projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
softpipe: enable ARB_texture_view
[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 float s,
1011 float t,
1012 float p,
1013 unsigned level,
1014 unsigned face_id,
1015 float *rgba)
1016 {
1017 unsigned xpot = pot_level_size(sp_sview->xpot, level);
1018 unsigned ypot = pot_level_size(sp_sview->ypot, level);
1019 int xmax = (xpot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
1020 int ymax = (ypot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
1021 union tex_tile_address addr;
1022 int c;
1023
1024 float u = s * xpot - 0.5F;
1025 float v = t * ypot - 0.5F;
1026
1027 int uflr = util_ifloor(u);
1028 int vflr = util_ifloor(v);
1029
1030 float xw = u - (float)uflr;
1031 float yw = v - (float)vflr;
1032
1033 int x0 = uflr & (xpot - 1);
1034 int y0 = vflr & (ypot - 1);
1035
1036 const float *tx[4];
1037
1038 addr.value = 0;
1039 addr.bits.level = level;
1040
1041 /* Can we fetch all four at once:
1042 */
1043 if (x0 < xmax && y0 < ymax) {
1044 get_texel_quad_2d_no_border_single_tile(sp_sview, addr, x0, y0, tx);
1045 }
1046 else {
1047 unsigned x1 = (x0 + 1) & (xpot - 1);
1048 unsigned y1 = (y0 + 1) & (ypot - 1);
1049 get_texel_quad_2d_no_border(sp_sview, addr, x0, y0, x1, y1, tx);
1050 }
1051
1052 /* interpolate R, G, B, A */
1053 for (c = 0; c < TGSI_QUAD_SIZE; c++) {
1054 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1055 tx[0][c], tx[1][c],
1056 tx[2][c], tx[3][c]);
1057 }
1058
1059 if (DEBUG_TEX) {
1060 print_sample(__FUNCTION__, rgba);
1061 }
1062 }
1063
1064
1065 static INLINE void
1066 img_filter_2d_nearest_repeat_POT(struct sp_sampler_view *sp_sview,
1067 struct sp_sampler *sp_samp,
1068 float s,
1069 float t,
1070 float p,
1071 unsigned level,
1072 unsigned face_id,
1073 float rgba[TGSI_QUAD_SIZE])
1074 {
1075 unsigned xpot = pot_level_size(sp_sview->xpot, level);
1076 unsigned ypot = pot_level_size(sp_sview->ypot, level);
1077 const float *out;
1078 union tex_tile_address addr;
1079 int c;
1080
1081 float u = s * xpot;
1082 float v = t * ypot;
1083
1084 int uflr = util_ifloor(u);
1085 int vflr = util_ifloor(v);
1086
1087 int x0 = uflr & (xpot - 1);
1088 int y0 = vflr & (ypot - 1);
1089
1090 addr.value = 0;
1091 addr.bits.level = level;
1092
1093 out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
1094 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1095 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1096
1097 if (DEBUG_TEX) {
1098 print_sample(__FUNCTION__, rgba);
1099 }
1100 }
1101
1102
1103 static INLINE void
1104 img_filter_2d_nearest_clamp_POT(struct sp_sampler_view *sp_sview,
1105 struct sp_sampler *sp_samp,
1106 float s,
1107 float t,
1108 float p,
1109 unsigned level,
1110 unsigned face_id,
1111 float rgba[TGSI_QUAD_SIZE])
1112 {
1113 unsigned xpot = pot_level_size(sp_sview->xpot, level);
1114 unsigned ypot = pot_level_size(sp_sview->ypot, level);
1115 union tex_tile_address addr;
1116 int c;
1117
1118 float u = s * xpot;
1119 float v = t * ypot;
1120
1121 int x0, y0;
1122 const float *out;
1123
1124 addr.value = 0;
1125 addr.bits.level = level;
1126
1127 x0 = util_ifloor(u);
1128 if (x0 < 0)
1129 x0 = 0;
1130 else if (x0 > (int) xpot - 1)
1131 x0 = xpot - 1;
1132
1133 y0 = util_ifloor(v);
1134 if (y0 < 0)
1135 y0 = 0;
1136 else if (y0 > (int) ypot - 1)
1137 y0 = ypot - 1;
1138
1139 out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
1140 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1141 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1142
1143 if (DEBUG_TEX) {
1144 print_sample(__FUNCTION__, rgba);
1145 }
1146 }
1147
1148
1149 static void
1150 img_filter_1d_nearest(struct sp_sampler_view *sp_sview,
1151 struct sp_sampler *sp_samp,
1152 float s,
1153 float t,
1154 float p,
1155 unsigned level,
1156 unsigned face_id,
1157 float rgba[TGSI_QUAD_SIZE])
1158 {
1159 const struct pipe_resource *texture = sp_sview->base.texture;
1160 int width;
1161 int x;
1162 union tex_tile_address addr;
1163 const float *out;
1164 int c;
1165
1166 width = u_minify(texture->width0, level);
1167
1168 assert(width > 0);
1169
1170 addr.value = 0;
1171 addr.bits.level = level;
1172
1173 sp_samp->nearest_texcoord_s(s, width, &x);
1174
1175 out = get_texel_2d(sp_sview, sp_samp, addr, x, 0);
1176 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1177 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1178
1179 if (DEBUG_TEX) {
1180 print_sample(__FUNCTION__, rgba);
1181 }
1182 }
1183
1184
1185 static void
1186 img_filter_1d_array_nearest(struct sp_sampler_view *sp_sview,
1187 struct sp_sampler *sp_samp,
1188 float s,
1189 float t,
1190 float p,
1191 unsigned level,
1192 unsigned face_id,
1193 float *rgba)
1194 {
1195 const struct pipe_resource *texture = sp_sview->base.texture;
1196 int width;
1197 int x, layer;
1198 union tex_tile_address addr;
1199 const float *out;
1200 int c;
1201
1202 width = u_minify(texture->width0, level);
1203
1204 assert(width > 0);
1205
1206 addr.value = 0;
1207 addr.bits.level = level;
1208
1209 sp_samp->nearest_texcoord_s(s, width, &x);
1210 layer = coord_to_layer(t, sp_sview->base.u.tex.first_layer,
1211 sp_sview->base.u.tex.last_layer);
1212
1213 out = get_texel_1d_array(sp_sview, sp_samp, addr, x, layer);
1214 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1215 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1216
1217 if (DEBUG_TEX) {
1218 print_sample(__FUNCTION__, rgba);
1219 }
1220 }
1221
1222
1223 static void
1224 img_filter_2d_nearest(struct sp_sampler_view *sp_sview,
1225 struct sp_sampler *sp_samp,
1226 float s,
1227 float t,
1228 float p,
1229 unsigned level,
1230 unsigned face_id,
1231 float *rgba)
1232 {
1233 const struct pipe_resource *texture = sp_sview->base.texture;
1234 int width, height;
1235 int x, y;
1236 union tex_tile_address addr;
1237 const float *out;
1238 int c;
1239
1240 width = u_minify(texture->width0, level);
1241 height = u_minify(texture->height0, level);
1242
1243 assert(width > 0);
1244 assert(height > 0);
1245
1246 addr.value = 0;
1247 addr.bits.level = level;
1248
1249 sp_samp->nearest_texcoord_s(s, width, &x);
1250 sp_samp->nearest_texcoord_t(t, height, &y);
1251
1252 out = get_texel_2d(sp_sview, sp_samp, addr, x, y);
1253 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1254 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1255
1256 if (DEBUG_TEX) {
1257 print_sample(__FUNCTION__, rgba);
1258 }
1259 }
1260
1261
1262 static void
1263 img_filter_2d_array_nearest(struct sp_sampler_view *sp_sview,
1264 struct sp_sampler *sp_samp,
1265 float s,
1266 float t,
1267 float p,
1268 unsigned level,
1269 unsigned face_id,
1270 float *rgba)
1271 {
1272 const struct pipe_resource *texture = sp_sview->base.texture;
1273 int width, height;
1274 int x, y, layer;
1275 union tex_tile_address addr;
1276 const float *out;
1277 int c;
1278
1279 width = u_minify(texture->width0, level);
1280 height = u_minify(texture->height0, level);
1281
1282 assert(width > 0);
1283 assert(height > 0);
1284
1285 addr.value = 0;
1286 addr.bits.level = level;
1287
1288 sp_samp->nearest_texcoord_s(s, width, &x);
1289 sp_samp->nearest_texcoord_t(t, height, &y);
1290 layer = coord_to_layer(p, sp_sview->base.u.tex.first_layer,
1291 sp_sview->base.u.tex.last_layer);
1292
1293 out = get_texel_2d_array(sp_sview, sp_samp, addr, x, y, layer);
1294 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1295 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1296
1297 if (DEBUG_TEX) {
1298 print_sample(__FUNCTION__, rgba);
1299 }
1300 }
1301
1302
1303 static void
1304 img_filter_cube_nearest(struct sp_sampler_view *sp_sview,
1305 struct sp_sampler *sp_samp,
1306 float s,
1307 float t,
1308 float p,
1309 unsigned level,
1310 unsigned face_id,
1311 float *rgba)
1312 {
1313 const struct pipe_resource *texture = sp_sview->base.texture;
1314 int width, height;
1315 int x, y, layerface;
1316 union tex_tile_address addr;
1317 const float *out;
1318 int c;
1319
1320 width = u_minify(texture->width0, level);
1321 height = u_minify(texture->height0, level);
1322
1323 assert(width > 0);
1324 assert(height > 0);
1325
1326 addr.value = 0;
1327 addr.bits.level = level;
1328
1329 /*
1330 * If NEAREST filtering is done within a miplevel, always apply wrap
1331 * mode CLAMP_TO_EDGE.
