1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2008 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 **************************************************************************/
36 #include "lp_context.h"
38 #include "lp_texture.h"
39 #include "lp_tex_sample.h"
40 #include "lp_tex_cache.h"
41 #include "pipe/p_context.h"
42 #include "pipe/p_defines.h"
43 #include "pipe/p_shader_tokens.h"
44 #include "util/u_math.h"
45 #include "util/u_memory.h"
50 * Note, the FRAC macro has to work perfectly. Otherwise you'll sometimes
51 * see 1-pixel bands of improperly weighted linear-filtered textures.
52 * The tests/texwrap.c demo is a good test.
53 * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0.
54 * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x).
56 #define FRAC(f) ((f) - util_ifloor(f))
60 * Linear interpolation macro
63 lerp(float a
, float v0
, float v1
)
65 return v0
+ a
* (v1
- v0
);
70 * Do 2D/biliner interpolation of float values.
71 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
72 * a and b are the horizontal and vertical interpolants.
73 * It's important that this function is inlined when compiled with
74 * optimization! If we find that's not true on some systems, convert
78 lerp_2d(float a
, float b
,
79 float v00
, float v10
, float v01
, float v11
)
81 const float temp0
= lerp(a
, v00
, v10
);
82 const float temp1
= lerp(a
, v01
, v11
);
83 return lerp(b
, temp0
, temp1
);
88 * As above, but 3D interpolation of 8 values.
91 lerp_3d(float a
, float b
, float c
,
92 float v000
, float v100
, float v010
, float v110
,
93 float v001
, float v101
, float v011
, float v111
)
95 const float temp0
= lerp_2d(a
, b
, v000
, v100
, v010
, v110
);
96 const float temp1
= lerp_2d(a
, b
, v001
, v101
, v011
, v111
);
97 return lerp(c
, temp0
, temp1
);
103 * If A is a signed integer, A % B doesn't give the right value for A < 0
104 * (in terms of texture repeat). Just casting to unsigned fixes that.
106 #define REMAINDER(A, B) ((unsigned) (A) % (unsigned) (B))
110 * Apply texture coord wrapping mode and return integer texture indexes
111 * for a vector of four texcoords (S or T or P).
112 * \param wrapMode PIPE_TEX_WRAP_x
113 * \param s the incoming texcoords
114 * \param size the texture image size
115 * \param icoord returns the integer texcoords
116 * \return integer texture index
119 nearest_texcoord_4(unsigned wrapMode
, const float s
[4], unsigned size
,
124 case PIPE_TEX_WRAP_REPEAT
:
125 /* s limited to [0,1) */
126 /* i limited to [0,size-1] */
127 for (ch
= 0; ch
< 4; ch
++) {
128 int i
= util_ifloor(s
[ch
] * size
);
129 icoord
[ch
] = REMAINDER(i
, size
);
132 case PIPE_TEX_WRAP_CLAMP
:
133 /* s limited to [0,1] */
134 /* i limited to [0,size-1] */
135 for (ch
= 0; ch
< 4; ch
++) {
138 else if (s
[ch
] >= 1.0F
)
139 icoord
[ch
] = size
- 1;
141 icoord
[ch
] = util_ifloor(s
[ch
] * size
);
144 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
146 /* s limited to [min,max] */
147 /* i limited to [0, size-1] */
148 const float min
= 1.0F
/ (2.0F
* size
);
149 const float max
= 1.0F
- min
;
150 for (ch
= 0; ch
< 4; ch
++) {
153 else if (s
[ch
] > max
)
154 icoord
[ch
] = size
- 1;
156 icoord
[ch
] = util_ifloor(s
[ch
] * size
);
160 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
162 /* s limited to [min,max] */
163 /* i limited to [-1, size] */
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
)
172 icoord
[ch
] = util_ifloor(s
[ch
] * size
);
176 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
178 const float min
= 1.0F
/ (2.0F
* size
);
179 const float max
= 1.0F
- min
;
180 for (ch
= 0; ch
< 4; ch
++) {
181 const int flr
= util_ifloor(s
[ch
]);
184 u
= 1.0F
- (s
[ch
] - (float) flr
);
186 u
= s
[ch
] - (float) flr
;
190 icoord
[ch
] = size
- 1;
192 icoord
[ch
] = util_ifloor(u
* size
);
196 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
197 for (ch
= 0; ch
< 4; ch
++) {
198 /* s limited to [0,1] */
199 /* i limited to [0,size-1] */
200 const float u
= fabsf(s
[ch
]);
204 icoord
[ch
] = size
- 1;
206 icoord
[ch
] = util_ifloor(u
* size
);
209 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
211 /* s limited to [min,max] */
212 /* i limited to [0, size-1] */
213 const float min
= 1.0F
/ (2.0F
* size
);
214 const float max
= 1.0F
- min
;
215 for (ch
= 0; ch
< 4; ch
++) {
216 const float u
= fabsf(s
[ch
]);
220 icoord
[ch
] = size
- 1;
222 icoord
[ch
] = util_ifloor(u
* size
);
226 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
228 /* s limited to [min,max] */
229 /* i limited to [0, size-1] */
230 const float min
= -1.0F
/ (2.0F
* size
);
231 const float max
= 1.0F
- min
;
232 for (ch
= 0; ch
< 4; ch
++) {
233 const float u
= fabsf(s
[ch
]);
239 icoord
[ch
] = util_ifloor(u
* size
);
250 * Used to compute texel locations for linear sampling for four texcoords.
