projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
softpipe: get rid of tgsi_sampler_control param in img_filter
[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_format.h"
42 #include "util/u_memory.h"
43 #include "sp_quad.h" /* only for #define QUAD_* tokens */
44 #include "sp_tex_sample.h"
45 #include "sp_tex_tile_cache.h"
46
47
48 /** Set to one to help debug texture sampling */
49 #define DEBUG_TEX 0
50
51
52 /*
53 * Return fractional part of 'f'. Used for computing interpolation weights.
54 * Need to be careful with negative values.
55 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
56 * of improperly weighted linear-filtered textures.
57 * The tests/texwrap.c demo is a good test.
58 */
59 static INLINE float
60 frac(float f)
61 {
62 return f - floorf(f);
63 }
64
65
66
67 /**
68 * Linear interpolation macro
69 */
70 static INLINE float
71 lerp(float a, float v0, float v1)
72 {
73 return v0 + a * (v1 - v0);
74 }
75
76
77 /**
78 * Do 2D/bilinear interpolation of float values.
79 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
80 * a and b are the horizontal and vertical interpolants.
81 * It's important that this function is inlined when compiled with
82 * optimization! If we find that's not true on some systems, convert
83 * to a macro.
84 */
85 static INLINE float
86 lerp_2d(float a, float b,
87 float v00, float v10, float v01, float v11)
88 {
89 const float temp0 = lerp(a, v00, v10);
90 const float temp1 = lerp(a, v01, v11);
91 return lerp(b, temp0, temp1);
92 }
93
94
95 /**
96 * As above, but 3D interpolation of 8 values.
97 */
98 static INLINE float
99 lerp_3d(float a, float b, float c,
100 float v000, float v100, float v010, float v110,
101 float v001, float v101, float v011, float v111)
102 {
103 const float temp0 = lerp_2d(a, b, v000, v100, v010, v110);
104 const float temp1 = lerp_2d(a, b, v001, v101, v011, v111);
105 return lerp(c, temp0, temp1);
106 }
107
108
109
110 /**
111 * Compute coord % size for repeat wrap modes.
112 * Note that if coord is negative, coord % size doesn't give the right
113 * value. To avoid that problem we add a large multiple of the size
114 * (rather than using a conditional).
115 */
116 static INLINE int
117 repeat(int coord, unsigned size)
118 {
119 return (coord + size * 1024) % size;
120 }
121
122
123 /**
124 * Apply texture coord wrapping mode and return integer texture indexes
125 * for a vector of four texcoords (S or T or P).
126 * \param wrapMode PIPE_TEX_WRAP_x
127 * \param s the incoming texcoords
128 * \param size the texture image size
129 * \param icoord returns the integer texcoords
130 * \return integer texture index
131 */
132 static void
133 wrap_nearest_repeat(float s, unsigned size, int *icoord)
134 {
135 /* s limited to [0,1) */
136 /* i limited to [0,size-1] */
137 int i = util_ifloor(s * size);
138 *icoord = repeat(i, size);
139 }
140
141
142 static void
143 wrap_nearest_clamp(float s, unsigned size, int *icoord)
144 {
145 /* s limited to [0,1] */
146 /* i limited to [0,size-1] */
147 if (s <= 0.0F)
148 *icoord = 0;
149 else if (s >= 1.0F)
150 *icoord = size - 1;
151 else
152 *icoord = util_ifloor(s * size);
153 }
154
155
156 static void
157 wrap_nearest_clamp_to_edge(float s, unsigned size, int *icoord)
158 {
159 /* s limited to [min,max] */
160 /* i limited to [0, size-1] */
161 const float min = 1.0F / (2.0F * size);
162 const float max = 1.0F - min;
163 if (s < min)
164 *icoord = 0;
165 else if (s > max)
166 *icoord = size - 1;
167 else
168 *icoord = util_ifloor(s * size);
169 }
170
171
172 static void
173 wrap_nearest_clamp_to_border(float s, unsigned size, int *icoord)
174 {
175 /* s limited to [min,max] */
176 /* i limited to [-1, size] */
177 const float min = -1.0F / (2.0F * size);
178 const float max = 1.0F - min;
179 if (s <= min)
180 *icoord = -1;
181 else if (s >= max)
182 *icoord = size;
183 else
184 *icoord = util_ifloor(s * size);
185 }
186
187
188 static void
189 wrap_nearest_mirror_repeat(float s, unsigned size, int *icoord)
190 {
191 const float min = 1.0F / (2.0F * size);
192 const float max = 1.0F - min;
193 const int flr = util_ifloor(s);
194 float u = frac(s);
195 if (flr & 1)
196 u = 1.0F - u;
197 if (u < min)
198 *icoord = 0;
199 else if (u > max)
200 *icoord = size - 1;
201 else
202 *icoord = util_ifloor(u * size);
203 }
204
205
206 static void
207 wrap_nearest_mirror_clamp(float s, unsigned size, int *icoord)
208 {
209 /* s limited to [0,1] */
210 /* i limited to [0,size-1] */
211 const float u = fabsf(s);
212 if (u <= 0.0F)
213 *icoord = 0;
214 else if (u >= 1.0F)
215 *icoord = size - 1;
216 else
217 *icoord = util_ifloor(u * size);
218 }
219
220
221 static void
222 wrap_nearest_mirror_clamp_to_edge(float s, unsigned size, int *icoord)
223 {
224 /* s limited to [min,max] */
225 /* i limited to [0, size-1] */
226 const float min = 1.0F / (2.0F * size);
227 const float max = 1.0F - min;
228 const float u = fabsf(s);
229 if (u < min)
230 *icoord = 0;
231 else if (u > max)
232 *icoord = size - 1;
233 else
234 *icoord = util_ifloor(u * size);
235 }
236
237
238 static void
239 wrap_nearest_mirror_clamp_to_border(float s, unsigned size, int *icoord)
240 {
241 /* s limited to [min,max] */
242 /* i limited to [0, size-1] */
243 const float min = -1.0F / (2.0F * size);
244 const float max = 1.0F - min;
245 const float u = fabsf(s);
246 if (u < min)
247 *icoord = -1;
248 else if (u > max)
249 *icoord = size;
250 else
251 *icoord = util_ifloor(u * size);
252 }
253
254
255 /**
256 * Used to compute texel locations for linear sampling
257 * \param wrapMode PIPE_TEX_WRAP_x
258 * \param s the texcoord
259 * \param size the texture image size
260 * \param icoord0 returns first texture index
261 * \param icoord1 returns second texture index (usually icoord0 + 1)
262 * \param w returns blend factor/weight between texture indices
263 * \param icoord returns the computed integer texture coord
264 */
265 static void
266 wrap_linear_repeat(float s, unsigned size,
267 int *icoord0, int *icoord1, float *w)
268 {
269 float u = s * size - 0.5F;
270 *icoord0 = repeat(util_ifloor(u), size);
271 *icoord1 = repeat(*icoord0 + 1, size);
272 *w = frac(u);
273 }
274
275
276 static void
277 wrap_linear_clamp(float s, unsigned size,
278 int *icoord0, int *icoord1, float *w)
279 {
280 float u = CLAMP(s, 0.0F, 1.0F);
281 u = u * size - 0.5f;
282 *icoord0 = util_ifloor(u);
283 *icoord1 = *icoord0 + 1;
284 *w = frac(u);
285 }
286
287
288 static void
289 wrap_linear_clamp_to_edge(float s, unsigned size,
290 int *icoord0, int *icoord1, float *w)
291 {
292 float u = CLAMP(s, 0.0F, 1.0F);
293 u = u * size - 0.5f;
294 *icoord0 = util_ifloor(u);
295 *icoord1 = *icoord0 + 1;
296 if (*icoord0 < 0)
297 *icoord0 = 0;
298 if (*icoord1 >= (int) size)
299 *icoord1 = size - 1;
300 *w = frac(u);
301 }
302
303
304 static void
305 wrap_linear_clamp_to_border(float s, unsigned size,
306 int *icoord0, int *icoord1, float *w)
307 {
308 const float min = -1.0F / (2.0F * size);
309 const float max = 1.0F - min;
310 float u = CLAMP(s, min, max);
311 u = u * size - 0.5f;
312 *icoord0 = util_ifloor(u);
313 *icoord1 = *icoord0 + 1;
314 *w = frac(u);
315 }
316
317
318 static void
319 wrap_linear_mirror_repeat(float s, unsigned size,
320 int *icoord0, int *icoord1, float *w)
321 {
322 const int flr = util_ifloor(s);
323 float u = frac(s);
324 if (flr & 1)
325 u = 1.0F - u;
326 u = u * size - 0.5F;
327 *icoord0 = util_ifloor(u);
328 *icoord1 = *icoord0 + 1;
329 if (*icoord0 < 0)
330 *icoord0 = 0;
331 if (*icoord1 >= (int) size)
332 *icoord1 = size - 1;
333 *w = frac(u);
334 }
335
336
337 static void
338 wrap_linear_mirror_clamp(float s, unsigned size,
339 int *icoord0, int *icoord1, float *w)
340 {
341 float u = fabsf(s);
342 if (u >= 1.0F)
343 u = (float) size;
344 else
345 u *= size;
346 u -= 0.5F;
347 *icoord0 = util_ifloor(u);
348 *icoord1 = *icoord0 + 1;
349 *w = frac(u);
350 }
351
352
353 static void
354 wrap_linear_mirror_clamp_to_edge(float s, unsigned size,
355 int *icoord0, int *icoord1, float *w)
356 {
357 float u = fabsf(s);
358 if (u >= 1.0F)
359 u = (float) size;
360 else
361 u *= size;
362 u -= 0.5F;
363 *icoord0 = util_ifloor(u);
364 *icoord1 = *icoord0 + 1;
365 if (*icoord0 < 0)
366 *icoord0 = 0;
367 if (*icoord1 >= (int) size)
368 *icoord1 = size - 1;
369 *w = frac(u);
370 }
371
372
373 static void
374 wrap_linear_mirror_clamp_to_border(float s, unsigned size,
375 int *icoord0, int *icoord1, float *w)
376 {
377 const float min = -1.0F / (2.0F * size);
378 const float max = 1.0F - min;
379 float u = fabsf(s);
380 if (u <= min)
381 u = min * size;
382 else if (u >= max)
383 u = max * size;
384 else
385 u *= size;
386 u -= 0.5F;
387 *icoord0 = util_ifloor(u);
388 *icoord1 = *icoord0 + 1;
389 *w = frac(u);
390 }
391
392
393 /**
394 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
395 */
396 static void
397 wrap_nearest_unorm_clamp(float s, unsigned size, int *icoord)
398 {
399 int i = util_ifloor(s);
400 *icoord = CLAMP(i, 0, (int) size-1);
401 }
402
403
404 /**
405 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
406 */
407 static void
408 wrap_nearest_unorm_clamp_to_border(float s, unsigned size, int *icoord)
409 {
410 *icoord = util_ifloor( CLAMP(s, -0.5F, (float) size + 0.5F) );
411 }
412
413
414 /**
415 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
416 */
417 static void
418 wrap_nearest_unorm_clamp_to_edge(float s, unsigned size, int *icoord)
419 {
420 *icoord = util_ifloor( CLAMP(s, 0.5F, (float) size - 0.5F) );
421 }
422
423
424 /**
425 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
426 */
427 static void
428 wrap_linear_unorm_clamp(float s, unsigned size,
429 int *icoord0, int *icoord1, float *w)
430 {
431 /* Not exactly what the spec says, but it matches NVIDIA output */
432 float u = CLAMP(s - 0.5F, 0.0f, (float) size - 1.0f);
433 *icoord0 = util_ifloor(u);
434 *icoord1 = *icoord0 + 1;
435 *w = frac(u);
436 }
437
438
439 /**
440 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
441 */
442 static void
443 wrap_linear_unorm_clamp_to_border(float s, unsigned size,
444 int *icoord0, int *icoord1, float *w)
445 {
446 float u = CLAMP(s, -0.5F, (float) size + 0.5F);
447 u -= 0.5F;
448 *icoord0 = util_ifloor(u);
449 *icoord1 = *icoord0 + 1;
450 if (*icoord1 > (int) size - 1)
451 *icoord1 = size - 1;
452 *w = frac(u);
453 }
454
455
456 /**
457 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
458 */
459 static void
460 wrap_linear_unorm_clamp_to_edge(float s, unsigned size,
461 int *icoord0, int *icoord1, float *w)
462 {
463 float u = CLAMP(s, +0.5F, (float) size - 0.5F);
464 u -= 0.5F;
465 *icoord0 = util_ifloor(u);
466 *icoord1 = *icoord0 + 1;
467 if (*icoord1 > (int) size - 1)
468 *icoord1 = size - 1;
469 *w = frac(u);
470 }
471
472
473 /**
474 * Do coordinate to array index conversion. For array textures.
