(Vector4f &dst, const Vector4f &src) 644 { 645 dst.x = As<Float4>(CmpNEQ(As<Int4>(src.x), Int4(0))); 646 dst.y = As<Float4>(CmpNEQ(As<Int4>(src.y), Int4(0))); 647 dst.z = As<Float4>(CmpNEQ(As<Int4>(src.z), Int4(0))); 648 dst.w = As<Float4>(CmpNEQ(As<Int4>(src.w), Int4(0))); 649 } 650 651 void ShaderCore::b2i(Vector4f &dst, const Vector4f &src) 652 { 653 dst.x = As<Float4>(As<Int4>(src.x) & Int4(1)); 654 dst.y = As<Float4>(As<Int4>(src.y) & Int4(1)); 655 dst.z = As<Float4>(As<Int4>(src.z) & Int4(1)); 656 dst.w = As<Float4>(As<Int4>(src.w) & Int4(1)); 657 } 658 659 void ShaderCore::add(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 660 { 661 dst.x = src0.x + src1.x; 662 dst.y = src0.y + src1.y; 663 dst.z = src0.z + src1.z; 664 dst.w = src0.w + src1.w; 665 } 666 667 void ShaderCore::iadd(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 668 { 669 dst.x = As<Float4>(As<Int4>(src0.x) + As<Int4>(src1.x)); 670 dst.y = As<Float4>(As<Int4>(src0.y) + As<Int4>(src1.y)); 671 dst.z = As<Float4>(As<Int4>(src0.z) + As<Int4>(src1.z)); 672 dst.w = As<Float4>(As<Int4>(src0.w) + As<Int4>(src1.w)); 673 } 674 675 void ShaderCore::sub(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 676 { 677 dst.x = src0.x - src1.x; 678 dst.y = src0.y - src1.y; 679 dst.z = src0.z - src1.z; 680 dst.w = src0.w - src1.w; 681 } 682 683 void ShaderCore::isub(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 684 { 685 dst.x = As<Float4>(As<Int4>(src0.x) - As<Int4>(src1.x)); 686 dst.y = As<Float4>(As<Int4>(src0.y) - As<Int4>(src1.y)); 687 dst.z = As<Float4>(As<Int4>(src0.z) - As<Int4>(src1.z)); 688 dst.w = As<Float4>(As<Int4>(src0.w) - As<Int4>(src1.w)); 689 } 690 691 void ShaderCore::mad(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2) 692 { 693 dst.x = src0.x * src1.x + src2.x; 694 dst.y = src0.y * src1.y + src2.y; 695 dst.z = src0.z * src1.z + src2.z; 696 dst.w = src0.w * src1.w + src2.w; 697 } 698 699 void ShaderCore::imad(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2) 700 { 701 dst.x = As<Float4>(As<Int4>(src0.x) * As<Int4>(src1.x) + As<Int4>(src2.x)); 702 dst.y = As<Float4>(As<Int4>(src0.y) * As<Int4>(src1.y) + As<Int4>(src2.y)); 703 dst.z = As<Float4>(As<Int4>(src0.z) * As<Int4>(src1.z) + As<Int4>(src2.z)); 704 dst.w = As<Float4>(As<Int4>(src0.w) * As<Int4>(src1.w) + As<Int4>(src2.w)); 705 } 706 707 void ShaderCore::mul(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 708 { 709 dst.x = src0.x * src1.x; 710 dst.y = src0.y * src1.y; 711 dst.z = src0.z * src1.z; 712 dst.w = src0.w * src1.w; 713 } 714 715 void ShaderCore::imul(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 716 { 717 dst.x = As<Float4>(As<Int4>(src0.x) * As<Int4>(src1.x)); 718 dst.y = As<Float4>(As<Int4>(src0.y) * As<Int4>(src1.y)); 719 dst.z = As<Float4>(As<Int4>(src0.z) * As<Int4>(src1.z)); 720 dst.w = As<Float4>(As<Int4>(src0.w) * As<Int4>(src1.w)); 721 } 722 723 void ShaderCore::rcpx(Vector4f &dst, const Vector4f &src, bool pp) 724 { 725 Float4 rcp = reciprocal(src.x, pp, true, true); 726 727 dst.x = rcp; 728 dst.y = rcp; 729 dst.z = rcp; 730 dst.w = rcp; 731 } 732 733 void ShaderCore::div(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 734 { 735 dst.x = src0.x / src1.x; 736 dst.y = src0.y / src1.y; 737 dst.z = src0.z / src1.z; 738 dst.w = src0.w / src1.w; 739 } 740 741 void ShaderCore::idiv(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 742 { 743 Float4 intMax(As<Float4>(Int4(INT_MAX))); 744 cmp0i(dst.x, src1.x, intMax, src1.x); 745 dst.x = As<Float4>(As<Int4>(src0.x) / As<Int4>(dst.x)); 746 cmp0i(dst.y, src1.y, intMax, src1.y); 747 dst.y = As<Float4>(As<Int4>(src0.y) / As<Int4>(dst.y)); 748 cmp0i(dst.z, src1.z, intMax, src1.z); 749 dst.z = As<Float4>(As<Int4>(src0.z) / As<Int4>(dst.z)); 750 cmp0i(dst.w, src1.w, intMax, src1.w); 751 dst.w = As<Float4>(As<Int4>(src0.w) / As<Int4>(dst.w)); 752 } 753 754 void ShaderCore::udiv(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 755 { 756 Float4 uintMax(As<Float4>(UInt4(UINT_MAX))); 757 cmp0i(dst.x, src1.x, uintMax, src1.x); 758 dst.x = As<Float4>(As<UInt4>(src0.x) / As<UInt4>(dst.x)); 759 cmp0i(dst.y, src1.y, uintMax, src1.y); 760 dst.y = As<Float4>(As<UInt4>(src0.y) / As<UInt4>(dst.y)); 761 cmp0i(dst.z, src1.z, uintMax, src1.z); 762 dst.z = As<Float4>(As<UInt4>(src0.z) / As<UInt4>(dst.z)); 763 cmp0i(dst.w, src1.w, uintMax, src1.w); 764 dst.w = As<Float4>(As<UInt4>(src0.w) / As<UInt4>(dst.w)); 765 } 766 767 void ShaderCore::mod(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 768 { 769 dst.x = modulo(src0.x, src1.x); 770 dst.y = modulo(src0.y, src1.y); 771 dst.z = modulo(src0.z, src1.z); 772 dst.w = modulo(src0.w, src1.w); 773 } 774 775 void ShaderCore::imod(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 776 { 777 Float4 intMax(As<Float4>(Int4(INT_MAX))); 778 cmp0i(dst.x, src1.x, intMax, src1.x); 779 dst.x = As<Float4>(As<Int4>(src0.x) % As<Int4>(dst.x)); 780 cmp0i(dst.y, src1.y, intMax, src1.y); 781 dst.y = As<Float4>(As<Int4>(src0.y) % As<Int4>(dst.y)); 782 cmp0i(dst.z, src1.z, intMax, src1.z); 783 dst.z = As<Float4>(As<Int4>(src0.z) % As<Int4>(dst.z)); 784 cmp0i(dst.w, src1.w, intMax, src1.w); 785 dst.w = As<Float4>(As<Int4>(src0.w) % As<Int4>(dst.w)); 786 } 787 788 void ShaderCore::umod(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 789 { 790 Float4 uintMax(As<Float4>(UInt4(UINT_MAX))); 791 cmp0i(dst.x, src1.x, uintMax, src1.x); 792 dst.x = As<Float4>(As<UInt4>(src0.x) % As<UInt4>(dst.x)); 793 cmp0i(dst.y, src1.y, uintMax, src1.y); 794 dst.y = As<Float4>(As<UInt4>(src0.y) % As<UInt4>(dst.y)); 795 cmp0i(dst.z, src1.z, uintMax, src1.z); 796 dst.z = As<Float4>(As<UInt4>(src0.z) % As<UInt4>(dst.z)); 797 cmp0i(dst.w, src1.w, uintMax, src1.w); 798 dst.w = As<Float4>(As<UInt4>(src0.w) % As<UInt4>(dst.w)); 799 } 800 801 void ShaderCore::shl(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 802 { 803 dst.x = As<Float4>(As<Int4>(src0.x) << As<Int4>(src1.x)); 804 dst.y = As<Float4>(As<Int4>(src0.y) << As<Int4>(src1.y)); 805 dst.z = As<Float4>(As<Int4>(src0.z) << As<Int4>(src1.z)); 806 dst.w = As<Float4>(As<Int4>(src0.w) << As<Int4>(src1.w)); 807 } 808 809 void ShaderCore::ishr(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 810 { 811 dst.x = As<Float4>(As<Int4>(src0.x) >> As<Int4>(src1.x)); 812 dst.y = As<Float4>(As<Int4>(src0.y) >> As<Int4>(src1.y)); 813 dst.z = As<Float4>(As<Int4>(src0.z) >> As<Int4>(src1.z)); 814 dst.w = As<Float4>(As<Int4>(src0.w) >> As<Int4>(src1.w)); 815 } 816 817 void ShaderCore::ushr(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 818 { 819 dst.x = As<Float4>(As<UInt4>(src0.x) >> As<UInt4>(src1.x)); 820 dst.y = As<Float4>(As<UInt4>(src0.y) >> As<UInt4>(src1.y)); 821 dst.z = As<Float4>(As<UInt4>(src0.z) >> As<UInt4>(src1.z)); 822 dst.w = As<Float4>(As<UInt4>(src0.w) >> As<UInt4>(src1.w)); 823 } 824 825 void ShaderCore::rsqx(Vector4f &dst, const Vector4f &src, bool pp) 826 { 827 Float4 rsq = reciprocalSquareRoot(src.x, true, pp); 828 829 dst.x = rsq; 830 dst.y = rsq; 831 dst.z = rsq; 832 dst.w = rsq; 833 } 834 835 void ShaderCore::sqrt(Vector4f &dst, const Vector4f &src, bool pp) 836 { 837 dst.x = Sqrt(src.x); 838 dst.y = Sqrt(src.y); 839 dst.z = Sqrt(src.z); 840 dst.w = Sqrt(src.w); 841 } 842 843 void ShaderCore::rsq(Vector4f &dst, const Vector4f &src, bool pp) 844 { 845 dst.x = reciprocalSquareRoot(src.x, false, pp); 846 dst.y = reciprocalSquareRoot(src.y, false, pp); 847 dst.z = reciprocalSquareRoot(src.z, false, pp); 848 dst.w = reciprocalSquareRoot(src.w, false, pp); 849 } 850 851 void ShaderCore::len2(Float4 &dst, const Vector4f &src, bool pp) 852 { 853 dst = Sqrt(dot2(src, src)); 854 } 855 856 void ShaderCore::len3(Float4 &dst, const Vector4f &src, bool pp) 857 { 858 dst = Sqrt(dot3(src, src)); 859 } 860 861 void ShaderCore::len4(Float4 &dst, const Vector4f &src, bool pp) 862 { 863 dst = Sqrt(dot4(src, src)); 864 } 865 866 void ShaderCore::dist1(Float4 &dst, const Vector4f &src0, const Vector4f &src1, bool pp) 867 { 868 dst = Abs(src0.x - src1.x); 869 } 870 871 void ShaderCore::dist2(Float4 &dst, const Vector4f &src0, const Vector4f &src1, bool pp) 872 { 873 Float4 dx = src0.x - src1.x; 874 Float4 dy = src0.y - src1.y; 875 Float4 dot2 = dx * dx + dy * dy; 876 dst = Sqrt(dot2); 877 } 878 879 void ShaderCore::dist3(Float4 &dst, const Vector4f &src0, const Vector4f &src1, bool pp) 880 { 881 Float4 dx = src0.x - src1.x; 882 Float4 dy = src0.y - src1.y; 883 Float4 dz = src0.z - src1.z; 884 Float4 dot3 = dx * dx + dy * dy + dz * dz; 885 dst = Sqrt(dot3); 886 } 887 888 void ShaderCore::dist4(Float4 &dst, const Vector4f &src0, const Vector4f &src1, bool pp) 889 { 890 Float4 dx = src0.x - src1.x; 891 Float4 dy = src0.y - src1.y; 892 Float4 dz = src0.z - src1.z; 893 Float4 dw = src0.w - src1.w; 894 Float4 dot4 = dx * dx + dy * dy + dz * dz + dw * dw; 895 dst = Sqrt(dot4); 896 } 897 898 void ShaderCore::dp1(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 899 { 900 Float4 t = src0.x * src1.x; 901 902 dst.x = t; 903 dst.y = t; 904 dst.z = t; 905 dst.w = t; 906 } 907 908 void ShaderCore::dp2(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 909 { 910 Float4 t = dot2(src0, src1); 911 912 dst.x = t; 913 dst.y = t; 914 dst.z = t; 915 dst.w = t; 916 } 917 918 void ShaderCore::dp2add(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2) 919 { 920 Float4 t = dot2(src0, src1) + src2.x; 921 922 dst.x = t; 923 dst.y = t; 924 dst.z = t; 925 dst.w = t; 926 } 927 928 void ShaderCore::dp3(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 929 { 930 Float4 dot = dot3(src0, src1); 931 932 dst.x = dot; 933 dst.y = dot; 934 dst.z = dot; 935 dst.w = dot; 936 } 937 938 void ShaderCore::dp4(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 939 { 940 Float4 dot = dot4(src0, src1); 941 942 dst.x = dot; 943 dst.y = dot; 944 dst.z = dot; 945 dst.w = dot; 946 } 947 948 void ShaderCore::min(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 949 { 950 dst.x = Min(src0.x, src1.x); 951 dst.y = Min(src0.y, src1.y); 952 dst.z = Min(src0.z, src1.z); 953 dst.w = Min(src0.w, src1.w); 954 } 955 956 void ShaderCore::imin(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 957 { 958 dst.x = As<Float4>(Min(As<Int4>(src0.x), As<Int4>(src1.x))); 959 dst.y = As<Float4>(Min(As<Int4>(src0.y), As<Int4>(src1.y))); 960 dst.z = As<Float4>(Min(As<Int4>(src0.z), As<Int4>(src1.z))); 961 dst.w = As<Float4>(Min(As<Int4>(src0.w), As<Int4>(src1.w))); 962 } 963 964 void ShaderCore::umin(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 965 { 966 dst.x = As<Float4>(Min(As<UInt4>(src0.x), As<UInt4>(src1.x))); 967 dst.y = As<Float4>(Min(As<UInt4>(src0.y), As<UInt4>(src1.y))); 968 dst.z = As<Float4>(Min(As<UInt4>(src0.z), As<UInt4>(src1.z))); 969 dst.w = As<Float4>(Min(As<UInt4>(src0.w), As<UInt4>(src1.w))); 970 } 971 972 void ShaderCore::max(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 973 { 974 dst.x = Max(src0.x, src1.x); 975 dst.y = Max(src0.y, src1.y); 976 dst.z = Max(src0.z, src1.z); 977 dst.w = Max(src0.w, src1.w); 978 } 979 980 void ShaderCore::imax(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 981 { 982 dst.x = As<Float4>(Max(As<Int4>(src0.x), As<Int4>(src1.x))); 983 dst.y = As<Float4>(Max(As<Int4>(src0.y), As<Int4>(src1.y))); 984 dst.z = As<Float4>(Max(As<Int4>(src0.z), As<Int4>(src1.z))); 985 dst.w = As<Float4>(Max(As<Int4>(src0.w), As<Int4>(src1.w))); 986 } 987 988 void ShaderCore::umax(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 989 { 990 dst.x = As<Float4>(Max(As<Int4>(src0.x), As<Int4>(src1.x))); 991 dst.y = As<Float4>(Max(As<Int4>(src0.y), As<Int4>(src1.y))); 992 dst.z = As<Float4>(Max(As<Int4>(src0.z), As<Int4>(src1.z))); 993 dst.w = As<Float4>(Max(As<Int4>(src0.w), As<Int4>(src1.w))); 994 } 995 996 void ShaderCore::slt(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 997 { 998 dst.x = As<Float4>(As<Int4>(CmpLT(src0.x, src1.x)) & As<Int4>(Float4(1.0f))); 999 dst.y = As<Float4>(As<Int4>(CmpLT(src0.y, src1.y)) & As<Int4>(Float4(1.0f))); 1000 dst.z = As<Float4>(As<Int4>(CmpLT(src0.z, src1.z)) & As<Int4>(Float4(1.0f))); 1001 dst.w = As<Float4>(As<Int4>(CmpLT(src0.w, src1.w)) & As<Int4>(Float4(1.0f))); 1002 } 1003 1004 void ShaderCore::step(Vector4f &dst, const Vector4f &edge, const Vector4f &x) 1005 { 1006 dst.x = As<Float4>(CmpNLT(x.x, edge.x) & As<Int4>(Float4(1.0f))); 1007 dst.y = As<Float4>(CmpNLT(x.y, edge.y) & As<Int4>(Float4(1.0f))); 1008 dst.z = As<Float4>(CmpNLT(x.z, edge.z) & As<Int4>(Float4(1.0f))); 1009 dst.w = As<Float4>(CmpNLT(x.w, edge.w) & As<Int4>(Float4(1.0f))); 1010 } 1011 1012 void ShaderCore::exp2x(Vector4f &dst, const Vector4f &src, bool pp) 1013 { 1014 Float4 exp = exponential2(src.x, pp); 1015 1016 dst.x = exp; 1017 dst.y = exp; 1018 dst.z = exp; 1019 dst.w = exp; 1020 } 1021 1022 void ShaderCore::exp2(Vector4f &dst, const Vector4f &src, bool pp) 1023 { 1024 dst.x = exponential2(src.x, pp); 1025 dst.y = exponential2(src.y, pp); 1026 dst.z = exponential2(src.z, pp); 1027 dst.w = exponential2(src.w, pp); 1028 } 1029 1030 void ShaderCore::exp(Vector4f &dst, const Vector4f &src, bool pp) 1031 { 1032 dst.x = exponential(src.x, pp); 1033 dst.y = exponential(src.y, pp); 1034 dst.z = exponential(src.z, pp); 1035 dst.w = exponential(src.w, pp); 1036 } 1037 1038 void ShaderCore::log2x(Vector4f &dst, const Vector4f &src, bool pp) 1039 { 1040 Float4 log = logarithm2(src.x, true, pp); 1041 1042 dst.x = log; 1043 dst.y = log; 1044 dst.z = log; 1045 dst.w = log; 1046 } 1047 1048 void ShaderCore::log2(Vector4f &dst, const Vector4f &src, bool pp) 1049 { 1050 dst.x = logarithm2(src.x, false, pp); 1051 dst.y = logarithm2(src.y, false, pp); 1052 dst.z = logarithm2(src.z, false, pp); 1053 dst.w = logarithm2(src.w, false, pp); 1054 } 1055 1056 void ShaderCore::log(Vector4f &dst, const Vector4f &src, bool pp) 1057 { 1058 dst.x = logarithm(src.x, false, pp); 1059 dst.y = logarithm(src.y, false, pp); 1060 dst.z = logarithm(src.z, false, pp); 1061 dst.w = logarithm(src.w, false, pp); 1062 } 1063 1064 void ShaderCore::lit(Vector4f &dst, const Vector4f &src) 1065 { 1066 dst.x = Float4(1.0f); 1067 dst.y = Max(src.x, Float4(0.0f)); 1068 1069 Float4 pow; 1070 1071 pow = src.w; 1072 pow = Min(pow, Float4(127.9961f)); 1073 pow = Max(pow, Float4(-127.9961f)); 1074 1075 dst.z = power(src.y, pow); 1076 dst.z = As<Float4>(As<Int4>(dst.z) & CmpNLT(src.x, Float4(0.0f))); 1077 dst.z = As<Float4>(As<Int4>(dst.z) & CmpNLT(src.y, Float4(0.0f))); 1078 1079 dst.w = Float4(1.0f); 1080 } 1081 1082 void ShaderCore::att(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1083 { 1084 // Computes attenuation factors (1, d, d^2, 1/d) assuming src0 = d^2, src1 = 1/d 1085 dst.x = 1; 1086 dst.y = src0.y * src1.y; 1087 dst.z = src0.z; 1088 dst.w = src1.w; 1089 } 1090 1091 void ShaderCore::lrp(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2) 1092 { 1093 dst.x = src0.x * (src1.x - src2.x) + src2.x; 1094 dst.y = src0.y * (src1.y - src2.y) + src2.y; 1095 dst.z = src0.z * (src1.z - src2.z) + src2.z; 1096 dst.w = src0.w * (src1.w - src2.w) + src2.w; 1097 } 1098 1099 void ShaderCore::isinf(Vector4f &dst, const Vector4f &src) 1100 { 1101 dst.x = As<Float4>(IsInf(src.x)); 1102 dst.y = As<Float4>(IsInf(src.y)); 1103 dst.z = As<Float4>(IsInf(src.z)); 1104 dst.w = As<Float4>(IsInf(src.w)); 1105 } 1106 1107 void ShaderCore::isnan(Vector4f &dst, const Vector4f &src) 1108 { 1109 dst.x = As<Float4>(IsNan(src.x)); 1110 dst.y = As<Float4>(IsNan(src.y)); 1111 dst.z = As<Float4>(IsNan(src.z)); 1112 dst.w = As<Float4>(IsNan(src.w)); 1113 } 1114 1115 void ShaderCore::smooth(Vector4f &dst, const Vector4f &edge0, const Vector4f &edge1, const Vector4f &x) 1116 { 1117 Float4 tx = Min(Max((x.x - edge0.x) / (edge1.x - edge0.x), Float4(0.0f)), Float4(1.0f)); dst.x = tx * tx * (Float4(3.0f) - Float4(2.0f) * tx); 1118 Float4 ty = Min(Max((x.y - edge0.y) / (edge1.y - edge0.y), Float4(0.0f)), Float4(1.0f)); dst.y = ty * ty * (Float4(3.0f) - Float4(2.0f) * ty); 1119 Float4 tz = Min(Max((x.z - edge0.z) / (edge1.z - edge0.z), Float4(0.0f)), Float4(1.0f)); dst.z = tz * tz * (Float4(3.0f) - Float4(2.0f) * tz); 1120 Float4 tw = Min(Max((x.w - edge0.w) / (edge1.w - edge0.w), Float4(0.0f)), Float4(1.0f)); dst.w = tw * tw * (Float4(3.0f) - Float4(2.0f) * tw); 1121 } 1122 1123 void ShaderCore::floatToHalfBits(Float4& dst, const Float4& floatBits, bool storeInUpperBits) 1124 { 1125 static const uint32_t mask_sign = 0x80000000u; 1126 static const uint32_t mask_round = ~0xfffu; 1127 static const uint32_t c_f32infty = 255 << 23; 1128 static const uint32_t c_magic = 15 << 23; 1129 static const uint32_t c_nanbit = 0x200; 1130 static const uint32_t c_infty_as_fp16 = 0x7c00; 1131 static const uint32_t c_clamp = (31 << 23) - 0x1000; 1132 1133 UInt4 justsign = UInt4(mask_sign) & As<UInt4>(floatBits); 1134 UInt4 absf = As<UInt4>(floatBits) ^ justsign; 1135 UInt4 b_isnormal = CmpNLE(UInt4(c_f32infty), absf); 