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"
49 * Return fractional part of 'f'. Used for computing interpolation weights.
50 * Need to be careful with negative values.
51 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
52 * of improperly weighted linear-filtered textures.
53 * The tests/texwrap.c demo is a good test.
64 * Linear interpolation macro
67 lerp(float a, float v0, float v1)
69 return v0 + a * (v1 - v0);
74 * Do 2D/biliner interpolation of float values.
75 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
76 * a and b are the horizontal and vertical interpolants.
77 * It's important that this function is inlined when compiled with
78 * optimization! If we find that's not true on some systems, convert
82 lerp_2d(float a, float b,
83 float v00, float v10, float v01, float v11)
85 const float temp0 = lerp(a, v00, v10);
86 const float temp1 = lerp(a, v01, v11);
87 return lerp(b, temp0, temp1);
92 * As above, but 3D interpolation of 8 values.
95 lerp_3d(float a, float b, float c,
96 float v000, float v100, float v010, float v110,
97 float v001, float v101, float v011, float v111)
99 const float temp0 = lerp_2d(a, b, v000, v100, v010, v110);
100 const float temp1 = lerp_2d(a, b, v001, v101, v011, v111);
101 return lerp(c, temp0, temp1);
107 * Compute coord % size for repeat wrap modes.
108 * Note that if coord is a signed integer, coord % size doesn't give
109 * the right value for coord < 0 (in terms of texture repeat). Just
110 * casting to unsigned fixes that.
113 repeat(int coord, unsigned size)
115 return (int) ((unsigned) coord % size);
120 * Apply texture coord wrapping mode and return integer texture indexes
121 * for a vector of four texcoords (S or T or P).
122 * \param wrapMode PIPE_TEX_WRAP_x
123 * \param s the incoming texcoords
124 * \param size the texture image size
125 * \param icoord returns the integer texcoords
126 * \return integer texture index
129 wrap_nearest_repeat(const float s[4], unsigned size, int icoord[4])
132 /* s limited to [0,1) */
133 /* i limited to [0,size-1] */
134 for (ch = 0; ch < 4; ch++) {
135 int i = util_ifloor(s[ch] * size);
136 icoord[ch] = repeat(i, size);
142 wrap_nearest_clamp(const float s[4], unsigned size, int icoord[4])
145 /* s limited to [0,1] */
146 /* i limited to [0,size-1] */
147 for (ch = 0; ch < 4; ch++) {
150 else if (s[ch] >= 1.0F)
151 icoord[ch] = size - 1;
153 icoord[ch] = util_ifloor(s[ch] * size);
159 wrap_nearest_clamp_to_edge(const float s[4], unsigned size, int icoord[4])
162 /* s limited to [min,max] */
163 /* i limited to [0, size-1] */
164 const float min = 1.0F / (2.0F * size);
165 const float max = 1.0F - min;
166 for (ch = 0; ch < 4; ch++) {
169 else if (s[ch] > max)
170 icoord[ch] = size - 1;
172 icoord[ch] = util_ifloor(s[ch] * size);
178 wrap_nearest_clamp_to_border(const float s[4], unsigned size, int icoord[4])
181 /* s limited to [min,max] */
182 /* i limited to [-1, size] */
183 const float min = -1.0F / (2.0F * size);
184 const float max = 1.0F - min;
185 for (ch = 0; ch < 4; ch++) {
188 else if (s[ch] >= max)
191 icoord[ch] = util_ifloor(s[ch] * size);
197 wrap_nearest_mirror_repeat(const float s[4], unsigned size, int icoord[4])
200 const float min = 1.0F / (2.0F * size);
201 const float max = 1.0F - min;
202 for (ch = 0; ch < 4; ch++) {
203 const int flr = util_ifloor(s[ch]);
204 float u = frac(s[ch]);
210 icoord[ch] = size - 1;
212 icoord[ch] = util_ifloor(u * size);
218 wrap_nearest_mirror_clamp(const float s[4], unsigned size, int icoord[4])
221 for (ch = 0; ch < 4; ch++) {
222 /* s limited to [0,1] */
223 /* i limited to [0,size-1] */
224 const float u = fabsf(s[ch]);
228 icoord[ch] = size - 1;
230 icoord[ch] = util_ifloor(u * size);
236 wrap_nearest_mirror_clamp_to_edge(const float s[4], unsigned size,
240 /* s limited to [min,max] */
241 /* i limited to [0, size-1] */
242 const float min = 1.0F / (2.0F * size);
243 const float max = 1.0F - min;
244 for (ch = 0; ch < 4; ch++) {
245 const float u = fabsf(s[ch]);
249 icoord[ch] = size - 1;
251 icoord[ch] = util_ifloor(u * size);
257 wrap_nearest_mirror_clamp_to_border(const float s[4], unsigned size,
261 /* s limited to [min,max] */
262 /* i limited to [0, size-1] */
263 const float min = -1.0F / (2.0F * size);
264 const float max = 1.0F - min;
265 for (ch = 0; ch < 4; ch++) {
266 const float u = fabsf(s[ch]);
272 icoord[ch] = util_ifloor(u * size);
278 * Used to compute texel locations for linear sampling for four texcoords.
