projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
gallium: replace INLINE with inline
[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 offset, 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 + offset, size);
140 }
141
142
143 static void
144 wrap_nearest_clamp(float s, unsigned size, int offset, int *icoord)
145 {
146 /* s limited to [0,1] */
147 /* i limited to [0,size-1] */
148 s *= size;
149 s += offset;
150 if (s <= 0.0F)
151 *icoord = 0;
152 else if (s >= size)
153 *icoord = size - 1;
154 else
155 *icoord = util_ifloor(s);
156 }
157
158
159 static void
160 wrap_nearest_clamp_to_edge(float s, unsigned size, int offset, int *icoord)
161 {
162 /* s limited to [min,max] */
163 /* i limited to [0, size-1] */
164 const float min = 0.5F;
165 const float max = (float)size - 0.5F;
166
167 s *= size;
168 s += offset;
169
170 if (s < min)
171 *icoord = 0;
172 else if (s > max)
173 *icoord = size - 1;
174 else
175 *icoord = util_ifloor(s);
176 }
177
178
179 static void
180 wrap_nearest_clamp_to_border(float s, unsigned size, int offset, int *icoord)
181 {
182 /* s limited to [min,max] */
183 /* i limited to [-1, size] */
184 const float min = -0.5F;
185 const float max = size + 0.5F;
186
187 s *= size;
188 s += offset;
189 if (s <= min)
190 *icoord = -1;
191 else if (s >= max)
192 *icoord = size;
193 else
194 *icoord = util_ifloor(s);
195 }
196
197 static void
198 wrap_nearest_mirror_repeat(float s, unsigned size, int offset, int *icoord)
199 {
200 const float min = 1.0F / (2.0F * size);
201 const float max = 1.0F - min;
202 int flr;
203 float u;
204
205 s += (float)offset / size;
206 flr = util_ifloor(s);
207 u = frac(s);
208 if (flr & 1)
209 u = 1.0F - u;
210 if (u < min)
211 *icoord = 0;
212 else if (u > max)
213 *icoord = size - 1;
214 else
215 *icoord = util_ifloor(u * size);
216 }
217
218
219 static void
220 wrap_nearest_mirror_clamp(float s, unsigned size, int offset, int *icoord)
221 {
222 /* s limited to [0,1] */
223 /* i limited to [0,size-1] */
224 const float u = fabsf(s * size + offset);
225 if (u <= 0.0F)
226 *icoord = 0;
227 else if (u >= size)
228 *icoord = size - 1;
229 else
230 *icoord = util_ifloor(u);
231 }
232
233
234 static void
235 wrap_nearest_mirror_clamp_to_edge(float s, unsigned size, int offset, int *icoord)
236 {
237 /* s limited to [min,max] */
238 /* i limited to [0, size-1] */
239 const float min = 0.5F;
240 const float max = (float)size - 0.5F;
241 const float u = fabsf(s * size + offset);
242
243 if (u < min)
244 *icoord = 0;
245 else if (u > max)
246 *icoord = size - 1;
247 else
248 *icoord = util_ifloor(u);
249 }
250
251
252 static void
253 wrap_nearest_mirror_clamp_to_border(float s, unsigned size, int offset, int *icoord)
254 {
255 /* u limited to [-0.5, size-0.5] */
256 const float min = -0.5F;
257 const float max = (float)size + 0.5F;
258 const float u = fabsf(s * size + offset);
259
260 if (u < min)
261 *icoord = -1;
262 else if (u > max)
263 *icoord = size;
264 else
265 *icoord = util_ifloor(u);
266 }
267
268
269 /**
270 * Used to compute texel locations for linear sampling
271 * \param wrapMode PIPE_TEX_WRAP_x
272 * \param s the texcoord
273 * \param size the texture image size
274 * \param icoord0 returns first texture index
275 * \param icoord1 returns second texture index (usually icoord0 + 1)
276 * \param w returns blend factor/weight between texture indices
277 * \param icoord returns the computed integer texture coord
278 */
279 static void
280 wrap_linear_repeat(float s, unsigned size, int offset,
281 int *icoord0, int *icoord1, float *w)
282 {
283 float u = s * size - 0.5F;
284 *icoord0 = repeat(util_ifloor(u) + offset, size);
285 *icoord1 = repeat(*icoord0 + 1, size);
286 *w = frac(u);
287 }
288
289
290 static void
291 wrap_linear_clamp(float s, unsigned size, int offset,
292 int *icoord0, int *icoord1, float *w)
293 {
294 float u = CLAMP(s * size + offset, 0.0F, (float)size);
295
296 u = u - 0.5f;
297 *icoord0 = util_ifloor(u);
298 *icoord1 = *icoord0 + 1;
299 *w = frac(u);
300 }
301
302
303 static void
304 wrap_linear_clamp_to_edge(float s, unsigned size, int offset,
305 int *icoord0, int *icoord1, float *w)
306 {
307 float u = CLAMP(s * size + offset, 0.0F, (float)size);
308 u = u - 0.5f;
309 *icoord0 = util_ifloor(u);
310 *icoord1 = *icoord0 + 1;
311 if (*icoord0 < 0)
312 *icoord0 = 0;
313 if (*icoord1 >= (int) size)
314 *icoord1 = size - 1;
315 *w = frac(u);
316 }
317
318
319 static void
320 wrap_linear_clamp_to_border(float s, unsigned size, int offset,
321 int *icoord0, int *icoord1, float *w)
322 {
323 const float min = -0.5F;
324 const float max = (float)size + 0.5F;
325 float u = CLAMP(s * size + offset, min, max);
326 u = u - 0.5f;
327 *icoord0 = util_ifloor(u);
328 *icoord1 = *icoord0 + 1;
329 *w = frac(u);
330 }
331
332
333 static void
334 wrap_linear_mirror_repeat(float s, unsigned size, int offset,
335 int *icoord0, int *icoord1, float *w)
336 {
337 int flr;
338 float u;
339
340 s += (float)offset / size;
341 flr = util_ifloor(s);
342 u = frac(s);
343 if (flr & 1)
344 u = 1.0F - u;
345 u = u * size - 0.5F;
346 *icoord0 = util_ifloor(u);
347 *icoord1 = *icoord0 + 1;
348 if (*icoord0 < 0)
349 *icoord0 = 0;
350 if (*icoord1 >= (int) size)
351 *icoord1 = size - 1;
352 *w = frac(u);
353 }
354
355
356 static void
357 wrap_linear_mirror_clamp(float s, unsigned size, int offset,
358 int *icoord0, int *icoord1, float *w)
359 {
360 float u = fabsf(s * size + offset);
361 if (u >= size)
362 u = (float) size;
363 u -= 0.5F;
364 *icoord0 = util_ifloor(u);
365 *icoord1 = *icoord0 + 1;
366 *w = frac(u);
367 }
368
369
370 static void
371 wrap_linear_mirror_clamp_to_edge(float s, unsigned size, int offset,
372 int *icoord0, int *icoord1, float *w)
373 {
374 float u = fabsf(s * size + offset);
375 if (u >= size)
376 u = (float) size;
377 u -= 0.5F;
378 *icoord0 = util_ifloor(u);
379 *icoord1 = *icoord0 + 1;
380 if (*icoord0 < 0)
381 *icoord0 = 0;
382 if (*icoord1 >= (int) size)
383 *icoord1 = size - 1;
384 *w = frac(u);
385 }
386
387
388 static void
389 wrap_linear_mirror_clamp_to_border(float s, unsigned size, int offset,
390 int *icoord0, int *icoord1, float *w)
391 {
392 const float min = -0.5F;
393 const float max = size + 0.5F;
394 float u = fabsf(s * size + offset);
395 if (u <= min)
396 u = min;
397 else if (u >= max)
398 u = max;
399 u -= 0.5F;
400 *icoord0 = util_ifloor(u);
401 *icoord1 = *icoord0 + 1;
402 *w = frac(u);
403 }
404
405
406 /**
407 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
408 */
409 static void
410 wrap_nearest_unorm_clamp(float s, unsigned size, int offset, int *icoord)
411 {
412 int i = util_ifloor(s);
413 *icoord = CLAMP(i + offset, 0, (int) size-1);
414 }
415
416
417 /**
418 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
419 */
420 static void
421 wrap_nearest_unorm_clamp_to_border(float s, unsigned size, int offset, int *icoord)
422 {
423 *icoord = util_ifloor( CLAMP(s + offset, -0.5F, (float) size + 0.5F) );
424 }
425
426
427 /**
428 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
429 */
430 static void
431 wrap_nearest_unorm_clamp_to_edge(float s, unsigned size, int offset, int *icoord)
432 {
433 *icoord = util_ifloor( CLAMP(s + offset, 0.5F, (float) size - 0.5F) );
434 }
435
436
437 /**
438 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
439 */
440 static void
441 wrap_linear_unorm_clamp(float s, unsigned size, int offset,
442 int *icoord0, int *icoord1, float *w)
443 {
444 /* Not exactly what the spec says, but it matches NVIDIA output */
445 float u = CLAMP(s + offset - 0.5F, 0.0f, (float) size - 1.0f);
446 *icoord0 = util_ifloor(u);
447 *icoord1 = *icoord0 + 1;
448 *w = frac(u);
449 }
450
451
452 /**
453 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
454 */
455 static void
456 wrap_linear_unorm_clamp_to_border(float s, unsigned size, int offset,
457 int *icoord0, int *icoord1, float *w)
458 {
459 float u = CLAMP(s + offset, -0.5F, (float) size + 0.5F);
460 u -= 0.5F;
461 *icoord0 = util_ifloor(u);
462 *icoord1 = *icoord0 + 1;
463 if (*icoord1 > (int) size - 1)
464 *icoord1 = size - 1;
465 *w = frac(u);
466 }
467
468
469 /**
470 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
471 */
472 static void
473 wrap_linear_unorm_clamp_to_edge(float s, unsigned size, int offset,
474 int *icoord0, int *icoord1, float *w)
475 {
476 float u = CLAMP(s + offset, +0.5F, (float) size - 0.5F);
477 u -= 0.5F;
478 *icoord0 = util_ifloor(u);
479 *icoord1 = *icoord0 + 1;
480 if (*icoord1 > (int) size - 1)
481 *icoord1 = size - 1;
482 *w = frac(u);
483 }
484
485
486 /**
487 * Do coordinate to array index conversion. For array textures.
488 */
489 static inline int
490 coord_to_layer(float coord, unsigned first_layer, unsigned last_layer)
491 {
492 int c = util_ifloor(coord + 0.5F);
493 return CLAMP(c, (int)first_layer, (int)last_layer);
494 }
495
496
497 /**
498 * Examine the quad's texture coordinates to compute the partial
499 * derivatives w.r.t X and Y, then compute lambda (level of detail).
500 */
501 static float
502 compute_lambda_1d(const struct sp_sampler_view *sview,
503 const float s[TGSI_QUAD_SIZE],
504 const float t[TGSI_QUAD_SIZE],
505 const float p[TGSI_QUAD_SIZE])
506 {
507 const struct pipe_resource *texture = sview->base.texture;
508 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
509 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
510 float rho = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
511
512 return util_fast_log2(rho);
513 }
514
515
516 static float
517 compute_lambda_2d(const struct sp_sampler_view *sview,
518 const float s[TGSI_QUAD_SIZE],
519 const float t[TGSI_QUAD_SIZE],
520 const float p[TGSI_QUAD_SIZE])
521 {
522 const struct pipe_resource *texture = sview->base.texture;
523 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
524 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
525 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
526 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
527 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
528 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
529 float rho = MAX2(maxx, maxy);
530
531 return util_fast_log2(rho);
532 }
533
534
535 static float
536 compute_lambda_3d(const struct sp_sampler_view *sview,
537 const float s[TGSI_QUAD_SIZE],
538 const float t[TGSI_QUAD_SIZE],
539 const float p[TGSI_QUAD_SIZE])
540 {
541 const struct pipe_resource *texture = sview->base.texture;
542 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
543 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
544 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
545 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
546 float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
547 float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
548 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, sview->base.u.tex.first_level);
549 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, sview->base.u.tex.first_level);
550 float maxz = MAX2(dpdx, dpdy) * u_minify(texture->depth0, sview->base.u.tex.first_level);
551 float rho;
552
553 rho = MAX2(maxx, maxy);
554 rho = MAX2(rho, maxz);
555
556 return util_fast_log2(rho);
557 }
558
559
560 /**
561 * Compute lambda for a vertex texture sampler.
562 * Since there aren't derivatives to use, just return 0.
563 */
564 static float
565 compute_lambda_vert(const struct sp_sampler_view *sview,
566 const float s[TGSI_QUAD_SIZE],
567 const float t[TGSI_QUAD_SIZE],
568 const float p[TGSI_QUAD_SIZE])
569 {
570 return 0.0f;
571 }
572
573
574
575 /**
576 * Get a texel from a texture, using the texture tile cache.
577 *
578 * \param addr the template tex address containing cube, z, face info.
579 * \param x the x coord of texel within 2D image
580 * \param y the y coord of texel within 2D image
581 * \param rgba the quad to put the texel/color into
582 *
583 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
584 * sp_get_cached_tile_tex() function.
585 */
586
587
588
589
590 static inline const float *
591 get_texel_2d_no_border(const struct sp_sampler_view *sp_sview,
592 union tex_tile_address addr, int x, int y)
593 {
594 const struct softpipe_tex_cached_tile *tile;
595 addr.bits.x = x / TEX_TILE_SIZE;
596 addr.bits.y = y / TEX_TILE_SIZE;
597 y %= TEX_TILE_SIZE;
598 x %= TEX_TILE_SIZE;
599
600 tile = sp_get_cached_tile_tex(sp_sview->cache, addr);
601
602 return &tile->data.color[y][x][0];
603 }
604
605
606 static inline const float *
607 get_texel_2d(const struct sp_sampler_view *sp_sview,
608 const struct sp_sampler *sp_samp,
609 union tex_tile_address addr, int x, int y)
610 {
611 const struct pipe_resource *texture = sp_sview->base.texture;
612 unsigned level = addr.bits.level;
613
614 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
615 y < 0 || y >= (int) u_minify(texture->height0, level)) {
616 return sp_samp->base.border_color.f;
617 }
618 else {
619 return get_texel_2d_no_border( sp_sview, addr, x, y );
620 }
621 }
622
623
624 /*
625 * Here's the complete logic (HOLY CRAP) for finding next face and doing the
626 * corresponding coord wrapping, implemented by get_next_face,
627 * get_next_xcoord, get_next_ycoord.
