projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
gallium: Pass per-element (not per-quad) LOD bias values down to texture sampler.
[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 2008-2010 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 * Keith Whitwell
35 */
36
37 #include "pipe/p_context.h"
38 #include "pipe/p_defines.h"
39 #include "pipe/p_shader_tokens.h"
40 #include "util/u_math.h"
41 #include "util/u_memory.h"
42 #include "sp_quad.h" /* only for #define QUAD_* tokens */
43 #include "sp_tex_sample.h"
44 #include "sp_tex_tile_cache.h"
45
46
47
48 /*
49 * Return fractional part of 'f'. Used for computing interpolation weights.
50 * Need to be careful with negative values.
51 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
52 * of improperly weighted linear-filtered textures.
53 * The tests/texwrap.c demo is a good test.
54 */
55 static INLINE float
56 frac(float f)
57 {
58 return f - util_ifloor(f);
59 }
60
61
62
63 /**
64 * Linear interpolation macro
65 */
66 static INLINE float
67 lerp(float a, float v0, float v1)
68 {
69 return v0 + a * (v1 - v0);
70 }
71
72
73 /**
74 * Do 2D/biliner interpolation of float values.
75 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
76 * a and b are the horizontal and vertical interpolants.
77 * It's important that this function is inlined when compiled with
78 * optimization! If we find that's not true on some systems, convert
79 * to a macro.
80 */
81 static INLINE float
82 lerp_2d(float a, float b,
83 float v00, float v10, float v01, float v11)
84 {
85 const float temp0 = lerp(a, v00, v10);
86 const float temp1 = lerp(a, v01, v11);
87 return lerp(b, temp0, temp1);
88 }
89
90
91 /**
92 * As above, but 3D interpolation of 8 values.
93 */
94 static INLINE float
95 lerp_3d(float a, float b, float c,
96 float v000, float v100, float v010, float v110,
97 float v001, float v101, float v011, float v111)
98 {
99 const float temp0 = lerp_2d(a, b, v000, v100, v010, v110);
100 const float temp1 = lerp_2d(a, b, v001, v101, v011, v111);
101 return lerp(c, temp0, temp1);
102 }
103
104
105
106 /**
107 * Compute coord % size for repeat wrap modes.
108 * Note that if coord is a signed integer, coord % size doesn't give
109 * the right value for coord < 0 (in terms of texture repeat). Just
110 * casting to unsigned fixes that.
111 */
112 static INLINE int
113 repeat(int coord, unsigned size)
114 {
115 return (int) ((unsigned) coord % size);
116 }
117
118
119 /**
120 * Apply texture coord wrapping mode and return integer texture indexes
121 * for a vector of four texcoords (S or T or P).
122 * \param wrapMode PIPE_TEX_WRAP_x
123 * \param s the incoming texcoords
124 * \param size the texture image size
125 * \param icoord returns the integer texcoords
126 * \return integer texture index
127 */
128 static void
129 wrap_nearest_repeat(const float s[4], unsigned size, int icoord[4])
130 {
131 uint ch;
132 /* s limited to [0,1) */
133 /* i limited to [0,size-1] */
134 for (ch = 0; ch < 4; ch++) {
135 int i = util_ifloor(s[ch] * size);
136 icoord[ch] = repeat(i, size);
137 }
138 }
139
140
141 static void
142 wrap_nearest_clamp(const float s[4], unsigned size, int icoord[4])
143 {
144 uint ch;
145 /* s limited to [0,1] */
146 /* i limited to [0,size-1] */
147 for (ch = 0; ch < 4; ch++) {
148 if (s[ch] <= 0.0F)
149 icoord[ch] = 0;
150 else if (s[ch] >= 1.0F)
151 icoord[ch] = size - 1;
152 else
153 icoord[ch] = util_ifloor(s[ch] * size);
154 }
155 }
156
157
158 static void
159 wrap_nearest_clamp_to_edge(const float s[4], unsigned size, int icoord[4])
160 {
161 uint ch;
162 /* s limited to [min,max] */
163 /* i limited to [0, size-1] */
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] = 0;
169 else if (s[ch] > max)
170 icoord[ch] = size - 1;
171 else
172 icoord[ch] = util_ifloor(s[ch] * size);
173 }
174 }
175
176
177 static void
178 wrap_nearest_clamp_to_border(const float s[4], unsigned size, int icoord[4])
179 {
180 uint ch;
181 /* s limited to [min,max] */
182 /* i limited to [-1, size] */
183 const float min = -1.0F / (2.0F * size);
184 const float max = 1.0F - min;
185 for (ch = 0; ch < 4; ch++) {
186 if (s[ch] <= min)
187 icoord[ch] = -1;
188 else if (s[ch] >= max)
189 icoord[ch] = size;
190 else
191 icoord[ch] = util_ifloor(s[ch] * size);
192 }
193 }
194
195
196 static void
197 wrap_nearest_mirror_repeat(const float s[4], unsigned size, int icoord[4])
198 {
199 uint ch;
200 const float min = 1.0F / (2.0F * size);
201 const float max = 1.0F - min;
202 for (ch = 0; ch < 4; ch++) {
203 const int flr = util_ifloor(s[ch]);
204 float u;
205 if (flr & 1)
206 u = 1.0F - (s[ch] - (float) flr);
207 else
208 u = s[ch] - (float) flr;
209 if (u < min)
210 icoord[ch] = 0;
211 else if (u > max)
212 icoord[ch] = size - 1;
213 else
214 icoord[ch] = util_ifloor(u * size);
215 }
216 }
217
218
219 static void
220 wrap_nearest_mirror_clamp(const float s[4], unsigned size, int icoord[4])
221 {
222 uint ch;
223 for (ch = 0; ch < 4; ch++) {
224 /* s limited to [0,1] */
225 /* i limited to [0,size-1] */
226 const float u = fabsf(s[ch]);
227 if (u <= 0.0F)
228 icoord[ch] = 0;
229 else if (u >= 1.0F)
230 icoord[ch] = size - 1;
231 else
232 icoord[ch] = util_ifloor(u * size);
233 }
234 }
235
236
237 static void
238 wrap_nearest_mirror_clamp_to_edge(const float s[4], unsigned size,
239 int icoord[4])
240 {
241 uint ch;
242 /* s limited to [min,max] */
243 /* i limited to [0, size-1] */
244 const float min = 1.0F / (2.0F * size);
245 const float max = 1.0F - min;
246 for (ch = 0; ch < 4; ch++) {
247 const float u = fabsf(s[ch]);
248 if (u < min)
249 icoord[ch] = 0;
250 else if (u > max)
251 icoord[ch] = size - 1;
252 else
253 icoord[ch] = util_ifloor(u * size);
254 }
255 }
256
257
258 static void
259 wrap_nearest_mirror_clamp_to_border(const float s[4], unsigned size,
260 int icoord[4])
261 {
262 uint ch;
263 /* s limited to [min,max] */
264 /* i limited to [0, size-1] */
265 const float min = -1.0F / (2.0F * size);
266 const float max = 1.0F - min;
267 for (ch = 0; ch < 4; ch++) {
268 const float u = fabsf(s[ch]);
269 if (u < min)
270 icoord[ch] = -1;
271 else if (u > max)
272 icoord[ch] = size;
273 else
274 icoord[ch] = util_ifloor(u * size);
275 }
276 }
277
278
279 /**
280 * Used to compute texel locations for linear sampling for four texcoords.