1332 */
1333 if (sp_samp->base.seamless_cube_map) {
1334 wrap_nearest_clamp_to_edge(s, width, &x);
1335 wrap_nearest_clamp_to_edge(t, height, &y);
1336 } else {
1337 /* Would probably make sense to ignore mode and just do edge clamp */
1338 sp_samp->nearest_texcoord_s(s, width, &x);
1339 sp_samp->nearest_texcoord_t(t, height, &y);
1340 }
1341
1342 layerface = face_id + sp_sview->base.u.tex.first_layer;
1343 out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
1344 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1345 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1346
1347 if (DEBUG_TEX) {
1348 print_sample(__FUNCTION__, rgba);
1349 }
1350 }
1351
1352 static void
1353 img_filter_cube_array_nearest(struct sp_sampler_view *sp_sview,
1354 struct sp_sampler *sp_samp,
1355 float s,
1356 float t,
1357 float p,
1358 unsigned level,
1359 unsigned face_id,
1360 float *rgba)
1361 {
1362 const struct pipe_resource *texture = sp_sview->base.texture;
1363 int width, height;
1364 int x, y, layerface;
1365 union tex_tile_address addr;
1366 const float *out;
1367 int c;
1368
1369 width = u_minify(texture->width0, level);
1370 height = u_minify(texture->height0, level);
1371
1372 assert(width > 0);
1373 assert(height > 0);
1374
1375 addr.value = 0;
1376 addr.bits.level = level;
1377
1378 sp_samp->nearest_texcoord_s(s, width, &x);
1379 sp_samp->nearest_texcoord_t(t, height, &y);
1380 layerface = coord_to_layer(6 * p + sp_sview->base.u.tex.first_layer,
1381 sp_sview->base.u.tex.first_layer,
1382 sp_sview->base.u.tex.last_layer - 5) + face_id;
1383
1384 out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
1385 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1386 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1387
1388 if (DEBUG_TEX) {
1389 print_sample(__FUNCTION__, rgba);
1390 }
1391 }
1392
1393 static void
1394 img_filter_3d_nearest(struct sp_sampler_view *sp_sview,
1395 struct sp_sampler *sp_samp,
1396 float s,
1397 float t,
1398 float p,
1399 unsigned level,
1400 unsigned face_id,
1401 float *rgba)
1402 {
1403 const struct pipe_resource *texture = sp_sview->base.texture;
1404 int width, height, depth;
1405 int x, y, z;
1406 union tex_tile_address addr;
1407 const float *out;
1408 int c;
1409
1410 width = u_minify(texture->width0, level);
1411 height = u_minify(texture->height0, level);
1412 depth = u_minify(texture->depth0, level);
1413
1414 assert(width > 0);
1415 assert(height > 0);
1416 assert(depth > 0);
1417
1418 sp_samp->nearest_texcoord_s(s, width, &x);
1419 sp_samp->nearest_texcoord_t(t, height, &y);
1420 sp_samp->nearest_texcoord_p(p, depth, &z);
1421
1422 addr.value = 0;
1423 addr.bits.level = level;
1424
1425 out = get_texel_3d(sp_sview, sp_samp, addr, x, y, z);
1426 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1427 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1428 }
1429
1430
1431 static void
1432 img_filter_1d_linear(struct sp_sampler_view *sp_sview,
1433 struct sp_sampler *sp_samp,
1434 float s,
1435 float t,
1436 float p,
1437 unsigned level,
1438 unsigned face_id,
1439 float *rgba)
1440 {
1441 const struct pipe_resource *texture = sp_sview->base.texture;
1442 int width;
1443 int x0, x1;
1444 float xw; /* weights */
1445 union tex_tile_address addr;
1446 const float *tx0, *tx1;
1447 int c;
1448
1449 width = u_minify(texture->width0, level);
1450
1451 assert(width > 0);
1452
1453 addr.value = 0;
1454 addr.bits.level = level;
1455
1456 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1457
1458 tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, 0);
1459 tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, 0);
1460
1461 /* interpolate R, G, B, A */
1462 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1463 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1464 }
1465
1466
1467 static void
1468 img_filter_1d_array_linear(struct sp_sampler_view *sp_sview,
1469 struct sp_sampler *sp_samp,
1470 float s,
1471 float t,
1472 float p,
1473 unsigned level,
1474 unsigned face_id,
1475 float *rgba)
1476 {
1477 const struct pipe_resource *texture = sp_sview->base.texture;
1478 int width;
1479 int x0, x1, layer;
1480 float xw; /* weights */
1481 union tex_tile_address addr;
1482 const float *tx0, *tx1;
1483 int c;
1484
1485 width = u_minify(texture->width0, level);
1486
1487 assert(width > 0);
1488
1489 addr.value = 0;
1490 addr.bits.level = level;
1491
1492 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1493 layer = coord_to_layer(t, sp_sview->base.u.tex.first_layer,
1494 sp_sview->base.u.tex.last_layer);
1495
1496 tx0 = get_texel_1d_array(sp_sview, sp_samp, addr, x0, layer);
1497 tx1 = get_texel_1d_array(sp_sview, sp_samp, addr, x1, layer);
1498
1499 /* interpolate R, G, B, A */
1500 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1501 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1502 }
1503
1504
1505 static void
1506 img_filter_2d_linear(struct sp_sampler_view *sp_sview,
1507 struct sp_sampler *sp_samp,
1508 float s,
1509 float t,
1510 float p,
1511 unsigned level,
1512 unsigned face_id,
1513 float *rgba)
1514 {
1515 const struct pipe_resource *texture = sp_sview->base.texture;
1516 int width, height;
1517 int x0, y0, x1, y1;
1518 float xw, yw; /* weights */
1519 union tex_tile_address addr;
1520 const float *tx0, *tx1, *tx2, *tx3;
1521 int c;
1522
1523 width = u_minify(texture->width0, level);
1524 height = u_minify(texture->height0, level);
1525
1526 assert(width > 0);
1527 assert(height > 0);
1528
1529 addr.value = 0;
1530 addr.bits.level = level;
1531
1532 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1533 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1534
1535 tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, y0);
1536 tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, y0);
1537 tx2 = get_texel_2d(sp_sview, sp_samp, addr, x0, y1);
1538 tx3 = get_texel_2d(sp_sview, sp_samp, addr, x1, y1);
1539
1540 /* interpolate R, G, B, A */
1541 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1542 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1543 tx0[c], tx1[c],
1544 tx2[c], tx3[c]);
1545 }
1546
1547
1548 static void
1549 img_filter_2d_array_linear(struct sp_sampler_view *sp_sview,
1550 struct sp_sampler *sp_samp,
1551 float s,
1552 float t,
1553 float p,
1554 unsigned level,
1555 unsigned face_id,
1556 float *rgba)
1557 {
1558 const struct pipe_resource *texture = sp_sview->base.texture;
1559 int width, height;
1560 int x0, y0, x1, y1, layer;
1561 float xw, yw; /* weights */
1562 union tex_tile_address addr;
1563 const float *tx0, *tx1, *tx2, *tx3;
1564 int c;
1565
1566 width = u_minify(texture->width0, level);
1567 height = u_minify(texture->height0, level);
1568
1569 assert(width > 0);
1570 assert(height > 0);
1571
1572 addr.value = 0;
1573 addr.bits.level = level;
1574
1575 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1576 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1577 layer = coord_to_layer(p, sp_sview->base.u.tex.first_layer,
1578 sp_sview->base.u.tex.last_layer);
1579
1580 tx0 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y0, layer);
1581 tx1 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y0, layer);
1582 tx2 = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y1, layer);
1583 tx3 = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y1, layer);
1584
1585 /* interpolate R, G, B, A */
1586 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1587 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1588 tx0[c], tx1[c],
1589 tx2[c], tx3[c]);
1590 }
1591
1592
1593 static void
1594 img_filter_cube_linear(struct sp_sampler_view *sp_sview,
1595 struct sp_sampler *sp_samp,
1596 float s,
1597 float t,
1598 float p,
1599 unsigned level,
1600 unsigned face_id,
1601 float *rgba)
1602 {
1603 const struct pipe_resource *texture = sp_sview->base.texture;
1604 int width, height;
1605 int x0, y0, x1, y1, layer;
1606 float xw, yw; /* weights */
1607 union tex_tile_address addr;
1608 const float *tx0, *tx1, *tx2, *tx3;
1609 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
1610 corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1611 int c;
1612
1613 width = u_minify(texture->width0, level);
1614 height = u_minify(texture->height0, level);
1615
1616 assert(width > 0);
1617 assert(height > 0);
1618
1619 addr.value = 0;
1620 addr.bits.level = level;
1621
1622 /*
1623 * For seamless if LINEAR filtering is done within a miplevel,
1624 * always apply wrap mode CLAMP_TO_BORDER.
1625 */
1626 if (sp_samp->base.seamless_cube_map) {
1627 /* Note this is a bit overkill, actual clamping is not required */
1628 wrap_linear_clamp_to_border(s, width, &x0, &x1, &xw);
1629 wrap_linear_clamp_to_border(t, height, &y0, &y1, &yw);
1630 } else {
1631 /* Would probably make sense to ignore mode and just do edge clamp */
1632 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1633 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1634 }
1635
1636 layer = sp_sview->base.u.tex.first_layer;
1637
1638 if (sp_samp->base.seamless_cube_map) {
1639 tx0 = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, face_id);
1640 tx1 = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, face_id);
1641 tx2 = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, face_id);
1642 tx3 = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, face_id);
1643 } else {
1644 tx0 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + face_id);
1645 tx1 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + face_id);
1646 tx2 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + face_id);
1647 tx3 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + face_id);
1648 }
1649
1650 /* interpolate R, G, B, A */
1651 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1652 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1653 tx0[c], tx1[c],
1654 tx2[c], tx3[c]);
1655 }
1656
1657
1658 static void
1659 img_filter_cube_array_linear(struct sp_sampler_view *sp_sview,
1660 struct sp_sampler *sp_samp,
1661 float s,
1662 float t,
1663 float p,
1664 unsigned level,
1665 unsigned face_id,
1666 float *rgba)
1667 {
1668 const struct pipe_resource *texture = sp_sview->base.texture;
1669 int width, height;
1670 int x0, y0, x1, y1, layer;
1671 float xw, yw; /* weights */
1672 union tex_tile_address addr;
1673 const float *tx0, *tx1, *tx2, *tx3;
1674 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
1675 corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1676 int c;
1677
1678 width = u_minify(texture->width0, level);
1679 height = u_minify(texture->height0, level);
1680
1681 assert(width > 0);
1682 assert(height > 0);
1683
1684 addr.value = 0;
1685 addr.bits.level = level;
1686
1687 /*
1688 * For seamless if LINEAR filtering is done within a miplevel,
1689 * always apply wrap mode CLAMP_TO_BORDER.