279 * \param wrapMode PIPE_TEX_WRAP_x
280 * \param s the texcoords
281 * \param size the texture image size
282 * \param icoord0 returns first texture indexes
283 * \param icoord1 returns second texture indexes (usually icoord0 + 1)
284 * \param w returns blend factor/weight between texture indexes
285 * \param icoord returns the computed integer texture coords
288 wrap_linear_repeat(const float s[4], unsigned size,
289 int icoord0[4], int icoord1[4], float w[4])
292 for (ch = 0; ch < 4; ch++) {
293 float u = s[ch] * size - 0.5F;
294 icoord0[ch] = repeat(util_ifloor(u), size);
295 icoord1[ch] = repeat(icoord0[ch] + 1, size);
302 wrap_linear_clamp(const float s[4], unsigned size,
303 int icoord0[4], int icoord1[4], float w[4])
306 for (ch = 0; ch < 4; ch++) {
307 float u = CLAMP(s[ch], 0.0F, 1.0F);
309 icoord0[ch] = util_ifloor(u);
310 icoord1[ch] = icoord0[ch] + 1;
317 wrap_linear_clamp_to_edge(const float s[4], unsigned size,
318 int icoord0[4], int icoord1[4], float w[4])
321 for (ch = 0; ch < 4; ch++) {
322 float u = CLAMP(s[ch], 0.0F, 1.0F);
324 icoord0[ch] = util_ifloor(u);
325 icoord1[ch] = icoord0[ch] + 1;
328 if (icoord1[ch] >= (int) size)
329 icoord1[ch] = size - 1;
336 wrap_linear_clamp_to_border(const float s[4], unsigned size,
337 int icoord0[4], int icoord1[4], float w[4])
339 const float min = -1.0F / (2.0F * size);
340 const float max = 1.0F - min;
342 for (ch = 0; ch < 4; ch++) {
343 float u = CLAMP(s[ch], min, max);
345 icoord0[ch] = util_ifloor(u);
346 icoord1[ch] = icoord0[ch] + 1;
353 wrap_linear_mirror_repeat(const float s[4], unsigned size,
354 int icoord0[4], int icoord1[4], float w[4])
357 for (ch = 0; ch < 4; ch++) {
358 const int flr = util_ifloor(s[ch]);
359 float u = frac(s[ch]);
363 icoord0[ch] = util_ifloor(u);
364 icoord1[ch] = icoord0[ch] + 1;
367 if (icoord1[ch] >= (int) size)
368 icoord1[ch] = size - 1;
375 wrap_linear_mirror_clamp(const float s[4], unsigned size,
376 int icoord0[4], int icoord1[4], float w[4])
379 for (ch = 0; ch < 4; ch++) {
380 float u = fabsf(s[ch]);
386 icoord0[ch] = util_ifloor(u);
387 icoord1[ch] = icoord0[ch] + 1;
394 wrap_linear_mirror_clamp_to_edge(const float s[4], unsigned size,
395 int icoord0[4], int icoord1[4], float w[4])
398 for (ch = 0; ch < 4; ch++) {
399 float u = fabsf(s[ch]);
405 icoord0[ch] = util_ifloor(u);
406 icoord1[ch] = icoord0[ch] + 1;
409 if (icoord1[ch] >= (int) size)
410 icoord1[ch] = size - 1;
417 wrap_linear_mirror_clamp_to_border(const float s[4], unsigned size,
418 int icoord0[4], int icoord1[4], float w[4])
420 const float min = -1.0F / (2.0F * size);
421 const float max = 1.0F - min;
423 for (ch = 0; ch < 4; ch++) {
424 float u = fabsf(s[ch]);
432 icoord0[ch] = util_ifloor(u);
433 icoord1[ch] = icoord0[ch] + 1;
440 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
443 wrap_nearest_unorm_clamp(const float s[4], unsigned size, int icoord[4])
446 for (ch = 0; ch < 4; ch++) {
447 int i = util_ifloor(s[ch]);
448 icoord[ch]= CLAMP(i, 0, (int) size-1);
454 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
457 wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size,
461 for (ch = 0; ch < 4; ch++) {
462 icoord[ch]= util_ifloor( CLAMP(s[ch], -0.5F, (float) size + 0.5F) );
468 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
471 wrap_nearest_unorm_clamp_to_edge(const float s[4], unsigned size,
475 for (ch = 0; ch < 4; ch++) {
476 icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
482 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
485 wrap_linear_unorm_clamp(const float s[4], unsigned size,
486 int icoord0[4], int icoord1[4], float w[4])
489 for (ch = 0; ch < 4; ch++) {
490 /* Not exactly what the spec says, but it matches NVIDIA output */
491 float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
492 icoord0[ch] = util_ifloor(u);
493 icoord1[ch] = icoord0[ch] + 1;
500 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
503 wrap_linear_unorm_clamp_to_border(const float s[4], unsigned size,
504 int icoord0[4], int icoord1[4], float w[4])
507 for (ch = 0; ch < 4; ch++) {
508 float u = CLAMP(s[ch], -0.5F, (float) size + 0.5F);
510 icoord0[ch] = util_ifloor(u);
511 icoord1[ch] = icoord0[ch] + 1;
512 if (icoord1[ch] > (int) size - 1)
513 icoord1[ch] = size - 1;
520 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
523 wrap_linear_unorm_clamp_to_edge(const float s[4], unsigned size,
524 int icoord0[4], int icoord1[4], float w[4])
527 for (ch = 0; ch < 4; ch++) {
528 float u = CLAMP(s[ch], +0.5F, (float) size - 0.5F);
530 icoord0[ch] = util_ifloor(u);
531 icoord1[ch] = icoord0[ch] + 1;
532 if (icoord1[ch] > (int) size - 1)
533 icoord1[ch] = size - 1;
541 * Examine the quad's texture coordinates to compute the partial
542 * derivatives w.r.t X and Y, then compute lambda (level of detail).
545 compute_lambda_1d(const struct sp_sampler_varient *samp,
546 const float s[QUAD_SIZE],
547 const float t[QUAD_SIZE],
548 const float p[QUAD_SIZE])
550 const struct pipe_texture *texture = samp->texture;
551 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
552 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
553 float rho = MAX2(dsdx, dsdy) * texture->width0;
555 return util_fast_log2(rho);
560 compute_lambda_2d(const struct sp_sampler_varient *samp,
561 const float s[QUAD_SIZE],
562 const float t[QUAD_SIZE],
563 const float p[QUAD_SIZE])
565 const struct pipe_texture *texture = samp->texture;
566 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
567 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
568 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
569 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
570 float maxx = MAX2(dsdx, dsdy) * texture->width0;
571 float maxy = MAX2(dtdx, dtdy) * texture->height0;
572 float rho = MAX2(maxx, maxy);
574 return util_fast_log2(rho);
579 compute_lambda_3d(const struct sp_sampler_varient *samp,
580 const float s[QUAD_SIZE],
581 const float t[QUAD_SIZE],
582 const float p[QUAD_SIZE])
584 const struct pipe_texture *texture = samp->texture;
585 float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
586 float dsdy = fabsf(s[QUAD_TOP_LEFT] - s[QUAD_BOTTOM_LEFT]);
587 float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
588 float dtdy = fabsf(t[QUAD_TOP_LEFT] - t[QUAD_BOTTOM_LEFT]);
589 float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
590 float dpdy = fabsf(p[QUAD_TOP_LEFT] - p[QUAD_BOTTOM_LEFT]);
591 float maxx = MAX2(dsdx, dsdy) * texture->width0;
592 float maxy = MAX2(dtdx, dtdy) * texture->height0;
593 float maxz = MAX2(dpdx, dpdy) * texture->depth0;
596 rho = MAX2(maxx, maxy);
597 rho = MAX2(rho, maxz);
599 return util_fast_log2(rho);
604 * Compute lambda for a vertex texture sampler.