281 * \param wrapMode PIPE_TEX_WRAP_x
282 * \param s the texcoords
283 * \param size the texture image size
284 * \param icoord0 returns first texture indexes
285 * \param icoord1 returns second texture indexes (usually icoord0 + 1)
286 * \param w returns blend factor/weight between texture indexes
287 * \param icoord returns the computed integer texture coords
288 */
289 static void
290 wrap_linear_repeat(const float s[4], unsigned size,
291 int icoord0[4], int icoord1[4], float w[4])
292 {
293 uint ch;
294 for (ch = 0; ch < 4; ch++) {
295 float u = s[ch] * size - 0.5F;
296 icoord0[ch] = repeat(util_ifloor(u), size);
297 icoord1[ch] = repeat(icoord0[ch] + 1, size);
298 w[ch] = frac(u);
299 }
300 }
301
302
303 static void
304 wrap_linear_clamp(const float s[4], unsigned size,
305 int icoord0[4], int icoord1[4], float w[4])
306 {
307 uint ch;
308 for (ch = 0; ch < 4; ch++) {
309 float u = CLAMP(s[ch], 0.0F, 1.0F);
310 u = u * size - 0.5f;
311 icoord0[ch] = util_ifloor(u);
312 icoord1[ch] = icoord0[ch] + 1;
313 w[ch] = frac(u);
314 }
315 }
316
317
318 static void
319 wrap_linear_clamp_to_edge(const float s[4], unsigned size,
320 int icoord0[4], int icoord1[4], float w[4])
321 {
322 uint ch;
323 for (ch = 0; ch < 4; ch++) {
324 float u = CLAMP(s[ch], 0.0F, 1.0F);
325 u = u * size - 0.5f;
326 icoord0[ch] = util_ifloor(u);
327 icoord1[ch] = icoord0[ch] + 1;
328 if (icoord0[ch] < 0)
329 icoord0[ch] = 0;
330 if (icoord1[ch] >= (int) size)
331 icoord1[ch] = size - 1;
332 w[ch] = frac(u);
333 }
334 }
335
336
337 static void
338 wrap_linear_clamp_to_border(const float s[4], unsigned size,
339 int icoord0[4], int icoord1[4], float w[4])
340 {
341 const float min = -1.0F / (2.0F * size);
342 const float max = 1.0F - min;
343 uint ch;
344 for (ch = 0; ch < 4; ch++) {
345 float u = CLAMP(s[ch], min, max);
346 u = u * size - 0.5f;
347 icoord0[ch] = util_ifloor(u);
348 icoord1[ch] = icoord0[ch] + 1;
349 w[ch] = frac(u);
350 }
351 }
352
353
354 static void
355 wrap_linear_mirror_repeat(const float s[4], unsigned size,
356 int icoord0[4], int icoord1[4], float w[4])
357 {
358 uint ch;
359 for (ch = 0; ch < 4; ch++) {
360 const int flr = util_ifloor(s[ch]);
361 float u;
362 if (flr & 1)
363 u = 1.0F - (s[ch] - (float) flr);
364 else
365 u = s[ch] - (float) flr;
366 u = u * size - 0.5F;
367 icoord0[ch] = util_ifloor(u);
368 icoord1[ch] = icoord0[ch] + 1;
369 if (icoord0[ch] < 0)
370 icoord0[ch] = 0;
371 if (icoord1[ch] >= (int) size)
372 icoord1[ch] = size - 1;
373 w[ch] = frac(u);
374 }
375 }
376
377
378 static void
379 wrap_linear_mirror_clamp(const float s[4], unsigned size,
380 int icoord0[4], int icoord1[4], float w[4])
381 {
382 uint ch;
383 for (ch = 0; ch < 4; ch++) {
384 float u = fabsf(s[ch]);
385 if (u >= 1.0F)
386 u = (float) size;
387 else
388 u *= size;
389 u -= 0.5F;
390 icoord0[ch] = util_ifloor(u);
391 icoord1[ch] = icoord0[ch] + 1;
392 w[ch] = frac(u);
393 }
394 }
395
396
397 static void
398 wrap_linear_mirror_clamp_to_edge(const float s[4], unsigned size,
399 int icoord0[4], int icoord1[4], float w[4])
400 {
401 uint ch;
402 for (ch = 0; ch < 4; ch++) {
403 float u = fabsf(s[ch]);
404 if (u >= 1.0F)
405 u = (float) size;
406 else
407 u *= size;
408 u -= 0.5F;
409 icoord0[ch] = util_ifloor(u);
410 icoord1[ch] = icoord0[ch] + 1;
411 if (icoord0[ch] < 0)
412 icoord0[ch] = 0;
413 if (icoord1[ch] >= (int) size)
414 icoord1[ch] = size - 1;
415 w[ch] = frac(u);
416 }
417 }
418
419
420 static void
421 wrap_linear_mirror_clamp_to_border(const float s[4], unsigned size,
422 int icoord0[4], int icoord1[4], float w[4])
423 {
424 const float min = -1.0F / (2.0F * size);
425 const float max = 1.0F - min;
426 uint ch;
427 for (ch = 0; ch < 4; ch++) {
428 float u = fabsf(s[ch]);
429 if (u <= min)
430 u = min * size;
431 else if (u >= max)
432 u = max * size;
433 else
434 u *= size;
435 u -= 0.5F;
436 icoord0[ch] = util_ifloor(u);
437 icoord1[ch] = icoord0[ch] + 1;
438 w[ch] = frac(u);
439 }
440 }
441
442
443 /**
444 * For RECT textures / unnormalized texcoords
445 * Only a subset of wrap modes supported.
446 */
447 static void
448 wrap_nearest_unorm_clamp(const float s[4], unsigned size, int icoord[4])
449 {
450 uint ch;
451 for (ch = 0; ch < 4; ch++) {
452 int i = util_ifloor(s[ch]);
453 icoord[ch]= CLAMP(i, 0, (int) size-1);
454 }
455 }
456
457
458 /**
459 * Handles clamp_to_edge and clamp_to_border:
460 */
461 static void
462 wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size,
463 int icoord[4])
464 {
465 uint ch;
466 for (ch = 0; ch < 4; ch++) {
467 icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
468 }
469 }
470
471
472 /**
473 * For RECT textures / unnormalized texcoords.
474 * Only a subset of wrap modes supported.
475 */
476 static void
477 wrap_linear_unorm_clamp(const float s[4], unsigned size,
478 int icoord0[4], int icoord1[4], float w[4])
479 {
480 uint ch;
481 for (ch = 0; ch < 4; ch++) {
482 /* Not exactly what the spec says, but it matches NVIDIA output */
483 float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
484 icoord0[ch] = util_ifloor(u);
485 icoord1[ch] = icoord0[ch] + 1;
486 w[ch] = frac(u);
487 }
488 }
489
490
491 static void
492 wrap_linear_unorm_clamp_to_border(const float s[4], unsigned size,
493 int icoord0[4], int icoord1[4], float w[4])
494 {
495 uint ch;
496 for (ch = 0; ch < 4; ch++) {
497 float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F);
498 u -= 0.5F;
499 icoord0[ch] = util_ifloor(u);
500 icoord1[ch] = icoord0[ch] + 1;
501 if (icoord1[ch] > (int) size - 1)
502 icoord1[ch] = size - 1;
503 w[ch] = frac(u);
504 }
505 }
506
507
508
509 /**
510 * Examine the quad's texture coordinates to compute the partial
511 * derivatives w.r.t X and Y, then compute lambda (level of detail).
