98e75f2d |
1 | /* |
2d262872 |
2 | * Mesa 3-D graphics library |
98e75f2d |
3 | * |
2d262872 |
4 | * Copyright (C) 1999-2008 Brian Paul All Rights Reserved. |
98e75f2d |
5 | * |
6 | * Permission is hereby granted, free of charge, to any person obtaining a |
7 | * copy of this software and associated documentation files (the "Software"), |
8 | * to deal in the Software without restriction, including without limitation |
9 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
10 | * and/or sell copies of the Software, and to permit persons to whom the |
11 | * Software is furnished to do so, subject to the following conditions: |
12 | * |
13 | * The above copyright notice and this permission notice shall be included |
14 | * in all copies or substantial portions of the Software. |
15 | * |
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS |
17 | * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
2d262872 |
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR |
20 | * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, |
21 | * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR |
22 | * OTHER DEALINGS IN THE SOFTWARE. |
98e75f2d |
23 | */ |
24 | |
2d262872 |
25 | /** |
26 | * \file texcompress_fxt1.c |
27 | * GL_3DFX_texture_compression_FXT1 support. |
98e75f2d |
28 | */ |
29 | |
30 | |
31 | #include <stdlib.h> |
32 | #include <string.h> |
2d262872 |
33 | #include <assert.h> |
98e75f2d |
34 | |
35 | #include "types.h" |
36 | #include "internal.h" |
37 | #include "fxt1.h" |
38 | |
39 | |
40 | /***************************************************************************\ |
41 | * FXT1 encoder |
42 | * |
43 | * The encoder was built by reversing the decoder, |
44 | * and is vaguely based on Texus2 by 3dfx. Note that this code |
45 | * is merely a proof of concept, since it is highly UNoptimized; |
46 | * moreover, it is sub-optimal due to initial conditions passed |
47 | * to Lloyd's algorithm (the interpolation modes are even worse). |
48 | \***************************************************************************/ |
49 | |
50 | |
51 | #define MAX_COMP 4 /* ever needed maximum number of components in texel */ |
52 | #define MAX_VECT 4 /* ever needed maximum number of base vectors to find */ |
53 | #define N_TEXELS 32 /* number of texels in a block (always 32) */ |
54 | #define LL_N_REP 50 /* number of iterations in lloyd's vq */ |
55 | #define LL_RMS_D 10 /* fault tolerance (maximum delta) */ |
56 | #define LL_RMS_E 255 /* fault tolerance (maximum error) */ |
57 | #define ALPHA_TS 2 /* alpha threshold: (255 - ALPHA_TS) deemed opaque */ |
58 | #define ISTBLACK(v) (*((dword *)(v)) == 0) |
98e75f2d |
59 | |
60 | |
61 | static int |
62 | fxt1_bestcol (float vec[][MAX_COMP], int nv, |
2d262872 |
63 | byte input[MAX_COMP], int nc) |
98e75f2d |
64 | { |
2d262872 |
65 | int i, j, best = -1; |
66 | float err = 1e9; /* big enough */ |
67 | |
68 | for (j = 0; j < nv; j++) { |
69 | float e = 0.0F; |
70 | for (i = 0; i < nc; i++) { |
71 | e += (vec[j][i] - input[i]) * (vec[j][i] - input[i]); |
72 | } |
73 | if (e < err) { |
74 | err = e; |
75 | best = j; |
76 | } |
77 | } |
78 | |
79 | return best; |
98e75f2d |
80 | } |
81 | |
82 | |
83 | static int |
84 | fxt1_worst (float vec[MAX_COMP], |
2d262872 |
85 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
98e75f2d |
86 | { |
2d262872 |
87 | int i, k, worst = -1; |
88 | float err = -1.0F; /* small enough */ |
89 | |
90 | for (k = 0; k < n; k++) { |
91 | float e = 0.0F; |
92 | for (i = 0; i < nc; i++) { |
93 | e += (vec[i] - input[k][i]) * (vec[i] - input[k][i]); |
94 | } |
95 | if (e > err) { |
96 | err = e; |
97 | worst = k; |
98 | } |
99 | } |
100 | |
101 | return worst; |
98e75f2d |
102 | } |
103 | |
104 | |
105 | static int |
106 | fxt1_variance (double variance[MAX_COMP], |
2d262872 |
107 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
98e75f2d |
108 | { |
2d262872 |
109 | int i, k, best = 0; |
110 | int sx, sx2; |
111 | double var, maxvar = -1; /* small enough */ |
112 | double teenth = 1.0 / n; |
113 | |
114 | for (i = 0; i < nc; i++) { |
115 | sx = sx2 = 0; |
116 | for (k = 0; k < n; k++) { |
117 | int t = input[k][i]; |
118 | sx += t; |
119 | sx2 += t * t; |
120 | } |
121 | var = sx2 * teenth - sx * sx * teenth * teenth; |
122 | if (maxvar < var) { |
123 | maxvar = var; |
124 | best = i; |
125 | } |
126 | if (variance) { |
127 | variance[i] = var; |
128 | } |
129 | } |
130 | |
131 | return best; |
98e75f2d |
132 | } |
133 | |
134 | |
135 | static int |
136 | fxt1_choose (float vec[][MAX_COMP], int nv, |
2d262872 |
137 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
98e75f2d |
138 | { |
139 | #if 0 |
2d262872 |
140 | /* Choose colors from a grid. |
141 | */ |
142 | int i, j; |
143 | |
144 | for (j = 0; j < nv; j++) { |
145 | int m = j * (n - 1) / (nv - 1); |
146 | for (i = 0; i < nc; i++) { |
147 | vec[j][i] = input[m][i]; |
148 | } |
149 | } |
98e75f2d |
150 | #else |
2d262872 |
151 | /* Our solution here is to find the darkest and brightest colors in |
152 | * the 8x4 tile and use those as the two representative colors. |
153 | * There are probably better algorithms to use (histogram-based). |
154 | */ |
155 | int i, j, k; |
156 | int minSum = 2000; /* big enough */ |
157 | int maxSum = -1; /* small enough */ |
158 | int minCol = 0; /* phoudoin: silent compiler! */ |
159 | int maxCol = 0; /* phoudoin: silent compiler! */ |
160 | |
161 | struct { |
162 | int flag; |
163 | int key; |
164 | int freq; |
165 | int idx; |
