projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
softpipe: fix xpot calculation typo in sp_get_samples_2d_nearest_clamp_POT
[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 unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */
897 unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */
898
899 for (j = 0; j < QUAD_SIZE; j++) {
900 int c;
901
902 float u = s[j] * xpot - 0.5F;
903 float v = t[j] * ypot - 0.5F;
904
905 int uflr = util_ifloor(u);
906 int vflr = util_ifloor(v);
907
908 float xw = u - (float)uflr;
909 float yw = v - (float)vflr;
910
911 int x0 = uflr & (xpot - 1);
912 int y0 = vflr & (ypot - 1);
913
914 const float *tx[4];
915
916
917 /* Can we fetch all four at once:
918 */
919 if (x0 < xmax && y0 < ymax)
920 {
921 get_texel_quad_2d(tgsi_sampler, 0, level, x0, y0, tx);
922 }
923 else
924 {
925 unsigned x1 = (x0 + 1) & (xpot - 1);
926 unsigned y1 = (y0 + 1) & (ypot - 1);
927 get_texel_quad_2d_mt(tgsi_sampler, 0, level,
928 x0, y0, x1, y1, tx);
929 }
930
931
932 /* interpolate R, G, B, A */
933 for (c = 0; c < 4; c++) {
934 rgba[c][j] = lerp_2d(xw, yw,
935 tx[0][c], tx[1][c],
936 tx[2][c], tx[3][c]);
937 }
938 }
939 }
940
941
942 static void
943 sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
944 const float s[QUAD_SIZE],
945 const float t[QUAD_SIZE],
946 const float p[QUAD_SIZE],
947 float lodbias,
948 float rgba[NUM_CHANNELS][QUAD_SIZE])
949 {
950 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
951 unsigned j;
952 unsigned level = samp->level;
953 unsigned xpot = 1 << (samp->xpot - level);
954 unsigned ypot = 1 << (samp->ypot - level);
955
956 for (j = 0; j < QUAD_SIZE; j++) {
957 int c;
958
959 float u = s[j] * xpot;
960 float v = t[j] * ypot;
961
962 int uflr = util_ifloor(u);
963 int vflr = util_ifloor(v);
964
965 int x0 = uflr & (xpot - 1);
966 int y0 = vflr & (ypot - 1);
967
968 const float *out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0);
969
970 for (c = 0; c < 4; c++) {
971 rgba[c][j] = out[c];
972 }
973 }
974 }
975
976
977 static void
978 sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
979 const float s[QUAD_SIZE],
980 const float t[QUAD_SIZE],
981 const float p[QUAD_SIZE],
982 float lodbias,
983 float rgba[NUM_CHANNELS][QUAD_SIZE])
984 {
985 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
986 unsigned j;
987 unsigned level = samp->level;
988 unsigned xpot = 1 << (samp->xpot - level);
989 unsigned ypot = 1 << (samp->ypot - level);
990
991 for (j = 0; j < QUAD_SIZE; j++) {
992 int c;
993
994 float u = s[j] * xpot;
995 float v = t[j] * ypot;
996
997 int x0, y0;
998 const float *out;
999
1000 x0 = util_ifloor(u);
1001 if (x0 < 0)
1002 x0 = 0;
1003 else if (x0 > xpot - 1)
1004 x0 = xpot - 1;
1005
1006 y0 = util_ifloor(v);
1007 if (y0 < 0)
1008 y0 = 0;
1009 else if (y0 > ypot - 1)
1010 y0 = ypot - 1;
1011
1012 out = get_texel_2d_ptr(tgsi_sampler, 0, level, x0, y0);
1013
1014 for (c = 0; c < 4; c++) {
1015 rgba[c][j] = out[c];
1016 }
1017 }
1018 }
1019
1020
1021 static void
1022 sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
1023 const float s[QUAD_SIZE],
1024 const float t[QUAD_SIZE],
1025 const float p[QUAD_SIZE],
1026 float lodbias,
1027 float rgba[NUM_CHANNELS][QUAD_SIZE])
1028 {
1029 struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1030 const struct pipe_texture *texture = samp->texture;
1031 const struct pipe_sampler_state *sampler = samp->sampler;
1032 int level0;
1033 float lambda;
1034
1035 lambda = compute_lambda(texture, sampler, s, t, p, lodbias);
1036 level0 = (int)lambda;
1037
1038 if (lambda < 0.0) {
1039 samp->level = 0;
1040 sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
1041 s, t, p, 0, 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, 0, rgba );
1047 }
1048 else {
1049 float levelBlend = lambda - level0;
1050 float rgba0[4][4];
1051 float rgba1[4][4];
1052 int c,j;
1053
1054 samp->level = level0;
1055 sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
1056 s, t, p, 0, rgba0 );
1057
1058 samp->level = level0+1;
1059 sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
1060 s, t, p, 0, rgba1 );
1061
1062 for (j = 0; j < QUAD_SIZE; j++) {
1063 for (c = 0; c < 4; c++) {
1064 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1065 }
1066 }
1067 }
1068 }
1069
1070 /**
1071 * Common code for sampling 1D/2D/cube textures.
