projects / mesa.git / blob
? search:


4651d781a954d2bb851e801be4f70db5747f1ffa
[mesa.git] / src / gallium / drivers / softpipe / sp_tex_sample.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 * Copyright 2008 VMware, Inc. 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
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29 /**
30 * Texture sampling
31 *
32 * Authors:
33 * Brian Paul
34 */
35
36 #include "sp_context.h"
37 #include "sp_quad.h"
38 #include "sp_surface.h"
39 #include "sp_texture.h"
40 #include "sp_tex_sample.h"
41 #include "sp_tile_cache.h"
42 #include "pipe/p_context.h"
43 #include "pipe/p_defines.h"
44 #include "util/u_math.h"
45 #include "util/u_memory.h"
46
47
48
49 /*
50 * Note, the FRAC macro has to work perfectly. Otherwise you'll sometimes
51 * see 1-pixel bands of improperly weighted linear-filtered textures.
52 * The tests/texwrap.c demo is a good test.
53 * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0.
54 * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x).
55 */
56 #define FRAC(f) ((f) - util_ifloor(f))
57
58
59 /**
60 * Linear interpolation macro
61 */
62 static INLINE float
63 lerp(float a, float v0, float v1)
64 {
65 return v0 + a * (v1 - v0);
66 }
67
68
69 /**
70 * Do 2D/biliner interpolation of float values.
71 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
72 * a and b are the horizontal and vertical interpolants.
73 * It's important that this function is inlined when compiled with
74 * optimization! If we find that's not true on some systems, convert
75 * to a macro.
76 */
77 static INLINE float
78 lerp_2d(float a, float b,
79 float v00, float v10, float v01, float v11)
80 {
81 const float temp0 = lerp(a, v00, v10);
82 const float temp1 = lerp(a, v01, v11);
83 return lerp(b, temp0, temp1);
84 }
85
86
87 /**
88 * As above, but 3D interpolation of 8 values.
89 */
90 static INLINE float
91 lerp_3d(float a, float b, float c,
92 float v000, float v100, float v010, float v110,
93 float v001, float v101, float v011, float v111)
94 {
95 const float temp0 = lerp_2d(a, b, v000, v100, v010, v110);
96 const float temp1 = lerp_2d(a, b, v001, v101, v011, v111);
97 return lerp(c, temp0, temp1);
98 }
99
100
101
102 /**
103 * If A is a signed integer, A % B doesn't give the right value for A < 0
104 * (in terms of texture repeat). Just casting to unsigned fixes that.
105 */
106 #define REMAINDER(A, B) ((unsigned) (A) % (unsigned) (B))
107
108
109 /**
110 * Apply texture coord wrapping mode and return integer texture indexes
111 * for a vector of four texcoords (S or T or P).
112 * \param wrapMode PIPE_TEX_WRAP_x
113 * \param s the incoming texcoords
114 * \param size the texture image size
115 * \param icoord returns the integer texcoords
116 * \return integer texture index
117 */
118 static INLINE void
119 nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
120 int icoord[4])
121 {
122 uint ch;
123 switch (wrapMode) {
124 case PIPE_TEX_WRAP_REPEAT:
125 /* s limited to [0,1) */
126 /* i limited to [0,size-1] */
127 for (ch = 0; ch < 4; ch++) {
128 int i = util_ifloor(s[ch] * size);
129 icoord[ch] = REMAINDER(i, size);
130 }
131 return;
132 case PIPE_TEX_WRAP_CLAMP:
133 /* s limited to [0,1] */
134 /* i limited to [0,size-1] */
135 for (ch = 0; ch < 4; ch++) {
136 if (s[ch] <= 0.0F)
137 icoord[ch] = 0;
138 else if (s[ch] >= 1.0F)
139 icoord[ch] = size - 1;
140 else
141 icoord[ch] = util_ifloor(s[ch] * size);
142 }
143 return;
144 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
145 {
146 /* s limited to [min,max] */
147 /* i limited to [0, size-1] */
148 const float min = 1.0F / (2.0F * size);
149 const float max = 1.0F - min;
150 for (ch = 0; ch < 4; ch++) {
151 if (s[ch] < min)
152 icoord[ch] = 0;
153 else if (s[ch] > max)
154 icoord[ch] = size - 1;
155 else
156 icoord[ch] = util_ifloor(s[ch] * size);
157 }
158 }
159 return;
160 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
161 {
162 /* s limited to [min,max] */
163 /* i limited to [-1, size] */
164 const float min = -1.0F / (2.0F * size);
165 const float max = 1.0F - min;
166 for (ch = 0; ch < 4; ch++) {
167 if (s[ch] <= min)
168 icoord[ch] = -1;
169 else if (s[ch] >= max)
170 icoord[ch] = size;
171 else
172 icoord[ch] = util_ifloor(s[ch] * size);
173 }
174 }
175 return;
176 case PIPE_TEX_WRAP_MIRROR_REPEAT:
177 {
178 const float min = 1.0F / (2.0F * size);
179 const float max = 1.0F - min;
180 for (ch = 0; ch < 4; ch++) {
181 const int flr = util_ifloor(s[ch]);
182 float u;
183 if (flr & 1)
184 u = 1.0F - (s[ch] - (float) flr);
185 else
186 u = s[ch] - (float) flr;
187 if (u < min)
188 icoord[ch] = 0;
189 else if (u > max)
190 icoord[ch] = size - 1;
191 else
192 icoord[ch] = util_ifloor(u * size);
193 }
194 }
195 return;
196 case PIPE_TEX_WRAP_MIRROR_CLAMP:
197 for (ch = 0; ch < 4; ch++) {
198 /* s limited to [0,1] */
199 /* i limited to [0,size-1] */
200 const float u = fabsf(s[ch]);
201 if (u <= 0.0F)
202 icoord[ch] = 0;
203 else if (u >= 1.0F)
204 icoord[ch] = size - 1;
205 else
206 icoord[ch] = util_ifloor(u * size);
207 }
208 return;
209 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
210 {
211 /* s limited to [min,max] */
212 /* i limited to [0, size-1] */
213 const float min = 1.0F / (2.0F * size);
214 const float max = 1.0F - min;
215 for (ch = 0; ch < 4; ch++) {
216 const float u = fabsf(s[ch]);
217 if (u < min)
218 icoord[ch] = 0;
219 else if (u > max)
220 icoord[ch] = size - 1;
221 else
222 icoord[ch] = util_ifloor(u * size);
223 }
224 }
225 return;
226 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
227 {
228 /* s limited to [min,max] */
229 /* i limited to [0, size-1] */
230 const float min = -1.0F / (2.0F * size);
231 const float max = 1.0F - min;
232 for (ch = 0; ch < 4; ch++) {
233 const float u = fabsf(s[ch]);
234 if (u < min)
235 icoord[ch] = -1;
236 else if (u > max)
237 icoord[ch] = size;
238 else
239 icoord[ch] = util_ifloor(u * size);
240 }
241 }
242 return;
243 default:
244 assert(0);
245 }
246 }
247
248
249 /**
250 * Used to compute texel locations for linear sampling for four texcoords.
