| 1 | /* |
| 2 | * DXTn codec |
| 3 | * Version: 1.1 |
| 4 | * |
| 5 | * Copyright (C) 2004 Daniel Borca All Rights Reserved. |
| 6 | * |
| 7 | * this is free software; you can redistribute it and/or modify |
| 8 | * it under the terms of the GNU General Public License as published by |
| 9 | * the Free Software Foundation; either version 2, or (at your option) |
| 10 | * any later version. |
| 11 | * |
| 12 | * this is distributed in the hope that it will be useful, |
| 13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 15 | * GNU General Public License for more details. |
| 16 | * |
| 17 | * You should have received a copy of the GNU General Public License |
| 18 | * along with GNU Make; see the file COPYING. If not, write to |
| 19 | * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. |
| 20 | */ |
| 21 | |
| 22 | /* Copyright (C) 2007 Hiroshi Morii <koolsmoky(at)users.sourceforge.net> |
| 23 | * Added support for ARGB inputs, DXT3,5 workaround for ATI Radeons, and |
| 24 | * YUV conversions to determine representative colors. |
| 25 | */ |
| 26 | |
| 27 | |
| 28 | #include <stdlib.h> |
| 29 | #include <string.h> |
| 30 | #include <assert.h> |
| 31 | |
| 32 | #include <stdio.h> |
| 33 | |
| 34 | #include "types.h" |
| 35 | #include "internal.h" |
| 36 | #include "dxtn.h" |
| 37 | |
| 38 | |
| 39 | /***************************************************************************\ |
| 40 | * DXTn encoder |
| 41 | * |
| 42 | * The encoder was built by reversing the decoder, |
| 43 | * and is vaguely based on FXT1 codec. Note that this code |
| 44 | * is merely a proof of concept, since it is highly UNoptimized! |
| 45 | \***************************************************************************/ |
| 46 | |
| 47 | |
| 48 | #define MAX_COMP 4 /* ever needed maximum number of components in texel */ |
| 49 | #define MAX_VECT 4 /* ever needed maximum number of base vectors to find */ |
| 50 | #define N_TEXELS 16 /* number of texels in a block (always 16) */ |
| 51 | #define COLOR565(v) (word)((((v)[RCOMP] & 0xf8) << 8) | (((v)[GCOMP] & 0xfc) << 3) | ((v)[BCOMP] >> 3)) |
| 52 | |
| 53 | |
| 54 | static const int dxtn_color_tlat[2][4] = { |
| 55 | { 0, 2, 3, 1 }, |
| 56 | { 0, 2, 1, 3 } |
| 57 | }; |
| 58 | |
| 59 | static const int dxtn_alpha_tlat[2][8] = { |
| 60 | { 0, 2, 3, 4, 5, 6, 7, 1 }, |
| 61 | { 0, 2, 3, 4, 5, 1, 6, 7 } |
| 62 | }; |
| 63 | |
| 64 | |
| 65 | static void |
| 66 | dxt1_rgb_quantize (dword *cc, const byte *lines[], int comps) |
| 67 | { |
| 68 | float b, iv[MAX_COMP]; /* interpolation vector */ |
| 69 | |
| 70 | dword hi; /* high doubleword */ |
| 71 | int color0, color1; |
| 72 | int n_vect; |
| 73 | const int n_comp = 3; |
| 74 | int black = 0; |
| 75 | |
| 76 | #ifndef YUV |
| 77 | int minSum = 2000; /* big enough */ |
| 78 | #else |
| 79 | int minSum = 2000000; |
| 80 | #endif |
| 81 | int maxSum = -1; /* small enough */ |
| 82 | int minCol = 0; /* phoudoin: silent compiler! */ |
| 83 | int maxCol = 0; /* phoudoin: silent compiler! */ |
| 84 | |
| 85 | byte input[N_TEXELS][MAX_COMP]; |
| 86 | int i, k, l; |
| 87 | |
| 88 | /* make the whole block opaque */ |
| 89 | /* we will NEVER reference ACOMP of any pixel */ |
| 90 | |
| 91 | /* 4 texels each line */ |
| 92 | #ifndef ARGB |
| 93 | for (l = 0; l < 4; l++) { |
| 94 | for (k = 0; k < 4; k++) { |
| 95 | for (i = 0; i < comps; i++) { |
| 96 | input[k + l * 4][i] = *lines[l]++; |
| 97 | } |
| 98 | } |
| 99 | } |
| 100 | #else |
| 101 | /* H.Morii - support for ARGB inputs */ |
| 102 | for (l = 0; l < 4; l++) { |
| 103 | for (k = 0; k < 4; k++) { |
| 104 | input[k + l * 4][2] = *lines[l]++; |
| 105 | input[k + l * 4][1] = *lines[l]++; |
| 106 | input[k + l * 4][0] = *lines[l]++; |