1072 * Could probably extend for 3D...
1073 */
1074 static void
1075 sp_get_samples_2d_common(const struct tgsi_sampler *tgsi_sampler,
1076 const float s[QUAD_SIZE],
1077 const float t[QUAD_SIZE],
1078 const float p[QUAD_SIZE],
1079 boolean computeLambda,
1080 float lodbias,
1081 float rgba[NUM_CHANNELS][QUAD_SIZE],
1082 const unsigned faces[4])
1083 {
1084 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1085 const struct pipe_texture *texture = samp->texture;
1086 const struct pipe_sampler_state *sampler = samp->sampler;
1087 unsigned level0, level1, j, imgFilter;
1088 int width, height;
1089 float levelBlend;
1090
1091 choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
1092 &level0, &level1, &levelBlend, &imgFilter);
1093
1094 assert(sampler->normalized_coords);
1095
1096 width = texture->width[level0];
1097 height = texture->height[level0];
1098
1099 assert(width > 0);
1100
1101 switch (imgFilter) {
1102 case PIPE_TEX_FILTER_NEAREST:
1103 {
1104 int x[4], y[4];
1105 nearest_texcoord_4(sampler->wrap_s, s, width, x);
1106 nearest_texcoord_4(sampler->wrap_t, t, height, y);
1107
1108 for (j = 0; j < QUAD_SIZE; j++) {
1109 get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j);
1110 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1111 shadow_compare(sampler, rgba, p, j);
1112 }
1113
1114 if (level0 != level1) {
1115 /* get texels from second mipmap level and blend */
1116 float rgba2[4][4];
1117 unsigned c;
1118 x[j] /= 2;
1119 y[j] /= 2;
1120 get_texel(tgsi_sampler, faces[j], level1, x[j], y[j], 0,
1121 rgba2, j);
1122 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
1123 shadow_compare(sampler, rgba2, p, j);
1124 }
1125
1126 for (c = 0; c < NUM_CHANNELS; c++) {
1127 rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
1128 }
1129 }
1130 }
1131 }
1132 break;
1133 case PIPE_TEX_FILTER_LINEAR:
1134 case PIPE_TEX_FILTER_ANISO:
1135 {
1136 int x0[4], y0[4], x1[4], y1[4];
1137 float xw[4], yw[4]; /* weights */
1138
1139 linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw);
1140 linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
1141
1142 for (j = 0; j < QUAD_SIZE; j++) {
1143 float tx[4][4]; /* texels */
1144 int c;
1145 get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0);
1146 get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1);
1147 get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2);
1148 get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3);
1149 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1150 shadow_compare4(sampler, tx, p);
1151 }
1152
1153 /* interpolate R, G, B, A */
1154 for (c = 0; c < 4; c++) {
1155 rgba[c][j] = lerp_2d(xw[j], yw[j],
1156 tx[c][0], tx[c][1],
1157 tx[c][2], tx[c][3]);
1158 }
1159
1160 if (level0 != level1) {
1161 /* get texels from second mipmap level and blend */
1162 float rgba2[4][4];
1163
1164 /* XXX: This is incorrect -- will often end up with (x0
1165 * == x1 && y0 == y1), meaning that we fetch the same
1166 * texel four times and linearly interpolate between
1167 * identical values. The correct approach would be to
1168 * call linear_texcoord again for the second level.
