projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
softpipe: move some image filter parameters into a struct
[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 int
478 coord_to_layer(float coord, unsigned first_layer, unsigned last_layer)
479 {
480 int c = util_ifloor(coord + 0.5F);
481 return CLAMP(c, (int)first_layer, (int)last_layer);
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 /* Gather a quad of adjacent texels within a tile:
761 */
762 static INLINE void
763 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_view *sp_sview,
764 union tex_tile_address addr,
765 unsigned x, unsigned y,
766 const float *out[4])
767 {
768 const struct softpipe_tex_cached_tile *tile;
769
770 addr.bits.x = x / TEX_TILE_SIZE;
771 addr.bits.y = y / TEX_TILE_SIZE;
772 y %= TEX_TILE_SIZE;
773 x %= TEX_TILE_SIZE;
774
775 tile = sp_get_cached_tile_tex(sp_sview->cache, addr);
776
777 out[0] = &tile->data.color[y ][x ][0];
778 out[1] = &tile->data.color[y ][x+1][0];
779 out[2] = &tile->data.color[y+1][x ][0];
780 out[3] = &tile->data.color[y+1][x+1][0];
781 }
782
783
784 /* Gather a quad of potentially non-adjacent texels:
785 */
786 static INLINE void
787 get_texel_quad_2d_no_border(const struct sp_sampler_view *sp_sview,
788 union tex_tile_address addr,
789 int x0, int y0,
790 int x1, int y1,
791 const float *out[4])
792 {
793 out[0] = get_texel_2d_no_border( sp_sview, addr, x0, y0 );
794 out[1] = get_texel_2d_no_border( sp_sview, addr, x1, y0 );
795 out[2] = get_texel_2d_no_border( sp_sview, addr, x0, y1 );
796 out[3] = get_texel_2d_no_border( sp_sview, addr, x1, y1 );
797 }
798
799 /* Can involve a lot of unnecessary checks for border color:
800 */
801 static INLINE void
802 get_texel_quad_2d(const struct sp_sampler_view *sp_sview,
803 const struct sp_sampler *sp_samp,
804 union tex_tile_address addr,
805 int x0, int y0,
806 int x1, int y1,
807 const float *out[4])
808 {
809 out[0] = get_texel_2d( sp_sview, sp_samp, addr, x0, y0 );
810 out[1] = get_texel_2d( sp_sview, sp_samp, addr, x1, y0 );
811 out[3] = get_texel_2d( sp_sview, sp_samp, addr, x1, y1 );
812 out[2] = get_texel_2d( sp_sview, sp_samp, addr, x0, y1 );
813 }
814
815
816
817 /* 3d variants:
818 */
819 static INLINE const float *
820 get_texel_3d_no_border(const struct sp_sampler_view *sp_sview,
821 union tex_tile_address addr, int x, int y, int z)
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 addr.bits.z = z;
828 y %= TEX_TILE_SIZE;
829 x %= TEX_TILE_SIZE;
830
831 tile = sp_get_cached_tile_tex(sp_sview->cache, addr);
832
833 return &tile->data.color[y][x][0];
834 }
835
836
837 static INLINE const float *
838 get_texel_3d(const struct sp_sampler_view *sp_sview,
839 const struct sp_sampler *sp_samp,
840 union tex_tile_address addr, int x, int y, int z)
841 {
842 const struct pipe_resource *texture = sp_sview->base.texture;
843 unsigned level = addr.bits.level;
844
845 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
846 y < 0 || y >= (int) u_minify(texture->height0, level) ||
847 z < 0 || z >= (int) u_minify(texture->depth0, level)) {
848 return sp_samp->base.border_color.f;
849 }
850 else {
851 return get_texel_3d_no_border( sp_sview, addr, x, y, z );
852 }
853 }
854
855
856 /* Get texel pointer for 1D array texture */
857 static INLINE const float *
858 get_texel_1d_array(const struct sp_sampler_view *sp_sview,
859 const struct sp_sampler *sp_samp,
860 union tex_tile_address addr, int x, int y)
861 {
862 const struct pipe_resource *texture = sp_sview->base.texture;
863 unsigned level = addr.bits.level;
864
865 if (x < 0 || x >= (int) u_minify(texture->width0, level)) {
866 return sp_samp->base.border_color.f;
867 }
868 else {
869 return get_texel_2d_no_border(sp_sview, addr, x, y);
870 }
871 }
872
873
874 /* Get texel pointer for 2D array texture */
875 static INLINE const float *
876 get_texel_2d_array(const struct sp_sampler_view *sp_sview,
877 const struct sp_sampler *sp_samp,
878 union tex_tile_address addr, int x, int y, int layer)
879 {
880 const struct pipe_resource *texture = sp_sview->base.texture;
881 unsigned level = addr.bits.level;
882
883 assert(layer < (int) texture->array_size);
884 assert(layer >= 0);
885
886 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
887 y < 0 || y >= (int) u_minify(texture->height0, level)) {
888 return sp_samp->base.border_color.f;
889 }
890 else {
891 return get_texel_3d_no_border(sp_sview, addr, x, y, layer);
892 }
893 }
894
895
896 static INLINE const float *
897 get_texel_cube_seamless(const struct sp_sampler_view *sp_sview,
898 union tex_tile_address addr, int x, int y,
899 float *corner, int layer, unsigned face)
900 {
901 const struct pipe_resource *texture = sp_sview->base.texture;
902 unsigned level = addr.bits.level;
903 int new_x, new_y, max_x;
904
905 max_x = (int) u_minify(texture->width0, level);
906
907 assert(texture->width0 == texture->height0);
908 new_x = x;
909 new_y = y;
910
911 /* change the face */
912 if (x < 0) {
913 /*
914 * Cheat with corners. They are difficult and I believe because we don't get
915 * per-pixel faces we can actually have multiple corner texels per pixel,
916 * which screws things up majorly in any case (as the per spec behavior is
917 * to average the 3 remaining texels, which we might not have).
918 * Hence just make sure that the 2nd coord is clamped, will simply pick the
919 * sample which would have fallen off the x coord, but not y coord.
920 * So the filter weight of the samples will be wrong, but at least this
921 * ensures that only valid texels near the corner are used.
922 */
923 if (y < 0 || y >= max_x) {
924 y = CLAMP(y, 0, max_x - 1);
925 }
926 new_x = get_next_xcoord(face, 0, max_x -1, x, y);
927 new_y = get_next_ycoord(face, 0, max_x -1, x, y);
928 face = get_next_face(face, 0);
929 } else if (x >= max_x) {
930 if (y < 0 || y >= max_x) {
931 y = CLAMP(y, 0, max_x - 1);
932 }
933 new_x = get_next_xcoord(face, 1, max_x -1, x, y);
934 new_y = get_next_ycoord(face, 1, max_x -1, x, y);
935 face = get_next_face(face, 1);
936 } else if (y < 0) {
937 new_x = get_next_xcoord(face, 2, max_x -1, x, y);
938 new_y = get_next_ycoord(face, 2, max_x -1, x, y);
939 face = get_next_face(face, 2);
940 } else if (y >= max_x) {
941 new_x = get_next_xcoord(face, 3, max_x -1, x, y);
942 new_y = get_next_ycoord(face, 3, max_x -1, x, y);
943 face = get_next_face(face, 3);
944 }
945
946 return get_texel_3d_no_border(sp_sview, addr, new_x, new_y, layer + face);
947 }
948
949
950 /* Get texel pointer for cube array texture */
951 static INLINE const float *
952 get_texel_cube_array(const struct sp_sampler_view *sp_sview,
953 const struct sp_sampler *sp_samp,
954 union tex_tile_address addr, int x, int y, int layer)
955 {
956 const struct pipe_resource *texture = sp_sview->base.texture;
957 unsigned level = addr.bits.level;
958
959 assert(layer < (int) texture->array_size);
960 assert(layer >= 0);
961
962 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
963 y < 0 || y >= (int) u_minify(texture->height0, level)) {
964 return sp_samp->base.border_color.f;
965 }
966 else {
967 return get_texel_3d_no_border(sp_sview, addr, x, y, layer);
968 }
969 }
970 /**
971 * Given the logbase2 of a mipmap's base level size and a mipmap level,
972 * return the size (in texels) of that mipmap level.
973 * For example, if level[0].width = 256 then base_pot will be 8.
974 * If level = 2, then we'll return 64 (the width at level=2).
975 * Return 1 if level > base_pot.
