1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2008-2010 VMware, Inc. All rights reserved.
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:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
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_memory.h"
42 #include "sp_quad.h" /* only for #define QUAD_* tokens */
43 #include "sp_tex_sample.h"
44 #include "sp_tex_tile_cache.h"
47 /** Set to one to help debug texture sampling */
52 * Return fractional part of 'f'. Used for computing interpolation weights.
53 * Need to be careful with negative values.
54 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
55 * of improperly weighted linear-filtered textures.
56 * The tests/texwrap.c demo is a good test.
67 * Linear interpolation macro
70 lerp(float a, float v0, float v1)
72 return v0 + a * (v1 - v0);
77 * Do 2D/bilinear interpolation of float values.
78 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
79 * a and b are the horizontal and vertical interpolants.
80 * It's important that this function is inlined when compiled with
81 * optimization! If we find that's not true on some systems, convert
85 lerp_2d(float a, float b,
86 float v00, float v10, float v01, float v11)
88 const float temp0 = lerp(a, v00, v10);
89 const float temp1 = lerp(a, v01, v11);
90 return lerp(b, temp0, temp1);
95 * As above, but 3D interpolation of 8 values.
98 lerp_3d(float a, float b, float c,
99 float v000, float v100, float v010, float v110,
100 float v001, float v101, float v011, float v111)
102 const float temp0 = lerp_2d(a, b, v000, v100, v010, v110);
103 const float temp1 = lerp_2d(a, b, v001, v101, v011, v111);
104 return lerp(c, temp0, temp1);
110 * Compute coord % size for repeat wrap modes.
111 * Note that if coord is negative, coord % size doesn't give the right
112 * value. To avoid that problem we add a large multiple of the size
113 * (rather than using a conditional).
116 repeat(int coord, unsigned size)
118 return (coord + size * 1024) % size;
123 * Apply texture coord wrapping mode and return integer texture indexes
124 * for a vector of four texcoords (S or T or P).
125 * \param wrapMode PIPE_TEX_WRAP_x
126 * \param s the incoming texcoords
127 * \param size the texture image size
128 * \param icoord returns the integer texcoords
129 * \return integer texture index
132 wrap_nearest_repeat(const float s[4], unsigned size, int icoord[4])
135 /* s limited to [0,1) */
136 /* i limited to [0,size-1] */
137 for (ch = 0; ch < 4; ch++) {
138 int i = util_ifloor(s[ch] * size);
139 icoord[ch] = repeat(i, size);
145 wrap_nearest_clamp(const float s[4], unsigned size, int icoord[4])
148 /* s limited to [0,1] */
149 /* i limited to [0,size-1] */
150 for (ch = 0; ch < 4; ch++) {
153 else if (s[ch] >= 1.0F)
154 icoord[ch] = size - 1;
156 icoord[ch] = util_ifloor(s[ch] * size);
162 wrap_nearest_clamp_to_edge(const float s[4], unsigned size, int icoord[4])
165 /* s limited to [min,max] */
166 /* i limited to [0, size-1] */
167 const float min = 1.0F / (2.0F * size);
168 const float max = 1.0F - min;
169 for (ch = 0; ch < 4; ch++) {
172 else if (s[ch] > max)
173 icoord[ch] = size - 1;
175 icoord[ch] = util_ifloor(s[ch] * size);
181 wrap_nearest_clamp_to_border(const float s[4], unsigned size, int icoord[4])
184 /* s limited to [min,max] */
185 /* i limited to [-1, size] */
186 const float min = -1.0F / (2.0F * size);
187 const float max = 1.0F - min;
188 for (ch = 0; ch < 4; ch++) {
191 else if (s[ch] >= max)
194 icoord[ch] = util_ifloor(s[ch] * size);
200 wrap_nearest_mirror_repeat(const float s[4], unsigned size, int icoord[4])
203 const float min = 1.0F / (2.0F * size);
204 const float max = 1.0F - min;
205 for (ch = 0; ch < 4; ch++) {
206 const int flr = util_ifloor(s[ch]);
207 float u = frac(s[ch]);
213 icoord[ch] = size - 1;
215 icoord[ch] = util_ifloor(u * size);
221 wrap_nearest_mirror_clamp(const float s[4], unsigned size, int icoord[4])
224 for (ch = 0; ch < 4; ch++) {
225 /* s limited to [0,1] */
226 /* i limited to [0,size-1] */
227 const float u = fabsf(s[ch]);
231 icoord[ch] = size - 1;
233 icoord[ch] = util_ifloor(u * size);
239 wrap_nearest_mirror_clamp_to_edge(const float s[4], unsigned size,
243 /* s limited to [min,max] */
244 /* i limited to [0, size-1] */
245 const float min = 1.0F / (2.0F * size);
246 const float max = 1.0F - min;
247 for (ch = 0; ch < 4; ch++) {
248 const float u = fabsf(s[ch]);
252 icoord[ch] = size - 1;
254 icoord[ch] = util_ifloor(u * size);
260 wrap_nearest_mirror_clamp_to_border(const float s[4], unsigned size,
264 /* s limited to [min,max] */
265 /* i limited to [0, size-1] */
266 const float min = -1.0F / (2.0F * size);
267 const float max = 1.0F - min;
268 for (ch = 0; ch < 4; ch++) {
269 const float u = fabsf(s[ch]);
275 icoord[ch] = util_ifloor(u * size);
281 * Used to compute texel locations for linear sampling for four texcoords.
282 * \param wrapMode PIPE_TEX_WRAP_x
283 * \param s the texcoords
284 * \param size the texture image size
285 * \param icoord0 returns first texture indexes
286 * \param icoord1 returns second texture indexes (usually icoord0 + 1)
287 * \param w returns blend factor/weight between texture indexes
288 * \param icoord returns the computed integer texture coords
291 wrap_linear_repeat(const float s[4], unsigned size,
292 int icoord0[4], int icoord1[4], float w[4])
295 for (ch = 0; ch < 4; ch++) {
296 float u = s[ch] * size - 0.5F;
297 icoord0[ch] = repeat(util_ifloor(u), size);
298 icoord1[ch] = repeat(icoord0[ch] + 1, size);
305 wrap_linear_clamp(const float s[4], unsigned size,
306 int icoord0[4], int icoord1[4], float w[4])
309 for (ch = 0; ch < 4; ch++) {
310 float u = CLAMP(s[ch], 0.0F, 1.0F);
312 icoord0[ch] = util_ifloor(u);
313 icoord1[ch] = icoord0[ch] + 1;
320 wrap_linear_clamp_to_edge(const float s[4], unsigned size,
321 int icoord0[4], int icoord1[4], float w[4])
324 for (ch = 0; ch < 4; ch++) {
325 float u = CLAMP(s[ch], 0.0F, 1.0F);
327 icoord0[ch] = util_ifloor(u);
328 icoord1[ch] = icoord0[ch] + 1;
331 if (icoord1[ch] >= (int) size)
332 icoord1[ch] = size - 1;
339 wrap_linear_clamp_to_border(const float s[4], unsigned size,
340 int icoord0[4], int icoord1[4], float w[4])
342 const float min = -1.0F / (2.0F * size);
343 const float max = 1.0F - min;
345 for (ch = 0; ch < 4; ch++) {
346 float u = CLAMP(s[ch], min, max);
348 icoord0[ch] = util_ifloor(u);
349 icoord1[ch] = icoord0[ch] + 1;
356 wrap_linear_mirror_repeat(const float s[4], unsigned size,
357 int icoord0[4], int icoord1[4], float w[4])
360 for (ch = 0; ch < 4; ch++) {
361 const int flr = util_ifloor(s[ch]);
362 float u = frac(s[ch]);
366 icoord0[ch] = util_ifloor(u);
367 icoord1[ch] = icoord0[ch] + 1;
370 if (icoord1[ch] >= (int) size)
371 icoord1[ch] = size - 1;
378 wrap_linear_mirror_clamp(const float s[4], unsigned size,
379 int icoord0[4], int icoord1[4], float w[4])
382 for (ch = 0; ch < 4; ch++) {
383 float u = fabsf(s[ch]);
389 icoord0[ch] = util_ifloor(u);
390 icoord1[ch] = icoord0[ch] + 1;
397 wrap_linear_mirror_clamp_to_edge(const float s[4], unsigned size,
398 int icoord0[4], int icoord1[4], float w[4])
401 for (ch = 0; ch < 4; ch++) {
402 float u = fabsf(s[ch]);
408 icoord0[ch] = util_ifloor(u);
409 icoord1[ch] = icoord0[ch] + 1;
412 if (icoord1[ch] >= (int) size)
413 icoord1[ch] = size - 1;
420 wrap_linear_mirror_clamp_to_border(const float s[4], unsigned size,
421 int icoord0[4], int icoord1[4], float w[4])
423 const float min = -1.0F / (2.0F * size);
424 const float max = 1.0F - min;
426 for (ch = 0; ch < 4; ch++) {
427 float u = fabsf(s[ch]);
435 icoord0[ch] = util_ifloor(u);
436 icoord1[ch] = icoord0[ch] + 1;
443 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
446 wrap_nearest_unorm_clamp(const float s[4], unsigned size, int icoord[4])
449 for (ch = 0; ch < 4; ch++) {
450 int i = util_ifloor(s[ch]);
451 icoord[ch]= CLAMP(i, 0, (int) size-1);
457 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
460 wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size,
464 for (ch = 0; ch < 4; ch++) {
465 icoord[ch]= util_ifloor( CLAMP(s[ch], -0.5F, (float) size + 0.5F) );
471 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
474 wrap_nearest_unorm_clamp_to_edge(const float s[4], unsigned size,
478 for (ch = 0; ch < 4; ch++) {
479 icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
485 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
488 wrap_linear_unorm_clamp(const float s[4], unsigned size,
489 int icoord0[4], int icoord1[4], float w[4])
492 for (ch = 0; ch < 4; ch++) {
493 /* Not exactly what the spec says, but it matches NVIDIA output */
494 float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
495 icoord0[ch] = util_ifloor(u);
496 icoord1[ch] = icoord0[ch] + 1;
503 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
506 wrap_linear_unorm_clamp_to_border(const float s[4], unsigned size,
507 int icoord0[4], int icoord1[4], float w[4])
510 for (ch = 0; ch < 4; ch++) {
511 float u = CLAMP(s[ch], -0.5F, (float) size + 0.5F);
513 icoord0[ch] = util_ifloor(u);
514 icoord1[ch] = icoord0[ch] + 1;
515 if (icoord1[ch] > (int) size - 1)
516 icoord1[ch] = size - 1;
523 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
526 wrap_linear_unorm_clamp_to_edge(const float s[4], unsigned size,
527 int icoord0[4], int icoord1[4], float w[4])
530 for (ch = 0; ch < 4; ch++) {
531 float u = CLAMP(s[ch], +0.5F, (float) size - 0.5F);
533 icoord0[ch] = util_ifloor(u);
534 icoord1[ch] = icoord0[ch] + 1;
535 if (icoord1[ch] > (int) size - 1)
536 icoord1[ch] = size - 1;
543 * Do coordinate to array index conversion. For array textures.