628 * Read like that (first line):
629 * If face is +x and s coord is below zero, then
630 * new face is +z, new s is max , new t is old t
631 * (max is always cube size - 1).
632 *
633 * +x s- -> +z: s = max, t = t
634 * +x s+ -> -z: s = 0, t = t
635 * +x t- -> +y: s = max, t = max-s
636 * +x t+ -> -y: s = max, t = s
637 *
638 * -x s- -> -z: s = max, t = t
639 * -x s+ -> +z: s = 0, t = t
640 * -x t- -> +y: s = 0, t = s
641 * -x t+ -> -y: s = 0, t = max-s
642 *
643 * +y s- -> -x: s = t, t = 0
644 * +y s+ -> +x: s = max-t, t = 0
645 * +y t- -> -z: s = max-s, t = 0
646 * +y t+ -> +z: s = s, t = 0
647 *
648 * -y s- -> -x: s = max-t, t = max
649 * -y s+ -> +x: s = t, t = max
650 * -y t- -> +z: s = s, t = max
651 * -y t+ -> -z: s = max-s, t = max
652
653 * +z s- -> -x: s = max, t = t
654 * +z s+ -> +x: s = 0, t = t
655 * +z t- -> +y: s = s, t = max
656 * +z t+ -> -y: s = s, t = 0
657
658 * -z s- -> +x: s = max, t = t
659 * -z s+ -> -x: s = 0, t = t
660 * -z t- -> +y: s = max-s, t = 0
661 * -z t+ -> -y: s = max-s, t = max
662 */
663
664
665 /*
666 * seamless cubemap neighbour array.
667 * this array is used to find the adjacent face in each of 4 directions,
668 * left, right, up, down. (or -x, +x, -y, +y).
669 */
670 static const unsigned face_array[PIPE_TEX_FACE_MAX][4] = {
671 /* pos X first then neg X is Z different, Y the same */
672 /* PIPE_TEX_FACE_POS_X,*/
673 { PIPE_TEX_FACE_POS_Z, PIPE_TEX_FACE_NEG_Z,
674 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y },
675 /* PIPE_TEX_FACE_NEG_X */
676 { PIPE_TEX_FACE_NEG_Z, PIPE_TEX_FACE_POS_Z,
677 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y },
678
679 /* pos Y first then neg Y is X different, X the same */
680 /* PIPE_TEX_FACE_POS_Y */
681 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
682 PIPE_TEX_FACE_NEG_Z, PIPE_TEX_FACE_POS_Z },
683
684 /* PIPE_TEX_FACE_NEG_Y */
685 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
686 PIPE_TEX_FACE_POS_Z, PIPE_TEX_FACE_NEG_Z },
687
688 /* pos Z first then neg Y is X different, X the same */
689 /* PIPE_TEX_FACE_POS_Z */
690 { PIPE_TEX_FACE_NEG_X, PIPE_TEX_FACE_POS_X,
691 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y },
692
693 /* PIPE_TEX_FACE_NEG_Z */
694 { PIPE_TEX_FACE_POS_X, PIPE_TEX_FACE_NEG_X,
695 PIPE_TEX_FACE_POS_Y, PIPE_TEX_FACE_NEG_Y }
696 };
697
698 static inline unsigned
699 get_next_face(unsigned face, int idx)
700 {
701 return face_array[face][idx];
702 }
703
704 /*
705 * return a new xcoord based on old face, old coords, cube size
706 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
707 */
708 static inline int
709 get_next_xcoord(unsigned face, unsigned fall_off_index, int max, int xc, int yc)
710 {
711 if ((face == 0 && fall_off_index != 1) ||
712 (face == 1 && fall_off_index == 0) ||
713 (face == 4 && fall_off_index == 0) ||
714 (face == 5 && fall_off_index == 0)) {
715 return max;
716 }
717 if ((face == 1 && fall_off_index != 0) ||
718 (face == 0 && fall_off_index == 1) ||
719 (face == 4 && fall_off_index == 1) ||
720 (face == 5 && fall_off_index == 1)) {
721 return 0;
722 }
723 if ((face == 4 && fall_off_index >= 2) ||
724 (face == 2 && fall_off_index == 3) ||
725 (face == 3 && fall_off_index == 2)) {
726 return xc;
727 }
728 if ((face == 5 && fall_off_index >= 2) ||
729 (face == 2 && fall_off_index == 2) ||
730 (face == 3 && fall_off_index == 3)) {
731 return max - xc;
732 }
733 if ((face == 2 && fall_off_index == 0) ||
734 (face == 3 && fall_off_index == 1)) {
735 return yc;
736 }
737 /* (face == 2 && fall_off_index == 1) ||
738 (face == 3 && fall_off_index == 0)) */
739 return max - yc;
740 }
741
742 /*
743 * return a new ycoord based on old face, old coords, cube size
744 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
745 */
746 static inline int
747 get_next_ycoord(unsigned face, unsigned fall_off_index, int max, int xc, int yc)
748 {
749 if ((fall_off_index <= 1) && (face <= 1 || face >= 4)) {
750 return yc;
751 }
752 if (face == 2 ||
753 (face == 4 && fall_off_index == 3) ||
754 (face == 5 && fall_off_index == 2)) {
755 return 0;
756 }
757 if (face == 3 ||
758 (face == 4 && fall_off_index == 2) ||
759 (face == 5 && fall_off_index == 3)) {
760 return max;
761 }
762 if ((face == 0 && fall_off_index == 3) ||
763 (face == 1 && fall_off_index == 2)) {
764 return xc;
765 }
766 /* (face == 0 && fall_off_index == 2) ||
767 (face == 1 && fall_off_index == 3) */
768 return max - xc;
769 }
770
771
772 /* Gather a quad of adjacent texels within a tile:
773 */
774 static inline void
775 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_view *sp_sview,
776 union tex_tile_address addr,
777 unsigned x, unsigned y,
778 const float *out[4])
779 {
780 const struct softpipe_tex_cached_tile *tile;
781
782 addr.bits.x = x / TEX_TILE_SIZE;
783 addr.bits.y = y / TEX_TILE_SIZE;
784 y %= TEX_TILE_SIZE;
785 x %= TEX_TILE_SIZE;
786
787 tile = sp_get_cached_tile_tex(sp_sview->cache, addr);
788
789 out[0] = &tile->data.color[y ][x ][0];
790 out[1] = &tile->data.color[y ][x+1][0];
791 out[2] = &tile->data.color[y+1][x ][0];
792 out[3] = &tile->data.color[y+1][x+1][0];
793 }
794
795
796 /* Gather a quad of potentially non-adjacent texels:
797 */
798 static inline void
799 get_texel_quad_2d_no_border(const struct sp_sampler_view *sp_sview,
800 union tex_tile_address addr,
801 int x0, int y0,
802 int x1, int y1,
803 const float *out[4])
804 {
805 out[0] = get_texel_2d_no_border( sp_sview, addr, x0, y0 );
806 out[1] = get_texel_2d_no_border( sp_sview, addr, x1, y0 );
807 out[2] = get_texel_2d_no_border( sp_sview, addr, x0, y1 );
808 out[3] = get_texel_2d_no_border( sp_sview, addr, x1, y1 );
809 }
810
811 /* Can involve a lot of unnecessary checks for border color:
812 */
813 static inline void
814 get_texel_quad_2d(const struct sp_sampler_view *sp_sview,
815 const struct sp_sampler *sp_samp,
816 union tex_tile_address addr,
817 int x0, int y0,
818 int x1, int y1,
819 const float *out[4])
820 {
821 out[0] = get_texel_2d( sp_sview, sp_samp, addr, x0, y0 );
822 out[1] = get_texel_2d( sp_sview, sp_samp, addr, x1, y0 );
823 out[3] = get_texel_2d( sp_sview, sp_samp, addr, x1, y1 );
824 out[2] = get_texel_2d( sp_sview, sp_samp, addr, x0, y1 );
825 }
826
827
828
829 /* 3d variants:
830 */
831 static inline const float *
832 get_texel_3d_no_border(const struct sp_sampler_view *sp_sview,
833 union tex_tile_address addr, int x, int y, int z)
834 {
835 const struct softpipe_tex_cached_tile *tile;
836
837 addr.bits.x = x / TEX_TILE_SIZE;
838 addr.bits.y = y / TEX_TILE_SIZE;
839 addr.bits.z = z;
840 y %= TEX_TILE_SIZE;
841 x %= TEX_TILE_SIZE;
842
843 tile = sp_get_cached_tile_tex(sp_sview->cache, addr);
844
845 return &tile->data.color[y][x][0];
846 }
847
848
849 static inline const float *
850 get_texel_3d(const struct sp_sampler_view *sp_sview,
851 const struct sp_sampler *sp_samp,
852 union tex_tile_address addr, int x, int y, int z)
853 {
854 const struct pipe_resource *texture = sp_sview->base.texture;
855 unsigned level = addr.bits.level;
856
857 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
858 y < 0 || y >= (int) u_minify(texture->height0, level) ||
859 z < 0 || z >= (int) u_minify(texture->depth0, level)) {
860 return sp_samp->base.border_color.f;
861 }
862 else {
863 return get_texel_3d_no_border( sp_sview, addr, x, y, z );
864 }
865 }
866
867
868 /* Get texel pointer for 1D array texture */
869 static inline const float *
870 get_texel_1d_array(const struct sp_sampler_view *sp_sview,
871 const struct sp_sampler *sp_samp,
872 union tex_tile_address addr, int x, int y)
873 {
874 const struct pipe_resource *texture = sp_sview->base.texture;
875 unsigned level = addr.bits.level;
876
877 if (x < 0 || x >= (int) u_minify(texture->width0, level)) {
878 return sp_samp->base.border_color.f;
879 }
880 else {
881 return get_texel_2d_no_border(sp_sview, addr, x, y);
882 }
883 }
884
885
886 /* Get texel pointer for 2D array texture */
887 static inline const float *
888 get_texel_2d_array(const struct sp_sampler_view *sp_sview,
889 const struct sp_sampler *sp_samp,
890 union tex_tile_address addr, int x, int y, int layer)
891 {
892 const struct pipe_resource *texture = sp_sview->base.texture;
893 unsigned level = addr.bits.level;
894
895 assert(layer < (int) texture->array_size);
896 assert(layer >= 0);
897
898 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
899 y < 0 || y >= (int) u_minify(texture->height0, level)) {
900 return sp_samp->base.border_color.f;
901 }
902 else {
903 return get_texel_3d_no_border(sp_sview, addr, x, y, layer);
904 }
905 }
906
907
908 static inline const float *
909 get_texel_cube_seamless(const struct sp_sampler_view *sp_sview,
910 union tex_tile_address addr, int x, int y,
911 float *corner, int layer, unsigned face)
912 {
913 const struct pipe_resource *texture = sp_sview->base.texture;
914 unsigned level = addr.bits.level;
915 int new_x, new_y, max_x;
916
917 max_x = (int) u_minify(texture->width0, level);
918
919 assert(texture->width0 == texture->height0);
920 new_x = x;
921 new_y = y;
922
923 /* change the face */
924 if (x < 0) {
925 /*
926 * Cheat with corners. They are difficult and I believe because we don't get
927 * per-pixel faces we can actually have multiple corner texels per pixel,
928 * which screws things up majorly in any case (as the per spec behavior is
929 * to average the 3 remaining texels, which we might not have).
930 * Hence just make sure that the 2nd coord is clamped, will simply pick the
931 * sample which would have fallen off the x coord, but not y coord.
932 * So the filter weight of the samples will be wrong, but at least this
933 * ensures that only valid texels near the corner are used.
934 */
935 if (y < 0 || y >= max_x) {
936 y = CLAMP(y, 0, max_x - 1);
937 }
938 new_x = get_next_xcoord(face, 0, max_x -1, x, y);
939 new_y = get_next_ycoord(face, 0, max_x -1, x, y);
940 face = get_next_face(face, 0);
941 } else if (x >= max_x) {
942 if (y < 0 || y >= max_x) {
943 y = CLAMP(y, 0, max_x - 1);
944 }
945 new_x = get_next_xcoord(face, 1, max_x -1, x, y);
946 new_y = get_next_ycoord(face, 1, max_x -1, x, y);
947 face = get_next_face(face, 1);
948 } else if (y < 0) {
949 new_x = get_next_xcoord(face, 2, max_x -1, x, y);
950 new_y = get_next_ycoord(face, 2, max_x -1, x, y);
951 face = get_next_face(face, 2);
952 } else if (y >= max_x) {
953 new_x = get_next_xcoord(face, 3, max_x -1, x, y);
954 new_y = get_next_ycoord(face, 3, max_x -1, x, y);
955 face = get_next_face(face, 3);
956 }
957
958 return get_texel_3d_no_border(sp_sview, addr, new_x, new_y, layer + face);
959 }
960
961
962 /* Get texel pointer for cube array texture */
963 static inline const float *
964 get_texel_cube_array(const struct sp_sampler_view *sp_sview,
965 const struct sp_sampler *sp_samp,
966 union tex_tile_address addr, int x, int y, int layer)
967 {
968 const struct pipe_resource *texture = sp_sview->base.texture;
969 unsigned level = addr.bits.level;
970
971 assert(layer < (int) texture->array_size);
972 assert(layer >= 0);
973
974 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
975 y < 0 || y >= (int) u_minify(texture->height0, level)) {
976 return sp_samp->base.border_color.f;
977 }
978 else {
979 return get_texel_3d_no_border(sp_sview, addr, x, y, layer);
980 }
981 }
982 /**
983 * Given the logbase2 of a mipmap's base level size and a mipmap level,
984 * return the size (in texels) of that mipmap level.
985 * For example, if level[0].width = 256 then base_pot will be 8.
986 * If level = 2, then we'll return 64 (the width at level=2).
987 * Return 1 if level > base_pot.