475 */
476 static INLINE void
477 wrap_array_layer(float coord, unsigned size, int *layer)
478 {
479 int c = util_ifloor(coord + 0.5F);
480 *layer = CLAMP(c, 0, size - 1);
481 }
482
483
484 /**
485 * Examine the quad's texture coordinates to compute the partial
486 * derivatives w.r.t X and Y, then compute lambda (level of detail).
487 */
488 static float
489 compute_lambda_1d(const struct sp_sampler_variant *samp,
490 const float s[TGSI_QUAD_SIZE],
491 const float t[TGSI_QUAD_SIZE],
492 const float p[TGSI_QUAD_SIZE])
493 {
494 const struct pipe_resource *texture = samp->view->texture;
495 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
496 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
497 float rho = MAX2(dsdx, dsdy) * u_minify(texture->width0, samp->view->u.tex.first_level);
498
499 return util_fast_log2(rho);
500 }
501
502
503 static float
504 compute_lambda_2d(const struct sp_sampler_variant *samp,
505 const float s[TGSI_QUAD_SIZE],
506 const float t[TGSI_QUAD_SIZE],
507 const float p[TGSI_QUAD_SIZE])
508 {
509 const struct pipe_resource *texture = samp->view->texture;
510 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
511 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
512 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
513 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
514 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, samp->view->u.tex.first_level);
515 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, samp->view->u.tex.first_level);
516 float rho = MAX2(maxx, maxy);
517
518 return util_fast_log2(rho);
519 }
520
521
522 static float
523 compute_lambda_3d(const struct sp_sampler_variant *samp,
524 const float s[TGSI_QUAD_SIZE],
525 const float t[TGSI_QUAD_SIZE],
526 const float p[TGSI_QUAD_SIZE])
527 {
528 const struct pipe_resource *texture = samp->view->texture;
529 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
530 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
531 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
532 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
533 float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
534 float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
535 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, samp->view->u.tex.first_level);
536 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, samp->view->u.tex.first_level);
537 float maxz = MAX2(dpdx, dpdy) * u_minify(texture->depth0, samp->view->u.tex.first_level);
538 float rho;
539
540 rho = MAX2(maxx, maxy);
541 rho = MAX2(rho, maxz);
542
543 return util_fast_log2(rho);
544 }
545
546
547 /**
548 * Compute lambda for a vertex texture sampler.
549 * Since there aren't derivatives to use, just return 0.
550 */
551 static float
552 compute_lambda_vert(const struct sp_sampler_variant *samp,
553 const float s[TGSI_QUAD_SIZE],
554 const float t[TGSI_QUAD_SIZE],
555 const float p[TGSI_QUAD_SIZE])
556 {
557 return 0.0f;
558 }
559
560
561
562 /**
563 * Get a texel from a texture, using the texture tile cache.
564 *
565 * \param addr the template tex address containing cube, z, face info.
566 * \param x the x coord of texel within 2D image
567 * \param y the y coord of texel within 2D image
568 * \param rgba the quad to put the texel/color into
569 *
570 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
571 * sp_get_cached_tile_tex() function.
572 */
573
574
575
576
577 static INLINE const float *
578 get_texel_2d_no_border(const struct sp_sampler_variant *samp,
579 union tex_tile_address addr, int x, int y)
580 {
581 const struct softpipe_tex_cached_tile *tile;
582
583 addr.bits.x = x / TILE_SIZE;
584 addr.bits.y = y / TILE_SIZE;
585 y %= TILE_SIZE;
586 x %= TILE_SIZE;
587
588 tile = sp_get_cached_tile_tex(samp->cache, addr);
589
590 return &tile->data.color[y][x][0];
591 }
592
593
594 static INLINE const float *
595 get_texel_2d(const struct sp_sampler_variant *samp,
596 union tex_tile_address addr, int x, int y)
597 {
598 const struct pipe_resource *texture = samp->view->texture;
599 unsigned level = addr.bits.level;
600
601 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
602 y < 0 || y >= (int) u_minify(texture->height0, level)) {
603 return samp->sampler->border_color.f;
604 }
605 else {
606 return get_texel_2d_no_border( samp, addr, x, y );
607 }
608 }
609
610 /*
611 * seamless cubemap neighbour array.
612 * this array is used to find the adjacent face in each of 4 directions,
613 * left, right, up, down. (or -x, +x, -y, +y).
614 */
615 static const unsigned face_array[PIPE_TEX_FACE_MAX][4] = {
616 /* pos X first then neg X is Z different, Y the same */
617 /* PIPE_TEX_FACE_POS_X,*/
618 { PIPE_TEX_FACE_POS_Z, PIPE_TEX_FACE_NEG_Z,
619 PIPE_TEX_FACE_NEG_Y, PIPE_TEX_FACE_POS_Y },
620 /* PIPE_TEX_FACE_NEG_X */
621 { PIPE_TEX_FACE_NEG_Z, PIPE_TEX_FACE_POS_Z,
622 PIPE_TEX_FACE_NEG_Y, PIPE_TEX_FACE_POS_Y },
623
624 /* pos Y first then neg Y is X different, X the same */
625 /* PIPE_TEX_FACE_POS_Y */
626 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
627 PIPE_TEX_FACE_POS_Z, PIPE_TEX_FACE_NEG_Z },
628
629 /* PIPE_TEX_FACE_NEG_Y */
630 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
631 PIPE_TEX_FACE_NEG_Z, PIPE_TEX_FACE_POS_Z },
632
633 /* pos Z first then neg Y is X different, X the same */
634 /* PIPE_TEX_FACE_POS_Z */
635 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
636 PIPE_TEX_FACE_NEG_Y, PIPE_TEX_FACE_POS_Y },
637
638 /* PIPE_TEX_FACE_NEG_Z */
639 { PIPE_TEX_FACE_POS_X, PIPE_TEX_FACE_NEG_X,
640 PIPE_TEX_FACE_NEG_Y, PIPE_TEX_FACE_POS_Y }
641 };
642
643 static INLINE unsigned
644 get_next_face(unsigned face, int x, int y)
645 {
646 int idx = 0;
647
648 if (x == 0 && y == 0)
649 return face;
650 if (x == -1)
651 idx = 0;
652 else if (x == 1)
653 idx = 1;
654 else if (y == -1)
655 idx = 2;
656 else if (y == 1)
657 idx = 3;
658
659 return face_array[face][idx];
660 }
661
662 static INLINE const float *
663 get_texel_cube_seamless(const struct sp_sampler_variant *samp,
664 union tex_tile_address addr, int x, int y,
665 float *corner)
666 {
667 const struct pipe_resource *texture = samp->view->texture;
668 unsigned level = addr.bits.level;
669 unsigned face = addr.bits.face;
670 int new_x, new_y;
671 int max_x, max_y;
672 int c;
673
674 max_x = (int) u_minify(texture->width0, level);
675 max_y = (int) u_minify(texture->height0, level);
676 new_x = x;
677 new_y = y;
678
679 /* the corner case */
680 if ((x < 0 || x >= max_x) &&
681 (y < 0 || y >= max_y)) {
682 const float *c1, *c2, *c3;
683 int fx = x < 0 ? 0 : max_x - 1;
684 int fy = y < 0 ? 0 : max_y - 1;
685 c1 = get_texel_2d_no_border( samp, addr, fx, fy);
686 addr.bits.face = get_next_face(face, (x < 0) ? -1 : 1, 0);
687 c2 = get_texel_2d_no_border( samp, addr, (x < 0) ? max_x - 1 : 0, fy);
688 addr.bits.face = get_next_face(face, 0, (y < 0) ? -1 : 1);
689 c3 = get_texel_2d_no_border( samp, addr, fx, (y < 0) ? max_y - 1 : 0);
690 for (c = 0; c < TGSI_QUAD_SIZE; c++)
691 corner[c] = CLAMP((c1[c] + c2[c] + c3[c]), 0.0F, 1.0F) / 3;
692
693 return corner;
694 }
695 /* change the face */
696 if (x < 0) {
697 new_x = max_x - 1;
698 face = get_next_face(face, -1, 0);
699 } else if (x >= max_x) {
700 new_x = 0;
701 face = get_next_face(face, 1, 0);
702 } else if (y < 0) {
703 new_y = max_y - 1;
704 face = get_next_face(face, 0, -1);
705 } else if (y >= max_y) {
706 new_y = 0;
707 face = get_next_face(face, 0, 1);
708 }
709
710 addr.bits.face = face;
711 return get_texel_2d_no_border( samp, addr, new_x, new_y );
712 }
713
714 /* Gather a quad of adjacent texels within a tile:
715 */
716 static INLINE void
717 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_variant *samp,
718 union tex_tile_address addr,
719 unsigned x, unsigned y,
720 const float *out[4])
721 {
722 const struct softpipe_tex_cached_tile *tile;
723
724 addr.bits.x = x / TILE_SIZE;
725 addr.bits.y = y / TILE_SIZE;
726 y %= TILE_SIZE;
727 x %= TILE_SIZE;
728
729 tile = sp_get_cached_tile_tex(samp->cache, addr);
730
731 out[0] = &tile->data.color[y ][x ][0];
732 out[1] = &tile->data.color[y ][x+1][0];
733 out[2] = &tile->data.color[y+1][x ][0];
734 out[3] = &tile->data.color[y+1][x+1][0];
735 }
736
737
738 /* Gather a quad of potentially non-adjacent texels:
739 */
740 static INLINE void
741 get_texel_quad_2d_no_border(const struct sp_sampler_variant *samp,
742 union tex_tile_address addr,
743 int x0, int y0,
744 int x1, int y1,
745 const float *out[4])
746 {
747 out[0] = get_texel_2d_no_border( samp, addr, x0, y0 );
748 out[1] = get_texel_2d_no_border( samp, addr, x1, y0 );
749 out[2] = get_texel_2d_no_border( samp, addr, x0, y1 );
750 out[3] = get_texel_2d_no_border( samp, addr, x1, y1 );
751 }
752
753 /* Can involve a lot of unnecessary checks for border color:
754 */
755 static INLINE void
756 get_texel_quad_2d(const struct sp_sampler_variant *samp,
757 union tex_tile_address addr,
758 int x0, int y0,
759 int x1, int y1,
760 const float *out[4])
761 {
762 out[0] = get_texel_2d( samp, addr, x0, y0 );
763 out[1] = get_texel_2d( samp, addr, x1, y0 );
764 out[3] = get_texel_2d( samp, addr, x1, y1 );
765 out[2] = get_texel_2d( samp, addr, x0, y1 );
766 }
767
768
769
770 /* 3d variants:
771 */
772 static INLINE const float *
773 get_texel_3d_no_border(const struct sp_sampler_variant *samp,
774 union tex_tile_address addr, int x, int y, int z)
775 {
776 const struct softpipe_tex_cached_tile *tile;
777
778 addr.bits.x = x / TILE_SIZE;
779 addr.bits.y = y / TILE_SIZE;
780 addr.bits.z = z;
781 y %= TILE_SIZE;
782 x %= TILE_SIZE;
783
784 tile = sp_get_cached_tile_tex(samp->cache, addr);
785
786 return &tile->data.color[y][x][0];
787 }
788
789
790 static INLINE const float *
791 get_texel_3d(const struct sp_sampler_variant *samp,
792 union tex_tile_address addr, int x, int y, int z)
793 {
794 const struct pipe_resource *texture = samp->view->texture;
795 unsigned level = addr.bits.level;
796
797 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
798 y < 0 || y >= (int) u_minify(texture->height0, level) ||
799 z < 0 || z >= (int) u_minify(texture->depth0, level)) {
800 return samp->sampler->border_color.f;
801 }
802 else {
803 return get_texel_3d_no_border( samp, addr, x, y, z );
804 }
805 }
806
807
808 /* Get texel pointer for 1D array texture */
809 static INLINE const float *
810 get_texel_1d_array(const struct sp_sampler_variant *samp,
811 union tex_tile_address addr, int x, int y)
812 {
813 const struct pipe_resource *texture = samp->view->texture;
814 unsigned level = addr.bits.level;
815
816 if (x < 0 || x >= (int) u_minify(texture->width0, level)) {
817 return samp->sampler->border_color.f;
818 }
819 else {
820 return get_texel_2d_no_border(samp, addr, x, y);
821 }
822 }
823
824
825 /* Get texel pointer for 2D array texture */
826 static INLINE const float *
827 get_texel_2d_array(const struct sp_sampler_variant *samp,
828 union tex_tile_address addr, int x, int y, int layer)
829 {
830 const struct pipe_resource *texture = samp->view->texture;
831 unsigned level = addr.bits.level;
832
833 assert(layer < (int) texture->array_size);
834 assert(layer >= 0);
835
836 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
837 y < 0 || y >= (int) u_minify(texture->height0, level)) {
838 return samp->sampler->border_color.f;
839 }
840 else {
841 return get_texel_3d_no_border(samp, addr, x, y, layer);
842 }
843 }
844
845
846 /* Get texel pointer for cube array texture */
847 static INLINE const float *
848 get_texel_cube_array(const struct sp_sampler_variant *samp,
849 union tex_tile_address addr, int x, int y, int layer)
850 {
851 const struct pipe_resource *texture = samp->view->texture;
852 unsigned level = addr.bits.level;
853
854 assert(layer < (int) texture->array_size);
855 assert(layer >= 0);
856
857 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
858 y < 0 || y >= (int) u_minify(texture->height0, level)) {
859 return samp->sampler->border_color.f;
860 }
861 else {
862 return get_texel_3d_no_border(samp, addr, x, y, layer);
863 }
864 }
865 /**
866 * Given the logbase2 of a mipmap's base level size and a mipmap level,
867 * return the size (in texels) of that mipmap level.