976 */
977 static INLINE unsigned
978 pot_level_size(unsigned base_pot, unsigned level)
979 {
980 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
981 }
982
983
984 static void
985 print_sample(const char *function, const float *rgba)
986 {
987 debug_printf("%s %g %g %g %g\n",
988 function,
989 rgba[0], rgba[TGSI_NUM_CHANNELS], rgba[2*TGSI_NUM_CHANNELS], rgba[3*TGSI_NUM_CHANNELS]);
990 }
991
992
993 static void
994 print_sample_4(const char *function, float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
995 {
996 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
997 function,
998 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
999 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
1000 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
1001 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
1002 }
1003
1004
1005 /* Some image-filter fastpaths:
1006 */
1007 static INLINE void
1008 img_filter_2d_linear_repeat_POT(struct sp_sampler_view *sp_sview,
1009 struct sp_sampler *sp_samp,
1010 const struct img_filter_args *args,
1011 float *rgba)
1012 {
1013 unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
1014 unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
1015 int xmax = (xpot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
1016 int ymax = (ypot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
1017 union tex_tile_address addr;
1018 int c;
1019
1020 float u = args->s * xpot - 0.5F;
1021 float v = args->t * ypot - 0.5F;
1022
1023 int uflr = util_ifloor(u);
1024 int vflr = util_ifloor(v);
1025
1026 float xw = u - (float)uflr;
1027 float yw = v - (float)vflr;
1028
1029 int x0 = uflr & (xpot - 1);
1030 int y0 = vflr & (ypot - 1);
1031
1032 const float *tx[4];
1033
1034 addr.value = 0;
1035 addr.bits.level = args->level;
1036
1037 /* Can we fetch all four at once:
1038 */
1039 if (x0 < xmax && y0 < ymax) {
1040 get_texel_quad_2d_no_border_single_tile(sp_sview, addr, x0, y0, tx);
1041 }
1042 else {
1043 unsigned x1 = (x0 + 1) & (xpot - 1);
1044 unsigned y1 = (y0 + 1) & (ypot - 1);
1045 get_texel_quad_2d_no_border(sp_sview, addr, x0, y0, x1, y1, tx);
1046 }
1047
1048 /* interpolate R, G, B, A */
1049 for (c = 0; c < TGSI_QUAD_SIZE; c++) {
1050 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1051 tx[0][c], tx[1][c],
1052 tx[2][c], tx[3][c]);
1053 }
1054
1055 if (DEBUG_TEX) {
1056 print_sample(__FUNCTION__, rgba);
1057 }
1058 }
1059
1060
1061 static INLINE void
1062 img_filter_2d_nearest_repeat_POT(struct sp_sampler_view *sp_sview,
1063 struct sp_sampler *sp_samp,
1064 const struct img_filter_args *args,
1065 float rgba[TGSI_QUAD_SIZE])
1066 {
1067 unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
1068 unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
1069 const float *out;
1070 union tex_tile_address addr;
1071 int c;
1072
1073 float u = args->s * xpot;
1074 float v = args->t * ypot;
1075
1076 int uflr = util_ifloor(u);
1077 int vflr = util_ifloor(v);
1078
1079 int x0 = uflr & (xpot - 1);
1080 int y0 = vflr & (ypot - 1);
1081
1082 addr.value = 0;
1083 addr.bits.level = args->level;
1084
1085 out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
1086 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1087 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1088
1089 if (DEBUG_TEX) {
1090 print_sample(__FUNCTION__, rgba);
1091 }
1092 }
1093
1094
1095 static INLINE void
1096 img_filter_2d_nearest_clamp_POT(struct sp_sampler_view *sp_sview,
1097 struct sp_sampler *sp_samp,
1098 const struct img_filter_args *args,
1099 float rgba[TGSI_QUAD_SIZE])
1100 {
1101 unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
1102 unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
1103 union tex_tile_address addr;
1104 int c;
1105
1106 float u = args->s * xpot;
1107 float v = args->t * ypot;
1108
1109 int x0, y0;
1110 const float *out;
1111
1112 addr.value = 0;
1113 addr.bits.level = args->level;
1114
1115 x0 = util_ifloor(u);
1116 if (x0 < 0)
1117 x0 = 0;
1118 else if (x0 > (int) xpot - 1)
1119 x0 = xpot - 1;
1120
1121 y0 = util_ifloor(v);
1122 if (y0 < 0)
1123 y0 = 0;
1124 else if (y0 > (int) ypot - 1)
1125 y0 = ypot - 1;
1126
1127 out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
1128 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1129 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1130
1131 if (DEBUG_TEX) {
1132 print_sample(__FUNCTION__, rgba);
1133 }
1134 }
1135
1136
1137 static void
1138 img_filter_1d_nearest(struct sp_sampler_view *sp_sview,
1139 struct sp_sampler *sp_samp,
1140 const struct img_filter_args *args,
1141 float rgba[TGSI_QUAD_SIZE])
1142 {
1143 const struct pipe_resource *texture = sp_sview->base.texture;
1144 int width;
1145 int x;
1146 union tex_tile_address addr;
1147 const float *out;
1148 int c;
1149
1150 width = u_minify(texture->width0, args->level);
1151
1152 assert(width > 0);
1153
1154 addr.value = 0;
1155 addr.bits.level = args->level;
1156
1157 sp_samp->nearest_texcoord_s(args->s, width, &x);
1158
1159 out = get_texel_2d(sp_sview, sp_samp, addr, x, 0);
1160 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1161 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1162
1163 if (DEBUG_TEX) {
1164 print_sample(__FUNCTION__, rgba);
1165 }
1166 }
1167
1168
1169 static void
1170 img_filter_1d_array_nearest(struct sp_sampler_view *sp_sview,
1171 struct sp_sampler *sp_samp,
1172 const struct img_filter_args *args,
1173 float *rgba)
1174 {
1175 const struct pipe_resource *texture = sp_sview->base.texture;
1176 int width;
1177 int x, layer;
1178 union tex_tile_address addr;
1179 const float *out;
1180 int c;
1181
1182 width = u_minify(texture->width0, args->level);
1183
1184 assert(width > 0);
1185
1186 addr.value = 0;
1187 addr.bits.level = args->level;
1188
1189 sp_samp->nearest_texcoord_s(args->s, width, &x);
1190 layer = coord_to_layer(args->t, sp_sview->base.u.tex.first_layer,
1191 sp_sview->base.u.tex.last_layer);
1192
1193 out = get_texel_1d_array(sp_sview, sp_samp, addr, x, layer);
1194 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1195 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1196
1197 if (DEBUG_TEX) {
1198 print_sample(__FUNCTION__, rgba);
1199 }
1200 }
1201
1202
1203 static void
1204 img_filter_2d_nearest(struct sp_sampler_view *sp_sview,
1205 struct sp_sampler *sp_samp,
1206 const struct img_filter_args *args,
1207 float *rgba)
1208 {
1209 const struct pipe_resource *texture = sp_sview->base.texture;
1210 int width, height;
1211 int x, y;
1212 union tex_tile_address addr;
1213 const float *out;
1214 int c;
1215
1216 width = u_minify(texture->width0, args->level);
1217 height = u_minify(texture->height0, args->level);
1218
1219 assert(width > 0);
1220 assert(height > 0);
1221
1222 addr.value = 0;
1223 addr.bits.level = args->level;
1224
1225 sp_samp->nearest_texcoord_s(args->s, width, &x);
1226 sp_samp->nearest_texcoord_t(args->t, height, &y);
1227
1228 out = get_texel_2d(sp_sview, sp_samp, addr, x, y);
1229 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1230 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1231
1232 if (DEBUG_TEX) {
1233 print_sample(__FUNCTION__, rgba);
1234 }
1235 }
1236
1237
1238 static void
1239 img_filter_2d_array_nearest(struct sp_sampler_view *sp_sview,
1240 struct sp_sampler *sp_samp,
1241 const struct img_filter_args *args,
1242 float *rgba)
1243 {
1244 const struct pipe_resource *texture = sp_sview->base.texture;
1245 int width, height;
1246 int x, y, layer;
1247 union tex_tile_address addr;
1248 const float *out;
1249 int c;
1250
1251 width = u_minify(texture->width0, args->level);
1252 height = u_minify(texture->height0, args->level);
1253
1254 assert(width > 0);
1255 assert(height > 0);
1256
1257 addr.value = 0;
1258 addr.bits.level = args->level;
1259
1260 sp_samp->nearest_texcoord_s(args->s, width, &x);
1261 sp_samp->nearest_texcoord_t(args->t, height, &y);
1262 layer = coord_to_layer(args->p, sp_sview->base.u.tex.first_layer,
1263 sp_sview->base.u.tex.last_layer);
1264
1265 out = get_texel_2d_array(sp_sview, sp_samp, addr, x, y, layer);
1266 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1267 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1268
1269 if (DEBUG_TEX) {
1270 print_sample(__FUNCTION__, rgba);
1271 }
1272 }
1273
1274
1275 static void
1276 img_filter_cube_nearest(struct sp_sampler_view *sp_sview,
1277 struct sp_sampler *sp_samp,
1278 const struct img_filter_args *args,
1279 float *rgba)
1280 {
1281 const struct pipe_resource *texture = sp_sview->base.texture;
1282 int width, height;
1283 int x, y, layerface;
1284 union tex_tile_address addr;
1285 const float *out;
1286 int c;
1287
1288 width = u_minify(texture->width0, args->level);
1289 height = u_minify(texture->height0, args->level);
1290
1291 assert(width > 0);
1292 assert(height > 0);
1293
1294 addr.value = 0;
1295 addr.bits.level = args->level;
1296
1297 /*
1298 * If NEAREST filtering is done within a miplevel, always apply wrap
1299 * mode CLAMP_TO_EDGE.
1300 */
1301 if (sp_samp->base.seamless_cube_map) {
1302 wrap_nearest_clamp_to_edge(args->s, width, &x);
1303 wrap_nearest_clamp_to_edge(args->t, height, &y);
1304 } else {
1305 /* Would probably make sense to ignore mode and just do edge clamp */
1306 sp_samp->nearest_texcoord_s(args->s, width, &x);
1307 sp_samp->nearest_texcoord_t(args->t, height, &y);
1308 }
1309
1310 layerface = args->face_id + sp_sview->base.u.tex.first_layer;
1311 out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
1312 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1313 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1314
1315 if (DEBUG_TEX) {
1316 print_sample(__FUNCTION__, rgba);
1317 }
1318 }
1319
1320 static void
1321 img_filter_cube_array_nearest(struct sp_sampler_view *sp_sview,
1322 struct sp_sampler *sp_samp,
1323 const struct img_filter_args *args,
1324 float *rgba)
1325 {
1326 const struct pipe_resource *texture = sp_sview->base.texture;
1327 int width, height;
1328 int x, y, layerface;
1329 union tex_tile_address addr;
1330 const float *out;
1331 int c;
1332
1333 width = u_minify(texture->width0, args->level);
1334 height = u_minify(texture->height0, args->level);
1335
1336 assert(width > 0);
1337 assert(height > 0);
1338
1339 addr.value = 0;
1340 addr.bits.level = args->level;
1341
1342 sp_samp->nearest_texcoord_s(args->s, width, &x);
1343 sp_samp->nearest_texcoord_t(args->t, height, &y);
1344 layerface = coord_to_layer(6 * args->p + sp_sview->base.u.tex.first_layer,
1345 sp_sview->base.u.tex.first_layer,
1346 sp_sview->base.u.tex.last_layer - 5) + args->face_id;
1347
1348 out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
1349 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1350 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1351
1352 if (DEBUG_TEX) {
1353 print_sample(__FUNCTION__, rgba);
1354 }
1355 }
1356
1357 static void
1358 img_filter_3d_nearest(struct sp_sampler_view *sp_sview,
1359 struct sp_sampler *sp_samp,
1360 const struct img_filter_args *args,
1361 float *rgba)
1362 {
1363 const struct pipe_resource *texture = sp_sview->base.texture;
1364 int width, height, depth;
1365 int x, y, z;
1366 union tex_tile_address addr;
1367 const float *out;
1368 int c;
1369
1370 width = u_minify(texture->width0, args->level);
1371 height = u_minify(texture->height0, args->level);
1372 depth = u_minify(texture->depth0, args->level);
1373
1374 assert(width > 0);
1375 assert(height > 0);
1376 assert(depth > 0);
1377
1378 sp_samp->nearest_texcoord_s(args->s, width, &x);
1379 sp_samp->nearest_texcoord_t(args->t, height, &y);
1380 sp_samp->nearest_texcoord_p(args->p, depth, &z);
1381
1382 addr.value = 0;
1383 addr.bits.level = args->level;
1384
1385 out = get_texel_3d(sp_sview, sp_samp, addr, x, y, z);
1386 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1387 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1388 }
1389
1390
1391 static void
1392 img_filter_1d_linear(struct sp_sampler_view *sp_sview,
1393 struct sp_sampler *sp_samp,
1394 const struct img_filter_args *args,
1395 float *rgba)
1396 {
1397 const struct pipe_resource *texture = sp_sview->base.texture;
1398 int width;
1399 int x0, x1;
1400 float xw; /* weights */
1401 union tex_tile_address addr;
1402 const float *tx0, *tx1;
1403 int c;
1404
