softpipe: fix shadow sampling
[mesa.git] / src / gallium / drivers / softpipe / sp_tex_sample.c
1 /**************************************************************************
2 *
3 * Copyright 2007 VMware, Inc.
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 VMWARE 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 "util/u_inlines.h"
44 #include "sp_quad.h" /* only for #define QUAD_* tokens */
45 #include "sp_tex_sample.h"
46 #include "sp_texture.h"
47 #include "sp_tex_tile_cache.h"
48
49
50 /** Set to one to help debug texture sampling */
51 #define DEBUG_TEX 0
52
53
54 /*
55 * Return fractional part of 'f'. Used for computing interpolation weights.
56 * Need to be careful with negative values.
57 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
58 * of improperly weighted linear-filtered textures.
59 * The tests/texwrap.c demo is a good test.
60 */
61 static INLINE float
62 frac(float f)
63 {
64 return f - floorf(f);
65 }
66
67
68
69 /**
70 * Linear interpolation macro
71 */
72 static INLINE float
73 lerp(float a, float v0, float v1)
74 {
75 return v0 + a * (v1 - v0);
76 }
77
78
79 /**
80 * Do 2D/bilinear interpolation of float values.
81 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
82 * a and b are the horizontal and vertical interpolants.
83 * It's important that this function is inlined when compiled with
84 * optimization! If we find that's not true on some systems, convert
85 * to a macro.
86 */
87 static INLINE float
88 lerp_2d(float a, float b,
89 float v00, float v10, float v01, float v11)
90 {
91 const float temp0 = lerp(a, v00, v10);
92 const float temp1 = lerp(a, v01, v11);
93 return lerp(b, temp0, temp1);
94 }
95
96
97 /**
98 * As above, but 3D interpolation of 8 values.
99 */
100 static INLINE float
101 lerp_3d(float a, float b, float c,
102 float v000, float v100, float v010, float v110,
103 float v001, float v101, float v011, float v111)
104 {
105 const float temp0 = lerp_2d(a, b, v000, v100, v010, v110);
106 const float temp1 = lerp_2d(a, b, v001, v101, v011, v111);
107 return lerp(c, temp0, temp1);
108 }
109
110
111
112 /**
113 * Compute coord % size for repeat wrap modes.
114 * Note that if coord is negative, coord % size doesn't give the right
115 * value. To avoid that problem we add a large multiple of the size
116 * (rather than using a conditional).
117 */
118 static INLINE int
119 repeat(int coord, unsigned size)
120 {
121 return (coord + size * 1024) % size;
122 }
123
124
125 /**
126 * Apply texture coord wrapping mode and return integer texture indexes
127 * for a vector of four texcoords (S or T or P).
128 * \param wrapMode PIPE_TEX_WRAP_x
129 * \param s the incoming texcoords
130 * \param size the texture image size
131 * \param icoord returns the integer texcoords
132 */
133 static void
134 wrap_nearest_repeat(float s, unsigned size, int *icoord)
135 {
136 /* s limited to [0,1) */
137 /* i limited to [0,size-1] */
138 int i = util_ifloor(s * size);
139 *icoord = repeat(i, size);
140 }
141
142
143 static void
144 wrap_nearest_clamp(float s, unsigned size, int *icoord)
145 {
146 /* s limited to [0,1] */
147 /* i limited to [0,size-1] */
148 if (s <= 0.0F)
149 *icoord = 0;
150 else if (s >= 1.0F)
151 *icoord = size - 1;
152 else
153 *icoord = util_ifloor(s * size);
154 }
155
156
157 static void
158 wrap_nearest_clamp_to_edge(float s, unsigned size, int *icoord)
159 {
160 /* s limited to [min,max] */
161 /* i limited to [0, size-1] */
162 const float min = 1.0F / (2.0F * size);
163 const float max = 1.0F - min;
164 if (s < min)
165 *icoord = 0;
166 else if (s > max)
167 *icoord = size - 1;
168 else
169 *icoord = util_ifloor(s * size);
170 }
171
172
173 static void
174 wrap_nearest_clamp_to_border(float s, unsigned size, int *icoord)
175 {
176 /* s limited to [min,max] */
177 /* i limited to [-1, size] */
178 const float min = -1.0F / (2.0F * size);
179 const float max = 1.0F - min;
180 if (s <= min)
181 *icoord = -1;
182 else if (s >= max)
183 *icoord = size;
184 else
185 *icoord = util_ifloor(s * size);
186 }
187
188
189 static void
190 wrap_nearest_mirror_repeat(float s, unsigned size, int *icoord)
191 {
192 const float min = 1.0F / (2.0F * size);
193 const float max = 1.0F - min;
194 const int flr = util_ifloor(s);
195 float u = frac(s);
196 if (flr & 1)
197 u = 1.0F - u;
198 if (u < min)
199 *icoord = 0;
200 else if (u > max)
201 *icoord = size - 1;
202 else
203 *icoord = util_ifloor(u * size);
204 }
205
206
207 static void
208 wrap_nearest_mirror_clamp(float s, unsigned size, int *icoord)
209 {
210 /* s limited to [0,1] */
211 /* i limited to [0,size-1] */
212 const float u = fabsf(s);
213 if (u <= 0.0F)
214 *icoord = 0;
215 else if (u >= 1.0F)
216 *icoord = size - 1;
217 else
218 *icoord = util_ifloor(u * size);
219 }
220
221
222 static void
223 wrap_nearest_mirror_clamp_to_edge(float s, unsigned size, int *icoord)
224 {
225 /* s limited to [min,max] */
226 /* i limited to [0, size-1] */
227 const float min = 1.0F / (2.0F * size);
228 const float max = 1.0F - min;
229 const float u = fabsf(s);
230 if (u < min)
231 *icoord = 0;
232 else if (u > max)
233 *icoord = size - 1;
234 else
235 *icoord = util_ifloor(u * size);
236 }
237
238
239 static void
240 wrap_nearest_mirror_clamp_to_border(float s, unsigned size, int *icoord)
241 {
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 const float u = fabsf(s);
247 if (u < min)
248 *icoord = -1;
249 else if (u > max)
250 *icoord = size;
251 else
252 *icoord = util_ifloor(u * size);
253 }
254
255
256 /**
257 * Used to compute texel locations for linear sampling
258 * \param wrapMode PIPE_TEX_WRAP_x
259 * \param s the texcoord
260 * \param size the texture image size
261 * \param icoord0 returns first texture index
262 * \param icoord1 returns second texture index (usually icoord0 + 1)
263 * \param w returns blend factor/weight between texture indices
264 * \param icoord returns the computed integer texture coord
265 */
266 static void
267 wrap_linear_repeat(float s, unsigned size,
268 int *icoord0, int *icoord1, float *w)
269 {
270 float u = s * size - 0.5F;
271 *icoord0 = repeat(util_ifloor(u), size);
272 *icoord1 = repeat(*icoord0 + 1, size);
273 *w = frac(u);
274 }
275
276
277 static void
278 wrap_linear_clamp(float s, unsigned size,
279 int *icoord0, int *icoord1, float *w)
280 {
281 float u = CLAMP(s, 0.0F, 1.0F);
282 u = u * size - 0.5f;
283 *icoord0 = util_ifloor(u);
284 *icoord1 = *icoord0 + 1;
285 *w = frac(u);
286 }
287
288
289 static void
290 wrap_linear_clamp_to_edge(float s, unsigned size,
291 int *icoord0, int *icoord1, float *w)
292 {
293 float u = CLAMP(s, 0.0F, 1.0F);
294 u = u * size - 0.5f;
295 *icoord0 = util_ifloor(u);
296 *icoord1 = *icoord0 + 1;
297 if (*icoord0 < 0)
298 *icoord0 = 0;
299 if (*icoord1 >= (int) size)
300 *icoord1 = size - 1;
301 *w = frac(u);
302 }
303
304
305 static void
306 wrap_linear_clamp_to_border(float s, unsigned size,
307 int *icoord0, int *icoord1, float *w)
308 {
309 const float min = -1.0F / (2.0F * size);
310 const float max = 1.0F - min;
311 float u = CLAMP(s, min, max);
312 u = u * size - 0.5f;
313 *icoord0 = util_ifloor(u);
314 *icoord1 = *icoord0 + 1;
315 *w = frac(u);
316 }
317
318
319 static void
320 wrap_linear_mirror_repeat(float s, unsigned size,
321 int *icoord0, int *icoord1, float *w)
322 {
323 const int flr = util_ifloor(s);
324 float u = frac(s);
325 if (flr & 1)
326 u = 1.0F - u;
327 u = u * size - 0.5F;
328 *icoord0 = util_ifloor(u);
329 *icoord1 = *icoord0 + 1;
330 if (*icoord0 < 0)
331 *icoord0 = 0;
332 if (*icoord1 >= (int) size)
333 *icoord1 = size - 1;
334 *w = frac(u);
335 }
336
337
338 static void
339 wrap_linear_mirror_clamp(float s, unsigned size,
340 int *icoord0, int *icoord1, float *w)
341 {
342 float u = fabsf(s);
343 if (u >= 1.0F)
344 u = (float) size;
345 else
346 u *= size;
347 u -= 0.5F;
348 *icoord0 = util_ifloor(u);
349 *icoord1 = *icoord0 + 1;
350 *w = frac(u);
351 }
352
353
354 static void
355 wrap_linear_mirror_clamp_to_edge(float s, unsigned size,
356 int *icoord0, int *icoord1, float *w)
357 {
358 float u = fabsf(s);
359 if (u >= 1.0F)
360 u = (float) size;
361 else
362 u *= size;
363 u -= 0.5F;
364 *icoord0 = util_ifloor(u);
365 *icoord1 = *icoord0 + 1;
366 if (*icoord0 < 0)
367 *icoord0 = 0;
368 if (*icoord1 >= (int) size)
369 *icoord1 = size - 1;
370 *w = frac(u);
371 }
372
373
374 static void
375 wrap_linear_mirror_clamp_to_border(float s, unsigned size,
376 int *icoord0, int *icoord1, float *w)
377 {
378 const float min = -1.0F / (2.0F * size);
379 const float max = 1.0F - min;
380 float u = fabsf(s);
381 if (u <= min)
382 u = min * size;
383 else if (u >= max)
384 u = max * size;
385 else
386 u *= size;
387 u -= 0.5F;
388 *icoord0 = util_ifloor(u);
389 *icoord1 = *icoord0 + 1;
390 *w = frac(u);
391 }
392
393
394 /**
395 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
396 */
397 static void
398 wrap_nearest_unorm_clamp(float s, unsigned size, int *icoord)
399 {
400 int i = util_ifloor(s);
401 *icoord = CLAMP(i, 0, (int) size-1);
402 }
403
404
405 /**
406 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
407 */
408 static void
409 wrap_nearest_unorm_clamp_to_border(float s, unsigned size, int *icoord)
410 {
411 *icoord = util_ifloor( CLAMP(s, -0.5F, (float) size + 0.5F) );
412 }
413
414
415 /**
416 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
417 */
418 static void
419 wrap_nearest_unorm_clamp_to_edge(float s, unsigned size, int *icoord)
420 {
421 *icoord = util_ifloor( CLAMP(s, 0.5F, (float) size - 0.5F) );
422 }
423
424
425 /**
426 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
427 */
428 static void
429 wrap_linear_unorm_clamp(float s, unsigned size,
430 int *icoord0, int *icoord1, float *w)
431 {
432 /* Not exactly what the spec says, but it matches NVIDIA output */
433 float u = CLAMP(s - 0.5F, 0.0f, (float) size - 1.0f);
434 *icoord0 = util_ifloor(u);
435 *icoord1 = *icoord0 + 1;
436 *w = frac(u);
437 }
438
439
440 /**
441 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
442 */
443 static void
444 wrap_linear_unorm_clamp_to_border(float s, unsigned size,
445 int *icoord0, int *icoord1, float *w)
446 {
447 float u = CLAMP(s, -0.5F, (float) size + 0.5F);
448 u -= 0.5F;
449 *icoord0 = util_ifloor(u);
450 *icoord1 = *icoord0 + 1;
451 if (*icoord1 > (int) size - 1)
452 *icoord1 = size - 1;
453 *w = frac(u);
454 }
455
456
457 /**
458 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
459 */
460 static void
461 wrap_linear_unorm_clamp_to_edge(float s, unsigned size,
462 int *icoord0, int *icoord1, float *w)
463 {
464 float u = CLAMP(s, +0.5F, (float) size - 0.5F);
465 u -= 0.5F;
466 *icoord0 = util_ifloor(u);
467 *icoord1 = *icoord0 + 1;
468 if (*icoord1 > (int) size - 1)
469 *icoord1 = size - 1;
470 *w = frac(u);
471 }
472
473
474 /**
475 * Do coordinate to array index conversion. For array textures.