166 | } hist[N_TEXELS]; |
167 | int lenh = 0; |
168 | |
169 | memset(hist, 0, sizeof(hist)); |
170 | |
171 | for (k = 0; k < n; k++) { |
172 | int l; |
173 | int key = 0; |
174 | int sum = 0; |
175 | for (i = 0; i < nc; i++) { |
176 | key <<= 8; |
177 | key |= input[k][i]; |
178 | sum += input[k][i]; |
179 | } |
180 | for (l = 0; l < n; l++) { |
181 | if (!hist[l].flag) { |
182 | /* alloc new slot */ |
183 | hist[l].flag = !0; |
184 | hist[l].key = key; |
185 | hist[l].freq = 1; |
186 | hist[l].idx = k; |
187 | lenh = l + 1; |
188 | break; |
189 | } else if (hist[l].key == key) { |
190 | hist[l].freq++; |
191 | break; |
192 | } |
193 | } |
194 | if (minSum > sum) { |
195 | minSum = sum; |
196 | minCol = k; |
197 | } |
198 | if (maxSum < sum) { |
199 | maxSum = sum; |
200 | maxCol = k; |
201 | } |
202 | } |
203 | |
204 | if (lenh <= nv) { |
205 | for (j = 0; j < lenh; j++) { |
206 | for (i = 0; i < nc; i++) { |
207 | vec[j][i] = (float)input[hist[j].idx][i]; |
208 | } |
209 | } |
210 | for (; j < nv; j++) { |
211 | for (i = 0; i < nc; i++) { |
212 | vec[j][i] = vec[0][i]; |
213 | } |
214 | } |
215 | return 0; |
216 | } |
217 | |
218 | for (j = 0; j < nv; j++) { |
219 | for (i = 0; i < nc; i++) { |
220 | vec[j][i] = ((nv - 1 - j) * input[minCol][i] + j * input[maxCol][i] + (nv - 1) / 2) / (float)(nv - 1); |
221 | } |
222 | } |
98e75f2d |
223 | #endif |
224 | |
2d262872 |
225 | return !0; |
98e75f2d |
226 | } |
227 | |
228 | |
229 | static int |
230 | fxt1_lloyd (float vec[][MAX_COMP], int nv, |
2d262872 |
231 | byte input[N_TEXELS][MAX_COMP], int nc, int n) |
98e75f2d |
232 | { |
2d262872 |
233 | /* Use the generalized lloyd's algorithm for VQ: |
234 | * find 4 color vectors. |
235 | * |
236 | * for each sample color |
237 | * sort to nearest vector. |
238 | * |
239 | * replace each vector with the centroid of its matching colors. |
240 | * |
241 | * repeat until RMS doesn't improve. |
242 | * |
243 | * if a color vector has no samples, or becomes the same as another |
244 | * vector, replace it with the color which is farthest from a sample. |
245 | * |
246 | * vec[][MAX_COMP] initial vectors and resulting colors |
247 | * nv number of resulting colors required |
248 | * input[N_TEXELS][MAX_COMP] input texels |
249 | * nc number of components in input / vec |
250 | * n number of input samples |
251 | */ |
252 | |
253 | int sum[MAX_VECT][MAX_COMP]; /* used to accumulate closest texels */ |
254 | int cnt[MAX_VECT]; /* how many times a certain vector was chosen */ |
255 | float error, lasterror = 1e9; |
256 | |
257 | int i, j, k, rep; |
258 | |
259 | /* the quantizer */ |
260 | for (rep = 0; rep < LL_N_REP; rep++) { |
261 | /* reset sums & counters */ |
262 | for (j = 0; j < nv; j++) { |
263 | for (i = 0; i < nc; i++) { |
264 | sum[j][i] = 0; |
265 | } |
266 | cnt[j] = 0; |
267 | } |
268 | error = 0; |
269 | |
270 | /* scan whole block */ |
271 | for (k = 0; k < n; k++) { |
98e75f2d |
272 | #if 1 |
2d262872 |
273 | int best = -1; |
274 | float err = 1e9; /* big enough */ |
275 | /* determine best vector */ |
276 | for (j = 0; j < nv; j++) { |
277 | float e = (vec[j][0] - input[k][0]) * (vec[j][0] - input[k][0]) + |
278 | (vec[j][1] - input[k][1]) * (vec[j][1] - input[k][1]) + |
279 | (vec[j][2] - input[k][2]) * (vec[j][2] - input[k][2]); |
280 | if (nc == 4) { |
281 | e += (vec[j][3] - input[k][3]) * (vec[j][3] - input[k][3]); |
282 | } |
283 | if (e < err) { |
284 | err = e; |
285 | best = j; |
286 | } |
287 | } |
98e75f2d |
288 | #else |
2d262872 |
289 | int best = fxt1_bestcol(vec, nv, input[k], nc, &err); |
98e75f2d |
290 | #endif |
2d262872 |
291 | assert(best >= 0); |
292 | /* add in closest color */ |
293 | for (i = 0; i < nc; i++) { |
294 | sum[best][i] += input[k][i]; |
295 | } |
296 | /* mark this vector as used */ |
297 | cnt[best]++; |
298 | /* accumulate error */ |
299 | error += err; |
300 | } |
301 | |
302 | /* check RMS */ |
303 | if ((error < LL_RMS_E) || |
304 | ((error < lasterror) && ((lasterror - error) < LL_RMS_D))) { |
305 | return !0; /* good match */ |
306 | } |
307 | lasterror = error; |
308 | |
309 | /* move each vector to the barycenter of its closest colors */ |
310 | for (j = 0; j < nv; j++) { |
311 | if (cnt[j]) { |
312 | float div = 1.0F / cnt[j]; |
313 | for (i = 0; i < nc; i++) { |
314 | vec[j][i] = div * sum[j][i]; |
315 | } |
316 | } else { |
317 | /* this vec has no samples or is identical with a previous vec */ |
318 | int worst = fxt1_worst(vec[j], input, nc, n); |
319 | for (i = 0; i < nc; i++) { |
320 | vec[j][i] = input[worst][i]; |
321 | } |
322 | } |
323 | } |
324 | } |
325 | |
326 | return 0; /* could not converge fast enough */ |
98e75f2d |
327 | } |
328 | |
329 | |
330 | static void |
331 | fxt1_quantize_CHROMA (dword *cc, |
2d262872 |
332 | byte input[N_TEXELS][MAX_COMP]) |
98e75f2d |
333 | { |
2d262872 |
334 | const int n_vect = 4; /* 4 base vectors to find */ |
335 | const int n_comp = 3; /* 3 components: R, G, B */ |
336 | float vec[MAX_VECT][MAX_COMP]; |
337 | int i, j, k; |
338 | qword hi; /* high quadword */ |
339 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
340 | |
341 | if (fxt1_choose(vec, n_vect, input, n_comp, N_TEXELS) != 0) { |
342 | fxt1_lloyd(vec, n_vect, input, n_comp, N_TEXELS); |
343 | } |
344 | |
345 | Q_MOV32(hi, 4); /* cc-chroma = "010" + unused bit */ |
346 | for (j = n_vect - 1; j >= 0; j--) { |
347 | for (i = 0; i < n_comp; i++) { |
348 | /* add in colors */ |
349 | Q_SHL(hi, 5); |