605 * Since there aren't derivatives to use, just return 0.
608 compute_lambda_vert(const struct sp_sampler_varient *samp,
609 const float s[QUAD_SIZE],
610 const float t[QUAD_SIZE],
611 const float p[QUAD_SIZE])
619 * Get a texel from a texture, using the texture tile cache.
621 * \param addr the template tex address containing cube, z, face info.
622 * \param x the x coord of texel within 2D image
623 * \param y the y coord of texel within 2D image
624 * \param rgba the quad to put the texel/color into
626 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
627 * sp_get_cached_tile_tex() function. Also, get 4 texels instead of 1...
633 static INLINE const float *
634 get_texel_2d_no_border(const struct sp_sampler_varient *samp,
635 union tex_tile_address addr, int x, int y)
637 const struct softpipe_tex_cached_tile *tile;
639 addr.bits.x = x / TILE_SIZE;
640 addr.bits.y = y / TILE_SIZE;
644 tile = sp_get_cached_tile_tex(samp->cache, addr);
646 return &tile->data.color[y][x][0];
650 static INLINE const float *
651 get_texel_2d(const struct sp_sampler_varient *samp,
652 union tex_tile_address addr, int x, int y)
654 const struct pipe_texture *texture = samp->texture;
655 unsigned level = addr.bits.level;
657 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
658 y < 0 || y >= (int) u_minify(texture->height0, level)) {
659 return samp->sampler->border_color;
662 return get_texel_2d_no_border( samp, addr, x, y );
667 /* Gather a quad of adjacent texels within a tile:
670 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_varient *samp,
671 union tex_tile_address addr,
672 unsigned x, unsigned y,
675 const struct softpipe_tex_cached_tile *tile;
677 addr.bits.x = x / TILE_SIZE;
678 addr.bits.y = y / TILE_SIZE;
682 tile = sp_get_cached_tile_tex(samp->cache, addr);
684 out[0] = &tile->data.color[y ][x ][0];
685 out[1] = &tile->data.color[y ][x+1][0];
686 out[2] = &tile->data.color[y+1][x ][0];
687 out[3] = &tile->data.color[y+1][x+1][0];
691 /* Gather a quad of potentially non-adjacent texels:
694 get_texel_quad_2d_no_border(const struct sp_sampler_varient *samp,
695 union tex_tile_address addr,
700 out[0] = get_texel_2d_no_border( samp, addr, x0, y0 );
701 out[1] = get_texel_2d_no_border( samp, addr, x1, y0 );
702 out[2] = get_texel_2d_no_border( samp, addr, x0, y1 );
703 out[3] = get_texel_2d_no_border( samp, addr, x1, y1 );
706 /* Can involve a lot of unnecessary checks for border color:
709 get_texel_quad_2d(const struct sp_sampler_varient *samp,
710 union tex_tile_address addr,
715 out[0] = get_texel_2d( samp, addr, x0, y0 );
716 out[1] = get_texel_2d( samp, addr, x1, y0 );
717 out[3] = get_texel_2d( samp, addr, x1, y1 );
718 out[2] = get_texel_2d( samp, addr, x0, y1 );
725 static INLINE const float *
726 get_texel_3d_no_border(const struct sp_sampler_varient *samp,
727 union tex_tile_address addr, int x, int y, int z)
729 const struct softpipe_tex_cached_tile *tile;
731 addr.bits.x = x / TILE_SIZE;
732 addr.bits.y = y / TILE_SIZE;
737 tile = sp_get_cached_tile_tex(samp->cache, addr);
739 return &tile->data.color[y][x][0];
743 static INLINE const float *
744 get_texel_3d(const struct sp_sampler_varient *samp,
745 union tex_tile_address addr, int x, int y, int z)
747 const struct pipe_texture *texture = samp->texture;
748 unsigned level = addr.bits.level;
750 if (x < 0 || x >= (int) u_minify(texture->width0, level) ||
751 y < 0 || y >= (int) u_minify(texture->height0, level) ||
752 z < 0 || z >= (int) u_minify(texture->depth0, level)) {
753 return samp->sampler->border_color;
756 return get_texel_3d_no_border( samp, addr, x, y, z );
762 * Given the logbase2 of a mipmap's base level size and a mipmap level,
763 * return the size (in texels) of that mipmap level.
764 * For example, if level[0].width = 256 then base_pot will be 8.
765 * If level = 2, then we'll return 64 (the width at level=2).
766 * Return 1 if level > base_pot.
