projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
softpipe: Use portable INLINE macro.
[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 * 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 float lodbias)
519 {
520 const struct pipe_texture *texture = samp->texture;
521 const struct pipe_sampler_state *sampler = samp->sampler;
522 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
523 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
524 float rho = MAX2(dsdx, dsdy) * texture->width[0];
525 float lambda;
526
527 lambda = util_fast_log2(rho);
528 lambda += lodbias + sampler->lod_bias;
529 lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
530
531 return lambda;
532 }
533
534
535 static float
536 compute_lambda_2d(const struct sp_sampler_varient *samp,
537 const float s[QUAD_SIZE],
538 const float t[QUAD_SIZE],
539 const float p[QUAD_SIZE],
540 float lodbias)
541 {
542 const struct pipe_texture *texture = samp->texture;
543 const struct pipe_sampler_state *sampler = samp->sampler;
544 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
545 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
546 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
547 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
548 float maxx = MAX2(dsdx, dsdy) * texture->width[0];
549 float maxy = MAX2(dtdx, dtdy) * texture->height[0];
550 float rho = MAX2(maxx, maxy);
551 float lambda;
552
553 lambda = util_fast_log2(rho);
554 lambda += lodbias + sampler->lod_bias;
555 lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
556
557 return lambda;
558 }
559
560
561 static float
562 compute_lambda_3d(const struct sp_sampler_varient *samp,
563 const float s[QUAD_SIZE],
564 const float t[QUAD_SIZE],
565 const float p[QUAD_SIZE],
566 float lodbias)
567 {
568 const struct pipe_texture *texture = samp->texture;
569 const struct pipe_sampler_state *sampler = samp->sampler;
570 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
571 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
572 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
573 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
574 float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
575 float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
576 float maxx = MAX2(dsdx, dsdy) * texture->width[0];
577 float maxy = MAX2(dtdx, dtdy) * texture->height[0];
578 float maxz = MAX2(dpdx, dpdy) * texture->depth[0];
579 float rho, lambda;
580
581 rho = MAX2(maxx, maxy);
582 rho = MAX2(rho, maxz);
583
584 lambda = util_fast_log2(rho);
585 lambda += lodbias + sampler->lod_bias;
586 lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
587
588 return lambda;
589 }
590
591
592 /**
593 * Compute lambda for a vertex texture sampler.
594 * Since there aren't derivatives to use, just return the LOD bias.
595 */
596 static float
597 compute_lambda_vert(const struct sp_sampler_varient *samp,
598 const float s[QUAD_SIZE],
599 const float t[QUAD_SIZE],
600 const float p[QUAD_SIZE],
601 float lodbias)
602 {
603 return lodbias;
604 }
605
606
607
608 /**
609 * Get a texel from a texture, using the texture tile cache.
610 *
611 * \param addr the template tex address containing cube, z, face info.
612 * \param x the x coord of texel within 2D image
613 * \param y the y coord of texel within 2D image
614 * \param rgba the quad to put the texel/color into
615 *
616 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
617 * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1...
618 */
619
620
621
622
623 static INLINE const float *
624 get_texel_2d_no_border(const struct sp_sampler_varient *samp,
625 union tex_tile_address addr, int x, int y)
626 {
627 const struct softpipe_tex_cached_tile *tile;
628
629 addr.bits.x = x / TILE_SIZE;
630 addr.bits.y = y / TILE_SIZE;
631 y %= TILE_SIZE;
632 x %= TILE_SIZE;
633
634 tile = sp_get_cached_tile_tex(samp->cache, addr);
635
636 return &tile->data.color[y][x][0];
637 }
638
639
640 static INLINE const float *
641 get_texel_2d(const struct sp_sampler_varient *samp,
642 union tex_tile_address addr, int x, int y)
643 {
644 const struct pipe_texture *texture = samp->texture;
645 unsigned level = addr.bits.level;
646
647 if (x < 0 || x >= (int) texture->width[level] ||
648 y < 0 || y >= (int) texture->height[level]) {
649 return samp->sampler->border_color;
650 }
651 else {
652 return get_texel_2d_no_border( samp, addr, x, y );
653 }
654 }
655
656
657 /* Gather a quad of adjacent texels within a tile:
658 */
659 static INLINE void
660 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_varient *samp,
661 union tex_tile_address addr,
662 unsigned x, unsigned y,
663 const float *out[4])
664 {
665 const struct softpipe_tex_cached_tile *tile;
666
667 addr.bits.x = x / TILE_SIZE;
668 addr.bits.y = y / TILE_SIZE;
669 y %= TILE_SIZE;
670 x %= TILE_SIZE;
671
672 tile = sp_get_cached_tile_tex(samp->cache, addr);
673
674 out[0] = &tile->data.color[y ][x ][0];
675 out[1] = &tile->data.color[y ][x+1][0];
676 out[2] = &tile->data.color[y+1][x ][0];
