| 1 | /* |
| 2 | * FXT1 codec |
| 3 | * Version: 1.1 |
| 4 | * |
| 5 | * Copyright (C) 2004 Daniel Borca All Rights Reserved. |
| 6 | * |
| 7 | * Permission is hereby granted, free of charge, to any person obtaining a |
| 8 | * copy of this software and associated documentation files (the "Software"), |
| 9 | * to deal in the Software without restriction, including without limitation |
| 10 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
| 11 | * and/or sell copies of the Software, and to permit persons to whom the |
| 12 | * Software is furnished to do so, subject to the following conditions: |
| 13 | * |
| 14 | * The above copyright notice and this permission notice shall be included |
| 15 | * in all copies or substantial portions of the Software. |
| 16 | * |
| 17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS |
| 18 | * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
| 19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
| 20 | * DANIEL BORCA BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN |
| 21 | * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN |
| 22 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. |
| 23 | */ |
| 24 | |
| 25 | /* Copyright (C) 2007 Hiroshi Morii <koolsmoky(at)users.sourceforge.net> |
| 26 | * Added support for ARGB inputs. |
| 27 | */ |
| 28 | |
| 29 | |
| 30 | #include <stdlib.h> |
| 31 | #include <string.h> |
| 32 | |
| 33 | #include "types.h" |
| 34 | #include "internal.h" |
| 35 | #include "fxt1.h" |
| 36 | |
| 37 | |
| 38 | /***************************************************************************\ |
| 39 | * FXT1 encoder |
| 40 | * |
| 41 | * The encoder was built by reversing the decoder, |
| 42 | * and is vaguely based on Texus2 by 3dfx. Note that this code |
| 43 | * is merely a proof of concept, since it is highly UNoptimized; |
| 44 | * moreover, it is sub-optimal due to initial conditions passed |
| 45 | * to Lloyd's algorithm (the interpolation modes are even worse). |
| 46 | \***************************************************************************/ |
| 47 | |
| 48 | |
| 49 | #define MAX_COMP 4 /* ever needed maximum number of components in texel */ |
| 50 | #define MAX_VECT 4 /* ever needed maximum number of base vectors to find */ |
| 51 | #define N_TEXELS 32 /* number of texels in a block (always 32) */ |
| 52 | #define LL_N_REP 50 /* number of iterations in lloyd's vq */ |
| 53 | #define LL_RMS_D 10 /* fault tolerance (maximum delta) */ |
| 54 | #define LL_RMS_E 255 /* fault tolerance (maximum error) */ |
| 55 | #define ALPHA_TS 2 /* alpha threshold: (255 - ALPHA_TS) deemed opaque */ |
| 56 | #define ISTBLACK(v) (*((dword *)(v)) == 0) |
| 57 | #define COPY_4UBV(DST, SRC) *((dword *)(DST)) = *((dword *)(SRC)) |
| 58 | |
| 59 | |
| 60 | static int |
| 61 | fxt1_bestcol (float vec[][MAX_COMP], int nv, |
| 62 | byte input[MAX_COMP], int nc) |
| 63 | { |
| 64 | int i, j, best = -1; |
| 65 | float err = 1e9; /* big enough */ |
| 66 | |
| 67 | for (j = 0; j < nv; j++) { |
| 68 | float e = 0.0F; |
| 69 | for (i = 0; i < nc; i++) { |
| 70 | e += (vec[j][i] - input[i]) * (vec[j][i] - input[i]); |
| 71 | } |
| 72 | if (e < err) { |
| 73 | err = e; |
| 74 | best = j; |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | return best; |
| 79 | } |
| 80 | |
| 81 | |
| 82 | static int |
| 83 | fxt1_worst (float vec[MAX_COMP], |
| 84 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
| 85 | { |
| 86 | int i, k, worst = -1; |
| 87 | float err = -1.0F; /* small enough */ |
| 88 | |
| 89 | for (k = 0; k < n; k++) { |
| 90 | float e = 0.0F; |
| 91 | for (i = 0; i < nc; i++) { |
| 92 | e += (vec[i] - input[k][i]) * (vec[i] - input[k][i]); |
| 93 | } |
| 94 | if (e > err) { |
| 95 | err = e; |
| 96 | worst = k; |
| 97 | } |
| 98 | } |
| 99 | |
| 100 | return worst; |
| 101 | } |
| 102 | |
| 103 | |
| 104 | static int |
| 105 | fxt1_variance (double variance[MAX_COMP], |
| 106 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
| 107 | { |
| 108 | int i, k, best = 0; |
| 109 | dword sx, sx2; |
| 110 | double var, maxvar = -1; /* small enough */ |
| 111 | double teenth = 1.0 / n; |
| 112 | |
| 113 | for (i = 0; i < nc; i++) { |
| 114 | sx = sx2 = 0; |
| 115 | for (k = 0; k < n; k++) { |
| 116 | int t = input[k][i]; |
| 117 | sx += t; |
| 118 | sx2 += t * t; |
| 119 | } |
| 120 | var = sx2 * teenth - sx * sx * teenth * teenth; |
| 121 | if (maxvar < var) { |
| 122 | maxvar = var; |
| 123 | best = i; |
| 124 | } |
| 125 | if (variance) { |
| 126 | variance[i] = var; |
| 127 | } |
| 128 | } |
| 129 | |
| 130 | return best; |
| 131 | } |
| 132 | |
| 133 | |
| 134 | static int |
| 135 | fxt1_choose (float vec[][MAX_COMP], int nv, |
| 136 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
| 137 | { |
| 138 | #if 0 |
| 139 | /* Choose colors from a grid. |
| 140 | */ |
| 141 | int i, j; |
| 142 | |
| 143 | for (j = 0; j < nv; j++) { |
| 144 | int m = j * (n - 1) / (nv - 1); |
| 145 | for (i = 0; i < nc; i++) { |
| 146 | vec[j][i] = input[m][i]; |
| 147 | } |
| 148 | } |
| 149 | #else |
| 150 | /* Our solution here is to find the darkest and brightest colors in |
| 151 | * the 8x4 tile and use those as the two representative colors. |
| 152 | * There are probably better algorithms to use (histogram-based). |
| 153 | */ |
| 154 | int i, j, k; |
| 155 | #ifndef YUV |
| 156 | int minSum = 2000; /* big enough */ |
| 157 | #else |
| 158 | int minSum = 2000000; |
| 159 | #endif |
| 160 | int maxSum = -1; /* small enough */ |
| 161 | int minCol = 0; /* phoudoin: silent compiler! */ |
| 162 | int maxCol = 0; /* phoudoin: silent compiler! */ |
| 163 | |
| 164 | struct { |
| 165 | int flag; |
| 166 | dword key; |
| 167 | int freq; |
| 168 | int idx; |
| 169 | } hist[N_TEXELS]; |
| 170 | int lenh = 0; |
| 171 | |
| 172 | memset(hist, 0, sizeof(hist)); |
| 173 | |