768 static INLINE unsigned
769 pot_level_size(unsigned base_pot, unsigned level)
771 return (base_pot >= level) ? (1 << (base_pot - level)) : 1;
775 /* Some image-filter fastpaths:
778 img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
779 const float s[QUAD_SIZE],
780 const float t[QUAD_SIZE],
781 const float p[QUAD_SIZE],
782 const float c0[QUAD_SIZE],
783 enum tgsi_sampler_control control,
784 float rgba[NUM_CHANNELS][QUAD_SIZE])
786 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
788 unsigned level = samp->level;
789 unsigned xpot = pot_level_size(samp->xpot, level);
790 unsigned ypot = pot_level_size(samp->ypot, level);
791 unsigned xmax = (xpot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, xpot) - 1; */
792 unsigned ymax = (ypot - 1) & (TILE_SIZE - 1); /* MIN2(TILE_SIZE, ypot) - 1; */
793 union tex_tile_address addr;
796 addr.bits.level = samp->level;
798 for (j = 0; j < QUAD_SIZE; j++) {
801 float u = s[j] * xpot - 0.5F;
802 float v = t[j] * ypot - 0.5F;
804 int uflr = util_ifloor(u);
805 int vflr = util_ifloor(v);
807 float xw = u - (float)uflr;
808 float yw = v - (float)vflr;
810 int x0 = uflr & (xpot - 1);
811 int y0 = vflr & (ypot - 1);
815 /* Can we fetch all four at once:
817 if (x0 < xmax && y0 < ymax) {
818 get_texel_quad_2d_no_border_single_tile(samp, addr, x0, y0, tx);
821 unsigned x1 = (x0 + 1) & (xpot - 1);
822 unsigned y1 = (y0 + 1) & (ypot - 1);
823 get_texel_quad_2d_no_border(samp, addr, x0, y0, x1, y1, tx);
826 /* interpolate R, G, B, A */
827 for (c = 0; c < 4; c++) {
828 rgba[c][j] = lerp_2d(xw, yw,
837 img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
838 const float s[QUAD_SIZE],
839 const float t[QUAD_SIZE],
840 const float p[QUAD_SIZE],
841 const float c0[QUAD_SIZE],
842 enum tgsi_sampler_control control,
843 float rgba[NUM_CHANNELS][QUAD_SIZE])
845 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
847 unsigned level = samp->level;
848 unsigned xpot = pot_level_size(samp->xpot, level);
849 unsigned ypot = pot_level_size(samp->ypot, level);
850 union tex_tile_address addr;
853 addr.bits.level = samp->level;
855 for (j = 0; j < QUAD_SIZE; j++) {
858 float u = s[j] * xpot;
859 float v = t[j] * ypot;
861 int uflr = util_ifloor(u);
862 int vflr = util_ifloor(v);
864 int x0 = uflr & (xpot - 1);
865 int y0 = vflr & (ypot - 1);
867 const float *out = get_texel_2d_no_border(samp, addr, x0, y0);
869 for (c = 0; c < 4; c++) {
877 img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
878 const float s[QUAD_SIZE],
879 const float t[QUAD_SIZE],
880 const float p[QUAD_SIZE],
881 const float c0[QUAD_SIZE],
882 enum tgsi_sampler_control control,
883 float rgba[NUM_CHANNELS][QUAD_SIZE])
885 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
887 unsigned level = samp->level;
888 unsigned xpot = pot_level_size(samp->xpot, level);
889 unsigned ypot = pot_level_size(samp->ypot, level);
890 union tex_tile_address addr;
893 addr.bits.level = samp->level;
895 for (j = 0; j < QUAD_SIZE; j++) {
898 float u = s[j] * xpot;
899 float v = t[j] * ypot;
907 else if (x0 > xpot - 1)
913 else if (y0 > ypot - 1)
916 out = get_texel_2d_no_border(samp, addr, x0, y0);
918 for (c = 0; c < 4; c++) {
926 img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
927 const float s[QUAD_SIZE],
928 const float t[QUAD_SIZE],
929 const float p[QUAD_SIZE],
930 const float c0[QUAD_SIZE],
931 enum tgsi_sampler_control control,
932 float rgba[NUM_CHANNELS][QUAD_SIZE])
934 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
935 const struct pipe_texture *texture = samp->texture;
939 union tex_tile_address addr;
941 level0 = samp->level;
942 width = u_minify(texture->width0, level0);
947 addr.bits.level = samp->level;
949 samp->nearest_texcoord_s(s, width, x);
951 for (j = 0; j < QUAD_SIZE; j++) {
952 const float *out = get_texel_2d(samp, addr, x[j], 0);
954 for (c = 0; c < 4; c++) {
962 img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
963 const float s[QUAD_SIZE],
964 const float t[QUAD_SIZE],
965 const float p[QUAD_SIZE],
966 const float c0[QUAD_SIZE],
967 enum tgsi_sampler_control control,
968 float rgba[NUM_CHANNELS][QUAD_SIZE])
970 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
971 const struct pipe_texture *texture = samp->texture;
975 union tex_tile_address addr;
978 level0 = samp->level;
979 width = u_minify(texture->width0, level0);
980 height = u_minify(texture->height0, level0);
986 addr.bits.level = samp->level;
988 samp->nearest_texcoord_s(s, width, x);
989 samp->nearest_texcoord_t(t, height, y);
991 for (j = 0; j < QUAD_SIZE; j++) {
992 const float *out = get_texel_2d(samp, addr, x[j], y[j]);
994 for (c = 0; c < 4; c++) {
1001 static INLINE union tex_tile_address
1002 face(union tex_tile_address addr, unsigned face )
1004 addr.bits.face = face;
1010 img_filter_cube_nearest(struct tgsi_sampler *tgsi_sampler,
1011 const float s[QUAD_SIZE],
1012 const float t[QUAD_SIZE],
1013 const float p[QUAD_SIZE],
1014 const float c0[QUAD_SIZE],
1015 enum tgsi_sampler_control control,
1016 float rgba[NUM_CHANNELS][QUAD_SIZE])
1018 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1019 const struct pipe_texture *texture = samp->texture;
1020 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
1024 union tex_tile_address addr;
1026 level0 = samp->level;
1027 width = u_minify(texture->width0, level0);
1028 height = u_minify(texture->height0, level0);
1034 addr.bits.level = samp->level;
1036 samp->nearest_texcoord_s(s, width, x);
1037 samp->nearest_texcoord_t(t, height, y);
1039 for (j = 0; j < QUAD_SIZE; j++) {
1040 const float *out = get_texel_2d(samp, face(addr, faces[j]), x[j], y[j]);
1042 for (c = 0; c < 4; c++) {
1043 rgba[c][j] = out[c];
1050 img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
1051 const float s[QUAD_SIZE],
1052 const float t[QUAD_SIZE],
1053 const float p[QUAD_SIZE],