677 out[3] = &tile->data.color[y+1][x+1][0];
678 }
679
680
681 /* Gather a quad of potentially non-adjacent texels:
682 */
683 static INLINE void
684 get_texel_quad_2d_no_border(const struct sp_sampler_varient *samp,
685 union tex_tile_address addr,
686 int x0, int y0,
687 int x1, int y1,
688 const float *out[4])
689 {
690 out[0] = get_texel_2d_no_border( samp, addr, x0, y0 );
691 out[1] = get_texel_2d_no_border( samp, addr, x1, y0 );
692 out[2] = get_texel_2d_no_border( samp, addr, x0, y1 );
693 out[3] = get_texel_2d_no_border( samp, addr, x1, y1 );
694 }
695
696 /* Can involve a lot of unnecessary checks for border color:
697 */
698 static INLINE void
699 get_texel_quad_2d(const struct sp_sampler_varient *samp,
700 union tex_tile_address addr,
701 int x0, int y0,
702 int x1, int y1,
703 const float *out[4])
704 {
705 out[0] = get_texel_2d( samp, addr, x0, y0 );
706 out[1] = get_texel_2d( samp, addr, x1, y0 );
707 out[3] = get_texel_2d( samp, addr, x1, y1 );
708 out[2] = get_texel_2d( samp, addr, x0, y1 );
709 }
710
711
712
713 /* 3d varients:
714 */
715 static INLINE const float *
716 get_texel_3d_no_border(const struct sp_sampler_varient *samp,
717 union tex_tile_address addr, int x, int y, int z)
718 {
719 const struct softpipe_tex_cached_tile *tile;
720
721 addr.bits.x = x / TILE_SIZE;
722 addr.bits.y = y / TILE_SIZE;
723 addr.bits.z = z;
724 y %= TILE_SIZE;
725 x %= TILE_SIZE;
726
727 tile = sp_get_cached_tile_tex(samp->cache, addr);
728
729 return &tile->data.color[y][x][0];
730 }
731
732
733 static INLINE const float *
734 get_texel_3d(const struct sp_sampler_varient *samp,
735 union tex_tile_address addr, int x, int y, int z)
736 {
737 const struct pipe_texture *texture = samp->texture;
738 unsigned level = addr.bits.level;
739
740 if (x < 0 || x >= (int) texture->width[level] ||
741 y < 0 || y >= (int) texture->height[level] ||
742 z < 0 || z >= (int) texture->depth[level]) {
743 return samp->sampler->border_color;
744 }
745 else {
746 return get_texel_3d_no_border( samp, addr, x, y, z );
747 }
748 }
749
750
751 /**
752 * Given the logbase2 of a mipmap's base level size and a mipmap level,
753 * return the size (in texels) of that mipmap level.
754 * For example, if level[0].width = 256 then base_pot will be 8.
755 * If level = 2, then we'll return 64 (the width at level=2).
756 * Return 1 if level > base_pot.
757 */
758 static INLINE unsigned
759 pot_level_size(unsigned base_pot, unsigned level)
760 {
761 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
762 }
763
764
765 /* Some image-filter fastpaths:
766 */
767 static INLINE void
768 img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
769 const float s[QUAD_SIZE],
770 const float t[QUAD_SIZE],
771 const float p[QUAD_SIZE],
772 float lodbias,
773 float rgba[NUM_CHANNELS][QUAD_SIZE])
774 {
775 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
776 unsigned j;
777 unsigned level = samp->level;
778 unsigned xpot = pot_level_size(samp->xpot, level);
779 unsigned ypot = pot_level_size(samp->ypot, level);
780 unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */
781 unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */
782 union tex_tile_address addr;
783
784 addr.value = 0;
785 addr.bits.level = samp->level;
786
787 for (j = 0; j < QUAD_SIZE; j++) {
788 int c;
789
790 float u = s[j] * xpot - 0.5F;
791 float v = t[j] * ypot - 0.5F;
792
793 int uflr = util_ifloor(u);
794 int vflr = util_ifloor(v);
795
796 float xw = u - (float)uflr;
797 float yw = v - (float)vflr;
798
799 int x0 = uflr & (xpot - 1);
800 int y0 = vflr & (ypot - 1);
801
802 const float *tx[4];
803
804 /* Can we fetch all four at once:
805 */
806 if (x0 < xmax && y0 < ymax) {
807 get_texel_quad_2d_no_border_single_tile(samp, addr, x0, y0, tx);
808 }
809 else {
810 unsigned x1 = (x0 + 1) & (xpot - 1);
811 unsigned y1 = (y0 + 1) & (ypot - 1);
812 get_texel_quad_2d_no_border(samp, addr, x0, y0, x1, y1, tx);
813 }
814
815 /* interpolate R, G, B, A */
816 for (c = 0; c < 4; c++) {
817 rgba[c][j] = lerp_2d(xw, yw,
818 tx[0][c], tx[1][c],
819 tx[2][c], tx[3][c]);
820 }
821 }
822 }
823
824
825 static INLINE void
826 img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
827 const float s[QUAD_SIZE],
828 const float t[QUAD_SIZE],
829 const float p[QUAD_SIZE],
830 float lodbias,
831 float rgba[NUM_CHANNELS][QUAD_SIZE])
832 {
833 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
834 unsigned j;
835 unsigned level = samp->level;
836 unsigned xpot = pot_level_size(samp->xpot, level);
837 unsigned ypot = pot_level_size(samp->ypot, level);
838 union tex_tile_address addr;
839
840 addr.value = 0;
841 addr.bits.level = samp->level;
842
843 for (j = 0; j < QUAD_SIZE; j++) {
844 int c;
845
846 float u = s[j] * xpot;
847 float v = t[j] * ypot;
848
849 int uflr = util_ifloor(u);
850 int vflr = util_ifloor(v);
851
852 int x0 = uflr & (xpot - 1);
853 int y0 = vflr & (ypot - 1);
854
855 const float *out = get_texel_2d_no_border(samp, addr, x0, y0);
856
857 for (c = 0; c < 4; c++) {