988 */
989 static inline unsigned
990 pot_level_size(unsigned base_pot, unsigned level)
991 {
992 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
993 }
994
995
996 static void
997 print_sample(const char *function, const float *rgba)
998 {
999 debug_printf("%s %g %g %g %g\n",
1000 function,
1001 rgba[0], rgba[TGSI_NUM_CHANNELS], rgba[2*TGSI_NUM_CHANNELS], rgba[3*TGSI_NUM_CHANNELS]);
1002 }
1003
1004
1005 static void
1006 print_sample_4(const char *function, float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1007 {
1008 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
1009 function,
1010 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
1011 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
1012 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
1013 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
1014 }
1015
1016
1017 /* Some image-filter fastpaths:
1018 */
1019 static inline void
1020 img_filter_2d_linear_repeat_POT(struct sp_sampler_view *sp_sview,
1021 struct sp_sampler *sp_samp,
1022 const struct img_filter_args *args,
1023 float *rgba)
1024 {
1025 unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
1026 unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
1027 int xmax = (xpot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
1028 int ymax = (ypot - 1) & (TEX_TILE_SIZE - 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
1029 union tex_tile_address addr;
1030 int c;
1031
1032 float u = (args->s * xpot - 0.5F) + args->offset[0];
1033 float v = (args->t * ypot - 0.5F) + args->offset[1];
1034
1035 int uflr = util_ifloor(u);
1036 int vflr = util_ifloor(v);
1037
1038 float xw = u - (float)uflr;
1039 float yw = v - (float)vflr;
1040
1041 int x0 = uflr & (xpot - 1);
1042 int y0 = vflr & (ypot - 1);
1043
1044 const float *tx[4];
1045
1046 addr.value = 0;
1047 addr.bits.level = args->level;
1048
1049 /* Can we fetch all four at once:
1050 */
1051 if (x0 < xmax && y0 < ymax) {
1052 get_texel_quad_2d_no_border_single_tile(sp_sview, addr, x0, y0, tx);
1053 }
1054 else {
1055 unsigned x1 = (x0 + 1) & (xpot - 1);
1056 unsigned y1 = (y0 + 1) & (ypot - 1);
1057 get_texel_quad_2d_no_border(sp_sview, addr, x0, y0, x1, y1, tx);
1058 }
1059
1060 /* interpolate R, G, B, A */
1061 for (c = 0; c < TGSI_QUAD_SIZE; c++) {
1062 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1063 tx[0][c], tx[1][c],
1064 tx[2][c], tx[3][c]);
1065 }
1066
1067 if (DEBUG_TEX) {
1068 print_sample(__FUNCTION__, rgba);
1069 }
1070 }
1071
1072
1073 static inline void
1074 img_filter_2d_nearest_repeat_POT(struct sp_sampler_view *sp_sview,
1075 struct sp_sampler *sp_samp,
1076 const struct img_filter_args *args,
1077 float rgba[TGSI_QUAD_SIZE])
1078 {
1079 unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
1080 unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
1081 const float *out;
1082 union tex_tile_address addr;
1083 int c;
1084
1085 float u = args->s * xpot + args->offset[0];
1086 float v = args->t * ypot + args->offset[1];
1087
1088 int uflr = util_ifloor(u);
1089 int vflr = util_ifloor(v);
1090
1091 int x0 = uflr & (xpot - 1);
1092 int y0 = vflr & (ypot - 1);
1093
1094 addr.value = 0;
1095 addr.bits.level = args->level;
1096
1097 out = get_texel_2d_no_border(sp_sview, addr, x0, y0);
1098 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1099 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1100
1101 if (DEBUG_TEX) {
1102 print_sample(__FUNCTION__, rgba);
1103 }
1104 }
1105
1106
1107 static inline void
1108 img_filter_2d_nearest_clamp_POT(struct sp_sampler_view *sp_sview,
1109 struct sp_sampler *sp_samp,
1110 const struct img_filter_args *args,
1111 float rgba[TGSI_QUAD_SIZE])
1112 {
1113 unsigned xpot = pot_level_size(sp_sview->xpot, args->level);
1114 unsigned ypot = pot_level_size(sp_sview->ypot, args->level);
1115 union tex_tile_address addr;
1116 int c;
1117
1118 float u = args->s * xpot + args->offset[0];
1119 float v = args->t * ypot + args->offset[1];
1120
1121 int x0, y0;
1122 const float *out;
1123
1124 addr.value = 0;
1125 addr.bits.level = args->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 const struct img_filter_args *args,
1153 float rgba[TGSI_QUAD_SIZE])
1154 {
1155 const struct pipe_resource *texture = sp_sview->base.texture;
1156 int width;
1157 int x;
1158 union tex_tile_address addr;
1159 const float *out;
1160 int c;
1161
1162 width = u_minify(texture->width0, args->level);
1163
1164 assert(width > 0);
1165
1166 addr.value = 0;
1167 addr.bits.level = args->level;
1168
1169 sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
1170
1171 out = get_texel_2d(sp_sview, sp_samp, addr, x, 0);
1172 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1173 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1174
1175 if (DEBUG_TEX) {
1176 print_sample(__FUNCTION__, rgba);
1177 }
1178 }
1179
1180
1181 static void
1182 img_filter_1d_array_nearest(struct sp_sampler_view *sp_sview,
1183 struct sp_sampler *sp_samp,
1184 const struct img_filter_args *args,
1185 float *rgba)
1186 {
1187 const struct pipe_resource *texture = sp_sview->base.texture;
1188 int width;
1189 int x, layer;
1190 union tex_tile_address addr;
1191 const float *out;
1192 int c;
1193
1194 width = u_minify(texture->width0, args->level);
1195
1196 assert(width > 0);
1197
1198 addr.value = 0;
1199 addr.bits.level = args->level;
1200
1201 sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
1202 layer = coord_to_layer(args->t, sp_sview->base.u.tex.first_layer,
1203 sp_sview->base.u.tex.last_layer);
1204
1205 out = get_texel_1d_array(sp_sview, sp_samp, addr, x, layer);
1206 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1207 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1208
1209 if (DEBUG_TEX) {
1210 print_sample(__FUNCTION__, rgba);
1211 }
1212 }
1213
1214
1215 static void
1216 img_filter_2d_nearest(struct sp_sampler_view *sp_sview,
1217 struct sp_sampler *sp_samp,
1218 const struct img_filter_args *args,
1219 float *rgba)
1220 {
1221 const struct pipe_resource *texture = sp_sview->base.texture;
1222 int width, height;
1223 int x, y;
1224 union tex_tile_address addr;
1225 const float *out;
1226 int c;
1227
1228 width = u_minify(texture->width0, args->level);
1229 height = u_minify(texture->height0, args->level);
1230
1231 assert(width > 0);
1232 assert(height > 0);
1233
1234 addr.value = 0;
1235 addr.bits.level = args->level;
1236
1237 sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
1238 sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
1239
1240 out = get_texel_2d(sp_sview, sp_samp, addr, x, y);
1241 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1242 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1243
1244 if (DEBUG_TEX) {
1245 print_sample(__FUNCTION__, rgba);
1246 }
1247 }
1248
1249
1250 static void
1251 img_filter_2d_array_nearest(struct sp_sampler_view *sp_sview,
1252 struct sp_sampler *sp_samp,
1253 const struct img_filter_args *args,
1254 float *rgba)
1255 {
1256 const struct pipe_resource *texture = sp_sview->base.texture;
1257 int width, height;
1258 int x, y, layer;
1259 union tex_tile_address addr;
1260 const float *out;
1261 int c;
1262
1263 width = u_minify(texture->width0, args->level);
1264 height = u_minify(texture->height0, args->level);
1265
1266 assert(width > 0);
1267 assert(height > 0);
1268
1269 addr.value = 0;
1270 addr.bits.level = args->level;
1271
1272 sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
1273 sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
1274 layer = coord_to_layer(args->p, sp_sview->base.u.tex.first_layer,
1275 sp_sview->base.u.tex.last_layer);
1276
1277 out = get_texel_2d_array(sp_sview, sp_samp, addr, x, y, layer);
1278 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1279 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1280
1281 if (DEBUG_TEX) {
1282 print_sample(__FUNCTION__, rgba);
1283 }
1284 }
1285
1286
1287 static void
1288 img_filter_cube_nearest(struct sp_sampler_view *sp_sview,
1289 struct sp_sampler *sp_samp,
1290 const struct img_filter_args *args,
1291 float *rgba)
1292 {
1293 const struct pipe_resource *texture = sp_sview->base.texture;
1294 int width, height;
1295 int x, y, layerface;
1296 union tex_tile_address addr;
1297 const float *out;
1298 int c;
1299
1300 width = u_minify(texture->width0, args->level);
1301 height = u_minify(texture->height0, args->level);
1302
1303 assert(width > 0);
1304 assert(height > 0);
1305
1306 addr.value = 0;
1307 addr.bits.level = args->level;
1308
1309 /*
1310 * If NEAREST filtering is done within a miplevel, always apply wrap
1311 * mode CLAMP_TO_EDGE.
1312 */
1313 if (sp_samp->base.seamless_cube_map) {
1314 wrap_nearest_clamp_to_edge(args->s, width, args->offset[0], &x);
1315 wrap_nearest_clamp_to_edge(args->t, height, args->offset[1], &y);
1316 } else {
1317 /* Would probably make sense to ignore mode and just do edge clamp */
1318 sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
1319 sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
1320 }
1321
1322 layerface = args->face_id + sp_sview->base.u.tex.first_layer;
1323 out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
1324 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1325 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1326
1327 if (DEBUG_TEX) {
1328 print_sample(__FUNCTION__, rgba);
1329 }
1330 }
1331
1332 static void
1333 img_filter_cube_array_nearest(struct sp_sampler_view *sp_sview,
1334 struct sp_sampler *sp_samp,
1335 const struct img_filter_args *args,
1336 float *rgba)
1337 {
1338 const struct pipe_resource *texture = sp_sview->base.texture;
1339 int width, height;
1340 int x, y, layerface;
1341 union tex_tile_address addr;
1342 const float *out;
1343 int c;
1344
1345 width = u_minify(texture->width0, args->level);
1346 height = u_minify(texture->height0, args->level);
1347
1348 assert(width > 0);
1349 assert(height > 0);
1350
1351 addr.value = 0;
1352 addr.bits.level = args->level;
1353
1354 sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
1355 sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
1356 layerface = coord_to_layer(6 * args->p + sp_sview->base.u.tex.first_layer,
1357 sp_sview->base.u.tex.first_layer,
1358 sp_sview->base.u.tex.last_layer - 5) + args->face_id;
1359
1360 out = get_texel_cube_array(sp_sview, sp_samp, addr, x, y, layerface);
1361 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1362 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1363
1364 if (DEBUG_TEX) {
1365 print_sample(__FUNCTION__, rgba);
1366 }
1367 }
1368
1369 static void
1370 img_filter_3d_nearest(struct sp_sampler_view *sp_sview,
1371 struct sp_sampler *sp_samp,
1372 const struct img_filter_args *args,
1373 float *rgba)
1374 {
1375 const struct pipe_resource *texture = sp_sview->base.texture;
1376 int width, height, depth;
1377 int x, y, z;
1378 union tex_tile_address addr;
1379 const float *out;
1380 int c;
1381
1382 width = u_minify(texture->width0, args->level);
1383 height = u_minify(texture->height0, args->level);
1384 depth = u_minify(texture->depth0, args->level);
1385
1386 assert(width > 0);
1387 assert(height > 0);
1388 assert(depth > 0);
1389
1390 sp_samp->nearest_texcoord_s(args->s, width, args->offset[0], &x);
1391 sp_samp->nearest_texcoord_t(args->t, height, args->offset[1], &y);
1392 sp_samp->nearest_texcoord_p(args->p, depth, args->offset[2], &z);
1393
1394 addr.value = 0;
1395 addr.bits.level = args->level;
1396
1397 out = get_texel_3d(sp_sview, sp_samp, addr, x, y, z);
1398 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1399 rgba[TGSI_NUM_CHANNELS*c] = out[c];
1400 }
1401
1402
1403 static void
1404 img_filter_1d_linear(struct sp_sampler_view *sp_sview,
1405 struct sp_sampler *sp_samp,
1406 const struct img_filter_args *args,
1407 float *rgba)
1408 {
1409 const struct pipe_resource *texture = sp_sview->base.texture;
1410 int width;
1411 int x0, x1;
1412 float xw; /* weights */
1413 union tex_tile_address addr;
1414 const float *tx0, *tx1;
1415 int c;
1416
1417 width = u_minify(texture->width0, args->level);
1418
1419 assert(width > 0);
1420
1421 addr.value = 0;
1422 addr.bits.level = args->level;
1423
1424 sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
1425
1426 tx0 = get_texel_2d(sp_sview, sp_samp, addr, x0, 0);
1427 tx1 = get_texel_2d(sp_sview, sp_samp, addr, x1, 0);
1428
1429 /* interpolate R, G, B, A */
1430 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1431 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1432 }
1433
1434
1435 static void
1436 img_filter_1d_array_linear(struct sp_sampler_view *sp_sview,
1437 struct sp_sampler *sp_samp,
1438 const struct img_filter_args *args,
1439 float *rgba)
1440 {
1441 const struct pipe_resource *texture = sp_sview->base.texture;
1442 int width;
1443 int x0, x1, layer;
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, args->level);
1450
1451 assert(width > 0);
1452
1453 addr.value = 0;
1454 addr.bits.level = args->level;
1455
1456 sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
1457 layer = coord_to_layer(args->t, sp_sview->base.u.tex.first_layer,
1458 sp_sview->base.u.tex.last_layer);
1459
1460 tx0 = get_texel_1d_array(sp_sview, sp_samp, addr, x0, layer);
1461 tx1 = get_texel_1d_array(sp_sview, sp_samp, addr, x1, layer);
1462
1463 /* interpolate R, G, B, A */
1464 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1465 rgba[TGSI_NUM_CHANNELS*c] = lerp(xw, tx0[c], tx1[c]);
1466 }
1467
1468 /*
1469 * Retrieve the gathered value, need to convert to the
1470 * TGSI expected interface, and take component select
1471 * and swizzling into account.
1472 */
1473 static float
1474 get_gather_value(const struct sp_sampler_view *sp_sview,
1475 int chan_in, int comp_sel,
1476 const float *tx[4])
1477 {
1478 int chan;
1479 unsigned swizzle;
1480
1481 /*
1482 * softpipe samples in a different order
1483 * to TGSI expects, so we need to swizzle,
1484 * the samples into the correct slots.