1169 */
1170 x0[j] /= 2;
1171 y0[j] /= 2;
1172 x1[j] /= 2;
1173 y1[j] /= 2;
1174 get_texel(tgsi_sampler, faces[j], level1, x0[j], y0[j], 0, tx, 0);
1175 get_texel(tgsi_sampler, faces[j], level1, x1[j], y0[j], 0, tx, 1);
1176 get_texel(tgsi_sampler, faces[j], level1, x0[j], y1[j], 0, tx, 2);
1177 get_texel(tgsi_sampler, faces[j], level1, x1[j], y1[j], 0, tx, 3);
1178 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
1179 shadow_compare4(sampler, tx, p);
1180 }
1181
1182 /* interpolate R, G, B, A */
1183 for (c = 0; c < 4; c++) {
1184 rgba2[c][j] = lerp_2d(xw[j], yw[j],
1185 tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
1186 }
1187
1188 for (c = 0; c < NUM_CHANNELS; c++) {
1189 rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
1190 }
1191 }
1192 }
1193 }
1194 break;
1195 default:
1196 assert(0);
1197 }
1198 }
1199
1200
1201 static INLINE void
1202 sp_get_samples_1d(const struct tgsi_sampler *sampler,
1203 const float s[QUAD_SIZE],
1204 const float t[QUAD_SIZE],
1205 const float p[QUAD_SIZE],
1206 boolean computeLambda,
1207 float lodbias,
1208 float rgba[NUM_CHANNELS][QUAD_SIZE])
1209 {
1210 static const unsigned faces[4] = {0, 0, 0, 0};
1211 static const float tzero[4] = {0, 0, 0, 0};
1212 sp_get_samples_2d_common(sampler, s, tzero, NULL,
1213 computeLambda, lodbias, rgba, faces);
1214 }
1215
1216
1217 static INLINE void
1218 sp_get_samples_2d(const struct tgsi_sampler *sampler,
1219 const float s[QUAD_SIZE],
1220 const float t[QUAD_SIZE],
1221 const float p[QUAD_SIZE],
1222 boolean computeLambda,
1223 float lodbias,
1224 float rgba[NUM_CHANNELS][QUAD_SIZE])
1225 {
1226 static const unsigned faces[4] = {0, 0, 0, 0};
1227 sp_get_samples_2d_common(sampler, s, t, p,
1228 computeLambda, lodbias, rgba, faces);
1229 }
1230
1231
1232 static INLINE void
1233 sp_get_samples_3d(const struct tgsi_sampler *tgsi_sampler,
1234 const float s[QUAD_SIZE],
1235 const float t[QUAD_SIZE],
1236 const float p[QUAD_SIZE],
1237 boolean computeLambda,
1238 float lodbias,
1239 float rgba[NUM_CHANNELS][QUAD_SIZE])
1240 {
1241 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1242 const struct pipe_texture *texture = samp->texture;
1243 const struct pipe_sampler_state *sampler = samp->sampler;
1244 /* get/map pipe_surfaces corresponding to 3D tex slices */
1245 unsigned level0, level1, j, imgFilter;
1246 int width, height, depth;
1247 float levelBlend;
1248 const uint face = 0;
1249
1250 choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
1251 &level0, &level1, &levelBlend, &imgFilter);
1252
1253 assert(sampler->normalized_coords);
1254
1255 width = texture->width[level0];
1256 height = texture->height[level0];
1257 depth = texture->depth[level0];
1258
1259 assert(width > 0);
1260 assert(height > 0);
1261 assert(depth > 0);
1262
1263 switch (imgFilter) {
1264 case PIPE_TEX_FILTER_NEAREST:
1265 {
1266 int x[4], y[4], z[4];
1267 nearest_texcoord_4(sampler->wrap_s, s, width, x);
1268 nearest_texcoord_4(sampler->wrap_t, t, height, y);
1269 nearest_texcoord_4(sampler->wrap_r, p, depth, z);
1270 for (j = 0; j < QUAD_SIZE; j++) {
1271 get_texel(tgsi_sampler, face, level0, x[j], y[j], z[j], rgba, j);
1272 if (level0 != level1) {
1273 /* get texels from second mipmap level and blend */
1274 float rgba2[4][4];
1275 unsigned c;
1276 x[j] /= 2;
1277 y[j] /= 2;
1278 z[j] /= 2;
1279 get_texel(tgsi_sampler, face, level1, x[j], y[j], z[j], rgba2, j);
1280 for (c = 0; c < NUM_CHANNELS; c++) {
1281 rgba[c][j] = lerp(levelBlend, rgba2[c][j], rgba[c][j]);
1282 }
1283 }
1284 }
1285 }
1286 break;
1287 case PIPE_TEX_FILTER_LINEAR:
1288 case PIPE_TEX_FILTER_ANISO:
1289 {
1290 int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
1291 float xw[4], yw[4], zw[4]; /* interpolation weights */