1405 width = u_minify(texture->width0, args->level);
1406
1407 assert(width > 0);
1408
1409 addr.value = 0;
1410 addr.bits.level = args->level;
1411
1412 sp_samp->linear_texcoord_s(args->s, width, &x0, &x1, &xw);
1413
1414 tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, 0);
1415 tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, 0);
1416
1417 /* interpolate R, G, B, A */
1418 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1419 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1420 }
1421
1422
1423 static void
1424 img_filter_1d_array_linear(struct sp_sampler_view *sp_sview,
1425 struct sp_sampler *sp_samp,
1426 const struct img_filter_args *args,
1427 float *rgba)
1428 {
1429 const struct pipe_resource *texture = sp_sview->base.texture;
1430 int width;
1431 int x0, x1, layer;
1432 float xw; /* weights */
1433 union tex_tile_address addr;
1434 const float *tx0, *tx1;
1435 int c;
1436
1437 width = u_minify(texture->width0, args->level);
1438
1439 assert(width > 0);
1440
1441 addr.value = 0;
1442 addr.bits.level = args->level;
1443
1444 sp_samp->linear_texcoord_s(args->s, width, &x0, &x1, &xw);
1445 layer = coord_to_layer(args->t, sp_sview->base.u.tex.first_layer,
1446 sp_sview->base.u.tex.last_layer);
1447
1448 tx0 = get_texel_1d_array(sp_sview, sp_samp, addr, x0, layer);
1449 tx1 = get_texel_1d_array(sp_sview, sp_samp, addr, x1, layer);
1450
1451 /* interpolate R, G, B, A */
1452 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1453 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1454 }
1455
1456
1457 static void
1458 img_filter_2d_linear(struct sp_sampler_view *sp_sview,
1459 struct sp_sampler *sp_samp,
1460 const struct img_filter_args *args,
1461 float *rgba)
1462 {
1463 const struct pipe_resource *texture = sp_sview->base.texture;
1464 int width, height;
1465 int x0, y0, x1, y1;
1466 float xw, yw; /* weights */
1467 union tex_tile_address addr;
1468 const float *tx0, *tx1, *tx2, *tx3;
1469 int c;
1470
1471 width = u_minify(texture->width0, args->level);
1472 height = u_minify(texture->height0, args->level);
1473
1474 assert(width > 0);
1475 assert(height > 0);
1476
1477 addr.value = 0;
1478 addr.bits.level = args->level;
1479
1480 sp_samp->linear_texcoord_s(args->s, width, &x0, &x1, &xw);
1481 sp_samp->linear_texcoord_t(args->t, height, &y0, &y1, &yw);
1482
1483 tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, y0);
1484 tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, y0);
1485 tx2 = get_texel_2d(sp_sview, sp_samp, addr, x0, y1);
1486 tx3 = get_texel_2d(sp_sview, sp_samp, addr, x1, y1);
1487
1488 /* interpolate R, G, B, A */
1489 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1490 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1491 tx0[c], tx1[c],
1492 tx2[c], tx3[c]);
1493 }
1494
1495
1496 static void
1497 img_filter_2d_array_linear(struct sp_sampler_view *sp_sview,
1498 struct sp_sampler *sp_samp,
1499 const struct img_filter_args *args,
1500 float *rgba)
1501 {
1502 const struct pipe_resource *texture = sp_sview->base.texture;
1503 int width, height;
1504 int x0, y0, x1, y1, layer;
1505 float xw, yw; /* weights */
1506 union tex_tile_address addr;
1507 const float *tx0, *tx1, *tx2, *tx3;
1508 int c;
1509
1510 width = u_minify(texture->width0, args->level);
1511 height = u_minify(texture->height0, args->level);
1512
1513 assert(width > 0);
1514 assert(height > 0);
1515
1516 addr.value = 0;
1517 addr.bits.level = args->level;
1518
1519 sp_samp->linear_texcoord_s(args->s, width, &x0, &x1, &xw);
1520 sp_samp->linear_texcoord_t(args->t, height, &y0, &y1, &yw);
1521 layer = coord_to_layer(args->p, sp_sview->base.u.tex.first_layer,
1522 sp_sview->base.u.tex.last_layer);
1523
1524 tx0 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y0, layer);
1525 tx1 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y0, layer);
1526 tx2 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y1, layer);
1527 tx3 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y1, layer);
1528
1529 /* interpolate R, G, B, A */
1530 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1531 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1532 tx0[c], tx1[c],
1533 tx2[c], tx3[c]);
1534 }
1535
1536
1537 static void
1538 img_filter_cube_linear(struct sp_sampler_view *sp_sview,
1539 struct sp_sampler *sp_samp,
1540 const struct img_filter_args *args,
1541 float *rgba)
1542 {
1543 const struct pipe_resource *texture = sp_sview->base.texture;
1544 int width, height;
1545 int x0, y0, x1, y1, layer;
1546 float xw, yw; /* weights */
1547 union tex_tile_address addr;
1548 const float *tx0, *tx1, *tx2, *tx3;
1549 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
1550 corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1551 int c;
1552
1553 width = u_minify(texture->width0, args->level);
1554 height = u_minify(texture->height0, args->level);
1555
1556 assert(width > 0);
1557 assert(height > 0);
1558
1559 addr.value = 0;
1560 addr.bits.level = args->level;
1561
1562 /*
1563 * For seamless if LINEAR filtering is done within a miplevel,
1564 * always apply wrap mode CLAMP_TO_BORDER.
1565 */
1566 if (sp_samp->base.seamless_cube_map) {
1567 /* Note this is a bit overkill, actual clamping is not required */
1568 wrap_linear_clamp_to_border(args->s, width, &x0, &x1, &xw);
1569 wrap_linear_clamp_to_border(args->t, height, &y0, &y1, &yw);
1570 } else {
1571 /* Would probably make sense to ignore mode and just do edge clamp */
1572 sp_samp->linear_texcoord_s(args->s, width, &x0, &x1, &xw);
1573 sp_samp->linear_texcoord_t(args->t, height, &y0, &y1, &yw);
1574 }
1575
1576 layer = sp_sview->base.u.tex.first_layer;
1577
1578 if (sp_samp->base.seamless_cube_map) {
1579 tx0 = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, args->face_id);
1580 tx1 = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, args->face_id);
1581 tx2 = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, args->face_id);
1582 tx3 = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, args->face_id);
1583 } else {
1584 tx0 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + args->face_id);
1585 tx1 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + args->face_id);
1586 tx2 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + args->face_id);
1587 tx3 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + args->face_id);
1588 }
1589
1590 /* interpolate R, G, B, A */
1591 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1592 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1593 tx0[c], tx1[c],
1594 tx2[c], tx3[c]);
1595 }
1596
1597
1598 static void
1599 img_filter_cube_array_linear(struct sp_sampler_view *sp_sview,
1600 struct sp_sampler *sp_samp,
1601 const struct img_filter_args *args,
1602 float *rgba)
1603 {
1604 const struct pipe_resource *texture = sp_sview->base.texture;
1605 int width, height;
1606 int x0, y0, x1, y1, layer;
1607 float xw, yw; /* weights */
1608 union tex_tile_address addr;
1609 const float *tx0, *tx1, *tx2, *tx3;
1610 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
1611 corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1612 int c;
1613
1614 width = u_minify(texture->width0, args->level);
1615 height = u_minify(texture->height0, args->level);
1616
1617 assert(width > 0);
1618 assert(height > 0);
1619
1620 addr.value = 0;
1621 addr.bits.level = args->level;
1622
1623 /*
1624 * For seamless if LINEAR filtering is done within a miplevel,
1625 * always apply wrap mode CLAMP_TO_BORDER.
1626 */
1627 if (sp_samp->base.seamless_cube_map) {
1628 /* Note this is a bit overkill, actual clamping is not required */
1629 wrap_linear_clamp_to_border(args->s, width, &x0, &x1, &xw);
1630 wrap_linear_clamp_to_border(args->t, height, &y0, &y1, &yw);
1631 } else {
1632 /* Would probably make sense to ignore mode and just do edge clamp */
1633 sp_samp->linear_texcoord_s(args->s, width, &x0, &x1, &xw);
1634 sp_samp->linear_texcoord_t(args->t, height, &y0, &y1, &yw);
1635 }
1636
1637 layer = coord_to_layer(6 * args->p + sp_sview->base.u.tex.first_layer,
1638 sp_sview->base.u.tex.first_layer,
1639 sp_sview->base.u.tex.last_layer - 5);
1640
1641 if (sp_samp->base.seamless_cube_map) {
1642 tx0 = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, args->face_id);
1643 tx1 = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, args->face_id);
1644 tx2 = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, args->face_id);
1645 tx3 = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, args->face_id);
1646 } else {
1647 tx0 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + args->face_id);
1648 tx1 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + args->face_id);
1649 tx2 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + args->face_id);
1650 tx3 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + args->face_id);
1651 }
1652
1653 /* interpolate R, G, B, A */
1654 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1655 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1656 tx0[c], tx1[c],
1657 tx2[c], tx3[c]);
1658 }
1659
1660 static void
1661 img_filter_3d_linear(struct sp_sampler_view *sp_sview,
1662 struct sp_sampler *sp_samp,
1663 const struct img_filter_args *args,
1664 float *rgba)
1665 {
1666 const struct pipe_resource *texture = sp_sview->base.texture;
1667 int width, height, depth;
1668 int x0, x1, y0, y1, z0, z1;
1669 float xw, yw, zw; /* interpolation weights */
1670 union tex_tile_address addr;
1671 const float *tx00, *tx01, *tx02, *tx03, *tx10, *tx11, *tx12, *tx13;
1672 int c;
1673
1674 width = u_minify(texture->width0, args->level);
1675 height = u_minify(texture->height0, args->level);
1676 depth = u_minify(texture->depth0, args->level);
1677
1678 addr.value = 0;
1679 addr.bits.level = args->level;
1680
1681 assert(width > 0);
1682 assert(height > 0);
1683 assert(depth > 0);
1684
1685 sp_samp->linear_texcoord_s(args->s, width, &x0, &x1, &xw);
1686 sp_samp->linear_texcoord_t(args->t, height, &y0, &y1, &yw);
1687 sp_samp->linear_texcoord_p(args->p, depth, &z0, &z1, &zw);
1688
1689 tx00 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z0);
1690 tx01 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z0);
1691 tx02 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z0);
1692 tx03 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z0);
1693
1694 tx10 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z1);
1695 tx11 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z1);
1696 tx12 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z1);
1697 tx13 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z1);
1698
1699 /* interpolate R, G, B, A */
1700 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1701 rgba[TGSI_NUM_CHANNELS*c] = lerp_3d(xw, yw, zw,
1702 tx00[c], tx01[c],
1703 tx02[c], tx03[c],
1704 tx10[c], tx11[c],
1705 tx12[c], tx13[c]);
1706 }
1707
1708
1709 /* Calculate level of detail for every fragment,
1710 * with lambda already computed.
1711 * Note that lambda has already been biased by global LOD bias.
1712 * \param biased_lambda per-quad lambda.
1713 * \param lod_in per-fragment lod_bias or explicit_lod.
1714 * \param lod returns the per-fragment lod.
1715 */
1716 static INLINE void
1717 compute_lod(const struct pipe_sampler_state *sampler,
1718 enum tgsi_sampler_control control,
1719 const float biased_lambda,
1720 const float lod_in[TGSI_QUAD_SIZE],
1721 float lod[TGSI_QUAD_SIZE])
1722 {
1723 float min_lod = sampler->min_lod;
1724 float max_lod = sampler->max_lod;
1725 uint i;
1726
1727 switch (control) {
1728 case tgsi_sampler_lod_none:
1729 case tgsi_sampler_lod_zero:
1730 /* XXX FIXME */
1731 case tgsi_sampler_derivs_explicit:
1732 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(biased_lambda, min_lod, max_lod);
1733 break;
1734 case tgsi_sampler_lod_bias:
1735 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1736 lod[i] = biased_lambda + lod_in[i];
1737 lod[i] = CLAMP(lod[i], min_lod, max_lod);
1738 }
1739 break;
1740 case tgsi_sampler_lod_explicit:
1741 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1742 lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
1743 }
1744 break;
1745 default:
1746 assert(0);
1747 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1748 }
1749 }
1750
1751
1752 /* Calculate level of detail for every fragment.