1485 */
1486 switch (chan_in) {
1487 case 0:
1488 chan = 2;
1489 break;
1490 case 1:
1491 chan = 3;
1492 break;
1493 case 2:
1494 chan = 1;
1495 break;
1496 case 3:
1497 chan = 0;
1498 break;
1499 default:
1500 assert(0);
1501 return 0.0;
1502 }
1503
1504 /* pick which component to use for the swizzle */
1505 switch (comp_sel) {
1506 case 0:
1507 swizzle = sp_sview->base.swizzle_r;
1508 break;
1509 case 1:
1510 swizzle = sp_sview->base.swizzle_g;
1511 break;
1512 case 2:
1513 swizzle = sp_sview->base.swizzle_b;
1514 break;
1515 case 3:
1516 swizzle = sp_sview->base.swizzle_a;
1517 break;
1518 default:
1519 assert(0);
1520 return 0.0;
1521 }
1522
1523 /* get correct result using the channel and swizzle */
1524 switch (swizzle) {
1525 case PIPE_SWIZZLE_ZERO:
1526 return 0.0;
1527 case PIPE_SWIZZLE_ONE:
1528 return 1.0;
1529 default:
1530 return tx[chan][swizzle];
1531 }
1532 }
1533
1534
1535 static void
1536 img_filter_2d_linear(struct sp_sampler_view *sp_sview,
1537 struct sp_sampler *sp_samp,
1538 const struct img_filter_args *args,
1539 float *rgba)
1540 {
1541 const struct pipe_resource *texture = sp_sview->base.texture;
1542 int width, height;
1543 int x0, y0, x1, y1;
1544 float xw, yw; /* weights */
1545 union tex_tile_address addr;
1546 const float *tx[4];
1547 int c;
1548
1549 width = u_minify(texture->width0, args->level);
1550 height = u_minify(texture->height0, args->level);
1551
1552 assert(width > 0);
1553 assert(height > 0);
1554
1555 addr.value = 0;
1556 addr.bits.level = args->level;
1557
1558 sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
1559 sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
1560
1561 tx[0] = get_texel_2d(sp_sview, sp_samp, addr, x0, y0);
1562 tx[1] = get_texel_2d(sp_sview, sp_samp, addr, x1, y0);
1563 tx[2] = get_texel_2d(sp_sview, sp_samp, addr, x0, y1);
1564 tx[3] = get_texel_2d(sp_sview, sp_samp, addr, x1, y1);
1565
1566 if (args->gather_only) {
1567 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1568 rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
1569 args->gather_comp,
1570 tx);
1571 } else {
1572 /* interpolate R, G, B, A */
1573 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1574 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1575 tx[0][c], tx[1][c],
1576 tx[2][c], tx[3][c]);
1577 }
1578 }
1579
1580
1581 static void
1582 img_filter_2d_array_linear(struct sp_sampler_view *sp_sview,
1583 struct sp_sampler *sp_samp,
1584 const struct img_filter_args *args,
1585 float *rgba)
1586 {
1587 const struct pipe_resource *texture = sp_sview->base.texture;
1588 int width, height;
1589 int x0, y0, x1, y1, layer;
1590 float xw, yw; /* weights */
1591 union tex_tile_address addr;
1592 const float *tx[4];
1593 int c;
1594
1595 width = u_minify(texture->width0, args->level);
1596 height = u_minify(texture->height0, args->level);
1597
1598 assert(width > 0);
1599 assert(height > 0);
1600
1601 addr.value = 0;
1602 addr.bits.level = args->level;
1603
1604 sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
1605 sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
1606 layer = coord_to_layer(args->p, sp_sview->base.u.tex.first_layer,
1607 sp_sview->base.u.tex.last_layer);
1608
1609 tx[0] = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y0, layer);
1610 tx[1] = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y0, layer);
1611 tx[2] = get_texel_2d_array(sp_sview, sp_samp, addr, x0, y1, layer);
1612 tx[3] = get_texel_2d_array(sp_sview, sp_samp, addr, x1, y1, layer);
1613
1614 if (args->gather_only) {
1615 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1616 rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
1617 args->gather_comp,
1618 tx);
1619 } else {
1620 /* interpolate R, G, B, A */
1621 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1622 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1623 tx[0][c], tx[1][c],
1624 tx[2][c], tx[3][c]);
1625 }
1626 }
1627
1628
1629 static void
1630 img_filter_cube_linear(struct sp_sampler_view *sp_sview,
1631 struct sp_sampler *sp_samp,
1632 const struct img_filter_args *args,
1633 float *rgba)
1634 {
1635 const struct pipe_resource *texture = sp_sview->base.texture;
1636 int width, height;
1637 int x0, y0, x1, y1, layer;
1638 float xw, yw; /* weights */
1639 union tex_tile_address addr;
1640 const float *tx[4];
1641 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
1642 corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1643 int c;
1644
1645 width = u_minify(texture->width0, args->level);
1646 height = u_minify(texture->height0, args->level);
1647
1648 assert(width > 0);
1649 assert(height > 0);
1650
1651 addr.value = 0;
1652 addr.bits.level = args->level;
1653
1654 /*
1655 * For seamless if LINEAR filtering is done within a miplevel,
1656 * always apply wrap mode CLAMP_TO_BORDER.
1657 */
1658 if (sp_samp->base.seamless_cube_map) {
1659 /* Note this is a bit overkill, actual clamping is not required */
1660 wrap_linear_clamp_to_border(args->s, width, args->offset[0], &x0, &x1, &xw);
1661 wrap_linear_clamp_to_border(args->t, height, args->offset[1], &y0, &y1, &yw);
1662 } else {
1663 /* Would probably make sense to ignore mode and just do edge clamp */
1664 sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
1665 sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
1666 }
1667
1668 layer = sp_sview->base.u.tex.first_layer;
1669
1670 if (sp_samp->base.seamless_cube_map) {
1671 tx[0] = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, args->face_id);
1672 tx[1] = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, args->face_id);
1673 tx[2] = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, args->face_id);
1674 tx[3] = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, args->face_id);
1675 } else {
1676 tx[0] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + args->face_id);
1677 tx[1] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + args->face_id);
1678 tx[2] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + args->face_id);
1679 tx[3] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + args->face_id);
1680 }
1681
1682 if (args->gather_only) {
1683 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1684 rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
1685 args->gather_comp,
1686 tx);
1687 } else {
1688 /* interpolate R, G, B, A */
1689 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1690 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1691 tx[0][c], tx[1][c],
1692 tx[2][c], tx[3][c]);
1693 }
1694 }
1695
1696
1697 static void
1698 img_filter_cube_array_linear(struct sp_sampler_view *sp_sview,
1699 struct sp_sampler *sp_samp,
1700 const struct img_filter_args *args,
1701 float *rgba)
1702 {
1703 const struct pipe_resource *texture = sp_sview->base.texture;
1704 int width, height;
1705 int x0, y0, x1, y1, layer;
1706 float xw, yw; /* weights */
1707 union tex_tile_address addr;
1708 const float *tx[4];
1709 float corner0[TGSI_QUAD_SIZE], corner1[TGSI_QUAD_SIZE],
1710 corner2[TGSI_QUAD_SIZE], corner3[TGSI_QUAD_SIZE];
1711 int c;
1712
1713 width = u_minify(texture->width0, args->level);
1714 height = u_minify(texture->height0, args->level);
1715
1716 assert(width > 0);
1717 assert(height > 0);
1718
1719 addr.value = 0;
1720 addr.bits.level = args->level;
1721
1722 /*
1723 * For seamless if LINEAR filtering is done within a miplevel,
1724 * always apply wrap mode CLAMP_TO_BORDER.
1725 */
1726 if (sp_samp->base.seamless_cube_map) {
1727 /* Note this is a bit overkill, actual clamping is not required */
1728 wrap_linear_clamp_to_border(args->s, width, args->offset[0], &x0, &x1, &xw);
1729 wrap_linear_clamp_to_border(args->t, height, args->offset[1], &y0, &y1, &yw);
1730 } else {
1731 /* Would probably make sense to ignore mode and just do edge clamp */
1732 sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
1733 sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
1734 }
1735
1736 layer = coord_to_layer(6 * args->p + sp_sview->base.u.tex.first_layer,
1737 sp_sview->base.u.tex.first_layer,
1738 sp_sview->base.u.tex.last_layer - 5);
1739
1740 if (sp_samp->base.seamless_cube_map) {
1741 tx[0] = get_texel_cube_seamless(sp_sview, addr, x0, y0, corner0, layer, args->face_id);
1742 tx[1] = get_texel_cube_seamless(sp_sview, addr, x1, y0, corner1, layer, args->face_id);
1743 tx[2] = get_texel_cube_seamless(sp_sview, addr, x0, y1, corner2, layer, args->face_id);
1744 tx[3] = get_texel_cube_seamless(sp_sview, addr, x1, y1, corner3, layer, args->face_id);
1745 } else {
1746 tx[0] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y0, layer + args->face_id);
1747 tx[1] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y0, layer + args->face_id);
1748 tx[2] = get_texel_cube_array(sp_sview, sp_samp, addr, x0, y1, layer + args->face_id);
1749 tx[3] = get_texel_cube_array(sp_sview, sp_samp, addr, x1, y1, layer + args->face_id);
1750 }
1751
1752 if (args->gather_only) {
1753 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1754 rgba[TGSI_NUM_CHANNELS*c] = get_gather_value(sp_sview, c,
1755 args->gather_comp,
1756 tx);
1757 } else {
1758 /* interpolate R, G, B, A */
1759 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1760 rgba[TGSI_NUM_CHANNELS*c] = lerp_2d(xw, yw,
1761 tx[0][c], tx[1][c],
1762 tx[2][c], tx[3][c]);
1763 }
1764 }
1765
1766 static void
1767 img_filter_3d_linear(struct sp_sampler_view *sp_sview,
1768 struct sp_sampler *sp_samp,
1769 const struct img_filter_args *args,
1770 float *rgba)
1771 {
1772 const struct pipe_resource *texture = sp_sview->base.texture;
1773 int width, height, depth;
1774 int x0, x1, y0, y1, z0, z1;
1775 float xw, yw, zw; /* interpolation weights */
1776 union tex_tile_address addr;
1777 const float *tx00, *tx01, *tx02, *tx03, *tx10, *tx11, *tx12, *tx13;
1778 int c;
1779
1780 width = u_minify(texture->width0, args->level);
1781 height = u_minify(texture->height0, args->level);
1782 depth = u_minify(texture->depth0, args->level);
1783
1784 addr.value = 0;
1785 addr.bits.level = args->level;
1786
1787 assert(width > 0);
1788 assert(height > 0);
1789 assert(depth > 0);
1790
1791 sp_samp->linear_texcoord_s(args->s, width, args->offset[0], &x0, &x1, &xw);
1792 sp_samp->linear_texcoord_t(args->t, height, args->offset[1], &y0, &y1, &yw);
1793 sp_samp->linear_texcoord_p(args->p, depth, args->offset[2], &z0, &z1, &zw);
1794
1795 tx00 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z0);
1796 tx01 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z0);
1797 tx02 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z0);
1798 tx03 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z0);
1799
1800 tx10 = get_texel_3d(sp_sview, sp_samp, addr, x0, y0, z1);
1801 tx11 = get_texel_3d(sp_sview, sp_samp, addr, x1, y0, z1);
1802 tx12 = get_texel_3d(sp_sview, sp_samp, addr, x0, y1, z1);
1803 tx13 = get_texel_3d(sp_sview, sp_samp, addr, x1, y1, z1);
1804
1805 /* interpolate R, G, B, A */
1806 for (c = 0; c < TGSI_QUAD_SIZE; c++)
1807 rgba[TGSI_NUM_CHANNELS*c] = lerp_3d(xw, yw, zw,
1808 tx00[c], tx01[c],
1809 tx02[c], tx03[c],
1810 tx10[c], tx11[c],
1811 tx12[c], tx13[c]);
1812 }
1813
1814
1815 /* Calculate level of detail for every fragment,
1816 * with lambda already computed.
1817 * Note that lambda has already been biased by global LOD bias.
1818 * \param biased_lambda per-quad lambda.
1819 * \param lod_in per-fragment lod_bias or explicit_lod.
1820 * \param lod returns the per-fragment lod.
1821 */
1822 static inline void
1823 compute_lod(const struct pipe_sampler_state *sampler,
1824 enum tgsi_sampler_control control,
1825 const float biased_lambda,
1826 const float lod_in[TGSI_QUAD_SIZE],
1827 float lod[TGSI_QUAD_SIZE])
1828 {
1829 float min_lod = sampler->min_lod;
1830 float max_lod = sampler->max_lod;
1831 uint i;
1832
1833 switch (control) {
1834 case tgsi_sampler_lod_none:
1835 case tgsi_sampler_lod_zero:
1836 /* XXX FIXME */
1837 case tgsi_sampler_derivs_explicit:
1838 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(biased_lambda, min_lod, max_lod);
1839 break;
1840 case tgsi_sampler_lod_bias:
1841 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1842 lod[i] = biased_lambda + lod_in[i];
1843 lod[i] = CLAMP(lod[i], min_lod, max_lod);
1844 }
1845 break;
1846 case tgsi_sampler_lod_explicit:
1847 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1848 lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
1849 }
1850 break;
1851 default:
1852 assert(0);
1853 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1854 }
1855 }
1856
1857
1858 /* Calculate level of detail for every fragment.
1859 * \param lod_in per-fragment lod_bias or explicit_lod.
1860 * \param lod results per-fragment lod.