1292 linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw);
1293 linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
1294 linear_texcoord_4(sampler->wrap_r, p, depth, z0, z1, zw);
1295
1296 for (j = 0; j < QUAD_SIZE; j++) {
1297 int c;
1298 float tx0[4][4], tx1[4][4];
1299 get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z0[j], tx0, 0);
1300 get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z0[j], tx0, 1);
1301 get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z0[j], tx0, 2);
1302 get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z0[j], tx0, 3);
1303 get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z1[j], tx1, 0);
1304 get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z1[j], tx1, 1);
1305 get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z1[j], tx1, 2);
1306 get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z1[j], tx1, 3);
1307
1308 /* interpolate R, G, B, A */
1309 for (c = 0; c < 4; c++) {
1310 rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1311 tx0[c][0], tx0[c][1],
1312 tx0[c][2], tx0[c][3],
1313 tx1[c][0], tx1[c][1],
1314 tx1[c][2], tx1[c][3]);
1315 }
1316
1317 if (level0 != level1) {
1318 /* get texels from second mipmap level and blend */
1319 float rgba2[4][4];
1320 x0[j] /= 2;
1321 y0[j] /= 2;
1322 z0[j] /= 2;
1323 x1[j] /= 2;
1324 y1[j] /= 2;
1325 z1[j] /= 2;
1326 get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z0[j], tx0, 0);
1327 get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z0[j], tx0, 1);
1328 get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z0[j], tx0, 2);
1329 get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z0[j], tx0, 3);
1330 get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z1[j], tx1, 0);
1331 get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z1[j], tx1, 1);
1332 get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z1[j], tx1, 2);
1333 get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z1[j], tx1, 3);
1334
1335 /* interpolate R, G, B, A */
1336 for (c = 0; c < 4; c++) {
1337 rgba2[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1338 tx0[c][0], tx0[c][1],
1339 tx0[c][2], tx0[c][3],
1340 tx1[c][0], tx1[c][1],
1341 tx1[c][2], tx1[c][3]);
1342 }
1343
1344 /* blend mipmap levels */
1345 for (c = 0; c < NUM_CHANNELS; c++) {
1346 rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
1347 }
1348 }
1349 }
1350 }
1351 break;
1352 default:
1353 assert(0);
1354 }
1355 }
1356
1357
1358 static void
1359 sp_get_samples_cube(const struct tgsi_sampler *sampler,
1360 const float s[QUAD_SIZE],
1361 const float t[QUAD_SIZE],
1362 const float p[QUAD_SIZE],
1363 boolean computeLambda,
1364 float lodbias,
1365 float rgba[NUM_CHANNELS][QUAD_SIZE])
1366 {
1367 unsigned faces[QUAD_SIZE], j;
1368 float ssss[4], tttt[4];
1369 for (j = 0; j < QUAD_SIZE; j++) {
1370 faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j);
1371 }
1372 sp_get_samples_2d_common(sampler, ssss, tttt, NULL,
1373 computeLambda, lodbias, rgba, faces);
1374 }
1375
1376
1377 static void
1378 sp_get_samples_rect(const struct tgsi_sampler *tgsi_sampler,
1379 const float s[QUAD_SIZE],
1380 const float t[QUAD_SIZE],
1381 const float p[QUAD_SIZE],
1382 boolean computeLambda,
1383 float lodbias,
1384 float rgba[NUM_CHANNELS][QUAD_SIZE])
1385 {
1386 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1387 const struct pipe_texture *texture = samp->texture;
1388 const struct pipe_sampler_state *sampler = samp->sampler;
1389 const uint face = 0;
1390 unsigned level0, level1, j, imgFilter;
1391 int width, height;
1392 float levelBlend;
1393
1394 choose_mipmap_levels(texture, sampler, s, t, p, computeLambda, lodbias,