350 | Q_OR32(hi, (dword)(vec[j][i] / 8.0F)); |
351 | } |
352 | } |
353 | ((qword *)cc)[1] = hi; |
354 | |
355 | lohi = lolo = 0; |
356 | /* right microtile */ |
357 | for (k = N_TEXELS - 1; k >= N_TEXELS/2; k--) { |
358 | lohi <<= 2; |
359 | lohi |= fxt1_bestcol(vec, n_vect, input[k], n_comp); |
360 | } |
361 | /* left microtile */ |
362 | for (; k >= 0; k--) { |
363 | lolo <<= 2; |
364 | lolo |= fxt1_bestcol(vec, n_vect, input[k], n_comp); |
365 | } |
366 | cc[1] = lohi; |
367 | cc[0] = lolo; |
98e75f2d |
368 | } |
369 | |
370 | |
371 | static void |
372 | fxt1_quantize_ALPHA0 (dword *cc, |
2d262872 |
373 | byte input[N_TEXELS][MAX_COMP], |
374 | byte reord[N_TEXELS][MAX_COMP], int n) |
98e75f2d |
375 | { |
2d262872 |
376 | const int n_vect = 3; /* 3 base vectors to find */ |
377 | const int n_comp = 4; /* 4 components: R, G, B, A */ |
378 | float vec[MAX_VECT][MAX_COMP]; |
379 | int i, j, k; |
380 | qword hi; /* high quadword */ |
381 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
382 | |
383 | /* the last vector indicates zero */ |
384 | for (i = 0; i < n_comp; i++) { |
385 | vec[n_vect][i] = 0; |
386 | } |
387 | |
388 | /* the first n texels in reord are guaranteed to be non-zero */ |
389 | if (fxt1_choose(vec, n_vect, reord, n_comp, n) != 0) { |
390 | fxt1_lloyd(vec, n_vect, reord, n_comp, n); |
391 | } |
392 | |
393 | Q_MOV32(hi, 6); /* alpha = "011" + lerp = 0 */ |
394 | for (j = n_vect - 1; j >= 0; j--) { |
395 | /* add in alphas */ |
396 | Q_SHL(hi, 5); |
397 | Q_OR32(hi, (dword)(vec[j][ACOMP] / 8.0F)); |
398 | } |
399 | for (j = n_vect - 1; j >= 0; j--) { |
400 | for (i = 0; i < n_comp - 1; i++) { |
401 | /* add in colors */ |
402 | Q_SHL(hi, 5); |
403 | Q_OR32(hi, (dword)(vec[j][i] / 8.0F)); |
404 | } |
405 | } |
406 | ((qword *)cc)[1] = hi; |
407 | |
408 | lohi = lolo = 0; |
409 | /* right microtile */ |
410 | for (k = N_TEXELS - 1; k >= N_TEXELS/2; k--) { |
411 | lohi <<= 2; |
412 | lohi |= fxt1_bestcol(vec, n_vect + 1, input[k], n_comp); |
413 | } |
414 | /* left microtile */ |
415 | for (; k >= 0; k--) { |
416 | lolo <<= 2; |
417 | lolo |= fxt1_bestcol(vec, n_vect + 1, input[k], n_comp); |
418 | } |
419 | cc[1] = lohi; |
420 | cc[0] = lolo; |
98e75f2d |
421 | } |
422 | |
423 | |
424 | static void |
425 | fxt1_quantize_ALPHA1 (dword *cc, |
2d262872 |
426 | byte input[N_TEXELS][MAX_COMP]) |
98e75f2d |
427 | { |
2d262872 |
428 | const int n_vect = 3; /* highest vector number in each microtile */ |
429 | const int n_comp = 4; /* 4 components: R, G, B, A */ |
430 | float vec[1 + 1 + 1][MAX_COMP]; /* 1.5 extrema for each sub-block */ |
431 | float b, iv[MAX_COMP]; /* interpolation vector */ |
432 | int i, j, k; |
433 | qword hi; /* high quadword */ |
434 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
435 | |
436 | int minSum; |
437 | int maxSum; |
438 | int minColL = 0, maxColL = 0; |
439 | int minColR = 0, maxColR = 0; |
440 | int sumL = 0, sumR = 0; |
441 | int nn_comp; |
442 | /* Our solution here is to find the darkest and brightest colors in |
443 | * the 4x4 tile and use those as the two representative colors. |
444 | * There are probably better algorithms to use (histogram-based). |
445 | */ |
446 | nn_comp = n_comp; |
447 | while ((minColL == maxColL) && nn_comp) { |
448 | minSum = 2000; /* big enough */ |
449 | maxSum = -1; /* small enough */ |
450 | for (k = 0; k < N_TEXELS / 2; k++) { |
451 | int sum = 0; |
452 | for (i = 0; i < nn_comp; i++) { |
453 | sum += input[k][i]; |
454 | } |
455 | if (minSum > sum) { |
456 | minSum = sum; |
457 | minColL = k; |
458 | } |
459 | if (maxSum < sum) { |
460 | maxSum = sum; |
461 | maxColL = k; |
462 | } |
463 | sumL += sum; |
464 | } |
465 | |
466 | nn_comp--; |
467 | } |
468 | |
469 | nn_comp = n_comp; |
470 | while ((minColR == maxColR) && nn_comp) { |
471 | minSum = 2000; /* big enough */ |
472 | maxSum = -1; /* small enough */ |
473 | for (k = N_TEXELS / 2; k < N_TEXELS; k++) { |
474 | int sum = 0; |
475 | for (i = 0; i < nn_comp; i++) { |
476 | sum += input[k][i]; |
477 | } |
478 | if (minSum > sum) { |
479 | minSum = sum; |
480 | minColR = k; |
481 | } |
482 | if (maxSum < sum) { |
483 | maxSum = sum; |
484 | maxColR = k; |
485 | } |
486 | sumR += sum; |
487 | } |
488 | |
489 | nn_comp--; |
490 | } |
491 | |
492 | /* choose the common vector (yuck!) */ |
493 | { |
494 | int j1, j2; |
495 | int v1 = 0, v2 = 0; |
496 | float err = 1e9; /* big enough */ |
497 | float tv[2 * 2][MAX_COMP]; /* 2 extrema for each sub-block */ |
498 | for (i = 0; i < n_comp; i++) { |
499 | tv[0][i] = input[minColL][i]; |
500 | tv[1][i] = input[maxColL][i]; |
501 | tv[2][i] = input[minColR][i]; |
502 | tv[3][i] = input[maxColR][i]; |
503 | } |
504 | for (j1 = 0; j1 < 2; j1++) { |
505 | for (j2 = 2; j2 < 4; j2++) { |
506 | float e = 0.0F; |
507 | for (i = 0; i < n_comp; i++) { |
508 | e += (tv[j1][i] - tv[j2][i]) * (tv[j1][i] - tv[j2][i]); |
509 | } |
510 | if (e < err) { |
511 | err = e; |
512 | v1 = j1; |
513 | v2 = j2; |
514 | } |
515 | } |
516 | } |
517 | for (i = 0; i < n_comp; i++) { |
518 | vec[0][i] = tv[1 - v1][i]; |
519 | vec[1][i] = (tv[v1][i] * sumL + tv[v2][i] * sumR) / (sumL + sumR); |
520 | vec[2][i] = tv[5 - v2][i]; |
521 | } |
522 | } |
523 | |
524 | /* left microtile */ |
525 | cc[0] = 0; |
526 | if (minColL != maxColL) { |
527 | /* compute interpolation vector */ |
528 | MAKEIVEC(n_vect, n_comp, iv, b, vec[0], vec[1]); |
529 | |
530 | /* add in texels */ |
531 | lolo = 0; |
532 | for (k = N_TEXELS / 2 - 1; k >= 0; k--) { |