251 * \param wrapMode PIPE_TEX_WRAP_x
252 * \param s the texcoords
253 * \param size the texture image size
254 * \param icoord0 returns first texture indexes
255 * \param icoord1 returns second texture indexes (usually icoord0 + 1)
256 * \param w returns blend factor/weight between texture indexes
257 * \param icoord returns the computed integer texture coords
258 */
259 static INLINE void
260 linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
261 int icoord0[4], int icoord1[4], float w[4])
262 {
263 uint ch;
264
265 switch (wrapMode) {
266 case PIPE_TEX_WRAP_REPEAT:
267 for (ch = 0; ch < 4; ch++) {
268 float u = s[ch] * size - 0.5F;
269 icoord0[ch] = REMAINDER(util_ifloor(u), size);
270 icoord1[ch] = REMAINDER(icoord0[ch] + 1, size);
271 w[ch] = FRAC(u);
272 }
273 break;;
274 case PIPE_TEX_WRAP_CLAMP:
275 for (ch = 0; ch < 4; ch++) {
276 float u = CLAMP(s[ch], 0.0F, 1.0F);
277 u = u * size - 0.5f;
278 icoord0[ch] = util_ifloor(u);
279 icoord1[ch] = icoord0[ch] + 1;
280 w[ch] = FRAC(u);
281 }
282 break;;
283 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
284 for (ch = 0; ch < 4; ch++) {
285 float u = CLAMP(s[ch], 0.0F, 1.0F);
286 u = u * size - 0.5f;
287 icoord0[ch] = util_ifloor(u);
288 icoord1[ch] = icoord0[ch] + 1;
289 if (icoord0[ch] < 0)
290 icoord0[ch] = 0;
291 if (icoord1[ch] >= (int) size)
292 icoord1[ch] = size - 1;
293 w[ch] = FRAC(u);
294 }
295 break;;
296 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
297 {
298 const float min = -1.0F / (2.0F * size);
299 const float max = 1.0F - min;
300 for (ch = 0; ch < 4; ch++) {
301 float u = CLAMP(s[ch], min, max);
302 u = u * size - 0.5f;
303 icoord0[ch] = util_ifloor(u);
304 icoord1[ch] = icoord0[ch] + 1;
305 w[ch] = FRAC(u);
306 }
307 }
308 break;;
309 case PIPE_TEX_WRAP_MIRROR_REPEAT:
310 for (ch = 0; ch < 4; ch++) {
311 const int flr = util_ifloor(s[ch]);
312 float u;
313 if (flr & 1)
314 u = 1.0F - (s[ch] - (float) flr);
315 else
316 u = s[ch] - (float) flr;
317 u = u * size - 0.5F;
318 icoord0[ch] = util_ifloor(u);
319 icoord1[ch] = icoord0[ch] + 1;
320 if (icoord0[ch] < 0)
321 icoord0[ch] = 0;
322 if (icoord1[ch] >= (int) size)
323 icoord1[ch] = size - 1;
324 w[ch] = FRAC(u);
325 }
326 break;;
327 case PIPE_TEX_WRAP_MIRROR_CLAMP:
328 for (ch = 0; ch < 4; ch++) {
329 float u = fabsf(s[ch]);
330 if (u >= 1.0F)
331 u = (float) size;
332 else
333 u *= size;
334 u -= 0.5F;
335 icoord0[ch] = util_ifloor(u);
336 icoord1[ch] = icoord0[ch] + 1;
337 w[ch] = FRAC(u);
338 }
339 break;;
340 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
341 for (ch = 0; ch < 4; ch++) {
342 float u = fabsf(s[ch]);
343 if (u >= 1.0F)
344 u = (float) size;
345 else
346 u *= size;
347 u -= 0.5F;
348 icoord0[ch] = util_ifloor(u);
349 icoord1[ch] = icoord0[ch] + 1;
350 if (icoord0[ch] < 0)
351 icoord0[ch] = 0;
352 if (icoord1[ch] >= (int) size)
353 icoord1[ch] = size - 1;
354 w[ch] = FRAC(u);
355 }
356 break;;
357 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
358 {
359 const float min = -1.0F / (2.0F * size);
360 const float max = 1.0F - min;
361 for (ch = 0; ch < 4; ch++) {
362 float u = fabsf(s[ch]);
363 if (u <= min)
364 u = min * size;
365 else if (u >= max)
366 u = max * size;
367 else
368 u *= size;
369 u -= 0.5F;
370 icoord0[ch] = util_ifloor(u);
371 icoord1[ch] = icoord0[ch] + 1;
372 w[ch] = FRAC(u);
373 }
374 }
375 break;;
376 default:
377 assert(0);
378 }
379 }
380
381
382 /**
383 * For RECT textures / unnormalized texcoords
384 * Only a subset of wrap modes supported.
385 */
386 static INLINE void
387 nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
388 int icoord[4])
389 {
390 uint ch;
391 switch (wrapMode) {
392 case PIPE_TEX_WRAP_CLAMP:
393 for (ch = 0; ch < 4; ch++) {
394 int i = util_ifloor(s[ch]);
395 icoord[ch]= CLAMP(i, 0, (int) size-1);
396 }
397 return;
398 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
399 /* fall-through */
400 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
401 for (ch = 0; ch < 4; ch++) {
402 icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
403 }
404 return;
405 default:
406 assert(0);
407 }
408 }
409
410
411 /**
412 * For RECT textures / unnormalized texcoords.
413 * Only a subset of wrap modes supported.