1690 */
1691 if (sp_samp->base.seamless_cube_map) {
1692 /* Note this is a bit overkill, actual clamping is not required */
1693 wrap_linear_clamp_to_border(s, width, &x0, &x1, &xw);
1694 wrap_linear_clamp_to_border(t, height, &y0, &y1, &yw);
1695 } else {
1696 /* Would probably make sense to ignore mode and just do edge clamp */
1697 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1698 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1699 }
1700
1701 layer = coord_to_layer(6 * p + sp_sview->base.u.tex.first_layer,
1702 sp_sview->base.u.tex.first_layer,
1703 sp_sview->base.u.tex.last_layer - 5);
1704
1705 if (sp_samp->base.seamless_cube_map) {
1706 tx0 = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, face_id);
1707 tx1 = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, face_id);
1708 tx2 = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, face_id);
1709 tx3 = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, face_id);
1710 } else {
1711 tx0 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + face_id);
1712 tx1 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + face_id);
1713 tx2 = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + face_id);
1714 tx3 = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + face_id);
1715 }
1716
1717 /* interpolate R, G, B, A */
1718 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1719 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1720 tx0[c], tx1[c],
1721 tx2[c], tx3[c]);
1722 }
1723
1724 static void
1725 img_filter_3d_linear(struct sp_sampler_view *sp_sview,
1726 struct sp_sampler *sp_samp,
1727 float s,
1728 float t,
1729 float p,
1730 unsigned level,
1731 unsigned face_id,
1732 float *rgba)
1733 {
1734 const struct pipe_resource *texture = sp_sview->base.texture;
1735 int width, height, depth;
1736 int x0, x1, y0, y1, z0, z1;
1737 float xw, yw, zw; /* interpolation weights */
1738 union tex_tile_address addr;
1739 const float *tx00, *tx01, *tx02, *tx03, *tx10, *tx11, *tx12, *tx13;
1740 int c;
1741
1742 width = u_minify(texture->width0, level);
1743 height = u_minify(texture->height0, level);
1744 depth = u_minify(texture->depth0, level);
1745
1746 addr.value = 0;
1747 addr.bits.level = level;
1748
1749 assert(width > 0);
1750 assert(height > 0);
1751 assert(depth > 0);
1752
1753 sp_samp->linear_texcoord_s(s, width, &x0, &x1, &xw);
1754 sp_samp->linear_texcoord_t(t, height, &y0, &y1, &yw);
1755 sp_samp->linear_texcoord_p(p, depth, &z0, &z1, &zw);
1756
1757
1758 tx00 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z0);
1759 tx01 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z0);
1760 tx02 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z0);
1761 tx03 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z0);
1762
1763 tx10 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z1);
1764 tx11 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z1);
1765 tx12 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z1);
1766 tx13 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z1);
1767
1768 /* interpolate R, G, B, A */
1769 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1770 rgba[TGSI_NUM_CHANNELS*c] = lerp_3d(xw, yw, zw,
1771 tx00[c], tx01[c],
1772 tx02[c], tx03[c],
1773 tx10[c], tx11[c],
1774 tx12[c], tx13[c]);
1775 }
1776
1777
1778 /* Calculate level of detail for every fragment,
1779 * with lambda already computed.
1780 * Note that lambda has already been biased by global LOD bias.
1781 * \param biased_lambda per-quad lambda.
1782 * \param lod_in per-fragment lod_bias or explicit_lod.
1783 * \param lod returns the per-fragment lod.
1784 */
1785 static INLINE void
1786 compute_lod(const struct pipe_sampler_state *sampler,
1787 enum tgsi_sampler_control control,
1788 const float biased_lambda,
1789 const float lod_in[TGSI_QUAD_SIZE],
1790 float lod[TGSI_QUAD_SIZE])
1791 {
1792 float min_lod = sampler->min_lod;
1793 float max_lod = sampler->max_lod;
1794 uint i;
1795
1796 switch (control) {
1797 case tgsi_sampler_lod_none:
1798 case tgsi_sampler_lod_zero:
1799 /* XXX FIXME */
1800 case tgsi_sampler_derivs_explicit:
1801 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(biased_lambda, min_lod, max_lod);
1802 break;
1803 case tgsi_sampler_lod_bias:
1804 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1805 lod[i] = biased_lambda + lod_in[i];
1806 lod[i] = CLAMP(lod[i], min_lod, max_lod);
1807 }
1808 break;
1809 case tgsi_sampler_lod_explicit:
1810 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1811 lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
1812 }
1813 break;
1814 default:
1815 assert(0);
1816 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1817 }
1818 }
1819
1820
1821 /* Calculate level of detail for every fragment.
1822 * \param lod_in per-fragment lod_bias or explicit_lod.
1823 * \param lod results per-fragment lod.
1824 */
1825 static INLINE void
1826 compute_lambda_lod(struct sp_sampler_view *sp_sview,
1827 struct sp_sampler *sp_samp,
1828 const float s[TGSI_QUAD_SIZE],
1829 const float t[TGSI_QUAD_SIZE],
1830 const float p[TGSI_QUAD_SIZE],
1831 const float lod_in[TGSI_QUAD_SIZE],
1832 enum tgsi_sampler_control control,
1833 float lod[TGSI_QUAD_SIZE])
1834 {
1835 const struct pipe_sampler_state *sampler = &sp_samp->base;
1836 float lod_bias = sampler->lod_bias;
1837 float min_lod = sampler->min_lod;
1838 float max_lod = sampler->max_lod;
1839 float lambda;
1840 uint i;
1841
1842 switch (control) {
1843 case tgsi_sampler_lod_none:
1844 /* XXX FIXME */
1845 case tgsi_sampler_derivs_explicit:
1846 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1847 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lambda, min_lod, max_lod);
1848 break;
1849 case tgsi_sampler_lod_bias:
1850 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1851 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1852 lod[i] = lambda + lod_in[i];
1853 lod[i] = CLAMP(lod[i], min_lod, max_lod);
1854 }
1855 break;
1856 case tgsi_sampler_lod_explicit:
1857 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1858 lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
1859 }
1860 break;
1861 case tgsi_sampler_lod_zero:
1862 /* this is all static state in the sampler really need clamp here? */
1863 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lod_bias, min_lod, max_lod);
1864 break;
1865 default:
1866 assert(0);
1867 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1868 }
1869 }
1870
1871
1872 static void
1873 mip_filter_linear(struct sp_sampler_view *sp_sview,
1874 struct sp_sampler *sp_samp,
1875 img_filter_func min_filter,
1876 img_filter_func mag_filter,
1877 const float s[TGSI_QUAD_SIZE],
1878 const float t[TGSI_QUAD_SIZE],
1879 const float p[TGSI_QUAD_SIZE],
1880 const float c0[TGSI_QUAD_SIZE],
1881 const float lod_in[TGSI_QUAD_SIZE],
1882 enum tgsi_sampler_control control,
1883 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1884 {
1885 const struct pipe_sampler_view *psview = &sp_sview->base;
1886 int j;
1887 float lod[TGSI_QUAD_SIZE];
1888
1889 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1890
1891 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1892 int level0 = psview->u.tex.first_level + (int)lod[j];
1893
1894 if (lod[j] < 0.0)
1895 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1896 psview->u.tex.first_level,
1897 sp_sview->faces[j], &rgba[0][j]);
1898
1899 else if (level0 >= (int) psview->u.tex.last_level)
1900 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], psview->u.tex.last_level,
1901 sp_sview->faces[j], &rgba[0][j]);
1902
1903 else {
1904 float levelBlend = frac(lod[j]);
1905 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
1906 int c;
1907
1908 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level0,
1909 sp_sview->faces[j], &rgbax[0][0]);
1910 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level0+1,
1911 sp_sview->faces[j], &rgbax[0][1]);
1912
1913 for (c = 0; c < 4; c++) {
1914 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
1915 }
1916 }
1917 }
1918
1919 if (DEBUG_TEX) {
1920 print_sample_4(__FUNCTION__, rgba);
1921 }
1922 }
1923
1924
1925 /**
1926 * Compute nearest mipmap level from texcoords.
1927 * Then sample the texture level for four elements of a quad.