546 wrap_array_layer(const float coord[4], unsigned size, int layer[4])
549 for (ch = 0; ch < 4; ch++) {
550 int c = util_ifloor(coord[ch] + 0.5F);
551 layer[ch] = CLAMP(c, 0, size - 1);
557 * Examine the quad's texture coordinates to compute the partial
558 * derivatives w.r.t X and Y, then compute lambda (level of detail).
561 compute_lambda_1d(const struct sp_sampler_variant *samp,
562 const float s[QUAD_SIZE],
563 const float t[QUAD_SIZE],
564 const float p[QUAD_SIZE])
566 const struct pipe_resource *texture = samp->view->texture;
567 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
568 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
569 float rho = MAX2(dsdx, dsdy) * u_minify(texture->width0, samp->view->u.tex.first_level);
571 return util_fast_log2(rho);
576 compute_lambda_2d(const struct sp_sampler_variant *samp,
577 const float s[QUAD_SIZE],
578 const float t[QUAD_SIZE],
579 const float p[QUAD_SIZE])
581 const struct pipe_resource *texture = samp->view->texture;
582 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
583 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
584 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
585 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
586 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, samp->view->u.tex.first_level);
587 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, samp->view->u.tex.first_level);
588 float rho = MAX2(maxx, maxy);
590 return util_fast_log2(rho);
595 compute_lambda_3d(const struct sp_sampler_variant *samp,
596 const float s[QUAD_SIZE],
597 const float t[QUAD_SIZE],
598 const float p[QUAD_SIZE])
600 const struct pipe_resource *texture = samp->view->texture;
601 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
602 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
603 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
604 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
605 float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
606 float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
607 float maxx = MAX2(dsdx, dsdy) * u_minify(texture->width0, samp->view->u.tex.first_level);
608 float maxy = MAX2(dtdx, dtdy) * u_minify(texture->height0, samp->view->u.tex.first_level);
609 float maxz = MAX2(dpdx, dpdy) * u_minify(texture->depth0, samp->view->u.tex.first_level);
612 rho = MAX2(maxx, maxy);
613 rho = MAX2(rho, maxz);
615 return util_fast_log2(rho);
620 * Compute lambda for a vertex texture sampler.
621 * Since there aren't derivatives to use, just return 0.
624 compute_lambda_vert(const struct sp_sampler_variant *samp,
625 const float s[QUAD_SIZE],
626 const float t[QUAD_SIZE],
627 const float p[QUAD_SIZE])
635 * Get a texel from a texture, using the texture tile cache.
637 * \param addr the template tex address containing cube, z, face info.
638 * \param x the x coord of texel within 2D image
639 * \param y the y coord of texel within 2D image
640 * \param rgba the quad to put the texel/color into
642 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
643 * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1...
649 static INLINE const float *
650 get_texel_2d_no_border(const struct sp_sampler_variant *samp,
651 union tex_tile_address addr, int x, int y)
653 const struct softpipe_tex_cached_tile *tile;
655 addr.bits.x = x / TILE_SIZE;
656 addr.bits.y = y / TILE_SIZE;
660 tile = sp_get_cached_tile_tex(samp->cache, addr);
662 return &tile->data.color[y][x][0];
666 static INLINE const float *
667 get_texel_2d(const struct sp_sampler_variant *samp,
668 union tex_tile_address addr, int x, int y)
670 const struct pipe_resource *texture = samp->view->texture;
671 unsigned level = addr.bits.level;
673 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
674 y < 0 || y >= (int) u_minify(texture->height0, level)) {
675 return samp->sampler->border_color.f;
678 return get_texel_2d_no_border( samp, addr, x, y );
683 /* Gather a quad of adjacent texels within a tile:
686 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_variant *samp,
687 union tex_tile_address addr,
688 unsigned x, unsigned y,
691 const struct softpipe_tex_cached_tile *tile;
693 addr.bits.x = x / TILE_SIZE;
694 addr.bits.y = y / TILE_SIZE;
698 tile = sp_get_cached_tile_tex(samp->cache, addr);
700 out[0] = &tile->data.color[y ][x ][0];
701 out[1] = &tile->data.color[y ][x+1][0];
702 out[2] = &tile->data.color[y+1][x ][0];
703 out[3] = &tile->data.color[y+1][x+1][0];
707 /* Gather a quad of potentially non-adjacent texels:
710 get_texel_quad_2d_no_border(const struct sp_sampler_variant *samp,
711 union tex_tile_address addr,
716 out[0] = get_texel_2d_no_border( samp, addr, x0, y0 );
717 out[1] = get_texel_2d_no_border( samp, addr, x1, y0 );
718 out[2] = get_texel_2d_no_border( samp, addr, x0, y1 );
719 out[3] = get_texel_2d_no_border( samp, addr, x1, y1 );
722 /* Can involve a lot of unnecessary checks for border color:
725 get_texel_quad_2d(const struct sp_sampler_variant *samp,
726 union tex_tile_address addr,
731 out[0] = get_texel_2d( samp, addr, x0, y0 );
732 out[1] = get_texel_2d( samp, addr, x1, y0 );
733 out[3] = get_texel_2d( samp, addr, x1, y1 );
734 out[2] = get_texel_2d( samp, addr, x0, y1 );
741 static INLINE const float *
742 get_texel_3d_no_border(const struct sp_sampler_variant *samp,
743 union tex_tile_address addr, int x, int y, int z)
745 const struct softpipe_tex_cached_tile *tile;
747 addr.bits.x = x / TILE_SIZE;
748 addr.bits.y = y / TILE_SIZE;
753 tile = sp_get_cached_tile_tex(samp->cache, addr);
755 return &tile->data.color[y][x][0];
759 static INLINE const float *
760 get_texel_3d(const struct sp_sampler_variant *samp,
761 union tex_tile_address addr, int x, int y, int z)
763 const struct pipe_resource *texture = samp->view->texture;
764 unsigned level = addr.bits.level;
766 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
767 y < 0 || y >= (int) u_minify(texture->height0, level) ||
768 z < 0 || z >= (int) u_minify(texture->depth0, level)) {
769 return samp->sampler->border_color.f;
772 return get_texel_3d_no_border( samp, addr, x, y, z );
777 /* Get texel pointer for 1D array texture */
778 static INLINE const float *
779 get_texel_1d_array(const struct sp_sampler_variant *samp,
780 union tex_tile_address addr, int x, int y)
782 const struct pipe_resource *texture = samp->view->texture;
783 unsigned level = addr.bits.level;
785 if (x < 0 || x >= (int) u_minify(texture->width0, level)) {
786 return samp->sampler->border_color.f;
789 return get_texel_2d_no_border(samp, addr, x, y);
794 /* Get texel pointer for 2D array texture */
795 static INLINE const float *
796 get_texel_2d_array(const struct sp_sampler_variant *samp,
797 union tex_tile_address addr, int x, int y, int layer)
799 const struct pipe_resource *texture = samp->view->texture;
800 unsigned level = addr.bits.level;
802 assert(layer < texture->array_size);
804 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
805 y < 0 || y >= (int) u_minify(texture->height0, level)) {
806 return samp->sampler->border_color.f;
809 return get_texel_3d_no_border(samp, addr, x, y, layer);
815 * Given the logbase2 of a mipmap's base level size and a mipmap level,
816 * return the size (in texels) of that mipmap level.
817 * For example, if level[0].width = 256 then base_pot will be 8.
818 * If level = 2, then we'll return 64 (the width at level=2).
819 * Return 1 if level > base_pot.