858 rgba[c][j] = out[c];
859 }
860 }
861 }
862
863
864 static INLINE void
865 img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
866 const float s[QUAD_SIZE],
867 const float t[QUAD_SIZE],
868 const float p[QUAD_SIZE],
869 float lodbias,
870 float rgba[NUM_CHANNELS][QUAD_SIZE])
871 {
872 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
873 unsigned j;
874 unsigned level = samp->level;
875 unsigned xpot = pot_level_size(samp->xpot, level);
876 unsigned ypot = pot_level_size(samp->ypot, level);
877 union tex_tile_address addr;
878
879 addr.value = 0;
880 addr.bits.level = samp->level;
881
882 for (j = 0; j < QUAD_SIZE; j++) {
883 int c;
884
885 float u = s[j] * xpot;
886 float v = t[j] * ypot;
887
888 int x0, y0;
889 const float *out;
890
891 x0 = util_ifloor(u);
892 if (x0 < 0)
893 x0 = 0;
894 else if (x0 > xpot - 1)
895 x0 = xpot - 1;
896
897 y0 = util_ifloor(v);
898 if (y0 < 0)
899 y0 = 0;
900 else if (y0 > ypot - 1)
901 y0 = ypot - 1;
902
903 out = get_texel_2d_no_border(samp, addr, x0, y0);
904
905 for (c = 0; c < 4; c++) {
906 rgba[c][j] = out[c];
907 }
908 }
909 }
910
911
912 static void
913 img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
914 const float s[QUAD_SIZE],
915 const float t[QUAD_SIZE],
916 const float p[QUAD_SIZE],
917 float lodbias,
918 float rgba[NUM_CHANNELS][QUAD_SIZE])
919 {
920 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
921 const struct pipe_texture *texture = samp->texture;
922 unsigned level0, j;
923 int width;
924 int x[4];
925 union tex_tile_address addr;
926
927 level0 = samp->level;
928 width = texture->width[level0];
929
930 assert(width > 0);
931
932 addr.value = 0;
933 addr.bits.level = samp->level;
934
935 samp->nearest_texcoord_s(s, width, x);
936
937 for (j = 0; j < QUAD_SIZE; j++) {
938 const float *out = get_texel_2d(samp, addr, x[j], 0);
939 int c;
940 for (c = 0; c < 4; c++) {
941 rgba[c][j] = out[c];
942 }
943 }
944 }
945
946
947 static void
948 img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
949 const float s[QUAD_SIZE],
950 const float t[QUAD_SIZE],
951 const float p[QUAD_SIZE],
952 float lodbias,
953 float rgba[NUM_CHANNELS][QUAD_SIZE])
954 {
955 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
956 const struct pipe_texture *texture = samp->texture;
957 unsigned level0, j;
958 int width, height;
959 int x[4], y[4];
960 union tex_tile_address addr;
961
962
963 level0 = samp->level;
964 width = texture->width[level0];
965 height = texture->height[level0];
966
967 assert(width > 0);
968 assert(height > 0);
969
970 addr.value = 0;
971 addr.bits.level = samp->level;
972
973 samp->nearest_texcoord_s(s, width, x);
974 samp->nearest_texcoord_t(t, height, y);
975
976 for (j = 0; j < QUAD_SIZE; j++) {
977 const float *out = get_texel_2d(samp, addr, x[j], y[j]);
978 int c;
979 for (c = 0; c < 4; c++) {
980 rgba[c][j] = out[c];
981 }
982 }
983 }
984
985
986 static INLINE union tex_tile_address
987 face(union tex_tile_address addr, unsigned face )
988 {
989 addr.bits.face = face;
990 return addr;
991 }
992
993
994 static void
995 img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
996 const float s[QUAD_SIZE],
997 const float t[QUAD_SIZE],
998 const float p[QUAD_SIZE],
999 float lodbias,
1000 float rgba[NUM_CHANNELS][QUAD_SIZE])
1001 {
1002 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1003 const struct pipe_texture *texture = samp->texture;
1004 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
1005 unsigned level0, j;
1006 int width, height;
1007 int x[4], y[4];
1008 union tex_tile_address addr;
1009
1010 level0 = samp->level;
1011 width = texture->width[level0];
1012 height = texture->height[level0];
1013
1014 assert(width > 0);
1015 assert(height > 0);
1016
1017 addr.value = 0;
1018 addr.bits.level = samp->level;
1019
1020 samp->nearest_texcoord_s(s, width, x);
1021 samp->nearest_texcoord_t(t, height, y);
1022
1023 for (j = 0; j < QUAD_SIZE; j++) {
1024 const float *out = get_texel_2d(samp, face(addr, faces[j]), x[j], y[j]);
1025 int c;
1026 for (c = 0; c < 4; c++) {
1027 rgba[c][j] = out[c];
1028 }
1029 }
1030 }
1031
1032
1033 static void
1034 img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
1035 const float s[QUAD_SIZE],
1036 const float t[QUAD_SIZE],
1037 const float p[QUAD_SIZE],
1038 float lodbias,
1039 float rgba[NUM_CHANNELS][QUAD_SIZE])
1040 {
1041 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1042 const struct pipe_texture *texture = samp->texture;
1043 unsigned level0, j;
1044 int width, height, depth;
1045 int x[4], y[4], z[4];
1046 union tex_tile_address addr;
1047
1048 level0 = samp->level;
1049 width = texture->width[level0];
1050 height = texture->height[level0];
1051 depth = texture->depth[level0];
1052
1053 assert(width > 0);
1054 assert(height > 0);
1055 assert(depth > 0);
1056
1057 samp->nearest_texcoord_s(s, width, x);
1058 samp->nearest_texcoord_t(t, height, y);
1059 samp->nearest_texcoord_p(p, depth, z);
1060
1061 addr.value = 0;
1062 addr.bits.level = samp->level;
1063
1064 for (j = 0; j < QUAD_SIZE; j++) {