| 107 | if (comps == 4) input[k + l * 4][3] = *lines[l]++; |
| 108 | } |
| 109 | } |
| 110 | #endif |
| 111 | |
| 112 | /* Our solution here is to find the darkest and brightest colors in |
| 113 | * the 4x4 tile and use those as the two representative colors. |
| 114 | * There are probably better algorithms to use (histogram-based). |
| 115 | */ |
| 116 | for (k = 0; k < N_TEXELS; k++) { |
| 117 | int sum = 0; |
| 118 | #ifndef YUV |
| 119 | for (i = 0; i < n_comp; i++) { |
| 120 | sum += input[k][i]; |
| 121 | } |
| 122 | #else |
| 123 | /* RGB to YUV conversion according to CCIR 601 specs |
| 124 | * Y = 0.299R+0.587G+0.114B |
| 125 | * U = 0.713(R - Y) = 0.500R-0.419G-0.081B |
| 126 | * V = 0.564(B - Y) = -0.169R-0.331G+0.500B |
| 127 | */ |
| 128 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 129 | #endif |
| 130 | if (minSum > sum) { |
| 131 | minSum = sum; |
| 132 | minCol = k; |
| 133 | } |
| 134 | if (maxSum < sum) { |
| 135 | maxSum = sum; |
| 136 | maxCol = k; |
| 137 | } |
| 138 | if (sum == 0) { |
| 139 | black = 1; |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | color0 = COLOR565(input[minCol]); |
| 144 | color1 = COLOR565(input[maxCol]); |
| 145 | |
| 146 | if (color0 == color1) { |
| 147 | /* we'll use 3-vector */ |
| 148 | cc[0] = color0 | (color1 << 16); |
| 149 | hi = black ? -1 : 0; |
| 150 | } else { |
| 151 | if (black && ((color0 == 0) || (color1 == 0))) { |
| 152 | /* we still can use 4-vector */ |
| 153 | black = 0; |
| 154 | } |
| 155 | |
| 156 | if (black ^ (color0 <= color1)) { |
| 157 | int aux; |
| 158 | aux = color0; |
| 159 | color0 = color1; |
| 160 | color1 = aux; |
| 161 | aux = minCol; |
| 162 | minCol = maxCol; |
| 163 | maxCol = aux; |
| 164 | } |
| 165 | n_vect = (color0 <= color1) ? 2 : 3; |
| 166 | |
| 167 | MAKEIVEC(n_vect, n_comp, iv, b, input[minCol], input[maxCol]); |
| 168 | |
| 169 | /* add in texels */ |
| 170 | cc[0] = color0 | (color1 << 16); |
| 171 | hi = 0; |
| 172 | for (k = N_TEXELS - 1; k >= 0; k--) { |
| 173 | int texel = 3; |
| 174 | int sum = 0; |
| 175 | if (black) { |
| 176 | for (i = 0; i < n_comp; i++) { |
| 177 | sum += input[k][i]; |
| 178 | } |
| 179 | } |
| 180 | if (!black || sum) { |
| 181 | /* interpolate color */ |
| 182 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 183 | texel = dxtn_color_tlat[black][texel]; |
| 184 | } |
| 185 | /* add in texel */ |
| 186 | hi <<= 2; |
| 187 | hi |= texel; |
| 188 | } |
| 189 | } |
| 190 | cc[1] = hi; |
| 191 | } |
| 192 | |
| 193 | |
| 194 | static void |
| 195 | dxt1_rgba_quantize (dword *cc, const byte *lines[], int comps) |
| 196 | { |
| 197 | float b, iv[MAX_COMP]; /* interpolation vector */ |
| 198 | |
| 199 | dword hi; /* high doubleword */ |
| 200 | int color0, color1; |
| 201 | int n_vect; |
| 202 | const int n_comp = 3; |
| 203 | int transparent = 0; |
| 204 | |
| 205 | #ifndef YUV |
| 206 | int minSum = 2000; /* big enough */ |
| 207 | #else |
| 208 | int minSum = 2000000; |
| 209 | #endif |
| 210 | int maxSum = -1; /* small enough */ |
| 211 | int minCol = 0; /* phoudoin: silent compiler! */ |
| 212 | int maxCol = 0; /* phoudoin: silent compiler! */ |
| 213 | |
| 214 | byte input[N_TEXELS][MAX_COMP]; |
| 215 | int i, k, l; |
| 216 | |
| 217 | if (comps == 3) { |
| 218 | /* make the whole block opaque */ |
| 219 | memset(input, -1, sizeof(input)); |
| 220 | } |
| 221 | |
| 222 | /* 4 texels each line */ |
| 223 | #ifndef ARGB |
| 224 | for (l = 0; l < 4; l++) { |
| 225 | for (k = 0; k < 4; k++) { |
| 226 | for (i = 0; i < comps; i++) { |
| 227 | input[k + l * 4][i] = *lines[l]++; |
| 228 | } |
| 229 | } |
| 230 | } |
| 231 | #else |
| 232 | /* H.Morii - support for ARGB inputs */ |