1136 1137 // Note: this version doesn't round to the nearest even in case of a tie as defined by IEEE 754-2008, it rounds to +inf 1138 // instead of nearest even, since that's fine for GLSL ES 3.0's needs (see section 2.1.1 Floating-Point Computation) 1139 UInt4 joined = ((((As<UInt4>(Min(As<Float4>(absf & UInt4(mask_round)) * As<Float4>(UInt4(c_magic)), 1140 As<Float4>(UInt4(c_clamp))))) - UInt4(mask_round)) >> 13) & b_isnormal) | 1141 ((b_isnormal ^ UInt4(0xFFFFFFFF)) & ((CmpNLE(absf, UInt4(c_f32infty)) & UInt4(c_nanbit)) | 1142 UInt4(c_infty_as_fp16))); 1143 1144 dst = As<Float4>(storeInUpperBits ? As<UInt4>(dst) | ((joined << 16) | justsign) : joined | (justsign >> 16)); 1145 } 1146 1147 void ShaderCore::halfToFloatBits(Float4& dst, const Float4& halfBits) 1148 { 1149 static const uint32_t mask_nosign = 0x7FFF; 1150 static const uint32_t magic = (254 - 15) << 23; 1151 static const uint32_t was_infnan = 0x7BFF; 1152 static const uint32_t exp_infnan = 255 << 23; 1153 1154 UInt4 expmant = As<UInt4>(halfBits) & UInt4(mask_nosign); 1155 dst = As<Float4>(As<UInt4>(As<Float4>(expmant << 13) * As<Float4>(UInt4(magic))) | 1156 ((As<UInt4>(halfBits) ^ UInt4(expmant)) << 16) | 1157 (CmpNLE(As<UInt4>(expmant), UInt4(was_infnan)) & UInt4(exp_infnan))); 1158 } 1159 1160 void ShaderCore::packHalf2x16(Vector4f &d, const Vector4f &s0) 1161 { 1162 // half2 | half1 1163 floatToHalfBits(d.x, s0.x, false); 1164 floatToHalfBits(d.x, s0.y, true); 1165 } 1166 1167 void ShaderCore::unpackHalf2x16(Vector4f &dst, const Vector4f &s0) 1168 { 1169 // half2 | half1 1170 halfToFloatBits(dst.x, As<Float4>(As<UInt4>(s0.x) & UInt4(0x0000FFFF))); 1171 halfToFloatBits(dst.y, As<Float4>((As<UInt4>(s0.x) & UInt4(0xFFFF0000)) >> 16)); 1172 } 1173 1174 void ShaderCore::packSnorm2x16(Vector4f &d, const Vector4f &s0) 1175 { 1176 // round(clamp(c, -1.0, 1.0) * 32767.0) 1177 d.x = As<Float4>((Int4(Round(Min(Max(s0.x, Float4(-1.0f)), Float4(1.0f)) * Float4(32767.0f))) & Int4(0xFFFF)) | 1178 ((Int4(Round(Min(Max(s0.y, Float4(-1.0f)), Float4(1.0f)) * Float4(32767.0f))) & Int4(0xFFFF)) << 16)); 1179 } 1180 1181 void ShaderCore::packUnorm2x16(Vector4f &d, const Vector4f &s0) 1182 { 1183 // round(clamp(c, 0.0, 1.0) * 65535.0) 1184 d.x = As<Float4>((Int4(Round(Min(Max(s0.x, Float4(0.0f)), Float4(1.0f)) * Float4(65535.0f))) & Int4(0xFFFF)) | 1185 ((Int4(Round(Min(Max(s0.y, Float4(0.0f)), Float4(1.0f)) * Float4(65535.0f))) & Int4(0xFFFF)) << 16)); 1186 } 1187 1188 void ShaderCore::unpackSnorm2x16(Vector4f &dst, const Vector4f &s0) 1189 { 1190 // clamp(f / 32727.0, -1.0, 1.0) 1191 dst.x = Min(Max(Float4(As<Int4>((As<UInt4>(s0.x) & UInt4(0x0000FFFF)) << 16)) * Float4(1.0f / float(0x7FFF0000)), Float4(-1.0f)), Float4(1.0f)); 1192 dst.y = Min(Max(Float4(As<Int4>(As<UInt4>(s0.x) & UInt4(0xFFFF0000))) * Float4(1.0f / float(0x7FFF0000)), Float4(-1.0f)), Float4(1.0f)); 1193 } 1194 1195 void ShaderCore::unpackUnorm2x16(Vector4f &dst, const Vector4f &s0) 1196 { 1197 // f / 65535.0 1198 dst.x = Float4((As<UInt4>(s0.x) & UInt4(0x0000FFFF)) << 16) * Float4(1.0f / float(0xFFFF0000)); 1199 dst.y = Float4(As<UInt4>(s0.x) & UInt4(0xFFFF0000)) * Float4(1.0f / float(0xFFFF0000)); 1200 } 1201 1202 void ShaderCore::det2(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1203 { 1204 dst.x = src0.x * src1.y - src0.y * src1.x; 1205 dst.y = dst.z = dst.w = dst.x; 1206 } 1207 1208 void ShaderCore::det3(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2) 1209 { 1210 crs(dst, src1, src2); 1211 dp3(dst, dst, src0); 1212 } 1213 1214 void ShaderCore::det4(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2, const Vector4f &src3) 1215 { 1216 dst.x = src2.z * src3.w - src2.w * src3.z; 1217 dst.y = src1.w * src3.z - src1.z * src3.w; 1218 dst.z = src1.z * src2.w - src1.w * src2.z; 1219 dst.x = src0.x * (src1.y * dst.x + src2.y * dst.y + src3.y * dst.z) - 1220 src0.y * (src1.x * dst.x + src2.x * dst.y + src3.x * dst.z) + 1221 src0.z * (src1.x * (src2.y * src3.w - src2.w * src3.y) + 1222 src2.x * (src1.w * src3.y - src1.y * src3.w) + 1223 src3.x * (src1.y * src2.w - src1.w * src2.y)) + 1224 src0.w * (src1.x * (src2.z * src3.y - src2.y * src3.z) + 1225 src2.x * (src1.y * src3.z - src1.z * src3.y) + 1226 src3.x * (src1.z * src2.y - src1.y * src2.z)); 1227 dst.y = dst.z = dst.w = dst.x; 1228 } 1229 1230 void ShaderCore::frc(Vector4f &dst, const Vector4f &src) 1231 { 1232 dst.x = Frac(src.x); 1233 dst.y = Frac(src.y); 1234 dst.z = Frac(src.z); 1235 dst.w = Frac(src.w); 1236 } 1237 1238 void ShaderCore::trunc(Vector4f &dst, const Vector4f &src) 1239 { 1240 dst.x = Trunc(src.x); 1241 dst.y = Trunc(src.y); 1242 dst.z = Trunc(src.z); 1243 dst.w = Trunc(src.w); 1244 } 1245 1246 void ShaderCore::floor(Vector4f &dst, const Vector4f &src) 1247 { 1248 dst.x = Floor(src.x); 1249 dst.y = Floor(src.y); 1250 dst.z = Floor(src.z); 1251 dst.w = Floor(src.w); 1252 } 1253 1254 void ShaderCore::round(Vector4f &dst, const Vector4f &src) 1255 { 1256 dst.x = Round(src.x); 1257 dst.y = Round(src.y); 1258 dst.z = Round(src.z); 1259 dst.w = Round(src.w); 1260 } 1261 1262 void ShaderCore::roundEven(Vector4f &dst, const Vector4f &src) 1263 { 1264 // dst = round(src) + ((round(src) < src) * 2 - 1) * (fract(src) == 0.5) * isOdd(round(src)); 1265 // ex.: 1.5: 2 + (0 * 2 - 1) * 1 * 0 = 2 1266 // 2.5: 3 + (0 * 2 - 1) * 1 * 1 = 2 1267 // -1.5: -2 + (1 * 2 - 1) * 1 * 0 = -2 1268 // -2.5: -3 + (1 * 2 - 1) * 1 * 1 = -2 1269 // Even if the round implementation rounds the other way: 1270 // 1.5: 1 + (1 * 2 - 1) * 1 * 1 = 2 1271 // 2.5: 2 + (1 * 2 - 1) * 1 * 0 = 2 1272 // -1.5: -1 + (0 * 2 - 1) * 1 * 1 = -2 1273 // -2.5: -2 + (0 * 2 - 1) * 1 * 0 = -2 1274 round(dst, src); 1275 dst.x += ((Float4(CmpLT(dst.x, src.x) & Int4(1)) * Float4(2.0f)) - Float4(1.0f)) * Float4(CmpEQ(Frac(src.x), Float4(0.5f)) & Int4(1)) * Float4(Int4(dst.x) & Int4(1)); 1276 dst.y += ((Float4(CmpLT(dst.y, src.y) & Int4(1)) * Float4(2.0f)) - Float4(1.0f)) * Float4(CmpEQ(Frac(src.y), Float4(0.5f)) & Int4(1)) * Float4(Int4(dst.y) & Int4(1)); 1277 dst.z += ((Float4(CmpLT(dst.z, src.z) & Int4(1)) * Float4(2.0f)) - Float4(1.0f)) * Float4(CmpEQ(Frac(src.z), Float4(0.5f)) & Int4(1)) * Float4(Int4(dst.z) & Int4(1)); 1278 dst.w += ((Float4(CmpLT(dst.w, src.w) & Int4(1)) * Float4(2.0f)) - Float4(1.0f)) * Float4(CmpEQ(Frac(src.w), Float4(0.5f)) & Int4(1)) * Float4(Int4(dst.w) & Int4(1)); 1279 } 1280 1281 void ShaderCore::ceil(Vector4f &dst, const Vector4f &src) 1282 { 1283 dst.x = Ceil(src.x); 1284 dst.y = Ceil(src.y); 1285 dst.z = Ceil(src.z); 1286 dst.w = Ceil(src.w); 1287 } 1288 1289 void ShaderCore::powx(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, bool pp) 1290 { 1291 Float4 pow = power(src0.x, src1.x, pp); 1292 1293 dst.x = pow; 1294 dst.y = pow; 1295 dst.z = pow; 1296 dst.w = pow; 1297 } 1298 1299 void ShaderCore::pow(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, bool pp) 1300 { 1301 dst.x = power(src0.x, src1.x, pp); 1302 dst.y = power(src0.y, src1.y, pp); 1303 dst.z = power(src0.z, src1.z, pp); 1304 dst.w = power(src0.w, src1.w, pp); 1305 } 1306 1307 void ShaderCore::crs(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1308 { 1309 dst.x = src0.y * src1.z - src0.z * src1.y; 1310 dst.y = src0.z * src1.x - src0.x * src1.z; 1311 dst.z = src0.x * src1.y - src0.y * src1.x; 1312 } 1313 1314 void ShaderCore::forward1(Vector4f &dst, const Vector4f &N, const Vector4f &I, const Vector4f &Nref) 1315 { 1316 Int4 flip = CmpNLT(Nref.x * I.x, Float4(0.0f)) & Int4(0x80000000); 1317 1318 dst.x = As<Float4>(flip ^ As<Int4>(N.x)); 1319 } 1320 1321 void ShaderCore::forward2(Vector4f &dst, const Vector4f &N, const Vector4f &I, const Vector4f &Nref) 1322 { 1323 Int4 flip = CmpNLT(dot2(Nref, I), Float4(0.0f)) & Int4(0x80000000); 1324 1325 dst.x = As<Float4>(flip ^ As<Int4>(N.x)); 1326 dst.y = As<Float4>(flip ^ As<Int4>(N.y)); 1327 } 1328 1329 void ShaderCore::forward3(Vector4f &dst, const Vector4f &N, const Vector4f &I, const Vector4f &Nref) 1330 { 1331 Int4 flip = CmpNLT(dot3(Nref, I), Float4(0.0f)) & Int4(0x80000000); 1332 1333 dst.x = As<Float4>(flip ^ As<Int4>(N.x)); 1334 dst.y = As<Float4>(flip ^ As<Int4>(N.y)); 1335 dst.z = As<Float4>(flip ^ As<Int4>(N.z)); 1336 } 1337 1338 void ShaderCore::forward4(Vector4f &dst, const Vector4f &N, const Vector4f &I, const Vector4f &Nref) 1339 { 1340 Int4 flip = CmpNLT(dot4(Nref, I), Float4(0.0f)) & Int4(0x80000000); 1341 1342 dst.x = As<Float4>(flip ^ As<Int4>(N.x)); 1343 dst.y = As<Float4>(flip ^ As<Int4>(N.y)); 1344 dst.z = As<Float4>(flip ^ As<Int4>(N.z)); 1345 dst.w = As<Float4>(flip ^ As<Int4>(N.w)); 1346 } 1347 1348 void ShaderCore::reflect1(Vector4f &dst, const Vector4f &I, const Vector4f &N) 1349 { 1350 Float4 d = N.x * I.x; 1351 1352 dst.x = I.x - Float4(2.0f) * d * N.x; 1353 } 1354 1355 void ShaderCore::reflect2(Vector4f &dst, const Vector4f &I, const Vector4f &N) 1356 { 1357 Float4 d = dot2(N, I); 1358 1359 dst.x = I.x - Float4(2.0f) * d * N.x; 1360 dst.y = I.y - Float4(2.0f) * d * N.y; 1361 } 1362 1363 void ShaderCore::reflect3(Vector4f &dst, const Vector4f &I, const Vector4f &N) 1364 { 1365 Float4 d = dot3(N, I); 1366 1367 dst.x = I.x - Float4(2.0f) * d * N.x; 1368 dst.y = I.y - Float4(2.0f) * d * N.y; 1369 dst.z = I.z - Float4(2.0f) * d * N.z; 1370 } 1371 1372 void ShaderCore::reflect4(Vector4f &dst, const Vector4f &I, const Vector4f &N) 1373 { 1374 Float4 d = dot4(N, I); 1375 1376 dst.x = I.x - Float4(2.0f) * d * N.x; 1377 dst.y = I.y - Float4(2.0f) * d * N.y; 1378 dst.z = I.z - Float4(2.0f) * d * N.z; 1379 dst.w = I.w - Float4(2.0f) * d * N.w; 1380 } 1381 1382 void ShaderCore::refract1(Vector4f &dst, const Vector4f &I, const Vector4f &N, const Float4 &eta) 1383 { 1384 Float4 d = N.x * I.x; 1385 Float4 k = Float4(1.0f) - eta * eta * (Float4(1.0f) - d * d); 1386 Int4 pos = CmpNLT(k, Float4(0.0f)); 1387 Float4 t = (eta * d + Sqrt(k)); 1388 1389 dst.x = As<Float4>(pos & As<Int4>(eta * I.x - t * N.x)); 1390 } 1391 1392 void ShaderCore::refract2(Vector4f &dst, const Vector4f &I, const Vector4f &N, const Float4 &eta) 1393 { 1394 Float4 d = dot2(N, I); 1395 Float4 k = Float4(1.0f) - eta * eta * (Float4(1.0f) - d * d); 1396 Int4 pos = CmpNLT(k, Float4(0.0f)); 1397 Float4 t = (eta * d + Sqrt(k)); 1398 1399 dst.x = As<Float4>(pos & As<Int4>(eta * I.x - t * N.x)); 1400 dst.y = As<Float4>(pos & As<Int4>(eta * I.y - t * N.y)); 1401 } 1402 1403 void ShaderCore::refract3(Vector4f &dst, const Vector4f &I, const Vector4f &N, const Float4 &eta) 1404 { 1405 Float4 d = dot3(N, I); 1406 Float4 k = Float4(1.0f) - eta * eta * (Float4(1.0f) - d * d); 1407 Int4 pos = CmpNLT(k, Float4(0.0f)); 1408 Float4 t = (eta * d + Sqrt(k)); 1409 1410 dst.x = As<Float4>(pos & As<Int4>(eta * I.x - t * N.x)); 1411 dst.y = As<Float4>(pos & As<Int4>(eta * I.y - t * N.y)); 1412 dst.z = As<Float4>(pos & As<Int4>(eta * I.z - t * N.z)); 1413 } 1414 1415 void ShaderCore::refract4(Vector4f &dst, const Vector4f &I, const Vector4f &N, const Float4 &eta) 1416 { 1417 Float4 d = dot4(N, I); 1418 Float4 k = Float4(1.0f) - eta * eta * (Float4(1.0f) - d * d); 1419 Int4 pos = CmpNLT(k, Float4(0.0f)); 1420 Float4 t = (eta * d + Sqrt(k)); 1421 1422 dst.x = As<Float4>(pos & As<Int4>(eta * I.x - t * N.x)); 1423 dst.y = As<Float4>(pos & As<Int4>(eta * I.y - t * N.y)); 1424 