1054 const float c0[QUAD_SIZE],
1055 enum tgsi_sampler_control control,
1056 float rgba[NUM_CHANNELS][QUAD_SIZE])
1058 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1059 const struct pipe_texture *texture = samp->texture;
1061 int width, height, depth;
1062 int x[4], y[4], z[4];
1063 union tex_tile_address addr;
1065 level0 = samp->level;
1066 width = u_minify(texture->width0, level0);
1067 height = u_minify(texture->height0, level0);
1068 depth = u_minify(texture->depth0, level0);
1074 samp->nearest_texcoord_s(s, width, x);
1075 samp->nearest_texcoord_t(t, height, y);
1076 samp->nearest_texcoord_p(p, depth, z);
1079 addr.bits.level = samp->level;
1081 for (j = 0; j < QUAD_SIZE; j++) {
1082 const float *out = get_texel_3d(samp, addr, x[j], y[j], z[j]);
1084 for (c = 0; c < 4; c++) {
1085 rgba[c][j] = out[c];
1092 img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
1093 const float s[QUAD_SIZE],
1094 const float t[QUAD_SIZE],
1095 const float p[QUAD_SIZE],
1096 const float c0[QUAD_SIZE],
1097 enum tgsi_sampler_control control,
1098 float rgba[NUM_CHANNELS][QUAD_SIZE])
1100 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1101 const struct pipe_texture *texture = samp->texture;
1105 float xw[4]; /* weights */
1106 union tex_tile_address addr;
1108 level0 = samp->level;
1109 width = u_minify(texture->width0, level0);
1114 addr.bits.level = samp->level;
1116 samp->linear_texcoord_s(s, width, x0, x1, xw);
1118 for (j = 0; j < QUAD_SIZE; j++) {
1119 const float *tx0 = get_texel_2d(samp, addr, x0[j], 0);
1120 const float *tx1 = get_texel_2d(samp, addr, x1[j], 0);
1123 /* interpolate R, G, B, A */
1124 for (c = 0; c < 4; c++) {
1125 rgba[c][j] = lerp(xw[j], tx0[c], tx1[c]);
1132 img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
1133 const float s[QUAD_SIZE],
1134 const float t[QUAD_SIZE],
1135 const float p[QUAD_SIZE],
1136 const float c0[QUAD_SIZE],
1137 enum tgsi_sampler_control control,
1138 float rgba[NUM_CHANNELS][QUAD_SIZE])
1140 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1141 const struct pipe_texture *texture = samp->texture;
1144 int x0[4], y0[4], x1[4], y1[4];
1145 float xw[4], yw[4]; /* weights */
1146 union tex_tile_address addr;
1148 level0 = samp->level;
1149 width = u_minify(texture->width0, level0);
1150 height = u_minify(texture->height0, level0);
1156 addr.bits.level = samp->level;
1158 samp->linear_texcoord_s(s, width, x0, x1, xw);
1159 samp->linear_texcoord_t(t, height, y0, y1, yw);
1161 for (j = 0; j < QUAD_SIZE; j++) {
1162 const float *tx0 = get_texel_2d(samp, addr, x0[j], y0[j]);
1163 const float *tx1 = get_texel_2d(samp, addr, x1[j], y0[j]);
1164 const float *tx2 = get_texel_2d(samp, addr, x0[j], y1[j]);
1165 const float *tx3 = get_texel_2d(samp, addr, x1[j], y1[j]);
1168 /* interpolate R, G, B, A */
1169 for (c = 0; c < 4; c++) {
1170 rgba[c][j] = lerp_2d(xw[j], yw[j],
1179 img_filter_cube_linear(struct tgsi_sampler *tgsi_sampler,
1180 const float s[QUAD_SIZE],
1181 const float t[QUAD_SIZE],
1182 const float p[QUAD_SIZE],
1183 const float c0[QUAD_SIZE],
1184 enum tgsi_sampler_control control,
1185 float rgba[NUM_CHANNELS][QUAD_SIZE])
1187 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1188 const struct pipe_texture *texture = samp->texture;
1189 const unsigned *faces = samp->faces; /* zero when not cube-mapping */
1192 int x0[4], y0[4], x1[4], y1[4];
1193 float xw[4], yw[4]; /* weights */
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->linear_texcoord_s(s, width, x0, x1, xw);
1207 samp->linear_texcoord_t(t, height, y0, y1, yw);
1209 for (j = 0; j < QUAD_SIZE; j++) {
1210 union tex_tile_address addrj = face(addr, faces[j]);
1211 const float *tx0 = get_texel_2d(samp, addrj, x0[j], y0[j]);
1212 const float *tx1 = get_texel_2d(samp, addrj, x1[j], y0[j]);
1213 const float *tx2 = get_texel_2d(samp, addrj, x0[j], y1[j]);
1214 const float *tx3 = get_texel_2d(samp, addrj, x1[j], y1[j]);
1217 /* interpolate R, G, B, A */
1218 for (c = 0; c < 4; c++) {
1219 rgba[c][j] = lerp_2d(xw[j], yw[j],
1228 img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
1229 const float s[QUAD_SIZE],
1230 const float t[QUAD_SIZE],
1231 const float p[QUAD_SIZE],
1232 const float c0[QUAD_SIZE],
1233 enum tgsi_sampler_control control,
1234 float rgba[NUM_CHANNELS][QUAD_SIZE])
1236 const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1237 const struct pipe_texture *texture = samp->texture;
1239 int width, height, depth;
1240 int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
1241 float xw[4], yw[4], zw[4]; /* interpolation weights */
1242 union tex_tile_address addr;
1244 level0 = samp->level;
1245 width = u_minify(texture->width0, level0);
1246 height = u_minify(texture->height0, level0);
1247 depth = u_minify(texture->depth0, level0);
1250 addr.bits.level = level0;
1256 samp->linear_texcoord_s(s, width, x0, x1, xw);
1257 samp->linear_texcoord_t(t, height, y0, y1, yw);
1258 samp->linear_texcoord_p(p, depth, z0, z1, zw);
1260 for (j = 0; j < QUAD_SIZE; j++) {
1263 const float *tx00 = get_texel_3d(samp, addr, x0[j], y0[j], z0[j]);
1264 const float *tx01 = get_texel_3d(samp, addr, x1[j], y0[j], z0[j]);
1265 const float *tx02 = get_texel_3d(samp, addr, x0[j], y1[j], z0[j]);
1266 const float *tx03 = get_texel_3d(samp, addr, x1[j], y1[j], z0[j]);
1268 const float *tx10 = get_texel_3d(samp, addr, x0[j], y0[j], z1[j]);
1269 const float *tx11 = get_texel_3d(samp, addr, x1[j], y0[j], z1[j]);
1270 const float *tx12 = get_texel_3d(samp, addr, x0[j], y1[j], z1[j]);
1271 const float *tx13 = get_texel_3d(samp, addr, x1[j], y1[j], z1[j]);
1273 /* interpolate R, G, B, A */
1274 for (c = 0; c < 4; c++) {
1275 rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
1285 /* Calculate level of detail for every fragment.
1286 * Note that lambda has already been biased by global LOD bias.