| 233 | for (l = 0; l < 4; l++) { |
| 234 | for (k = 0; k < 4; k++) { |
| 235 | input[k + l * 4][2] = *lines[l]++; |
| 236 | input[k + l * 4][1] = *lines[l]++; |
| 237 | input[k + l * 4][0] = *lines[l]++; |
| 238 | if (comps == 4) input[k + l * 4][3] = *lines[l]++; |
| 239 | } |
| 240 | } |
| 241 | #endif |
| 242 | |
| 243 | /* Our solution here is to find the darkest and brightest colors in |
| 244 | * the 4x4 tile and use those as the two representative colors. |
| 245 | * There are probably better algorithms to use (histogram-based). |
| 246 | */ |
| 247 | for (k = 0; k < N_TEXELS; k++) { |
| 248 | int sum = 0; |
| 249 | #ifndef YUV |
| 250 | for (i = 0; i < n_comp; i++) { |
| 251 | sum += input[k][i]; |
| 252 | } |
| 253 | #else |
| 254 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 255 | #endif |
| 256 | if (minSum > sum) { |
| 257 | minSum = sum; |
| 258 | minCol = k; |
| 259 | } |
| 260 | if (maxSum < sum) { |
| 261 | maxSum = sum; |
| 262 | maxCol = k; |
| 263 | } |
| 264 | if (input[k][ACOMP] < 128) { |
| 265 | transparent = 1; |
| 266 | } |
| 267 | } |
| 268 | |
| 269 | color0 = COLOR565(input[minCol]); |
| 270 | color1 = COLOR565(input[maxCol]); |
| 271 | |
| 272 | if (color0 == color1) { |
| 273 | /* we'll use 3-vector */ |
| 274 | cc[0] = color0 | (color1 << 16); |
| 275 | hi = transparent ? -1 : 0; |
| 276 | } else { |
| 277 | if (transparent ^ (color0 <= color1)) { |
| 278 | int aux; |
| 279 | aux = color0; |
| 280 | color0 = color1; |
| 281 | color1 = aux; |
| 282 | aux = minCol; |
| 283 | minCol = maxCol; |
| 284 | maxCol = aux; |
| 285 | } |
| 286 | n_vect = (color0 <= color1) ? 2 : 3; |
| 287 | |
| 288 | MAKEIVEC(n_vect, n_comp, iv, b, input[minCol], input[maxCol]); |
| 289 | |
| 290 | /* add in texels */ |
| 291 | cc[0] = color0 | (color1 << 16); |
| 292 | hi = 0; |
| 293 | for (k = N_TEXELS - 1; k >= 0; k--) { |
| 294 | int texel = 3; |
| 295 | if (input[k][ACOMP] >= 128) { |
| 296 | /* interpolate color */ |
| 297 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 298 | texel = dxtn_color_tlat[transparent][texel]; |
| 299 | } |
| 300 | /* add in texel */ |
| 301 | hi <<= 2; |
| 302 | hi |= texel; |
| 303 | } |
| 304 | } |
| 305 | cc[1] = hi; |
| 306 | } |
| 307 | |
| 308 | |
| 309 | static void |
| 310 | dxt3_rgba_quantize (dword *cc, const byte *lines[], int comps) |
| 311 | { |
| 312 | float b, iv[MAX_COMP]; /* interpolation vector */ |
| 313 | |
| 314 | dword lolo, lohi; /* low quadword: lo dword, hi dword */ |
| 315 | dword hihi; /* high quadword: high dword */ |
| 316 | int color0, color1; |
| 317 | const int n_vect = 3; |
| 318 | const int n_comp = 3; |
| 319 | |
| 320 | #ifndef YUV |
| 321 | int minSum = 2000; /* big enough */ |
| 322 | #else |
| 323 | int minSum = 2000000; |
| 324 | #endif |
| 325 | int maxSum = -1; /* small enough */ |
| 326 | int minCol = 0; /* phoudoin: silent compiler! */ |
| 327 | int maxCol = 0; /* phoudoin: silent compiler! */ |
| 328 | |
| 329 | byte input[N_TEXELS][MAX_COMP]; |
| 330 | int i, k, l; |
| 331 | |
| 332 | if (comps == 3) { |
| 333 | /* make the whole block opaque */ |
| 334 | memset(input, -1, sizeof(input)); |
| 335 | } |
| 336 | |
| 337 | /* 4 texels each line */ |
| 338 | #ifndef ARGB |
| 339 | for (l = 0; l < 4; l++) { |
| 340 | for (k = 0; k < 4; k++) { |
| 341 | for (i = 0; i < comps; i++) { |
| 342 | input[k + l * 4][i] = *lines[l]++; |
| 343 | } |
| 344 | } |
| 345 | } |
| 346 | #else |
| 347 | /* H.Morii - support for ARGB inputs */ |
| 348 | for (l = 0; l < 4; l++) { |
| 349 | for (k = 0; k < 4; k++) { |
| 350 | input[k + l * 4][2] = *lines[l]++; |
| 351 | input[k + l * 4][1] = *lines[l]++; |
| 352 | input[k + l * 4][0] = *lines[l]++; |
| 353 | if (comps == 4) input[k + l * 4][3] = *lines[l]++; |