476 */
477 static INLINE void
478 wrap_array_layer(float coord, unsigned size, int *layer)
479 {
480 int c = util_ifloor(coord + 0.5F);
481 *layer = CLAMP(c, 0, (int) size - 1);
482 }
483
484
485 /**
486 * Examine the quad's texture coordinates to compute the partial
487 * derivatives w.r.t X and Y, then compute lambda (level of detail).
488 */
489 static float
490 compute_lambda_1d(const struct sp_sampler_view *sview,
491 const float s[TGSI_QUAD_SIZE],
492 const float t[TGSI_QUAD_SIZE],
493 const float p[TGSI_QUAD_SIZE])
494 {
495 const struct pipe_resource *texture = sview->base.texture;
496 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
497 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
498 float rho = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
499
500 return util_fast_log2(rho);
501 }
502
503
504 static float
505 compute_lambda_2d(const struct sp_sampler_view *sview,
506 const float s[TGSI_QUAD_SIZE],
507 const float t[TGSI_QUAD_SIZE],
508 const float p[TGSI_QUAD_SIZE])
509 {
510 const struct pipe_resource *texture = sview->base.texture;
511 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
512 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
513 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
514 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
515 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
516 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
517 float rho = MAX2(maxx, maxy);
518
519 return util_fast_log2(rho);
520 }
521
522
523 static float
524 compute_lambda_3d(const struct sp_sampler_view *sview,
525 const float s[TGSI_QUAD_SIZE],
526 const float t[TGSI_QUAD_SIZE],
527 const float p[TGSI_QUAD_SIZE])
528 {
529 const struct pipe_resource *texture = sview->base.texture;
530 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
531 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
532 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
533 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
534 float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
535 float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
536 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
537 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
538 float maxz = MAX2(dpdx, dpdy) * u_minify(texture->depth0, sview->base.u.tex.first_level);
539 float rho;
540
541 rho = MAX2(maxx, maxy);
542 rho = MAX2(rho, maxz);
543
544 return util_fast_log2(rho);
545 }
546
547
548 /**
549 * Compute lambda for a vertex texture sampler.
550 * Since there aren't derivatives to use, just return 0.
551 */
552 static float
553 compute_lambda_vert(const struct sp_sampler_view *sview,
554 const float s[TGSI_QUAD_SIZE],
555 const float t[TGSI_QUAD_SIZE],
556 const float p[TGSI_QUAD_SIZE])
557 {
558 return 0.0f;
559 }
560
561
562
563 /**
564 * Get a texel from a texture, using the texture tile cache.
565 *
566 * \param addr the template tex address containing cube, z, face info.
567 * \param x the x coord of texel within 2D image
568 * \param y the y coord of texel within 2D image
569 * \param rgba the quad to put the texel/color into
570 *
571 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
572 * sp_get_cached_tile_tex() function.
573 */
574
575
576
577
578 static INLINE const float *
579 get_texel_2d_no_border(const struct sp_sampler_view *sp_sview,
580 union tex_tile_address addr, int x, int y)
581 {
582 const struct softpipe_tex_cached_tile *tile;
583 addr.bits.x = x / TEX_TILE_SIZE;
584 addr.bits.y = y / TEX_TILE_SIZE;
585 y %= TEX_TILE_SIZE;
586 x %= TEX_TILE_SIZE;
587
588 tile = sp_get_cached_tile_tex(sp_sview->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_view *sp_sview,
596 const struct sp_sampler *sp_samp,
597 union tex_tile_address addr, int x, int y)
598 {
599 const struct pipe_resource *texture = sp_sview->base.texture;
600 unsigned level = addr.bits.level;
601
602 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
603 y < 0 || y >= (int) u_minify(texture->height0, level)) {
604 return sp_samp->base.border_color.f;
605 }
606 else {
607 return get_texel_2d_no_border( sp_sview, addr, x, y );
608 }
609 }
610
611
612 /*
613 * Here's the complete logic (HOLY CRAP) for finding next face and doing the
614 * corresponding coord wrapping, implemented by get_next_face,
615 * get_next_xcoord, get_next_ycoord.
616 * Read like that (first line):
617 * If face is +x and s coord is below zero, then
618 * new face is +z, new s is max , new t is old t
619 * (max is always cube size - 1).
620 *
621 * +x s- -> +z: s = max, t = t
622 * +x s+ -> -z: s = 0, t = t
623 * +x t- -> +y: s = max, t = max-s
624 * +x t+ -> -y: s = max, t = s
625 *
626 * -x s- -> -z: s = max, t = t
627 * -x s+ -> +z: s = 0, t = t
628 * -x t- -> +y: s = 0, t = s
629 * -x t+ -> -y: s = 0, t = max-s
630 *
631 * +y s- -> -x: s = t, t = 0
632 * +y s+ -> +x: s = max-t, t = 0
633 * +y t- -> -z: s = max-s, t = 0
634 * +y t+ -> +z: s = s, t = 0
635 *
636 * -y s- -> -x: s = max-t, t = max
637 * -y s+ -> +x: s = t, t = max
638 * -y t- -> +z: s = s, t = max
639 * -y t+ -> -z: s = max-s, t = max
640
641 * +z s- -> -x: s = max, t = t
642 * +z s+ -> +x: s = 0, t = t
643 * +z t- -> +y: s = s, t = max
644 * +z t+ -> -y: s = s, t = 0
645
646 * -z s- -> +x: s = max, t = t
647 * -z s+ -> -x: s = 0, t = t
648 * -z t- -> +y: s = max-s, t = 0
649 * -z t+ -> -y: s = max-s, t = max
650 */
651
652
653 /*
654 * seamless cubemap neighbour array.
655 * this array is used to find the adjacent face in each of 4 directions,
656 * left, right, up, down. (or -x, +x, -y, +y).
657 */
658 static const unsigned face_array[PIPE_TEX_FACE_MAX][4] = {
659 /* pos X first then neg X is Z different, Y the same */
660 /* PIPE_TEX_FACE_POS_X,*/
661 { PIPE_TEX_FACE_POS_Z, PIPE_TEX_FACE_NEG_Z,
662 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y },
663 /* PIPE_TEX_FACE_NEG_X */
664 { PIPE_TEX_FACE_NEG_Z, PIPE_TEX_FACE_POS_Z,
665 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y },
666
667 /* pos Y first then neg Y is X different, X the same */
668 /* PIPE_TEX_FACE_POS_Y */
669 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
670 PIPE_TEX_FACE_NEG_Z, PIPE_TEX_FACE_POS_Z },
671
672 /* PIPE_TEX_FACE_NEG_Y */
673 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
674 PIPE_TEX_FACE_POS_Z, PIPE_TEX_FACE_NEG_Z },
675
676 /* pos Z first then neg Y is X different, X the same */
677 /* PIPE_TEX_FACE_POS_Z */
678 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
679 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y },
680
681 /* PIPE_TEX_FACE_NEG_Z */
682 { PIPE_TEX_FACE_POS_X, PIPE_TEX_FACE_NEG_X,
683 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y }
684 };
685
686 static INLINE unsigned
687 get_next_face(unsigned face, int idx)
688 {
689 return face_array[face][idx];
690 }
691
692 /*
693 * return a new xcoord based on old face, old coords, cube size
694 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
695 */
696 static INLINE int
697 get_next_xcoord(unsigned face, unsigned fall_off_index, int max, int xc, int yc)
698 {
699 if ((face == 0 && fall_off_index != 1) ||
700 (face == 1 && fall_off_index == 0) ||
701 (face == 4 && fall_off_index == 0) ||
702 (face == 5 && fall_off_index == 0)) {
703 return max;
704 }
705 if ((face == 1 && fall_off_index != 0) ||
706 (face == 0 && fall_off_index == 1) ||
707 (face == 4 && fall_off_index == 1) ||
708 (face == 5 && fall_off_index == 1)) {
709 return 0;
710 }
711 if ((face == 4 && fall_off_index >= 2) ||
712 (face == 2 && fall_off_index == 3) ||
713 (face == 3 && fall_off_index == 2)) {
714 return xc;
715 }
716 if ((face == 5 && fall_off_index >= 2) ||
717 (face == 2 && fall_off_index == 2) ||
718 (face == 3 && fall_off_index == 3)) {
719 return max - xc;
720 }
721 if ((face == 2 && fall_off_index == 0) ||
722 (face == 3 && fall_off_index == 1)) {
723 return yc;
724 }
725 /* (face == 2 && fall_off_index == 1) ||
726 (face == 3 && fall_off_index == 0)) */
727 return max - yc;
728 }
729
730 /*
731 * return a new ycoord based on old face, old coords, cube size
732 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
733 */
734 static INLINE int
735 get_next_ycoord(unsigned face, unsigned fall_off_index, int max, int xc, int yc)
736 {
737 if ((fall_off_index <= 1) && (face <= 1 || face >= 4)) {
738 return yc;
739 }
740 if (face == 2 ||
741 (face == 4 && fall_off_index == 3) ||
742 (face == 5 && fall_off_index == 2)) {
743 return 0;
744 }
745 if (face == 3 ||
746 (face == 4 && fall_off_index == 2) ||
747 (face == 5 && fall_off_index == 3)) {
748 return max;
749 }
750 if ((face == 0 && fall_off_index == 3) ||
751 (face == 1 && fall_off_index == 2)) {
752 return xc;
753 }
754 /* (face == 0 && fall_off_index == 2) ||
755 (face == 1 && fall_off_index == 3) */
756 return max - xc;
757 }
758
759
760 static INLINE const float *
761 get_texel_cube_seamless(const struct sp_sampler_view *sp_sview,
762 union tex_tile_address addr, int x, int y,
763 float *corner)
764 {
765 const struct pipe_resource *texture = sp_sview->base.texture;
766 unsigned level = addr.bits.level;
767 unsigned face = addr.bits.face;
768 int new_x, new_y, max_x;
769
770 max_x = (int) u_minify(texture->width0, level);
771
772 assert(texture->width0 == texture->height0);
773 new_x = x;
774 new_y = y;
775
776 /* change the face */
777 if (x < 0) {
778 /*
779 * Cheat with corners. They are difficult and I believe because we don't get
780 * per-pixel faces we can actually have multiple corner texels per pixel,
781 * which screws things up majorly in any case (as the per spec behavior is
782 * to average the 3 remaining texels, which we might not have).
783 * Hence just make sure that the 2nd coord is clamped, will simply pick the
784 * sample which would have fallen off the x coord, but not y coord.
785 * So the filter weight of the samples will be wrong, but at least this
786 * ensures that only valid texels near the corner are used.