1928 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1929 */
1930 static void
1931 mip_filter_nearest(struct sp_sampler_view *sp_sview,
1932 struct sp_sampler *sp_samp,
1933 img_filter_func min_filter,
1934 img_filter_func mag_filter,
1935 const float s[TGSI_QUAD_SIZE],
1936 const float t[TGSI_QUAD_SIZE],
1937 const float p[TGSI_QUAD_SIZE],
1938 const float c0[TGSI_QUAD_SIZE],
1939 const float lod_in[TGSI_QUAD_SIZE],
1940 enum tgsi_sampler_control control,
1941 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1942 {
1943 const struct pipe_sampler_view *psview = &sp_sview->base;
1944 float lod[TGSI_QUAD_SIZE];
1945 int j;
1946
1947 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1948
1949 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1950 if (lod[j] < 0.0)
1951 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1952 psview->u.tex.first_level,
1953 sp_sview->faces[j], &rgba[0][j]);
1954 else {
1955 int level = psview->u.tex.first_level + (int)(lod[j] + 0.5F);
1956 level = MIN2(level, (int)psview->u.tex.last_level);
1957 min_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1958 level, sp_sview->faces[j], &rgba[0][j]);
1959 }
1960 }
1961
1962 if (DEBUG_TEX) {
1963 print_sample_4(__FUNCTION__, rgba);
1964 }
1965 }
1966
1967
1968 static void
1969 mip_filter_none(struct sp_sampler_view *sp_sview,
1970 struct sp_sampler *sp_samp,
1971 img_filter_func min_filter,
1972 img_filter_func mag_filter,
1973 const float s[TGSI_QUAD_SIZE],
1974 const float t[TGSI_QUAD_SIZE],
1975 const float p[TGSI_QUAD_SIZE],
1976 const float c0[TGSI_QUAD_SIZE],
1977 const float lod_in[TGSI_QUAD_SIZE],
1978 enum tgsi_sampler_control control,
1979 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1980 {
1981 float lod[TGSI_QUAD_SIZE];
1982 int j;
1983
1984 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
1985
1986 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1987 if (lod[j] < 0.0) {
1988 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1989 sp_sview->base.u.tex.first_level,
1990 sp_sview->faces[j], &rgba[0][j]);
1991 }
1992 else {
1993 min_filter(sp_sview, sp_samp, s[j], t[j], p[j],
1994 sp_sview->base.u.tex.first_level,
1995 sp_sview->faces[j], &rgba[0][j]);
1996 }
1997 }
1998 }
1999
2000
2001 static void
2002 mip_filter_none_no_filter_select(struct sp_sampler_view *sp_sview,
2003 struct sp_sampler *sp_samp,
2004 img_filter_func min_filter,
2005 img_filter_func mag_filter,
2006 const float s[TGSI_QUAD_SIZE],
2007 const float t[TGSI_QUAD_SIZE],
2008 const float p[TGSI_QUAD_SIZE],
2009 const float c0[TGSI_QUAD_SIZE],
2010 const float lod_in[TGSI_QUAD_SIZE],
2011 enum tgsi_sampler_control control,
2012 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2013 {
2014 int j;
2015
2016 for (j = 0; j < TGSI_QUAD_SIZE; j++)
2017 mag_filter(sp_sview, sp_samp, s[j], t[j], p[j],
2018 sp_sview->base.u.tex.first_level,
2019 sp_sview->faces[j], &rgba[0][j]);
2020 }
2021
2022
2023 /* For anisotropic filtering */
2024 #define WEIGHT_LUT_SIZE 1024
2025
2026 static float *weightLut = NULL;
2027
2028 /**
2029 * Creates the look-up table used to speed-up EWA sampling
2030 */
2031 static void
2032 create_filter_table(void)
2033 {
2034 unsigned i;
2035 if (!weightLut) {
2036 weightLut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
2037
2038 for (i = 0; i < WEIGHT_LUT_SIZE; ++i) {
2039 float alpha = 2;
2040 float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
2041 float weight = (float) exp(-alpha * r2);
2042 weightLut[i] = weight;
2043 }
2044 }
2045 }
2046
2047
2048 /**
2049 * Elliptical weighted average (EWA) filter for producing high quality
2050 * anisotropic filtered results.
2051 * Based on the Higher Quality Elliptical Weighted Average Filter
2052 * published by Paul S. Heckbert in his Master's Thesis
2053 * "Fundamentals of Texture Mapping and Image Warping" (1989)
2054 */
2055 static void
2056 img_filter_2d_ewa(struct sp_sampler_view *sp_sview,
2057 struct sp_sampler *sp_samp,
2058 img_filter_func min_filter,
2059 img_filter_func mag_filter,
2060 const float s[TGSI_QUAD_SIZE],
2061 const float t[TGSI_QUAD_SIZE],
2062 const float p[TGSI_QUAD_SIZE],
2063 unsigned level,
2064 const float dudx, const float dvdx,
2065 const float dudy, const float dvdy,
2066 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2067 {
2068 const struct pipe_resource *texture = sp_sview->base.texture;
2069
2070 // ??? Won't the image filters blow up if level is negative?
2071 unsigned level0 = level > 0 ? level : 0;
2072 float scaling = 1.0f / (1 << level0);
2073 int width = u_minify(texture->width0, level0);
2074 int height = u_minify(texture->height0, level0);
2075
2076 float ux = dudx * scaling;
2077 float vx = dvdx * scaling;
2078 float uy = dudy * scaling;
2079 float vy = dvdy * scaling;
2080
2081 /* compute ellipse coefficients to bound the region:
2082 * A*x*x + B*x*y + C*y*y = F.
2083 */
2084 float A = vx*vx+vy*vy+1;
2085 float B = -2*(ux*vx+uy*vy);
2086 float C = ux*ux+uy*uy+1;
2087 float F = A*C-B*B/4.0f;
2088
2089 /* check if it is an ellipse */
2090 /* assert(F > 0.0); */
2091
2092 /* Compute the ellipse's (u,v) bounding box in texture space */
2093 float d = -B*B+4.0f*C*A;
2094 float box_u = 2.0f / d * sqrtf(d*C*F); /* box_u -> half of bbox with */
2095 float box_v = 2.0f / d * sqrtf(A*d*F); /* box_v -> half of bbox height */
2096
2097 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2098 float s_buffer[TGSI_QUAD_SIZE];
2099 float t_buffer[TGSI_QUAD_SIZE];
2100 float weight_buffer[TGSI_QUAD_SIZE];
2101 unsigned buffer_next;
2102 int j;
2103 float den; /* = 0.0F; */
2104 float ddq;
2105 float U; /* = u0 - tex_u; */
2106 int v;
2107
2108 /* Scale ellipse formula to directly index the Filter Lookup Table.
2109 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2110 */
2111 double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
2112 A *= formScale;
2113 B *= formScale;
2114 C *= formScale;
2115 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2116
2117 /* For each quad, the du and dx values are the same and so the ellipse is
2118 * also the same. Note that texel/image access can only be performed using
2119 * a quad, i.e. it is not possible to get the pixel value for a single
2120 * tex coord. In order to have a better performance, the access is buffered
2121 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2122 * full, then the pixel values are read from the image.
2123 */
2124 ddq = 2 * A;
2125
2126 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2127 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2128 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2129 * value, q, is less than F, we're inside the ellipse
2130 */
2131 float tex_u = -0.5F + s[j] * texture->width0 * scaling;
2132 float tex_v = -0.5F + t[j] * texture->height0 * scaling;
2133
2134 int u0 = (int) floorf(tex_u - box_u);
2135 int u1 = (int) ceilf(tex_u + box_u);
2136 int v0 = (int) floorf(tex_v - box_v);
2137 int v1 = (int) ceilf(tex_v + box_v);
2138
2139 float num[4] = {0.0F, 0.0F, 0.0F, 0.0F};
2140 buffer_next = 0;
2141 den = 0;
2142 U = u0 - tex_u;
2143 for (v = v0; v <= v1; ++v) {
2144 float V = v - tex_v;
2145 float dq = A * (2 * U + 1) + B * V;
2146 float q = (C * V + B * U) * V + A * U * U;
2147
2148 int u;
2149 for (u = u0; u <= u1; ++u) {
2150 /* Note that the ellipse has been pre-scaled so F =
2151 * WEIGHT_LUT_SIZE - 1
2152 */
2153 if (q < WEIGHT_LUT_SIZE) {
2154 /* as a LUT is used, q must never be negative;
2155 * should not happen, though
2156 */
2157 const int qClamped = q >= 0.0F ? q : 0;
2158 float weight = weightLut[qClamped];
2159
2160 weight_buffer[buffer_next] = weight;
2161 s_buffer[buffer_next] = u / ((float) width);
2162 t_buffer[buffer_next] = v / ((float) height);
2163
2164 buffer_next++;
2165 if (buffer_next == TGSI_QUAD_SIZE) {
2166 /* 4 texel coords are in the buffer -> read it now */
2167 unsigned jj;
2168 /* it is assumed that samp->min_img_filter is set to
2169 * img_filter_2d_nearest or one of the
2170 * accelerated img_filter_2d_nearest_XXX functions.
2171 */
2172 for (jj = 0; jj < buffer_next; jj++) {
2173 min_filter(sp_sview, sp_samp, s_buffer[jj], t_buffer[jj], p[jj],
2174 level, sp_sview->faces[j], &rgba_temp[0][jj]);
2175 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2176 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2177 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2178 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2179 }
2180
2181 buffer_next = 0;
2182 }
2183
2184 den += weight;
2185 }
2186 q += dq;
2187 dq += ddq;
2188 }
2189 }
2190
2191 /* if the tex coord buffer contains unread values, we will read
2192 * them now.
2193 */
2194 if (buffer_next > 0) {
2195 unsigned jj;
2196 /* it is assumed that samp->min_img_filter is set to
2197 * img_filter_2d_nearest or one of the
2198 * accelerated img_filter_2d_nearest_XXX functions.
2199 */
2200 for (jj = 0; jj < buffer_next; jj++) {
2201 min_filter(sp_sview, sp_samp, s_buffer[jj], t_buffer[jj], p[jj],
2202 level, sp_sview->faces[j], &rgba_temp[0][jj]);
2203 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2204 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2205 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2206 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2207 }
2208 }
2209
2210 if (den <= 0.0F) {
2211 /* Reaching this place would mean that no pixels intersected
2212 * the ellipse. This should never happen because the filter
2213 * we use always intersects at least one pixel.
2214 */
2215
2216 /*rgba[0]=0;
2217 rgba[1]=0;
2218 rgba[2]=0;
2219 rgba[3]=0;*/
2220 /* not enough pixels in resampling, resort to direct interpolation */
2221 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], level,
2222 sp_sview->faces[j], &rgba_temp[0][j]);
2223 den = 1;
2224 num[0] = rgba_temp[0][j];
2225 num[1] = rgba_temp[1][j];
2226 num[2] = rgba_temp[2][j];
2227 num[3] = rgba_temp[3][j];
2228 }
2229
2230 rgba[0][j] = num[0] / den;
2231 rgba[1][j] = num[1] / den;
2232 rgba[2][j] = num[2] / den;
2233 rgba[3][j] = num[3] / den;
2234 }
2235 }
2236
2237
2238 /**
2239 * Sample 2D texture using an anisotropic filter.