1395 &level0, &level1, &levelBlend, &imgFilter);
1396
1397 /* texture RECTS cannot be mipmapped */
1398 assert(level0 == level1);
1399
1400 width = texture->width[level0];
1401 height = texture->height[level0];
1402
1403 assert(width > 0);
1404
1405 switch (imgFilter) {
1406 case PIPE_TEX_FILTER_NEAREST:
1407 {
1408 int x[4], y[4];
1409 nearest_texcoord_unnorm_4(sampler->wrap_s, s, width, x);
1410 nearest_texcoord_unnorm_4(sampler->wrap_t, t, height, y);
1411 for (j = 0; j < QUAD_SIZE; j++) {
1412 get_texel(tgsi_sampler, face, level0, x[j], y[j], 0, rgba, j);
1413 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1414 shadow_compare(sampler, rgba, p, j);
1415 }
1416 }
1417 }
1418 break;
1419 case PIPE_TEX_FILTER_LINEAR:
1420 case PIPE_TEX_FILTER_ANISO:
1421 {
1422 int x0[4], y0[4], x1[4], y1[4];
1423 float xw[4], yw[4]; /* weights */
1424 linear_texcoord_unnorm_4(sampler->wrap_s, s, width, x0, x1, xw);
1425 linear_texcoord_unnorm_4(sampler->wrap_t, t, height, y0, y1, yw);
1426 for (j = 0; j < QUAD_SIZE; j++) {
1427 float tx[4][4]; /* texels */
1428 int c;
1429 get_texel(tgsi_sampler, face, level0, x0[j], y0[j], 0, tx, 0);
1430 get_texel(tgsi_sampler, face, level0, x1[j], y0[j], 0, tx, 1);
1431 get_texel(tgsi_sampler, face, level0, x0[j], y1[j], 0, tx, 2);
1432 get_texel(tgsi_sampler, face, level0, x1[j], y1[j], 0, tx, 3);
1433 if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
1434 shadow_compare4(sampler, tx, p);
1435 }
1436 for (c = 0; c < 4; c++) {
1437 rgba[c][j] = lerp_2d(xw[j], yw[j],
1438 tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
1439 }
1440 }
1441 }
1442 break;
1443 default:
1444 assert(0);
1445 }
1446 }
1447
1448
1449 /**
1450 * Common code for vertex/fragment program texture sampling.
1451 */
1452 static INLINE void
1453 sp_get_samples(struct tgsi_sampler *tgsi_sampler,
1454 const float s[QUAD_SIZE],
1455 const float t[QUAD_SIZE],
1456 const float p[QUAD_SIZE],
1457 boolean computeLambda,
1458 float lodbias,
1459 float rgba[NUM_CHANNELS][QUAD_SIZE])
1460 {
1461 const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1462 const struct pipe_texture *texture = samp->texture;
1463 const struct pipe_sampler_state *sampler = samp->sampler;
1464
1465 if (!texture)
1466 return;
1467
1468 switch (texture->target) {
1469 case PIPE_TEXTURE_1D:
1470 assert(sampler->normalized_coords);
1471 sp_get_samples_1d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1472 break;
1473 case PIPE_TEXTURE_2D:
1474 if (sampler->normalized_coords)
1475 sp_get_samples_2d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1476 else
1477 sp_get_samples_rect(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1478 break;
1479 case PIPE_TEXTURE_3D:
1480 assert(sampler->normalized_coords);
1481 sp_get_samples_3d(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1482 break;
1483 case PIPE_TEXTURE_CUBE:
1484 assert(sampler->normalized_coords);
1485 sp_get_samples_cube(tgsi_sampler, s, t, p, computeLambda, lodbias, rgba);
1486 break;
1487 default:
1488 assert(0);
1489 }
1490
1491 #if 0 /* DEBUG */
1492 {
1493 int i;
1494 printf("Sampled at %f, %f, %f:\n", s[0], t[0], p[0]);
1495 for (i = 0; i < 4; i++) {
1496 printf("Frag %d: %f %f %f %f\n", i,
1497 rgba[0][i],
1498 rgba[1][i],
1499 rgba[2][i],
1500 rgba[3][i]);
1501 }
1502 }
1503 #endif
1504 }
1505
1506 static void
1507 sp_get_samples_fallback(struct tgsi_sampler *tgsi_sampler,
1508 const float s[QUAD_SIZE],
1509 const float t[QUAD_SIZE],
1510 const float p[QUAD_SIZE],
1511 float lodbias,
1512 float rgba[NUM_CHANNELS][QUAD_SIZE])
1513 {
1514 sp_get_samples(tgsi_sampler, s, t, p, TRUE, lodbias, rgba);
1515 }
1516
1517 /**
1518 * Called via tgsi_sampler::get_samples() when running a fragment shader.