787 */
788 if (y < 0 || y >= max_x) {
789 y = CLAMP(y, 0, max_x - 1);
790 }
791 new_x = get_next_xcoord(face, 0, max_x -1, x, y);
792 new_y = get_next_ycoord(face, 0, max_x -1, x, y);
793 face = get_next_face(face, 0);
794 } else if (x >= max_x) {
795 if (y < 0 || y >= max_x) {
796 y = CLAMP(y, 0, max_x - 1);
797 }
798 new_x = get_next_xcoord(face, 1, max_x -1, x, y);
799 new_y = get_next_ycoord(face, 1, max_x -1, x, y);
800 face = get_next_face(face, 1);
801 } else if (y < 0) {
802 new_x = get_next_xcoord(face, 2, max_x -1, x, y);
803 new_y = get_next_ycoord(face, 2, max_x -1, x, y);
804 face = get_next_face(face, 2);
805 } else if (y >= max_x) {
806 new_x = get_next_xcoord(face, 3, max_x -1, x, y);
807 new_y = get_next_ycoord(face, 3, max_x -1, x, y);
808 face = get_next_face(face, 3);
809 }
810
811 addr.bits.face = face;
812 return get_texel_2d_no_border( sp_sview, addr, new_x, new_y );
813 }
814
815 /* Gather a quad of adjacent texels within a tile:
816 */
817 static INLINE void
818 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_view *sp_sview,
819 union tex_tile_address addr,
820 unsigned x, unsigned y,
821 const float *out[4])
822 {
823 const struct softpipe_tex_cached_tile *tile;
824
825 addr.bits.x = x / TEX_TILE_SIZE;
826 addr.bits.y = y / TEX_TILE_SIZE;
827 y %= TEX_TILE_SIZE;
828 x %= TEX_TILE_SIZE;
829
830 tile = sp_get_cached_tile_tex(sp_sview->cache, addr);
831
832 out[0] = &tile->data.color[y ][x ][0];
833 out[1] = &tile->data.color[y ][x+1][0];
834 out[2] = &tile->data.color[y+1][x ][0];
835 out[3] = &tile->data.color[y+1][x+1][0];
836 }
837
838
839 /* Gather a quad of potentially non-adjacent texels:
840 */
841 static INLINE void
842 get_texel_quad_2d_no_border(const struct sp_sampler_view *sp_sview,
843 union tex_tile_address addr,
844 int x0, int y0,
845 int x1, int y1,
846 const float *out[4])
847 {
848 out[0] = get_texel_2d_no_border( sp_sview, addr, x0, y0 );
849 out[1] = get_texel_2d_no_border( sp_sview, addr, x1, y0 );
850 out[2] = get_texel_2d_no_border( sp_sview, addr, x0, y1 );
851 out[3] = get_texel_2d_no_border( sp_sview, addr, x1, y1 );
852 }
853
854 /* Can involve a lot of unnecessary checks for border color:
855 */
856 static INLINE void
857 get_texel_quad_2d(const struct sp_sampler_view *sp_sview,
858 const struct sp_sampler *sp_samp,
859 union tex_tile_address addr,
860 int x0, int y0,
861 int x1, int y1,
862 const float *out[4])
863 {
864 out[0] = get_texel_2d( sp_sview, sp_samp, addr, x0, y0 );
865 out[1] = get_texel_2d( sp_sview, sp_samp, addr, x1, y0 );
866 out[3] = get_texel_2d( sp_sview, sp_samp, addr, x1, y1 );
867 out[2] = get_texel_2d( sp_sview, sp_samp, addr, x0, y1 );
868 }
869
870
871
872 /* 3d variants:
873 */
874 static INLINE const float *
875 get_texel_3d_no_border(const struct sp_sampler_view *sp_sview,
876 union tex_tile_address addr, int x, int y, int z)
877 {
878 const struct softpipe_tex_cached_tile *tile;
879
880 addr.bits.x = x / TEX_TILE_SIZE;
881 addr.bits.y = y / TEX_TILE_SIZE;
882 addr.bits.z = z;
883 y %= TEX_TILE_SIZE;
884 x %= TEX_TILE_SIZE;
885
886 tile = sp_get_cached_tile_tex(sp_sview->cache, addr);
887
888 return &tile->data.color[y][x][0];
889 }
890
891
892 static INLINE const float *
893 get_texel_3d(const struct sp_sampler_view *sp_sview,
894 const struct sp_sampler *sp_samp,
895 union tex_tile_address addr, int x, int y, int z)
896 {
897 const struct pipe_resource *texture = sp_sview->base.texture;
898 unsigned level = addr.bits.level;
899
900 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
901 y < 0 || y >= (int) u_minify(texture->height0, level) ||
902 z < 0 || z >= (int) u_minify(texture->depth0, level)) {
903 return sp_samp->base.border_color.f;
904 }
905 else {
906 return get_texel_3d_no_border( sp_sview, addr, x, y, z );
907 }
908 }
909
910
911 /* Get texel pointer for 1D array texture */
912 static INLINE const float *
913 get_texel_1d_array(const struct sp_sampler_view *sp_sview,
914 const struct sp_sampler *sp_samp,
915 union tex_tile_address addr, int x, int y)
916 {
917 const struct pipe_resource *texture = sp_sview->base.texture;
918 unsigned level = addr.bits.level;
919
920 if (x < 0 || x >= (int) u_minify(texture->width0, level)) {
921 return sp_samp->base.border_color.f;
922 }
923 else {
924 return get_texel_2d_no_border(sp_sview, addr, x, y);
925 }
926 }
927
928
929 /* Get texel pointer for 2D array texture */
930 static INLINE const float *
931 get_texel_2d_array(const struct sp_sampler_view *sp_sview,
932 const struct sp_sampler *sp_samp,
933 union tex_tile_address addr, int x, int y, int layer)
934 {
935 const struct pipe_resource *texture = sp_sview->base.texture;
936 unsigned level = addr.bits.level;
937
938 assert(layer < (int) texture->array_size);
939 assert(layer >= 0);
940
941 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
942 y < 0 || y >= (int) u_minify(texture->height0, level)) {
943 return sp_samp->base.border_color.f;
944 }
945 else {
946 return get_texel_3d_no_border(sp_sview, addr, x, y, layer);
947 }
948 }
949
950
951 /* Get texel pointer for cube array texture */
952 static INLINE const float *
953 get_texel_cube_array(const struct sp_sampler_view *sp_sview,
954 const struct sp_sampler *sp_samp,
955 union tex_tile_address addr, int x, int y, int layer)
956 {
957 const struct pipe_resource *texture = sp_sview->base.texture;
958 unsigned level = addr.bits.level;
959
960 assert(layer < (int) texture->array_size);
961 assert(layer >= 0);
962
963 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
964 y < 0 || y >= (int) u_minify(texture->height0, level)) {
965 return sp_samp->base.border_color.f;
966 }
967 else {
968 return get_texel_3d_no_border(sp_sview, addr, x, y, layer);
969 }
970 }
971 /**
972 * Given the logbase2 of a mipmap's base level size and a mipmap level,
973 * return the size (in texels) of that mipmap level.
974 * For example, if level[0].width = 256 then base_pot will be 8.
975 * If level = 2, then we'll return 64 (the width at level=2).
976 * Return 1 if level > base_pot.
977 */
978 static INLINE unsigned
979 pot_level_size(unsigned base_pot, unsigned level)
980 {
981 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
982 }
983
984
985 static void
986 print_sample(const char *function, const float *rgba)
987 {
988 debug_printf("%s %g %g %g %g\n",
989 function,
990 rgba[0], rgba[TGSI_NUM_CHANNELS], rgba[2*TGSI_NUM_CHANNELS], rgba[3*TGSI_NUM_CHANNELS]);
991 }
992
993
994 static void
995 print_sample_4(const char *function, float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
996 {
997 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
998 function,
999 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
1000 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
1001 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
1002 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
1003 }
1004
1005
1006 /* Some image-filter fastpaths:
1007 */
1008 static INLINE void
1009 img_filter_2d_linear_repeat_POT(struct sp_sampler_view *sp_sview,
1010 struct sp_sampler *sp_samp,
1011 float s,
1012 float t,
1013 float p,
1014 unsigned level,
1015 unsigned face_id,
1016 float *rgba)
1017 {
1018 unsigned xpot = pot_level_size(sp_sview->xpot, level);
1019 unsigned ypot = pot_level_size(sp_sview->ypot, level);
1020 int xmax = (xpot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
1021 int ymax = (ypot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
1022 union tex_tile_address addr;
1023 int c;
1024
1025 float u = s * xpot - 0.5F;
1026 float v = t * ypot - 0.5F;
1027
1028 int uflr = util_ifloor(u);
1029 int vflr = util_ifloor(v);
1030
1031 float xw = u - (float)uflr;
1032 float yw = v - (float)vflr;
1033
1034 int x0 = uflr & (xpot - 1);
1035 int y0 = vflr & (ypot - 1);
1036
1037 const float *tx[4];
1038
1039 addr.value = 0;
1040 addr.bits.level = level;
1041
1042 /* Can we fetch all four at once:
1043 */
1044 if (x0 < xmax && y0 < ymax) {
1045 get_texel_quad_2d_no_border_single_tile(sp_sview, addr, x0, y0, tx);
1046 }
1047 else {
1048 unsigned x1 = (x0 + 1) & (xpot - 1);
1049 unsigned y1 = (y0 + 1) & (ypot - 1);
1050 get_texel_quad_2d_no_border(sp_sview, addr, x0, y0, x1, y1, tx);
1051 }
1052
1053 /* interpolate R, G, B, A */
1054 for (c = 0; c < TGSI_QUAD_SIZE; c++) {
1055 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1056 tx[0][c], tx[1][c],
1057 tx[2][c], tx[3][c]);
1058 }
1059
1060 if (DEBUG_TEX) {
1061 print_sample(__FUNCTION__, rgba);
1062 }
1063 }
1064
1065
1066 static INLINE void
1067 img_filter_2d_nearest_repeat_POT(struct sp_sampler_view *sp_sview,
1068 struct sp_sampler *sp_samp,
1069 float s,
1070 float t,
1071 float p,
1072 unsigned level,
1073 unsigned face_id,
1074 float rgba[TGSI_QUAD_SIZE])
1075 {
1076 unsigned xpot = pot_level_size(sp_sview->xpot, level);
1077 unsigned ypot = pot_level_size(sp_sview->ypot, level);
1078 const float *out;
1079 union tex_tile_address addr;
1080 int c;
1081
1082 float u = s * xpot;
1083 float v = t * ypot;
1084
1085 int uflr = util_ifloor(u);
1086 int vflr = util_ifloor(v);
1087
1088 int x0 = uflr & (xpot - 1);
1089 int y0 = vflr & (ypot - 1);
1090
1091 addr.value = 0;
1092 addr.bits.level = level;
1093
1094 out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
1095 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1096 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1097
1098 if (DEBUG_TEX) {
1099 print_sample(__FUNCTION__, rgba);
1100 }
1101 }
1102
1103
1104 static INLINE void
1105 img_filter_2d_nearest_clamp_POT(struct sp_sampler_view *sp_sview,
1106 struct sp_sampler *sp_samp,
1107 float s,
1108 float t,
1109 float p,
1110 unsigned level,
1111 unsigned face_id,
1112 float rgba[TGSI_QUAD_SIZE])
1113 {
1114 unsigned xpot = pot_level_size(sp_sview->xpot, level);
1115 unsigned ypot = pot_level_size(sp_sview->ypot, level);
1116 union tex_tile_address addr;
1117 int c;
1118
1119 float u = s * xpot;
1120 float v = t * ypot;
1121
1122 int x0, y0;
1123 const float *out;
1124
1125 addr.value = 0;
1126 addr.bits.level = level;
1127
1128 x0 = util_ifloor(u);
1129 if (x0 < 0)
1130 x0 = 0;
1131 else if (x0 > (int) xpot - 1)
1132 x0 = xpot - 1;
1133
1134 y0 = util_ifloor(v);
1135 if (y0 < 0)
1136 y0 = 0;
1137 else if (y0 > (int) ypot - 1)
1138 y0 = ypot - 1;
1139
1140 out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
1141 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1142 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1143
1144 if (DEBUG_TEX) {
1145 print_sample(__FUNCTION__, rgba);
1146 }
1147 }
1148
1149
1150 static void
1151 img_filter_1d_nearest(struct sp_sampler_view *sp_sview,
1152 struct sp_sampler *sp_samp,
1153 float s,
1154 float t,
1155 float p,
1156 unsigned level,
1157 unsigned face_id,
1158 float rgba[TGSI_QUAD_SIZE])
1159 {
1160 const struct pipe_resource *texture = sp_sview->base.texture;
1161 int width;
1162 int x;
1163 union tex_tile_address addr;
1164 const float *out;
1165 int c;
1166
1167 width = u_minify(texture->width0, level);
1168
1169 assert(width > 0);
1170
1171 addr.value = 0;
1172 addr.bits.level = level;
1173
1174 sp_samp->nearest_texcoord_s(s, width, &x);
1175
1176 out = get_texel_2d(sp_sview, sp_samp, addr, x, 0);
1177 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1178 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1179
1180 if (DEBUG_TEX) {
1181 print_sample(__FUNCTION__, rgba);
1182 }
1183 }
1184
1185
1186 static void
1187 img_filter_1d_array_nearest(struct sp_sampler_view *sp_sview,
1188 struct sp_sampler *sp_samp,
1189 float s,
1190 float t,
1191 float p,
1192 unsigned level,
1193 unsigned face_id,
1194 float *rgba)
1195 {
1196 const struct pipe_resource *texture = sp_sview->base.texture;
1197 int width;
1198 int x, layer;