414 */
415 static INLINE void
416 linear_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
417 int icoord0[4], int icoord1[4], float w[4])
418 {
419 uint ch;
420 switch (wrapMode) {
421 case PIPE_TEX_WRAP_CLAMP:
422 for (ch = 0; ch < 4; ch++) {
423 /* Not exactly what the spec says, but it matches NVIDIA output */
424 float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
425 icoord0[ch] = util_ifloor(u);
426 icoord1[ch] = icoord0[ch] + 1;
427 w[ch] = FRAC(u);
428 }
429 return;
430 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
431 /* fall-through */
432 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
433 for (ch = 0; ch < 4; ch++) {
434 float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F);
435 u -= 0.5F;
436 icoord0[ch] = util_ifloor(u);
437 icoord1[ch] = icoord0[ch] + 1;
438 if (icoord1[ch] > (int) size - 1)
439 icoord1[ch] = size - 1;
440 w[ch] = FRAC(u);
441 }
442 break;
443 default:
444 assert(0);
445 }
446 }
447
448
449 static unsigned
450 choose_cube_face(float rx, float ry, float rz, float *newS, float *newT)
451 {
452 /*
453 major axis
454 direction target sc tc ma
455 ---------- ------------------------------- --- --- ---
456 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
457 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
458 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
459 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
460 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
461 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
462 */
463 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
464 unsigned face;
465 float sc, tc, ma;
466
467 if (arx > ary && arx > arz) {
468 if (rx >= 0.0F) {
469 face = PIPE_TEX_FACE_POS_X;
470 sc = -rz;
471 tc = -ry;
472 ma = arx;
473 }
474 else {
475 face = PIPE_TEX_FACE_NEG_X;
476 sc = rz;
477 tc = -ry;
478 ma = arx;
479 }
480 }
481 else if (ary > arx && ary > arz) {
482 if (ry >= 0.0F) {
483 face = PIPE_TEX_FACE_POS_Y;
484 sc = rx;
485 tc = rz;
486 ma = ary;
487 }
488 else {
489 face = PIPE_TEX_FACE_NEG_Y;
490 sc = rx;
491 tc = -rz;
492 ma = ary;
493 }
494 }
495 else {
496 if (rz > 0.0F) {
497 face = PIPE_TEX_FACE_POS_Z;
498 sc = rx;
499 tc = -ry;
500 ma = arz;
501 }
502 else {
503 face = PIPE_TEX_FACE_NEG_Z;
504 sc = -rx;
505 tc = -ry;
506 ma = arz;
507 }
508 }
509
510 *newS = ( sc / ma + 1.0F ) * 0.5F;
511 *newT = ( tc / ma + 1.0F ) * 0.5F;
512
513 return face;
514 }
515
516
517 /**
518 * Examine the quad's texture coordinates to compute the partial
519 * derivatives w.r.t X and Y, then compute lambda (level of detail).
520 *
521 * This is only done for fragment shaders, not vertex shaders.
522 */
523 static float
524 compute_lambda(const struct pipe_texture *tex,
525 const struct pipe_sampler_state *sampler,
526 const float s[QUAD_SIZE],
527 const float t[QUAD_SIZE],
528 const float p[QUAD_SIZE],
529 float lodbias)
530 {
531 float rho, lambda;
532
533 assert(sampler->normalized_coords);
534
535 assert(s);
536 {
537 float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT];
538 float dsdy = s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT];
539 dsdx = fabsf(dsdx);
540 dsdy = fabsf(dsdy);
541 rho = MAX2(dsdx, dsdy) * tex->width[0];
542 }
543 if (t) {
544 float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT];
545 float dtdy = t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT];
546 float max;
547 dtdx = fabsf(dtdx);
548 dtdy = fabsf(dtdy);
549 max = MAX2(dtdx, dtdy) * tex->height[0];
550 rho = MAX2(rho, max);
551 }
552 if (p) {
553 float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT];
554 float dpdy = p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT];
555 float max;
556 dpdx = fabsf(dpdx);
557 dpdy = fabsf(dpdy);
558 max = MAX2(dpdx, dpdy) * tex->depth[0];
559 rho = MAX2(rho, max);
560 }
561
562 lambda = util_fast_log2(rho);
563 lambda += lodbias + sampler->lod_bias;
564 lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
565
566 return lambda;
567 }
568
569
570 /**
571 * Do several things here:
572 * 1. Compute lambda from the texcoords, if needed
573 * 2. Determine if we're minifying or magnifying
574 * 3. If minifying, choose mipmap levels
575 * 4. Return image filter to use within mipmap images
576 * \param level0 Returns first mipmap level to sample from
577 * \param level1 Returns second mipmap level to sample from
578 * \param levelBlend Returns blend factor between levels, in [0,1]
579 * \param imgFilter Returns either the min or mag filter, depending on lambda
580 */
581 static void
582 choose_mipmap_levels(const struct pipe_texture *texture,
583 const struct pipe_sampler_state *sampler,
584 const float s[QUAD_SIZE],
585 const float t[QUAD_SIZE],
586 const float p[QUAD_SIZE],
587 boolean computeLambda,
588 float lodbias,
589 unsigned *level0, unsigned *level1, float *levelBlend,
590 unsigned *imgFilter)
591 {
592 if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
593 /* no mipmap selection needed */
594 *level0 = *level1 = CLAMP((int) sampler->min_lod,
595 0, (int) texture->last_level);
596
597 if (sampler->min_img_filter != sampler->mag_img_filter) {
598 /* non-mipmapped texture, but still need to determine if doing
599 * minification or magnification.
600 */
601 float lambda = compute_lambda(texture, sampler, s, t, p, lodbias);
602 if (lambda <= 0.0) {
603 *imgFilter = sampler->mag_img_filter;
604 }
605 else {
606 *imgFilter = sampler->min_img_filter;
607 }
608 }
609 else {
610 *imgFilter = sampler->mag_img_filter;
611 }
612 }
613 else {
614 float lambda;
615
616 if (computeLambda)
617 /* fragment shader */
618 lambda = compute_lambda(texture, sampler, s, t, p, lodbias);
619 else
620 /* vertex shader */
621 lambda = lodbias; /* not really a bias, but absolute LOD */
622
623 if (lambda <= 0.0) { /* XXX threshold depends on the filter */
624 /* magnifying */
625 *imgFilter = sampler->mag_img_filter;
626 *level0 = *level1 = 0;
627 }
628 else {
629 /* minifying */
630 *imgFilter = sampler->min_img_filter;
631
632 /* choose mipmap level(s) and compute the blend factor between them */
633 if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
634 /* Nearest mipmap level */
635 const int lvl = (int) (lambda + 0.5);
636 *level0 =
637 *level1 = CLAMP(lvl, 0, (int) texture->last_level);
638 }
639 else {
640 /* Linear interpolation between mipmap levels */
641 const int lvl = (int) lambda;
642 *level0 = CLAMP(lvl, 0, (int) texture->last_level);
643 *level1 = CLAMP(lvl + 1, 0, (int) texture->last_level);
644 *levelBlend = FRAC(lambda); /* blending weight between levels */
645 }
646 }
647 }
648 }
649
650
651 /**
652 * Get a texel from a texture, using the texture tile cache.