512 */
513 static float
514 compute_lambda_1d(const struct sp_sampler_varient *samp,
515 const float s[QUAD_SIZE],
516 const float t[QUAD_SIZE],
517 const float p[QUAD_SIZE])
518 {
519 const struct pipe_texture *texture = samp->texture;
520 const struct pipe_sampler_state *sampler = samp->sampler;
521 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
522 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
523 float rho = MAX2(dsdx, dsdy) * texture->width0;
524
525 return util_fast_log2(rho);
526 }
527
528
529 static float
530 compute_lambda_2d(const struct sp_sampler_varient *samp,
531 const float s[QUAD_SIZE],
532 const float t[QUAD_SIZE],
533 const float p[QUAD_SIZE])
534 {
535 const struct pipe_texture *texture = samp->texture;
536 const struct pipe_sampler_state *sampler = samp->sampler;
537 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
538 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
539 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
540 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
541 float maxx = MAX2(dsdx, dsdy) * texture->width0;
542 float maxy = MAX2(dtdx, dtdy) * texture->height0;
543 float rho = MAX2(maxx, maxy);
544
545 return util_fast_log2(rho);
546 }
547
548
549 static float
550 compute_lambda_3d(const struct sp_sampler_varient *samp,
551 const float s[QUAD_SIZE],
552 const float t[QUAD_SIZE],
553 const float p[QUAD_SIZE])
554 {
555 const struct pipe_texture *texture = samp->texture;
556 const struct pipe_sampler_state *sampler = samp->sampler;
557 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
558 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
559 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
560 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
561 float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
562 float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
563 float maxx = MAX2(dsdx, dsdy) * texture->width0;
564 float maxy = MAX2(dtdx, dtdy) * texture->height0;
565 float maxz = MAX2(dpdx, dpdy) * texture->depth0;
566 float rho;
567
568 rho = MAX2(maxx, maxy);
569 rho = MAX2(rho, maxz);
570
571 return util_fast_log2(rho);
572 }
573
574
575 /**
576 * Compute lambda for a vertex texture sampler.
577 * Since there aren't derivatives to use, just return 0.
578 */
579 static float
580 compute_lambda_vert(const struct sp_sampler_varient *samp,
581 const float s[QUAD_SIZE],
582 const float t[QUAD_SIZE],
583 const float p[QUAD_SIZE])
584 {
585 return 0.0f;
586 }
587
588
589
590 /**
591 * Get a texel from a texture, using the texture tile cache.
592 *
593 * \param addr the template tex address containing cube, z, face info.
594 * \param x the x coord of texel within 2D image
595 * \param y the y coord of texel within 2D image
596 * \param rgba the quad to put the texel/color into
597 *
598 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
599 * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1...
600 */
601
602
603
604
605 static INLINE const float *
606 get_texel_2d_no_border(const struct sp_sampler_varient *samp,
607 union tex_tile_address addr, int x, int y)
608 {
609 const struct softpipe_tex_cached_tile *tile;
610
611 addr.bits.x = x / TILE_SIZE;
612 addr.bits.y = y / TILE_SIZE;
613 y %= TILE_SIZE;
614 x %= TILE_SIZE;
615
616 tile = sp_get_cached_tile_tex(samp->cache, addr);
617
618 return &tile->data.color[y][x][0];
619 }
620
621
622 static INLINE const float *
623 get_texel_2d(const struct sp_sampler_varient *samp,
624 union tex_tile_address addr, int x, int y)
625 {
626 const struct pipe_texture *texture = samp->texture;
627 unsigned level = addr.bits.level;
628
629 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
630 y < 0 || y >= (int) u_minify(texture->height0, level)) {
631 return samp->sampler->border_color;
632 }
633 else {
634 return get_texel_2d_no_border( samp, addr, x, y );
635 }
636 }
637
638
639 /* Gather a quad of adjacent texels within a tile:
640 */
641 static INLINE void
642 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_varient *samp,
643 union tex_tile_address addr,
644 unsigned x, unsigned y,
645 const float *out[4])
646 {
647 const struct softpipe_tex_cached_tile *tile;
648
649 addr.bits.x = x / TILE_SIZE;
650 addr.bits.y = y / TILE_SIZE;
651 y %= TILE_SIZE;
652 x %= TILE_SIZE;
653
654 tile = sp_get_cached_tile_tex(samp->cache, addr);
655
656 out[0] = &tile->data.color[y ][x ][0];
657 out[1] = &tile->data.color[y ][x+1][0];
658 out[2] = &tile->data.color[y+1][x ][0];
659 out[3] = &tile->data.color[y+1][x+1][0];
660 }
661
662
663 /* Gather a quad of potentially non-adjacent texels:
664 */
665 static INLINE void
666 get_texel_quad_2d_no_border(const struct sp_sampler_varient *samp,
667 union tex_tile_address addr,
668 int x0, int y0,
669 int x1, int y1,
670 const float *out[4])
671 {
672 out[0] = get_texel_2d_no_border( samp, addr, x0, y0 );
673 out[1] = get_texel_2d_no_border( samp, addr, x1, y0 );
674 out[2] = get_texel_2d_no_border( samp, addr, x0, y1 );
675 out[3] = get_texel_2d_no_border( samp, addr, x1, y1 );
676 }
677
678 /* Can involve a lot of unnecessary checks for border color:
679 */
680 static INLINE void
681 get_texel_quad_2d(const struct sp_sampler_varient *samp,
682 union tex_tile_address addr,
683 int x0, int y0,
684 int x1, int y1,
685 const float *out[4])
686 {
687 out[0] = get_texel_2d( samp, addr, x0, y0 );
688 out[1] = get_texel_2d( samp, addr, x1, y0 );
689 out[3] = get_texel_2d( samp, addr, x1, y1 );
690 out[2] = get_texel_2d( samp, addr, x0, y1 );
691 }
692
693
694
695 /* 3d varients:
696 */
697 static INLINE const float *
698 get_texel_3d_no_border(const struct sp_sampler_varient *samp,
699 union tex_tile_address addr, int x, int y, int z)
700 {
701 const struct softpipe_tex_cached_tile *tile;
702
703 addr.bits.x = x / TILE_SIZE;
704 addr.bits.y = y / TILE_SIZE;
705 addr.bits.z = z;
706 y %= TILE_SIZE;
707 x %= TILE_SIZE;
708
709 tile = sp_get_cached_tile_tex(samp->cache, addr);
710
711 return &tile->data.color[y][x][0];
712 }
713
714
715 static INLINE const float *
716 get_texel_3d(const struct sp_sampler_varient *samp,
717 union tex_tile_address addr, int x, int y, int z)
718 {
719 const struct pipe_texture *texture = samp->texture;
720 unsigned level = addr.bits.level;
721
722 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
723 y < 0 || y >= (int) u_minify(texture->height0, level) ||
724 z < 0 || z >= (int) u_minify(texture->depth0, level)) {
725 return samp->sampler->border_color;
726 }
727 else {
728 return get_texel_3d_no_border( samp, addr, x, y, z );
729 }
730 }
731
732
733 /**
734 * Given the logbase2 of a mipmap's base level size and a mipmap level,
735 * return the size (in texels) of that mipmap level.
736 * For example, if level[0].width = 256 then base_pot will be 8.
737 * If level = 2, then we'll return 64 (the width at level=2).
738 * Return 1 if level > base_pot.