1065 const float *out = get_texel_3d(samp, addr, x[j], y[j], z[j]);
1066 int c;
1067 for (c = 0; c < 4; c++) {
1068 rgba[c][j] = out[c];
1069 }
1070 }
1071 }
1072
1073
1074 static void
1075 img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
1076 const float s[QUAD_SIZE],
1077 const float t[QUAD_SIZE],
1078 const float p[QUAD_SIZE],
1079 float lodbias,
1080 float rgba[NUM_CHANNELS][QUAD_SIZE])
1081 {
1082 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1083 const struct pipe_texture *texture = samp->texture;
1084 unsigned level0, j;
1085 int width;
1086 int x0[4], x1[4];
1087 float xw[4]; /* weights */
1088 union tex_tile_address addr;
1089
1090 level0 = samp->level;
1091 width = texture->width[level0];
1092
1093 assert(width > 0);
1094
1095 addr.value = 0;
1096 addr.bits.level = samp->level;
1097
1098 samp->linear_texcoord_s(s, width, x0, x1, xw);
1099
1100 for (j = 0; j < QUAD_SIZE; j++) {
1101 const float *tx0 = get_texel_2d(samp, addr, x0[j], 0);
1102 const float *tx1 = get_texel_2d(samp, addr, x1[j], 0);
1103 int c;
1104
1105 /* interpolate R, G, B, A */
1106 for (c = 0; c < 4; c++) {
1107 rgba[c][j] = lerp(xw[j], tx0[c], tx1[c]);
1108 }
1109 }
1110 }
1111
1112
1113 static void
1114 img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
1115 const float s[QUAD_SIZE],
1116 const float t[QUAD_SIZE],
1117 const float p[QUAD_SIZE],
1118 float lodbias,
1119 float rgba[NUM_CHANNELS][QUAD_SIZE])
1120 {
1121 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1122 const struct pipe_texture *texture = samp->texture;
1123 unsigned level0, j;
1124 int width, height;
1125 int x0[4], y0[4], x1[4], y1[4];
1126 float xw[4], yw[4]; /* weights */
1127 union tex_tile_address addr;
1128
1129 level0 = samp->level;
1130 width = texture->width[level0];
1131 height = texture->height[level0];
1132
1133 assert(width > 0);
1134 assert(height > 0);
1135
1136 addr.value = 0;
1137 addr.bits.level = samp->level;
1138
1139 samp->linear_texcoord_s(s, width, x0, x1, xw);
1140 samp->linear_texcoord_t(t, height, y0, y1, yw);
1141
1142 for (j = 0; j < QUAD_SIZE; j++) {
1143 const float *tx0 = get_texel_2d(samp, addr, x0[j], y0[j]);
1144 const float *tx1 = get_texel_2d(samp, addr, x1[j], y0[j]);
1145 const float *tx2 = get_texel_2d(samp, addr, x0[j], y1[j]);
1146 const float *tx3 = get_texel_2d(samp, addr, x1[j], y1[j]);
1147 int c;
1148
1149 /* interpolate R, G, B, A */
1150 for (c = 0; c < 4; c++) {
1151 rgba[c][j] = lerp_2d(xw[j], yw[j],
1152 tx0[c], tx1[c],
1153 tx2[c], tx3[c]);
1154 }
1155 }
1156 }
1157
1158
1159 static void
1160 img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler,
1161 const float s[QUAD_SIZE],
1162 const float t[QUAD_SIZE],
1163 const float p[QUAD_SIZE],
1164 float lodbias,
1165 float rgba[NUM_CHANNELS][QUAD_SIZE])
1166 {
1167 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1168 const struct pipe_texture *texture = samp->texture;
1169 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
1170 unsigned level0, j;
1171 int width, height;
1172 int x0[4], y0[4], x1[4], y1[4];
1173 float xw[4], yw[4]; /* weights */
1174 union tex_tile_address addr;
1175
1176 level0 = samp->level;
1177 width = texture->width[level0];
1178 height = texture->height[level0];
1179
1180 assert(width > 0);
1181 assert(height > 0);
1182
1183 addr.value = 0;
1184 addr.bits.level = samp->level;
1185
1186 samp->linear_texcoord_s(s, width, x0, x1, xw);
1187 samp->linear_texcoord_t(t, height, y0, y1, yw);
1188
1189 for (j = 0; j < QUAD_SIZE; j++) {
1190 union tex_tile_address addrj = face(addr, faces[j]);
1191 const float *tx0 = get_texel_2d(samp, addrj, x0[j], y0[j]);
1192 const float *tx1 = get_texel_2d(samp, addrj, x1[j], y0[j]);
1193 const float *tx2 = get_texel_2d(samp, addrj, x0[j], y1[j]);
1194 const float *tx3 = get_texel_2d(samp, addrj, x1[j], y1[j]);
1195 int c;
1196
1197 /* interpolate R, G, B, A */
1198 for (c = 0; c < 4; c++) {
1199 rgba[c][j] = lerp_2d(xw[j], yw[j],
1200 tx0[c], tx1[c],
1201 tx2[c], tx3[c]);
1202 }
1203 }
1204 }
1205
1206
1207 static void
1208 img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
1209 const float s[QUAD_SIZE],
1210 const float t[QUAD_SIZE],
1211 const float p[QUAD_SIZE],
1212 float lodbias,
1213 float rgba[NUM_CHANNELS][QUAD_SIZE])
1214 {
1215 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1216 const struct pipe_texture *texture = samp->texture;
1217 unsigned level0, j;
1218 int width, height, depth;
1219 int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
1220 float xw[4], yw[4], zw[4]; /* interpolation weights */
1221 union tex_tile_address addr;
1222
1223 level0 = samp->level;
1224 width = texture->width[level0];
1225 height = texture->height[level0];
1226 depth = texture->depth[level0];
1227
1228 addr.value = 0;
1229 addr.bits.level = level0;
1230
1231 assert(width > 0);
1232 assert(height > 0);
1233 assert(depth > 0);
1234
1235 samp->linear_texcoord_s(s, width, x0, x1, xw);
1236 samp->linear_texcoord_t(t, height, y0, y1, yw);
1237 samp->linear_texcoord_p(p, depth, z0, z1, zw);