1199 union tex_tile_address addr;
1200 const float *out;
1201 int c;
1202
1203 width = u_minify(texture->width0, level);
1204
1205 assert(width > 0);
1206
1207 addr.value = 0;
1208 addr.bits.level = level;
1209
1210 sp_samp->nearest_texcoord_s(s, width, &x);
1211 wrap_array_layer(t, texture->array_size, &layer);
1212
1213 out = get_texel_1d_array(sp_sview, sp_samp, addr, x, layer);
1214 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1215 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1216
1217 if (DEBUG_TEX) {
1218 print_sample(__FUNCTION__, rgba);
1219 }
1220 }
1221
1222
1223 static void
1224 img_filter_2d_nearest(struct sp_sampler_view *sp_sview,
1225 struct sp_sampler *sp_samp,
1226 float s,
1227 float t,
1228 float p,
1229 unsigned level,
1230 unsigned face_id,
1231 float *rgba)
1232 {
1233 const struct pipe_resource *texture = sp_sview->base.texture;
1234 int width, height;
1235 int x, y;
1236 union tex_tile_address addr;
1237 const float *out;
1238 int c;
1239
1240 width = u_minify(texture->width0, level);
1241 height = u_minify(texture->height0, level);
1242
1243 assert(width > 0);
1244 assert(height > 0);
1245
1246 addr.value = 0;
1247 addr.bits.level = level;
1248
1249 sp_samp->nearest_texcoord_s(s, width, &x);
1250 sp_samp->nearest_texcoord_t(t, height, &y);
1251
1252 out = get_texel_2d(sp_sview, sp_samp, addr, x, y);
1253 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1254 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1255
1256 if (DEBUG_TEX) {
1257 print_sample(__FUNCTION__, rgba);
1258 }
1259 }
1260
1261
1262 static void
1263 img_filter_2d_array_nearest(struct sp_sampler_view *sp_sview,
1264 struct sp_sampler *sp_samp,
1265 float s,
1266 float t,
1267 float p,
1268 unsigned level,
1269 unsigned face_id,
1270 float *rgba)
1271 {
1272 const struct pipe_resource *texture = sp_sview->base.texture;
1273 int width, height;
1274 int x, y, layer;
1275 union tex_tile_address addr;
1276 const float *out;
1277 int c;
1278
1279 width = u_minify(texture->width0, level);
1280 height = u_minify(texture->height0, level);
1281
1282 assert(width > 0);
1283 assert(height > 0);
1284
1285 addr.value = 0;
1286 addr.bits.level = level;
1287
1288 sp_samp->nearest_texcoord_s(s, width, &x);
1289 sp_samp->nearest_texcoord_t(t, height, &y);
1290 wrap_array_layer(p, texture->array_size, &layer);
1291
1292 out = get_texel_2d_array(sp_sview, sp_samp, addr, x, y, layer);
1293 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1294 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1295
1296 if (DEBUG_TEX) {
1297 print_sample(__FUNCTION__, rgba);
1298 }
1299 }
1300
1301
1302 static INLINE union tex_tile_address
1303 face(union tex_tile_address addr, unsigned face )
1304 {
1305 addr.bits.face = face;
1306 return addr;
1307 }
1308
1309
1310 static void
1311 img_filter_cube_nearest(struct sp_sampler_view *sp_sview,
1312 struct sp_sampler *sp_samp,
1313 float s,
1314 float t,
1315 float p,
1316 unsigned level,
1317 unsigned face_id,
1318 float *rgba)
1319 {
1320 const struct pipe_resource *texture = sp_sview->base.texture;
1321 int width, height;
1322 int x, y;
1323 union tex_tile_address addr;
1324 const float *out;
1325 int c;
1326
1327 width = u_minify(texture->width0, level);
1328 height = u_minify(texture->height0, level);
1329
1330 assert(width > 0);
1331 assert(height > 0);
1332
1333 addr.value = 0;
1334 addr.bits.level = level;
1335
1336 /*
1337 * If NEAREST filtering is done within a miplevel, always apply wrap
1338 * mode CLAMP_TO_EDGE.
1339 */
1340 if (sp_samp->base.seamless_cube_map) {
1341 wrap_nearest_clamp_to_edge(s, width, &x);
1342 wrap_nearest_clamp_to_edge(t, height, &y);
1343 } else {
1344 /* Would probably make sense to ignore mode and just do edge clamp */
1345 sp_samp->nearest_texcoord_s(s, width, &x);
1346 sp_samp->nearest_texcoord_t(t, height, &y);
1347 }
1348
1349 out = get_texel_2d(sp_sview, sp_samp, face(addr, face_id), x, y);
1350 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1351 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1352
1353 if (DEBUG_TEX) {
1354 print_sample(__FUNCTION__, rgba);
1355 }
1356 }
1357
1358 static void
1359 img_filter_cube_array_nearest(struct sp_sampler_view *sp_sview,
1360 struct sp_sampler *sp_samp,
1361 float s,
1362 float t,
1363 float p,
1364 unsigned level,
1365 unsigned face_id,
1366 float *rgba)
1367 {
1368 const struct pipe_resource *texture = sp_sview->base.texture;
1369 int width, height;
1370 int x, y, layer;
1371 union tex_tile_address addr;
1372 const float *out;
1373 int c;
1374
1375 width = u_minify(texture->width0, level);
1376 height = u_minify(texture->height0, level);
1377
1378 assert(width > 0);
1379 assert(height > 0);
1380
1381 addr.value = 0;
1382 addr.bits.level = level;
1383
1384 sp_samp->nearest_texcoord_s(s, width, &x);
1385 sp_samp->nearest_texcoord_t(t, height, &y);
1386 wrap_array_layer(p, texture->array_size, &layer);
1387
1388 out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layer * 6 + face_id);
1389 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1390 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1391
1392 if (DEBUG_TEX) {
1393 print_sample(__FUNCTION__, rgba);
1394 }
1395 }
1396
1397 static void
1398 img_filter_3d_nearest(struct sp_sampler_view *sp_sview,
1399 struct sp_sampler *sp_samp,
1400 float s,
1401 float t,
1402 float p,
1403 unsigned level,
1404 unsigned face_id,
1405 float *rgba)
1406 {
1407 const struct pipe_resource *texture = sp_sview->base.texture;
1408 int width, height, depth;
1409 int x, y, z;
1410 union tex_tile_address addr;
1411 const float *out;
1412 int c;
1413
1414 width = u_minify(texture->width0, level);
1415 height = u_minify(texture->height0, level);
1416 depth = u_minify(texture->depth0, level);
1417
1418 assert(width > 0);
1419 assert(height > 0);
1420 assert(depth > 0);
1421
1422 sp_samp->nearest_texcoord_s(s, width, &x);
1423 sp_samp->nearest_texcoord_t(t, height, &y);
1424 sp_samp->nearest_texcoord_p(p, depth, &z);
1425
1426 addr.value = 0;
1427 addr.bits.level = level;
1428
1429 out = get_texel_3d(sp_sview, sp_samp, addr, x, y, z);
1430 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1431 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1432 }
1433
1434
1435 static void
1436 img_filter_1d_linear(struct sp_sampler_view *sp_sview,
1437 struct sp_sampler *sp_samp,
1438 float s,
1439 float t,
1440 float p,
1441 unsigned level,
1442 unsigned face_id,
1443 float *rgba)
1444 {
1445 const struct pipe_resource *texture = sp_sview->base.texture;
1446 int width;
1447 int x0, x1;
1448 float xw; /* weights */
1449 union tex_tile_address addr;
1450 const float *tx0, *tx1;
1451 int c;
1452
1453 width = u_minify(texture->width0, level);
1454
1455 assert(width > 0);
1456
1457 addr.value = 0;
1458 addr.bits.level = level;
1459
1460 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1461
1462 tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, 0);
1463 tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, 0);
1464
1465 /* interpolate R, G, B, A */
1466 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1467 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1468 }
1469
1470
1471 static void
1472 img_filter_1d_array_linear(struct sp_sampler_view *sp_sview,
1473 struct sp_sampler *sp_samp,
1474 float s,
1475 float t,
1476 float p,
1477 unsigned level,
1478 unsigned face_id,
1479 float *rgba)
1480 {
1481 const struct pipe_resource *texture = sp_sview->base.texture;
1482 int width;
1483 int x0, x1, layer;
1484 float xw; /* weights */
1485 union tex_tile_address addr;
1486 const float *tx0, *tx1;
1487 int c;
1488
1489 width = u_minify(texture->width0, level);
1490
1491 assert(width > 0);
1492
1493 addr.value = 0;
1494 addr.bits.level = level;
1495
1496 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1497 wrap_array_layer(t, texture->array_size, &layer);
1498
1499 tx0 = get_texel_1d_array(sp_sview, sp_samp, addr, x0, layer);
1500 tx1 = get_texel_1d_array(sp_sview, sp_samp, addr, x1, layer);
1501
1502 /* interpolate R, G, B, A */
1503 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1504 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1505 }
1506
1507
1508 static void
1509 img_filter_2d_linear(struct sp_sampler_view *sp_sview,
1510 struct sp_sampler *sp_samp,
1511 float s,
1512 float t,
1513 float p,
1514 unsigned level,
1515 unsigned face_id,
1516 float *rgba)
1517 {
1518 const struct pipe_resource *texture = sp_sview->base.texture;
1519 int width, height;
1520 int x0, y0, x1, y1;
1521 float xw, yw; /* weights */
1522 union tex_tile_address addr;
1523 const float *tx0, *tx1, *tx2, *tx3;
1524 int c;
1525
1526 width = u_minify(texture->width0, level);
1527 height = u_minify(texture->height0, level);
1528
1529 assert(width > 0);
1530 assert(height > 0);
1531
1532 addr.value = 0;
1533 addr.bits.level = level;
1534
1535 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1536 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1537
1538 tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, y0);
1539 tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, y0);
1540 tx2 = get_texel_2d(sp_sview, sp_samp, addr, x0, y1);
1541 tx3 = get_texel_2d(sp_sview, sp_samp, addr, x1, y1);
1542
1543 /* interpolate R, G, B, A */
1544 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1545 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1546 tx0[c], tx1[c],
1547 tx2[c], tx3[c]);
1548 }
1549
1550
1551 static void
1552 img_filter_2d_array_linear(struct sp_sampler_view *sp_sview,
1553 struct sp_sampler *sp_samp,
1554 float s,
1555 float t,
1556 float p,
1557 unsigned level,
1558 unsigned face_id,
1559 float *rgba)
1560 {
1561 const struct pipe_resource *texture = sp_sview->base.texture;
1562 int width, height;
1563 int x0, y0, x1, y1, layer;
1564 float xw, yw; /* weights */
1565 union tex_tile_address addr;
1566 const float *tx0, *tx1, *tx2, *tx3;
1567 int c;
1568
1569 width = u_minify(texture->width0, level);
1570 height = u_minify(texture->height0, level);
1571
1572 assert(width > 0);
1573 assert(height > 0);
1574
1575 addr.value = 0;
1576 addr.bits.level = level;
1577
1578 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1579 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1580 wrap_array_layer(p, texture->array_size, &layer);
1581
1582 tx0 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y0, layer);
1583 tx1 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y0, layer);
1584 tx2 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y1, layer);
1585 tx3 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y1, layer);
1586
1587 /* interpolate R, G, B, A */
1588 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1589 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1590 tx0[c], tx1[c],
1591 tx2[c], tx3[c]);
1592 }
1593
1594
1595 static void
1596 img_filter_cube_linear(struct sp_sampler_view *sp_sview,
1597 struct sp_sampler *sp_samp,
1598 float s,
1599 float t,
1600 float p,
1601 unsigned level,
1602 unsigned face_id,
1603 float *rgba)
1604 {
1605 const struct pipe_resource *texture = sp_sview->base.texture;
1606 int width, height;
1607 int x0, y0, x1, y1;
1608 float xw, yw; /* weights */
1609 union tex_tile_address addr, addrj;
1610 const float *tx0, *tx1, *tx2, *tx3;
1611 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
1612 corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1613 int c;
1614
1615 width = u_minify(texture->width0, level);
1616 height = u_minify(texture->height0, level);
1617
1618 assert(width > 0);
1619 assert(height > 0);
1620
1621 addr.value = 0;
1622 addr.bits.level = level;
1623
1624 /*
1625 * For seamless if LINEAR filtering is done within a miplevel,
1626 * always apply wrap mode CLAMP_TO_BORDER.