653 *
654 * \param face the cube face in 0..5
655 * \param level the mipmap level
656 * \param x the x coord of texel within 2D image
657 * \param y the y coord of texel within 2D image
658 * \param z which slice of a 3D texture
659 * \param rgba the quad to put the texel/color into
660 * \param j which element of the rgba quad to write to
661 *
662 * XXX maybe move this into sp_tile_cache.c and merge with the
663 * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1...
664 */
665 static void
666 get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler,
667 unsigned face, unsigned level, int x, int y,
668 const float *out[4])
669 {
670 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
671
672 const struct softpipe_cached_tile *tile
673 = sp_get_cached_tile_tex(samp->cache,
674 tile_address(x, y, 0, face, level));
675
676 y %= TILE_SIZE;
677 x %= TILE_SIZE;
678
679 out[0] = &tile->data.color[y ][x ][0];
680 out[1] = &tile->data.color[y ][x+1][0];
681 out[2] = &tile->data.color[y+1][x ][0];
682 out[3] = &tile->data.color[y+1][x+1][0];
683 }
684
685 static INLINE const float *
686 get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler,
687 unsigned face, unsigned level, int x, int y)
688 {
689 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
690
691 const struct softpipe_cached_tile *tile
692 = sp_get_cached_tile_tex(samp->cache,
693 tile_address(x, y, 0, face, level));
694
695 y %= TILE_SIZE;
696 x %= TILE_SIZE;
697
698 return &tile->data.color[y][x][0];
699 }
700
701
702 static void
703 get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler,
704 unsigned face, unsigned level,
705 int x0, int y0,
706 int x1, int y1,
707 const float *out[4])
708 {
709 unsigned i;
710
711 for (i = 0; i < 4; i++) {
712 unsigned tx = (i & 1) ? x1 : x0;
713 unsigned ty = (i >> 1) ? y1 : y0;
714
715 out[i] = get_texel_2d_ptr( tgsi_sampler, face, level, tx, ty );
716 }
717 }
718
719 static void
720 get_texel(const struct tgsi_sampler *tgsi_sampler,
721 unsigned face, unsigned level, int x, int y, int z,
722 float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j)
723 {
724 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
725 const struct pipe_texture *texture = samp->texture;
726 const struct pipe_sampler_state *sampler = samp->sampler;
727
728 if (x < 0 || x >= (int) texture->width[level] ||
729 y < 0 || y >= (int) texture->height[level] ||
730 z < 0 || z >= (int) texture->depth[level]) {
731 rgba[0][j] = sampler->border_color[0];
732 rgba[1][j] = sampler->border_color[1];
733 rgba[2][j] = sampler->border_color[2];
734 rgba[3][j] = sampler->border_color[3];
735 }
736 else {
737 const unsigned tx = x % TILE_SIZE;
738 const unsigned ty = y % TILE_SIZE;
739 const struct softpipe_cached_tile *tile;
740
741 tile = sp_get_cached_tile_tex(samp->cache,
742 tile_address(x, y, z, face, level));
743
744 rgba[0][j] = tile->data.color[ty][tx][0];
745 rgba[1][j] = tile->data.color[ty][tx][1];
746 rgba[2][j] = tile->data.color[ty][tx][2];
747 rgba[3][j] = tile->data.color[ty][tx][3];
748 if (0)
749 {
750 debug_printf("Get texel %f %f %f %f from %s\n",
751 rgba[0][j], rgba[1][j], rgba[2][j], rgba[3][j],
752 pf_name(texture->format));
753 }
754 }
755 }
756
757
758 /**
759 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
760 * When we sampled the depth texture, the depth value was put into all
761 * RGBA channels. We look at the red channel here.
762 * \param rgba quad of (depth) texel values
763 * \param p texture 'P' components for four pixels in quad
764 * \param j which pixel in the quad to test [0..3]
765 */
766 static INLINE void
767 shadow_compare(const struct pipe_sampler_state *sampler,
768 float rgba[NUM_CHANNELS][QUAD_SIZE],
769 const float p[QUAD_SIZE],
770 uint j)
771 {
772 int k;
773 switch (sampler->compare_func) {
774 case PIPE_FUNC_LESS:
775 k = p[j] < rgba[0][j];
776 break;
777 case PIPE_FUNC_LEQUAL:
778 k = p[j] <= rgba[0][j];
779 break;
780 case PIPE_FUNC_GREATER:
781 k = p[j] > rgba[0][j];
782 break;
783 case PIPE_FUNC_GEQUAL:
784 k = p[j] >= rgba[0][j];
785 break;
786 case PIPE_FUNC_EQUAL:
787 k = p[j] == rgba[0][j];
788 break;
789 case PIPE_FUNC_NOTEQUAL:
790 k = p[j] != rgba[0][j];
791 break;
792 case PIPE_FUNC_ALWAYS:
793 k = 1;
794 break;
795 case PIPE_FUNC_NEVER:
796 k = 0;
797 break;
798 default:
799 k = 0;
800 assert(0);
801 break;
802 }
803
804 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
805 rgba[0][j] = rgba[1][j] = rgba[2][j] = (float) k;
806 rgba[3][j] = 1.0F;
807 }
808
809
810 /**
811 * As above, but do four z/texture comparisons.