| 354 | } |
| 355 | } |
| 356 | #endif |
| 357 | |
| 358 | /* Our solution here is to find the darkest and brightest colors in |
| 359 | * the 4x4 tile and use those as the two representative colors. |
| 360 | * There are probably better algorithms to use (histogram-based). |
| 361 | */ |
| 362 | for (k = 0; k < N_TEXELS; k++) { |
| 363 | int sum = 0; |
| 364 | #ifndef YUV |
| 365 | for (i = 0; i < n_comp; i++) { |
| 366 | sum += input[k][i]; |
| 367 | } |
| 368 | #else |
| 369 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 370 | #endif |
| 371 | if (minSum > sum) { |
| 372 | minSum = sum; |
| 373 | minCol = k; |
| 374 | } |
| 375 | if (maxSum < sum) { |
| 376 | maxSum = sum; |
| 377 | maxCol = k; |
| 378 | } |
| 379 | } |
| 380 | |
| 381 | /* add in alphas */ |
| 382 | lolo = lohi = 0; |
| 383 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
| 384 | /* add in alpha */ |
| 385 | lohi <<= 4; |
| 386 | lohi |= input[k][ACOMP] >> 4; |
| 387 | } |
| 388 | cc[1] = lohi; |
| 389 | for (; k >= 0; k--) { |
| 390 | /* add in alpha */ |
| 391 | lolo <<= 4; |
| 392 | lolo |= input[k][ACOMP] >> 4; |
| 393 | } |
| 394 | cc[0] = lolo; |
| 395 | |
| 396 | color0 = COLOR565(input[minCol]); |
| 397 | color1 = COLOR565(input[maxCol]); |
| 398 | |
| 399 | #ifdef RADEON |
| 400 | /* H.Morii - Workaround for ATI Radeon |
| 401 | * According to the OpenGL EXT_texture_compression_s3tc specs, |
| 402 | * the encoding of the RGB components for DXT3 and DXT5 formats |
| 403 | * use the non-transparent encodings of DXT1 but treated as |
| 404 | * though color0 > color1, regardless of the actual values of |
| 405 | * color0 and color1. ATI Radeons however require the values to |
| 406 | * be color0 > color1. |
| 407 | */ |
| 408 | if (color0 < color1) { |
| 409 | int aux; |
| 410 | aux = color0; |
| 411 | color0 = color1; |
| 412 | color1 = aux; |
| 413 | aux = minCol; |
| 414 | minCol = maxCol; |
| 415 | maxCol = aux; |
| 416 | } |
| 417 | #endif |
| 418 | |
| 419 | cc[2] = color0 | (color1 << 16); |
| 420 | |
| 421 | hihi = 0; |
| 422 | if (color0 != color1) { |
| 423 | MAKEIVEC(n_vect, n_comp, iv, b, input[minCol], input[maxCol]); |
| 424 | |
| 425 | /* add in texels */ |
| 426 | for (k = N_TEXELS - 1; k >= 0; k--) { |
| 427 | int texel; |
| 428 | /* interpolate color */ |
| 429 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 430 | texel = dxtn_color_tlat[0][texel]; |
| 431 | /* add in texel */ |
| 432 | hihi <<= 2; |
| 433 | hihi |= texel; |
| 434 | } |
| 435 | } |
| 436 | cc[3] = hihi; |
| 437 | } |
| 438 | |
| 439 | |
| 440 | static void |
| 441 | dxt5_rgba_quantize (dword *cc, const byte *lines[], int comps) |
| 442 | { |
| 443 | float b, iv[MAX_COMP]; /* interpolation vector */ |
| 444 | |
| 445 | qword lo; /* low quadword */ |
| 446 | dword hihi; /* high quadword: high dword */ |
| 447 | int color0, color1; |
| 448 | const int n_vect = 3; |
| 449 | const int n_comp = 3; |
| 450 | |
| 451 | #ifndef YUV |
| 452 | int minSum = 2000; /* big enough */ |
| 453 | #else |
| 454 | int minSum = 2000000; |
| 455 | #endif |
| 456 | int maxSum = -1; /* small enough */ |
| 457 | int minCol = 0; /* phoudoin: silent compiler! */ |
| 458 | int maxCol = 0; /* phoudoin: silent compiler! */ |
| 459 | int alpha0 = 2000; /* big enough */ |
| 460 | int alpha1 = -1; /* small enough */ |
| 461 | int anyZero = 0, anyOne = 0; |
| 462 | int a_vect; |
| 463 | |
| 464 | byte input[N_TEXELS][MAX_COMP]; |
| 465 | int i, k, l; |
| 466 | |
| 467 | if (comps == 3) { |
| 468 | /* make the whole block opaque */ |
| 469 | memset(input, -1, sizeof(input)); |
| 470 | } |
| 471 | |
| 472 | /* 4 texels each line */ |
| 473 | #ifndef ARGB |
| 474 | for (l = 0; l < 4; l++) { |