533 | int texel; |
534 | /* interpolate color */ |
535 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
536 | /* add in texel */ |
537 | lolo <<= 2; |
538 | lolo |= texel; |
539 | } |
540 | |
541 | cc[0] = lolo; |
542 | } |
543 | |
544 | /* right microtile */ |
545 | cc[1] = 0; |
546 | if (minColR != maxColR) { |
547 | /* compute interpolation vector */ |
548 | MAKEIVEC(n_vect, n_comp, iv, b, vec[2], vec[1]); |
549 | |
550 | /* add in texels */ |
551 | lohi = 0; |
552 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
553 | int texel; |
554 | /* interpolate color */ |
555 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
556 | /* add in texel */ |
557 | lohi <<= 2; |
558 | lohi |= texel; |
559 | } |
560 | |
561 | cc[1] = lohi; |
562 | } |
563 | |
564 | Q_MOV32(hi, 7); /* alpha = "011" + lerp = 1 */ |
565 | for (j = n_vect - 1; j >= 0; j--) { |
566 | /* add in alphas */ |
567 | Q_SHL(hi, 5); |
568 | Q_OR32(hi, (dword)(vec[j][ACOMP] / 8.0F)); |
569 | } |
570 | for (j = n_vect - 1; j >= 0; j--) { |
571 | for (i = 0; i < n_comp - 1; i++) { |
572 | /* add in colors */ |
573 | Q_SHL(hi, 5); |
574 | Q_OR32(hi, (dword)(vec[j][i] / 8.0F)); |
575 | } |
576 | } |
577 | ((qword *)cc)[1] = hi; |
98e75f2d |
578 | } |
579 | |
580 | |
581 | static void |
582 | fxt1_quantize_HI (dword *cc, |
2d262872 |
583 | byte input[N_TEXELS][MAX_COMP], |
584 | byte reord[N_TEXELS][MAX_COMP], int n) |
98e75f2d |
585 | { |
2d262872 |
586 | const int n_vect = 6; /* highest vector number */ |
587 | const int n_comp = 3; /* 3 components: R, G, B */ |
588 | float b = 0.0F; /* phoudoin: silent compiler! */ |
589 | float iv[MAX_COMP]; /* interpolation vector */ |
590 | int i, k; |
591 | dword hihi; /* high quadword: hi dword */ |
592 | |
593 | int minSum = 2000; /* big enough */ |
594 | int maxSum = -1; /* small enough */ |
595 | int minCol = 0; /* phoudoin: silent compiler! */ |
596 | int maxCol = 0; /* phoudoin: silent compiler! */ |
597 | |
598 | /* Our solution here is to find the darkest and brightest colors in |
599 | * the 8x4 tile and use those as the two representative colors. |
600 | * There are probably better algorithms to use (histogram-based). |
601 | */ |
602 | for (k = 0; k < n; k++) { |
603 | int sum = 0; |
604 | for (i = 0; i < n_comp; i++) { |
605 | sum += reord[k][i]; |
606 | } |
607 | if (minSum > sum) { |
608 | minSum = sum; |
609 | minCol = k; |
610 | } |
611 | if (maxSum < sum) { |
612 | maxSum = sum; |
613 | maxCol = k; |
614 | } |
615 | } |
616 | |
617 | hihi = 0; /* cc-hi = "00" */ |
618 | for (i = 0; i < n_comp; i++) { |
619 | /* add in colors */ |
620 | hihi <<= 5; |
621 | hihi |= reord[maxCol][i] >> 3; |
622 | } |
623 | for (i = 0; i < n_comp; i++) { |
624 | /* add in colors */ |
625 | hihi <<= 5; |
626 | hihi |= reord[minCol][i] >> 3; |
627 | } |
628 | cc[3] = hihi; |
629 | cc[0] = cc[1] = cc[2] = 0; |
630 | |
631 | /* compute interpolation vector */ |
632 | if (minCol != maxCol) { |
633 | MAKEIVEC(n_vect, n_comp, iv, b, reord[minCol], reord[maxCol]); |
634 | } |
635 | |
636 | /* add in texels */ |
637 | for (k = N_TEXELS - 1; k >= 0; k--) { |
638 | int t = k * 3; |
639 | dword *kk = (dword *)((char *)cc + t / 8); |
640 | int texel = n_vect + 1; /* transparent black */ |
641 | |
642 | if (!ISTBLACK(input[k])) { |
643 | if (minCol != maxCol) { |
644 | /* interpolate color */ |
645 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
646 | /* add in texel */ |
647 | kk[0] |= texel << (t & 7); |
648 | } |
649 | } else { |
650 | /* add in texel */ |
651 | kk[0] |= texel << (t & 7); |
652 | } |
653 | } |
98e75f2d |
654 | } |
655 | |
656 | |
657 | static void |
658 | fxt1_quantize_MIXED1 (dword *cc, |
2d262872 |
659 | byte input[N_TEXELS][MAX_COMP]) |
98e75f2d |
660 | { |
2d262872 |
661 | const int n_vect = 2; /* highest vector number in each microtile */ |
662 | const int n_comp = 3; /* 3 components: R, G, B */ |
663 | byte vec[2 * 2][MAX_COMP]; /* 2 extrema for each sub-block */ |
664 | float b, iv[MAX_COMP]; /* interpolation vector */ |
665 | int i, j, k; |
666 | qword hi; /* high quadword */ |
667 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
668 | |
669 | int minSum; |
670 | int maxSum; |
671 | int minColL = 0, maxColL = -1; |
672 | int minColR = 0, maxColR = -1; |
673 | |
674 | /* Our solution here is to find the darkest and brightest colors in |
675 | * the 4x4 tile and use those as the two representative colors. |
676 | * There are probably better algorithms to use (histogram-based). |
677 | */ |
678 | minSum = 2000; /* big enough */ |
679 | maxSum = -1; /* small enough */ |
680 | for (k = 0; k < N_TEXELS / 2; k++) { |
681 | if (!ISTBLACK(input[k])) { |
682 | int sum = 0; |
683 | for (i = 0; i < n_comp; i++) { |
684 | sum += input[k][i]; |
685 | } |
686 | if (minSum > sum) { |
687 | minSum = sum; |
688 | minColL = k; |
689 | } |
690 | if (maxSum < sum) { |
691 | maxSum = sum; |
692 | maxColL = k; |
693 | } |
694 | } |
695 | } |
696 | minSum = 2000; /* big enough */ |
697 | maxSum = -1; /* small enough */ |
698 | for (; k < N_TEXELS; k++) { |
699 | if (!ISTBLACK(input[k])) { |
700 | int sum = 0; |
701 | for (i = 0; i < n_comp; i++) { |
702 | sum += input[k][i]; |
703 | } |
704 | if (minSum > sum) { |
705 | minSum = sum; |
706 | minColR = k; |
707 | } |
708 | if (maxSum < sum) { |
709 | maxSum = sum; |
710 | maxColR = k; |
711 | } |
712 | } |
713 | } |
714 | |
715 | /* left microtile */ |
716 | if (maxColL == -1) { |
717 | /* all transparent black */ |
718 | cc[0] = ~0u; |