868 * For example, if level[0].width = 256 then base_pot will be 8.
869 * If level = 2, then we'll return 64 (the width at level=2).
870 * Return 1 if level > base_pot.
871 */
872 static INLINE unsigned
873 pot_level_size(unsigned base_pot, unsigned level)
874 {
875 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
876 }
877
878
879 static void
880 print_sample(const char *function, const float *rgba)
881 {
882 debug_printf("%s %g %g %g %g\n",
883 function,
884 rgba[0], rgba[TGSI_NUM_CHANNELS], rgba[2*TGSI_NUM_CHANNELS], rgba[3*TGSI_NUM_CHANNELS]);
885 }
886
887
888 static void
889 print_sample_4(const char *function, float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
890 {
891 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
892 function,
893 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
894 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
895 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
896 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
897 }
898
899 /* Some image-filter fastpaths:
900 */
901 static INLINE void
902 img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
903 float s,
904 float t,
905 float p,
906 unsigned level,
907 unsigned face_id,
908 float *rgba)
909 {
910 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
911 unsigned xpot = pot_level_size(samp->xpot, level);
912 unsigned ypot = pot_level_size(samp->ypot, level);
913 unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */
914 unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */
915 union tex_tile_address addr;
916 int c;
917
918 float u = s * xpot - 0.5F;
919 float v = t * ypot - 0.5F;
920
921 int uflr = util_ifloor(u);
922 int vflr = util_ifloor(v);
923
924 float xw = u - (float)uflr;
925 float yw = v - (float)vflr;
926
927 int x0 = uflr & (xpot - 1);
928 int y0 = vflr & (ypot - 1);
929
930 const float *tx[4];
931
932 addr.value = 0;
933 addr.bits.level = level;
934
935 /* Can we fetch all four at once:
936 */
937 if (x0 < xmax && y0 < ymax) {
938 get_texel_quad_2d_no_border_single_tile(samp, addr, x0, y0, tx);
939 }
940 else {
941 unsigned x1 = (x0 + 1) & (xpot - 1);
942 unsigned y1 = (y0 + 1) & (ypot - 1);
943 get_texel_quad_2d_no_border(samp, addr, x0, y0, x1, y1, tx);
944 }
945
946 /* interpolate R, G, B, A */
947 for (c = 0; c < TGSI_QUAD_SIZE; c++) {
948 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
949 tx[0][c], tx[1][c],
950 tx[2][c], tx[3][c]);
951 }
952
953 if (DEBUG_TEX) {
954 print_sample(__FUNCTION__, rgba);
955 }
956 }
957
958
959 static INLINE void
960 img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
961 float s,
962 float t,
963 float p,
964 unsigned level,
965 unsigned face_id,
966 float rgba[TGSI_QUAD_SIZE])
967 {
968 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
969 unsigned xpot = pot_level_size(samp->xpot, level);
970 unsigned ypot = pot_level_size(samp->ypot, level);
971 const float *out;
972 union tex_tile_address addr;
973 int c;
974
975 float u = s * xpot;
976 float v = t * ypot;
977
978 int uflr = util_ifloor(u);
979 int vflr = util_ifloor(v);
980
981 int x0 = uflr & (xpot - 1);
982 int y0 = vflr & (ypot - 1);
983
984 addr.value = 0;
985 addr.bits.level = level;
986
987 out = get_texel_2d_no_border(samp, addr, x0, y0);
988 for (c = 0; c < TGSI_QUAD_SIZE; c++)
989 rgba[TGSI_NUM_CHANNELS*c] = out[c];
990
991 if (DEBUG_TEX) {
992 print_sample(__FUNCTION__, rgba);
993 }
994 }
995
996
997 static INLINE void
998 img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
999 float s,
1000 float t,
1001 float p,
1002 unsigned level,
1003 unsigned face_id,
1004 float rgba[TGSI_QUAD_SIZE])
1005 {
1006 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1007 unsigned xpot = pot_level_size(samp->xpot, level);
1008 unsigned ypot = pot_level_size(samp->ypot, level);
1009 union tex_tile_address addr;
1010 int c;
1011
1012 float u = s * xpot;
1013 float v = t * ypot;
1014
1015 int x0, y0;
1016 const float *out;
1017
1018 addr.value = 0;
1019 addr.bits.level = level;
1020
1021 x0 = util_ifloor(u);
1022 if (x0 < 0)
1023 x0 = 0;
1024 else if (x0 > xpot - 1)
1025 x0 = xpot - 1;
1026
1027 y0 = util_ifloor(v);
1028 if (y0 < 0)
1029 y0 = 0;
1030 else if (y0 > ypot - 1)
1031 y0 = ypot - 1;
1032
1033 out = get_texel_2d_no_border(samp, addr, x0, y0);
1034 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1035 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1036
1037 if (DEBUG_TEX) {
1038 print_sample(__FUNCTION__, rgba);
1039 }
1040 }
1041
1042
1043 static void
1044 img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
1045 float s,
1046 float t,
1047 float p,
1048 unsigned level,
1049 unsigned face_id,
1050 float rgba[TGSI_QUAD_SIZE])
1051 {
1052 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1053 const struct pipe_resource *texture = samp->view->texture;
1054 int width;
1055 int x;
1056 union tex_tile_address addr;
1057 const float *out;
1058 int c;
1059
1060 width = u_minify(texture->width0, level);
1061
1062 assert(width > 0);
1063
1064 addr.value = 0;
1065 addr.bits.level = level;
1066
1067 samp->nearest_texcoord_s(s, width, &x);
1068
1069 out = get_texel_2d(samp, addr, x, 0);
1070 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1071 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1072
1073 if (DEBUG_TEX) {
1074 print_sample(__FUNCTION__, rgba);
1075 }
1076 }
1077
1078
1079 static void
1080 img_filter_1d_array_nearest(struct tgsi_sampler *tgsi_sampler,
1081 float s,
1082 float t,
1083 float p,
1084 unsigned level,
1085 unsigned face_id,
1086 float *rgba)
1087 {
1088 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1089 const struct pipe_resource *texture = samp->view->texture;
1090 int width;
1091 int x, layer;
1092 union tex_tile_address addr;
1093 const float *out;
1094 int c;
1095
1096 width = u_minify(texture->width0, level);
1097
1098 assert(width > 0);
1099
1100 addr.value = 0;
1101 addr.bits.level = level;
1102
1103 samp->nearest_texcoord_s(s, width, &x);
1104 wrap_array_layer(t, texture->array_size, &layer);
1105
1106 out = get_texel_1d_array(samp, addr, x, layer);
1107 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1108 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1109
1110 if (DEBUG_TEX) {
1111 print_sample(__FUNCTION__, rgba);
1112 }
1113 }
1114
1115
1116 static void
1117 img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
1118 float s,
1119 float t,
1120 float p,
1121 unsigned level,
1122 unsigned face_id,
1123 float *rgba)
1124 {
1125 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1126 const struct pipe_resource *texture = samp->view->texture;
1127 int width, height;
1128 int x, y;
1129 union tex_tile_address addr;
1130 const float *out;
1131 int c;
1132
1133 width = u_minify(texture->width0, level);
1134 height = u_minify(texture->height0, level);
1135
1136 assert(width > 0);
1137 assert(height > 0);
1138
1139 addr.value = 0;
1140 addr.bits.level = level;
1141
1142 samp->nearest_texcoord_s(s, width, &x);
1143 samp->nearest_texcoord_t(t, height, &y);
1144
1145 out = get_texel_2d(samp, addr, x, y);
1146 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1147 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1148
1149 if (DEBUG_TEX) {
1150 print_sample(__FUNCTION__, rgba);
1151 }
1152 }
1153
1154
1155 static void
1156 img_filter_2d_array_nearest(struct tgsi_sampler *tgsi_sampler,
1157 float s,
1158 float t,
1159 float p,
1160 unsigned level,
1161 unsigned face_id,
1162 float *rgba)
1163 {
1164 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1165 const struct pipe_resource *texture = samp->view->texture;
1166 int width, height;
1167 int x, y, layer;
1168 union tex_tile_address addr;
1169 const float *out;
1170 int c;
1171
1172 width = u_minify(texture->width0, level);
1173 height = u_minify(texture->height0, level);
1174
1175 assert(width > 0);
1176 assert(height > 0);
1177
1178 addr.value = 0;
1179 addr.bits.level = level;
1180
1181 samp->nearest_texcoord_s(s, width, &x);
1182 samp->nearest_texcoord_t(t, height, &y);
1183 wrap_array_layer(p, texture->array_size, &layer);
1184
1185 out = get_texel_2d_array(samp, addr, x, y, layer);
1186 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1187 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1188
1189 if (DEBUG_TEX) {
1190 print_sample(__FUNCTION__, rgba);
1191 }
1192 }
1193
1194
1195 static INLINE union tex_tile_address
1196 face(union tex_tile_address addr, unsigned face )
1197 {
1198 addr.bits.face = face;
1199 return addr;
1200 }
1201
1202
1203 static void
1204 img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
1205 float s,
1206 float t,
1207 float p,
1208 unsigned level,
1209 unsigned face_id,
1210 float *rgba)
1211 {
1212 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1213 const struct pipe_resource *texture = samp->view->texture;
1214 int width, height;
1215 int x, y;
1216 union tex_tile_address addr;
1217 const float *out;
1218 int c;
1219
1220 width = u_minify(texture->width0, level);
1221 height = u_minify(texture->height0, level);
1222
1223 assert(width > 0);
1224 assert(height > 0);
1225
1226 addr.value = 0;
1227 addr.bits.level = level;
1228
1229 /*
1230 * If NEAREST filtering is done within a miplevel, always apply wrap
1231 * mode CLAMP_TO_EDGE.