739 */
740 static INLINE unsigned
741 pot_level_size(unsigned base_pot, unsigned level)
742 {
743 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
744 }
745
746
747 /* Some image-filter fastpaths:
748 */
749 static INLINE void
750 img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
751 const float s[QUAD_SIZE],
752 const float t[QUAD_SIZE],
753 const float p[QUAD_SIZE],
754 const float lodbias[QUAD_SIZE],
755 float rgba[NUM_CHANNELS][QUAD_SIZE])
756 {
757 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
758 unsigned j;
759 unsigned level = samp->level;
760 unsigned xpot = pot_level_size(samp->xpot, level);
761 unsigned ypot = pot_level_size(samp->ypot, level);
762 unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */
763 unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */
764 union tex_tile_address addr;
765
766 addr.value = 0;
767 addr.bits.level = samp->level;
768
769 for (j = 0; j < QUAD_SIZE; j++) {
770 int c;
771
772 float u = s[j] * xpot - 0.5F;
773 float v = t[j] * ypot - 0.5F;
774
775 int uflr = util_ifloor(u);
776 int vflr = util_ifloor(v);
777
778 float xw = u - (float)uflr;
779 float yw = v - (float)vflr;
780
781 int x0 = uflr & (xpot - 1);
782 int y0 = vflr & (ypot - 1);
783
784 const float *tx[4];
785
786 /* Can we fetch all four at once:
787 */
788 if (x0 < xmax && y0 < ymax) {
789 get_texel_quad_2d_no_border_single_tile(samp, addr, x0, y0, tx);
790 }
791 else {
792 unsigned x1 = (x0 + 1) & (xpot - 1);
793 unsigned y1 = (y0 + 1) & (ypot - 1);
794 get_texel_quad_2d_no_border(samp, addr, x0, y0, x1, y1, tx);
795 }
796
797 /* interpolate R, G, B, A */
798 for (c = 0; c < 4; c++) {
799 rgba[c][j] = lerp_2d(xw, yw,
800 tx[0][c], tx[1][c],
801 tx[2][c], tx[3][c]);
802 }
803 }
804 }
805
806
807 static INLINE void
808 img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
809 const float s[QUAD_SIZE],
810 const float t[QUAD_SIZE],
811 const float p[QUAD_SIZE],
812 const float lodbias[QUAD_SIZE],
813 float rgba[NUM_CHANNELS][QUAD_SIZE])
814 {
815 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
816 unsigned j;
817 unsigned level = samp->level;
818 unsigned xpot = pot_level_size(samp->xpot, level);
819 unsigned ypot = pot_level_size(samp->ypot, level);
820 union tex_tile_address addr;
821
822 addr.value = 0;
823 addr.bits.level = samp->level;
824
825 for (j = 0; j < QUAD_SIZE; j++) {
826 int c;
827
828 float u = s[j] * xpot;
829 float v = t[j] * ypot;
830
831 int uflr = util_ifloor(u);
832 int vflr = util_ifloor(v);
833
834 int x0 = uflr & (xpot - 1);
835 int y0 = vflr & (ypot - 1);
836
837 const float *out = get_texel_2d_no_border(samp, addr, x0, y0);
838
839 for (c = 0; c < 4; c++) {
840 rgba[c][j] = out[c];
841 }
842 }
843 }
844
845
846 static INLINE void
847 img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
848 const float s[QUAD_SIZE],
849 const float t[QUAD_SIZE],
850 const float p[QUAD_SIZE],
851 const float lodbias[QUAD_SIZE],
852 float rgba[NUM_CHANNELS][QUAD_SIZE])
853 {
854 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
855 unsigned j;
856 unsigned level = samp->level;
857 unsigned xpot = pot_level_size(samp->xpot, level);
858 unsigned ypot = pot_level_size(samp->ypot, level);
859 union tex_tile_address addr;
860
861 addr.value = 0;
862 addr.bits.level = samp->level;
863
864 for (j = 0; j < QUAD_SIZE; j++) {
865 int c;
866
867 float u = s[j] * xpot;
868 float v = t[j] * ypot;
869
870 int x0, y0;
871 const float *out;
872
873 x0 = util_ifloor(u);
874 if (x0 < 0)
875 x0 = 0;
876 else if (x0 > xpot - 1)
877 x0 = xpot - 1;
878
879 y0 = util_ifloor(v);
880 if (y0 < 0)
881 y0 = 0;
882 else if (y0 > ypot - 1)
883 y0 = ypot - 1;
884
885 out = get_texel_2d_no_border(samp, addr, x0, y0);
886
887 for (c = 0; c < 4; c++) {
888 rgba[c][j] = out[c];
889 }
890 }
891 }
892
893
894 static void
895 img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
896 const float s[QUAD_SIZE],
897 const float t[QUAD_SIZE],
898 const float p[QUAD_SIZE],
899 const float lodbias[QUAD_SIZE],
900 float rgba[NUM_CHANNELS][QUAD_SIZE])
901 {
902 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
903 const struct pipe_texture *texture = samp->texture;
904 unsigned level0, j;
905 int width;
906 int x[4];
907 union tex_tile_address addr;
908
909 level0 = samp->level;
910 width = u_minify(texture->width0, level0);
911
912 assert(width > 0);
913
914 addr.value = 0;
915 addr.bits.level = samp->level;
916
917 samp->nearest_texcoord_s(s, width, x);
918
919 for (j = 0; j < QUAD_SIZE; j++) {
920 const float *out = get_texel_2d(samp, addr, x[j], 0);
921 int c;
922 for (c = 0; c < 4; c++) {
923 rgba[c][j] = out[c];
924 }
925 }
926 }
927
928
929 static void
930 img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
931 const float s[QUAD_SIZE],
932 const float t[QUAD_SIZE],
933 const float p[QUAD_SIZE],
934 const float lodbias[QUAD_SIZE],
935 float rgba[NUM_CHANNELS][QUAD_SIZE])
936 {
937 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
938 const struct pipe_texture *texture = samp->texture;
939 unsigned level0, j;
940 int width, height;
941 int x[4], y[4];
942 union tex_tile_address addr;
943
944
945 level0 = samp->level;
946 width = u_minify(texture->width0, level0);
947 height = u_minify(texture->height0, level0);
948
949 assert(width > 0);
950 assert(height > 0);
951
952 addr.value = 0;
953 addr.bits.level = samp->level;
954
955 samp->nearest_texcoord_s(s, width, x);
956 samp->nearest_texcoord_t(t, height, y);
957
958 for (j = 0; j < QUAD_SIZE; j++) {
959 const float *out = get_texel_2d(samp, addr, x[j], y[j]);
960 int c;
961 for (c = 0; c < 4; c++) {
962 rgba[c][j] = out[c];
963 }
964 }
965 }
966
967
968 static INLINE union tex_tile_address
969 face(union tex_tile_address addr, unsigned face )
970 {
971 addr.bits.face = face;
972 return addr;
973 }
974
975
976 static void
977 img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
978 const float s[QUAD_SIZE],
979 const float t[QUAD_SIZE],
980 const float p[QUAD_SIZE],
981 const float lodbias[QUAD_SIZE],
982 float rgba[NUM_CHANNELS][QUAD_SIZE])
983 {
984 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
985 const struct pipe_texture *texture = samp->texture;
986 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
987 unsigned level0, j;
988 int width, height;
989 int x[4], y[4];
990 union tex_tile_address addr;
991
992 level0 = samp->level;
993 width = u_minify(texture->width0, level0);
994 height = u_minify(texture->height0, level0);
995
996 assert(width > 0);
997 assert(height > 0);
998
999 addr.value = 0;
1000 addr.bits.level = samp->level;
1001
1002 samp->nearest_texcoord_s(s, width, x);
1003 samp->nearest_texcoord_t(t, height, y);
1004
1005 for (j = 0; j < QUAD_SIZE; j++) {
1006 const float *out = get_texel_2d(samp, face(addr, faces[j]), x[j], y[j]);
1007 int c;
1008 for (c = 0; c < 4; c++) {
1009 rgba[c][j] = out[c];
1010 }
1011 }
1012 }
1013
1014
1015 static void
1016 img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
1017 const float s[QUAD_SIZE],