821 static INLINE unsigned
822 pot_level_size(unsigned base_pot, unsigned level)
824 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
829 print_sample(const char *function, float rgba[NUM_CHANNELS][QUAD_SIZE])
831 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
833 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
834 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
835 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
836 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
840 /* Some image-filter fastpaths:
843 img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
844 const float s[QUAD_SIZE],
845 const float t[QUAD_SIZE],
846 const float p[QUAD_SIZE],
847 const float c0[QUAD_SIZE],
848 enum tgsi_sampler_control control,
849 float rgba[NUM_CHANNELS][QUAD_SIZE])
851 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
853 unsigned level = samp->level;
854 unsigned xpot = pot_level_size(samp->xpot, level);
855 unsigned ypot = pot_level_size(samp->ypot, level);
856 unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */
857 unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */
858 union tex_tile_address addr;
861 addr.bits.level = samp->level;
863 for (j = 0; j < QUAD_SIZE; j++) {
866 float u = s[j] * xpot - 0.5F;
867 float v = t[j] * ypot - 0.5F;
869 int uflr = util_ifloor(u);
870 int vflr = util_ifloor(v);
872 float xw = u - (float)uflr;
873 float yw = v - (float)vflr;
875 int x0 = uflr & (xpot - 1);
876 int y0 = vflr & (ypot - 1);
880 /* Can we fetch all four at once:
882 if (x0 < xmax && y0 < ymax) {
883 get_texel_quad_2d_no_border_single_tile(samp, addr, x0, y0, tx);
886 unsigned x1 = (x0 + 1) & (xpot - 1);
887 unsigned y1 = (y0 + 1) & (ypot - 1);
888 get_texel_quad_2d_no_border(samp, addr, x0, y0, x1, y1, tx);
891 /* interpolate R, G, B, A */
892 for (c = 0; c < 4; c++) {
893 rgba[c][j] = lerp_2d(xw, yw,
900 print_sample(__FUNCTION__, rgba);
906 img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
907 const float s[QUAD_SIZE],
908 const float t[QUAD_SIZE],
909 const float p[QUAD_SIZE],
910 const float c0[QUAD_SIZE],
911 enum tgsi_sampler_control control,
912 float rgba[NUM_CHANNELS][QUAD_SIZE])
914 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
916 unsigned level = samp->level;
917 unsigned xpot = pot_level_size(samp->xpot, level);
918 unsigned ypot = pot_level_size(samp->ypot, level);
919 union tex_tile_address addr;
922 addr.bits.level = samp->level;
924 for (j = 0; j < QUAD_SIZE; j++) {
927 float u = s[j] * xpot;
928 float v = t[j] * ypot;
930 int uflr = util_ifloor(u);
931 int vflr = util_ifloor(v);
933 int x0 = uflr & (xpot - 1);
934 int y0 = vflr & (ypot - 1);
936 const float *out = get_texel_2d_no_border(samp, addr, x0, y0);
938 for (c = 0; c < 4; c++) {
944 print_sample(__FUNCTION__, rgba);
950 img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
951 const float s[QUAD_SIZE],
952 const float t[QUAD_SIZE],
953 const float p[QUAD_SIZE],
954 const float c0[QUAD_SIZE],
955 enum tgsi_sampler_control control,
956 float rgba[NUM_CHANNELS][QUAD_SIZE])
958 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
960 unsigned level = samp->level;
961 unsigned xpot = pot_level_size(samp->xpot, level);
962 unsigned ypot = pot_level_size(samp->ypot, level);
963 union tex_tile_address addr;
966 addr.bits.level = samp->level;
968 for (j = 0; j < QUAD_SIZE; j++) {
971 float u = s[j] * xpot;
972 float v = t[j] * ypot;
980 else if (x0 > xpot - 1)
986 else if (y0 > ypot - 1)
989 out = get_texel_2d_no_border(samp, addr, x0, y0);
991 for (c = 0; c < 4; c++) {
997 print_sample(__FUNCTION__, rgba);
1003 img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
1004 const float s[QUAD_SIZE],
1005 const float t[QUAD_SIZE],
1006 const float p[QUAD_SIZE],
1007 const float c0[QUAD_SIZE],
1008 enum tgsi_sampler_control control,
1009 float rgba[NUM_CHANNELS][QUAD_SIZE])
1011 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1012 const struct pipe_resource *texture = samp->view->texture;
1016 union tex_tile_address addr;
1018 level0 = samp->level;
1019 width = u_minify(texture->width0, level0);
1024 addr.bits.level = samp->level;
1026 samp->nearest_texcoord_s(s, width, x);
1028 for (j = 0; j < QUAD_SIZE; j++) {
1029 const float *out = get_texel_2d(samp, addr, x[j], 0);
1031 for (c = 0; c < 4; c++) {
1032 rgba[c][j] = out[c];
1037 print_sample(__FUNCTION__, rgba);
1043 img_filter_1d_array_nearest(struct tgsi_sampler *tgsi_sampler,
1044 const float s[QUAD_SIZE],
1045 const float t[QUAD_SIZE],
1046 const float p[QUAD_SIZE],
1047 const float c0[QUAD_SIZE],
1048 enum tgsi_sampler_control control,
1049 float rgba[NUM_CHANNELS][QUAD_SIZE])
1051 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1052 const struct pipe_resource *texture = samp->view->texture;
1056 union tex_tile_address addr;
1058 level0 = samp->level;
1059 width = u_minify(texture->width0, level0);
1064 addr.bits.level = samp->level;
1066 samp->nearest_texcoord_s(s, width, x);
1067 wrap_array_layer(t, texture->array_size, layer);
1069 for (j = 0; j < QUAD_SIZE; j++) {
1070 const float *out = get_texel_1d_array(samp, addr, x[j], layer[j]);
1072 for (c = 0; c < 4; c++) {
1073 rgba[c][j] = out[c];
1078 print_sample(__FUNCTION__, rgba);
1084 img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
1085 const float s[QUAD_SIZE],
1086 const float t[QUAD_SIZE],
1087 const float p[QUAD_SIZE],
1088 const float c0[QUAD_SIZE],
1089 enum tgsi_sampler_control control,
1090 float rgba[NUM_CHANNELS][QUAD_SIZE])
1092 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1093 const struct pipe_resource *texture = samp->view->texture;
1097 union tex_tile_address addr;
1100 level0 = samp->level;
1101 width = u_minify(texture->width0, level0);
1102 height = u_minify(texture->height0, level0);
1108 addr.bits.level = samp->level;
1110 samp->nearest_texcoord_s(s, width, x);
1111 samp->nearest_texcoord_t(t, height, y);
1113 for (j = 0; j < QUAD_SIZE; j++) {
1114 const float *out = get_texel_2d(samp, addr, x[j], y[j]);
1116 for (c = 0; c < 4; c++) {
1117 rgba[c][j] = out[c];
1122 print_sample(__FUNCTION__, rgba);
1128 img_filter_2d_array_nearest(struct tgsi_sampler *tgsi_sampler,
1129 const float s[QUAD_SIZE],
1130 const float t[QUAD_SIZE],
1131 const float p[QUAD_SIZE],
1132 const float c0[QUAD_SIZE],
1133 enum tgsi_sampler_control control,
1134 float rgba[NUM_CHANNELS][QUAD_SIZE])
1136 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1137 const struct pipe_resource *texture = samp->view->texture;
1140 int x[4], y[4], layer[4];
1141 union tex_tile_address addr;
1143 level0 = samp->level;
1144 width = u_minify(texture->width0, level0);
1145 height = u_minify(texture->height0, level0);
1151 addr.bits.level = samp->level;
1153 samp->nearest_texcoord_s(s, width, x);
1154 samp->nearest_texcoord_t(t, height, y);
1155 wrap_array_layer(p, texture->array_size, layer);
1157 for (j = 0; j < QUAD_SIZE; j++) {
1158 const float *out = get_texel_2d_array(samp, addr, x[j], y[j], layer[j]);
1160 for (c = 0; c < 4; c++) {
1161 rgba[c][j] = out[c];
1166 print_sample(__FUNCTION__, rgba);
1171 static INLINE union tex_tile_address
1172 face(union tex_tile_address addr, unsigned face )
1174 addr.bits.face = face;
1180 img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
1181 const float s[QUAD_SIZE],
1182 const float t[QUAD_SIZE],
1183 const float p[QUAD_SIZE],
1184 const float c0[QUAD_SIZE],
1185 enum tgsi_sampler_control control,
1186 float rgba[NUM_CHANNELS][QUAD_SIZE])
1188 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1189 const struct pipe_resource *texture = samp->view->texture;
1190 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
1194 union tex_tile_address addr;
1196 level0 = samp->level;
1197 width = u_minify(texture->width0, level0);
1198 height = u_minify(texture->height0, level0);
1204 addr.bits.level = samp->level;
1206 samp->nearest_texcoord_s(s, width, x);
1207 samp->nearest_texcoord_t(t, height, y);
1209 for (j = 0; j < QUAD_SIZE; j++) {
1210 const float *out = get_texel_2d(samp, face(addr, faces[j]), x[j], y[j]);
1212 for (c = 0; c < 4; c++) {
1213 rgba[c][j] = out[c];
1218 print_sample(__FUNCTION__, rgba);
1224 img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
1225 const float s[QUAD_SIZE],
1226 const float t[QUAD_SIZE],
1227 const float p[QUAD_SIZE],