251 * \param wrapMode PIPE_TEX_WRAP_x
252 * \param s the texcoords
253 * \param size the texture image size
254 * \param icoord0 returns first texture indexes
255 * \param icoord1 returns second texture indexes (usually icoord0 + 1)
256 * \param w returns blend factor/weight between texture indexes
257 * \param icoord returns the computed integer texture coords
260 linear_texcoord_4(unsigned wrapMode
, const float s
[4], unsigned size
,
261 int icoord0
[4], int icoord1
[4], float w
[4])
266 case PIPE_TEX_WRAP_REPEAT
:
267 for (ch
= 0; ch
< 4; ch
++) {
268 float u
= s
[ch
] * size
- 0.5F
;
269 icoord0
[ch
] = REMAINDER(util_ifloor(u
), size
);
270 icoord1
[ch
] = REMAINDER(icoord0
[ch
] + 1, size
);
274 case PIPE_TEX_WRAP_CLAMP
:
275 for (ch
= 0; ch
< 4; ch
++) {
276 float u
= CLAMP(s
[ch
], 0.0F
, 1.0F
);
278 icoord0
[ch
] = util_ifloor(u
);
279 icoord1
[ch
] = icoord0
[ch
] + 1;
283 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
284 for (ch
= 0; ch
< 4; ch
++) {
285 float u
= CLAMP(s
[ch
], 0.0F
, 1.0F
);
287 icoord0
[ch
] = util_ifloor(u
);
288 icoord1
[ch
] = icoord0
[ch
] + 1;
291 if (icoord1
[ch
] >= (int) size
)
292 icoord1
[ch
] = size
- 1;
296 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
298 const float min
= -1.0F
/ (2.0F
* size
);
299 const float max
= 1.0F
- min
;
300 for (ch
= 0; ch
< 4; ch
++) {
301 float u
= CLAMP(s
[ch
], min
, max
);
303 icoord0
[ch
] = util_ifloor(u
);
304 icoord1
[ch
] = icoord0
[ch
] + 1;
309 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
310 for (ch
= 0; ch
< 4; ch
++) {
311 const int flr
= util_ifloor(s
[ch
]);
314 u
= 1.0F
- (s
[ch
] - (float) flr
);
316 u
= s
[ch
] - (float) flr
;
318 icoord0
[ch
] = util_ifloor(u
);
319 icoord1
[ch
] = icoord0
[ch
] + 1;
322 if (icoord1
[ch
] >= (int) size
)
323 icoord1
[ch
] = size
- 1;
327 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
328 for (ch
= 0; ch
< 4; ch
++) {
329 float u
= fabsf(s
[ch
]);
335 icoord0
[ch
] = util_ifloor(u
);
336 icoord1
[ch
] = icoord0
[ch
] + 1;
340 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
341 for (ch
= 0; ch
< 4; ch
++) {
342 float u
= fabsf(s
[ch
]);
348 icoord0
[ch
] = util_ifloor(u
);
349 icoord1
[ch
] = icoord0
[ch
] + 1;
352 if (icoord1
[ch
] >= (int) size
)
353 icoord1
[ch
] = size
- 1;
357 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
359 const float min
= -1.0F
/ (2.0F
* size
);
360 const float max
= 1.0F
- min
;
361 for (ch
= 0; ch
< 4; ch
++) {
362 float u
= fabsf(s
[ch
]);
370 icoord0
[ch
] = util_ifloor(u
);
371 icoord1
[ch
] = icoord0
[ch
] + 1;
383 * For RECT textures / unnormalized texcoords
384 * Only a subset of wrap modes supported.
387 nearest_texcoord_unnorm_4(unsigned wrapMode
, const float s
[4], unsigned size
,
392 case PIPE_TEX_WRAP_CLAMP
:
393 for (ch
= 0; ch
< 4; ch
++) {
394 int i
= util_ifloor(s
[ch
]);
395 icoord
[ch
]= CLAMP(i
, 0, (int) size
-1);
398 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
400 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
401 for (ch
= 0; ch
< 4; ch
++) {
402 icoord
[ch
]= util_ifloor( CLAMP(s
[ch
], 0.5F
, (float) size
- 0.5F
) );
412 * For RECT textures / unnormalized texcoords.
413 * Only a subset of wrap modes supported.
416 linear_texcoord_unnorm_4(unsigned wrapMode
, const float s
[4], unsigned size
,
417 int icoord0
[4], int icoord1
[4], float w
[4])
421 case PIPE_TEX_WRAP_CLAMP
:
422 for (ch
= 0; ch
< 4; ch
++) {
423 /* Not exactly what the spec says, but it matches NVIDIA output */
424 float u
= CLAMP(s
[ch
] - 0.5F
, 0.0f
, (float) size
- 1.0f
);
425 icoord0
[ch
] = util_ifloor(u
);
426 icoord1
[ch
] = icoord0
[ch
] + 1;
430 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
432 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
433 for (ch
= 0; ch
< 4; ch
++) {
434 float u
= CLAMP(s
[ch
], 0.5F
, (float) size
- 0.5F
);
436 icoord0
[ch
] = util_ifloor(u
);
437 icoord1
[ch
] = icoord0
[ch
] + 1;
438 if (icoord1
[ch
] > (int) size
- 1)
439 icoord1
[ch
] = size
- 1;
450 choose_cube_face(float rx
, float ry
, float rz
, float *newS
, float *newT
)
454 direction target sc tc ma
455 ---------- ------------------------------- --- --- ---
456 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
457 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
458 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
459 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
460 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
461 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
463 const float arx
= fabsf(rx
), ary
= fabsf(ry
), arz
= fabsf(rz
);
467 if (arx
> ary
&& arx
> arz
) {
469 face
= PIPE_TEX_FACE_POS_X
;
475 face
= PIPE_TEX_FACE_NEG_X
;
481 else if (ary
> arx
&& ary
> arz
) {
483 face
= PIPE_TEX_FACE_POS_Y
;
489 face
= PIPE_TEX_FACE_NEG_Y
;
497 face
= PIPE_TEX_FACE_POS_Z
;
503 face
= PIPE_TEX_FACE_NEG_Z
;
510 *newS
= ( sc
/ ma
+ 1.0F
) * 0.5F
;
511 *newT
= ( tc
/ ma
+ 1.0F
) * 0.5F
;
518 * Examine the quad's texture coordinates to compute the partial
519 * derivatives w.r.t X and Y, then compute lambda (level of detail).
521 * This is only done for fragment shaders, not vertex shaders.
524 compute_lambda(struct tgsi_sampler
*tgsi_sampler
,
525 const float s
[QUAD_SIZE
],
526 const float t
[QUAD_SIZE
],
527 const float p
[QUAD_SIZE
],
530 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
531 const struct pipe_texture
*texture
= samp
->texture
;
532 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
535 if (samp
->processor
== TGSI_PROCESSOR_VERTEX
)
538 assert(sampler
->normalized_coords
);
542 float dsdx
= s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
];
543 float dsdy
= s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
];
546 rho
= MAX2(dsdx
, dsdy
) * texture
->width
[0];
549 float dtdx
= t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
];
550 float dtdy
= t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
];
554 max
= MAX2(dtdx
, dtdy
) * texture
->height
[0];
555 rho
= MAX2(rho
, max
);
558 float dpdx
= p
[QUAD_BOTTOM_RIGHT
] - p
[QUAD_BOTTOM_LEFT
];
559 float dpdy
= p
[QUAD_TOP_LEFT
] - p
[QUAD_BOTTOM_LEFT
];
563 max
= MAX2(dpdx
, dpdy
) * texture
->depth
[0];
564 rho
= MAX2(rho
, max
);
567 lambda
= util_fast_log2(rho
);
568 lambda
+= lodbias
+ sampler
->lod_bias
;
569 lambda
= CLAMP(lambda
, sampler
->min_lod
, sampler
->max_lod
);
576 * Do several things here:
577 * 1. Compute lambda from the texcoords, if needed
578 * 2. Determine if we're minifying or magnifying
579 * 3. If minifying, choose mipmap levels
580 * 4. Return image filter to use within mipmap images
581 * \param level0 Returns first mipmap level to sample from
582 * \param level1 Returns second mipmap level to sample from
583 * \param levelBlend Returns blend factor between levels, in [0,1]
584 * \param imgFilter Returns either the min or mag filter, depending on lambda
587 choose_mipmap_levels(struct tgsi_sampler
*tgsi_sampler
,
588 const float s
[QUAD_SIZE
],
589 const float t
[QUAD_SIZE
],
590 const float p
[QUAD_SIZE
],
592 unsigned *level0
, unsigned *level1
, float *levelBlend
,
595 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
596 const struct pipe_texture
*texture
= samp
->texture
;
597 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
599 if (sampler
->min_mip_filter
== PIPE_TEX_MIPFILTER_NONE
) {
600 /* no mipmap selection needed */
601 *level0
= *level1
= CLAMP((int) sampler
->min_lod
,
602 0, (int) texture
->last_level
);
604 if (sampler
->min_img_filter
!= sampler
->mag_img_filter
) {
605 /* non-mipmapped texture, but still need to determine if doing
606 * minification or magnification.