1753 * \param lod_in per-fragment lod_bias or explicit_lod.
1754 * \param lod results per-fragment lod.
1755 */
1756 static INLINE void
1757 compute_lambda_lod(struct sp_sampler_view *sp_sview,
1758 struct sp_sampler *sp_samp,
1759 const float s[TGSI_QUAD_SIZE],
1760 const float t[TGSI_QUAD_SIZE],
1761 const float p[TGSI_QUAD_SIZE],
1762 const float lod_in[TGSI_QUAD_SIZE],
1763 enum tgsi_sampler_control control,
1764 float lod[TGSI_QUAD_SIZE])
1765 {
1766 const struct pipe_sampler_state *sampler = &sp_samp->base;
1767 float lod_bias = sampler->lod_bias;
1768 float min_lod = sampler->min_lod;
1769 float max_lod = sampler->max_lod;
1770 float lambda;
1771 uint i;
1772
1773 switch (control) {
1774 case tgsi_sampler_lod_none:
1775 /* XXX FIXME */
1776 case tgsi_sampler_derivs_explicit:
1777 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1778 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lambda, min_lod, max_lod);
1779 break;
1780 case tgsi_sampler_lod_bias:
1781 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1782 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1783 lod[i] = 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 case tgsi_sampler_lod_zero:
1793 /* this is all static state in the sampler really need clamp here? */
1794 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lod_bias, min_lod, max_lod);
1795 break;
1796 default:
1797 assert(0);
1798 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1799 }
1800 }
1801
1802
1803 static void
1804 mip_filter_linear(struct sp_sampler_view *sp_sview,
1805 struct sp_sampler *sp_samp,
1806 img_filter_func min_filter,
1807 img_filter_func mag_filter,
1808 const float s[TGSI_QUAD_SIZE],
1809 const float t[TGSI_QUAD_SIZE],
1810 const float p[TGSI_QUAD_SIZE],
1811 const float c0[TGSI_QUAD_SIZE],
1812 const float lod_in[TGSI_QUAD_SIZE],
1813 enum tgsi_sampler_control control,
1814 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1815 {
1816 const struct pipe_sampler_view *psview = &sp_sview->base;
1817 int j;
1818 float lod[TGSI_QUAD_SIZE];
1819 struct img_filter_args args;
1820
1821 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1822
1823 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1824 int level0 = psview->u.tex.first_level + (int)lod[j];
1825
1826 args.s = s[j];
1827 args.t = t[j];
1828 args.p = p[j];
1829 args.face_id = sp_sview->faces[j];
1830
1831 if (lod[j] < 0.0) {
1832 args.level = psview->u.tex.first_level;
1833 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1834 }
1835 else if (level0 >= (int) psview->u.tex.last_level) {
1836 args.level = psview->u.tex.last_level;
1837 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1838 }
1839 else {
1840 float levelBlend = frac(lod[j]);
1841 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
1842 int c;
1843
1844 args.level = level0;
1845 min_filter(sp_sview, sp_samp, &args, &rgbax[0][0]);
1846 args.level = level0+1;
1847 min_filter(sp_sview, sp_samp, &args, &rgbax[0][1]);
1848
1849 for (c = 0; c < 4; c++) {
1850 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
1851 }
1852 }
1853 }
1854
1855 if (DEBUG_TEX) {
1856 print_sample_4(__FUNCTION__, rgba);
1857 }
1858 }
1859
1860
1861 /**
1862 * Compute nearest mipmap level from texcoords.
1863 * Then sample the texture level for four elements of a quad.
1864 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1865 */
1866 static void
1867 mip_filter_nearest(struct sp_sampler_view *sp_sview,
1868 struct sp_sampler *sp_samp,
1869 img_filter_func min_filter,
1870 img_filter_func mag_filter,
1871 const float s[TGSI_QUAD_SIZE],
1872 const float t[TGSI_QUAD_SIZE],
1873 const float p[TGSI_QUAD_SIZE],
1874 const float c0[TGSI_QUAD_SIZE],
1875 const float lod_in[TGSI_QUAD_SIZE],
1876 enum tgsi_sampler_control control,
1877 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1878 {
1879 const struct pipe_sampler_view *psview = &sp_sview->base;
1880 float lod[TGSI_QUAD_SIZE];
1881 int j;
1882 struct img_filter_args args;
1883 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1884
1885 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1886 args.s = s[j];
1887 args.t = t[j];
1888 args.p = p[j];
1889 args.face_id = sp_sview->faces[j];
1890
1891 if (lod[j] < 0.0) {
1892 args.level = psview->u.tex.first_level;
1893 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1894 } else {
1895 int level = psview->u.tex.first_level + (int)(lod[j] + 0.5F);
1896 args.level = MIN2(level, (int)psview->u.tex.last_level);
1897 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1898 }
1899 }
1900
1901 if (DEBUG_TEX) {
1902 print_sample_4(__FUNCTION__, rgba);
1903 }
1904 }
1905
1906
1907 static void
1908 mip_filter_none(struct sp_sampler_view *sp_sview,
1909 struct sp_sampler *sp_samp,
1910 img_filter_func min_filter,
1911 img_filter_func mag_filter,
1912 const float s[TGSI_QUAD_SIZE],
1913 const float t[TGSI_QUAD_SIZE],
1914 const float p[TGSI_QUAD_SIZE],
1915 const float c0[TGSI_QUAD_SIZE],
1916 const float lod_in[TGSI_QUAD_SIZE],
1917 enum tgsi_sampler_control control,
1918 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1919 {
1920 float lod[TGSI_QUAD_SIZE];
1921 int j;
1922 struct img_filter_args args;
1923
1924 args.level = sp_sview->base.u.tex.first_level;
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 args.s = s[j];
1929 args.t = t[j];
1930 args.p = p[j];
1931 args.face_id = sp_sview->faces[j];
1932 if (lod[j] < 0.0) {
1933 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1934 }
1935 else {
1936 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1937 }
1938 }
1939 }
1940
1941
1942 static void
1943 mip_filter_none_no_filter_select(struct sp_sampler_view *sp_sview,
1944 struct sp_sampler *sp_samp,
1945 img_filter_func min_filter,
1946 img_filter_func mag_filter,
1947 const float s[TGSI_QUAD_SIZE],
1948 const float t[TGSI_QUAD_SIZE],
1949 const float p[TGSI_QUAD_SIZE],
1950 const float c0[TGSI_QUAD_SIZE],
1951 const float lod_in[TGSI_QUAD_SIZE],
1952 enum tgsi_sampler_control control,
1953 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1954 {
1955 int j;
1956 struct img_filter_args args;
1957 args.level = sp_sview->base.u.tex.first_level;
1958 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1959 args.s = s[j];
1960 args.t = t[j];
1961 args.p = p[j];
1962 args.face_id = sp_sview->faces[j];
1963 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1964 }
1965 }
1966
1967
1968 /* For anisotropic filtering */
1969 #define WEIGHT_LUT_SIZE 1024
1970
1971 static float *weightLut = NULL;
1972
1973 /**
1974 * Creates the look-up table used to speed-up EWA sampling
1975 */
1976 static void
1977 create_filter_table(void)
1978 {
1979 unsigned i;
1980 if (!weightLut) {
1981 weightLut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
1982
1983 for (i = 0; i < WEIGHT_LUT_SIZE; ++i) {
1984 float alpha = 2;
1985 float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
1986 float weight = (float) exp(-alpha * r2);
1987 weightLut[i] = weight;
1988 }
1989 }
1990 }
1991
1992
1993 /**
1994 * Elliptical weighted average (EWA) filter for producing high quality
1995 * anisotropic filtered results.
1996 * Based on the Higher Quality Elliptical Weighted Average Filter
1997 * published by Paul S. Heckbert in his Master's Thesis
1998 * "Fundamentals of Texture Mapping and Image Warping" (1989)
1999 */
2000 static void
2001 img_filter_2d_ewa(struct sp_sampler_view *sp_sview,
2002 struct sp_sampler *sp_samp,
2003 img_filter_func min_filter,
2004 img_filter_func mag_filter,
2005 const float s[TGSI_QUAD_SIZE],
2006 const float t[TGSI_QUAD_SIZE],
2007 const float p[TGSI_QUAD_SIZE],
2008 unsigned level,
2009 const float dudx, const float dvdx,
2010 const float dudy, const float dvdy,
2011 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2012 {
2013 const struct pipe_resource *texture = sp_sview->base.texture;
2014
2015 // ??? Won't the image filters blow up if level is negative?
2016 unsigned level0 = level > 0 ? level : 0;
2017 float scaling = 1.0f / (1 << level0);
2018 int width = u_minify(texture->width0, level0);
2019 int height = u_minify(texture->height0, level0);
2020 struct img_filter_args args;
2021 float ux = dudx * scaling;
2022 float vx = dvdx * scaling;
2023 float uy = dudy * scaling;
2024 float vy = dvdy * scaling;
2025
2026 /* compute ellipse coefficients to bound the region:
2027 * A*x*x + B*x*y + C*y*y = F.
2028 */
2029 float A = vx*vx+vy*vy+1;
2030 float B = -2*(ux*vx+uy*vy);
2031 float C = ux*ux+uy*uy+1;
2032 float F = A*C-B*B/4.0f;
2033
2034 /* check if it is an ellipse */
2035 /* assert(F > 0.0); */
2036
2037 /* Compute the ellipse's (u,v) bounding box in texture space */
2038 float d = -B*B+4.0f*C*A;
2039 float box_u = 2.0f / d * sqrtf(d*C*F); /* box_u -> half of bbox with */
2040 float box_v = 2.0f / d * sqrtf(A*d*F); /* box_v -> half of bbox height */
2041
2042 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2043 float s_buffer[TGSI_QUAD_SIZE];
2044 float t_buffer[TGSI_QUAD_SIZE];
2045 float weight_buffer[TGSI_QUAD_SIZE];
2046 unsigned buffer_next;
2047 int j;
2048 float den; /* = 0.0F; */
2049 float ddq;
2050 float U; /* = u0 - tex_u; */
2051 int v;
2052
2053 /* Scale ellipse formula to directly index the Filter Lookup Table.
2054 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2055 */
2056 double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
2057 A *= formScale;
2058 B *= formScale;
2059 C *= formScale;
2060 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2061
2062 /* For each quad, the du and dx values are the same and so the ellipse is
2063 * also the same. Note that texel/image access can only be performed using
2064 * a quad, i.e. it is not possible to get the pixel value for a single
2065 * tex coord. In order to have a better performance, the access is buffered
2066 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2067 * full, then the pixel values are read from the image.