1861 */
1862 static inline void
1863 compute_lambda_lod(struct sp_sampler_view *sp_sview,
1864 struct sp_sampler *sp_samp,
1865 const float s[TGSI_QUAD_SIZE],
1866 const float t[TGSI_QUAD_SIZE],
1867 const float p[TGSI_QUAD_SIZE],
1868 const float lod_in[TGSI_QUAD_SIZE],
1869 enum tgsi_sampler_control control,
1870 float lod[TGSI_QUAD_SIZE])
1871 {
1872 const struct pipe_sampler_state *sampler = &sp_samp->base;
1873 float lod_bias = sampler->lod_bias;
1874 float min_lod = sampler->min_lod;
1875 float max_lod = sampler->max_lod;
1876 float lambda;
1877 uint i;
1878
1879 switch (control) {
1880 case tgsi_sampler_lod_none:
1881 /* XXX FIXME */
1882 case tgsi_sampler_derivs_explicit:
1883 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1884 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lambda, min_lod, max_lod);
1885 break;
1886 case tgsi_sampler_lod_bias:
1887 lambda = sp_sview->compute_lambda(sp_sview, s, t, p) + lod_bias;
1888 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1889 lod[i] = lambda + lod_in[i];
1890 lod[i] = CLAMP(lod[i], min_lod, max_lod);
1891 }
1892 break;
1893 case tgsi_sampler_lod_explicit:
1894 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
1895 lod[i] = CLAMP(lod_in[i], min_lod, max_lod);
1896 }
1897 break;
1898 case tgsi_sampler_lod_zero:
1899 case tgsi_sampler_gather:
1900 /* this is all static state in the sampler really need clamp here? */
1901 lod[0] = lod[1] = lod[2] = lod[3] = CLAMP(lod_bias, min_lod, max_lod);
1902 break;
1903 default:
1904 assert(0);
1905 lod[0] = lod[1] = lod[2] = lod[3] = 0.0f;
1906 }
1907 }
1908
1909 static inline unsigned
1910 get_gather_component(const float lod_in[TGSI_QUAD_SIZE])
1911 {
1912 /* gather component is stored in lod_in slot as unsigned */
1913 return (*(unsigned int *)lod_in) & 0x3;
1914 }
1915
1916 static void
1917 mip_filter_linear(struct sp_sampler_view *sp_sview,
1918 struct sp_sampler *sp_samp,
1919 img_filter_func min_filter,
1920 img_filter_func mag_filter,
1921 const float s[TGSI_QUAD_SIZE],
1922 const float t[TGSI_QUAD_SIZE],
1923 const float p[TGSI_QUAD_SIZE],
1924 const float c0[TGSI_QUAD_SIZE],
1925 const float lod_in[TGSI_QUAD_SIZE],
1926 const struct filter_args *filt_args,
1927 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1928 {
1929 const struct pipe_sampler_view *psview = &sp_sview->base;
1930 int j;
1931 float lod[TGSI_QUAD_SIZE];
1932 struct img_filter_args args;
1933
1934 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
1935
1936 args.offset = filt_args->offset;
1937 args.gather_only = filt_args->control == tgsi_sampler_gather;
1938 args.gather_comp = get_gather_component(lod_in);
1939
1940 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
1941 int level0 = psview->u.tex.first_level + (int)lod[j];
1942
1943 args.s = s[j];
1944 args.t = t[j];
1945 args.p = p[j];
1946 args.face_id = sp_sview->faces[j];
1947
1948 if (lod[j] < 0.0) {
1949 args.level = psview->u.tex.first_level;
1950 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1951 }
1952 else if (level0 >= (int) psview->u.tex.last_level) {
1953 args.level = psview->u.tex.last_level;
1954 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
1955 }
1956 else {
1957 float levelBlend = frac(lod[j]);
1958 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
1959 int c;
1960
1961 args.level = level0;
1962 min_filter(sp_sview, sp_samp, &args, &rgbax[0][0]);
1963 args.level = level0+1;
1964 min_filter(sp_sview, sp_samp, &args, &rgbax[0][1]);
1965
1966 for (c = 0; c < 4; c++) {
1967 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
1968 }
1969 }
1970 }
1971
1972 if (DEBUG_TEX) {
1973 print_sample_4(__FUNCTION__, rgba);
1974 }
1975 }
1976
1977
1978 /**
1979 * Compute nearest mipmap level from texcoords.
1980 * Then sample the texture level for four elements of a quad.
1981 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1982 */
1983 static void
1984 mip_filter_nearest(struct sp_sampler_view *sp_sview,
1985 struct sp_sampler *sp_samp,
1986 img_filter_func min_filter,
1987 img_filter_func mag_filter,
1988 const float s[TGSI_QUAD_SIZE],
1989 const float t[TGSI_QUAD_SIZE],
1990 const float p[TGSI_QUAD_SIZE],
1991 const float c0[TGSI_QUAD_SIZE],
1992 const float lod_in[TGSI_QUAD_SIZE],
1993 const struct filter_args *filt_args,
1994 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
1995 {
1996 const struct pipe_sampler_view *psview = &sp_sview->base;
1997 float lod[TGSI_QUAD_SIZE];
1998 int j;
1999 struct img_filter_args args;
2000
2001 args.offset = filt_args->offset;
2002 args.gather_only = filt_args->control == tgsi_sampler_gather;
2003 args.gather_comp = get_gather_component(lod_in);
2004
2005 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
2006
2007 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2008 args.s = s[j];
2009 args.t = t[j];
2010 args.p = p[j];
2011 args.face_id = sp_sview->faces[j];
2012
2013 if (lod[j] < 0.0) {
2014 args.level = psview->u.tex.first_level;
2015 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
2016 } else {
2017 int level = psview->u.tex.first_level + (int)(lod[j] + 0.5F);
2018 args.level = MIN2(level, (int)psview->u.tex.last_level);
2019 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
2020 }
2021 }
2022
2023 if (DEBUG_TEX) {
2024 print_sample_4(__FUNCTION__, rgba);
2025 }
2026 }
2027
2028
2029 static void
2030 mip_filter_none(struct sp_sampler_view *sp_sview,
2031 struct sp_sampler *sp_samp,
2032 img_filter_func min_filter,
2033 img_filter_func mag_filter,
2034 const float s[TGSI_QUAD_SIZE],
2035 const float t[TGSI_QUAD_SIZE],
2036 const float p[TGSI_QUAD_SIZE],
2037 const float c0[TGSI_QUAD_SIZE],
2038 const float lod_in[TGSI_QUAD_SIZE],
2039 const struct filter_args *filt_args,
2040 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2041 {
2042 float lod[TGSI_QUAD_SIZE];
2043 int j;
2044 struct img_filter_args args;
2045
2046 args.level = sp_sview->base.u.tex.first_level;
2047 args.offset = filt_args->offset;
2048 args.gather_only = filt_args->control == tgsi_sampler_gather;
2049
2050 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
2051
2052 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2053 args.s = s[j];
2054 args.t = t[j];
2055 args.p = p[j];
2056 args.face_id = sp_sview->faces[j];
2057 if (lod[j] < 0.0) {
2058 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
2059 }
2060 else {
2061 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
2062 }
2063 }
2064 }
2065
2066
2067 static void
2068 mip_filter_none_no_filter_select(struct sp_sampler_view *sp_sview,
2069 struct sp_sampler *sp_samp,
2070 img_filter_func min_filter,
2071 img_filter_func mag_filter,
2072 const float s[TGSI_QUAD_SIZE],
2073 const float t[TGSI_QUAD_SIZE],
2074 const float p[TGSI_QUAD_SIZE],
2075 const float c0[TGSI_QUAD_SIZE],
2076 const float lod_in[TGSI_QUAD_SIZE],
2077 const struct filter_args *filt_args,
2078 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2079 {
2080 int j;
2081 struct img_filter_args args;
2082 args.level = sp_sview->base.u.tex.first_level;
2083 args.offset = filt_args->offset;
2084 args.gather_only = filt_args->control == tgsi_sampler_gather;
2085 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2086 args.s = s[j];
2087 args.t = t[j];
2088 args.p = p[j];
2089 args.face_id = sp_sview->faces[j];
2090 mag_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
2091 }
2092 }
2093
2094
2095 /* For anisotropic filtering */
2096 #define WEIGHT_LUT_SIZE 1024
2097
2098 static float *weightLut = NULL;
2099
2100 /**
2101 * Creates the look-up table used to speed-up EWA sampling
2102 */
2103 static void
2104 create_filter_table(void)
2105 {
2106 unsigned i;
2107 if (!weightLut) {
2108 weightLut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
2109
2110 for (i = 0; i < WEIGHT_LUT_SIZE; ++i) {
2111 float alpha = 2;
2112 float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
2113 float weight = (float) exp(-alpha * r2);
2114 weightLut[i] = weight;
2115 }
2116 }
2117 }
2118
2119
2120 /**
2121 * Elliptical weighted average (EWA) filter for producing high quality
2122 * anisotropic filtered results.
2123 * Based on the Higher Quality Elliptical Weighted Average Filter
2124 * published by Paul S. Heckbert in his Master's Thesis
2125 * "Fundamentals of Texture Mapping and Image Warping" (1989)
2126 */
2127 static void
2128 img_filter_2d_ewa(struct sp_sampler_view *sp_sview,
2129 struct sp_sampler *sp_samp,
2130 img_filter_func min_filter,
2131 img_filter_func mag_filter,
2132 const float s[TGSI_QUAD_SIZE],
2133 const float t[TGSI_QUAD_SIZE],
2134 const float p[TGSI_QUAD_SIZE],
2135 unsigned level,
2136 const float dudx, const float dvdx,
2137 const float dudy, const float dvdy,
2138 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2139 {
2140 const struct pipe_resource *texture = sp_sview->base.texture;
2141
2142 // ??? Won't the image filters blow up if level is negative?
2143 unsigned level0 = level > 0 ? level : 0;
2144 float scaling = 1.0f / (1 << level0);
2145 int width = u_minify(texture->width0, level0);
2146 int height = u_minify(texture->height0, level0);
2147 struct img_filter_args args;
2148 float ux = dudx * scaling;
2149 float vx = dvdx * scaling;
2150 float uy = dudy * scaling;
2151 float vy = dvdy * scaling;
2152
2153 /* compute ellipse coefficients to bound the region:
2154 * A*x*x + B*x*y + C*y*y = F.
2155 */
2156 float A = vx*vx+vy*vy+1;
2157 float B = -2*(ux*vx+uy*vy);
2158 float C = ux*ux+uy*uy+1;
2159 float F = A*C-B*B/4.0f;
2160
2161 /* check if it is an ellipse */
2162 /* assert(F > 0.0); */
2163
2164 /* Compute the ellipse's (u,v) bounding box in texture space */
2165 float d = -B*B+4.0f*C*A;
2166 float box_u = 2.0f / d * sqrtf(d*C*F); /* box_u -> half of bbox with */
2167 float box_v = 2.0f / d * sqrtf(A*d*F); /* box_v -> half of bbox height */
2168
2169 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2170 float s_buffer[TGSI_QUAD_SIZE];
2171 float t_buffer[TGSI_QUAD_SIZE];
2172 float weight_buffer[TGSI_QUAD_SIZE];
2173 unsigned buffer_next;
2174 int j;
2175 float den; /* = 0.0F; */
2176 float ddq;
2177 float U; /* = u0 - tex_u; */
2178 int v;
2179
2180 /* Scale ellipse formula to directly index the Filter Lookup Table.
2181 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2182 */
2183 double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
2184 A *= formScale;
2185 B *= formScale;
2186 C *= formScale;
2187 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2188
2189 /* For each quad, the du and dx values are the same and so the ellipse is
2190 * also the same. Note that texel/image access can only be performed using
2191 * a quad, i.e. it is not possible to get the pixel value for a single
2192 * tex coord. In order to have a better performance, the access is buffered
2193 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2194 * full, then the pixel values are read from the image.
2195 */
2196 ddq = 2 * A;
2197
2198 args.level = level;
2199 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2200 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2201 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2202 * value, q, is less than F, we're inside the ellipse
2203 */
2204 float tex_u = -0.5F + s[j] * texture->width0 * scaling;
2205 float tex_v = -0.5F + t[j] * texture->height0 * scaling;
2206
2207 int u0 = (int) floorf(tex_u - box_u);
2208 int u1 = (int) ceilf(tex_u + box_u);
2209 int v0 = (int) floorf(tex_v - box_v);
2210 int v1 = (int) ceilf(tex_v + box_v);
2211
2212 float num[4] = {0.0F, 0.0F, 0.0F, 0.0F};
2213 buffer_next = 0;
2214 den = 0;
2215 args.face_id = sp_sview->faces[j];
2216
2217 U = u0 - tex_u;
2218 for (v = v0; v <= v1; ++v) {
2219 float V = v - tex_v;
2220 float dq = A * (2 * U + 1) + B * V;
2221 float q = (C * V + B * U) * V + A * U * U;
2222
2223 int u;
2224 for (u = u0; u <= u1; ++u) {
2225 /* Note that the ellipse has been pre-scaled so F =
2226 * WEIGHT_LUT_SIZE - 1
2227 */
2228 if (q < WEIGHT_LUT_SIZE) {
2229 /* as a LUT is used, q must never be negative;
2230 * should not happen, though
2231 */
2232 const int qClamped = q >= 0.0F ? q : 0;
2233 float weight = weightLut[qClamped];
2234
2235 weight_buffer[buffer_next] = weight;
2236 s_buffer[buffer_next] = u / ((float) width);
2237 t_buffer[buffer_next] = v / ((float) height);
2238
2239 buffer_next++;
2240 if (buffer_next == TGSI_QUAD_SIZE) {
2241 /* 4 texel coords are in the buffer -> read it now */
2242 unsigned jj;
2243 /* it is assumed that samp->min_img_filter is set to
2244 * img_filter_2d_nearest or one of the
2245 * accelerated img_filter_2d_nearest_XXX functions.
2246 */
2247 for (jj = 0; jj < buffer_next; jj++) {
2248 args.s = s_buffer[jj];
2249 args.t = t_buffer[jj];
2250 args.p = p[jj];
2251 min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][jj]);
2252 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2253 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2254 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2255 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2256 }
2257
2258 buffer_next = 0;
2259 }
2260
2261 den += weight;
2262 }
2263 q += dq;
2264 dq += ddq;
2265 }
2266 }
2267
2268 /* if the tex coord buffer contains unread values, we will read
2269 * them now.
2270 */
2271 if (buffer_next > 0) {
2272 unsigned jj;
2273 /* it is assumed that samp->min_img_filter is set to
2274 * img_filter_2d_nearest or one of the
2275 * accelerated img_filter_2d_nearest_XXX functions.
2276 */
2277 for (jj = 0; jj < buffer_next; jj++) {
2278 args.s = s_buffer[jj];
2279 args.t = t_buffer[jj];
2280 args.p = p[jj];
2281 min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][jj]);
2282 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
2283 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
2284 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
2285 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
2286 }
2287 }
2288
2289 if (den <= 0.0F) {
2290 /* Reaching this place would mean that no pixels intersected
2291 * the ellipse. This should never happen because the filter
2292 * we use always intersects at least one pixel.
2293 */
2294
2295 /*rgba[0]=0;
2296 rgba[1]=0;
2297 rgba[2]=0;
2298 rgba[3]=0;*/
2299 /* not enough pixels in resampling, resort to direct interpolation */
2300 args.s = s[j];
2301 args.t = t[j];
2302 args.p = p[j];
2303 min_filter(sp_sview, sp_samp, &args, &rgba_temp[0][j]);
2304 den = 1;
2305 num[0] = rgba_temp[0][j];
2306 num[1] = rgba_temp[1][j];
2307 num[2] = rgba_temp[2][j];
2308 num[3] = rgba_temp[3][j];
2309 }
2310
2311 rgba[0][j] = num[0] / den;
2312 rgba[1][j] = num[1] / den;
2313 rgba[2][j] = num[2] / den;
2314 rgba[3][j] = num[3] / den;
2315 }
2316 }
2317
2318
2319 /**
2320 * Sample 2D texture using an anisotropic filter.