608 float lambda
= compute_lambda(tgsi_sampler
, s
, t
, p
, lodbias
);
610 *imgFilter
= sampler
->mag_img_filter
;
613 *imgFilter
= sampler
->min_img_filter
;
617 *imgFilter
= sampler
->mag_img_filter
;
621 float lambda
= compute_lambda(tgsi_sampler
, s
, t
, p
, lodbias
);
623 if (lambda
<= 0.0) { /* XXX threshold depends on the filter */
625 *imgFilter
= sampler
->mag_img_filter
;
626 *level0
= *level1
= 0;
630 *imgFilter
= sampler
->min_img_filter
;
632 /* choose mipmap level(s) and compute the blend factor between them */
633 if (sampler
->min_mip_filter
== PIPE_TEX_MIPFILTER_NEAREST
) {
634 /* Nearest mipmap level */
635 const int lvl
= (int) (lambda
+ 0.5);
637 *level1
= CLAMP(lvl
, 0, (int) texture
->last_level
);
640 /* Linear interpolation between mipmap levels */
641 const int lvl
= (int) lambda
;
642 *level0
= CLAMP(lvl
, 0, (int) texture
->last_level
);
643 *level1
= CLAMP(lvl
+ 1, 0, (int) texture
->last_level
);
644 *levelBlend
= FRAC(lambda
); /* blending weight between levels */
652 * Get a texel from a texture, using the texture tile cache.
654 * \param face the cube face in 0..5
655 * \param level the mipmap level
656 * \param x the x coord of texel within 2D image
657 * \param y the y coord of texel within 2D image
658 * \param z which slice of a 3D texture
659 * \param rgba the quad to put the texel/color into
660 * \param j which element of the rgba quad to write to
662 * XXX maybe move this into lp_tile_cache.c and merge with the
663 * lp_get_cached_tile_tex() function. Also, get 4 texels instead of 1...
666 get_texel_quad_2d(const struct tgsi_sampler
*tgsi_sampler
,
667 unsigned face
, unsigned level
, int x
, int y
,
668 const uint8_t *out
[4])
670 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
672 const struct llvmpipe_cached_tex_tile
*tile
673 = lp_get_cached_tex_tile(samp
->cache
,
674 tex_tile_address(x
, y
, 0, face
, level
));
679 out
[0] = &tile
->color
[y
][x
][0];
680 out
[1] = &tile
->color
[y
][x
+1][0];
681 out
[2] = &tile
->color
[y
+1][x
][0];
682 out
[3] = &tile
->color
[y
+1][x
+1][0];
685 static INLINE
const uint8_t *
686 get_texel_2d_ptr(const struct tgsi_sampler
*tgsi_sampler
,
687 unsigned face
, unsigned level
, int x
, int y
)
689 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
691 const struct llvmpipe_cached_tex_tile
*tile
692 = lp_get_cached_tex_tile(samp
->cache
,
693 tex_tile_address(x
, y
, 0, face
, level
));
698 return &tile
->color
[y
][x
][0];
703 get_texel_quad_2d_mt(const struct tgsi_sampler
*tgsi_sampler
,
704 unsigned face
, unsigned level
,
707 const uint8_t *out
[4])
711 for (i
= 0; i
< 4; i
++) {
712 unsigned tx
= (i
& 1) ? x1
: x0
;
713 unsigned ty
= (i
>> 1) ? y1
: y0
;
715 out
[i
] = get_texel_2d_ptr( tgsi_sampler
, face
, level
, tx
, ty
);
720 get_texel(const struct tgsi_sampler
*tgsi_sampler
,
721 unsigned face
, unsigned level
, int x
, int y
, int z
,
722 float rgba
[NUM_CHANNELS
][QUAD_SIZE
], unsigned j
)
724 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
725 const struct pipe_texture
*texture
= samp
->texture
;
726 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
728 if (x
< 0 || x
>= (int) texture
->width
[level
] ||
729 y
< 0 || y
>= (int) texture
->height
[level
] ||
730 z
< 0 || z
>= (int) texture
->depth
[level
]) {
731 rgba
[0][j
] = sampler
->border_color
[0];
732 rgba
[1][j
] = sampler
->border_color
[1];
733 rgba
[2][j
] = sampler
->border_color
[2];
734 rgba
[3][j
] = sampler
->border_color
[3];
737 const unsigned tx
= x
% TEX_TILE_SIZE
;
738 const unsigned ty
= y
% TEX_TILE_SIZE
;
739 const struct llvmpipe_cached_tex_tile
*tile
;
741 tile
= lp_get_cached_tex_tile(samp
->cache
,
742 tex_tile_address(x
, y
, z
, face
, level
));
744 rgba
[0][j
] = ubyte_to_float(tile
->color
[ty
][tx
][0]);
745 rgba
[1][j
] = ubyte_to_float(tile
->color
[ty
][tx
][1]);
746 rgba
[2][j
] = ubyte_to_float(tile
->color
[ty
][tx
][2]);
747 rgba
[3][j
] = ubyte_to_float(tile
->color
[ty
][tx
][3]);
750 debug_printf("Get texel %f %f %f %f from %s\n",
751 rgba
[0][j
], rgba
[1][j
], rgba
[2][j
], rgba
[3][j
],
752 pf_name(texture
->format
));
759 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
760 * When we sampled the depth texture, the depth value was put into all
761 * RGBA channels. We look at the red channel here.