812 */
813 static INLINE void
814 shadow_compare4(const struct pipe_sampler_state *sampler,
815 float rgba[NUM_CHANNELS][QUAD_SIZE],
816 const float p[QUAD_SIZE])
817 {
818 int j, k0, k1, k2, k3;
819 float val;
820
821 /* compare four texcoords vs. four texture samples */
822 switch (sampler->compare_func) {
823 case PIPE_FUNC_LESS:
824 k0 = p[0] < rgba[0][0];
825 k1 = p[1] < rgba[0][1];
826 k2 = p[2] < rgba[0][2];
827 k3 = p[3] < rgba[0][3];
828 break;
829 case PIPE_FUNC_LEQUAL:
830 k0 = p[0] <= rgba[0][0];
831 k1 = p[1] <= rgba[0][1];
832 k2 = p[2] <= rgba[0][2];
833 k3 = p[3] <= rgba[0][3];
834 break;
835 case PIPE_FUNC_GREATER:
836 k0 = p[0] > rgba[0][0];
837 k1 = p[1] > rgba[0][1];
838 k2 = p[2] > rgba[0][2];
839 k3 = p[3] > rgba[0][3];
840 break;
841 case PIPE_FUNC_GEQUAL:
842 k0 = p[0] >= rgba[0][0];
843 k1 = p[1] >= rgba[0][1];
844 k2 = p[2] >= rgba[0][2];
845 k3 = p[3] >= rgba[0][3];
846 break;
847 case PIPE_FUNC_EQUAL:
848 k0 = p[0] == rgba[0][0];
849 k1 = p[1] == rgba[0][1];
850 k2 = p[2] == rgba[0][2];
851 k3 = p[3] == rgba[0][3];
852 break;
853 case PIPE_FUNC_NOTEQUAL:
854 k0 = p[0] != rgba[0][0];
855 k1 = p[1] != rgba[0][1];
856 k2 = p[2] != rgba[0][2];
857 k3 = p[3] != rgba[0][3];
858 break;
859 case PIPE_FUNC_ALWAYS:
860 k0 = k1 = k2 = k3 = 1;
861 break;
862 case PIPE_FUNC_NEVER:
863 k0 = k1 = k2 = k3 = 0;
864 break;
865 default:
866 k0 = k1 = k2 = k3 = 0;
867 assert(0);
868 break;
869 }
870
871 /* convert four pass/fail values to an intensity in [0,1] */
872 val = 0.25F * (k0 + k1 + k2 + k3);
873
874 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
875 for (j = 0; j < 4; j++) {
876 rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
877 rgba[3][j] = 1.0F;
878 }
879 }
880
881
882
883 static void
884 sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
885 const float s[QUAD_SIZE],
886 const float t[QUAD_SIZE],
887 const float p[QUAD_SIZE],
888 float lodbias,
889 float rgba[NUM_CHANNELS][QUAD_SIZE])
890 {
891 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
892 unsigned j;
893 unsigned level = samp->level;
894 unsigned xpot = 1 << (samp->xpot - level);
895 unsigned ypot = 1 << (samp->ypot - level);
896
897 for (j = 0; j < QUAD_SIZE; j++) {
898 int c;
899
900 float u = s[j] * xpot - 0.5F;
901 float v = t[j] * ypot - 0.5F;
902
903 int uflr = util_ifloor(u);
904 int vflr = util_ifloor(v);
905
906 float xw = u - (float)uflr;
907 float yw = v - (float)vflr;
908
909 int x0 = uflr & (xpot - 1);
910 int y0 = vflr & (ypot - 1);
911
912 const float *tx[4];
913
914
915 /* Can we fetch all four at once:
916 */
917 if (x0 % TILE_SIZE != TILE_SIZE-1 &&
918 y0 % TILE_SIZE != TILE_SIZE-1)
919 {
920 get_texel_quad_2d(tgsi_sampler, 0, level, x0, y0, tx);
921 }
922 else
923 {
924 unsigned x1 = (uflr + 1) & (xpot - 1);
925 unsigned y1 = (vflr + 1) & (ypot - 1);
926 get_texel_quad_2d_mt(tgsi_sampler, 0, level,
927 x0, y0, x1, y1, tx);
928 }
929
930
931 /* interpolate R, G, B, A */
932 for (c = 0; c < 4; c++) {
933 rgba[c][j] = lerp_2d(xw, yw,
934 tx[0][c], tx[1][c],
935 tx[2][c], tx[3][c]);
936 }
937 }
938 }
939
940
941 static void
942 sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
943 const float s[QUAD_SIZE],
944 const float t[QUAD_SIZE],
945 const float p[QUAD_SIZE],
946 float lodbias,
947 float rgba[NUM_CHANNELS][QUAD_SIZE])
948 {
949 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
950 unsigned j;
951 unsigned level = samp->level;
952 unsigned xpot = 1 << (samp->xpot - level);
953 unsigned ypot = 1 << (samp->ypot - level);
954
955 for (j = 0; j < QUAD_SIZE; j++) {
956 int c;
957
958 float u = s[j] * xpot;
959 float v = t[j] * ypot;
960
961 int uflr = util_ifloor(u);
962 int vflr = util_ifloor(v);
963
964 int x0 = uflr & (xpot - 1);
965 int y0 = vflr & (ypot - 1);
966
967 const float *out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0);
968
969 for (c = 0; c < 4; c++) {
970 rgba[c][j] = out[c];
971 }
972 }
973 }
974
975
976 static void
977 sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
978 const float s[QUAD_SIZE],
979 const float t[QUAD_SIZE],
980 const float p[QUAD_SIZE],
981 float lodbias,
982 float rgba[NUM_CHANNELS][QUAD_SIZE])
983 {
984 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
985 unsigned j;
986 unsigned level = samp->level;
987 unsigned xpot = (1<<samp->xpot);
988 unsigned ypot = (1<<samp->ypot);
989
990 for (j = 0; j < QUAD_SIZE; j++) {
991 int c;
992
993 float u = s[j] * xpot;
994 float v = t[j] * ypot;
995
996 int x0, y0;
997 const float *out;
998
999 x0 = util_ifloor(u);
1000 if (x0 < 0)
1001 x0 = 0;
1002 else if (x0 > xpot - 1)
1003 x0 = xpot - 1;
1004
1005 y0 = util_ifloor(v);
1006 if (y0 < 0)
1007 y0 = 0;
1008 else if (y0 > ypot - 1)
1009 y0 = ypot - 1;
1010
1011 out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0);
1012
1013 for (c = 0; c < 4; c++) {
1014 rgba[c][j] = out[c];
1015 }
1016 }
1017 }
1018
1019
1020 static void
1021 sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
1022 const float s[QUAD_SIZE],
1023 const float t[QUAD_SIZE],
1024 const float p[QUAD_SIZE],
1025 float lodbias,
1026 float rgba[NUM_CHANNELS][QUAD_SIZE])
1027 {
1028 struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1029 const struct pipe_texture *texture = samp->texture;
1030 const struct pipe_sampler_state *sampler = samp->sampler;
1031 int level0, level1;
1032 float lambda;
1033
1034 lambda = compute_lambda(texture, sampler, s, t, p, lodbias);
1035 level0 = (int)lambda;
1036 level1 = level0 + 1;
1037
1038 if (lambda < 0.0) {
1039 samp->level = 0;
1040 sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
1041 s, t, p, lodbias, rgba );
1042 }
1043 else if (level0 >= texture->last_level) {
1044 samp->level = texture->last_level;
1045 sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
1046 s, t, p, lodbias, rgba );
1047 }
1048 else {
1049 float rgba0[4][4];
1050 float rgba1[4][4];
1051 int c,j;
1052
1053 float levelBlend = lambda - level0; /* blending weight between levels */
1054
1055 samp->level = level0;
1056 sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
1057 s, t, p, lodbias, rgba0 );
1058
1059 samp->level++;
1060 sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
1061 s, t, p, lodbias, rgba1 );
1062
1063 for (c = 0; c < 4; c++)
1064 for (j = 0; j < 4; j++)
1065 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1066 }
1067 }
1068
1069 /**
1070 * Common code for sampling 1D/2D/cube textures.
1071 * Could probably extend for 3D...