2068 */
2069 ddq = 2 * A;
2070
2071 args.level = level;
2072 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2073 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2074 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2075 * value, q, is less than F, we're inside the ellipse
2076 */
2077 float tex_u = -0.5F + s[j] * texture->width0 * scaling;
2078 float tex_v = -0.5F + t[j] * texture->height0 * scaling;
2079
2080 int u0 = (int) floorf(tex_u - box_u);
2081 int u1 = (int) ceilf(tex_u + box_u);
2082 int v0 = (int) floorf(tex_v - box_v);
2083 int v1 = (int) ceilf(tex_v + box_v);
2084
2085 float num[4] = {0.0F, 0.0F, 0.0F, 0.0F};
2086 buffer_next = 0;
2087 den = 0;
2088 args.face_id = sp_sview->faces[j];
2089
2090 U = u0 - tex_u;
2091 for (v = v0; v <= v1; ++v) {
2092 float V = v - tex_v;
2093 float dq = A * (2 * U + 1) + B * V;
2094 float q = (C * V + B * U) * V + A * U * U;
2095
2096 int u;
2097 for (u = u0; u <= u1; ++u) {
2098 /* Note that the ellipse has been pre-scaled so F =
2099 * WEIGHT_LUT_SIZE - 1
2100 */
2101 if (q < WEIGHT_LUT_SIZE) {
2102 /* as a LUT is used, q must never be negative;
2103 * should not happen, though
2104 */
2105 const int qClamped = q >= 0.0F ? q : 0;
2106 float weight = weightLut[qClamped];
2107
2108 weight_buffer[buffer_next] = weight;
2109 s_buffer[buffer_next] = u / ((float) width);
2110 t_buffer[buffer_next] = v / ((float) height);
2111
2112 buffer_next++;
2113 if (buffer_next == TGSI_QUAD_SIZE) {
2114 /* 4 texel coords are in the buffer -> read it now */
2115 unsigned jj;
2116 /* it is assumed that samp->min_img_filter is set to
2117 * img_filter_2d_nearest or one of the
2118 * accelerated img_filter_2d_nearest_XXX functions.
2119 */
2120 for (jj = 0; jj < buffer_next; jj++) {
2121 args.s = s_buffer[jj];
2122 args.t = t_buffer[jj];
2123 args.p = p[jj];
2124 min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][jj]);
2125 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2126 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2127 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2128 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2129 }
2130
2131 buffer_next = 0;
2132 }
2133
2134 den += weight;
2135 }
2136 q += dq;
2137 dq += ddq;
2138 }
2139 }
2140
2141 /* if the tex coord buffer contains unread values, we will read
2142 * them now.
2143 */
2144 if (buffer_next > 0) {
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 args.s = s_buffer[jj];
2152 args.t = t_buffer[jj];
2153 args.p = p[jj];
2154 min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][jj]);
2155 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2156 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2157 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2158 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2159 }
2160 }
2161
2162 if (den <= 0.0F) {
2163 /* Reaching this place would mean that no pixels intersected
2164 * the ellipse. This should never happen because the filter
2165 * we use always intersects at least one pixel.
2166 */
2167
2168 /*rgba[0]=0;
2169 rgba[1]=0;
2170 rgba[2]=0;
2171 rgba[3]=0;*/
2172 /* not enough pixels in resampling, resort to direct interpolation */
2173 args.s = s[j];
2174 args.t = t[j];
2175 args.p = p[j];
2176 min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][j]);
2177 den = 1;
2178 num[0] = rgba_temp[0][j];
2179 num[1] = rgba_temp[1][j];
2180 num[2] = rgba_temp[2][j];
2181 num[3] = rgba_temp[3][j];
2182 }
2183
2184 rgba[0][j] = num[0] / den;
2185 rgba[1][j] = num[1] / den;
2186 rgba[2][j] = num[2] / den;
2187 rgba[3][j] = num[3] / den;
2188 }
2189 }
2190
2191
2192 /**
2193 * Sample 2D texture using an anisotropic filter.
2194 */
2195 static void
2196 mip_filter_linear_aniso(struct sp_sampler_view *sp_sview,
2197 struct sp_sampler *sp_samp,
2198 img_filter_func min_filter,
2199 img_filter_func mag_filter,
2200 const float s[TGSI_QUAD_SIZE],
2201 const float t[TGSI_QUAD_SIZE],
2202 const float p[TGSI_QUAD_SIZE],
2203 const float c0[TGSI_QUAD_SIZE],
2204 const float lod_in[TGSI_QUAD_SIZE],
2205 enum tgsi_sampler_control control,
2206 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2207 {
2208 const struct pipe_resource *texture = sp_sview->base.texture;
2209 const struct pipe_sampler_view *psview = &sp_sview->base;
2210 int level0;
2211 float lambda;
2212 float lod[TGSI_QUAD_SIZE];
2213
2214 float s_to_u = u_minify(texture->width0, psview->u.tex.first_level);
2215 float t_to_v = u_minify(texture->height0, psview->u.tex.first_level);
2216 float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2217 float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2218 float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2219 float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2220 struct img_filter_args args;
2221
2222 if (control == tgsi_sampler_lod_bias ||
2223 control == tgsi_sampler_lod_none ||
2224 /* XXX FIXME */
2225 control == tgsi_sampler_derivs_explicit) {
2226 /* note: instead of working with Px and Py, we will use the
2227 * squared length instead, to avoid sqrt.
2228 */
2229 float Px2 = dudx * dudx + dvdx * dvdx;
2230 float Py2 = dudy * dudy + dvdy * dvdy;
2231
2232 float Pmax2;
2233 float Pmin2;
2234 float e;
2235 const float maxEccentricity = sp_samp->base.max_anisotropy * sp_samp->base.max_anisotropy;
2236
2237 if (Px2 < Py2) {
2238 Pmax2 = Py2;
2239 Pmin2 = Px2;
2240 }
2241 else {
2242 Pmax2 = Px2;
2243 Pmin2 = Py2;
2244 }
2245
2246 /* if the eccentricity of the ellipse is too big, scale up the shorter
2247 * of the two vectors to limit the maximum amount of work per pixel
2248 */
2249 e = Pmax2 / Pmin2;
2250 if (e > maxEccentricity) {
2251 /* float s=e / maxEccentricity;
2252 minor[0] *= s;
2253 minor[1] *= s;
2254 Pmin2 *= s; */
2255 Pmin2 = Pmax2 / maxEccentricity;
2256 }
2257
2258 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2259 * this since 0.5*log(x) = log(sqrt(x))
2260 */
2261 lambda = 0.5F * util_fast_log2(Pmin2) + sp_samp->base.lod_bias;
2262 compute_lod(&sp_samp->base, control, lambda, lod_in, lod);
2263 }
2264 else {
2265 assert(control == tgsi_sampler_lod_explicit ||
2266 control == tgsi_sampler_lod_zero);
2267 compute_lod(&sp_samp->base, control, sp_samp->base.lod_bias, lod_in, lod);
2268 }
2269
2270 /* XXX: Take into account all lod values.
2271 */
2272 lambda = lod[0];
2273 level0 = psview->u.tex.first_level + (int)lambda;
2274
2275 /* If the ellipse covers the whole image, we can
2276 * simply return the average of the whole image.
2277 */
2278 if (level0 >= (int) psview->u.tex.last_level) {
2279 int j;
2280 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2281 args.s = s[j];
2282 args.t = t[j];
2283 args.p = p[j];
2284 args.level = psview->u.tex.last_level;
2285 args.face_id = sp_sview->faces[j];
2286 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
2287 }
2288 }
2289 else {
2290 /* don't bother interpolating between multiple LODs; it doesn't
2291 * seem to be worth the extra running time.
2292 */
2293 img_filter_2d_ewa(sp_sview, sp_samp, min_filter, mag_filter,
2294 s, t, p, level0,
2295 dudx, dvdx, dudy, dvdy, rgba);
2296 }
2297
2298 if (DEBUG_TEX) {
2299 print_sample_4(__FUNCTION__, rgba);
2300 }
2301 }
2302
2303
2304 /**
2305 * Specialized version of mip_filter_linear with hard-wired calls to
2306 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2307 */
2308 static void
2309 mip_filter_linear_2d_linear_repeat_POT(
2310 struct sp_sampler_view *sp_sview,
2311 struct sp_sampler *sp_samp,
2312 img_filter_func min_filter,
2313 img_filter_func mag_filter,
2314 const float s[TGSI_QUAD_SIZE],
2315 const float t[TGSI_QUAD_SIZE],
2316 const float p[TGSI_QUAD_SIZE],
2317 const float c0[TGSI_QUAD_SIZE],
2318 const float lod_in[TGSI_QUAD_SIZE],
2319 enum tgsi_sampler_control control,
2320 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2321 {
2322 const struct pipe_sampler_view *psview = &sp_sview->base;
2323 int j;
2324 float lod[TGSI_QUAD_SIZE];
2325
2326 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
2327
2328 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2329 int level0 = psview->u.tex.first_level + (int)lod[j];
2330 struct img_filter_args args;
2331 /* Catches both negative and large values of level0:
2332 */
2333 args.s = s[j];
2334 args.t = t[j];
2335 args.p = p[j];
2336 args.face_id = sp_sview->faces[j];
2337 if ((unsigned)level0 >= psview->u.tex.last_level) {
2338 if (level0 < 0)
2339 args.level = psview->u.tex.first_level;
2340 else
2341 args.level = psview->u.tex.last_level;
2342 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args,
2343 &rgba[0][j]);
2344
2345 }
2346 else {
2347 float levelBlend = frac(lod[j]);
2348 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2349 int c;
2350
2351 args.level = level0;
2352 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args, &rgbax[0][0]);
2353 args.level = level0+1;
2354 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args, &rgbax[0][1]);
2355
2356 for (c = 0; c < TGSI_NUM_CHANNELS; c++)
2357 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
2358 }
2359 }
2360
2361 if (DEBUG_TEX) {
2362 print_sample_4(__FUNCTION__, rgba);
2363 }
2364 }
2365
2366
2367 /**
2368 * Do shadow/depth comparisons.
2369 */
2370 static void
2371 sample_compare(struct sp_sampler_view *sp_sview,
2372 struct sp_sampler *sp_samp,
2373 const float s[TGSI_QUAD_SIZE],
2374 const float t[TGSI_QUAD_SIZE],
2375 const float p[TGSI_QUAD_SIZE],
2376 const float c0[TGSI_QUAD_SIZE],
2377 const float c1[TGSI_QUAD_SIZE],
2378 enum tgsi_sampler_control control,
2379 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2380 {
2381 const struct pipe_sampler_state *sampler = &sp_samp->base;
2382 int j;
2383 int k[4];
2384 float pc[4];
2385 const struct util_format_description *format_desc;
2386 unsigned chan_type;
2387
2388 /**
2389 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2390 * for 2D Array texture we need to use the 'c0' (aka Q).
2391 * When we sampled the depth texture, the depth value was put into all
2392 * RGBA channels. We look at the red channel here.