762 * \param rgba quad of (depth) texel values
763 * \param p texture 'P' components for four pixels in quad
764 * \param j which pixel in the quad to test [0..3]
767 shadow_compare(const struct pipe_sampler_state
*sampler
,
768 float rgba
[NUM_CHANNELS
][QUAD_SIZE
],
769 const float p
[QUAD_SIZE
],
773 switch (sampler
->compare_func
) {
775 k
= p
[j
] < rgba
[0][j
];
777 case PIPE_FUNC_LEQUAL
:
778 k
= p
[j
] <= rgba
[0][j
];
780 case PIPE_FUNC_GREATER
:
781 k
= p
[j
] > rgba
[0][j
];
783 case PIPE_FUNC_GEQUAL
:
784 k
= p
[j
] >= rgba
[0][j
];
786 case PIPE_FUNC_EQUAL
:
787 k
= p
[j
] == rgba
[0][j
];
789 case PIPE_FUNC_NOTEQUAL
:
790 k
= p
[j
] != rgba
[0][j
];
792 case PIPE_FUNC_ALWAYS
:
795 case PIPE_FUNC_NEVER
:
804 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
805 rgba
[0][j
] = rgba
[1][j
] = rgba
[2][j
] = (float) k
;
811 * As above, but do four z/texture comparisons.
814 shadow_compare4(const struct pipe_sampler_state
*sampler
,
815 float rgba
[NUM_CHANNELS
][QUAD_SIZE
],
816 const float p
[QUAD_SIZE
])
818 int j
, k0
, k1
, k2
, k3
;
821 /* compare four texcoords vs. four texture samples */
822 switch (sampler
->compare_func
) {
824 k0
= p
[0] < rgba
[0][0];
825 k1
= p
[1] < rgba
[0][1];
826 k2
= p
[2] < rgba
[0][2];
827 k3
= p
[3] < rgba
[0][3];
829 case PIPE_FUNC_LEQUAL
:
830 k0
= p
[0] <= rgba
[0][0];
831 k1
= p
[1] <= rgba
[0][1];
832 k2
= p
[2] <= rgba
[0][2];
833 k3
= p
[3] <= rgba
[0][3];
835 case PIPE_FUNC_GREATER
:
836 k0
= p
[0] > rgba
[0][0];
837 k1
= p
[1] > rgba
[0][1];
838 k2
= p
[2] > rgba
[0][2];
839 k3
= p
[3] > rgba
[0][3];
841 case PIPE_FUNC_GEQUAL
:
842 k0
= p
[0] >= rgba
[0][0];
843 k1
= p
[1] >= rgba
[0][1];
844 k2
= p
[2] >= rgba
[0][2];
845 k3
= p
[3] >= rgba
[0][3];
847 case PIPE_FUNC_EQUAL
:
848 k0
= p
[0] == rgba
[0][0];
849 k1
= p
[1] == rgba
[0][1];
850 k2
= p
[2] == rgba
[0][2];
851 k3
= p
[3] == rgba
[0][3];
853 case PIPE_FUNC_NOTEQUAL
:
854 k0
= p
[0] != rgba
[0][0];
855 k1
= p
[1] != rgba
[0][1];
856 k2
= p
[2] != rgba
[0][2];
857 k3
= p
[3] != rgba
[0][3];
859 case PIPE_FUNC_ALWAYS
:
860 k0
= k1
= k2
= k3
= 1;
862 case PIPE_FUNC_NEVER
:
863 k0
= k1
= k2
= k3
= 0;
866 k0
= k1
= k2
= k3
= 0;
871 /* convert four pass/fail values to an intensity in [0,1] */
872 val
= 0.25F
* (k0
+ k1
+ k2
+ k3
);
874 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
875 for (j
= 0; j
< 4; j
++) {
876 rgba
[0][j
] = rgba
[1][j
] = rgba
[2][j
] = val
;
884 lp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler
*tgsi_sampler
,
885 const float s
[QUAD_SIZE
],
886 const float t
[QUAD_SIZE
],
887 const float p
[QUAD_SIZE
],
889 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
891 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
893 unsigned level
= samp
->level
;
894 unsigned xpot
= 1 << (samp
->xpot
- level
);
895 unsigned ypot
= 1 << (samp
->ypot
- level
);
896 unsigned xmax
= (xpot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
897 unsigned ymax
= (ypot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
899 for (j
= 0; j
< QUAD_SIZE
; j
++) {
902 float u
= s
[j
] * xpot
- 0.5F
;
903 float v
= t
[j
] * ypot
- 0.5F
;
905 int uflr
= util_ifloor(u
);
906 int vflr
= util_ifloor(v
);
908 float xw
= u
- (float)uflr
;
909 float yw
= v
- (float)vflr
;
911 int x0
= uflr
& (xpot
- 1);
912 int y0
= vflr
& (ypot
- 1);
914 const uint8_t *tx
[4];
917 /* Can we fetch all four at once:
919 if (x0
< xmax
&& y0
< ymax
)
921 get_texel_quad_2d(tgsi_sampler
, 0, level
, x0
, y0
, tx
);
925 unsigned x1
= (x0
+ 1) & (xpot
- 1);
926 unsigned y1
= (y0
+ 1) & (ypot
- 1);
927 get_texel_quad_2d_mt(tgsi_sampler
, 0, level
,
932 /* interpolate R, G, B, A */
933 for (c
= 0; c
< 4; c
++) {
934 rgba
[c
][j
] = lerp_2d(xw
, yw
,
935 ubyte_to_float(tx
[0][c
]), ubyte_to_float(tx
[1][c
]),
936 ubyte_to_float(tx
[2][c
]), ubyte_to_float(tx
[3][c
]));
943 lp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler
*tgsi_sampler
,
944 const float s
[QUAD_SIZE
],
945 const float t
[QUAD_SIZE
],
946 const float p
[QUAD_SIZE
],
948 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
950 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
952 unsigned level
= samp
->level
;
953 unsigned xpot
= 1 << (samp
->xpot
- level
);
954 unsigned ypot
= 1 << (samp
->ypot
- level
);
956 for (j
= 0; j
< QUAD_SIZE
; j
++) {
959 float u
= s
[j
] * xpot
;
960 float v
= t
[j
] * ypot
;
962 int uflr
= util_ifloor(u
);
963 int vflr
= util_ifloor(v
);
965 int x0
= uflr
& (xpot
- 1);
966 int y0
= vflr
& (ypot
- 1);
968 const uint8_t *out
= get_texel_2d_ptr(tgsi_sampler
, 0, level
, x0
, y0
);
970 for (c
= 0; c
< 4; c
++) {
971 rgba
[c
][j
] = ubyte_to_float(out
[c
]);
978 lp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler
*tgsi_sampler
,
979 const float s
[QUAD_SIZE