| 174 | for (k = 0; k < n; k++) { |
| 175 | int l; |
| 176 | dword key = 0; |
| 177 | int sum = 0; |
| 178 | for (i = 0; i < nc; i++) { |
| 179 | key <<= 8; |
| 180 | key |= input[k][i]; |
| 181 | #ifndef YUV |
| 182 | sum += input[k][i]; |
| 183 | #else |
| 184 | /* RGB to YUV conversion according to CCIR 601 specs |
| 185 | * Y = 0.299R+0.587G+0.114B |
| 186 | * U = 0.713(R - Y) = 0.500R-0.419G-0.081B |
| 187 | * V = 0.564(B - Y) = -0.169R-0.331G+0.500B |
| 188 | */ |
| 189 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 190 | #endif |
| 191 | } |
| 192 | for (l = 0; l < n; l++) { |
| 193 | if (!hist[l].flag) { |
| 194 | /* alloc new slot */ |
| 195 | hist[l].flag = !0; |
| 196 | hist[l].key = key; |
| 197 | hist[l].freq = 1; |
| 198 | hist[l].idx = k; |
| 199 | lenh = l + 1; |
| 200 | break; |
| 201 | } else if (hist[l].key == key) { |
| 202 | hist[l].freq++; |
| 203 | break; |
| 204 | } |
| 205 | } |
| 206 | if (minSum > sum) { |
| 207 | minSum = sum; |
| 208 | minCol = k; |
| 209 | } |
| 210 | if (maxSum < sum) { |
| 211 | maxSum = sum; |
| 212 | maxCol = k; |
| 213 | } |
| 214 | } |
| 215 | |
| 216 | if (lenh <= nv) { |
| 217 | for (j = 0; j < lenh; j++) { |
| 218 | for (i = 0; i < nc; i++) { |
| 219 | vec[j][i] = (float)input[hist[j].idx][i]; |
| 220 | } |
| 221 | } |
| 222 | for (; j < nv; j++) { |
| 223 | for (i = 0; i < nc; i++) { |
| 224 | vec[j][i] = vec[0][i]; |
| 225 | } |
| 226 | } |
| 227 | return 0; |
| 228 | } |
| 229 | |
| 230 | for (j = 0; j < nv; j++) { |
| 231 | for (i = 0; i < nc; i++) { |
| 232 | vec[j][i] = ((nv - 1 - j) * input[minCol][i] + j * input[maxCol][i] + (nv - 1) / 2) / (float)(nv - 1); |
| 233 | } |
| 234 | } |
| 235 | #endif |
| 236 | |
| 237 | return !0; |
| 238 | } |
| 239 | |
| 240 | |
| 241 | static int |
| 242 | fxt1_lloyd (float vec[][MAX_COMP], int nv, |
| 243 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
| 244 | { |
| 245 | /* Use the generalized lloyd's algorithm for VQ: |
| 246 | * find 4 color vectors. |
| 247 | * |
| 248 | * for each sample color |
| 249 | * sort to nearest vector. |
| 250 | * |
| 251 | * replace each vector with the centroid of it's matching colors. |
| 252 | * |
| 253 | * repeat until RMS doesn't improve. |
| 254 | * |
| 255 | * if a color vector has no samples, or becomes the same as another |
| 256 | * vector, replace it with the color which is farthest from a sample. |
| 257 | * |
| 258 | * vec[][MAX_COMP] initial vectors and resulting colors |
| 259 | * nv number of resulting colors required |
| 260 | * input[N_TEXELS][MAX_COMP] input texels |
| 261 | * nc number of components in input / vec |
| 262 | * n number of input samples |
| 263 | */ |
| 264 | |
| 265 | int sum[MAX_VECT][MAX_COMP]; /* used to accumulate closest texels */ |
| 266 | int cnt[MAX_VECT]; /* how many times a certain vector was chosen */ |
| 267 | float error, lasterror = 1e9; |
| 268 | |
| 269 | int i, j, k, rep; |
| 270 | |
| 271 | /* the quantizer */ |
| 272 | for (rep = 0; rep < LL_N_REP; rep++) { |
| 273 | /* reset sums & counters */ |
| 274 | for (j = 0; j < nv; j++) { |
| 275 | for (i = 0; i < nc; i++) { |
| 276 | sum[j][i] = 0; |
| 277 | } |
| 278 | cnt[j] = 0; |
| 279 | } |
| 280 | error = 0; |
| 281 | |
| 282 | /* scan whole block */ |
| 283 | for (k = 0; k < n; k++) { |
| 284 | #if 1 |
| 285 | int best = -1; |
| 286 | float err = 1e9; /* big enough */ |
| 287 | /* determine best vector */ |
| 288 | for (j = 0; j < nv; j++) { |
| 289 | float e = (vec[j][0] - input[k][0]) * (vec[j][0] - input[k][0]) + |
| 290 | (vec[j][1] - input[k][1]) * (vec[j][1] - input[k][1]) + |
| 291 | (vec[j][2] - input[k][2]) * (vec[j][2] - input[k][2]); |
| 292 | if (nc == 4) { |
| 293 | e += (vec[j][3] - input[k][3]) * (vec[j][3] - input[k][3]); |
| 294 | } |
| 295 | if (e < err) { |
| 296 | err = e; |
| 297 | best = j; |
| 298 | } |
| 299 | } |
| 300 | #else |
| 301 | int best = fxt1_bestcol(vec, nv, input[k], nc, &err); |
| 302 | #endif |
| 303 | /* add in closest color */ |
| 304 | for (i = 0; i < nc; i++) { |
| 305 | sum[best][i] += input[k][i]; |
| 306 | } |
| 307 | /* mark this vector as used */ |
| 308 | cnt[best]++; |
| 309 | /* accumulate error */ |
| 310 | error += err; |
| 311 | } |
| 312 | |
| 313 | /* check RMS */ |
| 314 | if ((error < LL_RMS_E) || |
| 315 | ((error < lasterror) && ((lasterror - error) < LL_RMS_D))) { |
| 316 | return !0; /* good match */ |
| 317 | } |
| 318 | lasterror = error; |
| 319 | |
| 320 | /* move each vector to the barycenter of its closest colors */ |
| 321 | for (j = 0; j < nv; j++) { |
| 322 | if (cnt[j]) { |
| 323 | float div = 1.0F / cnt[j]; |
| 324 | for (i = 0; i < nc; i++) { |
| 325 | vec[j][i] = div * sum[j][i]; |
| 326 | } |
| 327 | } else { |
| 328 | /* this vec has no samples or is identical with a previous vec */ |
| 329 | int worst = fxt1_worst(vec[j], input, nc, n); |
| 330 | for (i = 0; i < nc; i++) { |
| 331 | vec[j][i] = input[worst][i]; |
| 332 | } |
| 333 | } |
| 334 | } |
| 335 | } |
| 336 | |
| 337 | return 0; /* could not converge fast enough */ |
| 338 | } |
| 339 | |
| 340 | |
| 341 | static void |
| 342 | fxt1_quantize_CHROMA (dword *cc, |
| 343 | byte input[N_TEXELS][MAX_COMP]) |
| 344 | { |
| 345 | const int n_vect = 4; /* 4 base vectors to find */ |
| 346 | const int n_comp = 3; /* 3 components: R, G, B */ |
| 347 | float vec[MAX_VECT][MAX_COMP]; |
| 348 | int i, j, k; |
| 349 | qword hi; /* high quadword */ |
| 350 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
| 351 | |
| 352 | if (fxt1_choose(vec, n_vect, input, n_comp, N_TEXELS) != 0) { |
| 353 | fxt1_lloyd(vec, n_vect, input, n_comp, N_TEXELS); |
| 354 | } |
| 355 | |
| 356 | Q_MOV32(hi, 4); /* cc-chroma = "010" + unused bit */ |
| 357 | for (j = n_vect - 1; j >= 0; j--) { |
| 358 | for (i = 0; i < n_comp; i++) { |
| 359 | /* add in colors */ |