1228 const float c0[QUAD_SIZE],
1229 enum tgsi_sampler_control control,
1230 float rgba[NUM_CHANNELS][QUAD_SIZE])
1232 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1233 const struct pipe_resource *texture = samp->view->texture;
1235 int width, height, depth;
1236 int x[4], y[4], z[4];
1237 union tex_tile_address addr;
1239 level0 = samp->level;
1240 width = u_minify(texture->width0, level0);
1241 height = u_minify(texture->height0, level0);
1242 depth = u_minify(texture->depth0, level0);
1248 samp->nearest_texcoord_s(s, width, x);
1249 samp->nearest_texcoord_t(t, height, y);
1250 samp->nearest_texcoord_p(p, depth, z);
1253 addr.bits.level = samp->level;
1255 for (j = 0; j < QUAD_SIZE; j++) {
1256 const float *out = get_texel_3d(samp, addr, x[j], y[j], z[j]);
1258 for (c = 0; c < 4; c++) {
1259 rgba[c][j] = out[c];
1266 img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
1267 const float s[QUAD_SIZE],
1268 const float t[QUAD_SIZE],
1269 const float p[QUAD_SIZE],
1270 const float c0[QUAD_SIZE],
1271 enum tgsi_sampler_control control,
1272 float rgba[NUM_CHANNELS][QUAD_SIZE])
1274 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1275 const struct pipe_resource *texture = samp->view->texture;
1279 float xw[4]; /* weights */
1280 union tex_tile_address addr;
1282 level0 = samp->level;
1283 width = u_minify(texture->width0, level0);
1288 addr.bits.level = samp->level;
1290 samp->linear_texcoord_s(s, width, x0, x1, xw);
1292 for (j = 0; j < QUAD_SIZE; j++) {
1293 const float *tx0 = get_texel_2d(samp, addr, x0[j], 0);
1294 const float *tx1 = get_texel_2d(samp, addr, x1[j], 0);
1297 /* interpolate R, G, B, A */
1298 for (c = 0; c < 4; c++) {
1299 rgba[c][j] = lerp(xw[j], tx0[c], tx1[c]);
1306 img_filter_1d_array_linear(struct tgsi_sampler *tgsi_sampler,
1307 const float s[QUAD_SIZE],
1308 const float t[QUAD_SIZE],
1309 const float p[QUAD_SIZE],
1310 const float c0[QUAD_SIZE],
1311 enum tgsi_sampler_control control,
1312 float rgba[NUM_CHANNELS][QUAD_SIZE])
1314 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1315 const struct pipe_resource *texture = samp->view->texture;
1318 int x0[4], x1[4], layer[4];
1319 float xw[4]; /* weights */
1320 union tex_tile_address addr;
1322 level0 = samp->level;
1323 width = u_minify(texture->width0, level0);
1328 addr.bits.level = samp->level;
1330 samp->linear_texcoord_s(s, width, x0, x1, xw);
1331 wrap_array_layer(t, texture->array_size, layer);
1333 for (j = 0; j < QUAD_SIZE; j++) {
1334 const float *tx0 = get_texel_1d_array(samp, addr, x0[j], layer[j]);
1335 const float *tx1 = get_texel_1d_array(samp, addr, x1[j], layer[j]);
1338 /* interpolate R, G, B, A */
1339 for (c = 0; c < 4; c++) {
1340 rgba[c][j] = lerp(xw[j], tx0[c], tx1[c]);
1347 img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
1348 const float s[QUAD_SIZE],
1349 const float t[QUAD_SIZE],
1350 const float p[QUAD_SIZE],
1351 const float c0[QUAD_SIZE],
1352 enum tgsi_sampler_control control,
1353 float rgba[NUM_CHANNELS][QUAD_SIZE])
1355 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1356 const struct pipe_resource *texture = samp->view->texture;
1359 int x0[4], y0[4], x1[4], y1[4];
1360 float xw[4], yw[4]; /* weights */
1361 union tex_tile_address addr;
1363 level0 = samp->level;
1364 width = u_minify(texture->width0, level0);
1365 height = u_minify(texture->height0, level0);
1371 addr.bits.level = samp->level;
1373 samp->linear_texcoord_s(s, width, x0, x1, xw);
1374 samp->linear_texcoord_t(t, height, y0, y1, yw);
1376 for (j = 0; j < QUAD_SIZE; j++) {
1377 const float *tx0 = get_texel_2d(samp, addr, x0[j], y0[j]);
1378 const float *tx1 = get_texel_2d(samp, addr, x1[j], y0[j]);
1379 const float *tx2 = get_texel_2d(samp, addr, x0[j], y1[j]);
1380 const float *tx3 = get_texel_2d(samp, addr, x1[j], y1[j]);
1383 /* interpolate R, G, B, A */
1384 for (c = 0; c < 4; c++) {
1385 rgba[c][j] = lerp_2d(xw[j], yw[j],
1394 img_filter_2d_array_linear(struct tgsi_sampler *tgsi_sampler,
1395 const float s[QUAD_SIZE],
1396 const float t[QUAD_SIZE],
1397 const float p[QUAD_SIZE],
1398 const float c0[QUAD_SIZE],
1399 enum tgsi_sampler_control control,
1400 float rgba[NUM_CHANNELS][QUAD_SIZE])
1402 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1403 const struct pipe_resource *texture = samp->view->texture;
1406 int x0[4], y0[4], x1[4], y1[4], layer[4];
1407 float xw[4], yw[4]; /* weights */
1408 union tex_tile_address addr;
1410 level0 = samp->level;
1411 width = u_minify(texture->width0, level0);
1412 height = u_minify(texture->height0, level0);
1418 addr.bits.level = samp->level;
1420 samp->linear_texcoord_s(s, width, x0, x1, xw);
1421 samp->linear_texcoord_t(t, height, y0, y1, yw);
1422 wrap_array_layer(p, texture->array_size, layer);
1424 for (j = 0; j < QUAD_SIZE; j++) {
1425 const float *tx0 = get_texel_2d_array(samp, addr, x0[j], y0[j], layer[j]);
1426 const float *tx1 = get_texel_2d_array(samp, addr, x1[j], y0[j], layer[j]);
1427 const float *tx2 = get_texel_2d_array(samp, addr, x0[j], y1[j], layer[j]);
1428 const float *tx3 = get_texel_2d_array(samp, addr, x1[j], y1[j], layer[j]);
1431 /* interpolate R, G, B, A */
1432 for (c = 0; c < 4; c++) {
1433 rgba[c][j] = lerp_2d(xw[j], yw[j],
1442 img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler,
1443 const float s[QUAD_SIZE],
1444 const float t[QUAD_SIZE],
1445 const float p[QUAD_SIZE],
1446 const float c0[QUAD_SIZE],
1447 enum tgsi_sampler_control control,
1448 float rgba[NUM_CHANNELS][QUAD_SIZE])
1450 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1451 const struct pipe_resource *texture = samp->view->texture;
1452 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
1455 int x0[4], y0[4], x1[4], y1[4];
1456 float xw[4], yw[4]; /* weights */
1457 union tex_tile_address addr;
1459 level0 = samp->level;
1460 width = u_minify(texture->width0, level0);
1461 height = u_minify(texture->height0, level0);
1467 addr.bits.level = samp->level;
1469 samp->linear_texcoord_s(s, width, x0, x1, xw);
1470 samp->linear_texcoord_t(t, height, y0, y1, yw);
1472 for (j = 0; j < QUAD_SIZE; j++) {
1473 union tex_tile_address addrj = face(addr, faces[j]);
1474 const float *tx0 = get_texel_2d(samp, addrj, x0[j], y0[j]);
1475 const float *tx1 = get_texel_2d(samp, addrj, x1[j], y0[j]);
1476 const float *tx2 = get_texel_2d(samp, addrj, x0[j], y1[j]);
1477 const float *tx3 = get_texel_2d(samp, addrj, x1[j], y1[j]);
1480 /* interpolate R, G, B, A */
1481 for (c = 0; c < 4; c++) {
1482 rgba[c][j] = lerp_2d(xw[j], yw[j],
1491 img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
1492 const float s[QUAD_SIZE],
1493 const float t[QUAD_SIZE],
1494 const float p[QUAD_SIZE],
1495 const float c0[QUAD_SIZE],
1496 enum tgsi_sampler_control control,
1497 float rgba[NUM_CHANNELS][QUAD_SIZE])
1499 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1500 const struct pipe_resource *texture = samp->view->texture;
1502 int width, height, depth;
1503 int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
1504 float xw[4], yw[4], zw[4]; /* interpolation weights */
1505 union tex_tile_address addr;
1507 level0 = samp->level;
1508 width = u_minify(texture->width0, level0);
1509 height = u_minify(texture->height0, level0);
1510 depth = u_minify(texture->depth0, level0);
1513 addr.bits.level = level0;
1519 samp->linear_texcoord_s(s, width, x0, x1, xw);
1520 samp->linear_texcoord_t(t, height, y0, y1, yw);
1521 samp->linear_texcoord_p(p, depth, z0, z1, zw);
1523 for (j = 0; j < QUAD_SIZE; j++) {
1526 const float *tx00 = get_texel_3d(samp, addr, x0[j], y0[j], z0[j]);
1527 const float *tx01 = get_texel_3d(samp, addr, x1[j], y0[j], z0[j]);
1528 const float *tx02 = get_texel_3d(samp, addr, x0[j], y1[j], z0[j]);
1529 const float *tx03 = get_texel_3d(samp, addr, x1[j], y1[j], z0[j]);
1531 const float *tx10 = get_texel_3d(samp, addr, x0[j], y0[j], z1[j]);
1532 const float *tx11 = get_texel_3d(samp, addr, x1[j], y0[j], z1[j]);
1533 const float *tx12 = get_texel_3d(samp, addr, x0[j], y1[j], z1[j]);
1534 const float *tx13 = get_texel_3d(samp, addr, x1[j], y1[j], z1[j]);
1536 /* interpolate R, G, B, A */
1537 for (c = 0; c < 4; c++) {
1538 rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1548 /* Calculate level of detail for every fragment.
1549 * Note that lambda has already been biased by global LOD bias.