2393 */
2394
2395 if (sp_sview->base.target == PIPE_TEXTURE_2D_ARRAY ||
2396 sp_sview->base.target == PIPE_TEXTURE_CUBE) {
2397 pc[0] = c0[0];
2398 pc[1] = c0[1];
2399 pc[2] = c0[2];
2400 pc[3] = c0[3];
2401 } else if (sp_sview->base.target == PIPE_TEXTURE_CUBE_ARRAY) {
2402 pc[0] = c1[0];
2403 pc[1] = c1[1];
2404 pc[2] = c1[2];
2405 pc[3] = c1[3];
2406 } else {
2407 pc[0] = p[0];
2408 pc[1] = p[1];
2409 pc[2] = p[2];
2410 pc[3] = p[3];
2411 }
2412
2413 format_desc = util_format_description(sp_sview->base.format);
2414 /* not entirely sure we couldn't end up with non-valid swizzle here */
2415 chan_type = format_desc->swizzle[0] <= UTIL_FORMAT_SWIZZLE_W ?
2416 format_desc->channel[format_desc->swizzle[0]].type :
2417 UTIL_FORMAT_TYPE_FLOAT;
2418 if (chan_type != UTIL_FORMAT_TYPE_FLOAT) {
2419 /*
2420 * clamping is a result of conversion to texture format, hence
2421 * doesn't happen with floats. Technically also should do comparison
2422 * in texture format (quantization!).
2423 */
2424 pc[0] = CLAMP(pc[0], 0.0F, 1.0F);
2425 pc[1] = CLAMP(pc[1], 0.0F, 1.0F);
2426 pc[2] = CLAMP(pc[2], 0.0F, 1.0F);
2427 pc[3] = CLAMP(pc[3], 0.0F, 1.0F);
2428 }
2429
2430 /* compare four texcoords vs. four texture samples */
2431 switch (sampler->compare_func) {
2432 case PIPE_FUNC_LESS:
2433 k[0] = pc[0] < rgba[0][0];
2434 k[1] = pc[1] < rgba[0][1];
2435 k[2] = pc[2] < rgba[0][2];
2436 k[3] = pc[3] < rgba[0][3];
2437 break;
2438 case PIPE_FUNC_LEQUAL:
2439 k[0] = pc[0] <= rgba[0][0];
2440 k[1] = pc[1] <= rgba[0][1];
2441 k[2] = pc[2] <= rgba[0][2];
2442 k[3] = pc[3] <= rgba[0][3];
2443 break;
2444 case PIPE_FUNC_GREATER:
2445 k[0] = pc[0] > rgba[0][0];
2446 k[1] = pc[1] > rgba[0][1];
2447 k[2] = pc[2] > rgba[0][2];
2448 k[3] = pc[3] > rgba[0][3];
2449 break;
2450 case PIPE_FUNC_GEQUAL:
2451 k[0] = pc[0] >= rgba[0][0];
2452 k[1] = pc[1] >= rgba[0][1];
2453 k[2] = pc[2] >= rgba[0][2];
2454 k[3] = pc[3] >= rgba[0][3];
2455 break;
2456 case PIPE_FUNC_EQUAL:
2457 k[0] = pc[0] == rgba[0][0];
2458 k[1] = pc[1] == rgba[0][1];
2459 k[2] = pc[2] == rgba[0][2];
2460 k[3] = pc[3] == rgba[0][3];
2461 break;
2462 case PIPE_FUNC_NOTEQUAL:
2463 k[0] = pc[0] != rgba[0][0];
2464 k[1] = pc[1] != rgba[0][1];
2465 k[2] = pc[2] != rgba[0][2];
2466 k[3] = pc[3] != rgba[0][3];
2467 break;
2468 case PIPE_FUNC_ALWAYS:
2469 k[0] = k[1] = k[2] = k[3] = 1;
2470 break;
2471 case PIPE_FUNC_NEVER:
2472 k[0] = k[1] = k[2] = k[3] = 0;
2473 break;
2474 default:
2475 k[0] = k[1] = k[2] = k[3] = 0;
2476 assert(0);
2477 break;
2478 }
2479
2480 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2481 rgba[0][j] = k[j];
2482 rgba[1][j] = k[j];
2483 rgba[2][j] = k[j];
2484 rgba[3][j] = 1.0F;
2485 }
2486 }
2487
2488
2489 static void
2490 do_swizzling(const struct pipe_sampler_view *sview,
2491 float in[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
2492 float out[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2493 {
2494 int j;
2495 const unsigned swizzle_r = sview->swizzle_r;
2496 const unsigned swizzle_g = sview->swizzle_g;
2497 const unsigned swizzle_b = sview->swizzle_b;
2498 const unsigned swizzle_a = sview->swizzle_a;
2499
2500 switch (swizzle_r) {
2501 case PIPE_SWIZZLE_ZERO:
2502 for (j = 0; j < 4; j++)
2503 out[0][j] = 0.0f;
2504 break;
2505 case PIPE_SWIZZLE_ONE:
2506 for (j = 0; j < 4; j++)
2507 out[0][j] = 1.0f;
2508 break;
2509 default:
2510 assert(swizzle_r < 4);
2511 for (j = 0; j < 4; j++)
2512 out[0][j] = in[swizzle_r][j];
2513 }
2514
2515 switch (swizzle_g) {
2516 case PIPE_SWIZZLE_ZERO:
2517 for (j = 0; j < 4; j++)
2518 out[1][j] = 0.0f;
2519 break;
2520 case PIPE_SWIZZLE_ONE:
2521 for (j = 0; j < 4; j++)
2522 out[1][j] = 1.0f;
2523 break;
2524 default:
2525 assert(swizzle_g < 4);
2526 for (j = 0; j < 4; j++)
2527 out[1][j] = in[swizzle_g][j];
2528 }
2529
2530 switch (swizzle_b) {
2531 case PIPE_SWIZZLE_ZERO:
2532 for (j = 0; j < 4; j++)
2533 out[2][j] = 0.0f;
2534 break;
2535 case PIPE_SWIZZLE_ONE:
2536 for (j = 0; j < 4; j++)
2537 out[2][j] = 1.0f;
2538 break;
2539 default:
2540 assert(swizzle_b < 4);
2541 for (j = 0; j < 4; j++)
2542 out[2][j] = in[swizzle_b][j];
2543 }
2544
2545 switch (swizzle_a) {
2546 case PIPE_SWIZZLE_ZERO:
2547 for (j = 0; j < 4; j++)
2548 out[3][j] = 0.0f;
2549 break;
2550 case PIPE_SWIZZLE_ONE:
2551 for (j = 0; j < 4; j++)
2552 out[3][j] = 1.0f;
2553 break;
2554 default:
2555 assert(swizzle_a < 4);
2556 for (j = 0; j < 4; j++)
2557 out[3][j] = in[swizzle_a][j];
2558 }
2559 }
2560
2561
2562 static wrap_nearest_func
2563 get_nearest_unorm_wrap(unsigned mode)
2564 {
2565 switch (mode) {
2566 case PIPE_TEX_WRAP_CLAMP:
2567 return wrap_nearest_unorm_clamp;
2568 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2569 return wrap_nearest_unorm_clamp_to_edge;
2570 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2571 return wrap_nearest_unorm_clamp_to_border;
2572 default:
2573 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
2574 return wrap_nearest_unorm_clamp;
2575 }
2576 }
2577
2578
2579 static wrap_nearest_func
2580 get_nearest_wrap(unsigned mode)
2581 {
2582 switch (mode) {
2583 case PIPE_TEX_WRAP_REPEAT:
2584 return wrap_nearest_repeat;
2585 case PIPE_TEX_WRAP_CLAMP:
2586 return wrap_nearest_clamp;
2587 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2588 return wrap_nearest_clamp_to_edge;
2589 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2590 return wrap_nearest_clamp_to_border;
2591 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2592 return wrap_nearest_mirror_repeat;
2593 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2594 return wrap_nearest_mirror_clamp;
2595 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2596 return wrap_nearest_mirror_clamp_to_edge;
2597 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2598 return wrap_nearest_mirror_clamp_to_border;
2599 default:
2600 assert(0);
2601 return wrap_nearest_repeat;
2602 }
2603 }
2604
2605
2606 static wrap_linear_func
2607 get_linear_unorm_wrap(unsigned mode)
2608 {
2609 switch (mode) {
2610 case PIPE_TEX_WRAP_CLAMP:
2611 return wrap_linear_unorm_clamp;
2612 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2613 return wrap_linear_unorm_clamp_to_edge;
2614 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2615 return wrap_linear_unorm_clamp_to_border;
2616 default:
2617 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
2618 return wrap_linear_unorm_clamp;
2619 }
2620 }
2621
2622
2623 static wrap_linear_func
2624 get_linear_wrap(unsigned mode)
2625 {
2626 switch (mode) {
2627 case PIPE_TEX_WRAP_REPEAT:
2628 return wrap_linear_repeat;
2629 case PIPE_TEX_WRAP_CLAMP:
2630 return wrap_linear_clamp;
2631 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2632 return wrap_linear_clamp_to_edge;
2633 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2634 return wrap_linear_clamp_to_border;
2635 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2636 return wrap_linear_mirror_repeat;
2637 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2638 return wrap_linear_mirror_clamp;
2639 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2640 return wrap_linear_mirror_clamp_to_edge;
2641 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2642 return wrap_linear_mirror_clamp_to_border;
2643 default:
2644 assert(0);
2645 return wrap_linear_repeat;
2646 }
2647 }
2648
2649
2650 /**
2651 * Is swizzling needed for the given state key?