| 360 | Q_SHL(hi, 5); |
| 361 | Q_OR32(hi, (dword)(vec[j][i] / 8.0F)); |
| 362 | } |
| 363 | } |
| 364 | ((qword *)cc)[1] = hi; |
| 365 | |
| 366 | lohi = lolo = 0; |
| 367 | /* right microtile */ |
| 368 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
| 369 | lohi <<= 2; |
| 370 | lohi |= fxt1_bestcol(vec, n_vect, input[k], n_comp); |
| 371 | } |
| 372 | /* left microtile */ |
| 373 | for (; k >= 0; k--) { |
| 374 | lolo <<= 2; |
| 375 | lolo |= fxt1_bestcol(vec, n_vect, input[k], n_comp); |
| 376 | } |
| 377 | cc[1] = lohi; |
| 378 | cc[0] = lolo; |
| 379 | } |
| 380 | |
| 381 | |
| 382 | static void |
| 383 | fxt1_quantize_ALPHA0 (dword *cc, |
| 384 | byte input[N_TEXELS][MAX_COMP], |
| 385 | byte reord[N_TEXELS][MAX_COMP], int n) |
| 386 | { |
| 387 | const int n_vect = 3; /* 3 base vectors to find */ |
| 388 | const int n_comp = 4; /* 4 components: R, G, B, A */ |
| 389 | float vec[MAX_VECT][MAX_COMP]; |
| 390 | int i, j, k; |
| 391 | qword hi; /* high quadword */ |
| 392 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
| 393 | |
| 394 | /* the last vector indicates zero */ |
| 395 | for (i = 0; i < n_comp; i++) { |
| 396 | vec[n_vect][i] = 0; |
| 397 | } |
| 398 | |
| 399 | /* the first n texels in reord are guaranteed to be non-zero */ |
| 400 | if (fxt1_choose(vec, n_vect, reord, n_comp, n) != 0) { |
| 401 | fxt1_lloyd(vec, n_vect, reord, n_comp, n); |
| 402 | } |
| 403 | |
| 404 | Q_MOV32(hi, 6); /* alpha = "011" + lerp = 0 */ |
| 405 | for (j = n_vect - 1; j >= 0; j--) { |
| 406 | /* add in alphas */ |
| 407 | Q_SHL(hi, 5); |
| 408 | Q_OR32(hi, (dword)(vec[j][ACOMP] / 8.0F)); |
| 409 | } |
| 410 | for (j = n_vect - 1; j >= 0; j--) { |
| 411 | for (i = 0; i < n_comp - 1; i++) { |
| 412 | /* add in colors */ |
| 413 | Q_SHL(hi, 5); |
| 414 | Q_OR32(hi, (dword)(vec[j][i] / 8.0F)); |
| 415 | } |
| 416 | } |
| 417 | ((qword *)cc)[1] = hi; |
| 418 | |
| 419 | lohi = lolo = 0; |
| 420 | /* right microtile */ |
| 421 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
| 422 | lohi <<= 2; |
| 423 | lohi |= fxt1_bestcol(vec, n_vect + 1, input[k], n_comp); |
| 424 | } |
| 425 | /* left microtile */ |
| 426 | for (; k >= 0; k--) { |
| 427 | lolo <<= 2; |
| 428 | lolo |= fxt1_bestcol(vec, n_vect + 1, input[k], n_comp); |
| 429 | } |
| 430 | cc[1] = lohi; |
| 431 | cc[0] = lolo; |
| 432 | } |
| 433 | |
| 434 | |
| 435 | static void |
| 436 | fxt1_quantize_ALPHA1 (dword *cc, |
| 437 | byte input[N_TEXELS][MAX_COMP]) |
| 438 | { |
| 439 | const int n_vect = 3; /* highest vector number in each microtile */ |
| 440 | const int n_comp = 4; /* 4 components: R, G, B, A */ |
| 441 | float vec[1 + 1 + 1][MAX_COMP]; /* 1.5 extrema for each sub-block */ |
| 442 | float b, iv[MAX_COMP]; /* interpolation vector */ |
| 443 | int i, j, k; |
| 444 | qword hi; /* high quadword */ |
| 445 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
| 446 | |
| 447 | int minSum; |
| 448 | int maxSum; |
| 449 | int minColL = 0, maxColL = 0; |
| 450 | int minColR = 0, maxColR = 0; |
| 451 | int sumL = 0, sumR = 0; |
| 452 | |
| 453 | /* Our solution here is to find the darkest and brightest colors in |
| 454 | * the 4x4 tile and use those as the two representative colors. |
| 455 | * There are probably better algorithms to use (histogram-based). |
| 456 | */ |
| 457 | #ifndef YUV |
| 458 | minSum = 2000; /* big enough */ |
| 459 | #else |
| 460 | minSum = 2000000; |
| 461 | #endif |
| 462 | maxSum = -1; /* small enough */ |
| 463 | for (k = 0; k < N_TEXELS / 2; k++) { |
| 464 | int sum = 0; |
| 465 | #ifndef YUV |
| 466 | for (i = 0; i < n_comp; i++) { |
| 467 | sum += input[k][i]; |
| 468 | } |
| 469 | #else |
| 470 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 471 | #endif |
| 472 | if (minSum > sum) { |
| 473 | minSum = sum; |
| 474 | minColL = k; |
| 475 | } |
| 476 | if (maxSum < sum) { |
| 477 | maxSum = sum; |
| 478 | maxColL = k; |
| 479 | } |
| 480 | sumL += sum; |
| 481 | } |
| 482 | #ifndef YUV |
| 483 | minSum = 2000; /* big enough */ |
| 484 | #else |
| 485 | minSum = 2000000; |
| 486 | #endif |
| 487 | maxSum = -1; /* small enough */ |
| 488 | for (; k < N_TEXELS; k++) { |
| 489 | int sum = 0; |
| 490 | #ifndef YUV |
| 491 | for (i = 0; i < n_comp; i++) { |
| 492 | sum += input[k][i]; |
| 493 | } |
| 494 | #else |
| 495 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 496 | #endif |
| 497 | if (minSum > sum) { |
| 498 | minSum = sum; |
| 499 | minColR = k; |
| 500 | } |
| 501 | if (maxSum < sum) { |
| 502 | maxSum = sum; |
| 503 | maxColR = k; |
| 504 | } |
| 505 | sumR += sum; |
| 506 | } |
| 507 | |
| 508 | /* choose the common vector (yuck!) */ |
| 509 | { |
| 510 | int j1, j2; |
| 511 | int v1 = 0, v2 = 0; |
| 512 | float err = 1e9; /* big enough */ |
| 513 | float tv[2 * 2][MAX_COMP]; /* 2 extrema for each sub-block */ |
| 514 | for (i = 0; i < n_comp; i++) { |
| 515 | tv[0][i] = input[minColL][i]; |
| 516 | tv[1][i] = input[maxColL][i]; |
| 517 | tv[2][i] = input[minColR][i]; |
| 518 | tv[3][i] = input[maxColR][i]; |
| 519 | } |
| 520 | for (j1 = 0; j1 < 2; j1++) { |
| 521 | for (j2 = 2; j2 < 4; j2++) { |
| 522 | float e = 0.0F; |
| 523 | for (i = 0; i < n_comp; i++) { |
| 524 | e += (tv[j1][i] - tv[j2][i]) * (tv[j1][i] - tv[j2][i]); |
| 525 | } |
| 526 | if (e < err) { |
| 527 | err = e; |
| 528 | v1 = j1; |
| 529 | v2 = j2; |
| 530 | } |
| 531 | } |
| 532 | } |
| 533 | for (i = 0; i < n_comp; i++) { |
| 534 | vec[0][i] = tv[1 - v1][i]; |
| 535 | vec[1][i] = (tv[v1][i] * sumL + tv[v2][i] * sumR) / (sumL + sumR); |
| 536 | vec[2][i] = tv[5 - v2][i]; |
| 537 | } |
| 538 | } |
| 539 | |
| 540 | /* left microtile */ |
| 541 | cc[0] = 0; |
| 542 | if (minColL != maxColL) { |
| 543 | /* compute interpolation vector */ |
| 544 | MAKEIVEC(n_vect, n_comp, iv, b, vec[0], vec[1]); |
| 545 | |
| 546 | /* add in texels */ |