2240 */
2241 static void
2242 mip_filter_linear_aniso(struct sp_sampler_view *sp_sview,
2243 struct sp_sampler *sp_samp,
2244 img_filter_func min_filter,
2245 img_filter_func mag_filter,
2246 const float s[TGSI_QUAD_SIZE],
2247 const float t[TGSI_QUAD_SIZE],
2248 const float p[TGSI_QUAD_SIZE],
2249 const float c0[TGSI_QUAD_SIZE],
2250 const float lod_in[TGSI_QUAD_SIZE],
2251 enum tgsi_sampler_control control,
2252 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2253 {
2254 const struct pipe_resource *texture = sp_sview->base.texture;
2255 const struct pipe_sampler_view *psview = &sp_sview->base;
2256 int level0;
2257 float lambda;
2258 float lod[TGSI_QUAD_SIZE];
2259
2260 float s_to_u = u_minify(texture->width0, psview->u.tex.first_level);
2261 float t_to_v = u_minify(texture->height0, psview->u.tex.first_level);
2262 float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2263 float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2264 float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2265 float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2266
2267 if (control == tgsi_sampler_lod_bias ||
2268 control == tgsi_sampler_lod_none ||
2269 /* XXX FIXME */
2270 control == tgsi_sampler_derivs_explicit) {
2271 /* note: instead of working with Px and Py, we will use the
2272 * squared length instead, to avoid sqrt.
2273 */
2274 float Px2 = dudx * dudx + dvdx * dvdx;
2275 float Py2 = dudy * dudy + dvdy * dvdy;
2276
2277 float Pmax2;
2278 float Pmin2;
2279 float e;
2280 const float maxEccentricity = sp_samp->base.max_anisotropy * sp_samp->base.max_anisotropy;
2281
2282 if (Px2 < Py2) {
2283 Pmax2 = Py2;
2284 Pmin2 = Px2;
2285 }
2286 else {
2287 Pmax2 = Px2;
2288 Pmin2 = Py2;
2289 }
2290
2291 /* if the eccentricity of the ellipse is too big, scale up the shorter
2292 * of the two vectors to limit the maximum amount of work per pixel
2293 */
2294 e = Pmax2 / Pmin2;
2295 if (e > maxEccentricity) {
2296 /* float s=e / maxEccentricity;
2297 minor[0] *= s;
2298 minor[1] *= s;
2299 Pmin2 *= s; */
2300 Pmin2 = Pmax2 / maxEccentricity;
2301 }
2302
2303 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2304 * this since 0.5*log(x) = log(sqrt(x))
2305 */
2306 lambda = 0.5F * util_fast_log2(Pmin2) + sp_samp->base.lod_bias;
2307 compute_lod(&sp_samp->base, control, lambda, lod_in, lod);
2308 }
2309 else {
2310 assert(control == tgsi_sampler_lod_explicit ||
2311 control == tgsi_sampler_lod_zero);
2312 compute_lod(&sp_samp->base, control, sp_samp->base.lod_bias, lod_in, lod);
2313 }
2314
2315 /* XXX: Take into account all lod values.
2316 */
2317 lambda = lod[0];
2318 level0 = psview->u.tex.first_level + (int)lambda;
2319
2320 /* If the ellipse covers the whole image, we can
2321 * simply return the average of the whole image.
2322 */
2323 if (level0 >= (int) psview->u.tex.last_level) {
2324 int j;
2325 for (j = 0; j < TGSI_QUAD_SIZE; j++)
2326 min_filter(sp_sview, sp_samp, s[j], t[j], p[j], psview->u.tex.last_level,
2327 sp_sview->faces[j], &rgba[0][j]);
2328 }
2329 else {
2330 /* don't bother interpolating between multiple LODs; it doesn't
2331 * seem to be worth the extra running time.
2332 */
2333 img_filter_2d_ewa(sp_sview, sp_samp, min_filter, mag_filter,
2334 s, t, p, level0,
2335 dudx, dvdx, dudy, dvdy, rgba);
2336 }
2337
2338 if (DEBUG_TEX) {
2339 print_sample_4(__FUNCTION__, rgba);
2340 }
2341 }
2342
2343
2344 /**
2345 * Specialized version of mip_filter_linear with hard-wired calls to
2346 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2347 */
2348 static void
2349 mip_filter_linear_2d_linear_repeat_POT(
2350 struct sp_sampler_view *sp_sview,
2351 struct sp_sampler *sp_samp,
2352 img_filter_func min_filter,
2353 img_filter_func mag_filter,
2354 const float s[TGSI_QUAD_SIZE],
2355 const float t[TGSI_QUAD_SIZE],
2356 const float p[TGSI_QUAD_SIZE],
2357 const float c0[TGSI_QUAD_SIZE],
2358 const float lod_in[TGSI_QUAD_SIZE],
2359 enum tgsi_sampler_control control,
2360 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2361 {
2362 const struct pipe_sampler_view *psview = &sp_sview->base;
2363 int j;
2364 float lod[TGSI_QUAD_SIZE];
2365
2366 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, control, lod);
2367
2368 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2369 int level0 = psview->u.tex.first_level + (int)lod[j];
2370
2371 /* Catches both negative and large values of level0:
2372 */
2373 if ((unsigned)level0 >= psview->u.tex.last_level) {
2374 if (level0 < 0)
2375 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j],
2376 psview->u.tex.first_level,
2377 sp_sview->faces[j], &rgba[0][j]);
2378 else
2379 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j],
2380 psview->u.tex.last_level,
2381 sp_sview->faces[j], &rgba[0][j]);
2382
2383 }
2384 else {
2385 float levelBlend = frac(lod[j]);
2386 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2387 int c;
2388
2389 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j], level0,
2390 sp_sview->faces[j], &rgbax[0][0]);
2391 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, s[j], t[j], p[j], level0+1,
2392 sp_sview->faces[j], &rgbax[0][1]);
2393
2394 for (c = 0; c < TGSI_NUM_CHANNELS; c++)
2395 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
2396 }
2397 }
2398
2399 if (DEBUG_TEX) {
2400 print_sample_4(__FUNCTION__, rgba);
2401 }
2402 }
2403
2404
2405 /**
2406 * Do shadow/depth comparisons.
2407 */
2408 static void
2409 sample_compare(struct sp_sampler_view *sp_sview,
2410 struct sp_sampler *sp_samp,
2411 const float s[TGSI_QUAD_SIZE],
2412 const float t[TGSI_QUAD_SIZE],
2413 const float p[TGSI_QUAD_SIZE],
2414 const float c0[TGSI_QUAD_SIZE],
2415 const float c1[TGSI_QUAD_SIZE],
2416 enum tgsi_sampler_control control,
2417 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2418 {
2419 const struct pipe_sampler_state *sampler = &sp_samp->base;
2420 int j;
2421 int k[4];
2422 float pc[4];
2423 const struct util_format_description *format_desc;
2424 unsigned chan_type;
2425
2426 /**
2427 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2428 * for 2D Array texture we need to use the 'c0' (aka Q).
2429 * When we sampled the depth texture, the depth value was put into all
2430 * RGBA channels. We look at the red channel here.
2431 */
2432
2433 if (sp_sview->base.target == PIPE_TEXTURE_2D_ARRAY ||
2434 sp_sview->base.target == PIPE_TEXTURE_CUBE) {
2435 pc[0] = c0[0];
2436 pc[1] = c0[1];
2437 pc[2] = c0[2];
2438 pc[3] = c0[3];
2439 } else if (sp_sview->base.target == PIPE_TEXTURE_CUBE_ARRAY) {
2440 pc[0] = c1[0];
2441 pc[1] = c1[1];
2442 pc[2] = c1[2];
2443 pc[3] = c1[3];
2444 } else {
2445 pc[0] = p[0];
2446 pc[1] = p[1];
2447 pc[2] = p[2];
2448 pc[3] = p[3];
2449 }
2450
2451 format_desc = util_format_description(sp_sview->base.format);
2452 /* not entirely sure we couldn't end up with non-valid swizzle here */
2453 chan_type = format_desc->swizzle[0] <= UTIL_FORMAT_SWIZZLE_W ?
2454 format_desc->channel[format_desc->swizzle[0]].type :
2455 UTIL_FORMAT_TYPE_FLOAT;
2456 if (chan_type != UTIL_FORMAT_TYPE_FLOAT) {
2457 /*
2458 * clamping is a result of conversion to texture format, hence
2459 * doesn't happen with floats. Technically also should do comparison
2460 * in texture format (quantization!).