],
980 const float t
[QUAD_SIZE
],
981 const float p
[QUAD_SIZE
],
983 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
985 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
987 unsigned level
= samp
->level
;
988 unsigned xpot
= 1 << (samp
->xpot
- level
);
989 unsigned ypot
= 1 << (samp
->ypot
- level
);
991 for (j
= 0; j
< QUAD_SIZE
; j
++) {
994 float u
= s
[j
] * xpot
;
995 float v
= t
[j
] * ypot
;
1000 x0
= util_ifloor(u
);
1003 else if (x0
> xpot
- 1)
1006 y0
= util_ifloor(v
);
1009 else if (y0
> ypot
- 1)
1012 out
= get_texel_2d_ptr(tgsi_sampler
, 0, level
, x0
, y0
);
1014 for (c
= 0; c
< 4; c
++) {
1015 rgba
[c
][j
] = ubyte_to_float(out
[c
]);
1022 lp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler
*tgsi_sampler
,
1023 const float s
[QUAD_SIZE
],
1024 const float t
[QUAD_SIZE
],
1025 const float p
[QUAD_SIZE
],
1027 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1029 struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
1030 const struct pipe_texture
*texture
= samp
->texture
;
1034 lambda
= compute_lambda(tgsi_sampler
, s
, t
, p
, lodbias
);
1035 level0
= (int)lambda
;
1039 lp_get_samples_2d_linear_repeat_POT( tgsi_sampler
,
1042 else if (level0
>= texture
->last_level
) {
1043 samp
->level
= texture
->last_level
;
1044 lp_get_samples_2d_linear_repeat_POT( tgsi_sampler
,
1048 float levelBlend
= lambda
- level0
;
1053 samp
->level
= level0
;
1054 lp_get_samples_2d_linear_repeat_POT( tgsi_sampler
,
1055 s
, t
, p
, 0, rgba0
);
1057 samp
->level
= level0
+1;
1058 lp_get_samples_2d_linear_repeat_POT( tgsi_sampler
,
1059 s
, t
, p
, 0, rgba1
);
1061 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1062 for (c
= 0; c
< 4; c
++) {
1063 rgba
[c
][j
] = lerp(levelBlend
, rgba0
[c
][j
], rgba1
[c
][j
]);
1070 * Common code for sampling 1D/2D/cube textures.
1071 * Could probably extend for 3D...
1074 lp_get_samples_2d_common(struct tgsi_sampler
*tgsi_sampler
,
1075 const float s
[QUAD_SIZE
],
1076 const float t
[QUAD_SIZE
],
1077 const float p
[QUAD_SIZE
],
1079 float rgba
[NUM_CHANNELS
][QUAD_SIZE
],
1080 const unsigned faces
[4])
1082 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
1083 const struct pipe_texture
*texture
= samp
->texture
;
1084 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
1085 unsigned level0
, level1
, j
, imgFilter
;
1087 float levelBlend
= 0.0F
;
1089 choose_mipmap_levels(tgsi_sampler
, s
, t
, p
,
1091 &level0
, &level1
, &levelBlend
, &imgFilter
);
1093 assert(sampler
->normalized_coords
);
1095 width
= texture
->width
[level0
];
1096 height
= texture
->height
[level0
];
1100 switch (imgFilter
) {
1101 case PIPE_TEX_FILTER_NEAREST
:
1104 nearest_texcoord_4(sampler
->wrap_s
, s
, width
, x
);
1105 nearest_texcoord_4(sampler
->wrap_t
, t
, height
, y
);
1107 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1108 get_texel(tgsi_sampler
, faces
[j
], level0
, x
[j
], y
[j
], 0, rgba
, j
);
1109 if (sampler
->compare_mode
== PIPE_TEX_COMPARE_R_TO_TEXTURE
) {
1110 shadow_compare(sampler
, rgba
, p
, j
);
1113 if (level0
!= level1
) {
1114 /* get texels from second mipmap level and blend */
1119 get_texel(tgsi_sampler
, faces
[j
], level1
, x
[j
], y
[j
], 0,
1121 if (sampler
->compare_mode
== PIPE_TEX_COMPARE_R_TO_TEXTURE
){
1122 shadow_compare(sampler
, rgba2
, p
, j
);
1125 for (c
= 0; c
< NUM_CHANNELS
; c
++) {
1126 rgba
[c
][j
] = lerp(levelBlend
, rgba
[c
][j
], rgba2
[c
][j
]);
1132 case PIPE_TEX_FILTER_LINEAR
:
1133 case PIPE_TEX_FILTER_ANISO
:
1135 int x0
[4], y0
[4], x1
[4], y1
[4];
1136 float xw
[4], yw
[4]; /* weights */
1138 linear_texcoord_4(sampler
->wrap_s
, s
, width
, x0
, x1
, xw
);
1139 linear_texcoord_4(sampler
->wrap_t
, t
, height
, y0
, y1
, yw
);
1141 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1142 float tx
[4][4]; /* texels */
1144 get_texel(tgsi_sampler
, faces
[j
], level0
, x0
[j
], y0
[j
], 0, tx
, 0);
1145 get_texel(tgsi_sampler
, faces
[j
], level0
, x1
[j
], y0
[j
], 0, tx
, 1);
1146 get_texel(tgsi_sampler
, faces
[j
], level0
, x0
[j
], y1
[j
], 0, tx
, 2);
1147 get_texel(tgsi_sampler
, faces
[j
], level0
, x1
[j
], y1
[j
], 0, tx
, 3);
1148 if (sampler
->compare_mode
== PIPE_TEX_COMPARE_R_TO_TEXTURE
) {
1149 shadow_compare4(sampler
, tx
, p
);
1152 /* interpolate R, G, B, A */
1153 for (c
= 0; c
< 4; c
++) {
1154 rgba
[c
][j
] = lerp_2d(xw
[j
], yw
[j
],
1156 tx
[c
][2], tx
[c
][3]);
1159 if (level0
!= level1
) {
1160 /* get texels from second mipmap level and blend */
1163 /* XXX: This is incorrect -- will often end up with (x0
1164 * == x1 && y0 == y1), meaning that we fetch the same
1165 * texel four times and linearly interpolate between
1166 * identical values. The correct approach would be to
1167 * call linear_texcoord again for the second level.