1552 compute_lod(const struct pipe_sampler_state *sampler,
1553 const float biased_lambda,
1554 const float lodbias[QUAD_SIZE],
1555 float lod[QUAD_SIZE])
1559 for (i = 0; i < QUAD_SIZE; i++) {
1560 lod[i] = biased_lambda + lodbias[i];
1561 lod[i] = CLAMP(lod[i], sampler->min_lod, sampler->max_lod);
1567 mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
1568 const float s[QUAD_SIZE],
1569 const float t[QUAD_SIZE],
1570 const float p[QUAD_SIZE],
1571 const float c0[QUAD_SIZE],
1572 enum tgsi_sampler_control control,
1573 float rgba[NUM_CHANNELS][QUAD_SIZE])
1575 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1576 const struct pipe_resource *texture = samp->view->texture;
1579 float lod[QUAD_SIZE];
1581 if (control == tgsi_sampler_lod_bias) {
1582 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1583 compute_lod(samp->sampler, lambda, c0, lod);
1585 assert(control == tgsi_sampler_lod_explicit);
1587 memcpy(lod, c0, sizeof(lod));
1590 /* XXX: Take into account all lod values.
1593 level0 = samp->view->u.tex.first_level + (int)lambda;
1596 samp->level = samp->view->u.tex.first_level;
1597 samp->mag_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1599 else if (level0 >= texture->last_level) {
1600 samp->level = texture->last_level;
1601 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1604 float levelBlend = frac(lambda);
1609 samp->level = level0;
1610 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba0);
1612 samp->level = level0+1;
1613 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba1);
1615 for (j = 0; j < QUAD_SIZE; j++) {
1616 for (c = 0; c < 4; c++) {
1617 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1623 print_sample(__FUNCTION__, rgba);
1629 * Compute nearest mipmap level from texcoords.
1630 * Then sample the texture level for four elements of a quad.
1631 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1634 mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
1635 const float s[QUAD_SIZE],
1636 const float t[QUAD_SIZE],
1637 const float p[QUAD_SIZE],
1638 const float c0[QUAD_SIZE],
1639 enum tgsi_sampler_control control,
1640 float rgba[NUM_CHANNELS][QUAD_SIZE])
1642 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1643 const struct pipe_resource *texture = samp->view->texture;
1645 float lod[QUAD_SIZE];
1647 if (control == tgsi_sampler_lod_bias) {
1648 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1649 compute_lod(samp->sampler, lambda, c0, lod);
1651 assert(control == tgsi_sampler_lod_explicit);
1653 memcpy(lod, c0, sizeof(lod));
1656 /* XXX: Take into account all lod values.
1661 samp->level = samp->view->u.tex.first_level;
1662 samp->mag_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1665 samp->level = samp->view->u.tex.first_level + (int)(lambda + 0.5F) ;
1666 samp->level = MIN2(samp->level, (int)texture->last_level);
1667 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1671 print_sample(__FUNCTION__, rgba);
1677 mip_filter_none(struct tgsi_sampler *tgsi_sampler,
1678 const float s[QUAD_SIZE],
1679 const float t[QUAD_SIZE],
1680 const float p[QUAD_SIZE],
1681 const float c0[QUAD_SIZE],
1682 enum tgsi_sampler_control control,
1683 float rgba[NUM_CHANNELS][QUAD_SIZE])
1685 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1687 float lod[QUAD_SIZE];
1689 if (control == tgsi_sampler_lod_bias) {
1690 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1691 compute_lod(samp->sampler, lambda, c0, lod);
1693 assert(control == tgsi_sampler_lod_explicit);
1695 memcpy(lod, c0, sizeof(lod));
1698 /* XXX: Take into account all lod values.
1702 samp->level = samp->view->u.tex.first_level;
1704 samp->mag_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1707 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1712 /* For anisotropic filtering */
1713 #define WEIGHT_LUT_SIZE 1024
1715 static float *weightLut = NULL;
1718 * Creates the look-up table used to speed-up EWA sampling
1721 create_filter_table(void)
1725 weightLut = (float *) MALLOC(WEIGHT_LUT_SIZE * sizeof(float));
1727 for (i = 0; i < WEIGHT_LUT_SIZE; ++i) {
1729 float r2 = (float) i / (float) (WEIGHT_LUT_SIZE - 1);
1730 float weight = (float) exp(-alpha * r2);
1731 weightLut[i] = weight;
1738 * Elliptical weighted average (EWA) filter for producing high quality
1739 * anisotropic filtered results.
1740 * Based on the Higher Quality Elliptical Weighted Avarage Filter
1741 * published by Paul S. Heckbert in his Master's Thesis
1742 * "Fundamentals of Texture Mapping and Image Warping" (1989)
1745 img_filter_2d_ewa(struct tgsi_sampler *tgsi_sampler,
1746 const float s[QUAD_SIZE],
1747 const float t[QUAD_SIZE],
1748 const float p[QUAD_SIZE],
1749 const float c0[QUAD_SIZE],
1750 enum tgsi_sampler_control control,
1751 const float dudx, const float dvdx,
1752 const float dudy, const float dvdy,
1753 float rgba[NUM_CHANNELS][QUAD_SIZE])
1755 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1756 const struct pipe_resource *texture = samp->view->texture;
1758 unsigned level0 = samp->level > 0 ? samp->level : 0;
1759 float scaling = 1.0 / (1 << level0);
1760 int width = u_minify(texture->width0, level0);
1761 int height = u_minify(texture->height0, level0);
1763 float ux = dudx * scaling;
1764 float vx = dvdx * scaling;
1765 float uy = dudy * scaling;
1766 float vy = dvdy * scaling;
1768 /* compute ellipse coefficients to bound the region:
1769 * A*x*x + B*x*y + C*y*y = F.
1771 float A = vx*vx+vy*vy+1;
1772 float B = -2*(ux*vx+uy*vy);
1773 float C = ux*ux+uy*uy+1;
1774 float F = A*C-B*B/4.0;
1776 /* check if it is an ellipse */
1777 /* ASSERT(F > 0.0); */
1779 /* Compute the ellipse's (u,v) bounding box in texture space */
1780 float d = -B*B+4.0*C*A;
1781 float box_u = 2.0 / d * sqrt(d*C*F); /* box_u -> half of bbox with */
1782 float box_v = 2.0 / d * sqrt(A*d*F); /* box_v -> half of bbox height */
1784 float rgba_temp[NUM_CHANNELS][QUAD_SIZE];
1785 float s_buffer[QUAD_SIZE];
1786 float t_buffer[QUAD_SIZE];
1787 float weight_buffer[QUAD_SIZE];
1788 unsigned buffer_next;
1790 float den;// = 0.0F;
1792 float U;// = u0 - tex_u;
1795 /* Scale ellipse formula to directly index the Filter Lookup Table.
1796 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
1798 double formScale = (double) (WEIGHT_LUT_SIZE - 1) / F;
1802 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
1804 /* For each quad, the du and dx values are the same and so the ellipse is
1805 * also the same. Note that texel/image access can only be performed using
1806 * a quad, i.e. it is not possible to get the pixel value for a single
1807 * tex coord. In order to have a better performance, the access is buffered
1808 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is full,
1809 * then the pixel values are read from the image.
1813 for (j = 0; j < QUAD_SIZE; j++) {
1814 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
1815 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
1816 * value, q, is less than F, we're inside the ellipse
1818 float tex_u = -0.5F + s[j] * texture->width0 * scaling;
1819 float tex_v = -0.5F + t[j] * texture->height0 * scaling;
1821 int u0 = (int) floorf(tex_u - box_u);
1822 int u1 = (int) ceilf(tex_u + box_u);
1823 int v0 = (int) floorf(tex_v - box_v);
1824 int v1 = (int) ceilf(tex_v + box_v);
1826 float num[4] = {0.0F, 0.0F, 0.0F, 0.0F};
1830 for (v = v0; v <= v1; ++v) {
1831 float V = v - tex_v;
1832 float dq = A * (2 * U + 1) + B * V;
1833 float q = (C * V + B * U) * V + A * U * U;
1836 for (u = u0; u <= u1; ++u) {
1837 /* Note that the ellipse has been pre-scaled so F = WEIGHT_LUT_SIZE - 1 */
1838 if (q < WEIGHT_LUT_SIZE) {
1839 /* as a LUT is used, q must never be negative;
1840 * should not happen, though
1842 const int qClamped = q >= 0.0F ? q : 0;
1843 float weight = weightLut[qClamped];
1845 weight_buffer[buffer_next] = weight;
1846 s_buffer[buffer_next] = u / ((float) width);
1847 t_buffer[buffer_next] = v / ((float) height);
1850 if (buffer_next == QUAD_SIZE) {
1851 /* 4 texel coords are in the buffer -> read it now */
1853 /* it is assumed that samp->min_img_filter is set to
1854 * img_filter_2d_nearest or one of the
1855 * accelerated img_filter_2d_nearest_XXX functions.
1857 samp->min_img_filter(tgsi_sampler, s_buffer, t_buffer, p, NULL,
1858 tgsi_sampler_lod_bias, rgba_temp);
1859 for (jj = 0; jj < buffer_next; jj++) {
1860 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
1861 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
1862 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
1863 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
1876 /* if the tex coord buffer contains unread values, we will read them now.
1877 * Note that in most cases we have to read more pixel values than required,
1878 * however, as the img_filter_2d_nearest function(s) does not have a count
1879 * parameter, we need to read the whole quad and ignore the unused values
1881 if (buffer_next > 0) {
1883 /* it is assumed that samp->min_img_filter is set to
1884 * img_filter_2d_nearest or one of the
1885 * accelerated img_filter_2d_nearest_XXX functions.