1238
1239 for (j = 0; j < QUAD_SIZE; j++) {
1240 int c;
1241
1242 const float *tx00 = get_texel_3d(samp, addr, x0[j], y0[j], z0[j]);
1243 const float *tx01 = get_texel_3d(samp, addr, x1[j], y0[j], z0[j]);
1244 const float *tx02 = get_texel_3d(samp, addr, x0[j], y1[j], z0[j]);
1245 const float *tx03 = get_texel_3d(samp, addr, x1[j], y1[j], z0[j]);
1246
1247 const float *tx10 = get_texel_3d(samp, addr, x0[j], y0[j], z1[j]);
1248 const float *tx11 = get_texel_3d(samp, addr, x1[j], y0[j], z1[j]);
1249 const float *tx12 = get_texel_3d(samp, addr, x0[j], y1[j], z1[j]);
1250 const float *tx13 = get_texel_3d(samp, addr, x1[j], y1[j], z1[j]);
1251
1252 /* interpolate R, G, B, A */
1253 for (c = 0; c < 4; c++) {
1254 rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1255 tx00[c], tx01[c],
1256 tx02[c], tx03[c],
1257 tx10[c], tx11[c],
1258 tx12[c], tx13[c]);
1259 }
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 float lodbias,
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
1277 lambda = samp->compute_lambda(samp, s, t, p, lodbias);
1278 level0 = (int)lambda;
1279
1280 if (lambda < 0.0) {
1281 samp->level = 0;
1282 samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
1283 }
1284 else if (level0 >= texture->last_level) {
1285 samp->level = texture->last_level;
1286 samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
1287 }
1288 else {
1289 float levelBlend = lambda - level0;
1290 float rgba0[4][4];
1291 float rgba1[4][4];
1292 int c,j;
1293
1294 samp->level = level0;
1295 samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba0 );
1296
1297 samp->level = level0+1;
1298 samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba1 );
1299
1300 for (j = 0; j < QUAD_SIZE; j++) {
1301 for (c = 0; c < 4; c++) {
1302 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1303 }
1304 }
1305 }
1306 }
1307
1308
1309 static void
1310 mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
1311 const float s[QUAD_SIZE],
1312 const float t[QUAD_SIZE],
1313 const float p[QUAD_SIZE],
1314 float lodbias,
1315 float rgba[NUM_CHANNELS][QUAD_SIZE])
1316 {
1317 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1318 const struct pipe_texture *texture = samp->texture;
1319 float lambda;
1320
1321 lambda = samp->compute_lambda(samp, s, t, p, lodbias);
1322
1323 if (lambda < 0.0) {
1324 samp->level = 0;
1325 samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
1326 }
1327 else {
1328 samp->level = (int)(lambda + 0.5) ;
1329 samp->level = MIN2(samp->level, (int)texture->last_level);
1330 samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
1331 }
1332
1333 #if 0
1334 printf("RGBA %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
1335 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
1336 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
1337 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
1338 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
1339 #endif
1340 }
1341
1342
1343 static void
1344 mip_filter_none(struct tgsi_sampler *tgsi_sampler,
1345 const float s[QUAD_SIZE],
1346 const float t[QUAD_SIZE],
1347 const float p[QUAD_SIZE],
1348 float lodbias,
1349 float rgba[NUM_CHANNELS][QUAD_SIZE])
1350 {
1351 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1352 float lambda = samp->compute_lambda(samp, s, t, p, lodbias);
1353
1354 if (lambda < 0.0) {
1355 samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
1356 }
1357 else {
1358 samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
1359 }
1360 }
1361
1362
1363
1364 /**
1365 * Specialized version of mip_filter_linear with hard-wired calls to
1366 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
1367 */
1368 static void
1369 mip_filter_linear_2d_linear_repeat_POT(
1370 struct tgsi_sampler *tgsi_sampler,
1371 const float s[QUAD_SIZE],
1372 const float t[QUAD_SIZE],
1373 const float p[QUAD_SIZE],
1374 float lodbias,
1375 float rgba[NUM_CHANNELS][QUAD_SIZE])
1376 {
1377 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1378 const struct pipe_texture *texture = samp->texture;
1379 int level0;
1380 float lambda;
1381
1382 lambda = compute_lambda_2d(samp, s, t, p, lodbias);
1383 level0 = (int)lambda;
1384
1385 /* Catches both negative and large values of level0:
1386 */
1387 if ((unsigned)level0 >= texture->last_level) {
1388 if (level0 < 0)
1389 samp->level = 0;
1390 else
1391 samp->level = texture->last_level;
1392
1393 img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba );
1394 }
1395 else {
1396 float levelBlend = lambda - level0;
1397 float rgba0[4][4];
1398 float rgba1[4][4];
1399 int c,j;
1400
1401 samp->level = level0;
1402 img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba0 );
1403
1404 samp->level = level0+1;
1405 img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba1 );
1406
1407 for (j = 0; j < QUAD_SIZE; j++) {
1408 for (c = 0; c < 4; c++) {
1409 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1410 }
1411 }
1412 }
1413 }
1414
1415
1416
1417 /**
1418 * Do shadow/depth comparisons.