719 | for (i = 0; i < n_comp; i++) { |
720 | vec[0][i] = 0; |
721 | vec[1][i] = 0; |
722 | } |
723 | } else { |
724 | cc[0] = 0; |
725 | for (i = 0; i < n_comp; i++) { |
726 | vec[0][i] = input[minColL][i]; |
727 | vec[1][i] = input[maxColL][i]; |
728 | } |
729 | if (minColL != maxColL) { |
730 | /* compute interpolation vector */ |
731 | MAKEIVEC(n_vect, n_comp, iv, b, vec[0], vec[1]); |
732 | |
733 | /* add in texels */ |
734 | lolo = 0; |
735 | for (k = N_TEXELS / 2 - 1; k >= 0; k--) { |
736 | int texel = n_vect + 1; /* transparent black */ |
737 | if (!ISTBLACK(input[k])) { |
738 | /* interpolate color */ |
739 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
740 | } |
741 | /* add in texel */ |
742 | lolo <<= 2; |
743 | lolo |= texel; |
744 | } |
745 | cc[0] = lolo; |
746 | } |
747 | } |
748 | |
749 | /* right microtile */ |
750 | if (maxColR == -1) { |
751 | /* all transparent black */ |
752 | cc[1] = ~0u; |
753 | for (i = 0; i < n_comp; i++) { |
754 | vec[2][i] = 0; |
755 | vec[3][i] = 0; |
756 | } |
757 | } else { |
758 | cc[1] = 0; |
759 | for (i = 0; i < n_comp; i++) { |
760 | vec[2][i] = input[minColR][i]; |
761 | vec[3][i] = input[maxColR][i]; |
762 | } |
763 | if (minColR != maxColR) { |
764 | /* compute interpolation vector */ |
765 | MAKEIVEC(n_vect, n_comp, iv, b, vec[2], vec[3]); |
766 | |
767 | /* add in texels */ |
768 | lohi = 0; |
769 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
770 | int texel = n_vect + 1; /* transparent black */ |
771 | if (!ISTBLACK(input[k])) { |
772 | /* interpolate color */ |
773 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
774 | } |
775 | /* add in texel */ |
776 | lohi <<= 2; |
777 | lohi |= texel; |
778 | } |
779 | cc[1] = lohi; |
780 | } |
781 | } |
782 | |
783 | Q_MOV32(hi, 9 | (vec[3][GCOMP] & 4) | ((vec[1][GCOMP] >> 1) & 2)); /* chroma = "1" */ |
784 | for (j = 2 * 2 - 1; j >= 0; j--) { |
785 | for (i = 0; i < n_comp; i++) { |
786 | /* add in colors */ |
787 | Q_SHL(hi, 5); |
788 | Q_OR32(hi, vec[j][i] >> 3); |
789 | } |
790 | } |
791 | ((qword *)cc)[1] = hi; |
98e75f2d |
792 | } |
793 | |
794 | |
795 | static void |
796 | fxt1_quantize_MIXED0 (dword *cc, |
2d262872 |
797 | byte input[N_TEXELS][MAX_COMP]) |
98e75f2d |
798 | { |
2d262872 |
799 | const int n_vect = 3; /* highest vector number in each microtile */ |
800 | const int n_comp = 3; /* 3 components: R, G, B */ |
801 | byte vec[2 * 2][MAX_COMP]; /* 2 extrema for each sub-block */ |
802 | float b, iv[MAX_COMP]; /* interpolation vector */ |
803 | int i, j, k; |
804 | qword hi; /* high quadword */ |
805 | dword lohi, lolo; /* low quadword: hi dword, lo dword */ |
806 | |
807 | int minColL = 0, maxColL = 0; |
808 | int minColR = 0, maxColR = 0; |
98e75f2d |
809 | #if 0 |
2d262872 |
810 | int minSum; |
811 | int maxSum; |
812 | |
813 | /* Our solution here is to find the darkest and brightest colors in |
814 | * the 4x4 tile and use those as the two representative colors. |
815 | * There are probably better algorithms to use (histogram-based). |
816 | */ |
817 | minSum = 2000; /* big enough */ |
818 | maxSum = -1; /* small enough */ |
819 | for (k = 0; k < N_TEXELS / 2; k++) { |
820 | int sum = 0; |
821 | for (i = 0; i < n_comp; i++) { |
822 | sum += input[k][i]; |
823 | } |
824 | if (minSum > sum) { |
825 | minSum = sum; |
826 | minColL = k; |
827 | } |
828 | if (maxSum < sum) { |
829 | maxSum = sum; |
830 | maxColL = k; |
831 | } |
832 | } |
833 | minSum = 2000; /* big enough */ |
834 | maxSum = -1; /* small enough */ |
835 | for (; k < N_TEXELS; k++) { |
836 | int sum = 0; |
837 | for (i = 0; i < n_comp; i++) { |
838 | sum += input[k][i]; |
839 | } |
840 | if (minSum > sum) { |
841 | minSum = sum; |
842 | minColR = k; |
843 | } |
844 | if (maxSum < sum) { |
845 | maxSum = sum; |
846 | maxColR = k; |
847 | } |
848 | } |
98e75f2d |
849 | #else |
2d262872 |
850 | int minVal; |
851 | int maxVal; |
852 | int maxVarL = fxt1_variance(NULL, input, n_comp, N_TEXELS / 2); |
853 | int maxVarR = fxt1_variance(NULL, &input[N_TEXELS / 2], n_comp, N_TEXELS / 2); |
854 | |
855 | /* Scan the channel with max variance for lo & hi |
856 | * and use those as the two representative colors. |
857 | */ |
858 | minVal = 2000; /* big enough */ |
859 | maxVal = -1; /* small enough */ |
860 | for (k = 0; k < N_TEXELS / 2; k++) { |
861 | int t = input[k][maxVarL]; |
862 | if (minVal > t) { |
863 | minVal = t; |
864 | minColL = k; |
865 | } |
866 | if (maxVal < t) { |
867 | maxVal = t; |
868 | maxColL = k; |
869 | } |
870 | } |
871 | minVal = 2000; /* big enough */ |
872 | maxVal = -1; /* small enough */ |
873 | for (; k < N_TEXELS; k++) { |
874 | int t = input[k][maxVarR]; |
875 | if (minVal > t) { |
876 | minVal = t; |
877 | minColR = k; |
878 | } |
879 | if (maxVal < t) { |
880 | maxVal = t; |
881 | maxColR = k; |
882 | } |
883 | } |
98e75f2d |
884 | #endif |
885 | |
2d262872 |
886 | /* left microtile */ |
887 | cc[0] = 0; |
888 | for (i = 0; i < n_comp; i++) { |
889 | vec[0][i] = input[minColL][i]; |
890 | vec[1][i] = input[maxColL][i]; |
891 | } |
892 | if (minColL != maxColL) { |
893 | /* compute interpolation vector */ |
894 | MAKEIVEC(n_vect, n_comp, iv, b, vec[0], vec[1]); |
895 | |
896 | /* add in texels */ |
897 | lolo = 0; |
898 | for (k = N_TEXELS / 2 - 1; k >= 0; k--) { |
899 | int texel; |
900 | /* interpolate color */ |
901 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
902 | /* add in texel */ |
903 | lolo <<= 2; |
904 | lolo |= texel; |
905 | } |
906 | |