| 547 | lolo = 0; |
| 548 | for (k = N_TEXELS / 2 - 1; k >= 0; k--) { |
| 549 | int texel; |
| 550 | /* interpolate color */ |
| 551 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 552 | /* add in texel */ |
| 553 | lolo <<= 2; |
| 554 | lolo |= texel; |
| 555 | } |
| 556 | |
| 557 | cc[0] = lolo; |
| 558 | } |
| 559 | |
| 560 | /* right microtile */ |
| 561 | cc[1] = 0; |
| 562 | if (minColR != maxColR) { |
| 563 | /* compute interpolation vector */ |
| 564 | MAKEIVEC(n_vect, n_comp, iv, b, vec[2], vec[1]); |
| 565 | |
| 566 | /* add in texels */ |
| 567 | lohi = 0; |
| 568 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
| 569 | int texel; |
| 570 | /* interpolate color */ |
| 571 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 572 | /* add in texel */ |
| 573 | lohi <<= 2; |
| 574 | lohi |= texel; |
| 575 | } |
| 576 | |
| 577 | cc[1] = lohi; |
| 578 | } |
| 579 | |
| 580 | Q_MOV32(hi, 7); /* alpha = "011" + lerp = 1 */ |
| 581 | for (j = n_vect - 1; j >= 0; j--) { |
| 582 | /* add in alphas */ |
| 583 | Q_SHL(hi, 5); |
| 584 | Q_OR32(hi, (dword)(vec[j][ACOMP] / 8.0F)); |
| 585 | } |
| 586 | for (j = n_vect - 1; j >= 0; j--) { |
| 587 | for (i = 0; i < n_comp - 1; i++) { |
| 588 | /* add in colors */ |
| 589 | Q_SHL(hi, 5); |
| 590 | Q_OR32(hi, (dword)(vec[j][i] / 8.0F)); |
| 591 | } |
| 592 | } |
| 593 | ((qword *)cc)[1] = hi; |
| 594 | } |
| 595 | |
| 596 | |
| 597 | static void |
| 598 | fxt1_quantize_HI (dword *cc, |
| 599 | byte input[N_TEXELS][MAX_COMP], |
| 600 | byte reord[N_TEXELS][MAX_COMP], int n) |
| 601 | { |
| 602 | const int n_vect = 6; /* highest vector number */ |
| 603 | const int n_comp = 3; /* 3 components: R, G, B */ |
| 604 | float b = 0.0F; /* phoudoin: silent compiler! */ |
| 605 | float iv[MAX_COMP]; /* interpolation vector */ |
| 606 | int i, k; |
| 607 | dword hihi; /* high quadword: hi dword */ |
| 608 | |
| 609 | #ifndef YUV |
| 610 | int minSum = 2000; /* big enough */ |
| 611 | #else |
| 612 | int minSum = 2000000; |
| 613 | #endif |
| 614 | int maxSum = -1; /* small enough */ |
| 615 | int minCol = 0; /* phoudoin: silent compiler! */ |
| 616 | int maxCol = 0; /* phoudoin: silent compiler! */ |
| 617 | |
| 618 | /* Our solution here is to find the darkest and brightest colors in |
| 619 | * the 8x4 tile and use those as the two representative colors. |
| 620 | * There are probably better algorithms to use (histogram-based). |
| 621 | */ |
| 622 | for (k = 0; k < n; k++) { |
| 623 | int sum = 0; |
| 624 | #ifndef YUV |
| 625 | for (i = 0; i < n_comp; i++) { |
| 626 | sum += reord[k][i]; |
| 627 | } |
| 628 | #else |
| 629 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 630 | #endif |
| 631 | if (minSum > sum) { |
| 632 | minSum = sum; |
| 633 | minCol = k; |
| 634 | } |
| 635 | if (maxSum < sum) { |
| 636 | maxSum = sum; |
| 637 | maxCol = k; |
| 638 | } |
| 639 | } |
| 640 | |
| 641 | hihi = 0; /* cc-hi = "00" */ |
| 642 | for (i = 0; i < n_comp; i++) { |
| 643 | /* add in colors */ |
| 644 | hihi <<= 5; |
| 645 | hihi |= reord[maxCol][i] >> 3; |
| 646 | } |
| 647 | for (i = 0; i < n_comp; i++) { |
| 648 | /* add in colors */ |
| 649 | hihi <<= 5; |
| 650 | hihi |= reord[minCol][i] >> 3; |
| 651 | } |
| 652 | cc[3] = hihi; |
| 653 | cc[0] = cc[1] = cc[2] = 0; |
| 654 | |
| 655 | /* compute interpolation vector */ |
| 656 | if (minCol != maxCol) { |
| 657 | MAKEIVEC(n_vect, n_comp, iv, b, reord[minCol], reord[maxCol]); |
| 658 | } |
| 659 | |
| 660 | /* add in texels */ |
| 661 | for (k = N_TEXELS - 1; k >= 0; k--) { |
| 662 | int t = k * 3; |
| 663 | dword *kk = (dword *)((byte *)cc + t / 8); |
| 664 | int texel = n_vect + 1; /* transparent black */ |
| 665 | |
| 666 | if (!ISTBLACK(input[k])) { |
| 667 | if (minCol != maxCol) { |
| 668 | /* interpolate color */ |
| 669 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 670 | /* add in texel */ |
| 671 | kk[0] |= texel << (t & 7); |
| 672 | } |
| 673 | } else { |
| 674 | /* add in texel */ |
| 675 | kk[0] |= texel << (t & 7); |
| 676 | } |
| 677 | } |
| 678 | } |
| 679 | |
| 680 | |
| 681 | static void |
| 682 | fxt1_quantize_MIXED1 (dword *cc, |
| 683 | byte input[N_TEXELS][MAX_COMP]) |
| 684 | { |
| 685 | const int n_vect = 2; /* highest vector number in each microtile */ |
| 686 | const int n_comp = 3; /* 3 components: R, G, B */ |
| 687 | byte vec[2 * 2][MAX_COMP]; /* 2 extrema for each sub-block */ |
| 688 | float b, iv[MAX_COMP]; /* interpolation vector */ |
| 689 | int i, j, k; |
| 690 | qword hi; /* high quadword */ |
| 691 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
| 692 | |
| 693 | int minSum; |
| 694 | int maxSum; |
| 695 | int minColL = 0, maxColL = -1; |
| 696 | int minColR = 0, maxColR = -1; |
| 697 | |
| 698 | /* Our solution here is to find the darkest and brightest colors in |
| 699 | * the 4x4 tile and use those as the two representative colors. |
| 700 | * There are probably better algorithms to use (histogram-based). |
| 701 | */ |
| 702 | #ifndef YUV |
| 703 | minSum = 2000; /* big enough */ |
| 704 | #else |
| 705 | minSum = 2000000; |
| 706 | #endif |
| 707 | maxSum = -1; /* small enough */ |
| 708 | for (k = 0; k < N_TEXELS / 2; k++) { |
| 709 | if (!ISTBLACK(input[k])) { |
| 710 | int sum = 0; |
| 711 | #ifndef YUV |
| 712 | for (i = 0; i < n_comp; i++) { |
| 713 | sum += input[k][i]; |
| 714 | } |
| 715 | #else |
| 716 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 717 | #endif |
| 718 | if (minSum > sum) { |
| 719 | minSum = sum; |
| 720 | minColL = k; |
| 721 | } |
| 722 | if (maxSum < sum) { |
| 723 | maxSum = sum; |
| 724 | maxColL = k; |
| 725 | } |
| 726 | } |
| 727 | } |
| 728 | #ifndef YUV |
| 729 | minSum = 2000; /* big enough */ |
| 730 | #else |
| 731 | minSum = 2000000; |
| 732 | #endif |
| 733 | maxSum = -1; /* small enough */ |