1887 samp->min_img_filter(tgsi_sampler, s_buffer, t_buffer, p, NULL,
1888 tgsi_sampler_lod_bias, rgba_temp);
1889 for (jj = 0; jj < buffer_next; jj++) {
1890 num[0] += weight_buffer[jj] * rgba_temp[0][jj];
1891 num[1] += weight_buffer[jj] * rgba_temp[1][jj];
1892 num[2] += weight_buffer[jj] * rgba_temp[2][jj];
1893 num[3] += weight_buffer[jj] * rgba_temp[3][jj];
1898 /* Reaching this place would mean
1899 * that no pixels intersected the ellipse.
1900 * This should never happen because
1901 * the filter we use always
1902 * intersects at least one pixel.
1909 /* not enough pixels in resampling, resort to direct interpolation */
1910 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba_temp);
1912 num[0] = rgba_temp[0][j];
1913 num[1] = rgba_temp[1][j];
1914 num[2] = rgba_temp[2][j];
1915 num[3] = rgba_temp[3][j];
1918 rgba[0][j] = num[0] / den;
1919 rgba[1][j] = num[1] / den;
1920 rgba[2][j] = num[2] / den;
1921 rgba[3][j] = num[3] / den;
1927 * Sample 2D texture using an anisotropic filter.
1930 mip_filter_linear_aniso(struct tgsi_sampler *tgsi_sampler,
1931 const float s[QUAD_SIZE],
1932 const float t[QUAD_SIZE],
1933 const float p[QUAD_SIZE],
1934 const float c0[QUAD_SIZE],
1935 enum tgsi_sampler_control control,
1936 float rgba[NUM_CHANNELS][QUAD_SIZE])
1938 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
1939 const struct pipe_resource *texture = samp->view->texture;
1942 float lod[QUAD_SIZE];
1944 float s_to_u = u_minify(texture->width0, samp->view->u.tex.first_level);
1945 float t_to_v = u_minify(texture->height0, samp->view->u.tex.first_level);
1946 float dudx = (s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
1947 float dudy = (s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]) * s_to_u;
1948 float dvdx = (t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
1949 float dvdy = (t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]) * t_to_v;
1951 if (control == tgsi_sampler_lod_bias) {
1952 /* note: instead of working with Px and Py, we will use the
1953 * squared length instead, to avoid sqrt.
1955 float Px2 = dudx * dudx + dvdx * dvdx;
1956 float Py2 = dudy * dudy + dvdy * dvdy;
1961 const float maxEccentricity = samp->sampler->max_anisotropy * samp->sampler->max_anisotropy;
1972 /* if the eccentricity of the ellipse is too big, scale up the shorter
1973 * of the two vectors to limit the maximum amount of work per pixel
1976 if (e > maxEccentricity) {
1977 /* float s=e / maxEccentricity;
1981 Pmin2 = Pmax2 / maxEccentricity;
1984 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
1985 * this since 0.5*log(x) = log(sqrt(x))
1987 lambda = 0.5F * util_fast_log2(Pmin2) + samp->sampler->lod_bias;
1988 compute_lod(samp->sampler, lambda, c0, lod);
1991 assert(control == tgsi_sampler_lod_explicit);
1993 memcpy(lod, c0, sizeof(lod));
1996 /* XXX: Take into account all lod values.
1999 level0 = samp->view->u.tex.first_level + (int)lambda;
2001 /* If the ellipse covers the whole image, we can
2002 * simply return the average of the whole image.
2004 if (level0 >= (int) texture->last_level) {
2005 samp->level = texture->last_level;
2006 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
2009 /* don't bother interpolating between multiple LODs; it doesn't
2010 * seem to be worth the extra running time.
2012 samp->level = level0;
2013 img_filter_2d_ewa(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias,
2014 dudx, dvdx, dudy, dvdy, rgba);
2018 print_sample(__FUNCTION__, rgba);
2025 * Specialized version of mip_filter_linear with hard-wired calls to
2026 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2029 mip_filter_linear_2d_linear_repeat_POT(
2030 struct tgsi_sampler *tgsi_sampler,
2031 const float s[QUAD_SIZE],
2032 const float t[QUAD_SIZE],
2033 const float p[QUAD_SIZE],
2034 const float c0[QUAD_SIZE],
2035 enum tgsi_sampler_control control,
2036 float rgba[NUM_CHANNELS][QUAD_SIZE])
2038 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2039 const struct pipe_resource *texture = samp->view->texture;
2042 float lod[QUAD_SIZE];
2044 if (control == tgsi_sampler_lod_bias) {
2045 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
2046 compute_lod(samp->sampler, lambda, c0, lod);
2048 assert(control == tgsi_sampler_lod_explicit);
2050 memcpy(lod, c0, sizeof(lod));
2053 /* XXX: Take into account all lod values.
2056 level0 = samp->view->u.tex.first_level + (int)lambda;
2058 /* Catches both negative and large values of level0:
2060 if ((unsigned)level0 >= texture->last_level) {
2062 samp->level = samp->view->u.tex.first_level;
2064 samp->level = texture->last_level;
2066 img_filter_2d_linear_repeat_POT(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
2069 float levelBlend = frac(lambda);
2074 samp->level = level0;
2075 img_filter_2d_linear_repeat_POT(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba0);
2077 samp->level = level0+1;
2078 img_filter_2d_linear_repeat_POT(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba1);
2080 for (j = 0; j < QUAD_SIZE; j++) {
2081 for (c = 0; c < 4; c++) {
2082 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
2088 print_sample(__FUNCTION__, rgba);
2095 * Do shadow/depth comparisons.
2098 sample_compare(struct tgsi_sampler *tgsi_sampler,
2099 const float s[QUAD_SIZE],
2100 const float t[QUAD_SIZE],
2101 const float p[QUAD_SIZE],
2102 const float c0[QUAD_SIZE],
2103 enum tgsi_sampler_control control,
2104 float rgba[NUM_CHANNELS][QUAD_SIZE])
2106 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2107 const struct pipe_sampler_state *sampler = samp->sampler;
2108 int j, k0, k1, k2, k3;
2110 float pc0, pc1, pc2, pc3;
2112 samp->mip_filter(tgsi_sampler, s, t, p, c0, control, rgba);
2115 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2116 * for 2D Array texture we need to use the 'c0' (aka Q).
2117 * When we sampled the depth texture, the depth value was put into all
2118 * RGBA channels. We look at the red channel here.
2121 if (samp->view->texture->target == PIPE_TEXTURE_2D_ARRAY ||
2122 samp->view->texture->target == PIPE_TEXTURE_CUBE) {
2123 pc0 = CLAMP(c0[0], 0.0F, 1.0F);
2124 pc1 = CLAMP(c0[1], 0.0F, 1.0F);
2125 pc2 = CLAMP(c0[2], 0.0F, 1.0F);
2126 pc3 = CLAMP(c0[3], 0.0F, 1.0F);
2128 pc0 = CLAMP(p[0], 0.0F, 1.0F);
2129 pc1 = CLAMP(p[1], 0.0F, 1.0F);
2130 pc2 = CLAMP(p[2], 0.0F, 1.0F);
2131 pc3 = CLAMP(p[3], 0.0F, 1.0F);
2133 /* compare four texcoords vs. four texture samples */
2134 switch (sampler->compare_func) {
2135 case PIPE_FUNC_LESS:
2136 k0 = pc0 < rgba[0][0];
2137 k1 = pc1 < rgba[0][1];
2138 k2 = pc2 < rgba[0][2];
2139 k3 = pc3 < rgba[0][3];
2141 case PIPE_FUNC_LEQUAL:
2142 k0 = pc0 <= rgba[0][0];
2143 k1 = pc1 <= rgba[0][1];
2144 k2 = pc2 <= rgba[0][2];
2145 k3 = pc3 <= rgba[0][3];
2147 case PIPE_FUNC_GREATER:
2148 k0 = pc0 > rgba[0][0];
2149 k1 = pc1 > rgba[0][1];
2150 k2 = pc2 > rgba[0][2];
2151 k3 = pc3 > rgba[0][3];
2153 case PIPE_FUNC_GEQUAL:
2154 k0 = pc0 >= rgba[0][0];
2155 k1 = pc1 >= rgba[0][1];
2156 k2 = pc2 >= rgba[0][2];
2157 k3 = pc3 >= rgba[0][3];
2159 case PIPE_FUNC_EQUAL:
2160 k0 = pc0 == rgba[0][0];
2161 k1 = pc1 == rgba[0][1];
2162 k2 = pc2 == rgba[0][2];
2163 k3 = pc3 == rgba[0][3];
2165 case PIPE_FUNC_NOTEQUAL:
2166 k0 = pc0 != rgba[0][0];
2167 k1 = pc1 != rgba[0][1];
2168 k2 = pc2 != rgba[0][2];
2169 k3 = pc3 != rgba[0][3];
2171 case PIPE_FUNC_ALWAYS:
2172 k0 = k1 = k2 = k3 = 1;
2174 case PIPE_FUNC_NEVER:
2175 k0 = k1 = k2 = k3 = 0;
2178 k0 = k1 = k2 = k3 = 0;
2183 if (sampler->mag_img_filter == PIPE_TEX_FILTER_LINEAR) {
2184 /* convert four pass/fail values to an intensity in [0,1] */
2185 val = 0.25F * (k0 + k1 + k2 + k3);
2187 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
2188 for (j = 0; j < 4; j++) {
2189 rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
2193 for (j = 0; j < 4; j++) {
2204 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2205 * Put face info into the sampler faces[] array.