| 475 | for (k = 0; k < 4; k++) { |
| 476 | for (i = 0; i < comps; i++) { |
| 477 | input[k + l * 4][i] = *lines[l]++; |
| 478 | } |
| 479 | } |
| 480 | } |
| 481 | #else |
| 482 | /* H.Morii - support for ARGB inputs */ |
| 483 | for (l = 0; l < 4; l++) { |
| 484 | for (k = 0; k < 4; k++) { |
| 485 | input[k + l * 4][2] = *lines[l]++; |
| 486 | input[k + l * 4][1] = *lines[l]++; |
| 487 | input[k + l * 4][0] = *lines[l]++; |
| 488 | if (comps == 4) input[k + l * 4][3] = *lines[l]++; |
| 489 | } |
| 490 | } |
| 491 | #endif |
| 492 | |
| 493 | /* Our solution here is to find the darkest and brightest colors in |
| 494 | * the 4x4 tile and use those as the two representative colors. |
| 495 | * There are probably better algorithms to use (histogram-based). |
| 496 | */ |
| 497 | for (k = 0; k < N_TEXELS; k++) { |
| 498 | int sum = 0; |
| 499 | #ifndef YUV |
| 500 | for (i = 0; i < n_comp; i++) { |
| 501 | sum += input[k][i]; |
| 502 | } |
| 503 | #else |
| 504 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 505 | #endif |
| 506 | if (minSum > sum) { |
| 507 | minSum = sum; |
| 508 | minCol = k; |
| 509 | } |
| 510 | if (maxSum < sum) { |
| 511 | maxSum = sum; |
| 512 | maxCol = k; |
| 513 | } |
| 514 | if (alpha0 > input[k][ACOMP]) { |
| 515 | alpha0 = input[k][ACOMP]; |
| 516 | } |
| 517 | if (alpha1 < input[k][ACOMP]) { |
| 518 | alpha1 = input[k][ACOMP]; |
| 519 | } |
| 520 | if (input[k][ACOMP] == 0) { |
| 521 | anyZero = 1; |
| 522 | } |
| 523 | if (input[k][ACOMP] == 255) { |
| 524 | anyOne = 1; |
| 525 | } |
| 526 | } |
| 527 | |
| 528 | /* add in alphas */ |
| 529 | if (alpha0 == alpha1) { |
| 530 | /* we'll use 6-vector */ |
| 531 | cc[0] = alpha0 | (alpha1 << 8); |
| 532 | cc[1] = 0; |
| 533 | } else { |
| 534 | if (anyZero && ((alpha0 == 0) || (alpha1 == 0))) { |
| 535 | /* we still might use 8-vector */ |
| 536 | anyZero = 0; |
| 537 | } |
| 538 | if (anyOne && ((alpha0 == 255) || (alpha1 == 255))) { |
| 539 | /* we still might use 8-vector */ |
| 540 | anyOne = 0; |
| 541 | } |
| 542 | if ((anyZero | anyOne) ^ (alpha0 <= alpha1)) { |
| 543 | int aux; |
| 544 | aux = alpha0; |
| 545 | alpha0 = alpha1; |
| 546 | alpha1 = aux; |
| 547 | } |
| 548 | a_vect = (alpha0 <= alpha1) ? 5 : 7; |
| 549 | |
| 550 | /* compute interpolation vector */ |
| 551 | iv[ACOMP] = (float)a_vect / (alpha1 - alpha0); |
| 552 | b = -iv[ACOMP] * alpha0 + 0.5F; |
| 553 | |
| 554 | /* add in alphas */ |
| 555 | Q_MOV32(lo, 0); |
| 556 | for (k = N_TEXELS - 1; k >= 0; k--) { |
| 557 | int texel = -1; |
| 558 | if (anyZero | anyOne) { |
| 559 | if (input[k][ACOMP] == 0) { |
| 560 | texel = 6; |
| 561 | } else if (input[k][ACOMP] == 255) { |
| 562 | texel = 7; |
| 563 | } |
| 564 | } |
| 565 | /* interpolate alpha */ |
| 566 | if (texel == -1) { |
| 567 | float dot = input[k][ACOMP] * iv[ACOMP]; |
| 568 | texel = (int)(dot + b); |
| 569 | #if SAFECDOT |
| 570 | if (texel < 0) { |
| 571 | texel = 0; |
| 572 | } else if (texel > a_vect) { |
| 573 | texel = a_vect; |
| 574 | } |
| 575 | #endif |
| 576 | texel = dxtn_alpha_tlat[anyZero | anyOne][texel]; |
| 577 | } |
| 578 | /* add in texel */ |
| 579 | Q_SHL(lo, 3); |
| 580 | Q_OR32(lo, texel); |
| 581 | } |
| 582 | Q_SHL(lo, 16); |
| 583 | Q_OR32(lo, alpha0 | (alpha1 << 8)); |
| 584 | ((qword *)cc)[0] = lo; |
| 585 | } |
| 586 | |
| 587 | color0 = COLOR565(input[minCol]); |
| 588 | color1 = COLOR565(input[maxCol]); |
| 589 | |
| 590 | #ifdef RADEON /* H.Morii - Workaround for ATI Radeon */ |
| 591 | if (color0 < color1) { |
| 592 | int aux; |
| 593 | aux = color0; |
| 594 | color0 = color1; |
| 595 | color1 = aux; |
| 596 | aux = minCol; |
| 597 | minCol = maxCol; |
| 598 | maxCol = aux; |
| 599 | } |