1018 const float t[QUAD_SIZE],
1019 const float p[QUAD_SIZE],
1020 const float lodbias[QUAD_SIZE],
1021 float rgba[NUM_CHANNELS][QUAD_SIZE])
1022 {
1023 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1024 const struct pipe_texture *texture = samp->texture;
1025 unsigned level0, j;
1026 int width, height, depth;
1027 int x[4], y[4], z[4];
1028 union tex_tile_address addr;
1029
1030 level0 = samp->level;
1031 width = u_minify(texture->width0, level0);
1032 height = u_minify(texture->height0, level0);
1033 depth = u_minify(texture->depth0, level0);
1034
1035 assert(width > 0);
1036 assert(height > 0);
1037 assert(depth > 0);
1038
1039 samp->nearest_texcoord_s(s, width, x);
1040 samp->nearest_texcoord_t(t, height, y);
1041 samp->nearest_texcoord_p(p, depth, z);
1042
1043 addr.value = 0;
1044 addr.bits.level = samp->level;
1045
1046 for (j = 0; j < QUAD_SIZE; j++) {
1047 const float *out = get_texel_3d(samp, addr, x[j], y[j], z[j]);
1048 int c;
1049 for (c = 0; c < 4; c++) {
1050 rgba[c][j] = out[c];
1051 }
1052 }
1053 }
1054
1055
1056 static void
1057 img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
1058 const float s[QUAD_SIZE],
1059 const float t[QUAD_SIZE],
1060 const float p[QUAD_SIZE],
1061 const float lodbias[QUAD_SIZE],
1062 float rgba[NUM_CHANNELS][QUAD_SIZE])
1063 {
1064 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1065 const struct pipe_texture *texture = samp->texture;
1066 unsigned level0, j;
1067 int width;
1068 int x0[4], x1[4];
1069 float xw[4]; /* weights */
1070 union tex_tile_address addr;
1071
1072 level0 = samp->level;
1073 width = u_minify(texture->width0, level0);
1074
1075 assert(width > 0);
1076
1077 addr.value = 0;
1078 addr.bits.level = samp->level;
1079
1080 samp->linear_texcoord_s(s, width, x0, x1, xw);
1081
1082 for (j = 0; j < QUAD_SIZE; j++) {
1083 const float *tx0 = get_texel_2d(samp, addr, x0[j], 0);
1084 const float *tx1 = get_texel_2d(samp, addr, x1[j], 0);
1085 int c;
1086
1087 /* interpolate R, G, B, A */
1088 for (c = 0; c < 4; c++) {
1089 rgba[c][j] = lerp(xw[j], tx0[c], tx1[c]);
1090 }
1091 }
1092 }
1093
1094
1095 static void
1096 img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
1097 const float s[QUAD_SIZE],
1098 const float t[QUAD_SIZE],
1099 const float p[QUAD_SIZE],
1100 const float lodbias[QUAD_SIZE],
1101 float rgba[NUM_CHANNELS][QUAD_SIZE])
1102 {
1103 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1104 const struct pipe_texture *texture = samp->texture;
1105 unsigned level0, j;
1106 int width, height;
1107 int x0[4], y0[4], x1[4], y1[4];
1108 float xw[4], yw[4]; /* weights */
1109 union tex_tile_address addr;
1110
1111 level0 = samp->level;
1112 width = u_minify(texture->width0, level0);
1113 height = u_minify(texture->height0, level0);
1114
1115 assert(width > 0);
1116 assert(height > 0);
1117
1118 addr.value = 0;
1119 addr.bits.level = samp->level;
1120
1121 samp->linear_texcoord_s(s, width, x0, x1, xw);
1122 samp->linear_texcoord_t(t, height, y0, y1, yw);
1123
1124 for (j = 0; j < QUAD_SIZE; j++) {
1125 const float *tx0 = get_texel_2d(samp, addr, x0[j], y0[j]);
1126 const float *tx1 = get_texel_2d(samp, addr, x1[j], y0[j]);
1127 const float *tx2 = get_texel_2d(samp, addr, x0[j], y1[j]);
1128 const float *tx3 = get_texel_2d(samp, addr, x1[j], y1[j]);
1129 int c;
1130
1131 /* interpolate R, G, B, A */
1132 for (c = 0; c < 4; c++) {
1133 rgba[c][j] = lerp_2d(xw[j], yw[j],
1134 tx0[c], tx1[c],
1135 tx2[c], tx3[c]);
1136 }
1137 }
1138 }
1139
1140
1141 static void
1142 img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler,
1143 const float s[QUAD_SIZE],
1144 const float t[QUAD_SIZE],
1145 const float p[QUAD_SIZE],
1146 const float lodbias[QUAD_SIZE],
1147 float rgba[NUM_CHANNELS][QUAD_SIZE])
1148 {
1149 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1150 const struct pipe_texture *texture = samp->texture;
1151 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
1152 unsigned level0, j;
1153 int width, height;
1154 int x0[4], y0[4], x1[4], y1[4];
1155 float xw[4], yw[4]; /* weights */
1156 union tex_tile_address addr;
1157
1158 level0 = samp->level;
1159 width = u_minify(texture->width0, level0);
1160 height = u_minify(texture->height0, level0);
1161
1162 assert(width > 0);
1163 assert(height > 0);
1164
1165 addr.value = 0;
1166 addr.bits.level = samp->level;
1167
1168 samp->linear_texcoord_s(s, width, x0, x1, xw);
1169 samp->linear_texcoord_t(t, height, y0, y1, yw);
1170
1171 for (j = 0; j < QUAD_SIZE; j++) {
1172 union tex_tile_address addrj = face(addr, faces[j]);
1173 const float *tx0 = get_texel_2d(samp, addrj, x0[j], y0[j]);
1174 const float *tx1 = get_texel_2d(samp, addrj, x1[j], y0[j]);
1175 const float *tx2 = get_texel_2d(samp, addrj, x0[j], y1[j]);
1176 const float *tx3 = get_texel_2d(samp, addrj, x1[j], y1[j]);
1177 int c;
1178
1179 /* interpolate R, G, B, A */
1180 for (c = 0; c < 4; c++) {
1181 rgba[c][j] = lerp_2d(xw[j], yw[j],
1182 tx0[c], tx1[c],
1183 tx2[c], tx3[c]);
1184 }
1185 }
1186 }
1187
1188
1189 static void
1190 img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
1191 const float s[QUAD_SIZE],
1192 const float t[QUAD_SIZE],
1193 const float p[QUAD_SIZE],
1194 const float lodbias[QUAD_SIZE],
1195 float rgba[NUM_CHANNELS][QUAD_SIZE])
1196 {
1197 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1198 const struct pipe_texture *texture = samp->texture;
1199 unsigned level0, j;
1200 int width, height, depth;
1201 int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
1202 float xw[4], yw[4], zw[4]; /* interpolation weights */
1203 union tex_tile_address addr;
1204
1205 level0 = samp->level;
1206 width = u_minify(texture->width0, level0);
1207 height = u_minify(texture->height0, level0);
1208 depth = u_minify(texture->depth0, level0);
1209
1210 addr.value = 0;
1211 addr.bits.level = level0;
1212
1213 assert(width > 0);
1214 assert(height > 0);
1215 assert(depth > 0);
1216
1217 samp->linear_texcoord_s(s, width, x0, x1, xw);
1218 samp->linear_texcoord_t(t, height, y0, y1, yw);
1219 samp->linear_texcoord_p(p, depth, z0, z1, zw);
1220
1221 for (j = 0; j < QUAD_SIZE; j++) {
1222 int c;
1223
1224 const float *tx00 = get_texel_3d(samp, addr, x0[j], y0[j], z0[j]);
1225 const float *tx01 = get_texel_3d(samp, addr, x1[j], y0[j], z0[j]);
1226 const float *tx02 = get_texel_3d(samp, addr, x0[j], y1[j], z0[j]);
1227 const float *tx03 = get_texel_3d(samp, addr, x1[j], y1[j], z0[j]);
1228
1229 const float *tx10 = get_texel_3d(samp, addr, x0[j], y0[j], z1[j]);
1230 const float *tx11 = get_texel_3d(samp, addr, x1[j], y0[j], z1[j]);
1231 const float *tx12 = get_texel_3d(samp, addr, x0[j], y1[j], z1[j]);
1232 const float *tx13 = get_texel_3d(samp, addr, x1[j], y1[j], z1[j]);
1233
1234 /* interpolate R, G, B, A */
1235 for (c = 0; c < 4; c++) {
1236 rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1237 tx00[c], tx01[c],
1238 tx02[c], tx03[c],
1239 tx10[c], tx11[c],
1240 tx12[c], tx13[c]);
1241 }
1242 }
1243 }
1244
1245
1246 /* Calculate level of detail for every fragment.