| 734 | for (; k < N_TEXELS; k++) { |
| 735 | if (!ISTBLACK(input[k])) { |
| 736 | int sum = 0; |
| 737 | #ifndef YUV |
| 738 | for (i = 0; i < n_comp; i++) { |
| 739 | sum += input[k][i]; |
| 740 | } |
| 741 | #else |
| 742 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 743 | #endif |
| 744 | if (minSum > sum) { |
| 745 | minSum = sum; |
| 746 | minColR = k; |
| 747 | } |
| 748 | if (maxSum < sum) { |
| 749 | maxSum = sum; |
| 750 | maxColR = k; |
| 751 | } |
| 752 | } |
| 753 | } |
| 754 | |
| 755 | /* left microtile */ |
| 756 | if (maxColL == -1) { |
| 757 | /* all transparent black */ |
| 758 | cc[0] = 0xFFFFFFFF; |
| 759 | for (i = 0; i < n_comp; i++) { |
| 760 | vec[0][i] = 0; |
| 761 | vec[1][i] = 0; |
| 762 | } |
| 763 | } else { |
| 764 | cc[0] = 0; |
| 765 | for (i = 0; i < n_comp; i++) { |
| 766 | vec[0][i] = input[minColL][i]; |
| 767 | vec[1][i] = input[maxColL][i]; |
| 768 | } |
| 769 | if (minColL != maxColL) { |
| 770 | /* compute interpolation vector */ |
| 771 | MAKEIVEC(n_vect, n_comp, iv, b, vec[0], vec[1]); |
| 772 | |
| 773 | /* add in texels */ |
| 774 | lolo = 0; |
| 775 | for (k = N_TEXELS / 2 - 1; k >= 0; k--) { |
| 776 | int texel = n_vect + 1; /* transparent black */ |
| 777 | if (!ISTBLACK(input[k])) { |
| 778 | /* interpolate color */ |
| 779 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 780 | } |
| 781 | /* add in texel */ |
| 782 | lolo <<= 2; |
| 783 | lolo |= texel; |
| 784 | } |
| 785 | cc[0] = lolo; |
| 786 | } |
| 787 | } |
| 788 | |
| 789 | /* right microtile */ |
| 790 | if (maxColR == -1) { |
| 791 | /* all transparent black */ |
| 792 | cc[1] = 0xFFFFFFFF; |
| 793 | for (i = 0; i < n_comp; i++) { |
| 794 | vec[2][i] = 0; |
| 795 | vec[3][i] = 0; |
| 796 | } |
| 797 | } else { |
| 798 | cc[1] = 0; |
| 799 | for (i = 0; i < n_comp; i++) { |
| 800 | vec[2][i] = input[minColR][i]; |
| 801 | vec[3][i] = input[maxColR][i]; |
| 802 | } |
| 803 | if (minColR != maxColR) { |
| 804 | /* compute interpolation vector */ |
| 805 | MAKEIVEC(n_vect, n_comp, iv, b, vec[2], vec[3]); |
| 806 | |
| 807 | /* add in texels */ |
| 808 | lohi = 0; |
| 809 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
| 810 | int texel = n_vect + 1; /* transparent black */ |
| 811 | if (!ISTBLACK(input[k])) { |
| 812 | /* interpolate color */ |
| 813 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 814 | } |
| 815 | /* add in texel */ |
| 816 | lohi <<= 2; |
| 817 | lohi |= texel; |
| 818 | } |
| 819 | cc[1] = lohi; |
| 820 | } |
| 821 | } |
| 822 | |
| 823 | Q_MOV32(hi, 9 | (vec[3][GCOMP] & 4) | ((vec[1][GCOMP] >> 1) & 2)); /* chroma = "1" */ |
| 824 | for (j = 2 * 2 - 1; j >= 0; j--) { |
| 825 | for (i = 0; i < n_comp; i++) { |
| 826 | /* add in colors */ |
| 827 | Q_SHL(hi, 5); |
| 828 | Q_OR32(hi, vec[j][i] >> 3); |
| 829 | } |
| 830 | } |
| 831 | ((qword *)cc)[1] = hi; |
| 832 | } |
| 833 | |
| 834 | |
| 835 | static void |
| 836 | fxt1_quantize_MIXED0 (dword *cc, |
| 837 | byte input[N_TEXELS][MAX_COMP]) |
| 838 | { |
| 839 | const int n_vect = 3; /* highest vector number in each microtile */ |
| 840 | const int n_comp = 3; /* 3 components: R, G, B */ |
| 841 | byte vec[2 * 2][MAX_COMP]; /* 2 extrema for each sub-block */ |
| 842 | float b, iv[MAX_COMP]; /* interpolation vector */ |
| 843 | int i, j, k; |
| 844 | qword hi; /* high quadword */ |
| 845 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
| 846 | |
| 847 | int minColL = 0, maxColL = 0; |
| 848 | int minColR = 0, maxColR = 0; |
| 849 | #if 0 |
| 850 | int minSum; |
| 851 | int maxSum; |
| 852 | |
| 853 | /* Our solution here is to find the darkest and brightest colors in |
| 854 | * the 4x4 tile and use those as the two representative colors. |
| 855 | * There are probably better algorithms to use (histogram-based). |
| 856 | */ |
| 857 | #ifndef YUV |
| 858 | minSum = 2000; /* big enough */ |
| 859 | #else |
| 860 | minSum = 2000000; |
| 861 | #endif |
| 862 | maxSum = -1; /* small enough */ |
| 863 | for (k = 0; k < N_TEXELS / 2; k++) { |
| 864 | int sum = 0; |
| 865 | #ifndef YUV |
| 866 | for (i = 0; i < n_comp; i++) { |
| 867 | sum += input[k][i]; |
| 868 | } |
| 869 | #else |
| 870 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 871 | #endif |
| 872 | if (minSum > sum) { |
| 873 | minSum = sum; |
| 874 | minColL = k; |
| 875 | } |
| 876 | if (maxSum < sum) { |
| 877 | maxSum = sum; |
| 878 | maxColL = k; |
| 879 | } |
| 880 | } |
| 881 | minSum = 2000; /* big enough */ |
| 882 | maxSum = -1; /* small enough */ |
| 883 | for (; k < N_TEXELS; k++) { |
| 884 | int sum = 0; |
| 885 | #ifndef YUV |
| 886 | for (i = 0; i < n_comp; i++) { |
| 887 | sum += input[k][i]; |
| 888 | } |
| 889 | #else |
| 890 | sum = 299 * input[k][RCOMP] + 587 * input[k][GCOMP] + 114 * input[k][BCOMP]; |
| 891 | #endif |
| 892 | if (minSum > sum) { |
| 893 | minSum = sum; |
| 894 | minColR = k; |
| 895 | } |
| 896 | if (maxSum < sum) { |
| 897 | maxSum = sum; |
| 898 | maxColR = k; |
| 899 | } |
| 900 | } |
| 901 | #else |
| 902 | int minVal; |
| 903 | int maxVal; |
| 904 | int maxVarL = fxt1_variance(NULL, input, n_comp, N_TEXELS / 2); |
| 905 | int maxVarR = fxt1_variance(NULL, &input[N_TEXELS / 2], n_comp, N_TEXELS / 2); |
| 906 | |
| 907 | /* Scan the channel with max variance for lo & hi |
| 908 | * and use those as the two representative colors. |
| 909 | */ |
| 910 | minVal = 2000; /* big enough */ |
| 911 | maxVal = -1; /* small enough */ |
| 912 | for (k = 0; k < N_TEXELS / 2; k++) { |
| 913 | int t = input[k][maxVarL]; |
| 914 | if (minVal > t) { |
| 915 | minVal = t; |
| 916 | minColL = k; |
| 917 | } |
| 918 | if (maxVal < t) { |
| 919 | maxVal = t; |
| 920 | maxColL = k; |
| 921 | } |
| 922 | } |
| 923 | minVal = 2000; /* big enough */ |
| 924 | maxVal = -1; /* small enough */ |
| 925 | for (; k < N_TEXELS; k++) { |