2208 sample_cube(struct tgsi_sampler *tgsi_sampler,
2209 const float s[QUAD_SIZE],
2210 const float t[QUAD_SIZE],
2211 const float p[QUAD_SIZE],
2212 const float c0[QUAD_SIZE],
2213 enum tgsi_sampler_control control,
2214 float rgba[NUM_CHANNELS][QUAD_SIZE])
2216 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2218 float ssss[4], tttt[4];
2222 direction target sc tc ma
2223 ---------- ------------------------------- --- --- ---
2224 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2225 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2226 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2227 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2228 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2229 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2232 /* Choose the cube face and compute new s/t coords for the 2D face.
2234 * Use the same cube face for all four pixels in the quad.
2236 * This isn't ideal, but if we want to use a different cube face
2237 * per pixel in the quad, we'd have to also compute the per-face
2238 * LOD here too. That's because the four post-face-selection
2239 * texcoords are no longer related to each other (they're
2240 * per-face!) so we can't use subtraction to compute the partial
2241 * deriviates to compute the LOD. Doing so (near cube edges
2242 * anyway) gives us pretty much random values.
2245 /* use the average of the four pixel's texcoords to choose the face */
2246 const float rx = 0.25F * (s[0] + s[1] + s[2] + s[3]);
2247 const float ry = 0.25F * (t[0] + t[1] + t[2] + t[3]);
2248 const float rz = 0.25F * (p[0] + p[1] + p[2] + p[3]);
2249 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
2251 if (arx >= ary && arx >= arz) {
2252 float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
2253 uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
2254 for (j = 0; j < QUAD_SIZE; j++) {
2255 const float ima = -0.5F / fabsf(s[j]);
2256 ssss[j] = sign * p[j] * ima + 0.5F;
2257 tttt[j] = t[j] * ima + 0.5F;
2258 samp->faces[j] = face;
2261 else if (ary >= arx && ary >= arz) {
2262 float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
2263 uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
2264 for (j = 0; j < QUAD_SIZE; j++) {
2265 const float ima = -0.5F / fabsf(t[j]);
2266 ssss[j] = -s[j] * ima + 0.5F;
2267 tttt[j] = sign * -p[j] * ima + 0.5F;
2268 samp->faces[j] = face;
2272 float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
2273 uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
2274 for (j = 0; j < QUAD_SIZE; j++) {
2275 const float ima = -0.5F / fabsf(p[j]);
2276 ssss[j] = sign * -s[j] * ima + 0.5F;
2277 tttt[j] = t[j] * ima + 0.5F;
2278 samp->faces[j] = face;
2283 /* In our little pipeline, the compare stage is next. If compare
2284 * is not active, this will point somewhere deeper into the
2285 * pipeline, eg. to mip_filter or even img_filter.
2287 samp->compare(tgsi_sampler, ssss, tttt, NULL, c0, control, rgba);
2290 static void do_swizzling(const struct sp_sampler_variant *samp,
2291 float in[NUM_CHANNELS][QUAD_SIZE],
2292 float out[NUM_CHANNELS][QUAD_SIZE])
2295 const unsigned swizzle_r = samp->key.bits.swizzle_r;
2296 const unsigned swizzle_g = samp->key.bits.swizzle_g;
2297 const unsigned swizzle_b = samp->key.bits.swizzle_b;
2298 const unsigned swizzle_a = samp->key.bits.swizzle_a;
2300 switch (swizzle_r) {
2301 case PIPE_SWIZZLE_ZERO:
2302 for (j = 0; j < 4; j++)
2305 case PIPE_SWIZZLE_ONE:
2306 for (j = 0; j < 4; j++)
2310 assert(swizzle_r < 4);
2311 for (j = 0; j < 4; j++)
2312 out[0][j] = in[swizzle_r][j];
2315 switch (swizzle_g) {
2316 case PIPE_SWIZZLE_ZERO:
2317 for (j = 0; j < 4; j++)
2320 case PIPE_SWIZZLE_ONE:
2321 for (j = 0; j < 4; j++)
2325 assert(swizzle_g < 4);
2326 for (j = 0; j < 4; j++)
2327 out[1][j] = in[swizzle_g][j];
2330 switch (swizzle_b) {
2331 case PIPE_SWIZZLE_ZERO:
2332 for (j = 0; j < 4; j++)
2335 case PIPE_SWIZZLE_ONE:
2336 for (j = 0; j < 4; j++)
2340 assert(swizzle_b < 4);
2341 for (j = 0; j < 4; j++)
2342 out[2][j] = in[swizzle_b][j];
2345 switch (swizzle_a) {
2346 case PIPE_SWIZZLE_ZERO:
2347 for (j = 0; j < 4; j++)
2350 case PIPE_SWIZZLE_ONE:
2351 for (j = 0; j < 4; j++)
2355 assert(swizzle_a < 4);
2356 for (j = 0; j < 4; j++)
2357 out[3][j] = in[swizzle_a][j];
2362 sample_swizzle(struct tgsi_sampler *tgsi_sampler,
2363 const float s[QUAD_SIZE],
2364 const float t[QUAD_SIZE],
2365 const float p[QUAD_SIZE],
2366 const float c0[QUAD_SIZE],
2367 enum tgsi_sampler_control control,
2368 float rgba[NUM_CHANNELS][QUAD_SIZE])
2370 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2371 float rgba_temp[NUM_CHANNELS][QUAD_SIZE];
2373 samp->sample_target(tgsi_sampler, s, t, p, c0, control, rgba_temp);
2375 do_swizzling(samp, rgba_temp, rgba);
2379 static wrap_nearest_func
2380 get_nearest_unorm_wrap(unsigned mode)
2383 case PIPE_TEX_WRAP_CLAMP:
2384 return wrap_nearest_unorm_clamp;
2385 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2386 return wrap_nearest_unorm_clamp_to_edge;
2387 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2388 return wrap_nearest_unorm_clamp_to_border;
2391 return wrap_nearest_unorm_clamp;
2396 static wrap_nearest_func
2397 get_nearest_wrap(unsigned mode)
2400 case PIPE_TEX_WRAP_REPEAT:
2401 return wrap_nearest_repeat;
2402 case PIPE_TEX_WRAP_CLAMP:
2403 return wrap_nearest_clamp;
2404 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2405 return wrap_nearest_clamp_to_edge;
2406 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2407 return wrap_nearest_clamp_to_border;
2408 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2409 return wrap_nearest_mirror_repeat;
2410 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2411 return wrap_nearest_mirror_clamp;
2412 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2413 return wrap_nearest_mirror_clamp_to_edge;
2414 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2415 return wrap_nearest_mirror_clamp_to_border;
2418 return wrap_nearest_repeat;
2423 static wrap_linear_func
2424 get_linear_unorm_wrap(unsigned mode)
2427 case PIPE_TEX_WRAP_CLAMP:
2428 return wrap_linear_unorm_clamp;
2429 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2430 return wrap_linear_unorm_clamp_to_edge;
2431 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2432 return wrap_linear_unorm_clamp_to_border;
2435 return wrap_linear_unorm_clamp;
2440 static wrap_linear_func
2441 get_linear_wrap(unsigned mode)
2444 case PIPE_TEX_WRAP_REPEAT:
2445 return wrap_linear_repeat;
2446 case PIPE_TEX_WRAP_CLAMP:
2447 return wrap_linear_clamp;
2448 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
2449 return wrap_linear_clamp_to_edge;
2450 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
2451 return wrap_linear_clamp_to_border;
2452 case PIPE_TEX_WRAP_MIRROR_REPEAT:
2453 return wrap_linear_mirror_repeat;
2454 case PIPE_TEX_WRAP_MIRROR_CLAMP:
2455 return wrap_linear_mirror_clamp;
2456 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
2457 return wrap_linear_mirror_clamp_to_edge;
2458 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
2459 return wrap_linear_mirror_clamp_to_border;
2462 return wrap_linear_repeat;
2467 static compute_lambda_func
2468 get_lambda_func(const union sp_sampler_key key)
2470 if (key.bits.processor == TGSI_PROCESSOR_VERTEX)
2471 return compute_lambda_vert;
2473 switch (key.bits.target) {
2474 case PIPE_TEXTURE_1D:
2475 case PIPE_TEXTURE_1D_ARRAY:
2476 return compute_lambda_1d;
2477 case PIPE_TEXTURE_2D:
2478 case PIPE_TEXTURE_2D_ARRAY:
2479 case PIPE_TEXTURE_RECT:
2480 case PIPE_TEXTURE_CUBE:
2481 return compute_lambda_2d;
2482 case PIPE_TEXTURE_3D:
2483 return compute_lambda_3d;
2486 return compute_lambda_1d;
2492 get_img_filter(const union sp_sampler_key key,
2494 const struct pipe_sampler_state *sampler)
2496 switch (key.bits.target) {
2497 case PIPE_TEXTURE_1D:
2498 if (filter == PIPE_TEX_FILTER_NEAREST)
2499 return img_filter_1d_nearest;
2501 return img_filter_1d_linear;
2503 case PIPE_TEXTURE_1D_ARRAY:
2504 if (filter == PIPE_TEX_FILTER_NEAREST)
2505 return img_filter_1d_array_nearest;
2507 return img_filter_1d_array_linear;
2509 case PIPE_TEXTURE_2D:
2510 case PIPE_TEXTURE_RECT:
2511 /* Try for fast path:
2513 if (key.bits.is_pot &&
2514 sampler->wrap_s == sampler->wrap_t &&
2515 sampler->normalized_coords)
2517 switch (sampler->wrap_s) {
2518 case PIPE_TEX_WRAP_REPEAT:
2520 case PIPE_TEX_FILTER_NEAREST:
2521 return img_filter_2d_nearest_repeat_POT;
2522 case PIPE_TEX_FILTER_LINEAR:
2523 return img_filter_2d_linear_repeat_POT;
2528 case PIPE_TEX_WRAP_CLAMP:
2530 case PIPE_TEX_FILTER_NEAREST:
2531 return img_filter_2d_nearest_clamp_POT;
2537 /* Otherwise use default versions:
2539 if (filter == PIPE_TEX_FILTER_NEAREST)
2540 return img_filter_2d_nearest;
2542 return img_filter_2d_linear;
2544 case PIPE_TEXTURE_2D_ARRAY:
2545 if (filter == PIPE_TEX_FILTER_NEAREST)
2546 return img_filter_2d_array_nearest;
2548 return img_filter_2d_array_linear;
2550 case PIPE_TEXTURE_CUBE:
2551 if (filter == PIPE_TEX_FILTER_NEAREST)
2552 return img_filter_cube_nearest;
2554 return img_filter_cube_linear;
2556 case PIPE_TEXTURE_3D:
2557 if (filter == PIPE_TEX_FILTER_NEAREST)
2558 return img_filter_3d_nearest;
2560 return img_filter_3d_linear;
2564 return img_filter_1d_nearest;
2570 * Bind the given texture object and texture cache to the sampler variant.