1247 * Note that lambda has already been biased by global LOD bias.
1248 */
1249 static INLINE void
1250 compute_lod(const struct pipe_sampler_state *sampler,
1251 const float biased_lambda,
1252 const float lodbias[QUAD_SIZE],
1253 float lod[QUAD_SIZE])
1254 {
1255 uint i;
1256
1257 for (i = 0; i < QUAD_SIZE; i++) {
1258 lod[i] = biased_lambda + lodbias[i];
1259 lod[i] = CLAMP(lod[i], sampler->min_lod, sampler->max_lod);
1260 }
1261 }
1262
1263
1264 static void
1265 mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
1266 const float s[QUAD_SIZE],
1267 const float t[QUAD_SIZE],
1268 const float p[QUAD_SIZE],
1269 const float lodbias[QUAD_SIZE],
1270 float rgba[NUM_CHANNELS][QUAD_SIZE])
1271 {
1272 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1273 const struct pipe_texture *texture = samp->texture;
1274 int level0;
1275 float lambda;
1276 float lod[QUAD_SIZE];
1277
1278 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1279
1280 compute_lod(samp->sampler, lambda, lodbias, lod);
1281
1282 /* XXX: Take into account all lod values.
1283 */
1284 lambda = lod[0];
1285 level0 = (int)lambda;
1286
1287 if (lambda < 0.0) {
1288 samp->level = 0;
1289 samp->mag_img_filter( tgsi_sampler, s, t, p, NULL, rgba );
1290 }
1291 else if (level0 >= texture->last_level) {
1292 samp->level = texture->last_level;
1293 samp->min_img_filter( tgsi_sampler, s, t, p, NULL, rgba );
1294 }
1295 else {
1296 float levelBlend = lambda - level0;
1297 float rgba0[4][4];
1298 float rgba1[4][4];
1299 int c,j;
1300
1301 samp->level = level0;
1302 samp->min_img_filter( tgsi_sampler, s, t, p, NULL, rgba0 );
1303
1304 samp->level = level0+1;
1305 samp->min_img_filter( tgsi_sampler, s, t, p, NULL, rgba1 );
1306
1307 for (j = 0; j < QUAD_SIZE; j++) {
1308 for (c = 0; c < 4; c++) {
1309 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1310 }
1311 }
1312 }
1313 }
1314
1315
1316 static void
1317 mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
1318 const float s[QUAD_SIZE],
1319 const float t[QUAD_SIZE],
1320 const float p[QUAD_SIZE],
1321 const float lodbias[QUAD_SIZE],
1322 float rgba[NUM_CHANNELS][QUAD_SIZE])
1323 {
1324 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1325 const struct pipe_texture *texture = samp->texture;
1326 float lambda;
1327 float lod[QUAD_SIZE];
1328
1329 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1330
1331 compute_lod(samp->sampler, lambda, lodbias, lod);
1332
1333 /* XXX: Take into account all lod values.
1334 */
1335 lambda = lod[0];
1336
1337 if (lambda < 0.0) {
1338 samp->level = 0;
1339 samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
1340 }
1341 else {
1342 samp->level = (int)(lambda + 0.5) ;
1343 samp->level = MIN2(samp->level, (int)texture->last_level);
1344 samp->min_img_filter( tgsi_sampler, s, t, p, NULL, rgba );
1345 }
1346
1347 #if 0
1348 printf("RGBA %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
1349 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
1350 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
1351 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
1352 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
1353 #endif
1354 }
1355
1356
1357 static void
1358 mip_filter_none(struct tgsi_sampler *tgsi_sampler,
1359 const float s[QUAD_SIZE],
1360 const float t[QUAD_SIZE],
1361 const float p[QUAD_SIZE],
1362 const float lodbias[QUAD_SIZE],
1363 float rgba[NUM_CHANNELS][QUAD_SIZE])
1364 {
1365 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1366 float lambda;
1367 float lod[QUAD_SIZE];
1368
1369 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1370
1371 compute_lod(samp->sampler, lambda, lodbias, lod);
1372
1373 /* XXX: Take into account all lod values.
1374 */
1375 lambda = lod[0];
1376
1377 if (lambda < 0.0) {
1378 samp->mag_img_filter( tgsi_sampler, s, t, p, NULL, rgba );
1379 }
1380 else {
1381 samp->min_img_filter( tgsi_sampler, s, t, p, NULL, rgba );
1382 }
1383 }
1384
1385
1386
1387 /**
1388 * Specialized version of mip_filter_linear with hard-wired calls to
1389 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
1390 */
1391 static void
1392 mip_filter_linear_2d_linear_repeat_POT(
1393 struct tgsi_sampler *tgsi_sampler,
1394 const float s[QUAD_SIZE],
1395 const float t[QUAD_SIZE],
1396 const float p[QUAD_SIZE],
1397 const float lodbias[QUAD_SIZE],
1398 float rgba[NUM_CHANNELS][QUAD_SIZE])
1399 {
1400 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1401 const struct pipe_texture *texture = samp->texture;
1402 int level0;
1403 float lambda;
1404 float lod[QUAD_SIZE];
1405
1406 lambda = compute_lambda_2d(samp, s, t, p) + samp->sampler->lod_bias;
1407
1408 compute_lod(samp->sampler, lambda, lodbias, lod);
1409
1410 /* XXX: Take into account all lod values.
1411 */
1412 lambda = lod[0];
1413 level0 = (int)lambda;
1414
1415 /* Catches both negative and large values of level0:
1416 */
1417 if ((unsigned)level0 >= texture->last_level) {
1418 if (level0 < 0)
1419 samp->level = 0;
1420 else
1421 samp->level = texture->last_level;
1422
1423 img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, NULL, rgba );
1424 }
1425 else {
1426 float levelBlend = lambda - level0;
1427 float rgba0[4][4];
1428 float rgba1[4][4];
1429 int c,j;
1430
1431 samp->level = level0;
1432 img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, NULL, rgba0 );
1433
1434 samp->level = level0+1;
1435 img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, NULL, rgba1 );
1436
1437 for (j = 0; j < QUAD_SIZE; j++) {
1438 for (c = 0; c < 4; c++) {
1439 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1440 }
1441 }
1442 }
1443 }
1444
1445
1446
1447 /**
1448 * Do shadow/depth comparisons.