1419 */
1420 static void
1421 sample_compare(struct tgsi_sampler *tgsi_sampler,
1422 const float s[QUAD_SIZE],
1423 const float t[QUAD_SIZE],
1424 const float p[QUAD_SIZE],
1425 float lodbias,
1426 float rgba[NUM_CHANNELS][QUAD_SIZE])
1427 {
1428 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1429 const struct pipe_sampler_state *sampler = samp->sampler;
1430 int j, k0, k1, k2, k3;
1431 float val;
1432
1433 samp->mip_filter( tgsi_sampler, s, t, p, lodbias, rgba );
1434
1435 /**
1436 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
1437 * When we sampled the depth texture, the depth value was put into all
1438 * RGBA channels. We look at the red channel here.
1439 */
1440
1441 /* compare four texcoords vs. four texture samples */
1442 switch (sampler->compare_func) {
1443 case PIPE_FUNC_LESS:
1444 k0 = p[0] < rgba[0][0];
1445 k1 = p[1] < rgba[0][1];
1446 k2 = p[2] < rgba[0][2];
1447 k3 = p[3] < rgba[0][3];
1448 break;
1449 case PIPE_FUNC_LEQUAL:
1450 k0 = p[0] <= rgba[0][0];
1451 k1 = p[1] <= rgba[0][1];
1452 k2 = p[2] <= rgba[0][2];
1453 k3 = p[3] <= rgba[0][3];
1454 break;
1455 case PIPE_FUNC_GREATER:
1456 k0 = p[0] > rgba[0][0];
1457 k1 = p[1] > rgba[0][1];
1458 k2 = p[2] > rgba[0][2];
1459 k3 = p[3] > rgba[0][3];
1460 break;
1461 case PIPE_FUNC_GEQUAL:
1462 k0 = p[0] >= rgba[0][0];
1463 k1 = p[1] >= rgba[0][1];
1464 k2 = p[2] >= rgba[0][2];
1465 k3 = p[3] >= rgba[0][3];
1466 break;
1467 case PIPE_FUNC_EQUAL:
1468 k0 = p[0] == rgba[0][0];
1469 k1 = p[1] == rgba[0][1];
1470 k2 = p[2] == rgba[0][2];
1471 k3 = p[3] == rgba[0][3];
1472 break;
1473 case PIPE_FUNC_NOTEQUAL:
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_ALWAYS:
1480 k0 = k1 = k2 = k3 = 1;
1481 break;
1482 case PIPE_FUNC_NEVER:
1483 k0 = k1 = k2 = k3 = 0;
1484 break;
1485 default:
1486 k0 = k1 = k2 = k3 = 0;
1487 assert(0);
1488 break;
1489 }
1490
1491 /* convert four pass/fail values to an intensity in [0,1] */
1492 val = 0.25F * (k0 + k1 + k2 + k3);
1493
1494 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
1495 for (j = 0; j < 4; j++) {
1496 rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
1497 rgba[3][j] = 1.0F;
1498 }
1499 }
1500
1501
1502 /**
1503 * Compute which cube face is referenced by each texcoord and put that
1504 * info into the sampler faces[] array. Then sample the cube faces
1505 */
1506 static void
1507 sample_cube(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 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1515 unsigned j;
1516 float ssss[4], tttt[4];
1517
1518 /*
1519 major axis
1520 direction target sc tc ma
1521 ---------- ------------------------------- --- --- ---
1522 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
1523 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
1524 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
1525 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
1526 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
1527 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
1528 */
1529 for (j = 0; j < QUAD_SIZE; j++) {
1530 float rx = s[j];
1531 float ry = t[j];
1532 float rz = p[j];
1533 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
1534 unsigned face;
1535 float sc, tc, ma;
1536
1537 if (arx >= ary && arx >= arz) {
1538 if (rx >= 0.0F) {
1539 face = PIPE_TEX_FACE_POS_X;
1540 sc = -rz;
1541 tc = -ry;
1542 ma = arx;
1543 }
1544 else {
1545 face = PIPE_TEX_FACE_NEG_X;
1546 sc = rz;
1547 tc = -ry;
1548 ma = arx;
1549 }
1550 }
1551 else if (ary >= arx && ary >= arz) {
1552 if (ry >= 0.0F) {
1553 face = PIPE_TEX_FACE_POS_Y;
1554 sc = rx;
1555 tc = rz;
1556 ma = ary;
1557 }
1558 else {
1559 face = PIPE_TEX_FACE_NEG_Y;
1560 sc = rx;
1561 tc = -rz;
1562 ma = ary;
1563 }
1564 }
1565 else {
1566 if (rz > 0.0F) {
1567 face = PIPE_TEX_FACE_POS_Z;
1568 sc = rx;
1569 tc = -ry;
1570 ma = arz;
1571 }
1572 else {
1573 face = PIPE_TEX_FACE_NEG_Z;
1574 sc = -rx;
1575 tc = -ry;
1576 ma = arz;
1577 }
1578 }
1579
1580 {
1581 const float ima = 1.0 / ma;
1582 ssss[j] = ( sc * ima + 1.0F ) * 0.5F;
1583 tttt[j] = ( tc * ima + 1.0F ) * 0.5F;
1584 samp->faces[j] = face;
1585 }
1586 }
1587
1588 /* In our little pipeline, the compare stage is next. If compare
1589 * is not active, this will point somewhere deeper into the
1590 * pipeline, eg. to mip_filter or even img_filter.