2573 sp_sampler_variant_bind_view( struct sp_sampler_variant *samp,
2574 struct softpipe_tex_tile_cache *tex_cache,
2575 const struct pipe_sampler_view *view )
2577 const struct pipe_resource *texture = view->texture;
2580 samp->cache = tex_cache;
2581 samp->xpot = util_logbase2( texture->width0 );
2582 samp->ypot = util_logbase2( texture->height0 );
2583 samp->level = view->u.tex.first_level;
2588 sp_sampler_variant_destroy( struct sp_sampler_variant *samp )
2594 sample_get_dims(struct tgsi_sampler *tgsi_sampler, int level,
2597 struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2598 const struct pipe_sampler_view *view = samp->view;
2599 const struct pipe_resource *texture = view->texture;
2601 /* undefined according to EXT_gpu_program */
2602 level += view->u.tex.first_level;
2603 if (level > view->u.tex.last_level)
2606 dims[0] = u_minify(texture->width0, level);
2608 switch(texture->target) {
2609 case PIPE_TEXTURE_1D_ARRAY:
2610 dims[1] = texture->array_size;
2612 case PIPE_TEXTURE_1D:
2615 case PIPE_TEXTURE_2D_ARRAY:
2616 dims[2] = texture->array_size;
2618 case PIPE_TEXTURE_2D:
2619 case PIPE_TEXTURE_CUBE:
2620 case PIPE_TEXTURE_RECT:
2621 dims[1] = u_minify(texture->height0, level);
2623 case PIPE_TEXTURE_3D:
2624 dims[1] = u_minify(texture->height0, level);
2625 dims[2] = u_minify(texture->depth0, level);
2628 assert(!"unexpected texture target in sample_get_dims()");
2633 /* this function is only used for unfiltered texel gets
2634 via the TGSI TXF opcode. */
2636 sample_get_texels(struct tgsi_sampler *tgsi_sampler,
2637 const int v_i[QUAD_SIZE],
2638 const int v_j[QUAD_SIZE],
2639 const int v_k[QUAD_SIZE],
2640 const int lod[QUAD_SIZE],
2641 const int8_t offset[3],
2642 float rgba[NUM_CHANNELS][QUAD_SIZE])
2644 const struct sp_sampler_variant *samp = sp_sampler_variant(tgsi_sampler);
2645 union tex_tile_address addr;
2646 const struct pipe_resource *texture = samp->view->texture;
2649 bool need_swizzle = (samp->key.bits.swizzle_r != PIPE_SWIZZLE_RED ||
2650 samp->key.bits.swizzle_g != PIPE_SWIZZLE_GREEN ||
2651 samp->key.bits.swizzle_b != PIPE_SWIZZLE_BLUE ||
2652 samp->key.bits.swizzle_a != PIPE_SWIZZLE_ALPHA);
2655 /* TODO write a better test for LOD */
2656 addr.bits.level = lod[0];
2658 switch(texture->target) {
2659 case PIPE_TEXTURE_1D:
2660 for (j = 0; j < QUAD_SIZE; j++) {
2661 tx = get_texel_2d(samp, addr, v_i[j] + offset[0], 0);
2662 for (c = 0; c < 4; c++) {
2667 case PIPE_TEXTURE_1D_ARRAY:
2668 for (j = 0; j < QUAD_SIZE; j++) {
2669 tx = get_texel_1d_array(samp, addr, v_i[j] + offset[0],
2670 v_j[j] + offset[1]);
2671 for (c = 0; c < 4; c++) {
2676 case PIPE_TEXTURE_2D:
2677 case PIPE_TEXTURE_RECT:
2678 for (j = 0; j < QUAD_SIZE; j++) {
2679 tx = get_texel_2d(samp, addr, v_i[j] + offset[0],
2680 v_j[j] + offset[1]);
2681 for (c = 0; c < 4; c++) {
2686 case PIPE_TEXTURE_2D_ARRAY:
2687 for (j = 0; j < QUAD_SIZE; j++) {
2688 tx = get_texel_2d_array(samp, addr, v_i[j] + offset[0],
2690 v_k[j] + offset[2]);
2691 for (c = 0; c < 4; c++) {
2696 case PIPE_TEXTURE_3D:
2697 for (j = 0; j < QUAD_SIZE; j++) {
2698 tx = get_texel_3d(samp, addr, v_i[j] + offset[0],
2700 v_k[j] + offset[2]);
2701 for (c = 0; c < 4; c++) {
2706 case PIPE_TEXTURE_CUBE: /* TXF can't work on CUBE according to spec */
2708 assert(!"Unknown or CUBE texture type in TXF processing\n");
2713 float rgba_temp[NUM_CHANNELS][QUAD_SIZE];
2714 memcpy(rgba_temp, rgba, sizeof(rgba_temp));
2715 do_swizzling(samp, rgba_temp, rgba);
2719 * Create a sampler variant for a given set of non-orthogonal state.
2721 struct sp_sampler_variant *
2722 sp_create_sampler_variant( const struct pipe_sampler_state *sampler,
2723 const union sp_sampler_key key )
2725 struct sp_sampler_variant *samp = CALLOC_STRUCT(sp_sampler_variant);
2729 samp->sampler = sampler;
2732 /* Note that (for instance) linear_texcoord_s and
2733 * nearest_texcoord_s may be active at the same time, if the
2734 * sampler min_img_filter differs from its mag_img_filter.
2736 if (sampler->normalized_coords) {
2737 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
2738 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
2739 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
2741 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
2742 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
2743 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
2746 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
2747 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
2748 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
2750 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
2751 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
2752 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
2755 samp->compute_lambda = get_lambda_func( key );
2757 samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
2758 samp->mag_img_filter = get_img_filter(key, sampler->mag_img_filter, sampler);
2760 switch (sampler->min_mip_filter) {
2761 case PIPE_TEX_MIPFILTER_NONE:
2762 if (sampler->min_img_filter == sampler->mag_img_filter)
2763 samp->mip_filter = samp->min_img_filter;
2765 samp->mip_filter = mip_filter_none;
2768 case PIPE_TEX_MIPFILTER_NEAREST:
2769 samp->mip_filter = mip_filter_nearest;
2772 case PIPE_TEX_MIPFILTER_LINEAR:
2773 if (key.bits.is_pot &&
2774 sampler->min_img_filter == sampler->mag_img_filter &&
2775 sampler->normalized_coords &&
2776 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
2777 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
2778 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR) {
2779 samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT;
2782 samp->mip_filter = mip_filter_linear;
2785 /* Anisotropic filtering extension. */
2786 if (sampler->max_anisotropy > 1) {
2787 samp->mip_filter = mip_filter_linear_aniso;
2789 /* Override min_img_filter:
2790 * min_img_filter needs to be set to NEAREST since we need to access
2791 * each texture pixel as it is and weight it later; using linear
2792 * filters will have incorrect results.
2793 * By setting the filter to NEAREST here, we can avoid calling the
2794 * generic img_filter_2d_nearest in the anisotropic filter function,
2795 * making it possible to use one of the accelerated implementations
2797 samp->min_img_filter = get_img_filter(key, PIPE_TEX_FILTER_NEAREST, sampler);
2799 /* on first access create the lookup table containing the filter weights. */
2801 create_filter_table();
2808 if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
2809 samp->compare = sample_compare;
2812 /* Skip compare operation by promoting the mip_filter function
2815 samp->compare = samp->mip_filter;
2818 if (key.bits.target == PIPE_TEXTURE_CUBE) {
2819 samp->sample_target = sample_cube;
2827 /* Skip cube face determination by promoting the compare
2830 samp->sample_target = samp->compare;
2833 if (key.bits.swizzle_r != PIPE_SWIZZLE_RED ||
2834 key.bits.swizzle_g != PIPE_SWIZZLE_GREEN ||
2835 key.bits.swizzle_b != PIPE_SWIZZLE_BLUE ||
2836 key.bits.swizzle_a != PIPE_SWIZZLE_ALPHA) {
2837 samp->base.get_samples = sample_swizzle;
2840 samp->base.get_samples = samp->sample_target;
2843 samp->base.get_dims = sample_get_dims;
2844 samp->base.get_texel = sample_get_texels;