| 600 | #endif |
| 601 | |
| 602 | cc[2] = color0 | (color1 << 16); |
| 603 | |
| 604 | hihi = 0; |
| 605 | if (color0 != color1) { |
| 606 | MAKEIVEC(n_vect, n_comp, iv, b, input[minCol], input[maxCol]); |
| 607 | |
| 608 | /* add in texels */ |
| 609 | for (k = N_TEXELS - 1; k >= 0; k--) { |
| 610 | int texel; |
| 611 | /* interpolate color */ |
| 612 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 613 | texel = dxtn_color_tlat[0][texel]; |
| 614 | /* add in texel */ |
| 615 | hihi <<= 2; |
| 616 | hihi |= texel; |
| 617 | } |
| 618 | } |
| 619 | cc[3] = hihi; |
| 620 | } |
| 621 | |
| 622 | |
| 623 | #define ENCODER(dxtn, n) \ |
| 624 | int TAPIENTRY \ |
| 625 | dxtn##_encode (int width, int height, int comps, \ |
| 626 | const void *source, int srcRowStride, \ |
| 627 | void *dest, int destRowStride) \ |
| 628 | { \ |
| 629 | int x, y; \ |
| 630 | const byte *data; \ |
| 631 | dword *encoded = (dword *)dest; \ |
| 632 | void *newSource = NULL; \ |
| 633 | \ |
| 634 | /* Replicate image if width is not M4 or height is not M4 */ \ |
| 635 | if ((width & 3) | (height & 3)) { \ |
| 636 | int newWidth = (width + 3) & ~3; \ |
| 637 | int newHeight = (height + 3) & ~3; \ |
| 638 | newSource = malloc(comps * newWidth * newHeight * sizeof(byte *));\ |
| 639 | _mesa_upscale_teximage2d(width, height, newWidth, newHeight, \ |
| 640 | comps, (const byte *)source, \ |
| 641 | srcRowStride, (byte *)newSource); \ |
| 642 | source = newSource; \ |
| 643 | width = newWidth; \ |
| 644 | height = newHeight; \ |
| 645 | srcRowStride = comps * newWidth; \ |
| 646 | } \ |
| 647 | \ |
| 648 | data = (const byte *)source; \ |
| 649 | destRowStride = (destRowStride - width * n) / 4; \ |
| 650 | for (y = 0; y < height; y += 4) { \ |
| 651 | unsigned int offs = 0 + (y + 0) * srcRowStride; \ |
| 652 | for (x = 0; x < width; x += 4) { \ |
| 653 | const byte *lines[4]; \ |
| 654 | lines[0] = &data[offs]; \ |
| 655 | lines[1] = lines[0] + srcRowStride; \ |
| 656 | lines[2] = lines[1] + srcRowStride; \ |
| 657 | lines[3] = lines[2] + srcRowStride; \ |
| 658 | offs += 4 * comps; \ |
| 659 | dxtn##_quantize(encoded, lines, comps); \ |
| 660 | /* 4x4 block */ \ |
| 661 | encoded += n; \ |
| 662 | } \ |
| 663 | encoded += destRowStride; \ |
| 664 | } \ |
| 665 | \ |
| 666 | if (newSource != NULL) { \ |
| 667 | free(newSource); \ |
| 668 | } \ |
| 669 | \ |
| 670 | return 0; \ |
| 671 | } |
| 672 | |
| 673 | ENCODER(dxt1_rgb, 2) |
| 674 | ENCODER(dxt1_rgba, 2) |
| 675 | ENCODER(dxt3_rgba, 4) |
| 676 | ENCODER(dxt5_rgba, 4) |
| 677 | |
| 678 | |
| 679 | /***************************************************************************\ |
| 680 | * DXTn decoder |
| 681 | * |
| 682 | * The decoder is based on GL_EXT_texture_compression_s3tc |
| 683 | * specification and serves as a concept for the encoder. |
| 684 | \***************************************************************************/ |
| 685 | |
| 686 | |
| 687 | /* lookup table for scaling 4 bit colors up to 8 bits */ |
| 688 | static const byte _rgb_scale_4[] = { |
| 689 | 0, 17, 34, 51, 68, 85, 102, 119, |
| 690 | 136, 153, 170, 187, 204, 221, 238, 255 |
| 691 | }; |
| 692 | |
| 693 | /* lookup table for scaling 5 bit colors up to 8 bits */ |
| 694 | static const byte _rgb_scale_5[] = { |
| 695 | 0, 8, 16, 25, 33, 41, 49, 58, |
| 696 | 66, 74, 82, 90, 99, 107, 115, 123, |
| 697 | 132, 140, 148, 156, 165, 173, 181, 189, |
| 698 | 197, 206, 214, 222, 230, 239, 247, 255 |
| 699 | }; |
| 700 | |
| 701 | /* lookup table for scaling 6 bit colors up to 8 bits */ |
| 702 | static const byte _rgb_scale_6[] = { |
| 703 | 0, 4, 8, 12, 16, 20, 24, 28, |
| 704 | 32, 36, 40, 45, 49, 53, 57, 61, |
| 705 | 65, 69, 73, 77, 81, 85, 89, 93, |