1232 */
1233 if (samp->sampler->seamless_cube_map) {
1234 wrap_nearest_clamp_to_edge(s, width, &x);
1235 wrap_nearest_clamp_to_edge(t, height, &y);
1236 } else {
1237 samp->nearest_texcoord_s(s, width, &x);
1238 samp->nearest_texcoord_t(t, height, &y);
1239 }
1240
1241 out = get_texel_2d(samp, face(addr, face_id), x, y);
1242 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1243 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1244
1245 if (DEBUG_TEX) {
1246 print_sample(__FUNCTION__, rgba);
1247 }
1248 }
1249
1250 static void
1251 img_filter_cube_array_nearest(struct tgsi_sampler *tgsi_sampler,
1252 float s,
1253 float t,
1254 float p,
1255 unsigned level,
1256 unsigned face_id,
1257 float *rgba)
1258 {
1259 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1260 const struct pipe_resource *texture = samp->view->texture;
1261 int width, height;
1262 int x, y, layer;
1263 union tex_tile_address addr;
1264 const float *out;
1265 int c;
1266
1267 width = u_minify(texture->width0, level);
1268 height = u_minify(texture->height0, level);
1269
1270 assert(width > 0);
1271 assert(height > 0);
1272
1273 addr.value = 0;
1274 addr.bits.level = level;
1275
1276 samp->nearest_texcoord_s(s, width, &x);
1277 samp->nearest_texcoord_t(t, height, &y);
1278 wrap_array_layer(p, texture->array_size, &layer);
1279
1280 out = get_texel_cube_array(samp, addr, x, y, layer * 6 + face_id);
1281 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1282 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1283
1284 if (DEBUG_TEX) {
1285 print_sample(__FUNCTION__, rgba);
1286 }
1287 }
1288
1289 static void
1290 img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
1291 float s,
1292 float t,
1293 float p,
1294 unsigned level,
1295 unsigned face_id,
1296 float *rgba)
1297 {
1298 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1299 const struct pipe_resource *texture = samp->view->texture;
1300 int width, height, depth;
1301 int x, y, z;
1302 union tex_tile_address addr;
1303 const float *out;
1304 int c;
1305
1306 width = u_minify(texture->width0, level);
1307 height = u_minify(texture->height0, level);
1308 depth = u_minify(texture->depth0, level);
1309
1310 assert(width > 0);
1311 assert(height > 0);
1312 assert(depth > 0);
1313
1314 samp->nearest_texcoord_s(s, width, &x);
1315 samp->nearest_texcoord_t(t, height, &y);
1316 samp->nearest_texcoord_p(p, depth, &z);
1317
1318 addr.value = 0;
1319 addr.bits.level = level;
1320
1321 out = get_texel_3d(samp, addr, x, y, z);
1322 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1323 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1324 }
1325
1326
1327 static void
1328 img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
1329 float s,
1330 float t,
1331 float p,
1332 unsigned level,
1333 unsigned face_id,
1334 float *rgba)
1335 {
1336 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1337 const struct pipe_resource *texture = samp->view->texture;
1338 int width;
1339 int x0, x1;
1340 float xw; /* weights */
1341 union tex_tile_address addr;
1342 const float *tx0, *tx1;
1343 int c;
1344
1345 width = u_minify(texture->width0, level);
1346
1347 assert(width > 0);
1348
1349 addr.value = 0;
1350 addr.bits.level = level;
1351
1352 samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1353
1354 tx0 = get_texel_2d(samp, addr, x0, 0);
1355 tx1 = get_texel_2d(samp, addr, x1, 0);
1356
1357 /* interpolate R, G, B, A */
1358 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1359 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1360 }
1361
1362
1363 static void
1364 img_filter_1d_array_linear(struct tgsi_sampler *tgsi_sampler,
1365 float s,
1366 float t,
1367 float p,
1368 unsigned level,
1369 unsigned face_id,
1370 float *rgba)
1371 {
1372 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1373 const struct pipe_resource *texture = samp->view->texture;
1374 int width;
1375 int x0, x1, layer;
1376 float xw; /* weights */
1377 union tex_tile_address addr;
1378 const float *tx0, *tx1;
1379 int c;
1380
1381 width = u_minify(texture->width0, level);
1382
1383 assert(width > 0);
1384
1385 addr.value = 0;
1386 addr.bits.level = level;
1387
1388 samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1389 wrap_array_layer(t, texture->array_size, &layer);
1390
1391 tx0 = get_texel_1d_array(samp, addr, x0, layer);
1392 tx1 = get_texel_1d_array(samp, addr, x1, layer);
1393
1394 /* interpolate R, G, B, A */
1395 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1396 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1397 }
1398
1399
1400 static void
1401 img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
1402 float s,
1403 float t,
1404 float p,
1405 unsigned level,
1406 unsigned face_id,
1407 float *rgba)
1408 {
1409 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1410 const struct pipe_resource *texture = samp->view->texture;
1411 int width, height;
1412 int x0, y0, x1, y1;
1413 float xw, yw; /* weights */
1414 union tex_tile_address addr;
1415 const float *tx0, *tx1, *tx2, *tx3;
1416 int c;
1417
1418 width = u_minify(texture->width0, level);
1419 height = u_minify(texture->height0, level);
1420
1421 assert(width > 0);
1422 assert(height > 0);
1423
1424 addr.value = 0;
1425 addr.bits.level = level;
1426
1427 samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1428 samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1429
1430 tx0 = get_texel_2d(samp, addr, x0, y0);
1431 tx1 = get_texel_2d(samp, addr, x1, y0);
1432 tx2 = get_texel_2d(samp, addr, x0, y1);
1433 tx3 = get_texel_2d(samp, addr, x1, y1);
1434
1435 /* interpolate R, G, B, A */
1436 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1437 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1438 tx0[c], tx1[c],
1439 tx2[c], tx3[c]);
1440 }
1441
1442
1443 static void
1444 img_filter_2d_array_linear(struct tgsi_sampler *tgsi_sampler,
1445 float s,
1446 float t,
1447 float p,
1448 unsigned level,
1449 unsigned face_id,
1450 float *rgba)
1451 {
1452 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1453 const struct pipe_resource *texture = samp->view->texture;
1454 int width, height;
1455 int x0, y0, x1, y1, layer;
1456 float xw, yw; /* weights */
1457 union tex_tile_address addr;
1458 const float *tx0, *tx1, *tx2, *tx3;
1459 int c;
1460
1461 width = u_minify(texture->width0, level);
1462 height = u_minify(texture->height0, level);
1463
1464 assert(width > 0);
1465 assert(height > 0);
1466
1467 addr.value = 0;
1468 addr.bits.level = level;
1469
1470 samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1471 samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1472 wrap_array_layer(p, texture->array_size, &layer);
1473
1474 tx0 = get_texel_2d_array(samp, addr, x0, y0, layer);
1475 tx1 = get_texel_2d_array(samp, addr, x1, y0, layer);
1476 tx2 = get_texel_2d_array(samp, addr, x0, y1, layer);
1477 tx3 = get_texel_2d_array(samp, addr, x1, y1, layer);
1478
1479 /* interpolate R, G, B, A */
1480 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1481 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1482 tx0[c], tx1[c],
1483 tx2[c], tx3[c]);
1484 }
1485
1486
1487 static void
1488 img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler,
1489 float s,
1490 float t,
1491 float p,
1492 unsigned level,
1493 unsigned face_id,
1494 float *rgba)
1495 {
1496 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1497 const struct pipe_resource *texture = samp->view->texture;
1498 int width, height;
1499 int x0, y0, x1, y1;
1500 float xw, yw; /* weights */
1501 union tex_tile_address addr, addrj;
1502 const float *tx0, *tx1, *tx2, *tx3;
1503 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE], corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1504 int c;
1505
1506 width = u_minify(texture->width0, level);
1507 height = u_minify(texture->height0, level);
1508
1509 assert(width > 0);
1510 assert(height > 0);
1511
1512 addr.value = 0;
1513 addr.bits.level = level;
1514
1515 /*
1516 * For seamless if LINEAR filtering is done within a miplevel,
1517 * always apply wrap mode CLAMP_TO_BORDER.
1518 */
1519 if (samp->sampler->seamless_cube_map) {
1520 wrap_linear_clamp_to_border(s, width, &x0, &x1, &xw);
1521 wrap_linear_clamp_to_border(t, height, &y0, &y1, &yw);
1522 } else {
1523 samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1524 samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1525 }
1526
1527 addrj = face(addr, face_id);
1528
1529 if (samp->sampler->seamless_cube_map) {
1530 tx0 = get_texel_cube_seamless(samp, addrj, x0, y0, corner0);
1531 tx1 = get_texel_cube_seamless(samp, addrj, x1, y0, corner1);
1532 tx2 = get_texel_cube_seamless(samp, addrj, x0, y1, corner2);
1533 tx3 = get_texel_cube_seamless(samp, addrj, x1, y1, corner3);
1534 } else {
1535 tx0 = get_texel_2d(samp, addrj, x0, y0);
1536 tx1 = get_texel_2d(samp, addrj, x1, y0);
1537 tx2 = get_texel_2d(samp, addrj, x0, y1);
1538 tx3 = get_texel_2d(samp, addrj, x1, y1);
1539 }
1540 /* interpolate R, G, B, A */
1541 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1542 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1543 tx0[c], tx1[c],
1544 tx2[c], tx3[c]);
1545 }
1546
1547
1548 static void
1549 img_filter_cube_array_linear(struct tgsi_sampler *tgsi_sampler,
1550 float s,
1551 float t,
1552 float p,
1553 unsigned level,
1554 unsigned face_id,
1555 float *rgba)
1556 {
1557 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1558 const struct pipe_resource *texture = samp->view->texture;
1559 int width, height;
1560 int x0, y0, x1, y1, layer;
1561 float xw, yw; /* weights */
1562 union tex_tile_address addr;
1563 const float *tx0, *tx1, *tx2, *tx3;
1564 int c;
1565
1566 width = u_minify(texture->width0, level);
1567 height = u_minify(texture->height0, level);
1568
1569 assert(width > 0);
1570 assert(height > 0);
1571
1572 addr.value = 0;
1573 addr.bits.level = level;
1574
1575 samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1576 samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1577 wrap_array_layer(p, texture->array_size, &layer);
1578
1579 tx0 = get_texel_cube_array(samp, addr, x0, y0, layer * 6 + face_id);
1580 tx1 = get_texel_cube_array(samp, addr, x1, y0, layer * 6 + face_id);
1581 tx2 = get_texel_cube_array(samp, addr, x0, y1, layer * 6 + face_id);
1582 tx3 = get_texel_cube_array(samp, addr, x1, y1, layer * 6 + face_id);
1583
1584 /* interpolate R, G, B, A */
1585 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1586 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1587 tx0[c], tx1[c],
1588 tx2[c], tx3[c]);
1589 }
1590
1591 static void
1592 img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
1593 float s,
1594 float t,
1595 float p,
1596 unsigned level,
1597 unsigned face_id,
1598 float *rgba)
1599 {
1600 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1601 const struct pipe_resource *texture = samp->view->texture;
1602 int width, height, depth;
1603 int x0, x1, y0, y1, z0, z1;
1604 float xw, yw, zw; /* interpolation weights */
1605 union tex_tile_address addr;
1606 const float *tx00, *tx01, *tx02, *tx03, *tx10, *tx11, *tx12, *tx13;
1607 int c;
1608
1609 width = u_minify(texture->width0, level);
1610 height = u_minify(texture->height0, level);
1611 depth = u_minify(texture->depth0, level);
1612
1613 addr.value = 0;
1614 addr.bits.level = level;
1615
1616 assert(width > 0);
1617 assert(height > 0);
1618 assert(depth > 0);
1619
1620 samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1621 samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1622 samp->linear_texcoord_p(p, depth, &z0, &z1, &zw);
1623
1624
1625 tx00 = get_texel_3d(samp, addr, x0, y0, z0);
1626 tx01 = get_texel_3d(samp, addr, x1, y0, z0);
1627 tx02 = get_texel_3d(samp, addr, x0, y1, z0);
1628 tx03 = get_texel_3d(samp, addr, x1, y1, z0);
1629
1630 tx10 = get_texel_3d(samp, addr, x0, y0, z1);
1631 tx11 = get_texel_3d(samp, addr, x1, y0, z1);
1632 tx12 = get_texel_3d(samp, addr, x0, y1, z1);
1633 tx13 = get_texel_3d(samp, addr, x1, y1, z1);
1634
1635 /* interpolate R, G, B, A */
1636 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1637 rgba[TGSI_NUM_CHANNELS*c] = lerp_3d(xw, yw, zw,
1638 tx00[c], tx01[c],
1639 tx02[c], tx03[c],
1640 tx10[c], tx11[c],
1641 tx12[c], tx13[c]);
1642 }
1643
1644
1645 /* Calculate level of detail for every fragment.
1646 * Note that lambda has already been biased by global LOD bias.