907 | /* funky encoding for LSB of green */ |
908 | if ((int)((lolo >> 1) & 1) != (((vec[1][GCOMP] ^ vec[0][GCOMP]) >> 2) & 1)) { |
909 | for (i = 0; i < n_comp; i++) { |
910 | vec[1][i] = input[minColL][i]; |
911 | vec[0][i] = input[maxColL][i]; |
912 | } |
913 | lolo = ~lolo; |
914 | } |
915 | |
916 | cc[0] = lolo; |
917 | } |
918 | |
919 | /* right microtile */ |
920 | cc[1] = 0; |
921 | for (i = 0; i < n_comp; i++) { |
922 | vec[2][i] = input[minColR][i]; |
923 | vec[3][i] = input[maxColR][i]; |
924 | } |
925 | if (minColR != maxColR) { |
926 | /* compute interpolation vector */ |
927 | MAKEIVEC(n_vect, n_comp, iv, b, vec[2], vec[3]); |
928 | |
929 | /* add in texels */ |
930 | lohi = 0; |
931 | for (k = N_TEXELS - 1; k >= N_TEXELS / 2; k--) { |
932 | int texel; |
933 | /* interpolate color */ |
934 | CALCCDOT(texel, n_vect, n_comp, iv, b, input[k]); |
935 | /* add in texel */ |
936 | lohi <<= 2; |
937 | lohi |= texel; |
938 | } |
939 | |
940 | /* funky encoding for LSB of green */ |
941 | if ((int)((lohi >> 1) & 1) != (((vec[3][GCOMP] ^ vec[2][GCOMP]) >> 2) & 1)) { |
942 | for (i = 0; i < n_comp; i++) { |
943 | vec[3][i] = input[minColR][i]; |
944 | vec[2][i] = input[maxColR][i]; |
945 | } |
946 | lohi = ~lohi; |
947 | } |
948 | |
949 | cc[1] = lohi; |
950 | } |
951 | |
952 | Q_MOV32(hi, 8 | (vec[3][GCOMP] & 4) | ((vec[1][GCOMP] >> 1) & 2)); /* chroma = "1" */ |
953 | for (j = 2 * 2 - 1; j >= 0; j--) { |
954 | for (i = 0; i < n_comp; i++) { |
955 | /* add in colors */ |
956 | Q_SHL(hi, 5); |
957 | Q_OR32(hi, vec[j][i] >> 3); |
958 | } |
959 | } |
960 | ((qword *)cc)[1] = hi; |
98e75f2d |
961 | } |
962 | |
963 | |
964 | static void |
965 | fxt1_quantize (dword *cc, const byte *lines[], int comps) |
966 | { |
2d262872 |
967 | int trualpha; |
968 | byte reord[N_TEXELS][MAX_COMP]; |
969 | |
970 | byte input[N_TEXELS][MAX_COMP]; |
971 | int i, k, l; |
972 | |
973 | if (comps == 3) { |
974 | /* make the whole block opaque */ |
975 | memset(input, -1, sizeof(input)); |
976 | } |
977 | |
978 | /* 8 texels each line */ |
979 | for (l = 0; l < 4; l++) { |
980 | for (k = 0; k < 4; k++) { |
981 | for (i = 0; i < comps; i++) { |
982 | input[k + l * 4][i] = *lines[l]++; |
983 | } |
984 | } |
985 | for (; k < 8; k++) { |
986 | for (i = 0; i < comps; i++) { |
987 | input[k + l * 4 + 12][i] = *lines[l]++; |
988 | } |
989 | } |
990 | } |
991 | |
992 | /* block layout: |
993 | * 00, 01, 02, 03, 08, 09, 0a, 0b |
994 | * 10, 11, 12, 13, 18, 19, 1a, 1b |
995 | * 04, 05, 06, 07, 0c, 0d, 0e, 0f |
996 | * 14, 15, 16, 17, 1c, 1d, 1e, 1f |
997 | */ |
998 | |
999 | /* [dBorca] |
1000 | * stupidity flows forth from this |
1001 | */ |
1002 | l = N_TEXELS; |
1003 | trualpha = 0; |
1004 | if (comps == 4) { |
1005 | /* skip all transparent black texels */ |
1006 | l = 0; |
1007 | for (k = 0; k < N_TEXELS; k++) { |
1008 | /* test all components against 0 */ |
1009 | if (!ISTBLACK(input[k])) { |
1010 | /* texel is not transparent black */ |
1011 | COPY_4UBV(reord[l], input[k]); |
1012 | if (reord[l][ACOMP] < (255 - ALPHA_TS)) { |
1013 | /* non-opaque texel */ |
1014 | trualpha = !0; |
1015 | } |
1016 | l++; |
1017 | } |
1018 | } |
1019 | } |
98e75f2d |
1020 | |
2d262872 |
1021 | #if 0 |
1022 | if (trualpha) { |
1023 | fxt1_quantize_ALPHA0(cc, input, reord, l); |
1024 | } else if (l == 0) { |
1025 | cc[0] = cc[1] = cc[2] = -1; |
1026 | cc[3] = 0; |
1027 | } else if (l < N_TEXELS) { |
1028 | fxt1_quantize_HI(cc, input, reord, l); |
1029 | } else { |
1030 | fxt1_quantize_CHROMA(cc, input); |
1031 | } |
1032 | (void)fxt1_quantize_ALPHA1; |
1033 | (void)fxt1_quantize_MIXED1; |
1034 | (void)fxt1_quantize_MIXED0; |
98e75f2d |
1035 | #else |
2d262872 |
1036 | if (trualpha) { |
1037 | fxt1_quantize_ALPHA1(cc, input); |
1038 | } else if (l == 0) { |
1039 | cc[0] = cc[1] = cc[2] = ~0u; |
1040 | cc[3] = 0; |
1041 | } else if (l < N_TEXELS) { |
1042 | fxt1_quantize_MIXED1(cc, input); |
1043 | } else { |
1044 | fxt1_quantize_MIXED0(cc, input); |
1045 | } |
1046 | (void)fxt1_quantize_ALPHA0; |
1047 | (void)fxt1_quantize_HI; |
1048 | (void)fxt1_quantize_CHROMA; |
98e75f2d |
1049 | #endif |
2d262872 |
1050 | } |
98e75f2d |
1051 | |
98e75f2d |
1052 | |
2d262872 |
1053 | |
1054 | /** |
1055 | * Upscale an image by replication, not (typical) stretching. |
1056 | * We use this when the image width or height is less than a |
1057 | * certain size (4, 8) and we need to upscale an image. |
1058 | */ |
1059 | static void |
1060 | upscale_teximage2d(int inWidth, int inHeight, |
1061 | int outWidth, int outHeight, |
1062 | int comps, const byte *src, int srcRowStride, |
1063 | byte *dest ) |
1064 | { |
1065 | int i, j, k; |
1066 | |
1067 | assert(outWidth >= inWidth); |
1068 | assert(outHeight >= inHeight); |
98e75f2d |
1069 | #if 0 |
2d262872 |
1070 | ASSERT(inWidth == 1 || inWidth == 2 || inHeight == 1 || inHeight == 2); |
1071 | ASSERT((outWidth & 3) == 0); |
1072 | ASSERT((outHeight & 3) == 0); |
98e75f2d |
1073 | #endif |
98e75f2d |
1074 | |
2d262872 |
1075 | for (i = 0; i < outHeight; i++) { |
1076 | const int ii = i % inHeight; |
1077 | for (j = 0; j < outWidth; j++) { |
1078 | const int jj = j % inWidth; |
1079 | for (k = 0; k < comps; k++) { |
1080 | dest[(i * outWidth + j) * comps + k] |
1081 | = src[ii * srcRowStride + jj * comps + k]; |
1082 | } |
1083 | } |
1084 | } |
1085 | } |
98e75f2d |
1086 | |
2d262872 |
1087 | TAPI void TAPIENTRY |
1088 | fxt1_encode (dword width, dword height, int comps, |
1089 | const void *source, int srcRowStride, |