1072 */
1073 static void
1074 sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler,
1075 const float s[QUAD_SIZE],
1076 const float t[QUAD_SIZE],
1077 const float p[QUAD_SIZE],
1078 boolean computeLambda,
1079 float lodbias,
1080 float rgba[NUM_CHANNELS][QUAD_SIZE],
1081 const unsigned faces[4])
1082 {
1083 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1084 const struct pipe_texture *texture = samp->texture;
1085 const struct pipe_sampler_state *sampler = samp->sampler;
1086 unsigned level0, level1, j, imgFilter;
1087 int width, height;
1088 float levelBlend;
1089
1090 choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
1091 &level0, &level1, &levelBlend, &imgFilter);
1092
1093 assert(sampler->normalized_coords);
1094
1095 width = texture->width[level0];
1096 height = texture->height[level0];
1097
1098 assert(width > 0);
1099
1100 switch (imgFilter) {
1101 case PIPE_TEX_FILTER_NEAREST:
1102 {
1103 int x[4], y[4];
1104 nearest_texcoord_4(sampler->wrap_s, s, width, x);
1105 nearest_texcoord_4(sampler->wrap_t, t, height, y);
1106
1107 for (j = 0; j < QUAD_SIZE; j++) {
1108 get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j);
1109 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1110 shadow_compare(sampler, rgba, p, j);
1111 }
1112
1113 if (level0 != level1) {
1114 /* get texels from second mipmap level and blend */
1115 float rgba2[4][4];
1116 unsigned c;
1117 x[j] /= 2;
1118 y[j] /= 2;
1119 get_texel(tgsi_sampler, faces[j], level1, x[j], y[j], 0,
1120 rgba2, j);
1121 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
1122 shadow_compare(sampler, rgba2, p, j);
1123 }
1124
1125 for (c = 0; c < NUM_CHANNELS; c++) {
1126 rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
1127 }
1128 }
1129 }
1130 }
1131 break;
1132 case PIPE_TEX_FILTER_LINEAR:
1133 case PIPE_TEX_FILTER_ANISO:
1134 {
1135 int x0[4], y0[4], x1[4], y1[4];
1136 float xw[4], yw[4]; /* weights */
1137
1138 linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw);
1139 linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
1140
1141 for (j = 0; j < QUAD_SIZE; j++) {
1142 float tx[4][4]; /* texels */
1143 int c;
1144 get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0);
1145 get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1);
1146 get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2);
1147 get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3);
1148 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1149 shadow_compare4(sampler, tx, p);
1150 }
1151
1152 /* interpolate R, G, B, A */
1153 for (c = 0; c < 4; c++) {
1154 rgba[c][j] = lerp_2d(xw[j], yw[j],
1155 tx[c][0], tx[c][1],
1156 tx[c][2], tx[c][3]);
1157 }
1158
1159 if (level0 != level1) {
1160 /* get texels from second mipmap level and blend */
1161 float rgba2[4][4];
1162 x0[j] /= 2;
1163 y0[j] /= 2;
1164 x1[j] /= 2;
1165 y1[j] /= 2;
1166 get_texel(tgsi_sampler, faces[j], level1, x0[j], y0[j], 0, tx, 0);
1167 get_texel(tgsi_sampler, faces[j], level1, x1[j], y0[j], 0, tx, 1);
1168 get_texel(tgsi_sampler, faces[j], level1, x0[j], y1[j], 0, tx, 2);
1169 get_texel(tgsi_sampler, faces[j], level1, x1[j], y1[j], 0, tx, 3);
1170 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
1171 shadow_compare4(sampler, tx, p);
1172 }
1173
1174 /* interpolate R, G, B, A */
1175 for (c = 0; c < 4; c++) {
1176 rgba2[c][j] = lerp_2d(xw[j], yw[j],
1177 tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
1178 }
1179
1180 for (c = 0; c < NUM_CHANNELS; c++) {
1181 rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
1182 }
1183 }
1184 }
1185 }
1186 break;
1187 default:
1188 assert(0);
1189 }
1190 }
1191
1192
1193 static INLINE void
1194 sp_get_samples_1d(const struct tgsi_sampler *sampler,
1195 const float s[QUAD_SIZE],
1196 const float t[QUAD_SIZE],
1197 const float p[QUAD_SIZE],
1198 boolean computeLambda,
1199 float lodbias,
1200 float rgba[NUM_CHANNELS][QUAD_SIZE])
1201 {
1202 static const unsigned faces[4] = {0, 0, 0, 0};
1203 static const float tzero[4] = {0, 0, 0, 0};
1204 sp_get_samples_2d_common(sampler, s, tzero, NULL,
1205 computeLambda, lodbias, rgba, faces);
1206 }
1207
1208
1209 static INLINE void
1210 sp_get_samples_2d(const struct tgsi_sampler *sampler,
1211 const float s[QUAD_SIZE],
1212 const float t[QUAD_SIZE],
1213 const float p[QUAD_SIZE],
1214 boolean computeLambda,
1215 float lodbias,
1216 float rgba[NUM_CHANNELS][QUAD_SIZE])
1217 {
1218 static const unsigned faces[4] = {0, 0, 0, 0};
1219 sp_get_samples_2d_common(sampler, s, t, p,
1220 computeLambda, lodbias, rgba, faces);
1221 }
1222
1223
1224 static INLINE void
1225 sp_get_samples_3d(const struct tgsi_sampler *tgsi_sampler,
1226 const float s[QUAD_SIZE],
1227 const float t[QUAD_SIZE],
1228 const float p[QUAD_SIZE],
1229 boolean computeLambda,
1230 float lodbias,
1231 float rgba[NUM_CHANNELS][QUAD_SIZE])
1232 {
1233 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1234 const struct pipe_texture *texture = samp->texture;
1235 const struct pipe_sampler_state *sampler = samp->sampler;
1236 /* get/map pipe_surfaces corresponding to 3D tex slices */
1237 unsigned level0, level1, j, imgFilter;
1238 int width, height, depth;
1239 float levelBlend;
1240 const uint face = 0;
1241
1242 choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
1243 &level0, &level1, &levelBlend, &imgFilter);
1244
1245 assert(sampler->normalized_coords);
1246
1247 width = texture->width[level0];
1248 height = texture->height[level0];
1249 depth = texture->depth[level0];
1250
1251 assert(width > 0);
1252 assert(height > 0);
1253 assert(depth > 0);
1254
1255 switch (imgFilter) {
1256 case PIPE_TEX_FILTER_NEAREST:
1257 {
1258 int x[4], y[4], z[4];
1259 nearest_texcoord_4(sampler->wrap_s, s, width, x);
1260 nearest_texcoord_4(sampler->wrap_t, t, height, y);
1261 nearest_texcoord_4(sampler->wrap_r, p, depth, z);
1262 for (j = 0; j < QUAD_SIZE; j++) {
1263 get_texel(tgsi_sampler, face, level0, x[j], y[j], z[j], rgba, j);
1264 if (level0 != level1) {