2321 */
2322 static void
2323 mip_filter_linear_aniso(struct sp_sampler_view *sp_sview,
2324 struct sp_sampler *sp_samp,
2325 img_filter_func min_filter,
2326 img_filter_func mag_filter,
2327 const float s[TGSI_QUAD_SIZE],
2328 const float t[TGSI_QUAD_SIZE],
2329 const float p[TGSI_QUAD_SIZE],
2330 const float c0[TGSI_QUAD_SIZE],
2331 const float lod_in[TGSI_QUAD_SIZE],
2332 const struct filter_args *filt_args,
2333 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2334 {
2335 const struct pipe_resource *texture = sp_sview->base.texture;
2336 const struct pipe_sampler_view *psview = &sp_sview->base;
2337 int level0;
2338 float lambda;
2339 float lod[TGSI_QUAD_SIZE];
2340
2341 float s_to_u = u_minify(texture->width0, psview->u.tex.first_level);
2342 float t_to_v = u_minify(texture->height0, psview->u.tex.first_level);
2343 float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2344 float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
2345 float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2346 float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
2347 struct img_filter_args args;
2348
2349 if (filt_args->control == tgsi_sampler_lod_bias ||
2350 filt_args->control == tgsi_sampler_lod_none ||
2351 /* XXX FIXME */
2352 filt_args->control == tgsi_sampler_derivs_explicit) {
2353 /* note: instead of working with Px and Py, we will use the
2354 * squared length instead, to avoid sqrt.
2355 */
2356 float Px2 = dudx * dudx + dvdx * dvdx;
2357 float Py2 = dudy * dudy + dvdy * dvdy;
2358
2359 float Pmax2;
2360 float Pmin2;
2361 float e;
2362 const float maxEccentricity = sp_samp->base.max_anisotropy * sp_samp->base.max_anisotropy;
2363
2364 if (Px2 < Py2) {
2365 Pmax2 = Py2;
2366 Pmin2 = Px2;
2367 }
2368 else {
2369 Pmax2 = Px2;
2370 Pmin2 = Py2;
2371 }
2372
2373 /* if the eccentricity of the ellipse is too big, scale up the shorter
2374 * of the two vectors to limit the maximum amount of work per pixel
2375 */
2376 e = Pmax2 / Pmin2;
2377 if (e > maxEccentricity) {
2378 /* float s=e / maxEccentricity;
2379 minor[0] *= s;
2380 minor[1] *= s;
2381 Pmin2 *= s; */
2382 Pmin2 = Pmax2 / maxEccentricity;
2383 }
2384
2385 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2386 * this since 0.5*log(x) = log(sqrt(x))
2387 */
2388 lambda = 0.5F * util_fast_log2(Pmin2) + sp_samp->base.lod_bias;
2389 compute_lod(&sp_samp->base, filt_args->control, lambda, lod_in, lod);
2390 }
2391 else {
2392 assert(filt_args->control == tgsi_sampler_lod_explicit ||
2393 filt_args->control == tgsi_sampler_lod_zero);
2394 compute_lod(&sp_samp->base, filt_args->control, sp_samp->base.lod_bias, lod_in, lod);
2395 }
2396
2397 /* XXX: Take into account all lod values.
2398 */
2399 lambda = lod[0];
2400 level0 = psview->u.tex.first_level + (int)lambda;
2401
2402 /* If the ellipse covers the whole image, we can
2403 * simply return the average of the whole image.
2404 */
2405 if (level0 >= (int) psview->u.tex.last_level) {
2406 int j;
2407 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2408 args.s = s[j];
2409 args.t = t[j];
2410 args.p = p[j];
2411 args.level = psview->u.tex.last_level;
2412 args.face_id = sp_sview->faces[j];
2413 min_filter(sp_sview, sp_samp, &args, &rgba[0][j]);
2414 }
2415 }
2416 else {
2417 /* don't bother interpolating between multiple LODs; it doesn't
2418 * seem to be worth the extra running time.
2419 */
2420 img_filter_2d_ewa(sp_sview, sp_samp, min_filter, mag_filter,
2421 s, t, p, level0,
2422 dudx, dvdx, dudy, dvdy, rgba);
2423 }
2424
2425 if (DEBUG_TEX) {
2426 print_sample_4(__FUNCTION__, rgba);
2427 }
2428 }
2429
2430
2431 /**
2432 * Specialized version of mip_filter_linear with hard-wired calls to
2433 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2434 */
2435 static void
2436 mip_filter_linear_2d_linear_repeat_POT(
2437 struct sp_sampler_view *sp_sview,
2438 struct sp_sampler *sp_samp,
2439 img_filter_func min_filter,
2440 img_filter_func mag_filter,
2441 const float s[TGSI_QUAD_SIZE],
2442 const float t[TGSI_QUAD_SIZE],
2443 const float p[TGSI_QUAD_SIZE],
2444 const float c0[TGSI_QUAD_SIZE],
2445 const float lod_in[TGSI_QUAD_SIZE],
2446 const struct filter_args *filt_args,
2447 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2448 {
2449 const struct pipe_sampler_view *psview = &sp_sview->base;
2450 int j;
2451 float lod[TGSI_QUAD_SIZE];
2452
2453 compute_lambda_lod(sp_sview, sp_samp, s, t, p, lod_in, filt_args->control, lod);
2454
2455 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2456 int level0 = psview->u.tex.first_level + (int)lod[j];
2457 struct img_filter_args args;
2458 /* Catches both negative and large values of level0:
2459 */
2460 args.s = s[j];
2461 args.t = t[j];
2462 args.p = p[j];
2463 args.face_id = sp_sview->faces[j];
2464 args.offset = filt_args->offset;
2465 args.gather_only = filt_args->control == tgsi_sampler_gather;
2466 if ((unsigned)level0 >= psview->u.tex.last_level) {
2467 if (level0 < 0)
2468 args.level = psview->u.tex.first_level;
2469 else
2470 args.level = psview->u.tex.last_level;
2471 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args,
2472 &rgba[0][j]);
2473
2474 }
2475 else {
2476 float levelBlend = frac(lod[j]);
2477 float rgbax[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2478 int c;
2479
2480 args.level = level0;
2481 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args, &rgbax[0][0]);
2482 args.level = level0+1;
2483 img_filter_2d_linear_repeat_POT(sp_sview, sp_samp, &args, &rgbax[0][1]);
2484
2485 for (c = 0; c < TGSI_NUM_CHANNELS; c++)
2486 rgba[c][j] = lerp(levelBlend, rgbax[c][0], rgbax[c][1]);
2487 }
2488 }
2489
2490 if (DEBUG_TEX) {
2491 print_sample_4(__FUNCTION__, rgba);
2492 }
2493 }
2494
2495
2496 /**
2497 * Do shadow/depth comparisons.
2498 */
2499 static void
2500 sample_compare(struct sp_sampler_view *sp_sview,
2501 struct sp_sampler *sp_samp,
2502 const float s[TGSI_QUAD_SIZE],
2503 const float t[TGSI_QUAD_SIZE],
2504 const float p[TGSI_QUAD_SIZE],
2505 const float c0[TGSI_QUAD_SIZE],
2506 const float c1[TGSI_QUAD_SIZE],
2507 enum tgsi_sampler_control control,
2508 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2509 {
2510 const struct pipe_sampler_state *sampler = &sp_samp->base;
2511 int j, v;
2512 int k[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2513 float pc[4];
2514 const struct util_format_description *format_desc;
2515 unsigned chan_type;
2516 bool is_gather = (control == tgsi_sampler_gather);
2517
2518 /**
2519 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2520 * for 2D Array texture we need to use the 'c0' (aka Q).
2521 * When we sampled the depth texture, the depth value was put into all
2522 * RGBA channels. We look at the red channel here.
2523 */
2524
2525 if (sp_sview->base.target == PIPE_TEXTURE_2D_ARRAY ||
2526 sp_sview->base.target == PIPE_TEXTURE_CUBE) {
2527 pc[0] = c0[0];
2528 pc[1] = c0[1];
2529 pc[2] = c0[2];
2530 pc[3] = c0[3];
2531 } else if (sp_sview->base.target == PIPE_TEXTURE_CUBE_ARRAY) {
2532 pc[0] = c1[0];
2533 pc[1] = c1[1];
2534 pc[2] = c1[2];
2535 pc[3] = c1[3];
2536 } else {
2537 pc[0] = p[0];
2538 pc[1] = p[1];
2539 pc[2] = p[2];
2540 pc[3] = p[3];
2541 }
2542
2543 format_desc = util_format_description(sp_sview->base.format);
2544 /* not entirely sure we couldn't end up with non-valid swizzle here */
2545 chan_type = format_desc->swizzle[0] <= UTIL_FORMAT_SWIZZLE_W ?
2546 format_desc->channel[format_desc->swizzle[0]].type :
2547 UTIL_FORMAT_TYPE_FLOAT;
2548 if (chan_type != UTIL_FORMAT_TYPE_FLOAT) {
2549 /*
2550 * clamping is a result of conversion to texture format, hence
2551 * doesn't happen with floats. Technically also should do comparison
2552 * in texture format (quantization!).
2553 */
2554 pc[0] = CLAMP(pc[0], 0.0F, 1.0F);
2555 pc[1] = CLAMP(pc[1], 0.0F, 1.0F);
2556 pc[2] = CLAMP(pc[2], 0.0F, 1.0F);
2557 pc[3] = CLAMP(pc[3], 0.0F, 1.0F);
2558 }
2559
2560 for (v = 0; v < (is_gather ? TGSI_NUM_CHANNELS : 1); v++) {
2561 /* compare four texcoords vs. four texture samples */
2562 switch (sampler->compare_func) {
2563 case PIPE_FUNC_LESS:
2564 k[v][0] = pc[0] < rgba[v][0];
2565 k[v][1] = pc[1] < rgba[v][1];
2566 k[v][2] = pc[2] < rgba[v][2];
2567 k[v][3] = pc[3] < rgba[v][3];
2568 break;
2569 case PIPE_FUNC_LEQUAL:
2570 k[v][0] = pc[0] <= rgba[v][0];
2571 k[v][1] = pc[1] <= rgba[v][1];
2572 k[v][2] = pc[2] <= rgba[v][2];
2573 k[v][3] = pc[3] <= rgba[v][3];
2574 break;
2575 case PIPE_FUNC_GREATER:
2576 k[v][0] = pc[0] > rgba[v][0];
2577 k[v][1] = pc[1] > rgba[v][1];
2578 k[v][2] = pc[2] > rgba[v][2];
2579 k[v][3] = pc[3] > rgba[v][3];
2580 break;
2581 case PIPE_FUNC_GEQUAL:
2582 k[v][0] = pc[0] >= rgba[v][0];
2583 k[v][1] = pc[1] >= rgba[v][1];
2584 k[v][2] = pc[2] >= rgba[v][2];
2585 k[v][3] = pc[3] >= rgba[v][3];
2586 break;
2587 case PIPE_FUNC_EQUAL:
2588 k[v][0] = pc[0] == rgba[v][0];
2589 k[v][1] = pc[1] == rgba[v][1];
2590 k[v][2] = pc[2] == rgba[v][2];
2591 k[v][3] = pc[3] == rgba[v][3];
2592 break;
2593 case PIPE_FUNC_NOTEQUAL:
2594 k[v][0] = pc[0] != rgba[v][0];
2595 k[v][1] = pc[1] != rgba[v][1];
2596 k[v][2] = pc[2] != rgba[v][2];
2597 k[v][3] = pc[3] != rgba[v][3];
2598 break;
2599 case PIPE_FUNC_ALWAYS:
2600 k[v][0] = k[v][1] = k[v][2] = k[v][3] = 1;
2601 break;
2602 case PIPE_FUNC_NEVER:
2603 k[v][0] = k[v][1] = k[v][2] = k[v][3] = 0;
2604 break;
2605 default:
2606 k[v][0] = k[v][1] = k[v][2] = k[v][3] = 0;
2607 assert(0);
2608 break;
2609 }
2610 }
2611
2612 if (is_gather) {
2613 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2614 for (v = 0; v < TGSI_NUM_CHANNELS; v++) {
2615 rgba[v][j] = k[v][j];
2616 }
2617 }
2618 } else {
2619 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2620 rgba[0][j] = k[0][j];
2621 rgba[1][j] = k[0][j];
2622 rgba[2][j] = k[0][j];
2623 rgba[3][j] = 1.0F;
2624 }
2625 }
2626 }
2627
2628 static void
2629 do_swizzling(const struct pipe_sampler_view *sview,
2630 float in[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
2631 float out[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2632 {
2633 int j;
2634 const unsigned swizzle_r = sview->swizzle_r;
2635 const unsigned swizzle_g = sview->swizzle_g;
2636 const unsigned swizzle_b = sview->swizzle_b;
2637 const unsigned swizzle_a = sview->swizzle_a;
2638
2639 switch (swizzle_r) {
2640 case PIPE_SWIZZLE_ZERO:
2641 for (j = 0; j < 4; j++)
2642 out[0][j] = 0.0f;
2643 break;
2644 case PIPE_SWIZZLE_ONE:
2645 for (j = 0; j < 4; j++)
2646 out[0][j] = 1.0f;
2647 break;
2648 default:
2649 assert(swizzle_r < 4);
2650 for (j = 0; j < 4; j++)
2651 out[0][j] = in[swizzle_r][j];
2652 }
2653
2654 switch (swizzle_g) {
2655 case PIPE_SWIZZLE_ZERO:
2656 for (j = 0; j < 4; j++)
2657 out[1][j] = 0.0f;
2658 break;
2659 case PIPE_SWIZZLE_ONE:
2660 for (j = 0; j < 4; j++)
2661 out[1][j] = 1.0f;
2662 break;
2663 default:
2664 assert(swizzle_g < 4);
2665 for (j = 0; j < 4; j++)
2666 out[1][j] = in[swizzle_g][j];
2667 }
2668
2669 switch (swizzle_b) {
2670 case PIPE_SWIZZLE_ZERO:
2671 for (j = 0; j < 4; j++)
2672 out[2][j] = 0.0f;
2673 break;
2674 case PIPE_SWIZZLE_ONE:
2675 for (j = 0; j < 4; j++)
2676 out[2][j] = 1.0f;
2677 break;
2678 default:
2679 assert(swizzle_b < 4);
2680 for (j = 0; j < 4; j++)
2681 out[2][j] = in[swizzle_b][j];
2682 }
2683
2684 switch (swizzle_a) {
2685 case PIPE_SWIZZLE_ZERO:
2686 for (j = 0; j < 4; j++)
2687 out[3][j] = 0.0f;
2688 break;
2689 case PIPE_SWIZZLE_ONE:
2690 for (j = 0; j < 4; j++)
2691 out[3][j] = 1.0f;
2692 break;
2693 default:
2694 assert(swizzle_a < 4);
2695 for (j = 0; j < 4; j++)
2696 out[3][j] = in[swizzle_a][j];
2697 }
2698 }
2699
2700
2701 static wrap_nearest_func
2702 get_nearest_unorm_wrap(unsigned mode)
2703 {
2704 switch (mode) {
2705 case PIPE_TEX_WRAP_CLAMP:
2706 return wrap_nearest_unorm_clamp;
2707 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2708 return wrap_nearest_unorm_clamp_to_edge;
2709 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2710 return wrap_nearest_unorm_clamp_to_border;
2711 default:
2712 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
2713 return wrap_nearest_unorm_clamp;
2714 }
2715 }
2716
2717
2718 static wrap_nearest_func
2719 get_nearest_wrap(unsigned mode)
2720 {
2721 switch (mode) {
2722 case PIPE_TEX_WRAP_REPEAT:
2723 return wrap_nearest_repeat;
2724 case PIPE_TEX_WRAP_CLAMP:
2725 return wrap_nearest_clamp;
2726 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2727 return wrap_nearest_clamp_to_edge;
2728 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2729 return wrap_nearest_clamp_to_border;
2730 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2731 return wrap_nearest_mirror_repeat;
2732 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2733 return wrap_nearest_mirror_clamp;
2734 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2735 return wrap_nearest_mirror_clamp_to_edge;
2736 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2737 return wrap_nearest_mirror_clamp_to_border;
2738 default:
2739 assert(0);
2740 return wrap_nearest_repeat;
2741 }
2742 }
2743
2744
2745 static wrap_linear_func
2746 get_linear_unorm_wrap(unsigned mode)
2747 {
2748 switch (mode) {
2749 case PIPE_TEX_WRAP_CLAMP:
2750 return wrap_linear_unorm_clamp;
2751 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2752 return wrap_linear_unorm_clamp_to_edge;
2753 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2754 return wrap_linear_unorm_clamp_to_border;
2755 default:
2756 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode);
2757 return wrap_linear_unorm_clamp;
2758 }
2759 }
2760
2761
2762 static wrap_linear_func
2763 get_linear_wrap(unsigned mode)
2764 {
2765 switch (mode) {
2766 case PIPE_TEX_WRAP_REPEAT:
2767 return wrap_linear_repeat;
2768 case PIPE_TEX_WRAP_CLAMP:
2769 return wrap_linear_clamp;
2770 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2771 return wrap_linear_clamp_to_edge;
2772 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2773 return wrap_linear_clamp_to_border;
2774 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2775 return wrap_linear_mirror_repeat;
2776 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2777 return wrap_linear_mirror_clamp;
2778 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2779 return wrap_linear_mirror_clamp_to_edge;
2780 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2781 return wrap_linear_mirror_clamp_to_border;
2782 default:
2783 assert(0);
2784 return wrap_linear_repeat;
2785 }
2786 }
2787
2788
2789 /**
2790 * Is swizzling needed for the given state key?