1090 | void *dest, int destRowStride) |
98e75f2d |
1091 | { |
2d262872 |
1092 | dword x, y; |
1093 | const byte *data; |
1094 | dword *encoded = (dword *)dest; |
1095 | void *newSource = NULL, *newSourcetmp = NULL; |
1096 | |
1097 | assert(comps == 3 || comps == 4); |
1098 | |
1099 | if (comps == 3) |
1100 | newSource = reorder_source_3_alloc(source, width, height, srcRowStride); |
1101 | if (comps == 4) |
1102 | newSource = reorder_source_4_alloc(source, width, height, srcRowStride); |
1103 | if (!newSource) |
1104 | goto cleanUp; |
1105 | source = newSource; |
1106 | |
1107 | /* Replicate image if width is not M8 or height is not M4 */ |
1108 | if ((width & 7) | (height & 3)) { |
1109 | int newWidth = (width + 7) & ~7; |
1110 | int newHeight = (height + 3) & ~3; |
1111 | newSourcetmp = malloc(comps * newWidth * newHeight * sizeof(byte)); |
1112 | free(newSource); |
1113 | newSource = newSourcetmp; |
1114 | if (!newSource) { |
1115 | goto cleanUp; |
1116 | } |
1117 | upscale_teximage2d(width, height, newWidth, newHeight, |
1118 | comps, (const byte *) source, |
1119 | srcRowStride, (byte *) newSource); |
1120 | source = newSource; |
1121 | width = newWidth; |
1122 | height = newHeight; |
1123 | srcRowStride = comps * newWidth; |
1124 | } |
1125 | |
1126 | data = (const byte *) source; |
1127 | destRowStride = (destRowStride - width * 2) / 4; |
1128 | for (y = 0; y < height; y += 4) { |
1129 | dword offs = 0 + (y + 0) * srcRowStride; |
1130 | for (x = 0; x < width; x += 8) { |
1131 | const byte *lines[4]; |
1132 | lines[0] = &data[offs]; |
1133 | lines[1] = lines[0] + srcRowStride; |
1134 | lines[2] = lines[1] + srcRowStride; |
1135 | lines[3] = lines[2] + srcRowStride; |
1136 | offs += 8 * comps; |
1137 | fxt1_quantize(encoded, lines, comps); |
1138 | /* 128 bits per 8x4 block */ |
1139 | encoded += 4; |
1140 | } |
1141 | encoded += destRowStride; |
1142 | } |
1143 | |
1144 | cleanUp: |
1145 | free(newSource); |
98e75f2d |
1146 | } |
1147 | |
1148 | |
1149 | /***************************************************************************\ |
1150 | * FXT1 decoder |
1151 | * |
1152 | * The decoder is based on GL_3DFX_texture_compression_FXT1 |
1153 | * specification and serves as a concept for the encoder. |
1154 | \***************************************************************************/ |
1155 | |
1156 | |
1157 | /* lookup table for scaling 5 bit colors up to 8 bits */ |
1158 | static const byte _rgb_scale_5[] = { |
2d262872 |
1159 | 0, 8, 16, 25, 33, 41, 49, 58, |
1160 | 66, 74, 82, 90, 99, 107, 115, 123, |
1161 | 132, 140, 148, 156, 165, 173, 181, 189, |
1162 | 197, 206, 214, 222, 230, 239, 247, 255 |
98e75f2d |
1163 | }; |
1164 | |
1165 | /* lookup table for scaling 6 bit colors up to 8 bits */ |
1166 | static const byte _rgb_scale_6[] = { |
2d262872 |
1167 | 0, 4, 8, 12, 16, 20, 24, 28, |
1168 | 32, 36, 40, 45, 49, 53, 57, 61, |
1169 | 65, 69, 73, 77, 81, 85, 89, 93, |
1170 | 97, 101, 105, 109, 113, 117, 121, 125, |
1171 | 130, 134, 138, 142, 146, 150, 154, 158, |
1172 | 162, 166, 170, 174, 178, 182, 186, 190, |
1173 | 194, 198, 202, 206, 210, 215, 219, 223, |
1174 | 227, 231, 235, 239, 243, 247, 251, 255 |
98e75f2d |
1175 | }; |
1176 | |
1177 | |
1178 | #define CC_SEL(cc, which) (((dword *)(cc))[(which) / 32] >> ((which) & 31)) |
1179 | #define UP5(c) _rgb_scale_5[(c) & 31] |
1180 | #define UP6(c, b) _rgb_scale_6[(((c) & 31) << 1) | ((b) & 1)] |
1181 | #define LERP(n, t, c0, c1) (((n) - (t)) * (c0) + (t) * (c1) + (n) / 2) / (n) |
98e75f2d |
1182 | |
1183 | |
1184 | static void |
1185 | fxt1_decode_1HI (const byte *code, int t, byte *rgba) |
1186 | { |
2d262872 |
1187 | const dword *cc; |
1188 | |
1189 | t *= 3; |
1190 | cc = (const dword *)(code + t / 8); |
1191 | t = (cc[0] >> (t & 7)) & 7; |
1192 | |
1193 | if (t == 7) { |
1194 | rgba[RCOMP] = rgba[GCOMP] = rgba[BCOMP] = rgba[ACOMP] = 0; |
1195 | } else { |
1196 | byte r, g, b; |
1197 | cc = (const dword *)(code + 12); |
1198 | if (t == 0) { |
1199 | b = UP5(CC_SEL(cc, 0)); |
1200 | g = UP5(CC_SEL(cc, 5)); |
1201 | r = UP5(CC_SEL(cc, 10)); |
1202 | } else if (t == 6) { |
1203 | b = UP5(CC_SEL(cc, 15)); |
1204 | g = UP5(CC_SEL(cc, 20)); |
1205 | r = UP5(CC_SEL(cc, 25)); |
1206 | } else { |
1207 | b = LERP(6, t, UP5(CC_SEL(cc, 0)), UP5(CC_SEL(cc, 15))); |
1208 | g = LERP(6, t, UP5(CC_SEL(cc, 5)), UP5(CC_SEL(cc, 20))); |
1209 | r = LERP(6, t, UP5(CC_SEL(cc, 10)), UP5(CC_SEL(cc, 25))); |
1210 | } |
1211 | rgba[RCOMP] = r; |
1212 | rgba[GCOMP] = g; |
1213 | rgba[BCOMP] = b; |
1214 | rgba[ACOMP] = 255; |
1215 | } |
98e75f2d |
1216 | } |
1217 | |
1218 | |
1219 | static void |
1220 | fxt1_decode_1CHROMA (const byte *code, int t, byte *rgba) |
1221 | { |
2d262872 |
1222 | const dword *cc; |
1223 | dword kk; |
1224 | |
1225 | cc = (const dword *)code; |
1226 | if (t & 16) { |
1227 | cc++; |
1228 | t &= 15; |
1229 | } |
1230 | t = (cc[0] >> (t * 2)) & 3; |
1231 | |
1232 | t *= 15; |
1233 | cc = (const dword *)(code + 8 + t / 8); |
1234 | kk = cc[0] >> (t & 7); |
1235 | rgba[BCOMP] = UP5(kk); |
1236 | rgba[GCOMP] = UP5(kk >> 5); |
1237 | rgba[RCOMP] = UP5(kk >> 10); |
1238 | rgba[ACOMP] = 255; |
98e75f2d |
1239 | } |
1240 | |
1241 | |
1242 | static void |
1243 | fxt1_decode_1MIXED (const byte *code, int t, byte *rgba) |
1244 | { |
2d262872 |
1245 | const dword *cc; |
1246 | dword col[2][3]; |
1247 | int glsb, selb; |
1248 | |
1249 | cc = (const dword *)code; |
1250 | if (t & 16) { |
1251 | t &= 15; |
1252 | t = (cc[1] >> (t * 2)) & 3; |