1289 compute_lod(const struct pipe_sampler_state *sampler,
1290 const float biased_lambda,
1291 const float lodbias[QUAD_SIZE],
1292 float lod[QUAD_SIZE])
1296 for (i = 0; i < QUAD_SIZE; i++) {
1297 lod[i] = biased_lambda + lodbias[i];
1298 lod[i] = CLAMP(lod[i], sampler->min_lod, sampler->max_lod);
1304 mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
1305 const float s[QUAD_SIZE],
1306 const float t[QUAD_SIZE],
1307 const float p[QUAD_SIZE],
1308 const float c0[QUAD_SIZE],
1309 enum tgsi_sampler_control control,
1310 float rgba[NUM_CHANNELS][QUAD_SIZE])
1312 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1313 const struct pipe_texture *texture = samp->texture;
1316 float lod[QUAD_SIZE];
1318 if (control == tgsi_sampler_lod_bias) {
1319 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1320 compute_lod(samp->sampler, lambda, c0, lod);
1322 assert(control == tgsi_sampler_lod_explicit);
1324 memcpy(lod, c0, sizeof(lod));
1327 /* XXX: Take into account all lod values.
1330 level0 = (int)lambda;
1334 samp->mag_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1336 else if (level0 >= texture->last_level) {
1337 samp->level = texture->last_level;
1338 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1341 float levelBlend = lambda - level0;
1346 samp->level = level0;
1347 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba0);
1349 samp->level = level0+1;
1350 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba1);
1352 for (j = 0; j < QUAD_SIZE; j++) {
1353 for (c = 0; c < 4; c++) {
1354 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1362 * Compute nearest mipmap level from texcoords.
1363 * Then sample the texture level for four elements of a quad.
1364 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1367 mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
1368 const float s[QUAD_SIZE],
1369 const float t[QUAD_SIZE],
1370 const float p[QUAD_SIZE],
1371 const float c0[QUAD_SIZE],
1372 enum tgsi_sampler_control control,
1373 float rgba[NUM_CHANNELS][QUAD_SIZE])
1375 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1376 const struct pipe_texture *texture = samp->texture;
1378 float lod[QUAD_SIZE];
1380 if (control == tgsi_sampler_lod_bias) {
1381 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1382 compute_lod(samp->sampler, lambda, c0, lod);
1384 assert(control == tgsi_sampler_lod_explicit);
1386 memcpy(lod, c0, sizeof(lod));
1389 /* XXX: Take into account all lod values.
1395 samp->mag_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1398 samp->level = (int)(lambda + 0.5) ;
1399 samp->level = MIN2(samp->level, (int)texture->last_level);
1400 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1404 printf("RGBA %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
1405 rgba[0][0], rgba[1][0], rgba[2][0], rgba[3][0],
1406 rgba[0][1], rgba[1][1], rgba[2][1], rgba[3][1],
1407 rgba[0][2], rgba[1][2], rgba[2][2], rgba[3][2],
1408 rgba[0][3], rgba[1][3], rgba[2][3], rgba[3][3]);
1414 mip_filter_none(struct tgsi_sampler *tgsi_sampler,
1415 const float s[QUAD_SIZE],
1416 const float t[QUAD_SIZE],
1417 const float p[QUAD_SIZE],
1418 const float c0[QUAD_SIZE],
1419 enum tgsi_sampler_control control,
1420 float rgba[NUM_CHANNELS][QUAD_SIZE])
1422 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1424 float lod[QUAD_SIZE];
1426 if (control == tgsi_sampler_lod_bias) {
1427 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1428 compute_lod(samp->sampler, lambda, c0, lod);
1430 assert(control == tgsi_sampler_lod_explicit);
1432 memcpy(lod, c0, sizeof(lod));
1435 /* XXX: Take into account all lod values.
1440 samp->mag_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1443 samp->min_img_filter(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1450 * Specialized version of mip_filter_linear with hard-wired calls to
1451 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
1454 mip_filter_linear_2d_linear_repeat_POT(
1455 struct tgsi_sampler *tgsi_sampler,
1456 const float s[QUAD_SIZE],
1457 const float t[QUAD_SIZE],
1458 const float p[QUAD_SIZE],
1459 const float c0[QUAD_SIZE],
1460 enum tgsi_sampler_control control,
1461 float rgba[NUM_CHANNELS][QUAD_SIZE])
1463 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1464 const struct pipe_texture *texture = samp->texture;
1467 float lod[QUAD_SIZE];
1469 if (control == tgsi_sampler_lod_bias) {
1470 lambda = samp->compute_lambda(samp, s, t, p) + samp->sampler->lod_bias;
1471 compute_lod(samp->sampler, lambda, c0, lod);
1473 assert(control == tgsi_sampler_lod_explicit);
1475 memcpy(lod, c0, sizeof(lod));
1478 /* XXX: Take into account all lod values.
1481 level0 = (int)lambda;
1483 /* Catches both negative and large values of level0:
1485 if ((unsigned)level0 >= texture->last_level) {
1489 samp->level = texture->last_level;
1491 img_filter_2d_linear_repeat_POT(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba);
1494 float levelBlend = lambda - level0;
1499 samp->level = level0;
1500 img_filter_2d_linear_repeat_POT(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba0);
1502 samp->level = level0+1;
1503 img_filter_2d_linear_repeat_POT(tgsi_sampler, s, t, p, NULL, tgsi_sampler_lod_bias, rgba1);
1505 for (j = 0; j < QUAD_SIZE; j++) {
1506 for (c = 0; c < 4; c++) {
1507 rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
1516 * Do shadow/depth comparisons.
1519 sample_compare(struct tgsi_sampler *tgsi_sampler,
1520 const float s[QUAD_SIZE],
1521 const float t[QUAD_SIZE],
1522 const float p[QUAD_SIZE],
1523 const float c0[QUAD_SIZE],
1524 enum tgsi_sampler_control control,
1525 float rgba[NUM_CHANNELS][QUAD_SIZE])
1527 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1528 const struct pipe_sampler_state *sampler = samp->sampler;
1529 int j, k0, k1, k2, k3;
1532 samp->mip_filter(tgsi_sampler, s, t, p, c0, control, rgba);
1535 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
1536 * When we sampled the depth texture, the depth value was put into all
1537 * RGBA channels. We look at the red channel here.