2461 */
2462 pc[0] = CLAMP(pc[0], 0.0F, 1.0F);
2463 pc[1] = CLAMP(pc[1], 0.0F, 1.0F);
2464 pc[2] = CLAMP(pc[2], 0.0F, 1.0F);
2465 pc[3] = CLAMP(pc[3], 0.0F, 1.0F);
2466 }
2467
2468 /* compare four texcoords vs. four texture samples */
2469 switch (sampler->compare_func) {
2470 case PIPE_FUNC_LESS:
2471 k[0] = pc[0] < rgba[0][0];
2472 k[1] = pc[1] < rgba[0][1];
2473 k[2] = pc[2] < rgba[0][2];
2474 k[3] = pc[3] < rgba[0][3];
2475 break;
2476 case PIPE_FUNC_LEQUAL:
2477 k[0] = pc[0] <= rgba[0][0];
2478 k[1] = pc[1] <= rgba[0][1];
2479 k[2] = pc[2] <= rgba[0][2];
2480 k[3] = pc[3] <= rgba[0][3];
2481 break;
2482 case PIPE_FUNC_GREATER:
2483 k[0] = pc[0] > rgba[0][0];
2484 k[1] = pc[1] > rgba[0][1];
2485 k[2] = pc[2] > rgba[0][2];
2486 k[3] = pc[3] > rgba[0][3];
2487 break;
2488 case PIPE_FUNC_GEQUAL:
2489 k[0] = pc[0] >= rgba[0][0];
2490 k[1] = pc[1] >= rgba[0][1];
2491 k[2] = pc[2] >= rgba[0][2];
2492 k[3] = pc[3] >= rgba[0][3];
2493 break;
2494 case PIPE_FUNC_EQUAL:
2495 k[0] = pc[0] == rgba[0][0];
2496 k[1] = pc[1] == rgba[0][1];
2497 k[2] = pc[2] == rgba[0][2];
2498 k[3] = pc[3] == rgba[0][3];
2499 break;
2500 case PIPE_FUNC_NOTEQUAL:
2501 k[0] = pc[0] != rgba[0][0];
2502 k[1] = pc[1] != rgba[0][1];
2503 k[2] = pc[2] != rgba[0][2];
2504 k[3] = pc[3] != rgba[0][3];
2505 break;
2506 case PIPE_FUNC_ALWAYS:
2507 k[0] = k[1] = k[2] = k[3] = 1;
2508 break;
2509 case PIPE_FUNC_NEVER:
2510 k[0] = k[1] = k[2] = k[3] = 0;
2511 break;
2512 default:
2513 k[0] = k[1] = k[2] = k[3] = 0;
2514 assert(0);
2515 break;
2516 }
2517
2518 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2519 rgba[0][j] = k[j];
2520 rgba[1][j] = k[j];
2521 rgba[2][j] = k[j];
2522 rgba[3][j] = 1.0F;
2523 }
2524 }
2525
2526
2527 static void
2528 do_swizzling(const struct pipe_sampler_view *sview,
2529 float in[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
2530 float out[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2531 {
2532 int j;
2533 const unsigned swizzle_r = sview->swizzle_r;
2534 const unsigned swizzle_g = sview->swizzle_g;
2535 const unsigned swizzle_b = sview->swizzle_b;
2536 const unsigned swizzle_a = sview->swizzle_a;
2537
2538 switch (swizzle_r) {
2539 case PIPE_SWIZZLE_ZERO:
2540 for (j = 0; j < 4; j++)
2541 out[0][j] = 0.0f;
2542 break;
2543 case PIPE_SWIZZLE_ONE:
2544 for (j = 0; j < 4; j++)
2545 out[0][j] = 1.0f;
2546 break;
2547 default:
2548 assert(swizzle_r < 4);
2549 for (j = 0; j < 4; j++)
2550 out[0][j] = in[swizzle_r][j];
2551 }
2552
2553 switch (swizzle_g) {
2554 case PIPE_SWIZZLE_ZERO:
2555 for (j = 0; j < 4; j++)
2556 out[1][j] = 0.0f;
2557 break;
2558 case PIPE_SWIZZLE_ONE:
2559 for (j = 0; j < 4; j++)
2560 out[1][j] = 1.0f;
2561 break;
2562 default:
2563 assert(swizzle_g < 4);
2564 for (j = 0; j < 4; j++)
2565 out[1][j] = in[swizzle_g][j];
2566 }
2567
2568 switch (swizzle_b) {
2569 case PIPE_SWIZZLE_ZERO:
2570 for (j = 0; j < 4; j++)
2571 out[2][j] = 0.0f;
2572 break;
2573 case PIPE_SWIZZLE_ONE:
2574 for (j = 0; j < 4; j++)
2575 out[2][j] = 1.0f;
2576 break;
2577 default:
2578 assert(swizzle_b < 4);
2579 for (j = 0; j < 4; j++)
2580 out[2][j] = in[swizzle_b][j];
2581 }
2582
2583 switch (swizzle_a) {
2584 case PIPE_SWIZZLE_ZERO:
2585 for (j = 0; j < 4; j++)
2586 out[3][j] = 0.0f;
2587 break;
2588 case PIPE_SWIZZLE_ONE:
2589 for (j = 0; j < 4; j++)
2590 out[3][j] = 1.0f;
2591 break;
2592 default:
2593 assert(swizzle_a < 4);
2594 for (j = 0; j < 4; j++)
2595 out[3][j] = in[swizzle_a][j];
2596 }
2597 }
2598
2599
2600 static wrap_nearest_func
2601 get_nearest_unorm_wrap(unsigned mode)
2602 {
2603 switch (mode) {
2604 case PIPE_TEX_WRAP_CLAMP:
2605 return wrap_nearest_unorm_clamp;
2606 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2607 return wrap_nearest_unorm_clamp_to_edge;
2608 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2609 return wrap_nearest_unorm_clamp_to_border;
2610 default:
2611 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
2612 return wrap_nearest_unorm_clamp;
2613 }
2614 }
2615
2616
2617 static wrap_nearest_func
2618 get_nearest_wrap(unsigned mode)
2619 {
2620 switch (mode) {
2621 case PIPE_TEX_WRAP_REPEAT:
2622 return wrap_nearest_repeat;
2623 case PIPE_TEX_WRAP_CLAMP:
2624 return wrap_nearest_clamp;
2625 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2626 return wrap_nearest_clamp_to_edge;
2627 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2628 return wrap_nearest_clamp_to_border;
2629 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2630 return wrap_nearest_mirror_repeat;
2631 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2632 return wrap_nearest_mirror_clamp;
2633 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2634 return wrap_nearest_mirror_clamp_to_edge;
2635 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2636 return wrap_nearest_mirror_clamp_to_border;
2637 default:
2638 assert(0);
2639 return wrap_nearest_repeat;
2640 }
2641 }
2642
2643
2644 static wrap_linear_func
2645 get_linear_unorm_wrap(unsigned mode)
2646 {
2647 switch (mode) {
2648 case PIPE_TEX_WRAP_CLAMP:
2649 return wrap_linear_unorm_clamp;
2650 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2651 return wrap_linear_unorm_clamp_to_edge;
2652 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2653 return wrap_linear_unorm_clamp_to_border;
2654 default:
2655 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
2656 return wrap_linear_unorm_clamp;
2657 }
2658 }
2659
2660
2661 static wrap_linear_func
2662 get_linear_wrap(unsigned mode)
2663 {
2664 switch (mode) {
2665 case PIPE_TEX_WRAP_REPEAT:
2666 return wrap_linear_repeat;
2667 case PIPE_TEX_WRAP_CLAMP:
2668 return wrap_linear_clamp;
2669 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2670 return wrap_linear_clamp_to_edge;
2671 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2672 return wrap_linear_clamp_to_border;
2673 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2674 return wrap_linear_mirror_repeat;
2675 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2676 return wrap_linear_mirror_clamp;
2677 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2678 return wrap_linear_mirror_clamp_to_edge;
2679 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2680 return wrap_linear_mirror_clamp_to_border;
2681 default:
2682 assert(0);
2683 return wrap_linear_repeat;
2684 }
2685 }
2686
2687
2688 /**
2689 * Is swizzling needed for the given state key?
2690 */
2691 static INLINE bool
2692 any_swizzle(const struct pipe_sampler_view *view)
2693 {
2694 return (view->swizzle_r != PIPE_SWIZZLE_RED ||
2695 view->swizzle_g != PIPE_SWIZZLE_GREEN ||
2696 view->swizzle_b != PIPE_SWIZZLE_BLUE ||
2697 view->swizzle_a != PIPE_SWIZZLE_ALPHA);
2698 }
2699
2700
2701 static img_filter_func
2702 get_img_filter(const struct sp_sampler_view *sp_sview,
2703 const struct pipe_sampler_state *sampler,
2704 unsigned filter)
2705 {
2706 switch (sp_sview->base.target) {
2707 case PIPE_BUFFER:
2708 case PIPE_TEXTURE_1D:
2709 if (filter == PIPE_TEX_FILTER_NEAREST)
2710 return img_filter_1d_nearest;
2711 else
2712 return img_filter_1d_linear;
2713 break;
2714 case PIPE_TEXTURE_1D_ARRAY:
2715 if (filter == PIPE_TEX_FILTER_NEAREST)
2716 return img_filter_1d_array_nearest;
2717 else
2718 return img_filter_1d_array_linear;
2719 break;
2720 case PIPE_TEXTURE_2D:
2721 case PIPE_TEXTURE_RECT:
2722 /* Try for fast path:
2723 */
2724 if (sp_sview->pot2d &&
2725 sampler->wrap_s == sampler->wrap_t &&
2726 sampler->normalized_coords)
2727 {
2728 switch (sampler->wrap_s) {
2729 case PIPE_TEX_WRAP_REPEAT:
2730 switch (filter) {
2731 case PIPE_TEX_FILTER_NEAREST:
2732 return img_filter_2d_nearest_repeat_POT;
2733 case PIPE_TEX_FILTER_LINEAR:
2734 return img_filter_2d_linear_repeat_POT;
2735 default:
2736 break;
2737 }
2738 break;
2739 case PIPE_TEX_WRAP_CLAMP:
2740 switch (filter) {
2741 case PIPE_TEX_FILTER_NEAREST:
2742 return img_filter_2d_nearest_clamp_POT;
2743 default:
2744 break;
2745 }
2746 }
2747 }
2748 /* Otherwise use default versions:
2749 */
2750 if (filter == PIPE_TEX_FILTER_NEAREST)
2751 return img_filter_2d_nearest;
2752 else
2753 return img_filter_2d_linear;
2754 break;
2755 case PIPE_TEXTURE_2D_ARRAY:
2756 if (filter == PIPE_TEX_FILTER_NEAREST)
2757 return img_filter_2d_array_nearest;
2758 else
2759 return img_filter_2d_array_linear;
2760 break;
2761 case PIPE_TEXTURE_CUBE:
2762 if (filter == PIPE_TEX_FILTER_NEAREST)
2763 return img_filter_cube_nearest;
2764 else
2765 return img_filter_cube_linear;
2766 break;
2767 case PIPE_TEXTURE_CUBE_ARRAY:
2768 if (filter == PIPE_TEX_FILTER_NEAREST)
2769 return img_filter_cube_array_nearest;
2770 else
2771 return img_filter_cube_array_linear;
2772 break;
2773 case PIPE_TEXTURE_3D:
2774 if (filter == PIPE_TEX_FILTER_NEAREST)
2775 return img_filter_3d_nearest;
2776 else
2777 return img_filter_3d_linear;
2778 break;
2779 default:
2780 assert(0);
2781 return img_filter_1d_nearest;
2782 }
2783 }
2784
2785
2786 static void
2787 sample_mip(struct sp_sampler_view *sp_sview,
2788 struct sp_sampler *sp_samp,
2789 const float s[TGSI_QUAD_SIZE],
2790 const float t[TGSI_QUAD_SIZE],
2791 const float p[TGSI_QUAD_SIZE],
2792 const float c0[TGSI_QUAD_SIZE],
2793 const float lod[TGSI_QUAD_SIZE],
2794 enum tgsi_sampler_control control,
2795 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2796 {
2797 mip_filter_func mip_filter;
2798 img_filter_func min_img_filter = NULL;
2799 img_filter_func mag_img_filter = NULL;
2800
2801 if (sp_sview->pot2d & sp_samp->min_mag_equal_repeat_linear) {
2802 mip_filter = mip_filter_linear_2d_linear_repeat_POT;
2803 }
2804 else {
2805 mip_filter = sp_samp->mip_filter;
2806 min_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->min_img_filter);
2807 if (sp_samp->min_mag_equal) {
2808 mag_img_filter = min_img_filter;
2809 }
2810 else {
2811 mag_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->base.mag_img_filter);
2812 }
2813 }
2814
2815 mip_filter(sp_sview, sp_samp, min_img_filter, mag_img_filter,
2816 s, t, p, c0, lod, control, rgba);
2817
2818 if (sp_samp->base.compare_mode != PIPE_TEX_COMPARE_NONE) {
2819 sample_compare(sp_sview, sp_samp, s, t, p, c0, lod, control, rgba);
2820 }
2821
2822 if (sp_sview->need_swizzle) {
2823 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2824 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
2825 do_swizzling(&sp_sview->base, rgba_temp, rgba);
2826 }
2827
2828 }
2829
2830
2831 /**
2832 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2833 * Put face info into the sampler faces[] array.