1253 | /* col 2 */ |
1254 | col[0][BCOMP] = (*(const dword *)(code + 11)) >> 6; |
1255 | col[0][GCOMP] = CC_SEL(cc, 99); |
1256 | col[0][RCOMP] = CC_SEL(cc, 104); |
1257 | /* col 3 */ |
1258 | col[1][BCOMP] = CC_SEL(cc, 109); |
1259 | col[1][GCOMP] = CC_SEL(cc, 114); |
1260 | col[1][RCOMP] = CC_SEL(cc, 119); |
1261 | glsb = CC_SEL(cc, 126); |
1262 | selb = CC_SEL(cc, 33); |
1263 | } else { |
1264 | t = (cc[0] >> (t * 2)) & 3; |
1265 | /* col 0 */ |
1266 | col[0][BCOMP] = CC_SEL(cc, 64); |
1267 | col[0][GCOMP] = CC_SEL(cc, 69); |
1268 | col[0][RCOMP] = CC_SEL(cc, 74); |
1269 | /* col 1 */ |
1270 | col[1][BCOMP] = CC_SEL(cc, 79); |
1271 | col[1][GCOMP] = CC_SEL(cc, 84); |
1272 | col[1][RCOMP] = CC_SEL(cc, 89); |
1273 | glsb = CC_SEL(cc, 125); |
1274 | selb = CC_SEL(cc, 1); |
1275 | } |
1276 | |
1277 | if (CC_SEL(cc, 124) & 1) { |
1278 | /* alpha[0] == 1 */ |
1279 | |
1280 | if (t == 3) { |
1281 | /* zero */ |
1282 | rgba[RCOMP] = rgba[BCOMP] = rgba[GCOMP] = rgba[ACOMP] = 0; |
1283 | } else { |
1284 | byte r, g, b; |
1285 | if (t == 0) { |
1286 | b = UP5(col[0][BCOMP]); |
1287 | g = UP5(col[0][GCOMP]); |
1288 | r = UP5(col[0][RCOMP]); |
1289 | } else if (t == 2) { |
1290 | b = UP5(col[1][BCOMP]); |
1291 | g = UP6(col[1][GCOMP], glsb); |
1292 | r = UP5(col[1][RCOMP]); |
1293 | } else { |
1294 | b = (UP5(col[0][BCOMP]) + UP5(col[1][BCOMP])) / 2; |
1295 | g = (UP5(col[0][GCOMP]) + UP6(col[1][GCOMP], glsb)) / 2; |
1296 | r = (UP5(col[0][RCOMP]) + UP5(col[1][RCOMP])) / 2; |
1297 | } |
1298 | rgba[RCOMP] = r; |
1299 | rgba[GCOMP] = g; |
1300 | rgba[BCOMP] = b; |
1301 | rgba[ACOMP] = 255; |
1302 | } |
1303 | } else { |
1304 | /* alpha[0] == 0 */ |
1305 | byte r, g, b; |
1306 | if (t == 0) { |
1307 | b = UP5(col[0][BCOMP]); |
1308 | g = UP6(col[0][GCOMP], glsb ^ selb); |
1309 | r = UP5(col[0][RCOMP]); |
1310 | } else if (t == 3) { |
1311 | b = UP5(col[1][BCOMP]); |
1312 | g = UP6(col[1][GCOMP], glsb); |
1313 | r = UP5(col[1][RCOMP]); |
1314 | } else { |
1315 | b = LERP(3, t, UP5(col[0][BCOMP]), UP5(col[1][BCOMP])); |
1316 | g = LERP(3, t, UP6(col[0][GCOMP], glsb ^ selb), |
1317 | UP6(col[1][GCOMP], glsb)); |
1318 | r = LERP(3, t, UP5(col[0][RCOMP]), UP5(col[1][RCOMP])); |
1319 | } |
1320 | rgba[RCOMP] = r; |
1321 | rgba[GCOMP] = g; |
1322 | rgba[BCOMP] = b; |
1323 | rgba[ACOMP] = 255; |
1324 | } |
98e75f2d |
1325 | } |
1326 | |
1327 | |
1328 | static void |
1329 | fxt1_decode_1ALPHA (const byte *code, int t, byte *rgba) |
1330 | { |
2d262872 |
1331 | const dword *cc; |
1332 | byte r, g, b, a; |
1333 | |
1334 | cc = (const dword *)code; |
1335 | if (CC_SEL(cc, 124) & 1) { |
1336 | /* lerp == 1 */ |
1337 | dword col0[4]; |
1338 | |
1339 | if (t & 16) { |
1340 | t &= 15; |
1341 | t = (cc[1] >> (t * 2)) & 3; |
1342 | /* col 2 */ |
1343 | col0[BCOMP] = (*(const dword *)(code + 11)) >> 6; |
1344 | col0[GCOMP] = CC_SEL(cc, 99); |
1345 | col0[RCOMP] = CC_SEL(cc, 104); |
1346 | col0[ACOMP] = CC_SEL(cc, 119); |
1347 | } else { |
1348 | t = (cc[0] >> (t * 2)) & 3; |
1349 | /* col 0 */ |
1350 | col0[BCOMP] = CC_SEL(cc, 64); |
1351 | col0[GCOMP] = CC_SEL(cc, 69); |
1352 | col0[RCOMP] = CC_SEL(cc, 74); |
1353 | col0[ACOMP] = CC_SEL(cc, 109); |
1354 | } |
1355 | |
1356 | if (t == 0) { |
1357 | b = UP5(col0[BCOMP]); |
1358 | g = UP5(col0[GCOMP]); |
1359 | r = UP5(col0[RCOMP]); |
1360 | a = UP5(col0[ACOMP]); |
1361 | } else if (t == 3) { |
1362 | b = UP5(CC_SEL(cc, 79)); |
1363 | g = UP5(CC_SEL(cc, 84)); |
1364 | r = UP5(CC_SEL(cc, 89)); |
1365 | a = UP5(CC_SEL(cc, 114)); |
1366 | } else { |
1367 | b = LERP(3, t, UP5(col0[BCOMP]), UP5(CC_SEL(cc, 79))); |
1368 | g = LERP(3, t, UP5(col0[GCOMP]), UP5(CC_SEL(cc, 84))); |
1369 | r = LERP(3, t, UP5(col0[RCOMP]), UP5(CC_SEL(cc, 89))); |
1370 | a = LERP(3, t, UP5(col0[ACOMP]), UP5(CC_SEL(cc, 114))); |
1371 | } |
1372 | } else { |
1373 | /* lerp == 0 */ |
1374 | |
1375 | if (t & 16) { |
1376 | cc++; |
1377 | t &= 15; |
1378 | } |
1379 | t = (cc[0] >> (t * 2)) & 3; |
1380 | |
1381 | if (t == 3) { |
1382 | /* zero */ |
1383 | r = g = b = a = 0; |
1384 | } else { |
1385 | dword kk; |
1386 | cc = (const dword *)code; |
1387 | a = UP5(cc[3] >> (t * 5 + 13)); |
1388 | t *= 15; |
1389 | cc = (const dword *)(code + 8 + t / 8); |
1390 | kk = cc[0] >> (t & 7); |
1391 | b = UP5(kk); |
1392 | g = UP5(kk >> 5); |
1393 | r = UP5(kk >> 10); |
1394 | } |
1395 | } |
1396 | rgba[RCOMP] = r; |
1397 | rgba[GCOMP] = g; |
1398 | rgba[BCOMP] = b; |
1399 | rgba[ACOMP] = a; |
98e75f2d |
1400 | } |
1401 | |
1402 | |
1403 | TAPI void TAPIENTRY |
2d262872 |
1404 | fxt1_decode_1 (const void *texture, int stride, /* in pixels */ |
1405 | int i, int j, byte *rgba) |
98e75f2d |
1406 | { |
2d262872 |
1407 | static void (*decode_1[]) (const byte *, int, byte *) = { |
1408 | fxt1_decode_1HI, /* cc-high = "00?" */ |
1409 | fxt1_decode_1HI, /* cc-high = "00?" */ |
1410 | fxt1_decode_1CHROMA, /* cc-chroma = "010" */ |
1411 | fxt1_decode_1ALPHA, /* alpha = "011" */ |
1412 | fxt1_decode_1MIXED, /* mixed = "1??" */ |
1413 | fxt1_decode_1MIXED, /* mixed = "1??" */ |
1414 | fxt1_decode_1MIXED, /* mixed = "1??" */ |
1415 | fxt1_decode_1MIXED /* mixed = "1??" */ |
1416 | }; |
1417 | |
1418 | const byte *code = (const byte *)texture + |
1419 | ((j / 4) * (stride / 8) + (i / 8)) * 16; |
1420 | int mode = CC_SEL(code, 125); |
1421 | int t = i & 7; |
1422 | |
1423 | if (t & 4) { |
1424 | t += 12; |
1425 | } |
1426 | t += (j & 3) * 4; |
1427 | |
1428 | decode_1[mode](code, t, rgba); |
98e75f2d |
1429 | } |