1449 */
1450 static void
1451 sample_compare(struct tgsi_sampler *tgsi_sampler,
1452 const float s[QUAD_SIZE],
1453 const float t[QUAD_SIZE],
1454 const float p[QUAD_SIZE],
1455 const float lodbias[QUAD_SIZE],
1456 float rgba[NUM_CHANNELS][QUAD_SIZE])
1457 {
1458 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1459 const struct pipe_sampler_state *sampler = samp->sampler;
1460 int j, k0, k1, k2, k3;
1461 float val;
1462
1463 samp->mip_filter( tgsi_sampler, s, t, p, lodbias, rgba );
1464
1465 /**
1466 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
1467 * When we sampled the depth texture, the depth value was put into all
1468 * RGBA channels. We look at the red channel here.
1469 */
1470
1471 /* compare four texcoords vs. four texture samples */
1472 switch (sampler->compare_func) {
1473 case PIPE_FUNC_LESS:
1474 k0 = p[0] < rgba[0][0];
1475 k1 = p[1] < rgba[0][1];
1476 k2 = p[2] < rgba[0][2];
1477 k3 = p[3] < rgba[0][3];
1478 break;
1479 case PIPE_FUNC_LEQUAL:
1480 k0 = p[0] <= rgba[0][0];
1481 k1 = p[1] <= rgba[0][1];
1482 k2 = p[2] <= rgba[0][2];
1483 k3 = p[3] <= rgba[0][3];
1484 break;
1485 case PIPE_FUNC_GREATER:
1486 k0 = p[0] > rgba[0][0];
1487 k1 = p[1] > rgba[0][1];
1488 k2 = p[2] > rgba[0][2];
1489 k3 = p[3] > rgba[0][3];
1490 break;
1491 case PIPE_FUNC_GEQUAL:
1492 k0 = p[0] >= rgba[0][0];
1493 k1 = p[1] >= rgba[0][1];
1494 k2 = p[2] >= rgba[0][2];
1495 k3 = p[3] >= rgba[0][3];
1496 break;
1497 case PIPE_FUNC_EQUAL:
1498 k0 = p[0] == rgba[0][0];
1499 k1 = p[1] == rgba[0][1];
1500 k2 = p[2] == rgba[0][2];
1501 k3 = p[3] == rgba[0][3];
1502 break;
1503 case PIPE_FUNC_NOTEQUAL:
1504 k0 = p[0] != rgba[0][0];
1505 k1 = p[1] != rgba[0][1];
1506 k2 = p[2] != rgba[0][2];
1507 k3 = p[3] != rgba[0][3];
1508 break;
1509 case PIPE_FUNC_ALWAYS:
1510 k0 = k1 = k2 = k3 = 1;
1511 break;
1512 case PIPE_FUNC_NEVER:
1513 k0 = k1 = k2 = k3 = 0;
1514 break;
1515 default:
1516 k0 = k1 = k2 = k3 = 0;
1517 assert(0);
1518 break;
1519 }
1520
1521 /* convert four pass/fail values to an intensity in [0,1] */
1522 val = 0.25F * (k0 + k1 + k2 + k3);
1523
1524 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
1525 for (j = 0; j < 4; j++) {
1526 rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
1527 rgba[3][j] = 1.0F;
1528 }
1529 }
1530
1531
1532 /**
1533 * Compute which cube face is referenced by each texcoord and put that
1534 * info into the sampler faces[] array. Then sample the cube faces
1535 */
1536 static void
1537 sample_cube(struct tgsi_sampler *tgsi_sampler,
1538 const float s[QUAD_SIZE],
1539 const float t[QUAD_SIZE],
1540 const float p[QUAD_SIZE],
1541 const float lodbias[QUAD_SIZE],
1542 float rgba[NUM_CHANNELS][QUAD_SIZE])
1543 {
1544 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1545 unsigned j;
1546 float ssss[4], tttt[4];
1547
1548 /*
1549 major axis
1550 direction target sc tc ma
1551 ---------- ------------------------------- --- --- ---
1552 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
1553 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
1554 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
1555 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
1556 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
1557 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
1558 */
1559 for (j = 0; j < QUAD_SIZE; j++) {
1560 float rx = s[j];
1561 float ry = t[j];
1562 float rz = p[j];
1563 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
1564 unsigned face;
1565 float sc, tc, ma;
1566
1567 if (arx >= ary && arx >= arz) {
1568 if (rx >= 0.0F) {
1569 face = PIPE_TEX_FACE_POS_X;
1570 sc = -rz;
1571 tc = -ry;
1572 ma = arx;
1573 }
1574 else {
1575 face = PIPE_TEX_FACE_NEG_X;
1576 sc = rz;
1577 tc = -ry;
1578 ma = arx;
1579 }
1580 }
1581 else if (ary >= arx && ary >= arz) {
1582 if (ry >= 0.0F) {
1583 face = PIPE_TEX_FACE_POS_Y;
1584 sc = rx;
1585 tc = rz;
1586 ma = ary;
1587 }
1588 else {
1589 face = PIPE_TEX_FACE_NEG_Y;
1590 sc = rx;
1591 tc = -rz;
1592 ma = ary;
1593 }
1594 }
1595 else {
1596 if (rz > 0.0F) {
1597 face = PIPE_TEX_FACE_POS_Z;
1598 sc = rx;
1599 tc = -ry;
1600 ma = arz;
1601 }
1602 else {
1603 face = PIPE_TEX_FACE_NEG_Z;
1604 sc = -rx;
1605 tc = -ry;
1606 ma = arz;
1607 }
1608 }
1609
1610 {
1611 const float ima = 1.0 / ma;
1612 ssss[j] = ( sc * ima + 1.0F ) * 0.5F;
1613 tttt[j] = ( tc * ima + 1.0F ) * 0.5F;
1614 samp->faces[j] = face;
1615 }
1616 }
1617
1618 /* In our little pipeline, the compare stage is next. If compare
1619 * is not active, this will point somewhere deeper into the
1620 * pipeline, eg. to mip_filter or even img_filter.