2791 */
2792 static inline bool
2793 any_swizzle(const struct pipe_sampler_view *view)
2794 {
2795 return (view->swizzle_r != PIPE_SWIZZLE_RED ||
2796 view->swizzle_g != PIPE_SWIZZLE_GREEN ||
2797 view->swizzle_b != PIPE_SWIZZLE_BLUE ||
2798 view->swizzle_a != PIPE_SWIZZLE_ALPHA);
2799 }
2800
2801
2802 static img_filter_func
2803 get_img_filter(const struct sp_sampler_view *sp_sview,
2804 const struct pipe_sampler_state *sampler,
2805 unsigned filter, bool gather)
2806 {
2807 switch (sp_sview->base.target) {
2808 case PIPE_BUFFER:
2809 case PIPE_TEXTURE_1D:
2810 if (filter == PIPE_TEX_FILTER_NEAREST)
2811 return img_filter_1d_nearest;
2812 else
2813 return img_filter_1d_linear;
2814 break;
2815 case PIPE_TEXTURE_1D_ARRAY:
2816 if (filter == PIPE_TEX_FILTER_NEAREST)
2817 return img_filter_1d_array_nearest;
2818 else
2819 return img_filter_1d_array_linear;
2820 break;
2821 case PIPE_TEXTURE_2D:
2822 case PIPE_TEXTURE_RECT:
2823 /* Try for fast path:
2824 */
2825 if (!gather && sp_sview->pot2d &&
2826 sampler->wrap_s == sampler->wrap_t &&
2827 sampler->normalized_coords)
2828 {
2829 switch (sampler->wrap_s) {
2830 case PIPE_TEX_WRAP_REPEAT:
2831 switch (filter) {
2832 case PIPE_TEX_FILTER_NEAREST:
2833 return img_filter_2d_nearest_repeat_POT;
2834 case PIPE_TEX_FILTER_LINEAR:
2835 return img_filter_2d_linear_repeat_POT;
2836 default:
2837 break;
2838 }
2839 break;
2840 case PIPE_TEX_WRAP_CLAMP:
2841 switch (filter) {
2842 case PIPE_TEX_FILTER_NEAREST:
2843 return img_filter_2d_nearest_clamp_POT;
2844 default:
2845 break;
2846 }
2847 }
2848 }
2849 /* Otherwise use default versions:
2850 */
2851 if (filter == PIPE_TEX_FILTER_NEAREST)
2852 return img_filter_2d_nearest;
2853 else
2854 return img_filter_2d_linear;
2855 break;
2856 case PIPE_TEXTURE_2D_ARRAY:
2857 if (filter == PIPE_TEX_FILTER_NEAREST)
2858 return img_filter_2d_array_nearest;
2859 else
2860 return img_filter_2d_array_linear;
2861 break;
2862 case PIPE_TEXTURE_CUBE:
2863 if (filter == PIPE_TEX_FILTER_NEAREST)
2864 return img_filter_cube_nearest;
2865 else
2866 return img_filter_cube_linear;
2867 break;
2868 case PIPE_TEXTURE_CUBE_ARRAY:
2869 if (filter == PIPE_TEX_FILTER_NEAREST)
2870 return img_filter_cube_array_nearest;
2871 else
2872 return img_filter_cube_array_linear;
2873 break;
2874 case PIPE_TEXTURE_3D:
2875 if (filter == PIPE_TEX_FILTER_NEAREST)
2876 return img_filter_3d_nearest;
2877 else
2878 return img_filter_3d_linear;
2879 break;
2880 default:
2881 assert(0);
2882 return img_filter_1d_nearest;
2883 }
2884 }
2885
2886
2887 static void
2888 sample_mip(struct sp_sampler_view *sp_sview,
2889 struct sp_sampler *sp_samp,
2890 const float s[TGSI_QUAD_SIZE],
2891 const float t[TGSI_QUAD_SIZE],
2892 const float p[TGSI_QUAD_SIZE],
2893 const float c0[TGSI_QUAD_SIZE],
2894 const float lod[TGSI_QUAD_SIZE],
2895 const struct filter_args *filt_args,
2896 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2897 {
2898 mip_filter_func mip_filter;
2899 img_filter_func min_img_filter = NULL;
2900 img_filter_func mag_img_filter = NULL;
2901
2902 if (filt_args->control == tgsi_sampler_gather) {
2903 mip_filter = mip_filter_nearest;
2904 min_img_filter = get_img_filter(sp_sview, &sp_samp->base, PIPE_TEX_FILTER_LINEAR, true);
2905 } else if (sp_sview->pot2d & sp_samp->min_mag_equal_repeat_linear) {
2906 mip_filter = mip_filter_linear_2d_linear_repeat_POT;
2907 }
2908 else {
2909 mip_filter = sp_samp->mip_filter;
2910 min_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->min_img_filter, false);
2911 if (sp_samp->min_mag_equal) {
2912 mag_img_filter = min_img_filter;
2913 }
2914 else {
2915 mag_img_filter = get_img_filter(sp_sview, &sp_samp->base, sp_samp->base.mag_img_filter, false);
2916 }
2917 }
2918
2919 mip_filter(sp_sview, sp_samp, min_img_filter, mag_img_filter,
2920 s, t, p, c0, lod, filt_args, rgba);
2921
2922 if (sp_samp->base.compare_mode != PIPE_TEX_COMPARE_NONE) {
2923 sample_compare(sp_sview, sp_samp, s, t, p, c0, lod, filt_args->control, rgba);
2924 }
2925
2926 if (sp_sview->need_swizzle && filt_args->control != tgsi_sampler_gather) {
2927 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
2928 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
2929 do_swizzling(&sp_sview->base, rgba_temp, rgba);
2930 }
2931
2932 }
2933
2934
2935 /**
2936 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2937 * Put face info into the sampler faces[] array.
2938 */
2939 static void
2940 sample_cube(struct sp_sampler_view *sp_sview,
2941 struct sp_sampler *sp_samp,
2942 const float s[TGSI_QUAD_SIZE],
2943 const float t[TGSI_QUAD_SIZE],
2944 const float p[TGSI_QUAD_SIZE],
2945 const float c0[TGSI_QUAD_SIZE],
2946 const float c1[TGSI_QUAD_SIZE],
2947 const struct filter_args *filt_args,
2948 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
2949 {
2950 unsigned j;
2951 float ssss[4], tttt[4];
2952
2953 /* Not actually used, but the intermediate steps that do the
2954 * dereferencing don't know it.
2955 */
2956 static float pppp[4] = { 0, 0, 0, 0 };
2957
2958 pppp[0] = c0[0];
2959 pppp[1] = c0[1];
2960 pppp[2] = c0[2];
2961 pppp[3] = c0[3];
2962 /*
2963 major axis
2964 direction target sc tc ma
2965 ---------- ------------------------------- --- --- ---
2966 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2967 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2968 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2969 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2970 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2971 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2972 */
2973
2974 /* Choose the cube face and compute new s/t coords for the 2D face.
2975 *
2976 * Use the same cube face for all four pixels in the quad.
2977 *
2978 * This isn't ideal, but if we want to use a different cube face
2979 * per pixel in the quad, we'd have to also compute the per-face
2980 * LOD here too. That's because the four post-face-selection
2981 * texcoords are no longer related to each other (they're
2982 * per-face!) so we can't use subtraction to compute the partial
2983 * deriviates to compute the LOD. Doing so (near cube edges
2984 * anyway) gives us pretty much random values.
2985 */
2986 {
2987 /* use the average of the four pixel's texcoords to choose the face */
2988 const float rx = 0.25F * (s[0] + s[1] + s[2] + s[3]);
2989 const float ry = 0.25F * (t[0] + t[1] + t[2] + t[3]);
2990 const float rz = 0.25F * (p[0] + p[1] + p[2] + p[3]);
2991 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
2992
2993 if (arx >= ary && arx >= arz) {
2994 float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
2995 uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
2996 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
2997 const float ima = -0.5F / fabsf(s[j]);
2998 ssss[j] = sign * p[j] * ima + 0.5F;
2999 tttt[j] = t[j] * ima + 0.5F;
3000 sp_sview->faces[j] = face;
3001 }
3002 }
3003 else if (ary >= arx && ary >= arz) {
3004 float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
3005 uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
3006 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3007 const float ima = -0.5F / fabsf(t[j]);
3008 ssss[j] = -s[j] * ima + 0.5F;
3009 tttt[j] = sign * -p[j] * ima + 0.5F;
3010 sp_sview->faces[j] = face;
3011 }
3012 }
3013 else {
3014 float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
3015 uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
3016 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3017 const float ima = -0.5F / fabsf(p[j]);
3018 ssss[j] = sign * -s[j] * ima + 0.5F;
3019 tttt[j] = t[j] * ima + 0.5F;
3020 sp_sview->faces[j] = face;
3021 }
3022 }
3023 }
3024
3025 sample_mip(sp_sview, sp_samp, ssss, tttt, pppp, c0, c1, filt_args, rgba);
3026 }
3027
3028
3029 static void
3030 sp_get_dims(struct sp_sampler_view *sp_sview, int level,
3031 int dims[4])
3032 {
3033 const struct pipe_sampler_view *view = &sp_sview->base;
3034 const struct pipe_resource *texture = view->texture;
3035
3036 if (view->target == PIPE_BUFFER) {
3037 dims[0] = (view->u.buf.last_element - view->u.buf.first_element) + 1;
3038 /* the other values are undefined, but let's avoid potential valgrind
3039 * warnings.
3040 */
3041 dims[1] = dims[2] = dims[3] = 0;
3042 return;
3043 }
3044
3045 /* undefined according to EXT_gpu_program */
3046 level += view->u.tex.first_level;
3047 if (level > view->u.tex.last_level)
3048 return;
3049
3050 dims[3] = view->u.tex.last_level - view->u.tex.first_level + 1;
3051 dims[0] = u_minify(texture->width0, level);
3052
3053 switch (view->target) {
3054 case PIPE_TEXTURE_1D_ARRAY:
3055 dims[1] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
3056 /* fallthrough */
3057 case PIPE_TEXTURE_1D:
3058 return;
3059 case PIPE_TEXTURE_2D_ARRAY:
3060 dims[2] = view->u.tex.last_layer - view->u.tex.first_layer + 1;
3061 /* fallthrough */
3062 case PIPE_TEXTURE_2D:
3063 case PIPE_TEXTURE_CUBE:
3064 case PIPE_TEXTURE_RECT:
3065 dims[1] = u_minify(texture->height0, level);
3066 return;
3067 case PIPE_TEXTURE_3D:
3068 dims[1] = u_minify(texture->height0, level);
3069 dims[2] = u_minify(texture->depth0, level);
3070 return;
3071 case PIPE_TEXTURE_CUBE_ARRAY:
3072 dims[1] = u_minify(texture->height0, level);
3073 dims[2] = (view->u.tex.last_layer - view->u.tex.first_layer + 1) / 6;
3074 break;
3075 default:
3076 assert(!"unexpected texture target in sp_get_dims()");
3077 return;
3078 }
3079 }
3080
3081 /**
3082 * This function is only used for getting unfiltered texels via the
3083 * TXF opcode. The GL spec says that out-of-bounds texel fetches
3084 * produce undefined results. Instead of crashing, lets just clamp
3085 * coords to the texture image size.