2652 */
2653 static INLINE bool
2654 any_swizzle(const struct pipe_sampler_view *view)
2655 {
2656 return (view->swizzle_r != PIPE_SWIZZLE_RED ||
2657 view->swizzle_g != PIPE_SWIZZLE_GREEN ||
2658 view->swizzle_b != PIPE_SWIZZLE_BLUE ||
2659 view->swizzle_a != PIPE_SWIZZLE_ALPHA);
2660 }
2661
2662
2663 static img_filter_func
2664 get_img_filter(const struct sp_sampler_view *sp_sview,
2665 const struct pipe_sampler_state *sampler,
2666 unsigned filter)
2667 {
2668 switch (sp_sview->base.target) {
2669 case PIPE_BUFFER:
2670 case PIPE_TEXTURE_1D:
2671 if (filter == PIPE_TEX_FILTER_NEAREST)
2672 return img_filter_1d_nearest;
2673 else
2674 return img_filter_1d_linear;
2675 break;
2676 case PIPE_TEXTURE_1D_ARRAY:
2677 if (filter == PIPE_TEX_FILTER_NEAREST)
2678 return img_filter_1d_array_nearest;
2679 else
2680 return img_filter_1d_array_linear;
2681 break;
2682 case PIPE_TEXTURE_2D:
2683 case PIPE_TEXTURE_RECT:
2684 /* Try for fast path:
2685 */
2686 if (sp_sview->pot2d &&
2687 sampler->wrap_s == sampler->wrap_t &&
2688 sampler->normalized_coords)
2689 {
2690 switch (sampler->wrap_s) {
2691 case PIPE_TEX_WRAP_REPEAT:
2692 switch (filter) {
2693 case PIPE_TEX_FILTER_NEAREST:
2694 return img_filter_2d_nearest_repeat_POT;
2695 case PIPE_TEX_FILTER_LINEAR:
2696 return img_filter_2d_linear_repeat_POT;
2697 default:
2698 break;
2699 }
2700 break;
2701 case PIPE_TEX_WRAP_CLAMP:
2702 switch (filter) {
2703 case PIPE_TEX_FILTER_NEAREST:
2704 return img_filter_2d_nearest_clamp_POT;
2705 default:
2706 break;
2707 }
2708 }
2709 }
2710 /* Otherwise use default versions:
2711 */
2712 if (filter == PIPE_TEX_FILTER_NEAREST)
2713 return img_filter_2d_nearest;
2714 else
2715 return img_filter_2d_linear;
2716 break;
2717 case PIPE_TEXTURE_2D_ARRAY:
2718 if (filter == PIPE_TEX_FILTER_NEAREST)
2719 return img_filter_2d_array_nearest;
2720 else
2721 return img_filter_2d_array_linear;
2722 break;
2723 case PIPE_TEXTURE_CUBE:
2724 if (filter == PIPE_TEX_FILTER_NEAREST)
2725 return img_filter_cube_nearest;
2726 else
2727 return img_filter_cube_linear;
2728 break;
2729 case PIPE_TEXTURE_CUBE_ARRAY:
2730 if (filter == PIPE_TEX_FILTER_NEAREST)
2731 return img_filter_cube_array_nearest;
2732 else
2733 return img_filter_cube_array_linear;
2734 break;
2735 case PIPE_TEXTURE_3D:
2736 if (filter == PIPE_TEX_FILTER_NEAREST)
2737 return img_filter_3d_nearest;
2738 else
2739 return img_filter_3d_linear;
2740 break;
2741 default:
2742 assert(0);
2743 return img_filter_1d_nearest;
2744 }
2745 }
2746
2747
2748 static void
2749 sample_mip(struct sp_sampler_view *sp_sview,
2750 struct sp_sampler *sp_samp,
2751 const float s[TGSI_QUAD_SIZE],
2752 const float t[TGSI_QUAD_SIZE],
2753 const float p[TGSI_QUAD_SIZE],
2754 const float c0[TGSI_QUAD_SIZE],
2755 const float lod[TGSI_QUAD_SIZE],
2756 enum tgsi_sampler_control control,
2757 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2758 {
2759 mip_filter_func mip_filter;
2760 img_filter_func min_img_filter = NULL;
2761 img_filter_func mag_img_filter = NULL;
2762
2763 if (sp_sview->pot2d & sp_samp->min_mag_equal_repeat_linear) {
2764 mip_filter = mip_filter_linear_2d_linear_repeat_POT;
2765 }
2766 else {
2767 mip_filter = sp_samp->mip_filter;
2768 min_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->min_img_filter);
2769 if (sp_samp->min_mag_equal) {
2770 mag_img_filter = min_img_filter;
2771 }
2772 else {
2773 mag_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->base.mag_img_filter);
2774 }
2775 }
2776
2777 mip_filter(sp_sview, sp_samp, min_img_filter, mag_img_filter,
2778 s, t, p, c0, lod, control, rgba);
2779
2780 if (sp_samp->base.compare_mode != PIPE_TEX_COMPARE_NONE) {
2781 sample_compare(sp_sview, sp_samp, s, t, p, c0, lod, control, rgba);
2782 }
2783
2784 if (sp_sview->need_swizzle) {
2785 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2786 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
2787 do_swizzling(&sp_sview->base, rgba_temp, rgba);
2788 }
2789
2790 }
2791
2792
2793 /**
2794 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2795 * Put face info into the sampler faces[] array.
2796 */
2797 static void
2798 sample_cube(struct sp_sampler_view *sp_sview,
2799 struct sp_sampler *sp_samp,
2800 const float s[TGSI_QUAD_SIZE],
2801 const float t[TGSI_QUAD_SIZE],
2802 const float p[TGSI_QUAD_SIZE],
2803 const float c0[TGSI_QUAD_SIZE],
2804 const float c1[TGSI_QUAD_SIZE],
2805 enum tgsi_sampler_control control,
2806 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2807 {
2808 unsigned j;
2809 float ssss[4], tttt[4];
2810
2811 /* Not actually used, but the intermediate steps that do the
2812 * dereferencing don't know it.
2813 */
2814 static float pppp[4] = { 0, 0, 0, 0 };
2815
2816 pppp[0] = c0[0];
2817 pppp[1] = c0[1];
2818 pppp[2] = c0[2];
2819 pppp[3] = c0[3];
2820 /*
2821 major axis
2822 direction target sc tc ma
2823 ---------- ------------------------------- --- --- ---
2824 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2825 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2826 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2827 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2828 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2829 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2830 */
2831
2832 /* Choose the cube face and compute new s/t coords for the 2D face.
2833 *
2834 * Use the same cube face for all four pixels in the quad.
2835 *
2836 * This isn't ideal, but if we want to use a different cube face
2837 * per pixel in the quad, we'd have to also compute the per-face
2838 * LOD here too. That's because the four post-face-selection
2839 * texcoords are no longer related to each other (they're
2840 * per-face!) so we can't use subtraction to compute the partial
2841 * deriviates to compute the LOD. Doing so (near cube edges
2842 * anyway) gives us pretty much random values.
2843 */
2844 {
2845 /* use the average of the four pixel's texcoords to choose the face */
2846 const float rx = 0.25F * (s[0] + s[1] + s[2] + s[3]);
2847 const float ry = 0.25F * (t[0] + t[1] + t[2] + t[3]);
2848 const float rz = 0.25F * (p[0] + p[1] + p[2] + p[3]);
2849 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
2850
2851 if (arx >= ary && arx >= arz) {
2852 float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
2853 uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
2854 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2855 const float ima = -0.5F / fabsf(s[j]);
2856 ssss[j] = sign * p[j] * ima + 0.5F;
2857 tttt[j] = t[j] * ima + 0.5F;
2858 sp_sview->faces[j] = face;
2859 }
2860 }
2861 else if (ary >= arx && ary >= arz) {
2862 float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
2863 uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
2864 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2865 const float ima = -0.5F / fabsf(t[j]);
2866 ssss[j] = -s[j] * ima + 0.5F;
2867 tttt[j] = sign * -p[j] * ima + 0.5F;
2868 sp_sview->faces[j] = face;
2869 }
2870 }
2871 else {
2872 float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
2873 uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
2874 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2875 const float ima = -0.5F / fabsf(p[j]);
2876 ssss[j] = sign * -s[j] * ima + 0.5F;
2877 tttt[j] = t[j] * ima + 0.5F;
2878 sp_sview->faces[j] = face;
2879 }
2880 }
2881 }
2882
2883 sample_mip(sp_sview, sp_samp, ssss, tttt, pppp, c0, c1, control, rgba);
2884 }
2885
2886
2887 static void
2888 sp_get_dims(struct sp_sampler_view *sp_sview, int level,
2889 int dims[4])
2890 {
2891 const struct pipe_sampler_view *view = &sp_sview->base;
2892 const struct pipe_resource *texture = view->texture;
2893
2894 if (view->target == PIPE_BUFFER) {
2895 dims[0] = (view->u.buf.last_element - view->u.buf.first_element) + 1;
2896 /* the other values are undefined, but let's avoid potential valgrind
2897 * warnings.
2898 */
2899 dims[1] = dims[2] = dims[3] = 0;
2900 return;
2901 }
2902
2903 /* undefined according to EXT_gpu_program */
2904 level += view->u.tex.first_level;
2905 if (level > view->u.tex.last_level)
2906 return;
2907
2908 dims[3] = view->u.tex.last_level - view->u.tex.first_level + 1;
2909 dims[0] = u_minify(texture->width0, level);
2910
2911 switch (view->target) {
2912 case PIPE_TEXTURE_1D_ARRAY:
2913 dims[1] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
2914 /* fallthrough */
2915 case PIPE_TEXTURE_1D:
2916 return;
2917 case PIPE_TEXTURE_2D_ARRAY:
2918 dims[2] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
2919 /* fallthrough */
2920 case PIPE_TEXTURE_2D:
2921 case PIPE_TEXTURE_CUBE:
2922 case PIPE_TEXTURE_RECT:
2923 dims[1] = u_minify(texture->height0, level);
2924 return;
2925 case PIPE_TEXTURE_3D:
2926 dims[1] = u_minify(texture->height0, level);
2927 dims[2] = u_minify(texture->depth0, level);
2928 return;
2929 case PIPE_TEXTURE_CUBE_ARRAY:
2930 dims[1] = u_minify(texture->height0, level);
2931 dims[2] = (view->u.tex.last_layer - view->u.tex.first_layer + 1) / 6;
2932 break;
2933 default:
2934 assert(!"unexpected texture target in sp_get_dims()");
2935 return;
2936 }
2937 }
2938
2939 /**
2940 * This function is only used for getting unfiltered texels via the
2941 * TXF opcode. The GL spec says that out-of-bounds texel fetches
2942 * produce undefined results. Instead of crashing, lets just clamp
2943 * coords to the texture image size.
2944 */
2945 static void
2946 sp_get_texels(struct sp_sampler_view *sp_sview,
2947 const int v_i[TGSI_QUAD_SIZE],
2948 const int v_j[TGSI_QUAD_SIZE],
2949 const int v_k[TGSI_QUAD_SIZE],
2950 const int lod[TGSI_QUAD_SIZE],
2951 const int8_t offset[3],
2952 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2953 {
2954 union tex_tile_address addr;
2955 const struct pipe_resource *texture = sp_sview->base.texture;
2956 int j, c;
2957 const float *tx;
2958 int width, height, depth;
2959
2960 addr.value = 0;
2961 /* TODO write a better test for LOD */
2962 addr.bits.level = sp_sview->base.target == PIPE_BUFFER ? 0 :
2963 CLAMP(lod[0] + sp_sview->base.u.tex.first_level,
2964 sp_sview->base.u.tex.first_level,
2965 sp_sview->base.u.tex.last_level);
2966
2967 width = u_minify(texture->width0, addr.bits.level);
2968 height = u_minify(texture->height0, addr.bits.level);
2969 depth = u_minify(texture->depth0, addr.bits.level);
2970
2971 switch (sp_sview->base.target) {
2972 case PIPE_BUFFER:
2973 case PIPE_TEXTURE_1D:
2974 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2975 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2976 tx = get_texel_2d_no_border(sp_sview, addr, x, 0);
2977 for (c = 0; c < 4; c++) {
2978 rgba[c][j] = tx[c];
2979 }
2980 }
2981 break;
2982 case PIPE_TEXTURE_1D_ARRAY:
2983 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2984 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2985 int y = CLAMP(v_j[j], sp_sview->base.u.tex.first_layer,
2986 sp_sview->base.u.tex.last_layer);
2987 tx = get_texel_2d_no_border(sp_sview, addr, x, y);
2988 for (c = 0; c < 4; c++) {
2989 rgba[c][j] = tx[c];
2990 }
2991 }
2992 break;
2993 case PIPE_TEXTURE_2D:
2994 case PIPE_TEXTURE_RECT:
2995 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2996 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
2997 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
2998 tx = get_texel_2d_no_border(sp_sview, addr, x, y);
2999 for (c = 0; c < 4; c++) {
3000 rgba[c][j] = tx[c];
3001 }
3002 }
3003 break;
3004 case PIPE_TEXTURE_2D_ARRAY:
3005 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3006 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3007 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3008 int layer = CLAMP(v_k[j], sp_sview->base.u.tex.first_layer,
3009 sp_sview->base.u.tex.last_layer);
3010 tx = get_texel_3d_no_border(sp_sview, addr, x, y, layer);
3011 for (c = 0; c < 4; c++) {
3012 rgba[c][j] = tx[c];
3013 }
3014 }
3015 break;
3016 case PIPE_TEXTURE_3D:
3017 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3018 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3019 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3020 int z = CLAMP(v_k[j] + offset[2], 0, depth - 1);
3021 tx = get_texel_3d_no_border(sp_sview, addr, x, y, z);
3022 for (c = 0; c < 4; c++) {
3023 rgba[c][j] = tx[c];
3024 }
3025 }
3026 break;
3027 case PIPE_TEXTURE_CUBE: /* TXF can't work on CUBE according to spec */
3028 default:
3029 assert(!"Unknown or CUBE texture type in TXF processing\n");
3030 break;
3031 }
3032
3033 if (sp_sview->need_swizzle) {
3034 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
3035 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
3036 do_swizzling(&sp_sview->base, rgba_temp, rgba);
3037 }
3038 }
3039
3040
3041 void *
3042 softpipe_create_sampler_state(struct pipe_context *pipe,
3043 const struct pipe_sampler_state *sampler)
3044 {
3045 struct sp_sampler *samp = CALLOC_STRUCT(sp_sampler);
3046
3047 samp->base = *sampler;
3048
3049 /* Note that (for instance) linear_texcoord_s and
3050 * nearest_texcoord_s may be active at the same time, if the
3051 * sampler min_img_filter differs from its mag_img_filter.