1173 get_texel(tgsi_sampler
, faces
[j
], level1
, x0
[j
], y0
[j
], 0, tx
, 0);
1174 get_texel(tgsi_sampler
, faces
[j
], level1
, x1
[j
], y0
[j
], 0, tx
, 1);
1175 get_texel(tgsi_sampler
, faces
[j
], level1
, x0
[j
], y1
[j
], 0, tx
, 2);
1176 get_texel(tgsi_sampler
, faces
[j
], level1
, x1
[j
], y1
[j
], 0, tx
, 3);
1177 if (sampler
->compare_mode
== PIPE_TEX_COMPARE_R_TO_TEXTURE
){
1178 shadow_compare4(sampler
, tx
, p
);
1181 /* interpolate R, G, B, A */
1182 for (c
= 0; c
< 4; c
++) {
1183 rgba2
[c
][j
] = lerp_2d(xw
[j
], yw
[j
],
1184 tx
[c
][0], tx
[c
][1], tx
[c
][2], tx
[c
][3]);
1187 for (c
= 0; c
< NUM_CHANNELS
; c
++) {
1188 rgba
[c
][j
] = lerp(levelBlend
, rgba
[c
][j
], rgba2
[c
][j
]);
1201 lp_get_samples_1d(struct tgsi_sampler
*sampler
,
1202 const float s
[QUAD_SIZE
],
1203 const float t
[QUAD_SIZE
],
1204 const float p
[QUAD_SIZE
],
1206 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1208 static const unsigned faces
[4] = {0, 0, 0, 0};
1209 static const float tzero
[4] = {0, 0, 0, 0};
1210 lp_get_samples_2d_common(sampler
, s
, tzero
, NULL
,
1211 lodbias
, rgba
, faces
);
1216 lp_get_samples_2d(struct tgsi_sampler
*sampler
,
1217 const float s
[QUAD_SIZE
],
1218 const float t
[QUAD_SIZE
],
1219 const float p
[QUAD_SIZE
],
1221 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1223 static const unsigned faces
[4] = {0, 0, 0, 0};
1224 lp_get_samples_2d_common(sampler
, s
, t
, p
,
1225 lodbias
, rgba
, faces
);
1230 lp_get_samples_3d(struct tgsi_sampler
*tgsi_sampler
,
1231 const float s
[QUAD_SIZE
],
1232 const float t
[QUAD_SIZE
],
1233 const float p
[QUAD_SIZE
],
1235 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1237 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
1238 const struct pipe_texture
*texture
= samp
->texture
;
1239 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
1240 /* get/map pipe_surfaces corresponding to 3D tex slices */
1241 unsigned level0
, level1
, j
, imgFilter
;
1242 int width
, height
, depth
;
1244 const uint face
= 0;
1246 choose_mipmap_levels(tgsi_sampler
, s
, t
, p
,
1248 &level0
, &level1
, &levelBlend
, &imgFilter
);
1250 assert(sampler
->normalized_coords
);
1252 width
= texture
->width
[level0
];
1253 height
= texture
->height
[level0
];
1254 depth
= texture
->depth
[level0
];
1260 switch (imgFilter
) {
1261 case PIPE_TEX_FILTER_NEAREST
:
1263 int x
[4], y
[4], z
[4];
1264 nearest_texcoord_4(sampler
->wrap_s
, s
, width
, x
);
1265 nearest_texcoord_4(sampler
->wrap_t
, t
, height
, y
);
1266 nearest_texcoord_4(sampler
->wrap_r
, p
, depth
, z
);
1267 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1268 get_texel(tgsi_sampler
, face
, level0
, x
[j
], y
[j
], z
[j
], rgba
, j
);
1269 if (level0
!= level1
) {
1270 /* get texels from second mipmap level and blend */
1276 get_texel(tgsi_sampler
, face
, level1
, x
[j
], y
[j
], z
[j
], rgba2
, j
);
1277 for (c
= 0; c
< NUM_CHANNELS
; c
++) {
1278 rgba
[c
][j
] = lerp(levelBlend
, rgba2
[c
][j
], rgba
[c
][j
]);
1284 case PIPE_TEX_FILTER_LINEAR
:
1285 case PIPE_TEX_FILTER_ANISO
:
1287 int x0
[4], x1
[4], y0
[4], y1
[4], z0
[4], z1
[4];
1288 float xw
[4], yw
[4], zw
[4]; /* interpolation weights */
1289 linear_texcoord_4(sampler
->wrap_s
, s
, width
, x0
, x1
, xw
);
1290 linear_texcoord_4(sampler
->wrap_t
, t
, height
, y0
, y1
, yw
);
1291 linear_texcoord_4(sampler
->wrap_r
, p
, depth
, z0
, z1
, zw
);
1293 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1295 float tx0
[4][4], tx1
[4][4];
1296 get_texel(tgsi_sampler
, face
, level0
, x0
[j
], y0
[j
], z0
[j
], tx0
, 0);
1297 get_texel(tgsi_sampler
, face
, level0
, x1
[j
], y0
[j
], z0
[j
], tx0
, 1);
1298 get_texel(tgsi_sampler
, face
, level0
, x0
[j
], y1
[j
], z0
[j
], tx0
, 2);
1299 get_texel(tgsi_sampler
, face
, level0
, x1
[j
], y1
[j
], z0
[j
], tx0
, 3);
1300 get_texel(tgsi_sampler
, face
, level0
, x0
[j
], y0
[j
], z1
[j
], tx1
, 0);
1301 get_texel(tgsi_sampler
, face
, level0
, x1
[j
], y0
[j
], z1
[j
], tx1
, 1);
1302 get_texel(tgsi_sampler
, face
, level0
, x0
[j
], y1
[j
], z1
[j
], tx1
, 2);
1303 get_texel(tgsi_sampler
, face
, level0
, x1
[j
], y1
[j
], z1
[j
], tx1
, 3);
1305 /* interpolate R, G, B, A */
1306 for (c
= 0; c
< 4; c
++) {
1307 rgba
[c
][j
] = lerp_3d(xw
[j
], yw
[j
], zw
[j
],
1308 tx0
[c
][0], tx0
[c
][1],
1309 tx0
[c
][2], tx0
[c
][3],
1310 tx1
[c
][0], tx1
[c
][1],
1311 tx1
[c
][2], tx1
[c
][3]);
1314 if (level0
!= level1
) {
1315 /* get texels from second mipmap level and blend */
1323 get_texel(tgsi_sampler
, face
, level1
, x0
[j
], y0
[j
], z0
[j
], tx0
, 0);
1324 get_texel(tgsi_sampler
, face
, level1
, x1
[j
], y0
[j
], z0
[j
], tx0
, 1);
1325 get_texel(tgsi_sampler
, face
, level1
, x0
[j
], y1
[j
], z0
[j
], tx0
, 2);
1326 get_texel(tgsi_sampler
, face
, level1
, x1
[j
], y1
[j
], z0
[j
], tx0
, 3);
1327 get_texel(tgsi_sampler
, face
, level1
, x0
[j
], y0
[j
], z1
[j
], tx1
, 0);
1328 get_texel(tgsi_sampler
, face
, level1
, x1
[j
], y0
[j
], z1
[j
], tx1
, 1);
1329 get_texel(tgsi_sampler