1627 */
1628 if (sp_samp->base.seamless_cube_map) {
1629 /* Note this is a bit overkill, actual clamping is not required */
1630 wrap_linear_clamp_to_border(s, width, &x0, &x1, &xw);
1631 wrap_linear_clamp_to_border(t, height, &y0, &y1, &yw);
1632 } else {
1633 /* Would probably make sense to ignore mode and just do edge clamp */
1634 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1635 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1636 }
1637
1638 addrj = face(addr, face_id);
1639
1640 if (sp_samp->base.seamless_cube_map) {
1641 tx0 = get_texel_cube_seamless(sp_sview, addrj, x0, y0, corner0);
1642 tx1 = get_texel_cube_seamless(sp_sview, addrj, x1, y0, corner1);
1643 tx2 = get_texel_cube_seamless(sp_sview, addrj, x0, y1, corner2);
1644 tx3 = get_texel_cube_seamless(sp_sview, addrj, x1, y1, corner3);
1645 } else {
1646 tx0 = get_texel_2d(sp_sview, sp_samp, addrj, x0, y0);
1647 tx1 = get_texel_2d(sp_sview, sp_samp, addrj, x1, y0);
1648 tx2 = get_texel_2d(sp_sview, sp_samp, addrj, x0, y1);
1649 tx3 = get_texel_2d(sp_sview, sp_samp, addrj, x1, y1);
1650 }
1651 /* interpolate R, G, B, A */
1652 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1653 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1654 tx0[c], tx1[c],
1655 tx2[c], tx3[c]);
1656 }
1657
1658
1659 static void
1660 img_filter_cube_array_linear(struct sp_sampler_view *sp_sview,
1661 struct sp_sampler *sp_samp,
1662 float s,
1663 float t,
1664 float p,
1665 unsigned level,
1666 unsigned face_id,
1667 float *rgba)
1668 {
1669 const struct pipe_resource *texture = sp_sview->base.texture;
1670 int width, height;
1671 int x0, y0, x1, y1, layer;
1672 float xw, yw; /* weights */
1673 union tex_tile_address addr;
1674 const float *tx0, *tx1, *tx2, *tx3;
1675 int c;
1676
1677 width = u_minify(texture->width0, level);
1678 height = u_minify(texture->height0, level);
1679
1680 assert(width > 0);
1681 assert(height > 0);
1682
1683 addr.value = 0;
1684 addr.bits.level = level;
1685
1686 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1687 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1688 wrap_array_layer(p, texture->array_size, &layer);
1689
1690 tx0 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer * 6 + face_id);
1691 tx1 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer * 6 + face_id);
1692 tx2 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer * 6 + face_id);
1693 tx3 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer * 6 + face_id);
1694
1695 /* interpolate R, G, B, A */
1696 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1697 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1698 tx0[c], tx1[c],
1699 tx2[c], tx3[c]);
1700 }
1701
1702 static void
1703 img_filter_3d_linear(struct sp_sampler_view *sp_sview,
1704 struct sp_sampler *sp_samp,
1705 float s,
1706 float t,
1707 float p,
1708 unsigned level,
1709 unsigned face_id,
1710 float *rgba)
1711 {
1712 const struct pipe_resource *texture = sp_sview->base.texture;
1713 int width, height, depth;
1714 int x0, x1, y0, y1, z0, z1;
1715 float xw, yw, zw; /* interpolation weights */
1716 union tex_tile_address addr;
1717 const float *tx00, *tx01, *tx02, *tx03, *tx10, *tx11, *tx12, *tx13;
1718 int c;
1719
1720 width = u_minify(texture->width0, level);
1721 height = u_minify(texture->height0, level);
1722 depth = u_minify(texture->depth0, level);
1723
1724 addr.value = 0;
1725 addr.bits.level = level;
1726
1727 assert(width > 0);
1728 assert(height > 0);
1729 assert(depth > 0);
1730
1731 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1732 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1733 sp_samp->linear_texcoord_p(p, depth, &z0, &z1, &zw);
1734
1735
1736 tx00 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z0);
1737 tx01 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z0);
1738 tx02 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z0);
1739 tx03 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z0);
1740
1741 tx10 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z1);
1742 tx11 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z1);
1743 tx12 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z1);
1744 tx13 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z1);
1745
1746 /* interpolate R, G, B, A */
1747 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1748 rgba[TGSI_NUM_CHANNELS*c] = lerp_3d(xw, yw, zw,
1749 tx00[c], tx01[c],
1750 tx02[c], tx03[c],
1751 tx10[c], tx11[c],
1752 tx12[c], tx13[c]);
1753 }
1754
1755
1756 /* Calculate level of detail for every fragment,
1757 * with lambda already computed.
1758 * Note that lambda has already been biased by global LOD bias.
1759 * \param biased_lambda per-quad lambda.
1760 * \param lod_in per-fragment lod_bias or explicit_lod.
1761 * \param lod returns the per-fragment lod.
1762 */
1763 static INLINE void
1764 compute_lod(const struct pipe_sampler_state *sampler,
1765 enum tgsi_sampler_control control,
1766 const float biased_lambda,
1767 const float lod_in[TGSI_QUAD_SIZE],
1768 float lod[TGSI_QUAD_SIZE])
1769 {
1770 float min_lod = sampler->min_lod;
1771 float max_lod = sampler->max_lod;
1772 uint i;
1773
1774 switch (control) {
1775 case tgsi_sampler_lod_none:
1776 case tgsi_sampler_lod_zero:
1777 /* XXX FIXME */
1778 case tgsi_sampler_derivs_explicit:
1779 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(biased_lambda, min_lod, max_lod);
1780 break;
1781 case tgsi_sampler_lod_bias:
1782 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1783 lod[i] = biased_lambda + lod_in[i];
1784 lod[i] = CLAMP(lod[i], min_lod, max_lod);
1785 }
1786 break;
1787 case tgsi_sampler_lod_explicit:
1788 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1789 lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
1790 }
1791 break;
1792 default:
1793 assert(0);
1794 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1795 }
1796 }
1797
1798
1799 /* Calculate level of detail for every fragment.
1800 * \param lod_in per-fragment lod_bias or explicit_lod.
1801 * \param lod results per-fragment lod.
1802 */
1803 static INLINE void
1804 compute_lambda_lod(struct sp_sampler_view *sp_sview,
1805 struct sp_sampler *sp_samp,
1806 const float s[TGSI_QUAD_SIZE],
1807 const float t[TGSI_QUAD_SIZE],
1808 const float p[TGSI_QUAD_SIZE],
1809 const float lod_in[TGSI_QUAD_SIZE],
1810 enum tgsi_sampler_control control,
1811 float lod[TGSI_QUAD_SIZE])
1812 {
1813 const struct pipe_sampler_state *sampler = &sp_samp->base;
1814 float lod_bias = sampler->lod_bias;
1815 float min_lod = sampler->min_lod;
1816 float max_lod = sampler->max_lod;
1817 float lambda;
1818 uint i;
1819
1820 switch (control) {
1821 case tgsi_sampler_lod_none:
1822 /* XXX FIXME */
1823 case tgsi_sampler_derivs_explicit:
1824 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1825 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lambda, min_lod, max_lod);
1826 break;
1827 case tgsi_sampler_lod_bias:
1828 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1829 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1830 lod[i] = lambda + lod_in[i];
1831 lod[i] = CLAMP(lod[i], min_lod, max_lod);
1832 }
1833 break;
1834 case tgsi_sampler_lod_explicit:
1835 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1836 lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
1837 }
1838 break;
1839 case tgsi_sampler_lod_zero:
1840 /* this is all static state in the sampler really need clamp here? */
1841 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lod_bias, min_lod, max_lod);
1842 break;
1843 default:
1844 assert(0);
1845 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1846 }
1847 }
1848
1849
1850 static void
1851 mip_filter_linear(struct sp_sampler_view *sp_sview,
1852 struct sp_sampler *sp_samp,
1853 img_filter_func min_filter,
1854 img_filter_func mag_filter,
1855 const float s[TGSI_QUAD_SIZE],
1856 const float t[TGSI_QUAD_SIZE],
1857 const float p[TGSI_QUAD_SIZE],
1858 const float c0[TGSI_QUAD_SIZE],
1859 const float lod_in[TGSI_QUAD_SIZE],
1860 enum tgsi_sampler_control control,
1861 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1862 {
1863 const struct pipe_resource *texture = sp_sview->base.texture;
1864 int j;
1865 float lod[TGSI_QUAD_SIZE];
1866
1867 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1868
1869 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1870 int level0 = sp_sview->base.u.tex.first_level + (int)lod[j];
1871
1872 if (lod[j] < 0.0)
1873 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1874 sp_sview->base.u.tex.first_level,
1875 sp_sview->faces[j], &rgba[0][j]);
1876
1877 else if (level0 >= (int) texture->last_level)
1878 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], texture->last_level,
1879 sp_sview->faces[j], &rgba[0][j]);
1880
1881 else {
1882 float levelBlend = frac(lod[j]);
1883 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
1884 int c;
1885
1886 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level0,
1887 sp_sview->faces[j], &rgbax[0][0]);
1888 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level0+1,
1889 sp_sview->faces[j], &rgbax[0][1]);
1890
1891 for (c = 0; c < 4; c++) {
1892 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
1893 }
1894 }
1895 }
1896
1897 if (DEBUG_TEX) {
1898 print_sample_4(__FUNCTION__, rgba);
1899 }
1900 }
1901
1902
1903 /**
1904 * Compute nearest mipmap level from texcoords.
1905 * Then sample the texture level for four elements of a quad.
1906 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1907 */
1908 static void
1909 mip_filter_nearest(struct sp_sampler_view *sp_sview,
1910 struct sp_sampler *sp_samp,
1911 img_filter_func min_filter,
1912 img_filter_func mag_filter,
1913 const float s[TGSI_QUAD_SIZE],
1914 const float t[TGSI_QUAD_SIZE],
1915 const float p[TGSI_QUAD_SIZE],
1916 const float c0[TGSI_QUAD_SIZE],
1917 const float lod_in[TGSI_QUAD_SIZE],
1918 enum tgsi_sampler_control control,
1919 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1920 {
1921 const struct pipe_resource *texture = sp_sview->base.texture;
1922 float lod[TGSI_QUAD_SIZE];
1923 int j;
1924
1925 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1926
1927 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1928 if (lod[j] < 0.0)
1929 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1930 sp_sview->base.u.tex.first_level,
1931 sp_sview->faces[j], &rgba[0][j]);
1932 else {
1933 int level = sp_sview->base.u.tex.first_level + (int)(lod[j] + 0.5F);
1934 level = MIN2(level, (int)texture->last_level);
1935 min_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1936 level, sp_sview->faces[j], &rgba[0][j]);
1937 }
1938 }
1939
1940 if (DEBUG_TEX) {
1941 print_sample_4(__FUNCTION__, rgba);
1942 }
1943 }
1944
1945
1946 static void
1947 mip_filter_none(struct sp_sampler_view *sp_sview,
1948 struct sp_sampler *sp_samp,
1949 img_filter_func min_filter,
1950 img_filter_func mag_filter,
1951 const float s[TGSI_QUAD_SIZE],
1952 const float t[TGSI_QUAD_SIZE],
1953 const float p[TGSI_QUAD_SIZE],
1954 const float c0[TGSI_QUAD_SIZE],
1955 const float lod_in[TGSI_QUAD_SIZE],
1956 enum tgsi_sampler_control control,
1957 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1958 {
1959 float lod[TGSI_QUAD_SIZE];
1960 int j;
1961
1962 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1963
1964 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1965 if (lod[j] < 0.0) {
1966 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1967 sp_sview->base.u.tex.first_level,
1968 sp_sview->faces[j], &rgba[0][j]);
1969 }
1970 else {
1971 min_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1972 sp_sview->base.u.tex.first_level,
1973 sp_sview->faces[j], &rgba[0][j]);
1974 }
1975 }
1976 }
1977
1978
1979 static void
1980 mip_filter_none_no_filter_select(struct sp_sampler_view *sp_sview,
1981 struct sp_sampler *sp_samp,
1982 img_filter_func min_filter,
1983 img_filter_func mag_filter,
1984 const float s[TGSI_QUAD_SIZE],
1985 const float t[TGSI_QUAD_SIZE],
1986 const float p[TGSI_QUAD_SIZE],
1987 const float c0[TGSI_QUAD_SIZE],
1988 const float lod_in[TGSI_QUAD_SIZE],
1989 enum tgsi_sampler_control control,
1990 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1991 {
1992 int j;
1993
1994 for (j = 0; j < TGSI_QUAD_SIZE; j++)
1995 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1996 sp_sview->base.u.tex.first_level,
1997 sp_sview->faces[j], &rgba[0][j]);
1998 }
1999
2000
2001 /* For anisotropic filtering */
2002 #define WEIGHT_LUT_SIZE 1024
2003
2004 static float *weightLut = NULL;
2005
2006 /**
2007 * Creates the look-up table used to speed-up EWA sampling
2008 */
2009 static void
2010 create_filter_table(void)
2011 {
2012 unsigned i;
2013 if (!weightLut) {
2014 weightLut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
2015
2016 for (i = 0; i < WEIGHT_LUT_SIZE; ++i) {
2017 float alpha = 2;
2018 float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
2019 float weight = (float) exp(-alpha * r2);
2020 weightLut[i] = weight;
2021 }
2022 }
2023 }
2024
2025
2026 /**
2027 * Elliptical weighted average (EWA) filter for producing high quality
2028 * anisotropic filtered results.