1265 /* get texels from second mipmap level and blend */
1266 float rgba2[4][4];
1267 unsigned c;
1268 x[j] /= 2;
1269 y[j] /= 2;
1270 z[j] /= 2;
1271 get_texel(tgsi_sampler, face, level1, x[j], y[j], z[j], rgba2, j);
1272 for (c = 0; c < NUM_CHANNELS; c++) {
1273 rgba[c][j] = lerp(levelBlend, rgba2[c][j], rgba[c][j]);
1274 }
1275 }
1276 }
1277 }
1278 break;
1279 case PIPE_TEX_FILTER_LINEAR:
1280 case PIPE_TEX_FILTER_ANISO:
1281 {
1282 int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
1283 float xw[4], yw[4], zw[4]; /* interpolation weights */
1284 linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw);
1285 linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
1286 linear_texcoord_4(sampler->wrap_r, p, depth, z0, z1, zw);
1287
1288 for (j = 0; j < QUAD_SIZE; j++) {
1289 int c;
1290 float tx0[4][4], tx1[4][4];
1291 get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z0[j], tx0, 0);
1292 get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z0[j], tx0, 1);
1293 get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z0[j], tx0, 2);
1294 get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z0[j], tx0, 3);
1295 get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z1[j], tx1, 0);
1296 get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z1[j], tx1, 1);
1297 get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z1[j], tx1, 2);
1298 get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z1[j], tx1, 3);
1299
1300 /* interpolate R, G, B, A */
1301 for (c = 0; c < 4; c++) {
1302 rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1303 tx0[c][0], tx0[c][1],
1304 tx0[c][2], tx0[c][3],
1305 tx1[c][0], tx1[c][1],
1306 tx1[c][2], tx1[c][3]);
1307 }
1308
1309 if (level0 != level1) {
1310 /* get texels from second mipmap level and blend */
1311 float rgba2[4][4];
1312 x0[j] /= 2;
1313 y0[j] /= 2;
1314 z0[j] /= 2;
1315 x1[j] /= 2;
1316 y1[j] /= 2;
1317 z1[j] /= 2;
1318 get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z0[j], tx0, 0);
1319 get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z0[j], tx0, 1);
1320 get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z0[j], tx0, 2);
1321 get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z0[j], tx0, 3);
1322 get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z1[j], tx1, 0);
1323 get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z1[j], tx1, 1);
1324 get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z1[j], tx1, 2);
1325 get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z1[j], tx1, 3);
1326
1327 /* interpolate R, G, B, A */
1328 for (c = 0; c < 4; c++) {
1329 rgba2[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1330 tx0[c][0], tx0[c][1],
1331 tx0[c][2], tx0[c][3],
1332 tx1[c][0], tx1[c][1],
1333 tx1[c][2], tx1[c][3]);
1334 }
1335
1336 /* blend mipmap levels */
1337 for (c = 0; c < NUM_CHANNELS; c++) {
1338 rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
1339 }
1340 }
1341 }
1342 }
1343 break;
1344 default:
1345 assert(0);
1346 }
1347 }
1348
1349
1350 static void
1351 sp_get_samples_cube(const struct tgsi_sampler *sampler,
1352 const float s[QUAD_SIZE],
1353 const float t[QUAD_SIZE],
1354 const float p[QUAD_SIZE],
1355 boolean computeLambda,
1356 float lodbias,
1357 float rgba[NUM_CHANNELS][QUAD_SIZE])
1358 {
1359 unsigned faces[QUAD_SIZE], j;
1360 float ssss[4], tttt[4];
1361 for (j = 0; j < QUAD_SIZE; j++) {
1362 faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j);
1363 }
1364 sp_get_samples_2d_common(sampler, ssss, tttt, NULL,
1365 computeLambda, lodbias, rgba, faces);
1366 }
1367
1368
1369 static void
1370 sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler,
1371 const float s[QUAD_SIZE],
1372 const float t[QUAD_SIZE],
1373 const float p[QUAD_SIZE],
1374 boolean computeLambda,
1375 float lodbias,
1376 float rgba[NUM_CHANNELS][QUAD_SIZE])
1377 {
1378 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1379 const struct pipe_texture *texture = samp->texture;
1380 const struct pipe_sampler_state *sampler = samp->sampler;
1381 const uint face = 0;
1382 unsigned level0, level1, j, imgFilter;
1383 int width, height;
1384 float levelBlend;
1385
1386 choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
1387 &level0, &level1, &levelBlend, &imgFilter);
1388
1389 /* texture RECTS cannot be mipmapped */
1390 assert(level0 == level1);
1391
1392 width = texture->width[level0];
1393 height = texture->height[level0];
1394
1395 assert(width > 0);
1396
1397 switch (imgFilter) {
1398 case PIPE_TEX_FILTER_NEAREST:
1399 {
1400 int x[4], y[4];
1401 nearest_texcoord_unnorm_4(sampler->wrap_s, s, width, x);
1402 nearest_texcoord_unnorm_4(sampler->wrap_t, t, height, y);
1403 for (j = 0; j < QUAD_SIZE; j++) {
1404 get_texel(tgsi_sampler, face, level0, x[j], y[j], 0, rgba, j);
1405 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1406 shadow_compare(sampler, rgba, p, j);
1407 }
1408 }
1409 }
1410 break;
1411 case PIPE_TEX_FILTER_LINEAR:
1412 case PIPE_TEX_FILTER_ANISO:
1413 {
1414 int x0[4], y0[4], x1[4], y1[4];
1415 float xw[4], yw[4]; /* weights */
1416 linear_texcoord_unnorm_4(sampler->wrap_s, s, width, x0, x1, xw);
1417 linear_texcoord_unnorm_4(sampler->wrap_t, t, height, y0, y1, yw);
1418 for (j = 0; j < QUAD_SIZE; j++) {
1419 float tx[4][4]; /* texels */
1420 int c;
1421 get_texel(tgsi_sampler, face, level0, x0[j], y0[j], 0, tx, 0);
1422 get_texel(tgsi_sampler, face, level0, x1[j], y0[j], 0, tx, 1);
1423 get_texel(tgsi_sampler, face, level0, x0[j], y1[j], 0, tx, 2);
1424 get_texel(tgsi_sampler, face, level0, x1[j], y1[j], 0, tx, 3);
1425 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1426 shadow_compare4(sampler, tx, p);
1427 }
1428 for (c = 0; c < 4; c++) {
1429 rgba[c][j] = lerp_2d(xw[j], yw[j],
1430 tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
1431 }
1432 }
1433 }
1434 break;
1435 default:
1436 assert(0);
1437 }
1438 }
1439
1440
1441 /**
1442 * Common code for vertex/fragment program texture sampling.