, face
, level1
, x0
[j
], y1
[j
], z1
[j
], tx1
, 2);
1330 get_texel(tgsi_sampler
, face
, level1
, x1
[j
], y1
[j
], z1
[j
], tx1
, 3);
1332 /* interpolate R, G, B, A */
1333 for (c
= 0; c
< 4; c
++) {
1334 rgba2
[c
][j
] = lerp_3d(xw
[j
], yw
[j
], zw
[j
],
1335 tx0
[c
][0], tx0
[c
][1],
1336 tx0
[c
][2], tx0
[c
][3],
1337 tx1
[c
][0], tx1
[c
][1],
1338 tx1
[c
][2], tx1
[c
][3]);
1341 /* blend mipmap levels */
1342 for (c
= 0; c
< NUM_CHANNELS
; c
++) {
1343 rgba
[c
][j
] = lerp(levelBlend
, rgba
[c
][j
], rgba2
[c
][j
]);
1356 lp_get_samples_cube(struct tgsi_sampler
*sampler
,
1357 const float s
[QUAD_SIZE
],
1358 const float t
[QUAD_SIZE
],
1359 const float p
[QUAD_SIZE
],
1361 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1363 unsigned faces
[QUAD_SIZE
], j
;
1364 float ssss
[4], tttt
[4];
1365 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1366 faces
[j
] = choose_cube_face(s
[j
], t
[j
], p
[j
], ssss
+ j
, tttt
+ j
);
1368 lp_get_samples_2d_common(sampler
, ssss
, tttt
, NULL
,
1369 lodbias
, rgba
, faces
);
1374 lp_get_samples_rect(struct tgsi_sampler
*tgsi_sampler
,
1375 const float s
[QUAD_SIZE
],
1376 const float t
[QUAD_SIZE
],
1377 const float p
[QUAD_SIZE
],
1379 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1381 const struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
1382 const struct pipe_texture
*texture
= samp
->texture
;
1383 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
1384 const uint face
= 0;
1385 unsigned level0
, level1
, j
, imgFilter
;
1389 choose_mipmap_levels(tgsi_sampler
, s
, t
, p
,
1391 &level0
, &level1
, &levelBlend
, &imgFilter
);
1393 /* texture RECTS cannot be mipmapped */
1394 assert(level0
== level1
);
1396 width
= texture
->width
[level0
];
1397 height
= texture
->height
[level0
];
1401 switch (imgFilter
) {
1402 case PIPE_TEX_FILTER_NEAREST
:
1405 nearest_texcoord_unnorm_4(sampler
->wrap_s
, s
, width
, x
);
1406 nearest_texcoord_unnorm_4(sampler
->wrap_t
, t
, height
, y
);
1407 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1408 get_texel(tgsi_sampler
, face
, level0
, x
[j
], y
[j
], 0, rgba
, j
);
1409 if (sampler
->compare_mode
== PIPE_TEX_COMPARE_R_TO_TEXTURE
) {
1410 shadow_compare(sampler
, rgba
, p
, j
);
1415 case PIPE_TEX_FILTER_LINEAR
:
1416 case PIPE_TEX_FILTER_ANISO
:
1418 int x0
[4], y0
[4], x1
[4], y1
[4];
1419 float xw
[4], yw
[4]; /* weights */
1420 linear_texcoord_unnorm_4(sampler
->wrap_s
, s
, width
, x0
, x1
, xw
);
1421 linear_texcoord_unnorm_4(sampler
->wrap_t
, t
, height
, y0
, y1
, yw
);
1422 for (j
= 0; j
< QUAD_SIZE
; j
++) {
1423 float tx
[4][4]; /* texels */
1425 get_texel(tgsi_sampler
, face
, level0
, x0
[j
], y0
[j
], 0, tx
, 0);
1426 get_texel(tgsi_sampler
, face
, level0
, x1
[j
], y0
[j
], 0, tx
, 1);
1427 get_texel(tgsi_sampler
, face
, level0
, x0
[j
], y1
[j
], 0, tx
, 2);
1428 get_texel(tgsi_sampler
, face
, level0
, x1
[j
], y1
[j
], 0, tx
, 3);
1429 if (sampler
->compare_mode
== PIPE_TEX_COMPARE_R_TO_TEXTURE
) {
1430 shadow_compare4(sampler
, tx
, p
);
1432 for (c
= 0; c
< 4; c
++) {
1433 rgba
[c
][j
] = lerp_2d(xw
[j
], yw
[j
],
1434 tx
[c
][0], tx
[c
][1], tx
[c
][2], tx
[c
][3]);
1446 * Error condition handler
1449 lp_get_samples_null(struct tgsi_sampler
*tgsi_sampler
,
1450 const float s
[QUAD_SIZE
],
1451 const float t
[QUAD_SIZE
],
1452 const float p
[QUAD_SIZE
],
1454 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1458 for (i
= 0; i
< 4; i
++)
1459 for (j
= 0; j
< 4; j
++)
1464 * Called via tgsi_sampler::get_samples() when using a sampler for the
1465 * first time. Determine the actual sampler function, link it in and
1469 lp_get_samples(struct tgsi_sampler
*tgsi_sampler
,
1470 const float s
[QUAD_SIZE
],
1471 const float t
[QUAD_SIZE
],
1472 const float p
[QUAD_SIZE
],
1474 float rgba
[NUM_CHANNELS
][QUAD_SIZE
])
1476 struct lp_shader_sampler
*samp
= lp_shader_sampler(tgsi_sampler
);
1477 const struct pipe_texture
*texture
= samp
->texture
;
1478 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
1480 /* Default to the 'undefined' case:
1482 tgsi_sampler
->get_samples
= lp_get_samples_null
;
1485 assert(0); /* is this legal?? */
1489 if (!sampler
->normalized_coords
) {
1490 assert (texture
->target
== PIPE_TEXTURE_2D
);
1491 tgsi_sampler
->get_samples
= lp_get_samples_rect
;
1495 switch (texture
->target
) {
1496 case PIPE_TEXTURE_1D
:
1497 tgsi_sampler
->get_samples
= lp_get_samples_1d
;
1499 case PIPE_TEXTURE_2D
:
1500 tgsi_sampler
->get_samples
= lp_get_samples_2d
;
1502 case PIPE_TEXTURE_3D
:
1503 tgsi_sampler
->get_samples
= lp_get_samples_3d
;
1505 case PIPE_TEXTURE_CUBE
:
1506 tgsi_sampler
->get_samples
= lp_get_samples_cube
;
1513 /* Do this elsewhere:
1515 samp
->xpot
= util_unsigned_logbase2( samp
->texture
->width
[0] );
1516 samp
->ypot
= util_unsigned_logbase2( samp
->texture
->height
[0] );
1518 /* Try to hook in a faster sampler. Ultimately we'll have to
1519 * code-generate these. Luckily most of this looks like it is
1520 * orthogonal state within the sampler.