dst.z = As<Float4>(pos & As<Int4>(eta * I.z - t * N.z)); 1425 dst.w = As<Float4>(pos & As<Int4>(eta * I.w - t * N.w)); 1426 } 1427 1428 void ShaderCore::sgn(Vector4f &dst, const Vector4f &src) 1429 { 1430 sgn(dst.x, src.x); 1431 sgn(dst.y, src.y); 1432 sgn(dst.z, src.z); 1433 sgn(dst.w, src.w); 1434 } 1435 1436 void ShaderCore::isgn(Vector4f &dst, const Vector4f &src) 1437 { 1438 isgn(dst.x, src.x); 1439 isgn(dst.y, src.y); 1440 isgn(dst.z, src.z); 1441 isgn(dst.w, src.w); 1442 } 1443 1444 void ShaderCore::abs(Vector4f &dst, const Vector4f &src) 1445 { 1446 dst.x = Abs(src.x); 1447 dst.y = Abs(src.y); 1448 dst.z = Abs(src.z); 1449 dst.w = Abs(src.w); 1450 } 1451 1452 void ShaderCore::iabs(Vector4f &dst, const Vector4f &src) 1453 { 1454 dst.x = As<Float4>(Abs(As<Int4>(src.x))); 1455 dst.y = As<Float4>(Abs(As<Int4>(src.y))); 1456 dst.z = As<Float4>(Abs(As<Int4>(src.z))); 1457 dst.w = As<Float4>(Abs(As<Int4>(src.w))); 1458 } 1459 1460 void ShaderCore::nrm2(Vector4f &dst, const Vector4f &src, bool pp) 1461 { 1462 Float4 dot = dot2(src, src); 1463 Float4 rsq = reciprocalSquareRoot(dot, false, pp); 1464 1465 dst.x = src.x * rsq; 1466 dst.y = src.y * rsq; 1467 dst.z = src.z * rsq; 1468 dst.w = src.w * rsq; 1469 } 1470 1471 void ShaderCore::nrm3(Vector4f &dst, const Vector4f &src, bool pp) 1472 { 1473 Float4 dot = dot3(src, src); 1474 Float4 rsq = reciprocalSquareRoot(dot, false, pp); 1475 1476 dst.x = src.x * rsq; 1477 dst.y = src.y * rsq; 1478 dst.z = src.z * rsq; 1479 dst.w = src.w * rsq; 1480 } 1481 1482 void ShaderCore::nrm4(Vector4f &dst, const Vector4f &src, bool pp) 1483 { 1484 Float4 dot = dot4(src, src); 1485 Float4 rsq = reciprocalSquareRoot(dot, false, pp); 1486 1487 dst.x = src.x * rsq; 1488 dst.y = src.y * rsq; 1489 dst.z = src.z * rsq; 1490 dst.w = src.w * rsq; 1491 } 1492 1493 void ShaderCore::sincos(Vector4f &dst, const Vector4f &src, bool pp) 1494 { 1495 dst.x = cosine_pi(src.x, pp); 1496 dst.y = sine_pi(src.x, pp); 1497 } 1498 1499 void ShaderCore::cos(Vector4f &dst, const Vector4f &src, bool pp) 1500 { 1501 dst.x = cosine(src.x, pp); 1502 dst.y = cosine(src.y, pp); 1503 dst.z = cosine(src.z, pp); 1504 dst.w = cosine(src.w, pp); 1505 } 1506 1507 void ShaderCore::sin(Vector4f &dst, const Vector4f &src, bool pp) 1508 { 1509 dst.x = sine(src.x, pp); 1510 dst.y = sine(src.y, pp); 1511 dst.z = sine(src.z, pp); 1512 dst.w = sine(src.w, pp); 1513 } 1514 1515 void ShaderCore::tan(Vector4f &dst, const Vector4f &src, bool pp) 1516 { 1517 dst.x = tangent(src.x, pp); 1518 dst.y = tangent(src.y, pp); 1519 dst.z = tangent(src.z, pp); 1520 dst.w = tangent(src.w, pp); 1521 } 1522 1523 void ShaderCore::acos(Vector4f &dst, const Vector4f &src, bool pp) 1524 { 1525 dst.x = arccos(src.x, pp); 1526 dst.y = arccos(src.y, pp); 1527 dst.z = arccos(src.z, pp); 1528 dst.w = arccos(src.w, pp); 1529 } 1530 1531 void ShaderCore::asin(Vector4f &dst, const Vector4f &src, bool pp) 1532 { 1533 dst.x = arcsin(src.x, pp); 1534 dst.y = arcsin(src.y, pp); 1535 dst.z = arcsin(src.z, pp); 1536 dst.w = arcsin(src.w, pp); 1537 } 1538 1539 void ShaderCore::atan(Vector4f &dst, const Vector4f &src, bool pp) 1540 { 1541 dst.x = arctan(src.x, pp); 1542 dst.y = arctan(src.y, pp); 1543 dst.z = arctan(src.z, pp); 1544 dst.w = arctan(src.w, pp); 1545 } 1546 1547 void ShaderCore::atan2(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, bool pp) 1548 { 1549 dst.x = arctan(src0.x, src1.x, pp); 1550 dst.y = arctan(src0.y, src1.y, pp); 1551 dst.z = arctan(src0.z, src1.z, pp); 1552 dst.w = arctan(src0.w, src1.w, pp); 1553 } 1554 1555 void ShaderCore::cosh(Vector4f &dst, const Vector4f &src, bool pp) 1556 { 1557 dst.x = cosineh(src.x, pp); 1558 dst.y = cosineh(src.y, pp); 1559 dst.z = cosineh(src.z, pp); 1560 dst.w = cosineh(src.w, pp); 1561 } 1562 1563 void ShaderCore::sinh(Vector4f &dst, const Vector4f &src, bool pp) 1564 { 1565 dst.x = sineh(src.x, pp); 1566 dst.y = sineh(src.y, pp); 1567 dst.z = sineh(src.z, pp); 1568 dst.w = sineh(src.w, pp); 1569 } 1570 1571 void ShaderCore::tanh(Vector4f &dst, const Vector4f &src, bool pp) 1572 { 1573 dst.x = tangenth(src.x, pp); 1574 dst.y = tangenth(src.y, pp); 1575 dst.z = tangenth(src.z, pp); 1576 dst.w = tangenth(src.w, pp); 1577 } 1578 1579 void ShaderCore::acosh(Vector4f &dst, const Vector4f &src, bool pp) 1580 { 1581 dst.x = arccosh(src.x, pp); 1582 dst.y = arccosh(src.y, pp); 1583 dst.z = arccosh(src.z, pp); 1584 dst.w = arccosh(src.w, pp); 1585 } 1586 1587 void ShaderCore::asinh(Vector4f &dst, const Vector4f &src, bool pp) 1588 { 1589 dst.x = arcsinh(src.x, pp); 1590 dst.y = arcsinh(src.y, pp); 1591 dst.z = arcsinh(src.z, pp); 1592 dst.w = arcsinh(src.w, pp); 1593 } 1594 1595 void ShaderCore::atanh(Vector4f &dst, const Vector4f &src, bool pp) 1596 { 1597 dst.x = arctanh(src.x, pp); 1598 dst.y = arctanh(src.y, pp); 1599 dst.z = arctanh(src.z, pp); 1600 dst.w = arctanh(src.w, pp); 1601 } 1602 1603 void ShaderCore::expp(Vector4f &dst, const Vector4f &src, unsigned short shaderModel) 1604 { 1605 if(shaderModel < 0x0200) 1606 { 1607 Float4 frc = Frac(src.x); 1608 Float4 floor = src.x - frc; 1609 1610 dst.x = exponential2(floor, true); 1611 dst.y = frc; 1612 dst.z = exponential2(src.x, true); 1613 dst.w = Float4(1.0f); 1614 } 1615 else // Version >= 2.0 1616 { 1617 exp2x(dst, src, true); // FIXME: 10-bit precision suffices 1618 } 1619 } 1620 1621 void ShaderCore::logp(Vector4f &dst, const Vector4f &src, unsigned short shaderModel) 1622 { 1623 if(shaderModel < 0x0200) 1624 { 1625 Float4 tmp0; 1626 Float4 tmp1; 1627 Float4 t; 1628 Int4 r; 1629 1630 tmp0 = Abs(src.x); 1631 tmp1 = tmp0; 1632 1633 // X component 