1621 */
1622 samp->compare(tgsi_sampler, ssss, tttt, NULL, lodbias, rgba);
1623 }
1624
1625
1626
1627 static wrap_nearest_func
1628 get_nearest_unorm_wrap(unsigned mode)
1629 {
1630 switch (mode) {
1631 case PIPE_TEX_WRAP_CLAMP:
1632 return wrap_nearest_unorm_clamp;
1633 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1634 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1635 return wrap_nearest_unorm_clamp_to_border;
1636 default:
1637 assert(0);
1638 return wrap_nearest_unorm_clamp;
1639 }
1640 }
1641
1642
1643 static wrap_nearest_func
1644 get_nearest_wrap(unsigned mode)
1645 {
1646 switch (mode) {
1647 case PIPE_TEX_WRAP_REPEAT:
1648 return wrap_nearest_repeat;
1649 case PIPE_TEX_WRAP_CLAMP:
1650 return wrap_nearest_clamp;
1651 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1652 return wrap_nearest_clamp_to_edge;
1653 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1654 return wrap_nearest_clamp_to_border;
1655 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1656 return wrap_nearest_mirror_repeat;
1657 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1658 return wrap_nearest_mirror_clamp;
1659 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1660 return wrap_nearest_mirror_clamp_to_edge;
1661 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1662 return wrap_nearest_mirror_clamp_to_border;
1663 default:
1664 assert(0);
1665 return wrap_nearest_repeat;
1666 }
1667 }
1668
1669
1670 static wrap_linear_func
1671 get_linear_unorm_wrap(unsigned mode)
1672 {
1673 switch (mode) {
1674 case PIPE_TEX_WRAP_CLAMP:
1675 return wrap_linear_unorm_clamp;
1676 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1677 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1678 return wrap_linear_unorm_clamp_to_border;
1679 default:
1680 assert(0);
1681 return wrap_linear_unorm_clamp;
1682 }
1683 }
1684
1685
1686 static wrap_linear_func
1687 get_linear_wrap(unsigned mode)
1688 {
1689 switch (mode) {
1690 case PIPE_TEX_WRAP_REPEAT:
1691 return wrap_linear_repeat;
1692 case PIPE_TEX_WRAP_CLAMP:
1693 return wrap_linear_clamp;
1694 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1695 return wrap_linear_clamp_to_edge;
1696 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1697 return wrap_linear_clamp_to_border;
1698 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1699 return wrap_linear_mirror_repeat;
1700 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1701 return wrap_linear_mirror_clamp;
1702 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1703 return wrap_linear_mirror_clamp_to_edge;
1704 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1705 return wrap_linear_mirror_clamp_to_border;
1706 default:
1707 assert(0);
1708 return wrap_linear_repeat;
1709 }
1710 }
1711
1712
1713 static compute_lambda_func
1714 get_lambda_func(const union sp_sampler_key key)
1715 {
1716 if (key.bits.processor == TGSI_PROCESSOR_VERTEX)
1717 return compute_lambda_vert;
1718
1719 switch (key.bits.target) {
1720 case PIPE_TEXTURE_1D:
1721 return compute_lambda_1d;
1722 case PIPE_TEXTURE_2D:
1723 case PIPE_TEXTURE_CUBE:
1724 return compute_lambda_2d;
1725 case PIPE_TEXTURE_3D:
1726 return compute_lambda_3d;
1727 default:
1728 assert(0);
1729 return compute_lambda_1d;
1730 }
1731 }
1732
1733
1734 static filter_func
1735 get_img_filter(const union sp_sampler_key key,
1736 unsigned filter,
1737 const struct pipe_sampler_state *sampler)
1738 {
1739 switch (key.bits.target) {
1740 case PIPE_TEXTURE_1D:
1741 if (filter == PIPE_TEX_FILTER_NEAREST)
1742 return img_filter_1d_nearest;
1743 else
1744 return img_filter_1d_linear;
1745 break;
1746 case PIPE_TEXTURE_2D:
1747 /* Try for fast path:
1748 */
1749 if (key.bits.is_pot &&
1750 sampler->wrap_s == sampler->wrap_t &&
1751 sampler->normalized_coords)
1752 {
1753 switch (sampler->wrap_s) {
1754 case PIPE_TEX_WRAP_REPEAT:
1755 switch (filter) {
1756 case PIPE_TEX_FILTER_NEAREST:
1757 return img_filter_2d_nearest_repeat_POT;
1758 case PIPE_TEX_FILTER_LINEAR:
1759 return img_filter_2d_linear_repeat_POT;
1760 default:
1761 break;
1762 }
1763 break;
1764 case PIPE_TEX_WRAP_CLAMP:
1765 switch (filter) {
1766 case PIPE_TEX_FILTER_NEAREST:
1767 return img_filter_2d_nearest_clamp_POT;
1768 default:
1769 break;
1770 }
1771 }
1772 }
1773 /* Otherwise use default versions:
1774 */
1775 if (filter == PIPE_TEX_FILTER_NEAREST)
1776 return img_filter_2d_nearest;
1777 else
1778 return img_filter_2d_linear;
1779 break;
1780 case PIPE_TEXTURE_CUBE:
1781 if (filter == PIPE_TEX_FILTER_NEAREST)
1782 return img_filter_cube_nearest;
1783 else
1784 return img_filter_cube_linear;
1785 break;
1786 case PIPE_TEXTURE_3D:
1787 if (filter == PIPE_TEX_FILTER_NEAREST)
1788 return img_filter_3d_nearest;
1789 else
1790 return img_filter_3d_linear;
1791 break;
1792 default:
1793 assert(0);
1794 return img_filter_1d_nearest;
1795 }
1796 }
1797
1798
1799 /**
1800 * Bind the given texture object and texture cache to the sampler varient.
1801 */
1802 void
1803 sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp,
1804 struct softpipe_tex_tile_cache *tex_cache,
1805 const struct pipe_texture *texture )
1806 {
1807 const struct pipe_sampler_state *sampler = samp->sampler;
1808
1809 samp->texture = texture;
1810 samp->cache = tex_cache;
1811 samp->xpot = util_unsigned_logbase2( texture->width0 );
1812 samp->ypot = util_unsigned_logbase2( texture->height0 );
1813 samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level);
1814 }
1815
1816
1817 void
1818 sp_sampler_varient_destroy( struct sp_sampler_varient *samp )
1819 {
1820 FREE(samp);
1821 }
1822
1823
1824 /**
1825 * Create a sampler varient for a given set of non-orthogonal state.
1826 */
1827 struct sp_sampler_varient *
1828 sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
1829 const union sp_sampler_key key )
1830 {
1831 struct sp_sampler_varient *samp = CALLOC_STRUCT(sp_sampler_varient);
1832 if (!samp)
1833 return NULL;
1834
1835 samp->sampler = sampler;
1836 samp->key = key;
1837
1838 /* Note that (for instance) linear_texcoord_s and
1839 * nearest_texcoord_s may be active at the same time, if the
1840 * sampler min_img_filter differs from its mag_img_filter.
1841 */
1842 if (sampler->normalized_coords) {
1843 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
1844 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
1845 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
1846
1847 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
1848 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
1849 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
1850 }
1851 else {
1852 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
1853 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
1854 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
1855
1856 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
1857 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
1858 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
1859 }
1860
1861 samp->compute_lambda = get_lambda_func( key );
1862
1863 samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
1864 samp->mag_img_filter = get_img_filter(key, sampler->mag_img_filter, sampler);
1865
1866 switch (sampler->min_mip_filter) {
1867 case PIPE_TEX_MIPFILTER_NONE:
1868 if (sampler->min_img_filter == sampler->mag_img_filter)
1869 samp->mip_filter = samp->min_img_filter;
1870 else
1871 samp->mip_filter = mip_filter_none;
1872 break;
1873
1874 case PIPE_TEX_MIPFILTER_NEAREST:
1875 samp->mip_filter = mip_filter_nearest;
1876 break;
1877
1878 case PIPE_TEX_MIPFILTER_LINEAR:
1879 if (key.bits.is_pot &&
1880 sampler->min_img_filter == sampler->mag_img_filter &&
1881 sampler->normalized_coords &&
1882 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
1883 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
1884 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR)
1885 {
1886 samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT;
1887 }
1888 else
1889 {
1890 samp->mip_filter = mip_filter_linear;
1891 }
1892 break;
1893 }
1894
1895 if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
1896 samp->compare = sample_compare;
1897 }
1898 else {
1899 /* Skip compare operation by promoting the mip_filter function
1900 * pointer:
1901 */
1902 samp->compare = samp->mip_filter;
1903 }
1904
1905 if (key.bits.target == PIPE_TEXTURE_CUBE) {
1906 samp->base.get_samples = sample_cube;
1907 }
1908 else {
1909 samp->faces[0] = 0;
1910 samp->faces[1] = 0;
1911 samp->faces[2] = 0;
1912 samp->faces[3] = 0;
1913
1914 /* Skip cube face determination by promoting the compare
1915 * function pointer:
1916 */
1917 samp->base.get_samples = samp->compare;
1918 }
1919
1920 return samp;
1921 }