1522 if (texture
->target
== PIPE_TEXTURE_2D
&&
1523 sampler
->min_img_filter
== sampler
->mag_img_filter
&&
1524 sampler
->wrap_s
== sampler
->wrap_t
&&
1525 sampler
->compare_mode
== FALSE
&&
1526 sampler
->normalized_coords
)
1528 if (sampler
->min_mip_filter
== PIPE_TEX_MIPFILTER_NONE
) {
1529 samp
->level
= CLAMP((int) sampler
->min_lod
,
1530 0, (int) texture
->last_level
);
1532 if (sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
) {
1533 switch (sampler
->min_img_filter
) {
1534 case PIPE_TEX_FILTER_NEAREST
:
1535 tgsi_sampler
->get_samples
= lp_get_samples_2d_nearest_repeat_POT
;
1537 case PIPE_TEX_FILTER_LINEAR
:
1538 tgsi_sampler
->get_samples
= lp_get_samples_2d_linear_repeat_POT
;
1544 else if (sampler
->wrap_s
== PIPE_TEX_WRAP_CLAMP
) {
1545 switch (sampler
->min_img_filter
) {
1546 case PIPE_TEX_FILTER_NEAREST
:
1547 tgsi_sampler
->get_samples
= lp_get_samples_2d_nearest_clamp_POT
;
1554 else if (sampler
->min_mip_filter
== PIPE_TEX_MIPFILTER_LINEAR
) {
1555 if (sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
) {
1556 switch (sampler
->min_img_filter
) {
1557 case PIPE_TEX_FILTER_LINEAR
:
1558 tgsi_sampler
->get_samples
= lp_get_samples_2d_linear_mip_linear_repeat_POT
;
1567 _debug_printf("target %d/%d min_mip %d/%d min_img %d/%d wrap %d/%d compare %d/%d norm %d/%d\n",
1568 texture
->target
, PIPE_TEXTURE_2D
,
1569 sampler
->min_mip_filter
, PIPE_TEX_MIPFILTER_NONE
,
1570 sampler
->min_img_filter
, sampler
->mag_img_filter
,
1571 sampler
->wrap_s
, sampler
->wrap_t
,
1572 sampler
->compare_mode
, FALSE
,
1573 sampler
->normalized_coords
, TRUE
);
1577 tgsi_sampler
->get_samples( tgsi_sampler
, s
, t
, p
, lodbias
, rgba
);
1582 lp_fetch_texel_soa( struct tgsi_sampler
**samplers
,
1586 struct tgsi_sampler
*sampler
= samplers
[unit
];
1591 debug_printf("%s sampler: %p (%p) store: %p\n",
1596 debug_printf("lodbias %f\n", store
[12]);
1598 for (j
= 0; j
< 4; j
++)
1599 debug_printf("sample %d texcoord %f %f\n",
1606 float rgba
[NUM_CHANNELS
][QUAD_SIZE
];
1607 sampler
->get_samples(sampler
,
1611 0.0f
, /*store[12], lodbias */
1613 memcpy(store
, rgba
, sizeof rgba
);
1617 for (j
= 0; j
< 4; j
++)
1618 debug_printf("sample %d result %f %f %f %f\n",
1628 #include "lp_bld_type.h"
1629 #include "lp_bld_intr.h"
1630 #include "lp_bld_tgsi.h"
1633 struct lp_c_sampler_soa
1635 struct lp_build_sampler_soa base
;
1637 LLVMValueRef context_ptr
;
1639 LLVMValueRef samplers_ptr
;
1641 /** Coords/texels store */
1642 LLVMValueRef store_ptr
;
1647 lp_c_sampler_soa_destroy(struct lp_build_sampler_soa
*sampler
)
1654 lp_c_sampler_soa_emit_fetch_texel(struct lp_build_sampler_soa
*_sampler
,
1655 LLVMBuilderRef builder
,
1656 struct lp_type type
,
1658 unsigned num_coords
,
1659 const LLVMValueRef
*coords
,
1660 LLVMValueRef lodbias
,
1661 LLVMValueRef
*texel
)
1663 struct lp_c_sampler_soa
*sampler
= (struct lp_c_sampler_soa
*)_sampler
;
1664 LLVMTypeRef vec_type
= LLVMTypeOf(coords
[0]);
1665 LLVMValueRef args
[3];
1668 if(!sampler
->samplers_ptr
)
1669 sampler
->samplers_ptr
= lp_jit_context_samplers(builder
, sampler
->context_ptr
);
1671 if(!sampler
->store_ptr
)
1672 sampler
->store_ptr
= LLVMBuildArrayAlloca(builder
,
1674 LLVMConstInt(LLVMInt32Type(), 4, 0),
1677 for (i
= 0; i
< num_coords
; i
++) {
1678 LLVMValueRef index
= LLVMConstInt(LLVMInt32Type(), i
, 0);
1679 LLVMValueRef coord_ptr
= LLVMBuildGEP(builder
, sampler
->store_ptr
, &index
, 1, "");
1680 LLVMBuildStore(builder
, coords
[i
], coord_ptr
);
1683 args
[0] = sampler
->samplers_ptr
;
1684 args
[1] = LLVMConstInt(LLVMInt32Type(), unit
, 0);
1685 args
[2] = sampler
->store_ptr
;
1687 lp_build_intrinsic(builder
, "fetch_texel", LLVMVoidType(), args
, 3);
1689 for (i
= 0; i
< NUM_CHANNELS
; ++i
) {
1690 LLVMValueRef index
= LLVMConstInt(LLVMInt32Type(), i
, 0);
1691 LLVMValueRef texel_ptr
= LLVMBuildGEP(builder
, sampler
->store_ptr
, &index
, 1, "");
1692 texel
[i
] = LLVMBuildLoad(builder
, texel_ptr
, "");
1697 struct lp_build_sampler_soa
*
1698 lp_c_sampler_soa_create(LLVMValueRef context_ptr
)
1700 struct lp_c_sampler_soa
*sampler
;
1702 sampler
= CALLOC_STRUCT(lp_c_sampler_soa
);
1706 sampler
->base
.destroy
= lp_c_sampler_soa_destroy
;
1707 sampler
->base
.emit_fetch_texel
= lp_c_sampler_soa_emit_fetch_texel
;
1708 sampler
->context_ptr
= context_ptr
;
1710 return &sampler
->base
;