2029 * Based on the Higher Quality Elliptical Weighted Average Filter
2030 * published by Paul S. Heckbert in his Master's Thesis
2031 * "Fundamentals of Texture Mapping and Image Warping" (1989)
2032 */
2033 static void
2034 img_filter_2d_ewa(struct sp_sampler_view *sp_sview,
2035 struct sp_sampler *sp_samp,
2036 img_filter_func min_filter,
2037 img_filter_func mag_filter,
2038 const float s[TGSI_QUAD_SIZE],
2039 const float t[TGSI_QUAD_SIZE],
2040 const float p[TGSI_QUAD_SIZE],
2041 unsigned level,
2042 const float dudx, const float dvdx,
2043 const float dudy, const float dvdy,
2044 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2045 {
2046 const struct pipe_resource *texture = sp_sview->base.texture;
2047
2048 // ??? Won't the image filters blow up if level is negative?
2049 unsigned level0 = level > 0 ? level : 0;
2050 float scaling = 1.0f / (1 << level0);
2051 int width = u_minify(texture->width0, level0);
2052 int height = u_minify(texture->height0, level0);
2053
2054 float ux = dudx * scaling;
2055 float vx = dvdx * scaling;
2056 float uy = dudy * scaling;
2057 float vy = dvdy * scaling;
2058
2059 /* compute ellipse coefficients to bound the region:
2060 * A*x*x + B*x*y + C*y*y = F.
2061 */
2062 float A = vx*vx+vy*vy+1;
2063 float B = -2*(ux*vx+uy*vy);
2064 float C = ux*ux+uy*uy+1;
2065 float F = A*C-B*B/4.0f;
2066
2067 /* check if it is an ellipse */
2068 /* ASSERT(F > 0.0); */
2069
2070 /* Compute the ellipse's (u,v) bounding box in texture space */
2071 float d = -B*B+4.0f*C*A;
2072 float box_u = 2.0f / d * sqrtf(d*C*F); /* box_u -> half of bbox with */
2073 float box_v = 2.0f / d * sqrtf(A*d*F); /* box_v -> half of bbox height */
2074
2075 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2076 float s_buffer[TGSI_QUAD_SIZE];
2077 float t_buffer[TGSI_QUAD_SIZE];
2078 float weight_buffer[TGSI_QUAD_SIZE];
2079 unsigned buffer_next;
2080 int j;
2081 float den; /* = 0.0F; */
2082 float ddq;
2083 float U; /* = u0 - tex_u; */
2084 int v;
2085
2086 /* Scale ellipse formula to directly index the Filter Lookup Table.
2087 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2088 */
2089 double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
2090 A *= formScale;
2091 B *= formScale;
2092 C *= formScale;
2093 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2094
2095 /* For each quad, the du and dx values are the same and so the ellipse is
2096 * also the same. Note that texel/image access can only be performed using
2097 * a quad, i.e. it is not possible to get the pixel value for a single
2098 * tex coord. In order to have a better performance, the access is buffered
2099 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2100 * full, then the pixel values are read from the image.
2101 */
2102 ddq = 2 * A;
2103
2104 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2105 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2106 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2107 * value, q, is less than F, we're inside the ellipse
2108 */
2109 float tex_u = -0.5F + s[j] * texture->width0 * scaling;
2110 float tex_v = -0.5F + t[j] * texture->height0 * scaling;
2111
2112 int u0 = (int) floorf(tex_u - box_u);
2113 int u1 = (int) ceilf(tex_u + box_u);
2114 int v0 = (int) floorf(tex_v - box_v);
2115 int v1 = (int) ceilf(tex_v + box_v);
2116
2117 float num[4] = {0.0F, 0.0F, 0.0F, 0.0F};
2118 buffer_next = 0;
2119 den = 0;
2120 U = u0 - tex_u;
2121 for (v = v0; v <= v1; ++v) {
2122 float V = v - tex_v;
2123 float dq = A * (2 * U + 1) + B * V;
2124 float q = (C * V + B * U) * V + A * U * U;
2125
2126 int u;
2127 for (u = u0; u <= u1; ++u) {
2128 /* Note that the ellipse has been pre-scaled so F =
2129 * WEIGHT_LUT_SIZE - 1
2130 */
2131 if (q < WEIGHT_LUT_SIZE) {
2132 /* as a LUT is used, q must never be negative;
2133 * should not happen, though
2134 */
2135 const int qClamped = q >= 0.0F ? q : 0;
2136 float weight = weightLut[qClamped];
2137
2138 weight_buffer[buffer_next] = weight;
2139 s_buffer[buffer_next] = u / ((float) width);
2140 t_buffer[buffer_next] = v / ((float) height);
2141
2142 buffer_next++;
2143 if (buffer_next == TGSI_QUAD_SIZE) {
2144 /* 4 texel coords are in the buffer -> read it now */
2145 unsigned jj;
2146 /* it is assumed that samp->min_img_filter is set to
2147 * img_filter_2d_nearest or one of the
2148 * accelerated img_filter_2d_nearest_XXX functions.
2149 */
2150 for (jj = 0; jj < buffer_next; jj++) {
2151 min_filter(sp_sview, sp_samp, s_buffer[jj], t_buffer[jj], p[jj],
2152 level, sp_sview->faces[j], &rgba_temp[0][jj]);
2153 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2154 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2155 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2156 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2157 }
2158
2159 buffer_next = 0;
2160 }
2161
2162 den += weight;
2163 }
2164 q += dq;
2165 dq += ddq;
2166 }
2167 }
2168
2169 /* if the tex coord buffer contains unread values, we will read
2170 * them now.
2171 */
2172 if (buffer_next > 0) {
2173 unsigned jj;
2174 /* it is assumed that samp->min_img_filter is set to
2175 * img_filter_2d_nearest or one of the
2176 * accelerated img_filter_2d_nearest_XXX functions.
2177 */
2178 for (jj = 0; jj < buffer_next; jj++) {
2179 min_filter(sp_sview, sp_samp, s_buffer[jj], t_buffer[jj], p[jj],
2180 level, sp_sview->faces[j], &rgba_temp[0][jj]);
2181 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2182 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2183 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2184 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2185 }
2186 }
2187
2188 if (den <= 0.0F) {
2189 /* Reaching this place would mean that no pixels intersected
2190 * the ellipse. This should never happen because the filter
2191 * we use always intersects at least one pixel.
2192 */
2193
2194 /*rgba[0]=0;
2195 rgba[1]=0;
2196 rgba[2]=0;
2197 rgba[3]=0;*/
2198 /* not enough pixels in resampling, resort to direct interpolation */
2199 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level,
2200 sp_sview->faces[j], &rgba_temp[0][j]);
2201 den = 1;
2202 num[0] = rgba_temp[0][j];
2203 num[1] = rgba_temp[1][j];
2204 num[2] = rgba_temp[2][j];
2205 num[3] = rgba_temp[3][j];
2206 }
2207
2208 rgba[0][j] = num[0] / den;
2209 rgba[1][j] = num[1] / den;
2210 rgba[2][j] = num[2] / den;
2211 rgba[3][j] = num[3] / den;
2212 }
2213 }
2214
2215
2216 /**
2217 * Sample 2D texture using an anisotropic filter.
2218 */
2219 static void
2220 mip_filter_linear_aniso(struct sp_sampler_view *sp_sview,
2221 struct sp_sampler *sp_samp,
2222 img_filter_func min_filter,
2223 img_filter_func mag_filter,
2224 const float s[TGSI_QUAD_SIZE],
2225 const float t[TGSI_QUAD_SIZE],
2226 const float p[TGSI_QUAD_SIZE],
2227 const float c0[TGSI_QUAD_SIZE],
2228 const float lod_in[TGSI_QUAD_SIZE],
2229 enum tgsi_sampler_control control,
2230 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2231 {
2232 const struct pipe_resource *texture = sp_sview->base.texture;
2233 int level0;
2234 float lambda;
2235 float lod[TGSI_QUAD_SIZE];
2236
2237 float s_to_u = u_minify(texture->width0, sp_sview->base.u.tex.first_level);
2238 float t_to_v = u_minify(texture->height0, sp_sview->base.u.tex.first_level);
2239 float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2240 float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2241 float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2242 float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2243
2244 if (control == tgsi_sampler_lod_bias ||
2245 control == tgsi_sampler_lod_none ||
2246 /* XXX FIXME */
2247 control == tgsi_sampler_derivs_explicit) {
2248 /* note: instead of working with Px and Py, we will use the
2249 * squared length instead, to avoid sqrt.
2250 */
2251 float Px2 = dudx * dudx + dvdx * dvdx;
2252 float Py2 = dudy * dudy + dvdy * dvdy;
2253
2254 float Pmax2;
2255 float Pmin2;
2256 float e;
2257 const float maxEccentricity = sp_samp->base.max_anisotropy * sp_samp->base.max_anisotropy;
2258
2259 if (Px2 < Py2) {
2260 Pmax2 = Py2;
2261 Pmin2 = Px2;
2262 }
2263 else {
2264 Pmax2 = Px2;
2265 Pmin2 = Py2;
2266 }
2267
2268 /* if the eccentricity of the ellipse is too big, scale up the shorter
2269 * of the two vectors to limit the maximum amount of work per pixel
2270 */
2271 e = Pmax2 / Pmin2;
2272 if (e > maxEccentricity) {
2273 /* float s=e / maxEccentricity;
2274 minor[0] *= s;
2275 minor[1] *= s;
2276 Pmin2 *= s; */
2277 Pmin2 = Pmax2 / maxEccentricity;
2278 }
2279
2280 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2281 * this since 0.5*log(x) = log(sqrt(x))
2282 */
2283 lambda = 0.5F * util_fast_log2(Pmin2) + sp_samp->base.lod_bias;
2284 compute_lod(&sp_samp->base, control, lambda, lod_in, lod);
2285 }
2286 else {
2287 assert(control == tgsi_sampler_lod_explicit ||
2288 control == tgsi_sampler_lod_zero);
2289 compute_lod(&sp_samp->base, control, sp_samp->base.lod_bias, lod_in, lod);
2290 }
2291
2292 /* XXX: Take into account all lod values.
2293 */
2294 lambda = lod[0];
2295 level0 = sp_sview->base.u.tex.first_level + (int)lambda;
2296
2297 /* If the ellipse covers the whole image, we can
2298 * simply return the average of the whole image.