| 926 | int t = input[k][maxVarR]; |
| 927 | if (minVal > t) { |
| 928 | minVal = t; |
| 929 | minColR = k; |
| 930 | } |
| 931 | if (maxVal < t) { |
| 932 | maxVal = t; |
| 933 | maxColR = k; |
| 934 | } |
| 935 | } |
| 936 | #endif |
| 937 | |
| 938 | /* left microtile */ |
| 939 | cc[0] = 0; |
| 940 | for (i = 0; i < n_comp; i++) { |
| 941 | vec[0][i] = input[minColL][i]; |
| 942 | vec[1][i] = input[maxColL][i]; |
| 943 | } |
| 944 | if (minColL != maxColL) { |
| 945 | /* compute interpolation vector */ |
| 946 | MAKEIVEC(n_vect, n_comp, iv, b, vec[0], vec[1]); |
| 947 | |
| 948 | /* add in texels */ |
| 949 | lolo = 0; |
| 950 | for (k = N_TEXELS / 2 - 1; k >= 0; k--) { |
| 951 | int texel; |
| 952 | /* interpolate color */ |
| 953 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 954 | /* add in texel */ |
| 955 | lolo <<= 2; |
| 956 | lolo |= texel; |
| 957 | } |
| 958 | |
| 959 | /* funky encoding for LSB of green */ |
| 960 | if ((int)((lolo >> 1) & 1) != (((vec[1][GCOMP] ^ vec[0][GCOMP]) >> 2) & 1)) { |
| 961 | for (i = 0; i < n_comp; i++) { |
| 962 | vec[1][i] = input[minColL][i]; |
| 963 | vec[0][i] = input[maxColL][i]; |
| 964 | } |
| 965 | lolo = ~lolo; |
| 966 | } |
| 967 | |
| 968 | cc[0] = lolo; |
| 969 | } |
| 970 | |
| 971 | /* right microtile */ |
| 972 | cc[1] = 0; |
| 973 | for (i = 0; i < n_comp; i++) { |
| 974 | vec[2][i] = input[minColR][i]; |
| 975 | vec[3][i] = input[maxColR][i]; |
| 976 | } |
| 977 | if (minColR != maxColR) { |
| 978 | /* compute interpolation vector */ |
| 979 | MAKEIVEC(n_vect, n_comp, iv, b, vec[2], vec[3]); |
| 980 | |
| 981 | /* add in texels */ |
| 982 | lohi = 0; |
| 983 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
| 984 | int texel; |
| 985 | /* interpolate color */ |
| 986 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
| 987 | /* add in texel */ |
| 988 | lohi <<= 2; |
| 989 | lohi |= texel; |
| 990 | } |
| 991 | |
| 992 | /* funky encoding for LSB of green */ |
| 993 | if ((int)((lohi >> 1) & 1) != (((vec[3][GCOMP] ^ vec[2][GCOMP]) >> 2) & 1)) { |
| 994 | for (i = 0; i < n_comp; i++) { |
| 995 | vec[3][i] = input[minColR][i]; |
| 996 | vec[2][i] = input[maxColR][i]; |
| 997 | } |
| 998 | lohi = ~lohi; |
| 999 | } |
| 1000 | |
| 1001 | cc[1] = lohi; |
| 1002 | } |
| 1003 | |
| 1004 | Q_MOV32(hi, 8 | (vec[3][GCOMP] & 4) | ((vec[1][GCOMP] >> 1) & 2)); /* chroma = "1" */ |
| 1005 | for (j = 2 * 2 - 1; j >= 0; j--) { |
| 1006 | for (i = 0; i < n_comp; i++) { |
| 1007 | /* add in colors */ |
| 1008 | Q_SHL(hi, 5); |
| 1009 | Q_OR32(hi, vec[j][i] >> 3); |
| 1010 | } |
| 1011 | } |
| 1012 | ((qword *)cc)[1] = hi; |
| 1013 | } |
| 1014 | |
| 1015 | |
| 1016 | static void |
| 1017 | fxt1_quantize (dword *cc, const byte *lines[], int comps) |
| 1018 | { |
| 1019 | int trualpha; |
| 1020 | byte reord[N_TEXELS][MAX_COMP]; |
| 1021 | |
| 1022 | byte input[N_TEXELS][MAX_COMP]; |
| 1023 | #ifndef ARGB |
| 1024 | int i; |
| 1025 | #endif |
| 1026 | int k, l; |
| 1027 | |
| 1028 | if (comps == 3) { |
| 1029 | /* make the whole block opaque */ |
| 1030 | memset(input, -1, sizeof(input)); |
| 1031 | } |
| 1032 | |
| 1033 | /* 8 texels each line */ |
| 1034 | #ifndef ARGB |
| 1035 | for (l = 0; l < 4; l++) { |
| 1036 | for (k = 0; k < 4; k++) { |
| 1037 | for (i = 0; i < comps; i++) { |
| 1038 | input[k + l * 4][i] = *lines[l]++; |
| 1039 | } |
| 1040 | } |
| 1041 | for (; k < 8; k++) { |
| 1042 | for (i = 0; i < comps; i++) { |
| 1043 | input[k + l * 4 + 12][i] = *lines[l]++; |
| 1044 | } |
| 1045 | } |
| 1046 | } |
| 1047 | #else |
| 1048 | /* H.Morii - support for ARGB inputs */ |
| 1049 | for (l = 0; l < 4; l++) { |
| 1050 | for (k = 0; k < 4; k++) { |
| 1051 | input[k + l * 4][2] = *lines[l]++; |
| 1052 | input[k + l * 4][1] = *lines[l]++; |
| 1053 | input[k + l * 4][0] = *lines[l]++; |
| 1054 | if (comps == 4) input[k + l * 4][3] = *lines[l]++; |
| 1055 | } |
| 1056 | for (; k < 8; k++) { |
| 1057 | input[k + l * 4 + 12][2] = *lines[l]++; |
| 1058 | input[k + l * 4 + 12][1] = *lines[l]++; |
| 1059 | input[k + l * 4 + 12][0] = *lines[l]++; |
| 1060 | if (comps == 4) input[k + l * 4 + 12][3] = *lines[l]++; |
| 1061 | } |
| 1062 | } |
| 1063 | #endif |
| 1064 | |
| 1065 | /* block layout: |
| 1066 | * 00, 01, 02, 03, 08, 09, 0a, 0b |
| 1067 | * 10, 11, 12, 13, 18, 19, 1a, 1b |
| 1068 | * 04, 05, 06, 07, 0c, 0d, 0e, 0f |
| 1069 | * 14, 15, 16, 17, 1c, 1d, 1e, 1f |
| 1070 | */ |
| 1071 | |
| 1072 | /* [dBorca] |
| 1073 | * stupidity flows forth from this |
| 1074 | */ |
| 1075 | l = N_TEXELS; |
| 1076 | trualpha = 0; |
| 1077 | if (comps == 4) { |
| 1078 | /* skip all transparent black texels */ |
| 1079 | l = 0; |
| 1080 | for (k = 0; k < N_TEXELS; k++) { |
| 1081 | /* test all components against 0 */ |
| 1082 | if (!ISTBLACK(input[k])) { |
| 1083 | /* texel is not transparent black */ |
| 1084 | COPY_4UBV(reord[l], input[k]); |
| 1085 | if (reord[l][ACOMP] < (255 - ALPHA_TS)) { |
| 1086 | /* non-opaque texel */ |
| 1087 | trualpha = !0; |
| 1088 | } |
| 1089 | l++; |
| 1090 | } |
| 1091 | } |
| 1092 | } |
| 1093 | |
| 1094 | #if 0 |
| 1095 | if (trualpha) { |
| 1096 | fxt1_quantize_ALPHA0(cc, input, reord, l); |
| 1097 | } else if (l == 0) { |
| 1098 | cc[0] = cc[1] = cc[2] = -1; |
| 1099 | cc[3] = 0; |
| 1100 | } else if (l < N_TEXELS) { |
| 1101 | fxt1_quantize_HI(cc, input, reord, l); |
| 1102 | } else { |
| 1103 | fxt1_quantize_CHROMA(cc, input); |
| 1104 | } |
| 1105 | (void)fxt1_quantize_ALPHA1; |
| 1106 | (void)fxt1_quantize_MIXED1; |
| 1107 | (void)fxt1_quantize_MIXED0; |
| 1108 | #else |
| 1109 | if (trualpha) { |
| 1110 | fxt1_quantize_ALPHA1(cc, input); |
| 1111 | } else if (l == 0) { |
| 1112 | cc[0] = cc[1] = cc[2] = 0xFFFFFFFF; |
| 1113 | cc[3] = 0; |
| 1114 | } else if (l < N_TEXELS) { |
| 1115 | fxt1_quantize_MIXED1(cc, input); |
| 1116 | } else { |