1443 */
1444 static INLINE void
1445 sp_get_samples(struct tgsi_sampler *tgsi_sampler,
1446 const float s[QUAD_SIZE],
1447 const float t[QUAD_SIZE],
1448 const float p[QUAD_SIZE],
1449 boolean computeLambda,
1450 float lodbias,
1451 float rgba[NUM_CHANNELS][QUAD_SIZE])
1452 {
1453 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1454 const struct pipe_texture *texture = samp->texture;
1455 const struct pipe_sampler_state *sampler = samp->sampler;
1456
1457 if (!texture)
1458 return;
1459
1460 switch (texture->target) {
1461 case PIPE_TEXTURE_1D:
1462 assert(sampler->normalized_coords);
1463 sp_get_samples_1d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1464 break;
1465 case PIPE_TEXTURE_2D:
1466 if (sampler->normalized_coords)
1467 sp_get_samples_2d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1468 else
1469 sp_get_samples_rect(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1470 break;
1471 case PIPE_TEXTURE_3D:
1472 assert(sampler->normalized_coords);
1473 sp_get_samples_3d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1474 break;
1475 case PIPE_TEXTURE_CUBE:
1476 assert(sampler->normalized_coords);
1477 sp_get_samples_cube(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1478 break;
1479 default:
1480 assert(0);
1481 }
1482
1483 #if 0 /* DEBUG */
1484 {
1485 int i;
1486 printf("Sampled at %f, %f, %f:\n", s[0], t[0], p[0]);
1487 for (i = 0; i < 4; i++) {
1488 printf("Frag %d: %f %f %f %f\n", i,
1489 rgba[0][i],
1490 rgba[1][i],
1491 rgba[2][i],
1492 rgba[3][i]);
1493 }
1494 }
1495 #endif
1496 }
1497
1498 static void
1499 sp_get_samples_fallback(struct tgsi_sampler *tgsi_sampler,
1500 const float s[QUAD_SIZE],
1501 const float t[QUAD_SIZE],
1502 const float p[QUAD_SIZE],
1503 float lodbias,
1504 float rgba[NUM_CHANNELS][QUAD_SIZE])
1505 {
1506 sp_get_samples(tgsi_sampler, s, t, p, TRUE, lodbias, rgba);
1507 }
1508
1509 /**
1510 * Called via tgsi_sampler::get_samples() when running a fragment shader.
1511 * Get four filtered RGBA values from the sampler's texture.
1512 */
1513 void
1514 sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler,
1515 const float s[QUAD_SIZE],
1516 const float t[QUAD_SIZE],
1517 const float p[QUAD_SIZE],
1518 float lodbias,
1519 float rgba[NUM_CHANNELS][QUAD_SIZE])
1520 {
1521 struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1522 const struct pipe_texture *texture = samp->texture;
1523 const struct pipe_sampler_state *sampler = samp->sampler;
1524
1525 tgsi_sampler->get_samples = sp_get_samples_fallback;
1526
1527 /* Try to hook in a faster sampler. Ultimately we'll have to
1528 * code-generate these. Luckily most of this looks like it is
1529 * orthogonal state within the sampler.
1530 */
1531 if (texture->target == PIPE_TEXTURE_2D &&
1532 sampler->min_img_filter == sampler->mag_img_filter &&
1533 sampler->wrap_s == sampler->wrap_t &&
1534 sampler->compare_mode == FALSE &&
1535 sampler->normalized_coords)
1536 {
1537 samp->xpot = util_unsigned_logbase2( samp->texture->width[0] );
1538 samp->ypot = util_unsigned_logbase2( samp->texture->height[0] );
1539
1540 if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
1541 samp->level = CLAMP((int) sampler->min_lod,
1542 0, (int) texture->last_level);
1543
1544 if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) {
1545 switch (sampler->min_img_filter) {
1546 case PIPE_TEX_FILTER_NEAREST:
1547 tgsi_sampler->get_samples = sp_get_samples_2d_nearest_repeat_POT;
1548 break;
1549 case PIPE_TEX_FILTER_LINEAR:
1550 tgsi_sampler->get_samples = sp_get_samples_2d_linear_repeat_POT;
1551 break;
1552 default:
1553 break;
1554 }
1555 }
1556 else if (sampler->wrap_s == PIPE_TEX_WRAP_CLAMP) {
1557 switch (sampler->min_img_filter) {
1558 case PIPE_TEX_FILTER_NEAREST:
1559 tgsi_sampler->get_samples = sp_get_samples_2d_nearest_clamp_POT;
1560 break;
1561 default:
1562 break;
1563 }
1564 }
1565 }
1566 else if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
1567 if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) {
1568 switch (sampler->min_img_filter) {
1569 case PIPE_TEX_FILTER_LINEAR:
1570 /* This one not working yet:
1571 */
1572 if (0)
1573 tgsi_sampler->get_samples = sp_get_samples_2d_linear_mip_linear_repeat_POT;
1574 break;
1575 default:
1576 break;
1577 }
1578 }
1579 }
1580 }
1581 else if (0) {
1582 _debug_printf("target %d/%d min_mip %d/%d min_img %d/%d wrap %d/%d compare %d/%d norm %d/%d\n",
1583 texture->target, PIPE_TEXTURE_2D,
1584 sampler->min_mip_filter, PIPE_TEX_MIPFILTER_NONE,
1585 sampler->min_img_filter, sampler->mag_img_filter,
1586 sampler->wrap_s, sampler->wrap_t,
1587 sampler->compare_mode, FALSE,
1588 sampler->normalized_coords, TRUE);
1589 }
1590
1591 tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba );
1592 }
1593
1594
1595 /**
1596 * Called via tgsi_sampler::get_samples() when running a vertex shader.
1597 * Get four filtered RGBA values from the sampler's texture.
1598 */
1599 void
1600 sp_get_samples_vertex(struct tgsi_sampler *tgsi_sampler,
1601 const float s[QUAD_SIZE],
1602 const float t[QUAD_SIZE],
1603 const float p[QUAD_SIZE],
1604 float lodbias,
1605 float rgba[NUM_CHANNELS][QUAD_SIZE])
1606 {
1607 sp_get_samples(tgsi_sampler, s, t, p, FALSE, lodbias, rgba);
1608 }