2299 */
2300 if (level0 >= (int) texture->last_level) {
2301 int j;
2302 for (j = 0; j < TGSI_QUAD_SIZE; j++)
2303 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], texture->last_level,
2304 sp_sview->faces[j], &rgba[0][j]);
2305 }
2306 else {
2307 /* don't bother interpolating between multiple LODs; it doesn't
2308 * seem to be worth the extra running time.
2309 */
2310 img_filter_2d_ewa(sp_sview, sp_samp, min_filter, mag_filter,
2311 s, t, p, level0,
2312 dudx, dvdx, dudy, dvdy, rgba);
2313 }
2314
2315 if (DEBUG_TEX) {
2316 print_sample_4(__FUNCTION__, rgba);
2317 }
2318 }
2319
2320
2321 /**
2322 * Specialized version of mip_filter_linear with hard-wired calls to
2323 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2324 */
2325 static void
2326 mip_filter_linear_2d_linear_repeat_POT(
2327 struct sp_sampler_view *sp_sview,
2328 struct sp_sampler *sp_samp,
2329 img_filter_func min_filter,
2330 img_filter_func mag_filter,
2331 const float s[TGSI_QUAD_SIZE],
2332 const float t[TGSI_QUAD_SIZE],
2333 const float p[TGSI_QUAD_SIZE],
2334 const float c0[TGSI_QUAD_SIZE],
2335 const float lod_in[TGSI_QUAD_SIZE],
2336 enum tgsi_sampler_control control,
2337 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2338 {
2339 const struct pipe_resource *texture = sp_sview->base.texture;
2340 int j;
2341 float lod[TGSI_QUAD_SIZE];
2342
2343 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
2344
2345 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2346 int level0 = sp_sview->base.u.tex.first_level + (int)lod[j];
2347
2348 /* Catches both negative and large values of level0:
2349 */
2350 if ((unsigned)level0 >= texture->last_level) {
2351 if (level0 < 0)
2352 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j],
2353 sp_sview->base.u.tex.first_level,
2354 sp_sview->faces[j], &rgba[0][j]);
2355 else
2356 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j],
2357 sp_sview->base.texture->last_level,
2358 sp_sview->faces[j], &rgba[0][j]);
2359
2360 }
2361 else {
2362 float levelBlend = frac(lod[j]);
2363 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2364 int c;
2365
2366 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j], level0,
2367 sp_sview->faces[j], &rgbax[0][0]);
2368 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j], level0+1,
2369 sp_sview->faces[j], &rgbax[0][1]);
2370
2371 for (c = 0; c < TGSI_NUM_CHANNELS; c++)
2372 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
2373 }
2374 }
2375
2376 if (DEBUG_TEX) {
2377 print_sample_4(__FUNCTION__, rgba);
2378 }
2379 }
2380
2381
2382 /**
2383 * Do shadow/depth comparisons.
2384 */
2385 static void
2386 sample_compare(struct sp_sampler_view *sp_sview,
2387 struct sp_sampler *sp_samp,
2388 const float s[TGSI_QUAD_SIZE],
2389 const float t[TGSI_QUAD_SIZE],
2390 const float p[TGSI_QUAD_SIZE],
2391 const float c0[TGSI_QUAD_SIZE],
2392 const float c1[TGSI_QUAD_SIZE],
2393 enum tgsi_sampler_control control,
2394 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2395 {
2396 const struct pipe_sampler_state *sampler = &sp_samp->base;
2397 int j;
2398 int k[4];
2399 float pc[4];
2400 const struct util_format_description *format_desc;
2401 unsigned chan_type;
2402
2403 /**
2404 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2405 * for 2D Array texture we need to use the 'c0' (aka Q).
2406 * When we sampled the depth texture, the depth value was put into all
2407 * RGBA channels. We look at the red channel here.
2408 */
2409
2410 if (sp_sview->base.texture->target == PIPE_TEXTURE_2D_ARRAY ||
2411 sp_sview->base.texture->target == PIPE_TEXTURE_CUBE) {
2412 pc[0] = c0[0];
2413 pc[1] = c0[1];
2414 pc[2] = c0[2];
2415 pc[3] = c0[3];
2416 } else if (sp_sview->base.texture->target == PIPE_TEXTURE_CUBE_ARRAY) {
2417 pc[0] = c1[0];
2418 pc[1] = c1[1];
2419 pc[2] = c1[2];
2420 pc[3] = c1[3];
2421 } else {
2422 pc[0] = p[0];
2423 pc[1] = p[1];
2424 pc[2] = p[2];
2425 pc[3] = p[3];
2426 }
2427
2428 format_desc = util_format_description(sp_sview->base.format);
2429 /* not entirely sure we couldn't end up with non-valid swizzle here */
2430 chan_type = format_desc->swizzle[0] <= UTIL_FORMAT_SWIZZLE_W ?
2431 format_desc->channel[format_desc->swizzle[0]].type :
2432 UTIL_FORMAT_TYPE_FLOAT;
2433 if (chan_type != UTIL_FORMAT_TYPE_FLOAT) {
2434 /*
2435 * clamping is a result of conversion to texture format, hence
2436 * doesn't happen with floats. Technically also should do comparison
2437 * in texture format (quantization!).
2438 */
2439 pc[0] = CLAMP(pc[0], 0.0F, 1.0F);
2440 pc[1] = CLAMP(pc[1], 0.0F, 1.0F);
2441 pc[2] = CLAMP(pc[2], 0.0F, 1.0F);
2442 pc[3] = CLAMP(pc[3], 0.0F, 1.0F);
2443 }
2444
2445 /* compare four texcoords vs. four texture samples */
2446 switch (sampler->compare_func) {
2447 case PIPE_FUNC_LESS:
2448 k[0] = pc[0] < rgba[0][0];
2449 k[1] = pc[1] < rgba[0][1];
2450 k[2] = pc[2] < rgba[0][2];
2451 k[3] = pc[3] < rgba[0][3];
2452 break;
2453 case PIPE_FUNC_LEQUAL:
2454 k[0] = pc[0] <= rgba[0][0];
2455 k[1] = pc[1] <= rgba[0][1];
2456 k[2] = pc[2] <= rgba[0][2];
2457 k[3] = pc[3] <= rgba[0][3];
2458 break;
2459 case PIPE_FUNC_GREATER:
2460 k[0] = pc[0] > rgba[0][0];
2461 k[1] = pc[1] > rgba[0][1];
2462 k[2] = pc[2] > rgba[0][2];
2463 k[3] = pc[3] > rgba[0][3];
2464 break;
2465 case PIPE_FUNC_GEQUAL:
2466 k[0] = pc[0] >= rgba[0][0];
2467 k[1] = pc[1] >= rgba[0][1];
2468 k[2] = pc[2] >= rgba[0][2];
2469 k[3] = pc[3] >= rgba[0][3];
2470 break;
2471 case PIPE_FUNC_EQUAL:
2472 k[0] = pc[0] == rgba[0][0];
2473 k[1] = pc[1] == rgba[0][1];
2474 k[2] = pc[2] == rgba[0][2];
2475 k[3] = pc[3] == rgba[0][3];
2476 break;
2477 case PIPE_FUNC_NOTEQUAL:
2478 k[0] = pc[0] != rgba[0][0];
2479 k[1] = pc[1] != rgba[0][1];
2480 k[2] = pc[2] != rgba[0][2];
2481 k[3] = pc[3] != rgba[0][3];
2482 break;
2483 case PIPE_FUNC_ALWAYS:
2484 k[0] = k[1] = k[2] = k[3] = 1;
2485 break;
2486 case PIPE_FUNC_NEVER:
2487 k[0] = k[1] = k[2] = k[3] = 0;
2488 break;
2489 default:
2490 k[0] = k[1] = k[2] = k[3] = 0;
2491 assert(0);
2492 break;
2493 }
2494
2495 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2496 rgba[0][j] = k[j];
2497 rgba[1][j] = k[j];
2498 rgba[2][j] = k[j];
2499 rgba[3][j] = 1.0F;
2500 }
2501 }
2502
2503
2504 static void
2505 do_swizzling(const struct pipe_sampler_view *sview,
2506 float in[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
2507 float out[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2508 {
2509 int j;
2510 const unsigned swizzle_r = sview->swizzle_r;
2511 const unsigned swizzle_g = sview->swizzle_g;
2512 const unsigned swizzle_b = sview->swizzle_b;
2513 const unsigned swizzle_a = sview->swizzle_a;
2514
2515 switch (swizzle_r) {
2516 case PIPE_SWIZZLE_ZERO:
2517 for (j = 0; j < 4; j++)
2518 out[0][j] = 0.0f;
2519 break;
2520 case PIPE_SWIZZLE_ONE:
2521 for (j = 0; j < 4; j++)
2522 out[0][j] = 1.0f;
2523 break;
2524 default:
2525 assert(swizzle_r < 4);
2526 for (j = 0; j < 4; j++)
2527 out[0][j] = in[swizzle_r][j];
2528 }
2529
2530 switch (swizzle_g) {
2531 case PIPE_SWIZZLE_ZERO:
2532 for (j = 0; j < 4; j++)
2533 out[1][j] = 0.0f;
2534 break;
2535 case PIPE_SWIZZLE_ONE:
2536 for (j = 0; j < 4; j++)
2537 out[1][j] = 1.0f;
2538 break;
2539 default:
2540 assert(swizzle_g < 4);
2541 for (j = 0; j < 4; j++)
2542 out[1][j] = in[swizzle_g][j];
2543 }
2544
2545 switch (swizzle_b) {
2546 case PIPE_SWIZZLE_ZERO:
2547 for (j = 0; j < 4; j++)
2548 out[2][j] = 0.0f;
2549 break;
2550 case PIPE_SWIZZLE_ONE:
2551 for (j = 0; j < 4; j++)
2552 out[2][j] = 1.0f;
2553 break;
2554 default:
2555 assert(swizzle_b < 4);
2556 for (j = 0; j < 4; j++)
2557 out[2][j] = in[swizzle_b][j];
2558 }
2559
2560 switch (swizzle_a) {
2561 case PIPE_SWIZZLE_ZERO:
2562 for (j = 0; j < 4; j++)
2563 out[3][j] = 0.0f;
2564 break;
2565 case PIPE_SWIZZLE_ONE:
2566 for (j = 0; j < 4; j++)
2567 out[3][j] = 1.0f;
2568 break;
2569 default:
2570 assert(swizzle_a < 4);
2571 for (j = 0; j < 4; j++)
2572 out[3][j] = in[swizzle_a][j];
2573 }
2574 }
2575
2576
2577 static wrap_nearest_func
2578 get_nearest_unorm_wrap(unsigned mode)
2579 {
2580 switch (mode) {
2581 case PIPE_TEX_WRAP_CLAMP:
2582 return wrap_nearest_unorm_clamp;
2583 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2584 return wrap_nearest_unorm_clamp_to_edge;
2585 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2586 return wrap_nearest_unorm_clamp_to_border;
2587 default:
2588 assert(0);
2589 return wrap_nearest_unorm_clamp;
2590 }
2591 }
2592
2593
2594 static wrap_nearest_func
2595 get_nearest_wrap(unsigned mode)
2596 {
2597 switch (mode) {
2598 case PIPE_TEX_WRAP_REPEAT:
2599 return wrap_nearest_repeat;
2600 case PIPE_TEX_WRAP_CLAMP:
2601 return wrap_nearest_clamp;
2602 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2603 return wrap_nearest_clamp_to_edge;
2604 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2605 return wrap_nearest_clamp_to_border;
2606 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2607 return wrap_nearest_mirror_repeat;
2608 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2609 return wrap_nearest_mirror_clamp;
2610 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2611 return wrap_nearest_mirror_clamp_to_edge;
2612 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2613 return wrap_nearest_mirror_clamp_to_border;
2614 default:
2615 assert(0);
2616 return wrap_nearest_repeat;
2617 }
2618 }
2619
2620
2621 static wrap_linear_func
2622 get_linear_unorm_wrap(unsigned mode)
2623 {
2624 switch (mode) {
2625 case PIPE_TEX_WRAP_CLAMP:
2626 return wrap_linear_unorm_clamp;
2627 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2628 return wrap_linear_unorm_clamp_to_edge;
2629 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2630 return wrap_linear_unorm_clamp_to_border;
2631 default:
2632 assert(0);
2633 return wrap_linear_unorm_clamp;
2634 }
2635 }
2636
2637
2638 static wrap_linear_func
2639 get_linear_wrap(unsigned mode)
2640 {
2641 switch (mode) {
2642 case PIPE_TEX_WRAP_REPEAT:
2643 return wrap_linear_repeat;
2644 case PIPE_TEX_WRAP_CLAMP:
2645 return wrap_linear_clamp;
2646 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2647 return wrap_linear_clamp_to_edge;
2648 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2649 return wrap_linear_clamp_to_border;
2650 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2651 return wrap_linear_mirror_repeat;
2652 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2653 return wrap_linear_mirror_clamp;
2654 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2655 return wrap_linear_mirror_clamp_to_edge;
2656 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2657 return wrap_linear_mirror_clamp_to_border;
2658 default:
2659 assert(0);
2660 return wrap_linear_repeat;
2661 }
2662 }
2663
2664
2665 /**
2666 * Is swizzling needed for the given state key?