1591 */
1592 samp->compare(tgsi_sampler, ssss, tttt, NULL, lodbias, rgba);
1593 }
1594
1595
1596
1597 static wrap_nearest_func
1598 get_nearest_unorm_wrap(unsigned mode)
1599 {
1600 switch (mode) {
1601 case PIPE_TEX_WRAP_CLAMP:
1602 return wrap_nearest_unorm_clamp;
1603 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1604 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1605 return wrap_nearest_unorm_clamp_to_border;
1606 default:
1607 assert(0);
1608 return wrap_nearest_unorm_clamp;
1609 }
1610 }
1611
1612
1613 static wrap_nearest_func
1614 get_nearest_wrap(unsigned mode)
1615 {
1616 switch (mode) {
1617 case PIPE_TEX_WRAP_REPEAT:
1618 return wrap_nearest_repeat;
1619 case PIPE_TEX_WRAP_CLAMP:
1620 return wrap_nearest_clamp;
1621 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1622 return wrap_nearest_clamp_to_edge;
1623 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1624 return wrap_nearest_clamp_to_border;
1625 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1626 return wrap_nearest_mirror_repeat;
1627 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1628 return wrap_nearest_mirror_clamp;
1629 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1630 return wrap_nearest_mirror_clamp_to_edge;
1631 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1632 return wrap_nearest_mirror_clamp_to_border;
1633 default:
1634 assert(0);
1635 return wrap_nearest_repeat;
1636 }
1637 }
1638
1639
1640 static wrap_linear_func
1641 get_linear_unorm_wrap(unsigned mode)
1642 {
1643 switch (mode) {
1644 case PIPE_TEX_WRAP_CLAMP:
1645 return wrap_linear_unorm_clamp;
1646 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1647 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1648 return wrap_linear_unorm_clamp_to_border;
1649 default:
1650 assert(0);
1651 return wrap_linear_unorm_clamp;
1652 }
1653 }
1654
1655
1656 static wrap_linear_func
1657 get_linear_wrap(unsigned mode)
1658 {
1659 switch (mode) {
1660 case PIPE_TEX_WRAP_REPEAT:
1661 return wrap_linear_repeat;
1662 case PIPE_TEX_WRAP_CLAMP:
1663 return wrap_linear_clamp;
1664 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1665 return wrap_linear_clamp_to_edge;
1666 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1667 return wrap_linear_clamp_to_border;
1668 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1669 return wrap_linear_mirror_repeat;
1670 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1671 return wrap_linear_mirror_clamp;
1672 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1673 return wrap_linear_mirror_clamp_to_edge;
1674 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1675 return wrap_linear_mirror_clamp_to_border;
1676 default:
1677 assert(0);
1678 return wrap_linear_repeat;
1679 }
1680 }
1681
1682
1683 static compute_lambda_func
1684 get_lambda_func(const union sp_sampler_key key)
1685 {
1686 if (key.bits.processor == TGSI_PROCESSOR_VERTEX)
1687 return compute_lambda_vert;
1688
1689 switch (key.bits.target) {
1690 case PIPE_TEXTURE_1D:
1691 return compute_lambda_1d;
1692 case PIPE_TEXTURE_2D:
1693 case PIPE_TEXTURE_CUBE:
1694 return compute_lambda_2d;
1695 case PIPE_TEXTURE_3D:
1696 return compute_lambda_3d;
1697 default:
1698 assert(0);
1699 return compute_lambda_1d;
1700 }
1701 }
1702
1703
1704 static filter_func
1705 get_img_filter(const union sp_sampler_key key,
1706 unsigned filter,
1707 const struct pipe_sampler_state *sampler)
1708 {
1709 switch (key.bits.target) {
1710 case PIPE_TEXTURE_1D:
1711 if (filter == PIPE_TEX_FILTER_NEAREST)
1712 return img_filter_1d_nearest;
1713 else
1714 return img_filter_1d_linear;
1715 break;
1716 case PIPE_TEXTURE_2D:
1717 /* Try for fast path:
1718 */
1719 if (key.bits.is_pot &&
1720 sampler->wrap_s == sampler->wrap_t &&
1721 sampler->normalized_coords)
1722 {
1723 switch (sampler->wrap_s) {
1724 case PIPE_TEX_WRAP_REPEAT:
1725 switch (filter) {
1726 case PIPE_TEX_FILTER_NEAREST:
1727 return img_filter_2d_nearest_repeat_POT;
1728 case PIPE_TEX_FILTER_LINEAR:
1729 return img_filter_2d_linear_repeat_POT;
1730 default:
1731 break;
1732 }
1733 break;
1734 case PIPE_TEX_WRAP_CLAMP:
1735 switch (filter) {
1736 case PIPE_TEX_FILTER_NEAREST:
1737 return img_filter_2d_nearest_clamp_POT;
1738 default:
1739 break;
1740 }
1741 }
1742 }
1743 /* Otherwise use default versions:
1744 */
1745 if (filter == PIPE_TEX_FILTER_NEAREST)