2834 */
2835 static void
2836 sample_cube(struct sp_sampler_view *sp_sview,
2837 struct sp_sampler *sp_samp,
2838 const float s[TGSI_QUAD_SIZE],
2839 const float t[TGSI_QUAD_SIZE],
2840 const float p[TGSI_QUAD_SIZE],
2841 const float c0[TGSI_QUAD_SIZE],
2842 const float c1[TGSI_QUAD_SIZE],
2843 enum tgsi_sampler_control control,
2844 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2845 {
2846 unsigned j;
2847 float ssss[4], tttt[4];
2848
2849 /* Not actually used, but the intermediate steps that do the
2850 * dereferencing don't know it.
2851 */
2852 static float pppp[4] = { 0, 0, 0, 0 };
2853
2854 pppp[0] = c0[0];
2855 pppp[1] = c0[1];
2856 pppp[2] = c0[2];
2857 pppp[3] = c0[3];
2858 /*
2859 major axis
2860 direction target sc tc ma
2861 ---------- ------------------------------- --- --- ---
2862 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2863 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2864 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2865 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2866 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2867 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2868 */
2869
2870 /* Choose the cube face and compute new s/t coords for the 2D face.
2871 *
2872 * Use the same cube face for all four pixels in the quad.
2873 *
2874 * This isn't ideal, but if we want to use a different cube face
2875 * per pixel in the quad, we'd have to also compute the per-face
2876 * LOD here too. That's because the four post-face-selection
2877 * texcoords are no longer related to each other (they're
2878 * per-face!) so we can't use subtraction to compute the partial
2879 * deriviates to compute the LOD. Doing so (near cube edges
2880 * anyway) gives us pretty much random values.
2881 */
2882 {
2883 /* use the average of the four pixel's texcoords to choose the face */
2884 const float rx = 0.25F * (s[0] + s[1] + s[2] + s[3]);
2885 const float ry = 0.25F * (t[0] + t[1] + t[2] + t[3]);
2886 const float rz = 0.25F * (p[0] + p[1] + p[2] + p[3]);
2887 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
2888
2889 if (arx >= ary && arx >= arz) {
2890 float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
2891 uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
2892 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2893 const float ima = -0.5F / fabsf(s[j]);
2894 ssss[j] = sign * p[j] * ima + 0.5F;
2895 tttt[j] = t[j] * ima + 0.5F;
2896 sp_sview->faces[j] = face;
2897 }
2898 }
2899 else if (ary >= arx && ary >= arz) {
2900 float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
2901 uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
2902 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2903 const float ima = -0.5F / fabsf(t[j]);
2904 ssss[j] = -s[j] * ima + 0.5F;
2905 tttt[j] = sign * -p[j] * ima + 0.5F;
2906 sp_sview->faces[j] = face;
2907 }
2908 }
2909 else {
2910 float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
2911 uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
2912 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2913 const float ima = -0.5F / fabsf(p[j]);
2914 ssss[j] = sign * -s[j] * ima + 0.5F;
2915 tttt[j] = t[j] * ima + 0.5F;
2916 sp_sview->faces[j] = face;
2917 }
2918 }
2919 }
2920
2921 sample_mip(sp_sview, sp_samp, ssss, tttt, pppp, c0, c1, control, rgba);
2922 }
2923
2924
2925 static void
2926 sp_get_dims(struct sp_sampler_view *sp_sview, int level,
2927 int dims[4])
2928 {
2929 const struct pipe_sampler_view *view = &sp_sview->base;
2930 const struct pipe_resource *texture = view->texture;
2931
2932 if (view->target == PIPE_BUFFER) {
2933 dims[0] = (view->u.buf.last_element - view->u.buf.first_element) + 1;
2934 /* the other values are undefined, but let's avoid potential valgrind
2935 * warnings.
2936 */
2937 dims[1] = dims[2] = dims[3] = 0;
2938 return;
2939 }
2940
2941 /* undefined according to EXT_gpu_program */
2942 level += view->u.tex.first_level;
2943 if (level > view->u.tex.last_level)
2944 return;
2945
2946 dims[3] = view->u.tex.last_level - view->u.tex.first_level + 1;
2947 dims[0] = u_minify(texture->width0, level);
2948
2949 switch (view->target) {
2950 case PIPE_TEXTURE_1D_ARRAY:
2951 dims[1] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
2952 /* fallthrough */
2953 case PIPE_TEXTURE_1D:
2954 return;
2955 case PIPE_TEXTURE_2D_ARRAY:
2956 dims[2] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
2957 /* fallthrough */
2958 case PIPE_TEXTURE_2D:
2959 case PIPE_TEXTURE_CUBE:
2960 case PIPE_TEXTURE_RECT:
2961 dims[1] = u_minify(texture->height0, level);
2962 return;
2963 case PIPE_TEXTURE_3D:
2964 dims[1] = u_minify(texture->height0, level);
2965 dims[2] = u_minify(texture->depth0, level);
2966 return;
2967 case PIPE_TEXTURE_CUBE_ARRAY:
2968 dims[1] = u_minify(texture->height0, level);
2969 dims[2] = (view->u.tex.last_layer - view->u.tex.first_layer + 1) / 6;
2970 break;
2971 default:
2972 assert(!"unexpected texture target in sp_get_dims()");
2973 return;
2974 }
2975 }
2976
2977 /**
2978 * This function is only used for getting unfiltered texels via the
2979 * TXF opcode. The GL spec says that out-of-bounds texel fetches
2980 * produce undefined results. Instead of crashing, lets just clamp
2981 * coords to the texture image size.
2982 */
2983 static void
2984 sp_get_texels(struct sp_sampler_view *sp_sview,
2985 const int v_i[TGSI_QUAD_SIZE],
2986 const int v_j[TGSI_QUAD_SIZE],
2987 const int v_k[TGSI_QUAD_SIZE],
2988 const int lod[TGSI_QUAD_SIZE],
2989 const int8_t offset[3],
2990 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2991 {
2992 union tex_tile_address addr;
2993 const struct pipe_resource *texture = sp_sview->base.texture;
2994 int j, c;
2995 const float *tx;
2996 int width, height, depth;
2997
2998 addr.value = 0;
2999 /* TODO write a better test for LOD */
3000 addr.bits.level = sp_sview->base.target == PIPE_BUFFER ? 0 :
3001 CLAMP(lod[0] + sp_sview->base.u.tex.first_level,
3002 sp_sview->base.u.tex.first_level,
3003 sp_sview->base.u.tex.last_level);
3004
3005 width = u_minify(texture->width0, addr.bits.level);
3006 height = u_minify(texture->height0, addr.bits.level);
3007 depth = u_minify(texture->depth0, addr.bits.level);
3008
3009 switch (sp_sview->base.target) {
3010 case PIPE_BUFFER:
3011 case PIPE_TEXTURE_1D:
3012 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3013 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3014 tx = get_texel_2d_no_border(sp_sview, addr, x, 0);
3015 for (c = 0; c < 4; c++) {
3016 rgba[c][j] = tx[c];
3017 }
3018 }
3019 break;
3020 case PIPE_TEXTURE_1D_ARRAY:
3021 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3022 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3023 int y = CLAMP(v_j[j], sp_sview->base.u.tex.first_layer,
3024 sp_sview->base.u.tex.last_layer);
3025 tx = get_texel_2d_no_border(sp_sview, addr, x, y);
3026 for (c = 0; c < 4; c++) {
3027 rgba[c][j] = tx[c];
3028 }
3029 }
3030 break;
3031 case PIPE_TEXTURE_2D:
3032 case PIPE_TEXTURE_RECT:
3033 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3034 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3035 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3036 tx = get_texel_2d_no_border(sp_sview, addr, x, y);
3037 for (c = 0; c < 4; c++) {
3038 rgba[c][j] = tx[c];
3039 }
3040 }
3041 break;
3042 case PIPE_TEXTURE_2D_ARRAY:
3043 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3044 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3045 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3046 int layer = CLAMP(v_k[j], sp_sview->base.u.tex.first_layer,
3047 sp_sview->base.u.tex.last_layer);
3048 tx = get_texel_3d_no_border(sp_sview, addr, x, y, layer);
3049 for (c = 0; c < 4; c++) {
3050 rgba[c][j] = tx[c];
3051 }
3052 }
3053 break;
3054 case PIPE_TEXTURE_3D:
3055 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3056 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3057 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3058 int z = CLAMP(v_k[j] + offset[2], 0, depth - 1);
3059 tx = get_texel_3d_no_border(sp_sview, addr, x, y, z);
3060 for (c = 0; c < 4; c++) {
3061 rgba[c][j] = tx[c];
3062 }
3063 }
3064 break;
3065 case PIPE_TEXTURE_CUBE: /* TXF can't work on CUBE according to spec */
3066 default:
3067 assert(!"Unknown or CUBE texture type in TXF processing\n");
3068 break;
3069 }
3070
3071 if (sp_sview->need_swizzle) {
3072 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
3073 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
3074 do_swizzling(&sp_sview->base, rgba_temp, rgba);
3075 }
3076 }
3077
3078
3079 void *
3080 softpipe_create_sampler_state(struct pipe_context *pipe,
3081 const struct pipe_sampler_state *sampler)
3082 {
3083 struct sp_sampler *samp = CALLOC_STRUCT(sp_sampler);
3084
3085 samp->base = *sampler;
3086
3087 /* Note that (for instance) linear_texcoord_s and
3088 * nearest_texcoord_s may be active at the same time, if the
3089 * sampler min_img_filter differs from its mag_img_filter.