1540 /* compare four texcoords vs. four texture samples */
1541 switch (sampler->compare_func) {
1542 case PIPE_FUNC_LESS:
1543 k0 = p[0] < rgba[0][0];
1544 k1 = p[1] < rgba[0][1];
1545 k2 = p[2] < rgba[0][2];
1546 k3 = p[3] < rgba[0][3];
1548 case PIPE_FUNC_LEQUAL:
1549 k0 = p[0] <= rgba[0][0];
1550 k1 = p[1] <= rgba[0][1];
1551 k2 = p[2] <= rgba[0][2];
1552 k3 = p[3] <= rgba[0][3];
1554 case PIPE_FUNC_GREATER:
1555 k0 = p[0] > rgba[0][0];
1556 k1 = p[1] > rgba[0][1];
1557 k2 = p[2] > rgba[0][2];
1558 k3 = p[3] > rgba[0][3];
1560 case PIPE_FUNC_GEQUAL:
1561 k0 = p[0] >= rgba[0][0];
1562 k1 = p[1] >= rgba[0][1];
1563 k2 = p[2] >= rgba[0][2];
1564 k3 = p[3] >= rgba[0][3];
1566 case PIPE_FUNC_EQUAL:
1567 k0 = p[0] == rgba[0][0];
1568 k1 = p[1] == rgba[0][1];
1569 k2 = p[2] == rgba[0][2];
1570 k3 = p[3] == rgba[0][3];
1572 case PIPE_FUNC_NOTEQUAL:
1573 k0 = p[0] != rgba[0][0];
1574 k1 = p[1] != rgba[0][1];
1575 k2 = p[2] != rgba[0][2];
1576 k3 = p[3] != rgba[0][3];
1578 case PIPE_FUNC_ALWAYS:
1579 k0 = k1 = k2 = k3 = 1;
1581 case PIPE_FUNC_NEVER:
1582 k0 = k1 = k2 = k3 = 0;
1585 k0 = k1 = k2 = k3 = 0;
1590 /* convert four pass/fail values to an intensity in [0,1] */
1591 val = 0.25F * (k0 + k1 + k2 + k3);
1593 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
1594 for (j = 0; j < 4; j++) {
1595 rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
1602 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
1603 * Put face info into the sampler faces[] array.
1606 sample_cube(struct tgsi_sampler *tgsi_sampler,
1607 const float s[QUAD_SIZE],
1608 const float t[QUAD_SIZE],
1609 const float p[QUAD_SIZE],
1610 const float c0[QUAD_SIZE],
1611 enum tgsi_sampler_control control,
1612 float rgba[NUM_CHANNELS][QUAD_SIZE])
1614 struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
1616 float ssss[4], tttt[4];
1620 direction target sc tc ma
1621 ---------- ------------------------------- --- --- ---
1622 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
1623 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
1624 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
1625 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
1626 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
1627 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
1630 /* Choose the cube face and compute new s/t coords for the 2D face.
1632 * Use the same cube face for all four pixels in the quad.
1634 * This isn't ideal, but if we want to use a different cube face
1635 * per pixel in the quad, we'd have to also compute the per-face
1636 * LOD here too. That's because the four post-face-selection
1637 * texcoords are no longer related to each other (they're
1638 * per-face!) so we can't use subtraction to compute the partial
1639 * deriviates to compute the LOD. Doing so (near cube edges
1640 * anyway) gives us pretty much random values.
1643 /* use the average of the four pixel's texcoords to choose the face */
1644 const float rx = 0.25 * (s[0] + s[1] + s[2] + s[3]);
1645 const float ry = 0.25 * (t[0] + t[1] + t[2] + t[3]);
1646 const float rz = 0.25 * (p[0] + p[1] + p[2] + p[3]);
1647 const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
1649 if (arx >= ary && arx >= arz) {
1650 float sign = (rx >= 0.0F) ? 1.0F : -1.0F;
1651 uint face = (rx >= 0.0F) ? PIPE_TEX_FACE_POS_X : PIPE_TEX_FACE_NEG_X;
1652 for (j = 0; j < QUAD_SIZE; j++) {
1653 const float ima = -0.5F / fabsf(s[j]);
1654 ssss[j] = sign * p[j] * ima + 0.5F;
1655 tttt[j] = t[j] * ima + 0.5F;
1656 samp->faces[j] = face;
1659 else if (ary >= arx && ary >= arz) {
1660 float sign = (ry >= 0.0F) ? 1.0F : -1.0F;
1661 uint face = (ry >= 0.0F) ? PIPE_TEX_FACE_POS_Y : PIPE_TEX_FACE_NEG_Y;
1662 for (j = 0; j < QUAD_SIZE; j++) {
1663 const float ima = -0.5F / fabsf(t[j]);
1664 ssss[j] = -s[j] * ima + 0.5F;
1665 tttt[j] = sign * -p[j] * ima + 0.5F;
1666 samp->faces[j] = face;
1670 float sign = (rz >= 0.0F) ? 1.0F : -1.0F;
1671 uint face = (rz >= 0.0F) ? PIPE_TEX_FACE_POS_Z : PIPE_TEX_FACE_NEG_Z;
1672 for (j = 0; j < QUAD_SIZE; j++) {
1673 const float ima = -0.5 / fabsf(p[j]);
1674 ssss[j] = sign * -s[j] * ima + 0.5F;
1675 tttt[j] = t[j] * ima + 0.5F;
1676 samp->faces[j] = face;
1681 /* In our little pipeline, the compare stage is next. If compare
1682 * is not active, this will point somewhere deeper into the
1683 * pipeline, eg. to mip_filter or even img_filter.