1647 */
1648 static INLINE void
1649 compute_lod(const struct pipe_sampler_state *sampler,
1650 const float biased_lambda,
1651 const float lodbias[TGSI_QUAD_SIZE],
1652 float lod[TGSI_QUAD_SIZE])
1653 {
1654 uint i;
1655
1656 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1657 lod[i] = biased_lambda + lodbias[i];
1658 lod[i] = CLAMP(lod[i], sampler->min_lod, sampler->max_lod);
1659 }
1660 }
1661
1662
1663 static void
1664 mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
1665 const float s[TGSI_QUAD_SIZE],
1666 const float t[TGSI_QUAD_SIZE],
1667 const float p[TGSI_QUAD_SIZE],
1668 const float c0[TGSI_QUAD_SIZE],
1669 const float c1[TGSI_QUAD_SIZE],
1670 enum tgsi_sampler_control control,
1671 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1672 {
1673 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1674 const struct pipe_resource *texture = samp->view->texture;
1675 int j;
1676 float lod[TGSI_QUAD_SIZE];
1677
1678 if (control == tgsi_sampler_lod_bias) {
1679 float lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1680 if (samp->key.bits.target == PIPE_TEXTURE_CUBE_ARRAY)
1681 compute_lod(samp->sampler, lambda, c1, lod);
1682 else
1683 compute_lod(samp->sampler, lambda, c0, lod);
1684 } else {
1685 assert(control == tgsi_sampler_lod_explicit);
1686
1687 if (samp->key.bits.target == PIPE_TEXTURE_CUBE_ARRAY)
1688 memcpy(lod, c1, sizeof(lod));
1689 else
1690 memcpy(lod, c0, sizeof(lod));
1691
1692 }
1693
1694 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1695 int level0 = samp->view->u.tex.first_level + (int)lod[j];
1696
1697 if (lod[j] < 0.0)
1698 samp->mag_img_filter(tgsi_sampler, s[j], t[j], p[j], samp->view->u.tex.first_level,
1699 samp->faces[j], &rgba[0][j]);
1700
1701 else if (level0 >= texture->last_level)
1702 samp->min_img_filter(tgsi_sampler, s[j], t[j], p[j], texture->last_level,
1703 samp->faces[j], &rgba[0][j]);
1704
1705 else {
1706 float levelBlend = frac(lod[j]);
1707 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
1708 int c;
1709
1710 samp->min_img_filter(tgsi_sampler, s[j], t[j], p[j], level0,
1711 samp->faces[j], &rgbax[0][0]);
1712 samp->min_img_filter(tgsi_sampler, s[j], t[j], p[j], level0+1,
1713 samp->faces[j], &rgbax[0][1]);
1714
1715 for (c = 0; c < 4; c++) {
1716 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
1717 }
1718 }
1719 }
1720
1721 if (DEBUG_TEX) {
1722 print_sample_4(__FUNCTION__, rgba);
1723 }
1724 }
1725
1726
1727 /**
1728 * Compute nearest mipmap level from texcoords.
1729 * Then sample the texture level for four elements of a quad.
1730 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1731 */
1732 static void
1733 mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
1734 const float s[TGSI_QUAD_SIZE],
1735 const float t[TGSI_QUAD_SIZE],
1736 const float p[TGSI_QUAD_SIZE],
1737 const float c0[TGSI_QUAD_SIZE],
1738 const float c1[TGSI_QUAD_SIZE],
1739 enum tgsi_sampler_control control,
1740 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1741 {
1742 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1743 const struct pipe_resource *texture = samp->view->texture;
1744 float lod[TGSI_QUAD_SIZE];
1745 int j;
1746
1747 if (control == tgsi_sampler_lod_bias) {
1748 float lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1749 if (samp->key.bits.target == PIPE_TEXTURE_CUBE_ARRAY)
1750 compute_lod(samp->sampler, lambda, c1, lod);
1751 else
1752 compute_lod(samp->sampler, lambda, c0, lod);
1753 } else {
1754 assert(control == tgsi_sampler_lod_explicit);
1755
1756 if (samp->key.bits.target == PIPE_TEXTURE_CUBE_ARRAY)
1757 memcpy(lod, c1, sizeof(lod));
1758 else
1759 memcpy(lod, c0, sizeof(lod));
1760 }
1761
1762 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1763 if (lod[j] < 0.0)
1764 samp->mag_img_filter(tgsi_sampler, s[j], t[j], p[j], samp->view->u.tex.first_level,
1765 samp->faces[j], &rgba[0][j]);
1766 else {
1767 float level = samp->view->u.tex.first_level + (int)(lod[j] + 0.5F) ;
1768 level = MIN2(level, (int)texture->last_level);
1769 samp->min_img_filter(tgsi_sampler, s[j], t[j], p[j], level, samp->faces[j],
1770 &rgba[0][j]);
1771 }
1772 }
1773
1774 if (DEBUG_TEX) {
1775 print_sample_4(__FUNCTION__, rgba);
1776 }
1777 }
1778
1779
1780 static void
1781 mip_filter_none(struct tgsi_sampler *tgsi_sampler,
1782 const float s[TGSI_QUAD_SIZE],
1783 const float t[TGSI_QUAD_SIZE],
1784 const float p[TGSI_QUAD_SIZE],
1785 const float c0[TGSI_QUAD_SIZE],
1786 const float c1[TGSI_QUAD_SIZE],
1787 enum tgsi_sampler_control control,
1788 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1789 {
1790 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1791 float lod[TGSI_QUAD_SIZE];
1792 int j;
1793
1794 if (control == tgsi_sampler_lod_bias) {
1795 float lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1796 if (samp->key.bits.target == PIPE_TEXTURE_CUBE_ARRAY)
1797 compute_lod(samp->sampler, lambda, c1, lod);
1798 else
1799 compute_lod(samp->sampler, lambda, c0, lod);
1800 } else {
1801 assert(control == tgsi_sampler_lod_explicit);
1802
1803 if (samp->key.bits.target == PIPE_TEXTURE_CUBE_ARRAY)
1804 memcpy(lod, c1, sizeof(lod));
1805 else
1806 memcpy(lod, c0, sizeof(lod));
1807 }
1808
1809 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1810 if (lod[j] < 0.0) {
1811 samp->mag_img_filter(tgsi_sampler, s[j], t[j], p[j], samp->view->u.tex.first_level,
1812 samp->faces[j], &rgba[0][j]);
1813 }
1814 else {
1815 samp->min_img_filter(tgsi_sampler, s[j], t[j], p[j], samp->view->u.tex.first_level,
1816 samp->faces[j], &rgba[0][j]);
1817 }
1818 }
1819 }
1820
1821
1822 static void
1823 mip_filter_none_no_filter_select(struct tgsi_sampler *tgsi_sampler,
1824 const float s[TGSI_QUAD_SIZE],
1825 const float t[TGSI_QUAD_SIZE],
1826 const float p[TGSI_QUAD_SIZE],
1827 const float c0[TGSI_QUAD_SIZE],
1828 const float c1[TGSI_QUAD_SIZE],
1829 enum tgsi_sampler_control control,
1830 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1831 {
1832 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1833 int j;
1834
1835 for (j = 0; j < TGSI_QUAD_SIZE; j++)
1836 samp->mag_img_filter(tgsi_sampler, s[j], t[j], p[j], samp->view->u.tex.first_level,
1837 samp->faces[j], &rgba[0][j]);
1838 }
1839
1840
1841 /* For anisotropic filtering */
1842 #define WEIGHT_LUT_SIZE 1024
1843
1844 static float *weightLut = NULL;
1845
1846 /**
1847 * Creates the look-up table used to speed-up EWA sampling
1848 */
1849 static void
1850 create_filter_table(void)
1851 {
1852 unsigned i;
1853 if (!weightLut) {
1854 weightLut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
1855
1856 for (i = 0; i < WEIGHT_LUT_SIZE; ++i) {
1857 float alpha = 2;
1858 float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
1859 float weight = (float) exp(-alpha * r2);
1860 weightLut[i] = weight;
1861 }
1862 }
1863 }
1864
1865
1866 /**
1867 * Elliptical weighted average (EWA) filter for producing high quality
1868 * anisotropic filtered results.
1869 * Based on the Higher Quality Elliptical Weighted Average Filter
1870 * published by Paul S. Heckbert in his Master's Thesis
1871 * "Fundamentals of Texture Mapping and Image Warping" (1989)
1872 */
1873 static void
1874 img_filter_2d_ewa(struct tgsi_sampler *tgsi_sampler,
1875 const float s[TGSI_QUAD_SIZE],
1876 const float t[TGSI_QUAD_SIZE],
1877 const float p[TGSI_QUAD_SIZE],
1878 unsigned level,
1879 const float dudx, const float dvdx,
1880 const float dudy, const float dvdy,
1881 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1882 {
1883 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1884 const struct pipe_resource *texture = samp->view->texture;
1885
1886 // ??? Won't the image filters blow up if level is negative?
1887 unsigned level0 = level > 0 ? level : 0;
1888 float scaling = 1.0 / (1 << level0);
1889 int width = u_minify(texture->width0, level0);
1890 int height = u_minify(texture->height0, level0);
1891
1892 float ux = dudx * scaling;
1893 float vx = dvdx * scaling;
1894 float uy = dudy * scaling;
1895 float vy = dvdy * scaling;
1896
1897 /* compute ellipse coefficients to bound the region:
1898 * A*x*x + B*x*y + C*y*y = F.
1899 */
1900 float A = vx*vx+vy*vy+1;
1901 float B = -2*(ux*vx+uy*vy);
1902 float C = ux*ux+uy*uy+1;
1903 float F = A*C-B*B/4.0;
1904
1905 /* check if it is an ellipse */
1906 /* ASSERT(F > 0.0); */
1907
1908 /* Compute the ellipse's (u,v) bounding box in texture space */
1909 float d = -B*B+4.0*C*A;
1910 float box_u = 2.0 / d * sqrt(d*C*F); /* box_u -> half of bbox with */
1911 float box_v = 2.0 / d * sqrt(A*d*F); /* box_v -> half of bbox height */
1912
1913 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
1914 float s_buffer[TGSI_QUAD_SIZE];
1915 float t_buffer[TGSI_QUAD_SIZE];
1916 float weight_buffer[TGSI_QUAD_SIZE];
1917 unsigned buffer_next;
1918 int j;
1919 float den; /* = 0.0F; */
1920 float ddq;
1921 float U; /* = u0 - tex_u; */
1922 int v;
1923
1924 /* Scale ellipse formula to directly index the Filter Lookup Table.
1925 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
1926 */
1927 double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
1928 A *= formScale;
1929 B *= formScale;
1930 C *= formScale;
1931 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
1932
1933 /* For each quad, the du and dx values are the same and so the ellipse is
1934 * also the same. Note that texel/image access can only be performed using
1935 * a quad, i.e. it is not possible to get the pixel value for a single
1936 * tex coord. In order to have a better performance, the access is buffered
1937 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
1938 * full, then the pixel values are read from the image.
1939 */
1940 ddq = 2 * A;
1941
1942 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1943 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
1944 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
1945 * value, q, is less than F, we're inside the ellipse
1946 */
1947 float tex_u = -0.5F + s[j] * texture->width0 * scaling;
1948 float tex_v = -0.5F + t[j] * texture->height0 * scaling;
1949
1950 int u0 = (int) floorf(tex_u - box_u);
1951 int u1 = (int) ceilf(tex_u + box_u);
1952 int v0 = (int) floorf(tex_v - box_v);
1953 int v1 = (int) ceilf(tex_v + box_v);
1954
1955 float num[4] = {0.0F, 0.0F, 0.0F, 0.0F};
1956 buffer_next = 0;
1957 den = 0;
1958 U = u0 - tex_u;
1959 for (v = v0; v <= v1; ++v) {
1960 float V = v - tex_v;
1961 float dq = A * (2 * U + 1) + B * V;
1962 float q = (C * V + B * U) * V + A * U * U;
1963
1964 int u;
1965 for (u = u0; u <= u1; ++u) {
1966 /* Note that the ellipse has been pre-scaled so F =
1967 * WEIGHT_LUT_SIZE - 1
1968 */
1969 if (q < WEIGHT_LUT_SIZE) {
1970 /* as a LUT is used, q must never be negative;
1971 * should not happen, though
1972 */
1973 const int qClamped = q >= 0.0F ? q : 0;
1974 float weight = weightLut[qClamped];
1975
1976 weight_buffer[buffer_next] = weight;
1977 s_buffer[buffer_next] = u / ((float) width);
1978 t_buffer[buffer_next] = v / ((float) height);
1979
1980 buffer_next++;
1981 if (buffer_next == TGSI_QUAD_SIZE) {
1982 /* 4 texel coords are in the buffer -> read it now */
1983 unsigned jj;
1984 /* it is assumed that samp->min_img_filter is set to
1985 * img_filter_2d_nearest or one of the
1986 * accelerated img_filter_2d_nearest_XXX functions.