| 1117 | fxt1_quantize_MIXED0(cc, input); |
| 1118 | } |
| 1119 | (void)fxt1_quantize_ALPHA0; |
| 1120 | (void)fxt1_quantize_HI; |
| 1121 | (void)fxt1_quantize_CHROMA; |
| 1122 | #endif |
| 1123 | } |
| 1124 | |
| 1125 | |
| 1126 | TAPI int TAPIENTRY |
| 1127 | fxt1_encode (int width, int height, int comps, |
| 1128 | const void *source, int srcRowStride, |
| 1129 | void *dest, int destRowStride) |
| 1130 | { |
| 1131 | int x, y; |
| 1132 | const byte *data; |
| 1133 | dword *encoded = (dword *)dest; |
| 1134 | void *newSource = NULL; |
| 1135 | |
| 1136 | /* Replicate image if width is not M8 or height is not M4 */ |
| 1137 | if ((width & 7) | (height & 3)) { |
| 1138 | int newWidth = (width + 7) & ~7; |
| 1139 | int newHeight = (height + 3) & ~3; |
| 1140 | newSource = malloc(comps * newWidth * newHeight * sizeof(byte *)); |
| 1141 | _mesa_upscale_teximage2d(width, height, newWidth, newHeight, |
| 1142 | comps, (const byte *)source, |
| 1143 | srcRowStride, (byte *)newSource); |
| 1144 | source = newSource; |
| 1145 | width = newWidth; |
| 1146 | height = newHeight; |
| 1147 | srcRowStride = comps * newWidth; |
| 1148 | } |
| 1149 | |
| 1150 | data = (const byte *)source; |
| 1151 | destRowStride = (destRowStride - width * 2) / 4; |
| 1152 | for (y = 0; y < height; y += 4) { |
| 1153 | unsigned int offs = 0 + (y + 0) * srcRowStride; |
| 1154 | for (x = 0; x < width; x += 8) { |
| 1155 | const byte *lines[4]; |
| 1156 | lines[0] = &data[offs]; |
| 1157 | lines[1] = lines[0] + srcRowStride; |
| 1158 | lines[2] = lines[1] + srcRowStride; |
| 1159 | lines[3] = lines[2] + srcRowStride; |
| 1160 | offs += 8 * comps; |
| 1161 | fxt1_quantize(encoded, lines, comps); |
| 1162 | /* 128 bits per 8x4 block */ |
| 1163 | encoded += 4; |
| 1164 | } |
| 1165 | encoded += destRowStride; |
| 1166 | } |
| 1167 | |
| 1168 | if (newSource != NULL) { |
| 1169 | free(newSource); |
| 1170 | } |
| 1171 | |
| 1172 | return 0; |
| 1173 | } |
| 1174 | |
| 1175 | |
| 1176 | /***************************************************************************\ |
| 1177 | * FXT1 decoder |
| 1178 | * |
| 1179 | * The decoder is based on GL_3DFX_texture_compression_FXT1 |
| 1180 | * specification and serves as a concept for the encoder. |
| 1181 | \***************************************************************************/ |
| 1182 | |
| 1183 | |
| 1184 | /* lookup table for scaling 5 bit colors up to 8 bits */ |
| 1185 | static const byte _rgb_scale_5[] = { |
| 1186 | 0, 8, 16, 25, 33, 41, 49, 58, |
| 1187 | 66, 74, 82, 90, 99, 107, 115, 123, |
| 1188 | 132, 140, 148, 156, 165, 173, 181, 189, |
| 1189 | 197, 206, 214, 222, 230, 239, 247, 255 |
| 1190 | }; |
| 1191 | |
| 1192 | /* lookup table for scaling 6 bit colors up to 8 bits */ |
| 1193 | static const byte _rgb_scale_6[] = { |
| 1194 | 0, 4, 8, 12, 16, 20, 24, 28, |
| 1195 | 32, 36, 40, 45, 49, 53, 57, 61, |
| 1196 | 65, 69, 73, 77, 81, 85, 89, 93, |
| 1197 | 97, 101, 105, 109, 113, 117, 121, 125, |
| 1198 | 130, 134, 138, 142, 146, 150, 154, 158, |
| 1199 | 162, 166, 170, 174, 178, 182, 186, 190, |
| 1200 | 194, 198, 202, 206, 210, 215, 219, 223, |
| 1201 | 227, 231, 235, 239, 243, 247, 251, 255 |
| 1202 | }; |
| 1203 | |
| 1204 | |
| 1205 | #define CC_SEL(cc, which) (((dword *)(cc))[(which) / 32] >> ((which) & 31)) |
| 1206 | #define UP5(c) _rgb_scale_5[(c) & 31] |
| 1207 | #define UP6(c, b) _rgb_scale_6[(((c) & 31) << 1) | ((b) & 1)] |
| 1208 | #define LERP(n, t, c0, c1) (((n) - (t)) * (c0) + (t) * (c1) + (n) / 2) / (n) |
| 1209 | #define ZERO_4UBV(v) *((dword *)(v)) = 0 |
| 1210 | |
| 1211 | |
| 1212 | static void |
| 1213 | fxt1_decode_1HI (const byte *code, int t, byte *rgba) |
| 1214 | { |
| 1215 | const dword *cc; |
| 1216 | |
| 1217 | t *= 3; |
| 1218 | cc = (const dword *)(code + t / 8); |
| 1219 | t = (cc[0] >> (t & 7)) & 7; |
| 1220 | |
| 1221 | if (t == 7) { |
| 1222 | ZERO_4UBV(rgba); |
| 1223 | } else { |
| 1224 | cc = (const dword *)(code + 12); |
| 1225 | if (t == 0) { |
| 1226 | rgba[BCOMP] = UP5(CC_SEL(cc, 0)); |
| 1227 | rgba[GCOMP] = UP5(CC_SEL(cc, 5)); |
| 1228 | rgba[RCOMP] = UP5(CC_SEL(cc, 10)); |
| 1229 | } else if (t == 6) { |
| 1230 | rgba[BCOMP] = UP5(CC_SEL(cc, 15)); |
| 1231 | rgba[GCOMP] = UP5(CC_SEL(cc, 20)); |
| 1232 | rgba[RCOMP] = UP5(CC_SEL(cc, 25)); |
| 1233 | } else { |
| 1234 | rgba[BCOMP] = LERP(6, t, UP5(CC_SEL(cc, 0)), UP5(CC_SEL(cc, 15))); |
| 1235 | rgba[GCOMP] = LERP(6, t, UP5(CC_SEL(cc, 5)), UP5(CC_SEL(cc, 20))); |
| 1236 | rgba[RCOMP] = LERP(6, t, UP5(CC_SEL(cc, 10)), UP5(CC_SEL(cc, 25))); |
| 1237 | } |
| 1238 | rgba[ACOMP] = 255; |
| 1239 | } |
| 1240 | } |
| 1241 | |
| 1242 | |
| 1243 | static void |
| 1244 | fxt1_decode_1CHROMA (const byte *code, int t, byte *rgba) |
| 1245 | { |
| 1246 | const dword *cc; |
| 1247 | dword kk; |
| 1248 | |
| 1249 | cc = (const dword *)code; |
| 1250 | if (t & 16) { |
| 1251 | cc++; |
| 1252 | t &= 15; |
| 1253 | } |
| 1254 | t = (cc[0] >> (t * 2)) & 3; |
| 1255 | |
| 1256 | t *= 15; |
| 1257 | cc = (const dword *)(code + 8 + t / 8); |
| 1258 | kk = cc[0] >> (t & 7); |
| 1259 | rgba[BCOMP] = UP5(kk); |
| 1260 | rgba[GCOMP] = UP5(kk >> 5); |
| 1261 | rgba[RCOMP] = UP5(kk >> 10); |
| 1262 | rgba[ACOMP] = 255; |
| 1263 | } |
| 1264 | |
| 1265 | |
| 1266 | static void |
| 1267 | fxt1_decode_1MIXED (const byte *code, int t, byte *rgba) |
| 1268 | { |
| 1269 | const dword *cc; |
| 1270 | int col[2][3]; |
| 1271 | int glsb, selb; |
| 1272 | |
| 1273 | cc = (const dword *)code; |
| 1274 | if (t & 16) { |
| 1275 | t &= 15; |
| 1276 | t = (cc[1] >> (t * 2)) & 3; |
| 1277 | /* col 2 */ |
| 1278 | col[0][BCOMP] = (*(const dword *)(code + 11)) >> 6; |
| 1279 | col[0][GCOMP] = CC_SEL(cc, 99); |
| 1280 | col[0][RCOMP] = CC_SEL(cc, 104); |
| 1281 | /* col 3 */ |
| 1282 | col[1][BCOMP] = CC_SEL(cc, 109); |
| 1283 | col[1][GCOMP] = CC_SEL(cc, 114); |