3086 */
3087 static void
3088 sp_get_texels(struct sp_sampler_view *sp_sview,
3089 const int v_i[TGSI_QUAD_SIZE],
3090 const int v_j[TGSI_QUAD_SIZE],
3091 const int v_k[TGSI_QUAD_SIZE],
3092 const int lod[TGSI_QUAD_SIZE],
3093 const int8_t offset[3],
3094 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
3095 {
3096 union tex_tile_address addr;
3097 const struct pipe_resource *texture = sp_sview->base.texture;
3098 int j, c;
3099 const float *tx;
3100 int width, height, depth;
3101
3102 addr.value = 0;
3103 /* TODO write a better test for LOD */
3104 addr.bits.level = sp_sview->base.target == PIPE_BUFFER ? 0 :
3105 CLAMP(lod[0] + sp_sview->base.u.tex.first_level,
3106 sp_sview->base.u.tex.first_level,
3107 sp_sview->base.u.tex.last_level);
3108
3109 width = u_minify(texture->width0, addr.bits.level);
3110 height = u_minify(texture->height0, addr.bits.level);
3111 depth = u_minify(texture->depth0, addr.bits.level);
3112
3113 switch (sp_sview->base.target) {
3114 case PIPE_BUFFER:
3115 case PIPE_TEXTURE_1D:
3116 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3117 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3118 tx = get_texel_2d_no_border(sp_sview, addr, x, 0);
3119 for (c = 0; c < 4; c++) {
3120 rgba[c][j] = tx[c];
3121 }
3122 }
3123 break;
3124 case PIPE_TEXTURE_1D_ARRAY:
3125 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3126 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3127 int y = CLAMP(v_j[j], sp_sview->base.u.tex.first_layer,
3128 sp_sview->base.u.tex.last_layer);
3129 tx = get_texel_2d_no_border(sp_sview, addr, x, y);
3130 for (c = 0; c < 4; c++) {
3131 rgba[c][j] = tx[c];
3132 }
3133 }
3134 break;
3135 case PIPE_TEXTURE_2D:
3136 case PIPE_TEXTURE_RECT:
3137 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3138 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3139 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3140 tx = get_texel_2d_no_border(sp_sview, addr, x, y);
3141 for (c = 0; c < 4; c++) {
3142 rgba[c][j] = tx[c];
3143 }
3144 }
3145 break;
3146 case PIPE_TEXTURE_2D_ARRAY:
3147 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3148 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3149 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3150 int layer = CLAMP(v_k[j], sp_sview->base.u.tex.first_layer,
3151 sp_sview->base.u.tex.last_layer);
3152 tx = get_texel_3d_no_border(sp_sview, addr, x, y, layer);
3153 for (c = 0; c < 4; c++) {
3154 rgba[c][j] = tx[c];
3155 }
3156 }
3157 break;
3158 case PIPE_TEXTURE_3D:
3159 for (j = 0; j < TGSI_QUAD_SIZE; j++) {
3160 int x = CLAMP(v_i[j] + offset[0], 0, width - 1);
3161 int y = CLAMP(v_j[j] + offset[1], 0, height - 1);
3162 int z = CLAMP(v_k[j] + offset[2], 0, depth - 1);
3163 tx = get_texel_3d_no_border(sp_sview, addr, x, y, z);
3164 for (c = 0; c < 4; c++) {
3165 rgba[c][j] = tx[c];
3166 }
3167 }
3168 break;
3169 case PIPE_TEXTURE_CUBE: /* TXF can't work on CUBE according to spec */
3170 default:
3171 assert(!"Unknown or CUBE texture type in TXF processing\n");
3172 break;
3173 }
3174
3175 if (sp_sview->need_swizzle) {
3176 float rgba_temp[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
3177 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
3178 do_swizzling(&sp_sview->base, rgba_temp, rgba);
3179 }
3180 }
3181
3182
3183 void *
3184 softpipe_create_sampler_state(struct pipe_context *pipe,
3185 const struct pipe_sampler_state *sampler)
3186 {
3187 struct sp_sampler *samp = CALLOC_STRUCT(sp_sampler);
3188
3189 samp->base = *sampler;
3190
3191 /* Note that (for instance) linear_texcoord_s and
3192 * nearest_texcoord_s may be active at the same time, if the
3193 * sampler min_img_filter differs from its mag_img_filter.
3194 */
3195 if (sampler->normalized_coords) {
3196 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
3197 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
3198 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
3199
3200 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
3201 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
3202 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
3203 }
3204 else {
3205 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
3206 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
3207 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
3208
3209 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
3210 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
3211 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
3212 }
3213
3214 samp->min_img_filter = sampler->min_img_filter;
3215
3216 switch (sampler->min_mip_filter) {
3217 case PIPE_TEX_MIPFILTER_NONE:
3218 if (sampler->min_img_filter == sampler->mag_img_filter)
3219 samp->mip_filter = mip_filter_none_no_filter_select;
3220 else
3221 samp->mip_filter = mip_filter_none;
3222 break;
3223
3224 case PIPE_TEX_MIPFILTER_NEAREST:
3225 samp->mip_filter = mip_filter_nearest;
3226 break;
3227
3228 case PIPE_TEX_MIPFILTER_LINEAR:
3229 if (sampler->min_img_filter == sampler->mag_img_filter &&
3230 sampler->normalized_coords &&
3231 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
3232 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
3233 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR &&
3234 sampler->max_anisotropy <= 1) {
3235 samp->min_mag_equal_repeat_linear = TRUE;
3236 }
3237 samp->mip_filter = mip_filter_linear;
3238
3239 /* Anisotropic filtering extension. */
3240 if (sampler->max_anisotropy > 1) {
3241 samp->mip_filter = mip_filter_linear_aniso;
3242
3243 /* Override min_img_filter:
3244 * min_img_filter needs to be set to NEAREST since we need to access
3245 * each texture pixel as it is and weight it later; using linear
3246 * filters will have incorrect results.
3247 * By setting the filter to NEAREST here, we can avoid calling the
3248 * generic img_filter_2d_nearest in the anisotropic filter function,
3249 * making it possible to use one of the accelerated implementations
3250 */
3251 samp->min_img_filter = PIPE_TEX_FILTER_NEAREST;
3252
3253 /* on first access create the lookup table containing the filter weights. */
3254 if (!weightLut) {
3255 create_filter_table();
3256 }
3257 }
3258 break;
3259 }
3260 if (samp->min_img_filter == sampler->mag_img_filter) {
3261 samp->min_mag_equal = TRUE;
3262 }
3263
3264 return (void *)samp;
3265 }
3266
3267
3268 compute_lambda_func
3269 softpipe_get_lambda_func(const struct pipe_sampler_view *view, unsigned shader)
3270 {
3271 if (shader != PIPE_SHADER_FRAGMENT)
3272 return compute_lambda_vert;
3273
3274 switch (view->target) {
3275 case PIPE_BUFFER:
3276 case PIPE_TEXTURE_1D:
3277 case PIPE_TEXTURE_1D_ARRAY:
3278 return compute_lambda_1d;
3279 case PIPE_TEXTURE_2D:
3280 case PIPE_TEXTURE_2D_ARRAY:
3281 case PIPE_TEXTURE_RECT:
3282 case PIPE_TEXTURE_CUBE:
3283 case PIPE_TEXTURE_CUBE_ARRAY:
3284 return compute_lambda_2d;
3285 case PIPE_TEXTURE_3D:
3286 return compute_lambda_3d;
3287 default:
3288 assert(0);
3289 return compute_lambda_1d;
3290 }
3291 }
3292
3293
3294 struct pipe_sampler_view *
3295 softpipe_create_sampler_view(struct pipe_context *pipe,
3296 struct pipe_resource *resource,
3297 const struct pipe_sampler_view *templ)
3298 {
3299 struct sp_sampler_view *sview = CALLOC_STRUCT(sp_sampler_view);
3300 struct softpipe_resource *spr = (struct softpipe_resource *)resource;
3301
3302 if (sview) {
3303 struct pipe_sampler_view *view = &sview->base;
3304 *view = *templ;
3305 view->reference.count = 1;
3306 view->texture = NULL;
3307 pipe_resource_reference(&view->texture, resource);
3308 view->context = pipe;
3309
3310 #ifdef DEBUG
3311 /*
3312 * This is possibly too lenient, but the primary reason is just
3313 * to catch state trackers which forget to initialize this, so
3314 * it only catches clearly impossible view targets.
3315 */
3316 if (view->target != resource->target) {
3317 if (view->target == PIPE_TEXTURE_1D)
3318 assert(resource->target == PIPE_TEXTURE_1D_ARRAY);
3319 else if (view->target == PIPE_TEXTURE_1D_ARRAY)
3320 assert(resource->target == PIPE_TEXTURE_1D);
3321 else if (view->target == PIPE_TEXTURE_2D)
3322 assert(resource->target == PIPE_TEXTURE_2D_ARRAY ||
3323 resource->target == PIPE_TEXTURE_CUBE ||
3324 resource->target == PIPE_TEXTURE_CUBE_ARRAY);
3325 else if (view->target == PIPE_TEXTURE_2D_ARRAY)
3326 assert(resource->target == PIPE_TEXTURE_2D ||
3327 resource->target == PIPE_TEXTURE_CUBE ||
3328 resource->target == PIPE_TEXTURE_CUBE_ARRAY);
3329 else if (view->target == PIPE_TEXTURE_CUBE)
3330 assert(resource->target == PIPE_TEXTURE_CUBE_ARRAY ||
3331 resource->target == PIPE_TEXTURE_2D_ARRAY);
3332 else if (view->target == PIPE_TEXTURE_CUBE_ARRAY)
3333 assert(resource->target == PIPE_TEXTURE_CUBE ||
3334 resource->target == PIPE_TEXTURE_2D_ARRAY);
3335 else
3336 assert(0);
3337 }
3338 #endif
3339
3340 if (any_swizzle(view)) {
3341 sview->need_swizzle = TRUE;
3342 }
3343
3344 if (view->target == PIPE_TEXTURE_CUBE ||
3345 view->target == PIPE_TEXTURE_CUBE_ARRAY)
3346 sview->get_samples = sample_cube;
3347 else {
3348 sview->get_samples = sample_mip;
3349 }
3350 sview->pot2d = spr->pot &&
3351 (view->target == PIPE_TEXTURE_2D ||
3352 view->target == PIPE_TEXTURE_RECT);
3353
3354 sview->xpot = util_logbase2( resource->width0 );
3355 sview->ypot = util_logbase2( resource->height0 );
3356 }
3357
3358 return (struct pipe_sampler_view *) sview;
3359 }
3360
3361
3362 static void
3363 sp_tgsi_get_dims(struct tgsi_sampler *tgsi_sampler,
3364 const unsigned sview_index,
3365 int level, int dims[4])
3366 {
3367 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3368
3369 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3370 /* always have a view here but texture is NULL if no sampler view was set. */
3371 if (!sp_samp->sp_sview[sview_index].base.texture) {
3372 dims[0] = dims[1] = dims[2] = dims[3] = 0;
3373 return;
3374 }
3375 sp_get_dims(&sp_samp->sp_sview[sview_index], level, dims);
3376 }
3377
3378
3379 static void
3380 sp_tgsi_get_samples(struct tgsi_sampler *tgsi_sampler,
3381 const unsigned sview_index,
3382 const unsigned sampler_index,
3383 const float s[TGSI_QUAD_SIZE],
3384 const float t[TGSI_QUAD_SIZE],
3385 const float p[TGSI_QUAD_SIZE],
3386 const float c0[TGSI_QUAD_SIZE],
3387 const float lod[TGSI_QUAD_SIZE],
3388 float derivs[3][2][TGSI_QUAD_SIZE],
3389 const int8_t offset[3],
3390 enum tgsi_sampler_control control,
3391 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
3392 {
3393 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3394 struct filter_args filt_args;
3395 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3396 assert(sampler_index < PIPE_MAX_SAMPLERS);
3397 assert(sp_samp->sp_sampler[sampler_index]);
3398 /* always have a view here but texture is NULL if no sampler view was set. */
3399 if (!sp_samp->sp_sview[sview_index].base.texture) {
3400 int i, j;
3401 for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
3402 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
3403 rgba[j][i] = 0.0f;
3404 }
3405 }
3406 return;
3407 }
3408
3409 filt_args.control = control;
3410 filt_args.offset = offset;
3411 sp_samp->sp_sview[sview_index].get_samples(&sp_samp->sp_sview[sview_index],
3412 sp_samp->sp_sampler[sampler_index],
3413 s, t, p, c0, lod, &filt_args, rgba);
3414 }
3415
3416
3417 static void
3418 sp_tgsi_get_texel(struct tgsi_sampler *tgsi_sampler,
3419 const unsigned sview_index,
3420 const int i[TGSI_QUAD_SIZE],
3421 const int j[TGSI_QUAD_SIZE], const int k[TGSI_QUAD_SIZE],
3422 const int lod[TGSI_QUAD_SIZE], const int8_t offset[3],
3423 float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE])
3424 {
3425 struct sp_tgsi_sampler *sp_samp = (struct sp_tgsi_sampler *)tgsi_sampler;
3426
3427 assert(sview_index < PIPE_MAX_SHADER_SAMPLER_VIEWS);
3428 /* always have a view here but texture is NULL if no sampler view was set. */
3429 if (!sp_samp->sp_sview[sview_index].base.texture) {
3430 int i, j;
3431 for (j = 0; j < TGSI_NUM_CHANNELS; j++) {
3432 for (i = 0; i < TGSI_QUAD_SIZE; i++) {
3433 rgba[j][i] = 0.0f;
3434 }
3435 }
3436 return;
3437 }
3438 sp_get_texels(&sp_samp->sp_sview[sview_index], i, j, k, lod, offset, rgba);
3439 }
3440
3441
3442 struct sp_tgsi_sampler *
3443 sp_create_tgsi_sampler(void)
3444 {
3445 struct sp_tgsi_sampler *samp = CALLOC_STRUCT(sp_tgsi_sampler);
3446 if (!samp)
3447 return NULL;
3448
3449 samp->base.get_dims = sp_tgsi_get_dims;
3450 samp->base.get_samples = sp_tgsi_get_samples;
3451 samp->base.get_texel = sp_tgsi_get_texel;
3452
3453 return samp;
3454 }
3455