1987 */
1988 for (jj = 0; jj < buffer_next; jj++) {
1989 samp->min_img_filter(tgsi_sampler, s_buffer[jj], t_buffer[jj], p[jj],
1990 level, samp->faces[j], &rgba_temp[0][jj]);
1991 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
1992 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
1993 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
1994 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
1995 }
1996
1997 buffer_next = 0;
1998 }
1999
2000 den += weight;
2001 }
2002 q += dq;
2003 dq += ddq;
2004 }
2005 }
2006
2007 /* if the tex coord buffer contains unread values, we will read
2008 * them now.
2009 */
2010 if (buffer_next > 0) {
2011 unsigned jj;
2012 /* it is assumed that samp->min_img_filter is set to
2013 * img_filter_2d_nearest or one of the
2014 * accelerated img_filter_2d_nearest_XXX functions.
2015 */
2016 for (jj = 0; jj < buffer_next; jj++) {
2017 samp->min_img_filter(tgsi_sampler, s_buffer[jj], t_buffer[jj], p[jj], level,
2018 samp->faces[j], &rgba_temp[0][jj]);
2019 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2020 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2021 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2022 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2023 }
2024 }
2025
2026 if (den <= 0.0F) {
2027 /* Reaching this place would mean that no pixels intersected
2028 * the ellipse. This should never happen because the filter
2029 * we use always intersects at least one pixel.
2030 */
2031
2032 /*rgba[0]=0;
2033 rgba[1]=0;
2034 rgba[2]=0;
2035 rgba[3]=0;*/
2036 /* not enough pixels in resampling, resort to direct interpolation */
2037 samp->min_img_filter(tgsi_sampler, s[j], t[j], p[j], level, samp->faces[j],
2038 &rgba_temp[0][j]);
2039 den = 1;
2040 num[0] = rgba_temp[0][j];
2041 num[1] = rgba_temp[1][j];
2042 num[2] = rgba_temp[2][j];
2043 num[3] = rgba_temp[3][j];
2044 }
2045
2046 rgba[0][j] = num[0] / den;
2047 rgba[1][j] = num[1] / den;
2048 rgba[2][j] = num[2] / den;
2049 rgba[3][j] = num[3] / den;
2050 }
2051 }
2052
2053
2054 /**
2055 * Sample 2D texture using an anisotropic filter.
2056 */
2057 static void
2058 mip_filter_linear_aniso(struct tgsi_sampler *tgsi_sampler,
2059 const float s[TGSI_QUAD_SIZE],
2060 const float t[TGSI_QUAD_SIZE],
2061 const float p[TGSI_QUAD_SIZE],
2062 const float c0[TGSI_QUAD_SIZE],
2063 const float c1[TGSI_QUAD_SIZE],
2064 enum tgsi_sampler_control control,
2065 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2066 {
2067 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2068 const struct pipe_resource *texture = samp->view->texture;
2069 int level0;
2070 float lambda;
2071 float lod[TGSI_QUAD_SIZE];
2072
2073 float s_to_u = u_minify(texture->width0, samp->view->u.tex.first_level);
2074 float t_to_v = u_minify(texture->height0, samp->view->u.tex.first_level);
2075 float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2076 float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2077 float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2078 float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2079
2080 if (control == tgsi_sampler_lod_bias) {
2081 /* note: instead of working with Px and Py, we will use the
2082 * squared length instead, to avoid sqrt.
2083 */
2084 float Px2 = dudx * dudx + dvdx * dvdx;
2085 float Py2 = dudy * dudy + dvdy * dvdy;
2086
2087 float Pmax2;
2088 float Pmin2;
2089 float e;
2090 const float maxEccentricity = samp->sampler->max_anisotropy * samp->sampler->max_anisotropy;
2091
2092 if (Px2 < Py2) {
2093 Pmax2 = Py2;
2094 Pmin2 = Px2;
2095 }
2096 else {
2097 Pmax2 = Px2;
2098 Pmin2 = Py2;
2099 }
2100
2101 /* if the eccentricity of the ellipse is too big, scale up the shorter
2102 * of the two vectors to limit the maximum amount of work per pixel
2103 */
2104 e = Pmax2 / Pmin2;
2105 if (e > maxEccentricity) {
2106 /* float s=e / maxEccentricity;
2107 minor[0] *= s;
2108 minor[1] *= s;
2109 Pmin2 *= s; */
2110 Pmin2 = Pmax2 / maxEccentricity;
2111 }
2112
2113 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2114 * this since 0.5*log(x) = log(sqrt(x))
2115 */
2116 lambda = 0.5F * util_fast_log2(Pmin2) + samp->sampler->lod_bias;
2117 compute_lod(samp->sampler, lambda, c0, lod);
2118 }
2119 else {
2120 assert(control == tgsi_sampler_lod_explicit);
2121
2122 memcpy(lod, c0, sizeof(lod));
2123 }
2124
2125 /* XXX: Take into account all lod values.
2126 */
2127 lambda = lod[0];
2128 level0 = samp->view->u.tex.first_level + (int)lambda;
2129
2130 /* If the ellipse covers the whole image, we can
2131 * simply return the average of the whole image.
2132 */
2133 if (level0 >= (int) texture->last_level) {
2134 int j;
2135 for (j = 0; j < TGSI_QUAD_SIZE; j++)
2136 samp->min_img_filter(tgsi_sampler, s[j], t[j], p[j], texture->last_level,
2137 samp->faces[j], &rgba[0][j]);
2138 }
2139 else {
2140 /* don't bother interpolating between multiple LODs; it doesn't
2141 * seem to be worth the extra running time.
2142 */
2143 img_filter_2d_ewa(tgsi_sampler, s, t, p, level0,
2144 dudx, dvdx, dudy, dvdy, rgba);
2145 }
2146
2147 if (DEBUG_TEX) {
2148 print_sample_4(__FUNCTION__, rgba);
2149 }
2150 }
2151
2152
2153 /**
2154 * Specialized version of mip_filter_linear with hard-wired calls to
2155 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2156 */
2157 static void
2158 mip_filter_linear_2d_linear_repeat_POT(
2159 struct tgsi_sampler *tgsi_sampler,
2160 const float s[TGSI_QUAD_SIZE],
2161 const float t[TGSI_QUAD_SIZE],
2162 const float p[TGSI_QUAD_SIZE],
2163 const float c0[TGSI_QUAD_SIZE],
2164 const float c1[TGSI_QUAD_SIZE],
2165 enum tgsi_sampler_control control,
2166 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2167 {
2168 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2169 const struct pipe_resource *texture = samp->view->texture;
2170 int j;
2171 float lambda;
2172 float lod[TGSI_QUAD_SIZE];
2173
2174 if (control == tgsi_sampler_lod_bias) {
2175 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
2176 compute_lod(samp->sampler, lambda, c0, lod);
2177 } else {
2178 assert(control == tgsi_sampler_lod_explicit);
2179
2180 memcpy(lod, c0, sizeof(lod));
2181 }
2182
2183 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2184 int level0 = samp->view->u.tex.first_level + (int)lod[j];
2185
2186 /* Catches both negative and large values of level0:
2187 */
2188 if ((unsigned)level0 >= texture->last_level) {
2189 if (level0 < 0)
2190 img_filter_2d_linear_repeat_POT(tgsi_sampler, s[j], t[j], p[j],
2191 samp->view->u.tex.first_level,
2192 samp->faces[j], &rgba[0][j]);
2193 else
2194 img_filter_2d_linear_repeat_POT(tgsi_sampler, s[j], t[j], p[j],
2195 samp->view->texture->last_level,
2196 samp->faces[j], &rgba[0][j]);
2197
2198 }
2199 else {
2200 float levelBlend = frac(lod[j]);
2201 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2202 int c;
2203
2204 img_filter_2d_linear_repeat_POT(tgsi_sampler, s[j], t[j], p[j], level0,
2205 samp->faces[j], &rgbax[0][0]);
2206 img_filter_2d_linear_repeat_POT(tgsi_sampler, s[j], t[j], p[j], level0+1,
2207 samp->faces[j], &rgbax[0][1]);
2208
2209 for (c = 0; c < TGSI_NUM_CHANNELS; c++)
2210 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
2211 }
2212 }
2213
2214 if (DEBUG_TEX) {
2215 print_sample_4(__FUNCTION__, rgba);
2216 }
2217 }
2218
2219
2220 /**
2221 * Do shadow/depth comparisons.
2222 */
2223 static void
2224 sample_compare(struct tgsi_sampler *tgsi_sampler,
2225 const float s[TGSI_QUAD_SIZE],
2226 const float t[TGSI_QUAD_SIZE],
2227 const float p[TGSI_QUAD_SIZE],
2228 const float c0[TGSI_QUAD_SIZE],
2229 const float c1[TGSI_QUAD_SIZE],
2230 enum tgsi_sampler_control control,
2231 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2232 {
2233 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2234 const struct pipe_sampler_state *sampler = samp->sampler;
2235 int j, k0, k1, k2, k3;
2236 float val;
2237 float pc0, pc1, pc2, pc3;
2238
2239 samp->mip_filter(tgsi_sampler, s, t, p, c0, c1, control, rgba);
2240
2241 /**
2242 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2243 * for 2D Array texture we need to use the 'c0' (aka Q).
2244 * When we sampled the depth texture, the depth value was put into all
2245 * RGBA channels. We look at the red channel here.
2246 */
2247
2248 if (samp->view->texture->target == PIPE_TEXTURE_2D_ARRAY ||
2249 samp->view->texture->target == PIPE_TEXTURE_CUBE) {
2250 pc0 = CLAMP(c0[0], 0.0F, 1.0F);
2251 pc1 = CLAMP(c0[1], 0.0F, 1.0F);
2252 pc2 = CLAMP(c0[2], 0.0F, 1.0F);
2253 pc3 = CLAMP(c0[3], 0.0F, 1.0F);
2254 } else if (samp->view->texture->target == PIPE_TEXTURE_CUBE_ARRAY) {
2255 pc0 = CLAMP(c1[0], 0.0F, 1.0F);
2256 pc1 = CLAMP(c1[1], 0.0F, 1.0F);
2257 pc2 = CLAMP(c1[2], 0.0F, 1.0F);
2258 pc3 = CLAMP(c1[3], 0.0F, 1.0F);
2259 } else {
2260 pc0 = CLAMP(p[0], 0.0F, 1.0F);
2261 pc1 = CLAMP(p[1], 0.0F, 1.0F);
2262 pc2 = CLAMP(p[2], 0.0F, 1.0F);
2263 pc3 = CLAMP(p[3], 0.0F, 1.0F);
2264 }
2265 /* compare four texcoords vs. four texture samples */
2266 switch (sampler->compare_func) {
2267 case PIPE_FUNC_LESS:
2268 k0 = pc0 < rgba[0][0];
2269 k1 = pc1 < rgba[0][1];
2270 k2 = pc2 < rgba[0][2];
2271 k3 = pc3 < rgba[0][3];
2272 break;
2273 case PIPE_FUNC_LEQUAL:
2274 k0 = pc0 <= rgba[0][0];
2275 k1 = pc1 <= rgba[0][1];
2276 k2 = pc2 <= rgba[0][2];
2277 k3 = pc3 <= rgba[0][3];
2278 break;
2279 case PIPE_FUNC_GREATER:
2280 k0 = pc0 > rgba[0][0];
2281 k1 = pc1 > rgba[0][1];
2282 k2 = pc2 > rgba[0][2];
2283 k3 = pc3 > rgba[0][3];
2284 break;
2285 case PIPE_FUNC_GEQUAL:
2286 k0 = pc0 >= rgba[0][0];
2287 k1 = pc1 >= rgba[0][1];
2288 k2 = pc2 >= rgba[0][2];
2289 k3 = pc3 >= rgba[0][3];
2290 break;
2291 case PIPE_FUNC_EQUAL:
2292 k0 = pc0 == rgba[0][0];
2293 k1 = pc1 == rgba[0][1];
2294 k2 = pc2 == rgba[0][2];
2295 k3 = pc3 == rgba[0][3];
2296 break;
2297 case PIPE_FUNC_NOTEQUAL:
2298 k0 = pc0 != rgba[0][0];
2299 k1 = pc1 != rgba[0][1];
2300 k2 = pc2 != rgba[0][2];
2301 k3 = pc3 != rgba[0][3];
2302 break;
2303 case PIPE_FUNC_ALWAYS:
2304 k0 = k1 = k2 = k3 = 1;
2305 break;
2306 case PIPE_FUNC_NEVER:
2307 k0 = k1 = k2 = k3 = 0;
2308 break;
2309 default:
2310 k0 = k1 = k2 = k3 = 0;
2311 assert(0);
2312 break;
2313 }
2314
2315 if (sampler->mag_img_filter == PIPE_TEX_FILTER_LINEAR) {
2316 /* convert four pass/fail values to an intensity in [0,1] */
2317 val = 0.25F * (k0 + k1 + k2 + k3);
2318
2319 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
2320 for (j = 0; j < 4; j++) {
2321 rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
2322 rgba[3][j] = 1.0F;
2323 }
2324 } else {
2325 for (j = 0; j < 4; j++) {
2326 rgba[0][j] = k0;
2327 rgba[1][j] = k1;
2328 rgba[2][j] = k2;
2329 rgba[3][j] = 1.0F;
2330 }
2331 }
2332 }
2333
2334
2335 /**
2336 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2337 * Put face info into the sampler faces[] array.