1746 return img_filter_2d_nearest;
1747 else
1748 return img_filter_2d_linear;
1749 break;
1750 case PIPE_TEXTURE_CUBE:
1751 if (filter == PIPE_TEX_FILTER_NEAREST)
1752 return img_filter_cube_nearest;
1753 else
1754 return img_filter_cube_linear;
1755 break;
1756 case PIPE_TEXTURE_3D:
1757 if (filter == PIPE_TEX_FILTER_NEAREST)
1758 return img_filter_3d_nearest;
1759 else
1760 return img_filter_3d_linear;
1761 break;
1762 default:
1763 assert(0);
1764 return img_filter_1d_nearest;
1765 }
1766 }
1767
1768
1769 /**
1770 * Bind the given texture object and texture cache to the sampler varient.
1771 */
1772 void
1773 sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp,
1774 struct softpipe_tex_tile_cache *tex_cache,
1775 const struct pipe_texture *texture )
1776 {
1777 const struct pipe_sampler_state *sampler = samp->sampler;
1778
1779 samp->texture = texture;
1780 samp->cache = tex_cache;
1781 samp->xpot = util_unsigned_logbase2( texture->width[0] );
1782 samp->ypot = util_unsigned_logbase2( texture->height[0] );
1783 samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level);
1784 }
1785
1786
1787 void
1788 sp_sampler_varient_destroy( struct sp_sampler_varient *samp )
1789 {
1790 FREE(samp);
1791 }
1792
1793
1794 /**
1795 * Create a sampler varient for a given set of non-orthogonal state.
1796 */
1797 struct sp_sampler_varient *
1798 sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
1799 const union sp_sampler_key key )
1800 {
1801 struct sp_sampler_varient *samp = CALLOC_STRUCT(sp_sampler_varient);
1802 if (!samp)
1803 return NULL;
1804
1805 samp->sampler = sampler;
1806 samp->key = key;
1807
1808 /* Note that (for instance) linear_texcoord_s and
1809 * nearest_texcoord_s may be active at the same time, if the
1810 * sampler min_img_filter differs from its mag_img_filter.
1811 */
1812 if (sampler->normalized_coords) {
1813 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
1814 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
1815 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
1816
1817 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
1818 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
1819 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
1820 }
1821 else {
1822 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
1823 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
1824 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
1825
1826 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
1827 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
1828 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
1829 }
1830
1831 samp->compute_lambda = get_lambda_func( key );
1832
1833 samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
1834 samp->mag_img_filter = get_img_filter(key, sampler->mag_img_filter, sampler);
1835
1836 switch (sampler->min_mip_filter) {
1837 case PIPE_TEX_MIPFILTER_NONE:
1838 if (sampler->min_img_filter == sampler->mag_img_filter)
1839 samp->mip_filter = samp->min_img_filter;
1840 else
1841 samp->mip_filter = mip_filter_none;
1842 break;
1843
1844 case PIPE_TEX_MIPFILTER_NEAREST:
1845 samp->mip_filter = mip_filter_nearest;
1846 break;
1847
1848 case PIPE_TEX_MIPFILTER_LINEAR:
1849 if (key.bits.is_pot &&
1850 sampler->min_img_filter == sampler->mag_img_filter &&
1851 sampler->normalized_coords &&
1852 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
1853 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
1854 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR)
1855 {
1856 samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT;
1857 }
1858 else
1859 {
1860 samp->mip_filter = mip_filter_linear;
1861 }
1862 break;
1863 }
1864
1865 if (sampler->compare_mode != FALSE) {
1866 samp->compare = sample_compare;
1867 }
1868 else {
1869 /* Skip compare operation by promoting the mip_filter function
1870 * pointer:
1871 */
1872 samp->compare = samp->mip_filter;
1873 }
1874
1875 if (key.bits.target == PIPE_TEXTURE_CUBE) {
1876 samp->base.get_samples = sample_cube;
1877 }
1878 else {
1879 samp->faces[0] = 0;
1880 samp->faces[1] = 0;
1881 samp->faces[2] = 0;
1882 samp->faces[3] = 0;
1883
1884 /* Skip cube face determination by promoting the compare
1885 * function pointer:
1886 */
1887 samp->base.get_samples = samp->compare;
1888 }
1889
1890 return samp;
1891 }