| 706 | 97, 101, 105, 109, 113, 117, 121, 125, |
| 707 | 130, 134, 138, 142, 146, 150, 154, 158, |
| 708 | 162, 166, 170, 174, 178, 182, 186, 190, |
| 709 | 194, 198, 202, 206, 210, 215, 219, 223, |
| 710 | 227, 231, 235, 239, 243, 247, 251, 255 |
| 711 | }; |
| 712 | |
| 713 | |
| 714 | #define CC_SEL(cc, which) (((dword *)(cc))[(which) / 32] >> ((which) & 31)) |
| 715 | #define UP4(c) _rgb_scale_4[(c) & 15] |
| 716 | #define UP5(c) _rgb_scale_5[(c) & 31] |
| 717 | #define UP6(c) _rgb_scale_6[(c) & 63] |
| 718 | #define ZERO_4UBV(v) *((dword *)(v)) = 0 |
| 719 | |
| 720 | |
| 721 | void TAPIENTRY |
| 722 | dxt1_rgb_decode_1 (const void *texture, int stride, |
| 723 | int i, int j, byte *rgba) |
| 724 | { |
| 725 | const byte *src = (const byte *)texture |
| 726 | + ((j / 4) * ((stride + 3) / 4) + i / 4) * 8; |
| 727 | const int code = (src[4 + (j & 3)] >> ((i & 3) * 2)) & 0x3; |
| 728 | if (code == 0) { |
| 729 | rgba[RCOMP] = UP5(CC_SEL(src, 11)); |
| 730 | rgba[GCOMP] = UP6(CC_SEL(src, 5)); |
| 731 | rgba[BCOMP] = UP5(CC_SEL(src, 0)); |
| 732 | } else if (code == 1) { |
| 733 | rgba[RCOMP] = UP5(CC_SEL(src, 27)); |
| 734 | rgba[GCOMP] = UP6(CC_SEL(src, 21)); |
| 735 | rgba[BCOMP] = UP5(CC_SEL(src, 16)); |
| 736 | } else { |
| 737 | const word col0 = src[0] | (src[1] << 8); |
| 738 | const word col1 = src[2] | (src[3] << 8); |
| 739 | if (col0 > col1) { |
| 740 | if (code == 2) { |
| 741 | rgba[RCOMP] = (UP5(col0 >> 11) * 2 + UP5(col1 >> 11)) / 3; |
| 742 | rgba[GCOMP] = (UP6(col0 >> 5) * 2 + UP6(col1 >> 5)) / 3; |
| 743 | rgba[BCOMP] = (UP5(col0 ) * 2 + UP5(col1 )) / 3; |
| 744 | } else { |
| 745 | rgba[RCOMP] = (UP5(col0 >> 11) + 2 * UP5(col1 >> 11)) / 3; |
| 746 | rgba[GCOMP] = (UP6(col0 >> 5) + 2 * UP6(col1 >> 5)) / 3; |
| 747 | rgba[BCOMP] = (UP5(col0 ) + 2 * UP5(col1 )) / 3; |
| 748 | } |
| 749 | } else { |
| 750 | if (code == 2) { |
| 751 | rgba[RCOMP] = (UP5(col0 >> 11) + UP5(col1 >> 11)) / 2; |
| 752 | rgba[GCOMP] = (UP6(col0 >> 5) + UP6(col1 >> 5)) / 2; |
| 753 | rgba[BCOMP] = (UP5(col0 ) + UP5(col1 )) / 2; |
| 754 | } else { |
| 755 | ZERO_4UBV(rgba); |
| 756 | } |
| 757 | } |
| 758 | } |
| 759 | rgba[ACOMP] = 255; |
| 760 | } |
| 761 | |
| 762 | |
| 763 | void TAPIENTRY |
| 764 | dxt1_rgba_decode_1 (const void *texture, int stride, |
| 765 | int i, int j, byte *rgba) |
| 766 | { |
| 767 | /* Same as rgb_dxt1 above, except alpha=0 if col0<=col1 and code=3. */ |
| 768 | const byte *src = (const byte *)texture |
| 769 | + ((j / 4) * ((stride + 3) / 4) + i / 4) * 8; |
| 770 | const int code = (src[4 + (j & 3)] >> ((i & 3) * 2)) & 0x3; |
| 771 | if (code == 0) { |
| 772 | rgba[RCOMP] = UP5(CC_SEL(src, 11)); |
| 773 | rgba[GCOMP] = UP6(CC_SEL(src, 5)); |
| 774 | rgba[BCOMP] = UP5(CC_SEL(src, 0)); |
| 775 | rgba[ACOMP] = 255; |
| 776 | } else if (code == 1) { |
| 777 | rgba[RCOMP] = UP5(CC_SEL(src, 27)); |
| 778 | rgba[GCOMP] = UP6(CC_SEL(src, 21)); |
| 779 | rgba[BCOMP] = UP5(CC_SEL(src, 16)); |
| 780 | rgba[ACOMP] = 255; |
| 781 | } else { |
| 782 | const word col0 = src[0] | (src[1] << 8); |
| 783 | const word col1 = src[2] | (src[3] << 8); |
| 784 | if (col0 > col1) { |
| 785 | if (code == 2) { |
| 786 | rgba[RCOMP] = (UP5(col0 >> 11) * 2 + UP5(col1 >> 11)) / 3; |
| 787 | rgba[GCOMP] = (UP6(col0 >> 5) * 2 + UP6(col1 >> 5)) / 3; |
| 788 | rgba[BCOMP] = (UP5(col0 ) * 2 + UP5(col1 )) / 3; |
| 789 | } else { |
| 790 | rgba[RCOMP] = (UP5(col0 >> 11) + 2 * UP5(col1 >> 11)) / 3; |
| 791 | rgba[GCOMP] = (UP6(col0 >> 5) + 2 * UP6(col1 >> 5)) / 3; |
| 792 | rgba[BCOMP] = (UP5(col0 ) + 2 * UP5(col1 )) / 3; |
| 793 | } |
| 794 | rgba[ACOMP] = 255; |
| 795 | } else { |
| 796 | if (code == 2) { |
| 797 | rgba[RCOMP] = (UP5(col0 >> 11) + UP5(col1 >> 11)) / 2; |