3052 */
3053 if (sampler->normalized_coords) {
3054 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
3055 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
3056 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
3057
3058 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
3059 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
3060 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
3061 }
3062 else {
3063 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
3064 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
3065 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
3066
3067 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
3068 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
3069 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
3070 }
3071
3072 samp->min_img_filter = sampler->min_img_filter;
3073
3074 switch (sampler->min_mip_filter) {
3075 case PIPE_TEX_MIPFILTER_NONE:
3076 if (sampler->min_img_filter == sampler->mag_img_filter)
3077 samp->mip_filter = mip_filter_none_no_filter_select;
3078 else
3079 samp->mip_filter = mip_filter_none;
3080 break;
3081
3082 case PIPE_TEX_MIPFILTER_NEAREST:
3083 samp->mip_filter = mip_filter_nearest;
3084 break;
3085
3086 case PIPE_TEX_MIPFILTER_LINEAR:
3087 if (sampler->min_img_filter == sampler->mag_img_filter &&
3088 sampler->normalized_coords &&
3089 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
3090 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
3091 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR &&
3092 sampler->max_anisotropy <= 1) {
3093 samp->min_mag_equal_repeat_linear = TRUE;
3094 }
3095 samp->mip_filter = mip_filter_linear;
3096
3097 /* Anisotropic filtering extension. */
3098 if (sampler->max_anisotropy > 1) {
3099 samp->mip_filter = mip_filter_linear_aniso;
3100
3101 /* Override min_img_filter:
3102 * min_img_filter needs to be set to NEAREST since we need to access
3103 * each texture pixel as it is and weight it later; using linear
3104 * filters will have incorrect results.
3105 * By setting the filter to NEAREST here, we can avoid calling the
3106 * generic img_filter_2d_nearest in the anisotropic filter function,
3107 * making it possible to use one of the accelerated implementations
3108 */
3109 samp->min_img_filter = PIPE_TEX_FILTER_NEAREST;
3110
3111 /* on first access create the lookup table containing the filter weights. */
3112 if (!weightLut) {
3113 create_filter_table();
3114 }
3115 }
3116 break;
3117 }
3118 if (samp->min_img_filter == sampler->mag_img_filter) {
3119 samp->min_mag_equal = TRUE;
3120 }
3121
3122 return (void *)samp;
3123 }
3124
3125
3126 compute_lambda_func
3127 softpipe_get_lambda_func(const struct pipe_sampler_view *view, unsigned shader)
3128 {
3129 if (shader != PIPE_SHADER_FRAGMENT)
3130 return compute_lambda_vert;
3131
3132 switch (view->target) {
3133 case PIPE_BUFFER:
3134 case PIPE_TEXTURE_1D:
3135 case PIPE_TEXTURE_1D_ARRAY:
3136 return compute_lambda_1d;
3137 case PIPE_TEXTURE_2D:
3138 case PIPE_TEXTURE_2D_ARRAY:
3139 case PIPE_TEXTURE_RECT:
3140 case PIPE_TEXTURE_CUBE:
3141 case PIPE_TEXTURE_CUBE_ARRAY:
3142 return compute_lambda_2d;
3143 case PIPE_TEXTURE_3D:
3144 return compute_lambda_3d;
3145 default:
3146 assert(0);
3147 return compute_lambda_1d;
3148 }
3149 }
3150
3151
3152 struct pipe_sampler_view *
3153 softpipe_create_sampler_view(struct pipe_context *pipe,
3154 struct pipe_resource *resource,
3155 const struct pipe_sampler_view *templ)
3156 {
3157 struct sp_sampler_view *sview = CALLOC_STRUCT(sp_sampler_view);
3158 struct softpipe_resource *spr = (struct softpipe_resource *)resource;
3159
3160 if (sview) {
3161 struct pipe_sampler_view *view = &sview->base;
3162 *view = *templ;
3163 view->reference.count = 1;
3164 view->texture = NULL;
3165 pipe_resource_reference(&view->texture, resource);
3166 view->context = pipe;
3167
3168 #ifdef DEBUG
3169 /*
3170 * This is possibly too lenient, but the primary reason is just
3171 * to catch state trackers which forget to initialize this, so
3172 * it only catches clearly impossible view targets.
3173 */
3174 if (view->target != resource->target) {
3175 if (view->target == PIPE_TEXTURE_1D)
3176 assert(resource->target == PIPE_TEXTURE_1D_ARRAY);
3177 else if (view->target == PIPE_TEXTURE_1D_ARRAY)
3178 assert(resource->target == PIPE_TEXTURE_1D);
3179 else if (view->target == PIPE_TEXTURE_2D)
3180 assert(resource->target == PIPE_TEXTURE_2D_ARRAY ||
3181 resource->target == PIPE_TEXTURE_CUBE ||
3182 resource->target == PIPE_TEXTURE_CUBE_ARRAY);
3183 else if (view->target == PIPE_TEXTURE_2D_ARRAY)
3184 assert(resource->target == PIPE_TEXTURE_2D ||
3185 resource->target == PIPE_TEXTURE_CUBE ||
3186 resource->target == PIPE_TEXTURE_CUBE_ARRAY);
3187 else if (view->target == PIPE_TEXTURE_CUBE)
3188 assert(resource->target == PIPE_TEXTURE_CUBE_ARRAY ||
3189 resource->target == PIPE_TEXTURE_2D_ARRAY);
3190 else if (view->target == PIPE_TEXTURE_CUBE_ARRAY)
3191 assert(resource->target == PIPE_TEXTURE_CUBE ||
3192 resource->target == PIPE_TEXTURE_2D_ARRAY);
3193 else
3194 assert(0);
3195 }
3196 #endif
3197
3198 if (any_swizzle(view)) {
3199 sview->need_swizzle = TRUE;
3200 }
3201
3202 if (view->target == PIPE_TEXTURE_CUBE ||
3203 view->target == PIPE_TEXTURE_CUBE_ARRAY)
3204 sview->get_samples = sample_cube;
3205 else {
3206 sview->get_samples = sample_mip;
3207 }
3208 sview->pot2d = spr->pot &&
3209 (view->target == PIPE_TEXTURE_2D ||
3210 view->target == PIPE_TEXTURE_RECT);
3211
3212 sview->xpot = util_logbase2( resource->width0 );
3213 sview->ypot = util_logbase2( resource->height0 );
3214 }
3215
3216 return (struct pipe_sampler_view *) sview;
3217 }
3218
3219
3220 static void
3221 sp_tgsi_get_dims(struct tgsi_sampler *tgsi_sampler,
3222 const unsigned sview_index,
3223 int level, int dims[4])
3224 {
3225 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3226
3227 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3228 /* always have a view here but texture is NULL if no sampler view was set. */
3229 if (!sp_samp->sp_sview[sview_index].base.texture) {
3230 dims[0] = dims[1] = dims[2] = dims[3] = 0;
3231 return;
3232 }
3233 sp_get_dims(&sp_samp->sp_sview[sview_index], level, dims);
3234 }
3235
3236
3237 static void
3238 sp_tgsi_get_samples(struct tgsi_sampler *tgsi_sampler,
3239 const unsigned sview_index,
3240 const unsigned sampler_index,
3241 const float s[TGSI_QUAD_SIZE],
3242 const float t[TGSI_QUAD_SIZE],
3243 const float p[TGSI_QUAD_SIZE],
3244 const float c0[TGSI_QUAD_SIZE],
3245 const float lod[TGSI_QUAD_SIZE],
3246 float derivs[3][2][TGSI_QUAD_SIZE],
3247 const int8_t offset[3],
3248 enum tgsi_sampler_control control,
3249 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
3250 {
3251 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3252
3253 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3254 assert(sampler_index < PIPE_MAX_SAMPLERS);
3255 assert(sp_samp->sp_sampler[sampler_index]);
3256 /* always have a view here but texture is NULL if no sampler view was set. */
3257 if (!sp_samp->sp_sview[sview_index].base.texture) {
3258 int i, j;
3259 for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
3260 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
3261 rgba[j][i] = 0.0f;
3262 }
3263 }
3264 return;
3265 }
3266 sp_samp->sp_sview[sview_index].get_samples(&sp_samp->sp_sview[sview_index],
3267 sp_samp->sp_sampler[sampler_index],
3268 s, t, p, c0, lod, control, rgba);
3269 }
3270
3271
3272 static void
3273 sp_tgsi_get_texel(struct tgsi_sampler *tgsi_sampler,
3274 const unsigned sview_index,
3275 const int i[TGSI_QUAD_SIZE],
3276 const int j[TGSI_QUAD_SIZE], const int k[TGSI_QUAD_SIZE],
3277 const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
3278 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
3279 {
3280 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3281
3282 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3283 /* always have a view here but texture is NULL if no sampler view was set. */
3284 if (!sp_samp->sp_sview[sview_index].base.texture) {
3285 int i, j;
3286 for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
3287 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
3288 rgba[j][i] = 0.0f;
3289 }
3290 }
3291 return;
3292 }
3293 sp_get_texels(&sp_samp->sp_sview[sview_index], i, j, k, lod, offset, rgba);
3294 }
3295
3296
3297 struct sp_tgsi_sampler *
3298 sp_create_tgsi_sampler(void)
3299 {
3300 struct sp_tgsi_sampler *samp = CALLOC_STRUCT(sp_tgsi_sampler);
3301 if (!samp)
3302 return NULL;
3303
3304 samp->base.get_dims = sp_tgsi_get_dims;
3305 samp->base.get_samples = sp_tgsi_get_samples;
3306 samp->base.get_texel = sp_tgsi_get_texel;
3307
3308 return samp;
3309 }
3310