| 1284 | col[1][RCOMP] = CC_SEL(cc, 119); |
| 1285 | glsb = CC_SEL(cc, 126); |
| 1286 | selb = CC_SEL(cc, 33); |
| 1287 | } else { |
| 1288 | t = (cc[0] >> (t * 2)) & 3; |
| 1289 | /* col 0 */ |
| 1290 | col[0][BCOMP] = CC_SEL(cc, 64); |
| 1291 | col[0][GCOMP] = CC_SEL(cc, 69); |
| 1292 | col[0][RCOMP] = CC_SEL(cc, 74); |
| 1293 | /* col 1 */ |
| 1294 | col[1][BCOMP] = CC_SEL(cc, 79); |
| 1295 | col[1][GCOMP] = CC_SEL(cc, 84); |
| 1296 | col[1][RCOMP] = CC_SEL(cc, 89); |
| 1297 | glsb = CC_SEL(cc, 125); |
| 1298 | selb = CC_SEL(cc, 1); |
| 1299 | } |
| 1300 | |
| 1301 | if (CC_SEL(cc, 124) & 1) { |
| 1302 | /* alpha[0] == 1 */ |
| 1303 | |
| 1304 | if (t == 3) { |
| 1305 | ZERO_4UBV(rgba); |
| 1306 | } else { |
| 1307 | if (t == 0) { |
| 1308 | rgba[BCOMP] = UP5(col[0][BCOMP]); |
| 1309 | rgba[GCOMP] = UP5(col[0][GCOMP]); |
| 1310 | rgba[RCOMP] = UP5(col[0][RCOMP]); |
| 1311 | } else if (t == 2) { |
| 1312 | rgba[BCOMP] = UP5(col[1][BCOMP]); |
| 1313 | rgba[GCOMP] = UP6(col[1][GCOMP], glsb); |
| 1314 | rgba[RCOMP] = UP5(col[1][RCOMP]); |
| 1315 | } else { |
| 1316 | rgba[BCOMP] = (UP5(col[0][BCOMP]) + UP5(col[1][BCOMP])) / 2; |
| 1317 | rgba[GCOMP] = (UP5(col[0][GCOMP]) + UP6(col[1][GCOMP], glsb)) / 2; |
| 1318 | rgba[RCOMP] = (UP5(col[0][RCOMP]) + UP5(col[1][RCOMP])) / 2; |
| 1319 | } |
| 1320 | rgba[ACOMP] = 255; |
| 1321 | } |
| 1322 | } else { |
| 1323 | /* alpha[0] == 0 */ |
| 1324 | |
| 1325 | if (t == 0) { |
| 1326 | rgba[BCOMP] = UP5(col[0][BCOMP]); |
| 1327 | rgba[GCOMP] = UP6(col[0][GCOMP], glsb ^ selb); |
| 1328 | rgba[RCOMP] = UP5(col[0][RCOMP]); |
| 1329 | } else if (t == 3) { |
| 1330 | rgba[BCOMP] = UP5(col[1][BCOMP]); |
| 1331 | rgba[GCOMP] = UP6(col[1][GCOMP], glsb); |
| 1332 | rgba[RCOMP] = UP5(col[1][RCOMP]); |
| 1333 | } else { |
| 1334 | rgba[BCOMP] = LERP(3, t, UP5(col[0][BCOMP]), UP5(col[1][BCOMP])); |
| 1335 | rgba[GCOMP] = LERP(3, t, UP6(col[0][GCOMP], glsb ^ selb), |
| 1336 | UP6(col[1][GCOMP], glsb)); |
| 1337 | rgba[RCOMP] = LERP(3, t, UP5(col[0][RCOMP]), UP5(col[1][RCOMP])); |
| 1338 | } |
| 1339 | rgba[ACOMP] = 255; |
| 1340 | } |
| 1341 | } |
| 1342 | |
| 1343 | |
| 1344 | static void |
| 1345 | fxt1_decode_1ALPHA (const byte *code, int t, byte *rgba) |
| 1346 | { |
| 1347 | const dword *cc; |
| 1348 | |
| 1349 | cc = (const dword *)code; |
| 1350 | if (CC_SEL(cc, 124) & 1) { |
| 1351 | /* lerp == 1 */ |
| 1352 | int col0[4]; |
| 1353 | |
| 1354 | if (t & 16) { |
| 1355 | t &= 15; |
| 1356 | t = (cc[1] >> (t * 2)) & 3; |
| 1357 | /* col 2 */ |
| 1358 | col0[BCOMP] = (*(const dword *)(code + 11)) >> 6; |
| 1359 | col0[GCOMP] = CC_SEL(cc, 99); |
| 1360 | col0[RCOMP] = CC_SEL(cc, 104); |
| 1361 | col0[ACOMP] = CC_SEL(cc, 119); |
| 1362 | } else { |
| 1363 | t = (cc[0] >> (t * 2)) & 3; |
| 1364 | /* col 0 */ |
| 1365 | col0[BCOMP] = CC_SEL(cc, 64); |
| 1366 | col0[GCOMP] = CC_SEL(cc, 69); |
| 1367 | col0[RCOMP] = CC_SEL(cc, 74); |
| 1368 | col0[ACOMP] = CC_SEL(cc, 109); |
| 1369 | } |
| 1370 | |
| 1371 | if (t == 0) { |
| 1372 | rgba[BCOMP] = UP5(col0[BCOMP]); |
| 1373 | rgba[GCOMP] = UP5(col0[GCOMP]); |
| 1374 | rgba[RCOMP] = UP5(col0[RCOMP]); |
| 1375 | rgba[ACOMP] = UP5(col0[ACOMP]); |
| 1376 | } else if (t == 3) { |
| 1377 | rgba[BCOMP] = UP5(CC_SEL(cc, 79)); |
| 1378 | rgba[GCOMP] = UP5(CC_SEL(cc, 84)); |
| 1379 | rgba[RCOMP] = UP5(CC_SEL(cc, 89)); |
| 1380 | rgba[ACOMP] = UP5(CC_SEL(cc, 114)); |
| 1381 | } else { |
| 1382 | rgba[BCOMP] = LERP(3, t, UP5(col0[BCOMP]), UP5(CC_SEL(cc, 79))); |
| 1383 | rgba[GCOMP] = LERP(3, t, UP5(col0[GCOMP]), UP5(CC_SEL(cc, 84))); |
| 1384 | rgba[RCOMP] = LERP(3, t, UP5(col0[RCOMP]), UP5(CC_SEL(cc, 89))); |
| 1385 | rgba[ACOMP] = LERP(3, t, UP5(col0[ACOMP]), UP5(CC_SEL(cc, 114))); |
| 1386 | } |
| 1387 | } else { |
| 1388 | /* lerp == 0 */ |
| 1389 | |
| 1390 | if (t & 16) { |
| 1391 | cc++; |
| 1392 | t &= 15; |
| 1393 | } |
| 1394 | t = (cc[0] >> (t * 2)) & 3; |
| 1395 | |
| 1396 | if (t == 3) { |
| 1397 | ZERO_4UBV(rgba); |
| 1398 | } else { |
| 1399 | dword kk; |
| 1400 | cc = (const dword *)code; |
| 1401 | rgba[ACOMP] = UP5(cc[3] >> (t * 5 + 13)); |
| 1402 | t *= 15; |
| 1403 | cc = (const dword *)(code + 8 + t / 8); |
| 1404 | kk = cc[0] >> (t & 7); |
| 1405 | rgba[BCOMP] = UP5(kk); |
| 1406 | rgba[GCOMP] = UP5(kk >> 5); |
| 1407 | rgba[RCOMP] = UP5(kk >> 10); |
| 1408 | } |
| 1409 | } |
| 1410 | } |
| 1411 | |
| 1412 | |
| 1413 | TAPI void TAPIENTRY |
| 1414 | fxt1_decode_1 (const void *texture, int stride, |
| 1415 | int i, int j, byte *rgba) |
| 1416 | { |
| 1417 | static void (*decode_1[]) (const byte *, int, byte *) = { |
| 1418 | fxt1_decode_1HI, /* cc-high = "00?" */ |
| 1419 | fxt1_decode_1HI, /* cc-high = "00?" */ |
| 1420 | fxt1_decode_1CHROMA, /* cc-chroma = "010" */ |
| 1421 | fxt1_decode_1ALPHA, /* alpha = "011" */ |
| 1422 | fxt1_decode_1MIXED, /* mixed = "1??" */ |
| 1423 | fxt1_decode_1MIXED, /* mixed = "1??" */ |
| 1424 | fxt1_decode_1MIXED, /* mixed = "1??" */ |
| 1425 | fxt1_decode_1MIXED /* mixed = "1??" */ |
| 1426 | }; |
| 1427 | |
| 1428 | const byte *code = (const byte *)texture + |
| 1429 | ((j / 4) * (stride / 8) + (i / 8)) * 16; |
| 1430 | int mode = CC_SEL(code, 125); |
| 1431 | int t = i & 7; |
| 1432 | |
| 1433 | if (t & 4) { |
| 1434 | t += 12; |
| 1435 | } |
| 1436 | t += (j & 3) * 4; |
| 1437 | |
| 1438 | decode_1[mode](code, t, rgba); |
| 1439 | |
| 1440 | #if VERBOSE |
| 1441 | { |
| 1442 | extern int cc_chroma; |
| 1443 | extern int cc_alpha; |
| 1444 | extern int cc_high; |
| 1445 | extern int cc_mixed; |
| 1446 | static int *cctype[] = { |
| 1447 | &cc_high, |
| 1448 | &cc_high, |
| 1449 | &cc_chroma, |
| 1450 | &cc_alpha, |
| 1451 | &cc_mixed, |
| 1452 | &cc_mixed, |
| 1453 | &cc_mixed, |
| 1454 | &cc_mixed |
| 1455 | }; |
| 1456 | (*cctype[mode])++; |
| 1457 | } |
| 1458 | #endif |
| 1459 | } |