| 798 | rgba[GCOMP] = (UP6(col0 >> 5) + UP6(col1 >> 5)) / 2; |
| 799 | rgba[BCOMP] = (UP5(col0 ) + UP5(col1 )) / 2; |
| 800 | rgba[ACOMP] = 255; |
| 801 | } else { |
| 802 | ZERO_4UBV(rgba); |
| 803 | } |
| 804 | } |
| 805 | } |
| 806 | } |
| 807 | |
| 808 | |
| 809 | void TAPIENTRY |
| 810 | dxt3_rgba_decode_1 (const void *texture, int stride, |
| 811 | int i, int j, byte *rgba) |
| 812 | { |
| 813 | const byte *src = (const byte *)texture |
| 814 | + ((j / 4) * ((stride + 3) / 4) + i / 4) * 16; |
| 815 | const int code = (src[12 + (j & 3)] >> ((i & 3) * 2)) & 0x3; |
| 816 | const dword *cc = (const dword *)(src + 8); |
| 817 | if (code == 0) { |
| 818 | rgba[RCOMP] = UP5(CC_SEL(cc, 11)); |
| 819 | rgba[GCOMP] = UP6(CC_SEL(cc, 5)); |
| 820 | rgba[BCOMP] = UP5(CC_SEL(cc, 0)); |
| 821 | } else if (code == 1) { |
| 822 | rgba[RCOMP] = UP5(CC_SEL(cc, 27)); |
| 823 | rgba[GCOMP] = UP6(CC_SEL(cc, 21)); |
| 824 | rgba[BCOMP] = UP5(CC_SEL(cc, 16)); |
| 825 | } else if (code == 2) { |
| 826 | /* (col0 * (4 - code) + col1 * (code - 1)) / 3 */ |
| 827 | rgba[RCOMP] = (UP5(CC_SEL(cc, 11)) * 2 + UP5(CC_SEL(cc, 27))) / 3; |
| 828 | rgba[GCOMP] = (UP6(CC_SEL(cc, 5)) * 2 + UP6(CC_SEL(cc, 21))) / 3; |
| 829 | rgba[BCOMP] = (UP5(CC_SEL(cc, 0)) * 2 + UP5(CC_SEL(cc, 16))) / 3; |
| 830 | } else { |
| 831 | rgba[RCOMP] = (UP5(CC_SEL(cc, 11)) + 2 * UP5(CC_SEL(cc, 27))) / 3; |
| 832 | rgba[GCOMP] = (UP6(CC_SEL(cc, 5)) + 2 * UP6(CC_SEL(cc, 21))) / 3; |
| 833 | rgba[BCOMP] = (UP5(CC_SEL(cc, 0)) + 2 * UP5(CC_SEL(cc, 16))) / 3; |
| 834 | } |
| 835 | rgba[ACOMP] = UP4(src[((j & 3) * 4 + (i & 3)) / 2] >> ((i & 1) * 4)); |
| 836 | } |
| 837 | |
| 838 | |
| 839 | void TAPIENTRY |
| 840 | dxt5_rgba_decode_1 (const void *texture, int stride, |
| 841 | int i, int j, byte *rgba) |
| 842 | { |
| 843 | const byte *src = (const byte *)texture |
| 844 | + ((j / 4) * ((stride + 3) / 4) + i / 4) * 16; |
| 845 | const int code = (src[12 + (j & 3)] >> ((i & 3) * 2)) & 0x3; |
| 846 | const dword *cc = (const dword *)(src + 8); |
| 847 | const byte alpha0 = src[0]; |
| 848 | const byte alpha1 = src[1]; |
| 849 | const int alphaShift = (((j & 3) * 4) + (i & 3)) * 3 + 16; |
| 850 | const int acode = ((alphaShift == 31) |
| 851 | ? CC_SEL(src + 2, alphaShift - 16) |
| 852 | : CC_SEL(src, alphaShift)) & 0x7; |
| 853 | if (code == 0) { |
| 854 | rgba[RCOMP] = UP5(CC_SEL(cc, 11)); |
| 855 | rgba[GCOMP] = UP6(CC_SEL(cc, 5)); |
| 856 | rgba[BCOMP] = UP5(CC_SEL(cc, 0)); |
| 857 | } else if (code == 1) { |
| 858 | rgba[RCOMP] = UP5(CC_SEL(cc, 27)); |
| 859 | rgba[GCOMP] = UP6(CC_SEL(cc, 21)); |
| 860 | rgba[BCOMP] = UP5(CC_SEL(cc, 16)); |
| 861 | } else if (code == 2) { |
| 862 | /* (col0 * (4 - code) + col1 * (code - 1)) / 3 */ |
| 863 | rgba[RCOMP] = (UP5(CC_SEL(cc, 11)) * 2 + UP5(CC_SEL(cc, 27))) / 3; |
| 864 | rgba[GCOMP] = (UP6(CC_SEL(cc, 5)) * 2 + UP6(CC_SEL(cc, 21))) / 3; |
| 865 | rgba[BCOMP] = (UP5(CC_SEL(cc, 0)) * 2 + UP5(CC_SEL(cc, 16))) / 3; |
| 866 | } else { |
| 867 | rgba[RCOMP] = (UP5(CC_SEL(cc, 11)) + 2 * UP5(CC_SEL(cc, 27))) / 3; |
| 868 | rgba[GCOMP] = (UP6(CC_SEL(cc, 5)) + 2 * UP6(CC_SEL(cc, 21))) / 3; |
| 869 | rgba[BCOMP] = (UP5(CC_SEL(cc, 0)) + 2 * UP5(CC_SEL(cc, 16))) / 3; |
| 870 | } |
| 871 | if (acode == 0) { |
| 872 | rgba[ACOMP] = alpha0; |
| 873 | } else if (acode == 1) { |
| 874 | rgba[ACOMP] = alpha1; |
| 875 | } else if (alpha0 > alpha1) { |
| 876 | rgba[ACOMP] = ((8 - acode) * alpha0 + (acode - 1) * alpha1) / 7; |
| 877 | } else if (acode == 6) { |
| 878 | rgba[ACOMP] = 0; |
| 879 | } else if (acode == 7) { |
| 880 | rgba[ACOMP] = 255; |
| 881 | } else { |
| 882 | rgba[ACOMP] = ((6 - acode) * alpha0 + (acode - 1) * alpha1) / 5; |
| 883 | } |
| 884 | } |