1519 * Get four filtered RGBA values from the sampler's texture.
1520 */
1521 void
1522 sp_get_samples_fragment(struct tgsi_sampler *tgsi_sampler,
1523 const float s[QUAD_SIZE],
1524 const float t[QUAD_SIZE],
1525 const float p[QUAD_SIZE],
1526 float lodbias,
1527 float rgba[NUM_CHANNELS][QUAD_SIZE])
1528 {
1529 struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
1530 const struct pipe_texture *texture = samp->texture;
1531 const struct pipe_sampler_state *sampler = samp->sampler;
1532
1533 tgsi_sampler->get_samples = sp_get_samples_fallback;
1534
1535 /* Try to hook in a faster sampler. Ultimately we'll have to
1536 * code-generate these. Luckily most of this looks like it is
1537 * orthogonal state within the sampler.
1538 */
1539 if (texture->target == PIPE_TEXTURE_2D &&
1540 sampler->min_img_filter == sampler->mag_img_filter &&
1541 sampler->wrap_s == sampler->wrap_t &&
1542 sampler->compare_mode == FALSE &&
1543 sampler->normalized_coords)
1544 {
1545 samp->xpot = util_unsigned_logbase2( samp->texture->width[0] );
1546 samp->ypot = util_unsigned_logbase2( samp->texture->height[0] );
1547
1548 if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
1549 samp->level = CLAMP((int) sampler->min_lod,
1550 0, (int) texture->last_level);
1551
1552 if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) {
1553 switch (sampler->min_img_filter) {
1554 case PIPE_TEX_FILTER_NEAREST:
1555 tgsi_sampler->get_samples = sp_get_samples_2d_nearest_repeat_POT;
1556 break;
1557 case PIPE_TEX_FILTER_LINEAR:
1558 tgsi_sampler->get_samples = sp_get_samples_2d_linear_repeat_POT;
1559 break;
1560 default:
1561 break;
1562 }
1563 }
1564 else if (sampler->wrap_s == PIPE_TEX_WRAP_CLAMP) {
1565 switch (sampler->min_img_filter) {
1566 case PIPE_TEX_FILTER_NEAREST:
1567 tgsi_sampler->get_samples = sp_get_samples_2d_nearest_clamp_POT;
1568 break;
1569 default:
1570 break;
1571 }
1572 }
1573 }
1574 else if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
1575 if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) {
1576 switch (sampler->min_img_filter) {
1577 case PIPE_TEX_FILTER_LINEAR:
1578 tgsi_sampler->get_samples = sp_get_samples_2d_linear_mip_linear_repeat_POT;
1579 break;
1580 default:
1581 break;
1582 }
1583 }
1584 }
1585 }
1586 else if (0) {
1587 _debug_printf("target %d/%d min_mip %d/%d min_img %d/%d wrap %d/%d compare %d/%d norm %d/%d\n",
1588 texture->target, PIPE_TEXTURE_2D,
1589 sampler->min_mip_filter, PIPE_TEX_MIPFILTER_NONE,
1590 sampler->min_img_filter, sampler->mag_img_filter,
1591 sampler->wrap_s, sampler->wrap_t,
1592 sampler->compare_mode, FALSE,
1593 sampler->normalized_coords, TRUE);
1594 }
1595
1596 tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba );
1597 }
1598
1599
1600 /**
1601 * Called via tgsi_sampler::get_samples() when running a vertex shader.
1602 * Get four filtered RGBA values from the sampler's texture.
1603 */
1604 void
1605 sp_get_samples_vertex(struct tgsi_sampler *tgsi_sampler,
1606 const float s[QUAD_SIZE],
1607 const float t[QUAD_SIZE],
1608 const float p[QUAD_SIZE],
1609 float lodbias,
1610 float rgba[NUM_CHANNELS][QUAD_SIZE])
1611 {
1612 sp_get_samples(tgsi_sampler, s, t, p, FALSE, lodbias, rgba);
1613 }