1634 r = As<Int4>(As<UInt4>(tmp0) >> 23) - Int4(127); 1635 dst.x = Float4(r); 1636 1637 // Y component 1638 dst.y = As<Float4>((As<Int4>(tmp1) & Int4(0x007FFFFF)) | As<Int4>(Float4(1.0f))); 1639 1640 // Z component 1641 dst.z = logarithm2(src.x, true, true); 1642 1643 // W component 1644 dst.w = 1.0f; 1645 } 1646 else 1647 { 1648 log2x(dst, src, true); 1649 } 1650 } 1651 1652 void ShaderCore::cmp0(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2) 1653 { 1654 cmp0(dst.x, src0.x, src1.x, src2.x); 1655 cmp0(dst.y, src0.y, src1.y, src2.y); 1656 cmp0(dst.z, src0.z, src1.z, src2.z); 1657 cmp0(dst.w, src0.w, src1.w, src2.w); 1658 } 1659 1660 void ShaderCore::select(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, const Vector4f &src2) 1661 { 1662 select(dst.x, As<Int4>(src0.x), src1.x, src2.x); 1663 select(dst.y, As<Int4>(src0.y), src1.y, src2.y); 1664 select(dst.z, As<Int4>(src0.z), src1.z, src2.z); 1665 select(dst.w, As<Int4>(src0.w), src1.w, src2.w); 1666 } 1667 1668 void ShaderCore::extract(Float4 &dst, const Vector4f &src0, const Float4 &src1) 1669 { 1670 select(dst, CmpEQ(As<Int4>(src1), Int4(1)), src0.y, src0.x); 1671 select(dst, CmpEQ(As<Int4>(src1), Int4(2)), src0.z, dst); 1672 select(dst, CmpEQ(As<Int4>(src1), Int4(3)), src0.w, dst); 1673 } 1674 1675 void ShaderCore::insert(Vector4f &dst, const Vector4f &src, const Float4 &element, const Float4 &index) 1676 { 1677 select(dst.x, CmpEQ(As<Int4>(index), Int4(0)), element, src.x); 1678 select(dst.y, CmpEQ(As<Int4>(index), Int4(1)), element, src.y); 1679 select(dst.z, CmpEQ(As<Int4>(index), Int4(2)), element, src.z); 1680 select(dst.w, CmpEQ(As<Int4>(index), Int4(3)), element, src.w); 1681 } 1682 1683 void ShaderCore::sgn(Float4 &dst, const Float4 &src) 1684 { 1685 Int4 neg = As<Int4>(CmpLT(src, Float4(-0.0f))) & As<Int4>(Float4(-1.0f)); 1686 Int4 pos = As<Int4>(CmpNLE(src, Float4(+0.0f))) & As<Int4>(Float4(1.0f)); 1687 dst = As<Float4>(neg | pos); 1688 } 1689 1690 void ShaderCore::isgn(Float4 &dst, const Float4 &src) 1691 { 1692 Int4 neg = CmpLT(As<Int4>(src), Int4(0)) & Int4(-1); 1693 Int4 pos = CmpNLE(As<Int4>(src), Int4(0)) & Int4(1); 1694 dst = As<Float4>(neg | pos); 1695 } 1696 1697 void ShaderCore::cmp0(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2) 1698 { 1699 Int4 pos = CmpLE(Float4(0.0f), src0); 1700 select(dst, pos, src1, src2); 1701 } 1702 1703 void ShaderCore::cmp0i(Float4 &dst, const Float4 &src0, const Float4 &src1, const Float4 &src2) 1704 { 1705 Int4 pos = CmpEQ(Int4(0), As<Int4>(src0)); 1706 select(dst, pos, src1, src2); 1707 } 1708 1709 void ShaderCore::select(Float4 &dst, RValue<Int4> src0, const Float4 &src1, const Float4 &src2) 1710 { 1711 // FIXME: LLVM vector select 1712 dst = As<Float4>((src0 & As<Int4>(src1)) | (~src0 & As<Int4>(src2))); 1713 } 1714 1715 void ShaderCore::cmp(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, Control control) 1716 { 1717 switch(control) 1718 { 1719 case Shader::CONTROL_GT: 1720 dst.x = As<Float4>(CmpNLE(src0.x, src1.x)); 1721 dst.y = As<Float4>(CmpNLE(src0.y, src1.y)); 1722 dst.z = As<Float4>(CmpNLE(src0.z, src1.z)); 1723 dst.w = As<Float4>(CmpNLE(src0.w, src1.w)); 1724 break; 1725 case Shader::CONTROL_EQ: 1726 dst.x = As<Float4>(CmpEQ(src0.x, src1.x)); 1727 dst.y = As<Float4>(CmpEQ(src0.y, src1.y)); 1728 dst.z = As<Float4>(CmpEQ(src0.z, src1.z)); 1729 dst.w = As<Float4>(CmpEQ(src0.w, src1.w)); 1730 break; 1731 case Shader::CONTROL_GE: 1732 dst.x = As<Float4>(CmpNLT(src0.x, src1.x)); 1733 dst.y = As<Float4>(CmpNLT(src0.y, src1.y)); 1734 dst.z = As<Float4>(CmpNLT(src0.z, src1.z)); 1735 dst.w = As<Float4>(CmpNLT(src0.w, src1.w)); 1736 break; 1737 case Shader::CONTROL_LT: 1738 dst.x = As<Float4>(CmpLT(src0.x, src1.x)); 1739 dst.y = As<Float4>(CmpLT(src0.y, src1.y)); 1740 dst.z = As<Float4>(CmpLT(src0.z, src1.z)); 1741 dst.w = As<Float4>(CmpLT(src0.w, src1.w)); 1742 break; 1743 case Shader::CONTROL_NE: 1744 dst.x = As<Float4>(CmpNEQ(src0.x, src1.x)); 1745 dst.y = As<Float4>(CmpNEQ(src0.y, src1.y)); 1746 dst.z = As<Float4>(CmpNEQ(src0.z, src1.z)); 1747 dst.w = As<Float4>(CmpNEQ(src0.w, src1.w)); 1748 break; 1749 case Shader::CONTROL_LE: 1750 dst.x = As<Float4>(CmpLE(src0.x, src1.x)); 1751 dst.y = As<Float4>(CmpLE(src0.y, src1.y)); 1752 dst.z = As<Float4>(CmpLE(src0.z, src1.z)); 1753 dst.w = As<Float4>(CmpLE(src0.w, src1.w)); 1754 break; 1755 default: 1756 ASSERT(false); 1757 } 1758 } 1759 1760 void ShaderCore::icmp(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, Control control) 1761 { 1762 switch(control) 1763 { 1764 case Shader::CONTROL_GT: 1765 dst.x = As<Float4>(CmpNLE(As<Int4>(src0.x), As<Int4>(src1.x))); 1766 dst.y = As<Float4>(CmpNLE(As<Int4>(src0.y), As<Int4>(src1.y))); 1767 dst.z = As<Float4>(CmpNLE(As<Int4>(src0.z), As<Int4>(src1.z))); 1768 dst.w = As<Float4>(CmpNLE(As<Int4>(src0.w), As<Int4>(src1.w))); 1769 break; 1770 case Shader::CONTROL_EQ: 1771 dst.x = As<Float4>(CmpEQ(As<Int4>(src0.x), As<Int4>(src1.x))); 1772 dst.y = As<Float4>(CmpEQ(As<Int4>(src0.y), As<Int4>(src1.y))); 1773 dst.z = As<Float4>(CmpEQ(As<Int4>(src0.z), As<Int4>(src1.z))); 1774 dst.w = As<Float4>(CmpEQ(As<Int4>(src0.w), As<Int4>(src1.w))); 1775 break; 1776 case Shader::CONTROL_GE: 1777 dst.x = As<Float4>(CmpNLT(As<Int4>(src0.x), As<Int4>(src1.x))); 1778 dst.y = As<Float4>(CmpNLT(As<Int4>(src0.y), As<Int4>(src1.y))); 1779 dst.z = As<Float4>(CmpNLT(As<Int4>(src0.z), As<Int4>(src1.z))); 1780 dst.w = As<Float4>(CmpNLT(As<Int4>(src0.w), As<Int4>(src1.w))); 1781 break; 1782 case Shader::CONTROL_LT: 1783 dst.x = As<Float4>(CmpLT(As<Int4>(src0.x), As<Int4>(src1.x))); 1784 dst.y = As<Float4>(CmpLT(As<Int4>(src0.y), As<Int4>(src1.y))); 1785 dst.z = As<Float4>(CmpLT(As<Int4>(src0.z), As<Int4>(src1.z))); 1786 dst.w = As<Float4>(CmpLT(As<Int4>(src0.w), As<Int4>(src1.w))); 1787 break; 1788 case Shader::CONTROL_NE: 1789 dst.x = As<Float4>(CmpNEQ(As<Int4>(src0.x), As<Int4>(src1.x))); 1790 dst.y = As<Float4>(CmpNEQ(As<Int4>(src0.y), As<Int4>(src1.y))); 1791 dst.z = As<Float4>(CmpNEQ(As<Int4>(src0.z), As<Int4>(src1.z))); 1792 dst.w = As<Float4>(CmpNEQ(As<Int4>(src0.w), As<Int4>(src1.w))); 1793 break; 1794 case Shader::CONTROL_LE: 1795 dst.x = As<Float4>(CmpLE(As<Int4>(src0.x), As<Int4>(src1.x))); 1796 dst.y = As<Float4>(CmpLE(As<Int4>(src0.y), As<Int4>(src1.y))); 1797 dst.z = As<Float4>(CmpLE(As<Int4>(src0.z), As<Int4>(src1.z))); 1798 dst.w = As<Float4>(CmpLE(As<Int4>(src0.w), As<Int4>(src1.w))); 1799 break; 1800 default: 1801 ASSERT(false); 1802 } 1803 } 1804 1805 void ShaderCore::ucmp(Vector4f &dst, const Vector4f &src0, const Vector4f &src1, Control control) 1806 { 1807 switch(control) 1808 { 1809 case Shader::CONTROL_GT: 1810 dst.x = As<Float4>(CmpNLE(As<UInt4>(src0.x), As<UInt4>(src1.x))); 1811 dst.y = As<Float4>(CmpNLE(As<UInt4>(src0.y), As<UInt4>(src1.y))); 1812 dst.z = As<Float4>(CmpNLE(As<UInt4>(src0.z), As<UInt4>(src1.z))); 1813 dst.w = As<Float4>(CmpNLE(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1814 break; 1815 case Shader::CONTROL_EQ: 1816 dst.x = As<Float4>(CmpEQ(As<UInt4>(src0.x), As<UInt4>(src1.x))); 1817 dst.y = As<Float4>(CmpEQ(As<UInt4>(src0.y), As<UInt4>(src1.y))); 1818 dst.z = As<Float4>(CmpEQ(As<UInt4>(src0.z), As<UInt4>(src1.z))); 1819 dst.w = As<Float4>(CmpEQ(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1820 break; 1821 case Shader::CONTROL_GE: 1822 dst.x = As<Float4>(CmpNLT(As<UInt4>(src0.x), As<UInt4>(src1.x))); 1823 dst.y = As<Float4>(CmpNLT(As<UInt4>(src0.y), As<UInt4>(src1.y))); 1824 dst.z = As<Float4>(CmpNLT(As<UInt4>(src0.z), As<UInt4>(src1.z))); 1825 dst.w = As<Float4>(CmpNLT(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1826 break; 1827 case Shader::CONTROL_LT: 1828 dst.x = As<Float4>(CmpLT(As<UInt4>(src0.x), As<UInt4>(src1.x))); 1829 dst.y = As<Float4>(CmpLT(As<UInt4>(src0.y), As<UInt4>(src1.y))); 1830 dst.z = As<Float4>(CmpLT(As<UInt4>(src0.z), As<UInt4>(src1.z))); 1831 dst.w = As<Float4>(CmpLT(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1832 break; 1833 case Shader::CONTROL_NE: 1834 dst.x = As<Float4>(CmpNEQ(As<UInt4>(src0.x), As<UInt4>(src1.x))); 1835 dst.y = As<Float4>(CmpNEQ(As<UInt4>(src0.y), As<UInt4>(src1.y))); 1836 dst.z = As<Float4>(CmpNEQ(As<UInt4>(src0.z), As<UInt4>(src1.z))); 1837 dst.w = As<Float4>(CmpNEQ(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1838 break; 1839 case Shader::CONTROL_LE: 1840 dst.x = As<Float4>(CmpLE(As<UInt4>(src0.x), As<UInt4>(src1.x))); 1841 dst.y = As<Float4>(CmpLE(As<UInt4>(src0.y), As<UInt4>(src1.y))); 1842 dst.z = As<Float4>(CmpLE(As<UInt4>(src0.z), As<UInt4>(src1.z))); 1843 dst.w = As<Float4>(CmpLE(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1844 break; 1845 default: 1846 ASSERT(false); 1847 } 1848 } 1849 1850 void ShaderCore::all(Float4 &dst, const Vector4f &src) 1851 { 1852 dst = As<Float4>(As<Int4>(src.x) & As<Int4>(src.y) & As<Int4>(src.z) & As<Int4>(src.w)); 1853 } 1854 1855 void ShaderCore::any(Float4 &dst, const Vector4f &src) 1856 { 1857 dst = As<Float4>(As<Int4>(src.x) | As<Int4>(src.y) | As<Int4>(src.z) | As<Int4>(src.w)); 1858 } 1859 1860 void ShaderCore::bitwise_not(Vector4f &dst, const Vector4f &src) 1861 { 1862 dst.x = As<Float4>(As<Int4>(src.x) ^ Int4(0xFFFFFFFF)); 1863 dst.y = As<Float4>(As<Int4>(src.y) ^ Int4(0xFFFFFFFF)); 1864 dst.z = As<Float4>(As<Int4>(src.z) ^ Int4(0xFFFFFFFF)); 1865 dst.w = As<Float4>(As<Int4>(src.w) ^ Int4(0xFFFFFFFF)); 1866 } 1867 1868 void ShaderCore::bitwise_or(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1869 { 1870 dst.x = As<Float4>(As<Int4>(src0.x) | As<Int4>(src1.x)); 1871 dst.y = As<Float4>(As<Int4>(src0.y) | As<Int4>(src1.y)); 1872 dst.z = As<Float4>(As<Int4>(src0.z) | As<Int4>(src1.z)); 1873 dst.w = As<Float4>(As<Int4>(src0.w) | As<Int4>(src1.w)); 1874 } 1875 1876 void ShaderCore::bitwise_xor(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1877 { 1878 dst.x = As<Float4>(As<Int4>(src0.x) ^ As<Int4>(src1.x)); 1879 dst.y = As<Float4>(As<Int4>(src0.y) ^ As<Int4>(src1.y)); 1880 dst.z = As<Float4>(As<Int4>(src0.z) ^ As<Int4>(src1.z)); 1881 dst.w = As<Float4>(As<Int4>(src0.w) ^ As<Int4>(src1.w)); 1882 } 1883 1884 void ShaderCore::bitwise_and(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1885 { 1886 dst.x = As<Float4>(As<Int4>(src0.x) & As<Int4>(src1.x)); 1887 dst.y = As<Float4>(As<Int4>(src0.y) & As<Int4>(src1.y)); 1888 dst.z = As<Float4>(As<Int4>(src0.z) & As<Int4>(src1.z)); 1889 dst.w = As<Float4>(As<Int4>(src0.w) & As<Int4>(src1.w)); 1890 } 1891 1892 void ShaderCore::equal(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1893 { 1894 dst.x = As<Float4>(CmpEQ(As<UInt4>(src0.x), As<UInt4>(src1.x)) & 1895 CmpEQ(As<UInt4>(src0.y), As<UInt4>(src1.y)) & 1896 CmpEQ(As<UInt4>(src0.z), As<UInt4>(src1.z)) & 1897 CmpEQ(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1898 dst.y = dst.x; 1899 dst.z = dst.x; 1900 dst.w = dst.x; 1901 } 1902 1903 void ShaderCore::notEqual(Vector4f &dst, const Vector4f &src0, const Vector4f &src1) 1904 { 1905 dst.x = As<Float4>(CmpNEQ(As<UInt4>(src0.x), As<UInt4>(src1.x)) | 1906 CmpNEQ(As<UInt4>(src0.y), As<UInt4>(src1.y)) | 1907 CmpNEQ(As<UInt4>(src0.z), As<UInt4>(src1.z)) | 1908 CmpNEQ(As<UInt4>(src0.w), As<UInt4>(src1.w))); 1909 dst.y = dst.x; 1910 dst.z = dst.x; 1911 dst.w = dst.x; 1912 } 1913 } 1914