3090 */
3091 if (sampler->normalized_coords) {
3092 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
3093 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
3094 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
3095
3096 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
3097 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
3098 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
3099 }
3100 else {
3101 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
3102 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
3103 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
3104
3105 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
3106 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
3107 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
3108 }
3109
3110 samp->min_img_filter = sampler->min_img_filter;
3111
3112 switch (sampler->min_mip_filter) {
3113 case PIPE_TEX_MIPFILTER_NONE:
3114 if (sampler->min_img_filter == sampler->mag_img_filter)
3115 samp->mip_filter = mip_filter_none_no_filter_select;
3116 else
3117 samp->mip_filter = mip_filter_none;
3118 break;
3119
3120 case PIPE_TEX_MIPFILTER_NEAREST:
3121 samp->mip_filter = mip_filter_nearest;
3122 break;
3123
3124 case PIPE_TEX_MIPFILTER_LINEAR:
3125 if (sampler->min_img_filter == sampler->mag_img_filter &&
3126 sampler->normalized_coords &&
3127 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
3128 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
3129 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR &&
3130 sampler->max_anisotropy <= 1) {
3131 samp->min_mag_equal_repeat_linear = TRUE;
3132 }
3133 samp->mip_filter = mip_filter_linear;
3134
3135 /* Anisotropic filtering extension. */
3136 if (sampler->max_anisotropy > 1) {
3137 samp->mip_filter = mip_filter_linear_aniso;
3138
3139 /* Override min_img_filter:
3140 * min_img_filter needs to be set to NEAREST since we need to access
3141 * each texture pixel as it is and weight it later; using linear
3142 * filters will have incorrect results.
3143 * By setting the filter to NEAREST here, we can avoid calling the
3144 * generic img_filter_2d_nearest in the anisotropic filter function,
3145 * making it possible to use one of the accelerated implementations
3146 */
3147 samp->min_img_filter = PIPE_TEX_FILTER_NEAREST;
3148
3149 /* on first access create the lookup table containing the filter weights. */
3150 if (!weightLut) {
3151 create_filter_table();
3152 }
3153 }
3154 break;
3155 }
3156 if (samp->min_img_filter == sampler->mag_img_filter) {
3157 samp->min_mag_equal = TRUE;
3158 }
3159
3160 return (void *)samp;
3161 }
3162
3163
3164 compute_lambda_func
3165 softpipe_get_lambda_func(const struct pipe_sampler_view *view, unsigned shader)
3166 {
3167 if (shader != PIPE_SHADER_FRAGMENT)
3168 return compute_lambda_vert;
3169
3170 switch (view->target) {
3171 case PIPE_BUFFER:
3172 case PIPE_TEXTURE_1D:
3173 case PIPE_TEXTURE_1D_ARRAY:
3174 return compute_lambda_1d;
3175 case PIPE_TEXTURE_2D:
3176 case PIPE_TEXTURE_2D_ARRAY:
3177 case PIPE_TEXTURE_RECT:
3178 case PIPE_TEXTURE_CUBE:
3179 case PIPE_TEXTURE_CUBE_ARRAY:
3180 return compute_lambda_2d;
3181 case PIPE_TEXTURE_3D:
3182 return compute_lambda_3d;
3183 default:
3184 assert(0);
3185 return compute_lambda_1d;
3186 }
3187 }
3188
3189
3190 struct pipe_sampler_view *
3191 softpipe_create_sampler_view(struct pipe_context *pipe,
3192 struct pipe_resource *resource,
3193 const struct pipe_sampler_view *templ)
3194 {
3195 struct sp_sampler_view *sview = CALLOC_STRUCT(sp_sampler_view);
3196 struct softpipe_resource *spr = (struct softpipe_resource *)resource;
3197
3198 if (sview) {
3199 struct pipe_sampler_view *view = &sview->base;
3200 *view = *templ;
3201 view->reference.count = 1;
3202 view->texture = NULL;
3203 pipe_resource_reference(&view->texture, resource);
3204 view->context = pipe;
3205
3206 #ifdef DEBUG
3207 /*
3208 * This is possibly too lenient, but the primary reason is just
3209 * to catch state trackers which forget to initialize this, so
3210 * it only catches clearly impossible view targets.
3211 */
3212 if (view->target != resource->target) {
3213 if (view->target == PIPE_TEXTURE_1D)
3214 assert(resource->target == PIPE_TEXTURE_1D_ARRAY);
3215 else if (view->target == PIPE_TEXTURE_1D_ARRAY)
3216 assert(resource->target == PIPE_TEXTURE_1D);
3217 else if (view->target == PIPE_TEXTURE_2D)
3218 assert(resource->target == PIPE_TEXTURE_2D_ARRAY ||
3219 resource->target == PIPE_TEXTURE_CUBE ||
3220 resource->target == PIPE_TEXTURE_CUBE_ARRAY);
3221 else if (view->target == PIPE_TEXTURE_2D_ARRAY)
3222 assert(resource->target == PIPE_TEXTURE_2D ||
3223 resource->target == PIPE_TEXTURE_CUBE ||
3224 resource->target == PIPE_TEXTURE_CUBE_ARRAY);
3225 else if (view->target == PIPE_TEXTURE_CUBE)
3226 assert(resource->target == PIPE_TEXTURE_CUBE_ARRAY ||
3227 resource->target == PIPE_TEXTURE_2D_ARRAY);
3228 else if (view->target == PIPE_TEXTURE_CUBE_ARRAY)
3229 assert(resource->target == PIPE_TEXTURE_CUBE ||
3230 resource->target == PIPE_TEXTURE_2D_ARRAY);
3231 else
3232 assert(0);
3233 }
3234 #endif
3235
3236 if (any_swizzle(view)) {
3237 sview->need_swizzle = TRUE;
3238 }
3239
3240 if (view->target == PIPE_TEXTURE_CUBE ||
3241 view->target == PIPE_TEXTURE_CUBE_ARRAY)
3242 sview->get_samples = sample_cube;
3243 else {
3244 sview->get_samples = sample_mip;
3245 }
3246 sview->pot2d = spr->pot &&
3247 (view->target == PIPE_TEXTURE_2D ||
3248 view->target == PIPE_TEXTURE_RECT);
3249
3250 sview->xpot = util_logbase2( resource->width0 );
3251 sview->ypot = util_logbase2( resource->height0 );
3252 }
3253
3254 return (struct pipe_sampler_view *) sview;
3255 }
3256
3257
3258 static void
3259 sp_tgsi_get_dims(struct tgsi_sampler *tgsi_sampler,
3260 const unsigned sview_index,
3261 int level, int dims[4])
3262 {
3263 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3264
3265 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3266 /* always have a view here but texture is NULL if no sampler view was set. */
3267 if (!sp_samp->sp_sview[sview_index].base.texture) {
3268 dims[0] = dims[1] = dims[2] = dims[3] = 0;
3269 return;
3270 }
3271 sp_get_dims(&sp_samp->sp_sview[sview_index], level, dims);
3272 }
3273
3274
3275 static void
3276 sp_tgsi_get_samples(struct tgsi_sampler *tgsi_sampler,
3277 const unsigned sview_index,
3278 const unsigned sampler_index,
3279 const float s[TGSI_QUAD_SIZE],
3280 const float t[TGSI_QUAD_SIZE],
3281 const float p[TGSI_QUAD_SIZE],
3282 const float c0[TGSI_QUAD_SIZE],
3283 const float lod[TGSI_QUAD_SIZE],
3284 float derivs[3][2][TGSI_QUAD_SIZE],
3285 const int8_t offset[3],
3286 enum tgsi_sampler_control control,
3287 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
3288 {
3289 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3290
3291 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3292 assert(sampler_index < PIPE_MAX_SAMPLERS);
3293 assert(sp_samp->sp_sampler[sampler_index]);
3294 /* always have a view here but texture is NULL if no sampler view was set. */
3295 if (!sp_samp->sp_sview[sview_index].base.texture) {
3296 int i, j;
3297 for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
3298 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
3299 rgba[j][i] = 0.0f;
3300 }
3301 }
3302 return;
3303 }
3304 sp_samp->sp_sview[sview_index].get_samples(&sp_samp->sp_sview[sview_index],
3305 sp_samp->sp_sampler[sampler_index],
3306 s, t, p, c0, lod, control, rgba);
3307 }
3308
3309
3310 static void
3311 sp_tgsi_get_texel(struct tgsi_sampler *tgsi_sampler,
3312 const unsigned sview_index,
3313 const int i[TGSI_QUAD_SIZE],
3314 const int j[TGSI_QUAD_SIZE], const int k[TGSI_QUAD_SIZE],
3315 const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
3316 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
3317 {
3318 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3319
3320 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3321 /* always have a view here but texture is NULL if no sampler view was set. */
3322 if (!sp_samp->sp_sview[sview_index].base.texture) {
3323 int i, j;
3324 for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
3325 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
3326 rgba[j][i] = 0.0f;
3327 }
3328 }
3329 return;
3330 }
3331 sp_get_texels(&sp_samp->sp_sview[sview_index], i, j, k, lod, offset, rgba);
3332 }
3333
3334
3335 struct sp_tgsi_sampler *
3336 sp_create_tgsi_sampler(void)
3337 {
3338 struct sp_tgsi_sampler *samp = CALLOC_STRUCT(sp_tgsi_sampler);
3339 if (!samp)
3340 return NULL;
3341
3342 samp->base.get_dims = sp_tgsi_get_dims;
3343 samp->base.get_samples = sp_tgsi_get_samples;
3344 samp->base.get_texel = sp_tgsi_get_texel;
3345
3346 return samp;
3347 }
3348