2338 */
2339 static void
2340 sample_cube(struct tgsi_sampler *tgsi_sampler,
2341 const float s[TGSI_QUAD_SIZE],
2342 const float t[TGSI_QUAD_SIZE],
2343 const float p[TGSI_QUAD_SIZE],
2344 const float c0[TGSI_QUAD_SIZE],
2345 const float c1[TGSI_QUAD_SIZE],
2346 enum tgsi_sampler_control control,
2347 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2348 {
2349 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2350 unsigned j;
2351 float ssss[4], tttt[4];
2352
2353 /* Not actually used, but the intermediate steps that do the
2354 * dereferencing don't know it.
2355 */
2356 static float pppp[4] = { 0, 0, 0, 0 };
2357
2358 pppp[0] = c0[0];
2359 pppp[1] = c0[1];
2360 pppp[2] = c0[2];
2361 pppp[3] = c0[3];
2362 /*
2363 major axis
2364 direction target sc tc ma
2365 ---------- ------------------------------- --- --- ---
2366 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2367 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2368 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2369 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2370 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2371 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2372 */
2373
2374 /* Choose the cube face and compute new s/t coords for the 2D face.
2375 *
2376 * Use the same cube face for all four pixels in the quad.
2377 *
2378 * This isn't ideal, but if we want to use a different cube face
2379 * per pixel in the quad, we'd have to also compute the per-face
2380 * LOD here too. That's because the four post-face-selection
2381 * texcoords are no longer related to each other (they're
2382 * per-face!) so we can't use subtraction to compute the partial
2383 * deriviates to compute the LOD. Doing so (near cube edges
2384 * anyway) gives us pretty much random values.
2385 */
2386 {
2387 /* use the average of the four pixel's texcoords to choose the face */
2388 const float rx = 0.25F * (s[0] + s[1] + s[2] + s[3]);
2389 const float ry = 0.25F * (t[0] + t[1] + t[2] + t[3]);
2390 const float rz = 0.25F * (p[0] + p[1] + p[2] + p[3]);
2391 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
2392
2393 if (arx >= ary && arx >= arz) {
2394 float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
2395 uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
2396 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2397 const float ima = -0.5F / fabsf(s[j]);
2398 ssss[j] = sign * p[j] * ima + 0.5F;
2399 tttt[j] = t[j] * ima + 0.5F;
2400 samp->faces[j] = face;
2401 }
2402 }
2403 else if (ary >= arx && ary >= arz) {
2404 float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
2405 uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
2406 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2407 const float ima = -0.5F / fabsf(t[j]);
2408 ssss[j] = -s[j] * ima + 0.5F;
2409 tttt[j] = sign * -p[j] * ima + 0.5F;
2410 samp->faces[j] = face;
2411 }
2412 }
2413 else {
2414 float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
2415 uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
2416 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2417 const float ima = -0.5F / fabsf(p[j]);
2418 ssss[j] = sign * -s[j] * ima + 0.5F;
2419 tttt[j] = t[j] * ima + 0.5F;
2420 samp->faces[j] = face;
2421 }
2422 }
2423 }
2424
2425 /* In our little pipeline, the compare stage is next. If compare
2426 * is not active, this will point somewhere deeper into the
2427 * pipeline, eg. to mip_filter or even img_filter.
2428 */
2429 samp->compare(tgsi_sampler, ssss, tttt, pppp, c0, c1, control, rgba);
2430 }
2431
2432
2433 static void
2434 do_swizzling(const struct sp_sampler_variant *samp,
2435 float in[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
2436 float out[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2437 {
2438 int j;
2439 const unsigned swizzle_r = samp->key.bits.swizzle_r;
2440 const unsigned swizzle_g = samp->key.bits.swizzle_g;
2441 const unsigned swizzle_b = samp->key.bits.swizzle_b;
2442 const unsigned swizzle_a = samp->key.bits.swizzle_a;
2443
2444 switch (swizzle_r) {
2445 case PIPE_SWIZZLE_ZERO:
2446 for (j = 0; j < 4; j++)
2447 out[0][j] = 0.0f;
2448 break;
2449 case PIPE_SWIZZLE_ONE:
2450 for (j = 0; j < 4; j++)
2451 out[0][j] = 1.0f;
2452 break;
2453 default:
2454 assert(swizzle_r < 4);
2455 for (j = 0; j < 4; j++)
2456 out[0][j] = in[swizzle_r][j];
2457 }
2458
2459 switch (swizzle_g) {
2460 case PIPE_SWIZZLE_ZERO:
2461 for (j = 0; j < 4; j++)
2462 out[1][j] = 0.0f;
2463 break;
2464 case PIPE_SWIZZLE_ONE:
2465 for (j = 0; j < 4; j++)
2466 out[1][j] = 1.0f;
2467 break;
2468 default:
2469 assert(swizzle_g < 4);
2470 for (j = 0; j < 4; j++)
2471 out[1][j] = in[swizzle_g][j];
2472 }
2473
2474 switch (swizzle_b) {
2475 case PIPE_SWIZZLE_ZERO:
2476 for (j = 0; j < 4; j++)
2477 out[2][j] = 0.0f;
2478 break;
2479 case PIPE_SWIZZLE_ONE:
2480 for (j = 0; j < 4; j++)
2481 out[2][j] = 1.0f;
2482 break;
2483 default:
2484 assert(swizzle_b < 4);
2485 for (j = 0; j < 4; j++)
2486 out[2][j] = in[swizzle_b][j];
2487 }
2488
2489 switch (swizzle_a) {
2490 case PIPE_SWIZZLE_ZERO:
2491 for (j = 0; j < 4; j++)
2492 out[3][j] = 0.0f;
2493 break;
2494 case PIPE_SWIZZLE_ONE:
2495 for (j = 0; j < 4; j++)
2496 out[3][j] = 1.0f;
2497 break;
2498 default:
2499 assert(swizzle_a < 4);
2500 for (j = 0; j < 4; j++)
2501 out[3][j] = in[swizzle_a][j];
2502 }
2503 }
2504
2505
2506 static void
2507 sample_swizzle(struct tgsi_sampler *tgsi_sampler,
2508 const float s[TGSI_QUAD_SIZE],
2509 const float t[TGSI_QUAD_SIZE],
2510 const float p[TGSI_QUAD_SIZE],
2511 const float c0[TGSI_QUAD_SIZE],
2512 const float c1[TGSI_QUAD_SIZE],
2513 enum tgsi_sampler_control control,
2514 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2515 {
2516 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2517 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2518
2519 samp->sample_target(tgsi_sampler, s, t, p, c0, c1, control, rgba_temp);
2520
2521 do_swizzling(samp, rgba_temp, rgba);
2522 }
2523
2524
2525 static wrap_nearest_func
2526 get_nearest_unorm_wrap(unsigned mode)
2527 {
2528 switch (mode) {
2529 case PIPE_TEX_WRAP_CLAMP:
2530 return wrap_nearest_unorm_clamp;
2531 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2532 return wrap_nearest_unorm_clamp_to_edge;
2533 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2534 return wrap_nearest_unorm_clamp_to_border;
2535 default:
2536 assert(0);
2537 return wrap_nearest_unorm_clamp;
2538 }
2539 }
2540
2541
2542 static wrap_nearest_func
2543 get_nearest_wrap(unsigned mode)
2544 {
2545 switch (mode) {
2546 case PIPE_TEX_WRAP_REPEAT:
2547 return wrap_nearest_repeat;
2548 case PIPE_TEX_WRAP_CLAMP:
2549 return wrap_nearest_clamp;
2550 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2551 return wrap_nearest_clamp_to_edge;
2552 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2553 return wrap_nearest_clamp_to_border;
2554 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2555 return wrap_nearest_mirror_repeat;
2556 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2557 return wrap_nearest_mirror_clamp;
2558 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2559 return wrap_nearest_mirror_clamp_to_edge;
2560 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2561 return wrap_nearest_mirror_clamp_to_border;
2562 default:
2563 assert(0);
2564 return wrap_nearest_repeat;
2565 }
2566 }
2567
2568
2569 static wrap_linear_func
2570 get_linear_unorm_wrap(unsigned mode)
2571 {
2572 switch (mode) {
2573 case PIPE_TEX_WRAP_CLAMP:
2574 return wrap_linear_unorm_clamp;
2575 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2576 return wrap_linear_unorm_clamp_to_edge;
2577 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2578 return wrap_linear_unorm_clamp_to_border;
2579 default:
2580 assert(0);
2581 return wrap_linear_unorm_clamp;
2582 }
2583 }
2584
2585
2586 static wrap_linear_func
2587 get_linear_wrap(unsigned mode)
2588 {
2589 switch (mode) {
2590 case PIPE_TEX_WRAP_REPEAT:
2591 return wrap_linear_repeat;
2592 case PIPE_TEX_WRAP_CLAMP:
2593 return wrap_linear_clamp;
2594 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2595 return wrap_linear_clamp_to_edge;
2596 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2597 return wrap_linear_clamp_to_border;
2598 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2599 return wrap_linear_mirror_repeat;
2600 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2601 return wrap_linear_mirror_clamp;
2602 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2603 return wrap_linear_mirror_clamp_to_edge;
2604 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2605 return wrap_linear_mirror_clamp_to_border;
2606 default:
2607 assert(0);
2608 return wrap_linear_repeat;
2609 }
2610 }
2611
2612
2613 /**
2614 * Is swizzling needed for the given state key?