2667 */
2668 static INLINE bool
2669 any_swizzle(const struct pipe_sampler_view *view)
2670 {
2671 return (view->swizzle_r != PIPE_SWIZZLE_RED ||
2672 view->swizzle_g != PIPE_SWIZZLE_GREEN ||
2673 view->swizzle_b != PIPE_SWIZZLE_BLUE ||
2674 view->swizzle_a != PIPE_SWIZZLE_ALPHA);
2675 }
2676
2677
2678 static img_filter_func
2679 get_img_filter(const struct sp_sampler_view *sp_sview,
2680 const struct pipe_sampler_state *sampler,
2681 unsigned filter)
2682 {
2683 switch (sp_sview->base.texture->target) {
2684 case PIPE_BUFFER:
2685 case PIPE_TEXTURE_1D:
2686 if (filter == PIPE_TEX_FILTER_NEAREST)
2687 return img_filter_1d_nearest;
2688 else
2689 return img_filter_1d_linear;
2690 break;
2691 case PIPE_TEXTURE_1D_ARRAY:
2692 if (filter == PIPE_TEX_FILTER_NEAREST)
2693 return img_filter_1d_array_nearest;
2694 else
2695 return img_filter_1d_array_linear;
2696 break;
2697 case PIPE_TEXTURE_2D:
2698 case PIPE_TEXTURE_RECT:
2699 /* Try for fast path:
2700 */
2701 if (sp_sview->pot2d &&
2702 sampler->wrap_s == sampler->wrap_t &&
2703 sampler->normalized_coords)
2704 {
2705 switch (sampler->wrap_s) {
2706 case PIPE_TEX_WRAP_REPEAT:
2707 switch (filter) {
2708 case PIPE_TEX_FILTER_NEAREST:
2709 return img_filter_2d_nearest_repeat_POT;
2710 case PIPE_TEX_FILTER_LINEAR:
2711 return img_filter_2d_linear_repeat_POT;
2712 default:
2713 break;
2714 }
2715 break;
2716 case PIPE_TEX_WRAP_CLAMP:
2717 switch (filter) {
2718 case PIPE_TEX_FILTER_NEAREST:
2719 return img_filter_2d_nearest_clamp_POT;
2720 default:
2721 break;
2722 }
2723 }
2724 }
2725 /* Otherwise use default versions:
2726 */
2727 if (filter == PIPE_TEX_FILTER_NEAREST)
2728 return img_filter_2d_nearest;
2729 else
2730 return img_filter_2d_linear;
2731 break;
2732 case PIPE_TEXTURE_2D_ARRAY:
2733 if (filter == PIPE_TEX_FILTER_NEAREST)
2734 return img_filter_2d_array_nearest;
2735 else
2736 return img_filter_2d_array_linear;
2737 break;
2738 case PIPE_TEXTURE_CUBE:
2739 if (filter == PIPE_TEX_FILTER_NEAREST)
2740 return img_filter_cube_nearest;
2741 else
2742 return img_filter_cube_linear;
2743 break;
2744 case PIPE_TEXTURE_CUBE_ARRAY:
2745 if (filter == PIPE_TEX_FILTER_NEAREST)
2746 return img_filter_cube_array_nearest;
2747 else
2748 return img_filter_cube_array_linear;
2749 break;
2750 case PIPE_TEXTURE_3D:
2751 if (filter == PIPE_TEX_FILTER_NEAREST)
2752 return img_filter_3d_nearest;
2753 else
2754 return img_filter_3d_linear;
2755 break;
2756 default:
2757 assert(0);
2758 return img_filter_1d_nearest;
2759 }
2760 }
2761
2762
2763 static void
2764 sample_mip(struct sp_sampler_view *sp_sview,
2765 struct sp_sampler *sp_samp,
2766 const float s[TGSI_QUAD_SIZE],
2767 const float t[TGSI_QUAD_SIZE],
2768 const float p[TGSI_QUAD_SIZE],
2769 const float c0[TGSI_QUAD_SIZE],
2770 const float lod[TGSI_QUAD_SIZE],
2771 enum tgsi_sampler_control control,
2772 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2773 {
2774 mip_filter_func mip_filter;
2775 img_filter_func min_img_filter = NULL;
2776 img_filter_func mag_img_filter = NULL;
2777
2778 if (sp_sview->pot2d & sp_samp->min_mag_equal_repeat_linear) {
2779 mip_filter = mip_filter_linear_2d_linear_repeat_POT;
2780 }
2781 else {
2782 mip_filter = sp_samp->mip_filter;
2783 min_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->min_img_filter);
2784 if (sp_samp->min_mag_equal) {
2785 mag_img_filter = min_img_filter;
2786 }
2787 else {
2788 mag_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->base.mag_img_filter);
2789 }
2790 }
2791
2792 mip_filter(sp_sview, sp_samp, min_img_filter, mag_img_filter,
2793 s, t, p, c0, lod, control, rgba);
2794
2795 if (sp_samp->base.compare_mode != PIPE_TEX_COMPARE_NONE) {
2796 sample_compare(sp_sview, sp_samp, s, t, p, c0, lod, control, rgba);
2797 }
2798
2799 if (sp_sview->need_swizzle) {
2800 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2801 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
2802 do_swizzling(&sp_sview->base, rgba_temp, rgba);
2803 }
2804
2805 }
2806
2807
2808 /**
2809 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2810 * Put face info into the sampler faces[] array.
2811 */
2812 static void
2813 sample_cube(struct sp_sampler_view *sp_sview,
2814 struct sp_sampler *sp_samp,
2815 const float s[TGSI_QUAD_SIZE],
2816 const float t[TGSI_QUAD_SIZE],
2817 const float p[TGSI_QUAD_SIZE],
2818 const float c0[TGSI_QUAD_SIZE],
2819 const float c1[TGSI_QUAD_SIZE],
2820 enum tgsi_sampler_control control,
2821 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2822 {
2823 unsigned j;
2824 float ssss[4], tttt[4];
2825
2826 /* Not actually used, but the intermediate steps that do the
2827 * dereferencing don't know it.
2828 */
2829 static float pppp[4] = { 0, 0, 0, 0 };
2830
2831 pppp[0] = c0[0];
2832 pppp[1] = c0[1];
2833 pppp[2] = c0[2];
2834 pppp[3] = c0[3];
2835 /*
2836 major axis
2837 direction target sc tc ma
2838 ---------- ------------------------------- --- --- ---
2839 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2840 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2841 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2842 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2843 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2844 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2845 */
2846
2847 /* Choose the cube face and compute new s/t coords for the 2D face.
2848 *
2849 * Use the same cube face for all four pixels in the quad.
2850 *
2851 * This isn't ideal, but if we want to use a different cube face
2852 * per pixel in the quad, we'd have to also compute the per-face
2853 * LOD here too. That's because the four post-face-selection
2854 * texcoords are no longer related to each other (they're
2855 * per-face!) so we can't use subtraction to compute the partial
2856 * deriviates to compute the LOD. Doing so (near cube edges
2857 * anyway) gives us pretty much random values.
2858 */
2859 {
2860 /* use the average of the four pixel's texcoords to choose the face */
2861 const float rx = 0.25F * (s[0] + s[1] + s[2] + s[3]);
2862 const float ry = 0.25F * (t[0] + t[1] + t[2] + t[3]);
2863 const float rz = 0.25F * (p[0] + p[1] + p[2] + p[3]);
2864 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
2865
2866 if (arx >= ary && arx >= arz) {
2867 float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
2868 uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
2869 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2870 const float ima = -0.5F / fabsf(s[j]);
2871 ssss[j] = sign * p[j] * ima + 0.5F;
2872 tttt[j] = t[j] * ima + 0.5F;
2873 sp_sview->faces[j] = face;
2874 }
2875 }
2876 else if (ary >= arx && ary >= arz) {
2877 float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
2878 uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
2879 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2880 const float ima = -0.5F / fabsf(t[j]);
2881 ssss[j] = -s[j] * ima + 0.5F;
2882 tttt[j] = sign * -p[j] * ima + 0.5F;
2883 sp_sview->faces[j] = face;
2884 }
2885 }
2886 else {
2887 float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
2888 uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
2889 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2890 const float ima = -0.5F / fabsf(p[j]);
2891 ssss[j] = sign * -s[j] * ima + 0.5F;
2892 tttt[j] = t[j] * ima + 0.5F;
2893 sp_sview->faces[j] = face;
2894 }
2895 }
2896 }
2897
2898 sample_mip(sp_sview, sp_samp, ssss, tttt, pppp, c0, c1, control, rgba);
2899 }
2900
2901
2902 static void
2903 sp_get_dims(struct sp_sampler_view *sp_sview, int level,
2904 int dims[4])
2905 {
2906 const struct pipe_sampler_view *view = &sp_sview->base;
2907 const struct pipe_resource *texture = view->texture;
2908
2909 /* undefined according to EXT_gpu_program */
2910 level += view->u.tex.first_level;
2911 if (level > view->u.tex.last_level)
2912 return;
2913
2914 dims[0] = u_minify(texture->width0, level);
2915
2916 switch(texture->target) {
2917 case PIPE_TEXTURE_1D_ARRAY:
2918 dims[1] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
2919 /* fallthrough */
2920 case PIPE_TEXTURE_1D:
2921 return;
2922 case PIPE_TEXTURE_2D_ARRAY:
2923 dims[2] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
2924 /* fallthrough */
2925 case PIPE_TEXTURE_2D:
2926 case PIPE_TEXTURE_CUBE:
2927 case PIPE_TEXTURE_RECT:
2928 dims[1] = u_minify(texture->height0, level);
2929 return;
2930 case PIPE_TEXTURE_3D:
2931 dims[1] = u_minify(texture->height0, level);
2932 dims[2] = u_minify(texture->depth0, level);
2933 return;
2934 case PIPE_TEXTURE_CUBE_ARRAY:
2935 dims[1] = u_minify(texture->height0, level);
2936 dims[2] = (view->u.tex.last_layer - view->u.tex.first_layer + 1) / 6;
2937 break;
2938 case PIPE_BUFFER:
2939 dims[0] /= util_format_get_blocksize(view->format);
2940 return;
2941 default:
2942 assert(!"unexpected texture target in sp_get_dims()");
2943 return;
2944 }
2945 }
2946
2947 /**
2948 * This function is only used for getting unfiltered texels via the
2949 * TXF opcode. The GL spec says that out-of-bounds texel fetches
2950 * produce undefined results. Instead of crashing, lets just clamp
2951 * coords to the texture image size.
2952 */
2953 static void
2954 sp_get_texels(struct sp_sampler_view *sp_sview,
2955 const int v_i[TGSI_QUAD_SIZE],
2956 const int v_j[TGSI_QUAD_SIZE],
2957 const int v_k[TGSI_QUAD_SIZE],
2958 const int lod[TGSI_QUAD_SIZE],
2959 const int8_t offset[3],
2960 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2961 {
2962 union tex_tile_address addr;
2963 const struct pipe_resource *texture = sp_sview->base.texture;
2964 int j, c;
2965 const float *tx;
2966 int width, height, depth;
2967
2968 addr.value = 0;
2969 /* TODO write a better test for LOD */
2970 addr.bits.level = lod[0];
2971
2972 width = u_minify(texture->width0, addr.bits.level);