1685 samp->compare(tgsi_sampler, ssss, tttt, NULL, c0, control, rgba);
1690 static wrap_nearest_func
1691 get_nearest_unorm_wrap(unsigned mode)
1694 case PIPE_TEX_WRAP_CLAMP:
1695 return wrap_nearest_unorm_clamp;
1696 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1697 return wrap_nearest_unorm_clamp_to_edge;
1698 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1699 return wrap_nearest_unorm_clamp_to_border;
1702 return wrap_nearest_unorm_clamp;
1707 static wrap_nearest_func
1708 get_nearest_wrap(unsigned mode)
1711 case PIPE_TEX_WRAP_REPEAT:
1712 return wrap_nearest_repeat;
1713 case PIPE_TEX_WRAP_CLAMP:
1714 return wrap_nearest_clamp;
1715 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1716 return wrap_nearest_clamp_to_edge;
1717 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1718 return wrap_nearest_clamp_to_border;
1719 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1720 return wrap_nearest_mirror_repeat;
1721 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1722 return wrap_nearest_mirror_clamp;
1723 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1724 return wrap_nearest_mirror_clamp_to_edge;
1725 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1726 return wrap_nearest_mirror_clamp_to_border;
1729 return wrap_nearest_repeat;
1734 static wrap_linear_func
1735 get_linear_unorm_wrap(unsigned mode)
1738 case PIPE_TEX_WRAP_CLAMP:
1739 return wrap_linear_unorm_clamp;
1740 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1741 return wrap_linear_unorm_clamp_to_edge;
1742 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1743 return wrap_linear_unorm_clamp_to_border;
1746 return wrap_linear_unorm_clamp;
1751 static wrap_linear_func
1752 get_linear_wrap(unsigned mode)
1755 case PIPE_TEX_WRAP_REPEAT:
1756 return wrap_linear_repeat;
1757 case PIPE_TEX_WRAP_CLAMP:
1758 return wrap_linear_clamp;
1759 case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
1760 return wrap_linear_clamp_to_edge;
1761 case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
1762 return wrap_linear_clamp_to_border;
1763 case PIPE_TEX_WRAP_MIRROR_REPEAT:
1764 return wrap_linear_mirror_repeat;
1765 case PIPE_TEX_WRAP_MIRROR_CLAMP:
1766 return wrap_linear_mirror_clamp;
1767 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
1768 return wrap_linear_mirror_clamp_to_edge;
1769 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
1770 return wrap_linear_mirror_clamp_to_border;
1773 return wrap_linear_repeat;
1778 static compute_lambda_func
1779 get_lambda_func(const union sp_sampler_key key)
1781 if (key.bits.processor == TGSI_PROCESSOR_VERTEX)
1782 return compute_lambda_vert;
1784 switch (key.bits.target) {
1785 case PIPE_TEXTURE_1D:
1786 return compute_lambda_1d;
1787 case PIPE_TEXTURE_2D:
1788 case PIPE_TEXTURE_CUBE:
1789 return compute_lambda_2d;
1790 case PIPE_TEXTURE_3D:
1791 return compute_lambda_3d;
1794 return compute_lambda_1d;
1800 get_img_filter(const union sp_sampler_key key,
1802 const struct pipe_sampler_state *sampler)
1804 switch (key.bits.target) {
1805 case PIPE_TEXTURE_1D:
1806 if (filter == PIPE_TEX_FILTER_NEAREST)
1807 return img_filter_1d_nearest;
1809 return img_filter_1d_linear;
1811 case PIPE_TEXTURE_2D:
1812 /* Try for fast path:
1814 if (key.bits.is_pot &&
1815 sampler->wrap_s == sampler->wrap_t &&
1816 sampler->normalized_coords)
1818 switch (sampler->wrap_s) {
1819 case PIPE_TEX_WRAP_REPEAT:
1821 case PIPE_TEX_FILTER_NEAREST:
1822 return img_filter_2d_nearest_repeat_POT;
1823 case PIPE_TEX_FILTER_LINEAR:
1824 return img_filter_2d_linear_repeat_POT;
1829 case PIPE_TEX_WRAP_CLAMP:
1831 case PIPE_TEX_FILTER_NEAREST:
1832 return img_filter_2d_nearest_clamp_POT;
1838 /* Otherwise use default versions:
1840 if (filter == PIPE_TEX_FILTER_NEAREST)
1841 return img_filter_2d_nearest;
1843 return img_filter_2d_linear;
1845 case PIPE_TEXTURE_CUBE:
1846 if (filter == PIPE_TEX_FILTER_NEAREST)
1847 return img_filter_cube_nearest;
1849 return img_filter_cube_linear;
1851 case PIPE_TEXTURE_3D:
1852 if (filter == PIPE_TEX_FILTER_NEAREST)
1853 return img_filter_3d_nearest;
1855 return img_filter_3d_linear;
1859 return img_filter_1d_nearest;
1865 * Bind the given texture object and texture cache to the sampler varient.
1868 sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp,
1869 struct softpipe_tex_tile_cache *tex_cache,
1870 const struct pipe_texture *texture )
1872 const struct pipe_sampler_state *sampler = samp->sampler;
1874 samp->texture = texture;
1875 samp->cache = tex_cache;
1876 samp->xpot = util_unsigned_logbase2( texture->width0 );
1877 samp->ypot = util_unsigned_logbase2( texture->height0 );
1878 samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level);
1883 sp_sampler_varient_destroy( struct sp_sampler_varient *samp )
1890 * Create a sampler varient for a given set of non-orthogonal state.
1892 struct sp_sampler_varient *
1893 sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
1894 const union sp_sampler_key key )
1896 struct sp_sampler_varient *samp = CALLOC_STRUCT(sp_sampler_varient);
1900 samp->sampler = sampler;
1903 /* Note that (for instance) linear_texcoord_s and
1904 * nearest_texcoord_s may be active at the same time, if the
1905 * sampler min_img_filter differs from its mag_img_filter.
1907 if (sampler->normalized_coords) {
1908 samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
1909 samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
1910 samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
1912 samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
1913 samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
1914 samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
1917 samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
1918 samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
1919 samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
1921 samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
1922 samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
1923 samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
1926 samp->compute_lambda = get_lambda_func( key );
1928 samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
1929 samp->mag_img_filter = get_img_filter(key, sampler->mag_img_filter, sampler);
1931 switch (sampler->min_mip_filter) {
1932 case PIPE_TEX_MIPFILTER_NONE:
1933 if (sampler->min_img_filter == sampler->mag_img_filter)
1934 samp->mip_filter = samp->min_img_filter;
1936 samp->mip_filter = mip_filter_none;
1939 case PIPE_TEX_MIPFILTER_NEAREST:
1940 samp->mip_filter = mip_filter_nearest;
1943 case PIPE_TEX_MIPFILTER_LINEAR:
1944 if (key.bits.is_pot &&
1945 sampler->min_img_filter == sampler->mag_img_filter &&
1946 sampler->normalized_coords &&
1947 sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
1948 sampler->wrap_t == PIPE_TEX_WRAP_REPEAT &&
1949 sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR)
1951 samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT;
1955 samp->mip_filter = mip_filter_linear;
1960 if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
1961 samp->compare = sample_compare;
1964 /* Skip compare operation by promoting the mip_filter function
1967 samp->compare = samp->mip_filter;
1970 if (key.bits.target == PIPE_TEXTURE_CUBE) {
1971 samp->base.get_samples = sample_cube;
1979 /* Skip cube face determination by promoting the compare
1982 samp->base.get_samples = samp->compare;