2615 */
2616 static INLINE bool
2617 any_swizzle(union sp_sampler_key key)
2618 {
2619 return (key.bits.swizzle_r != PIPE_SWIZZLE_RED ||
2620 key.bits.swizzle_g != PIPE_SWIZZLE_GREEN ||
2621 key.bits.swizzle_b != PIPE_SWIZZLE_BLUE ||
2622 key.bits.swizzle_a != PIPE_SWIZZLE_ALPHA);
2623 }
2624
2625
2626 static compute_lambda_func
2627 get_lambda_func(const union sp_sampler_key key)
2628 {
2629 if (key.bits.processor == TGSI_PROCESSOR_VERTEX)
2630 return compute_lambda_vert;
2631
2632 switch (key.bits.target) {
2633 case PIPE_BUFFER:
2634 case PIPE_TEXTURE_1D:
2635 case PIPE_TEXTURE_1D_ARRAY:
2636 return compute_lambda_1d;
2637 case PIPE_TEXTURE_2D:
2638 case PIPE_TEXTURE_2D_ARRAY:
2639 case PIPE_TEXTURE_RECT:
2640 case PIPE_TEXTURE_CUBE:
2641 case PIPE_TEXTURE_CUBE_ARRAY:
2642 return compute_lambda_2d;
2643 case PIPE_TEXTURE_3D:
2644 return compute_lambda_3d;
2645 default:
2646 assert(0);
2647 return compute_lambda_1d;
2648 }
2649 }
2650
2651
2652 static img_filter_func
2653 get_img_filter(const union sp_sampler_key key,
2654 unsigned filter,
2655 const struct pipe_sampler_state *sampler)
2656 {
2657 switch (key.bits.target) {
2658 case PIPE_BUFFER:
2659 case PIPE_TEXTURE_1D:
2660 if (filter == PIPE_TEX_FILTER_NEAREST)
2661 return img_filter_1d_nearest;
2662 else
2663 return img_filter_1d_linear;
2664 break;
2665 case PIPE_TEXTURE_1D_ARRAY:
2666 if (filter == PIPE_TEX_FILTER_NEAREST)
2667 return img_filter_1d_array_nearest;
2668 else
2669 return img_filter_1d_array_linear;
2670 break;
2671 case PIPE_TEXTURE_2D:
2672 case PIPE_TEXTURE_RECT:
2673 /* Try for fast path:
2674 */
2675 if (key.bits.is_pot &&
2676 sampler->wrap_s == sampler->wrap_t &&
2677 sampler->normalized_coords)
2678 {
2679 switch (sampler->wrap_s) {
2680 case PIPE_TEX_WRAP_REPEAT:
2681 switch (filter) {
2682 case PIPE_TEX_FILTER_NEAREST:
2683 return img_filter_2d_nearest_repeat_POT;
2684 case PIPE_TEX_FILTER_LINEAR:
2685 return img_filter_2d_linear_repeat_POT;
2686 default:
2687 break;
2688 }
2689 break;
2690 case PIPE_TEX_WRAP_CLAMP:
2691 switch (filter) {
2692 case PIPE_TEX_FILTER_NEAREST:
2693 return img_filter_2d_nearest_clamp_POT;
2694 default:
2695 break;
2696 }
2697 }
2698 }
2699 /* Otherwise use default versions:
2700 */
2701 if (filter == PIPE_TEX_FILTER_NEAREST)
2702 return img_filter_2d_nearest;
2703 else
2704 return img_filter_2d_linear;
2705 break;
2706 case PIPE_TEXTURE_2D_ARRAY:
2707 if (filter == PIPE_TEX_FILTER_NEAREST)
2708 return img_filter_2d_array_nearest;
2709 else
2710 return img_filter_2d_array_linear;
2711 break;
2712 case PIPE_TEXTURE_CUBE:
2713 if (filter == PIPE_TEX_FILTER_NEAREST)
2714 return img_filter_cube_nearest;
2715 else
2716 return img_filter_cube_linear;
2717 break;
2718 case PIPE_TEXTURE_CUBE_ARRAY:
2719 if (filter == PIPE_TEX_FILTER_NEAREST)
2720 return img_filter_cube_array_nearest;
2721 else
2722 return img_filter_cube_array_linear;
2723 break;
2724 case PIPE_TEXTURE_3D:
2725 if (filter == PIPE_TEX_FILTER_NEAREST)
2726 return img_filter_3d_nearest;
2727 else
2728 return img_filter_3d_linear;
2729 break;
2730 default:
2731 assert(0);
2732 return img_filter_1d_nearest;
2733 }
2734 }
2735
2736
2737 /**
2738 * Bind the given texture object and texture cache to the sampler variant.
2739 */
2740 void
2741 sp_sampler_variant_bind_view( struct sp_sampler_variant *samp,
2742 struct softpipe_tex_tile_cache *tex_cache,
2743 const struct pipe_sampler_view *view )
2744 {
2745 const struct pipe_resource *texture = view->texture;
2746
2747 samp->view = view;
2748 samp->cache = tex_cache;
2749 samp->xpot = util_logbase2( texture->width0 );
2750 samp->ypot = util_logbase2( texture->height0 );
2751 }
2752
2753
2754 void
2755 sp_sampler_variant_destroy( struct sp_sampler_variant *samp )
2756 {
2757 FREE(samp);
2758 }
2759
2760
2761 static void
2762 sample_get_dims(struct tgsi_sampler *tgsi_sampler, int level,
2763 int dims[4])
2764 {
2765 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2766 const struct pipe_sampler_view *view = samp->view;
2767 const struct pipe_resource *texture = view->texture;
2768
2769 /* undefined according to EXT_gpu_program */
2770 level += view->u.tex.first_level;
2771 if (level > view->u.tex.last_level)
2772 return;
2773
2774 dims[0] = u_minify(texture->width0, level);
2775
2776 switch(texture->target) {
2777 case PIPE_TEXTURE_1D_ARRAY:
2778 dims[1] = texture->array_size;
2779 /* fallthrough */
2780 case PIPE_TEXTURE_1D:
2781 return;
2782 case PIPE_TEXTURE_2D_ARRAY:
2783 dims[2] = texture->array_size;
2784 /* fallthrough */
2785 case PIPE_TEXTURE_2D:
2786 case PIPE_TEXTURE_CUBE:
2787 case PIPE_TEXTURE_RECT:
2788 dims[1] = u_minify(texture->height0, level);
2789 return;
2790 case PIPE_TEXTURE_3D:
2791 dims[1] = u_minify(texture->height0, level);
2792 dims[2] = u_minify(texture->depth0, level);
2793 return;
2794 case PIPE_TEXTURE_CUBE_ARRAY:
2795 dims[1] = u_minify(texture->height0, level);
2796 dims[2] = texture->array_size / 6;
2797 break;
2798 case PIPE_BUFFER:
2799 dims[0] /= util_format_get_blocksize(view->format);
2800 return;
2801 default:
2802 assert(!"unexpected texture target in sample_get_dims()");
2803 return;
2804 }
2805 }
2806
2807 /**
2808 * This function is only used for getting unfiltered texels via the
2809 * TXF opcode. The GL spec says that out-of-bounds texel fetches
2810 * produce undefined results. Instead of crashing, lets just clamp
2811 * coords to the texture image size.
2812 */
2813 static void
2814 sample_get_texels(struct tgsi_sampler *tgsi_sampler,
2815 const int v_i[TGSI_QUAD_SIZE],
2816 const int v_j[TGSI_QUAD_SIZE],
2817 const int v_k[TGSI_QUAD_SIZE],
2818 const int lod[TGSI_QUAD_SIZE],
2819 const int8_t offset[3],
2820 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2821 {
2822 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2823 union tex_tile_address addr;
2824 const struct pipe_resource *texture = samp->view->texture;
2825 int j, c;
2826 const float *tx;
2827 const bool need_swizzle = any_swizzle(samp->key);
2828 int width, height, depth, layers;
2829
2830 addr.value = 0;
2831 /* TODO write a better test for LOD */
2832 addr.bits.level = lod[0];
2833
2834 width = u_minify(texture->width0, addr.bits.level);
2835 height = u_minify(texture->height0, addr.bits.level);
2836 depth = u_minify(texture->depth0, addr.bits.level);
2837 layers = texture->array_size;
2838
2839 switch(texture->target) {
2840 case PIPE_BUFFER:
2841 case PIPE_TEXTURE_1D:
2842 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2843 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2844 tx = get_texel_2d(samp, addr, x, 0);
2845 for (c = 0; c < 4; c++) {
2846 rgba[c][j] = tx[c];
2847 }
2848 }
2849 break;
2850 case PIPE_TEXTURE_1D_ARRAY:
2851 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2852 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2853 int y = CLAMP(v_j[j], 0, layers - 1);
2854 tx = get_texel_1d_array(samp, addr, x, y);
2855 for (c = 0; c < 4; c++) {
2856 rgba[c][j] = tx[c];
2857 }
2858 }
2859 break;
2860 case PIPE_TEXTURE_2D:
2861 case PIPE_TEXTURE_RECT:
2862 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2863 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2864 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
2865 tx = get_texel_2d(samp, addr, x, y);
2866 for (c = 0; c < 4; c++) {
2867 rgba[c][j] = tx[c];
2868 }
2869 }
2870 break;
2871 case PIPE_TEXTURE_2D_ARRAY:
2872 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2873 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2874 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
2875 int layer = CLAMP(v_k[j], 0, layers - 1);
2876 tx = get_texel_2d_array(samp, addr, x, y, layer);
2877 for (c = 0; c < 4; c++) {
2878 rgba[c][j] = tx[c];
2879 }
2880 }
2881 break;
2882 case PIPE_TEXTURE_3D:
2883 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2884 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2885 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
2886 int z = CLAMP(v_k[j] + offset[2], 0, depth - 1);
2887
2888 tx = get_texel_3d(samp, addr, x, y, z);
2889 for (c = 0; c < 4; c++) {
2890 rgba[c][j] = tx[c];
2891 }
2892 }
2893 break;
2894 case PIPE_TEXTURE_CUBE: /* TXF can't work on CUBE according to spec */
2895 default:
2896 assert(!"Unknown or CUBE texture type in TXF processing\n");
2897 break;
2898 }
2899
2900 if (need_swizzle) {
2901 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2902 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
2903 do_swizzling(samp, rgba_temp, rgba);
2904 }
2905 }
2906
2907
2908 /**
2909 * Create a sampler variant for a given set of non-orthogonal state.
2910 */
2911 struct sp_sampler_variant *
2912 sp_create_sampler_variant( const struct pipe_sampler_state *sampler,
2913 const union sp_sampler_key key )
2914 {
2915 struct sp_sampler_variant *samp = CALLOC_STRUCT(sp_sampler_variant);
2916 if (!samp)
2917 return NULL;
2918
2919 samp->sampler = sampler;
2920 samp->key = key;
2921
2922 /* Note that (for instance) linear_texcoord_s and
2923 * nearest_texcoord_s may be active at the same time, if the
2924 * sampler min_img_filter differs from its mag_img_filter.
2925 */
2926 if (sampler->normalized_coords) {
2927 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
2928 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
2929 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
2930
2931 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
2932 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
2933 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
2934 }
2935 else {
2936 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
2937 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
2938 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
2939
2940 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
2941 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
2942 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
2943 }
2944
2945 samp->compute_lambda = get_lambda_func( key );
2946
2947 samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
2948 samp->mag_img_filter = get_img_filter(key, sampler->mag_img_filter, sampler);
2949
2950 switch (sampler->min_mip_filter) {
2951 case PIPE_TEX_MIPFILTER_NONE:
2952 if (sampler->min_img_filter == sampler->mag_img_filter)
2953 samp->mip_filter = mip_filter_none_no_filter_select;
2954 else
2955 samp->mip_filter = mip_filter_none;
2956 break;
2957
2958 case PIPE_TEX_MIPFILTER_NEAREST:
2959 samp->mip_filter = mip_filter_nearest;
2960 break;
2961
2962 case PIPE_TEX_MIPFILTER_LINEAR:
2963 if (key.bits.is_pot &&
2964 key.bits.target == PIPE_TEXTURE_2D &&
2965 sampler->min_img_filter == sampler->mag_img_filter &&
2966 sampler->normalized_coords &&
2967 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
2968 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
2969 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR) {
2970 samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT;
2971 }
2972 else {
2973 samp->mip_filter = mip_filter_linear;
2974 }
2975
2976 /* Anisotropic filtering extension. */
2977 if (sampler->max_anisotropy > 1) {
2978 samp->mip_filter = mip_filter_linear_aniso;
2979
2980 /* Override min_img_filter:
2981 * min_img_filter needs to be set to NEAREST since we need to access
2982 * each texture pixel as it is and weight it later; using linear
2983 * filters will have incorrect results.
2984 * By setting the filter to NEAREST here, we can avoid calling the
2985 * generic img_filter_2d_nearest in the anisotropic filter function,
2986 * making it possible to use one of the accelerated implementations
2987 */
2988 samp->min_img_filter = get_img_filter(key, PIPE_TEX_FILTER_NEAREST, sampler);
2989
2990 /* on first access create the lookup table containing the filter weights. */
2991 if (!weightLut) {
2992 create_filter_table();
2993 }
2994 }
2995
2996 break;
2997 }
2998
2999 if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
3000 samp->compare = sample_compare;
3001 }
3002 else {
3003 /* Skip compare operation by promoting the mip_filter function
3004 * pointer:
3005 */
3006 samp->compare = samp->mip_filter;
3007 }
3008
3009 if (key.bits.target == PIPE_TEXTURE_CUBE || key.bits.target == PIPE_TEXTURE_CUBE_ARRAY) {
3010 samp->sample_target = sample_cube;
3011 }
3012 else {
3013 samp->faces[0] = 0;
3014 samp->faces[1] = 0;
3015 samp->faces[2] = 0;
3016 samp->faces[3] = 0;
3017
3018 /* Skip cube face determination by promoting the compare
3019 * function pointer:
3020 */
3021 samp->sample_target = samp->compare;
3022 }
3023
3024 if (any_swizzle(key)) {
3025 samp->base.get_samples = sample_swizzle;
3026 }
3027 else {
3028 samp->base.get_samples = samp->sample_target;
3029 }
3030
3031 samp->base.get_dims = sample_get_dims;
3032 samp->base.get_texel = sample_get_texels;
3033 return samp;
3034 }