1 /**************************************************************************
3 * Copyright 2007 VMware, Inc.
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 VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
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_format.h"
42 #include "util/u_memory.h"
43 #include "util/u_inlines.h"
44 #include "sp_quad.h" /* only for #define QUAD_* tokens */
45 #include "sp_tex_sample.h"
46 #include "sp_texture.h"
47 #include "sp_tex_tile_cache.h"
50 /** Set to one to help debug texture sampling */
55 * Return fractional part of 'f'. Used for computing interpolation weights.
56 * Need to be careful with negative values.
57 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
58 * of improperly weighted linear-filtered textures.
59 * The tests/texwrap.c demo is a good test.
70 * Linear interpolation macro
73 lerp(float a
, float v0
, float v1
)
75 return v0
+ a
* (v1
- v0
);
80 * Do 2D/bilinear interpolation of float values.
81 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
82 * a and b are the horizontal and vertical interpolants.
83 * It's important that this function is inlined when compiled with
84 * optimization! If we find that's not true on some systems, convert
88 lerp_2d(float a
, float b
,
89 float v00
, float v10
, float v01
, float v11
)
91 const float temp0
= lerp(a
, v00
, v10
);
92 const float temp1
= lerp(a
, v01
, v11
);
93 return lerp(b
, temp0
, temp1
);
98 * As above, but 3D interpolation of 8 values.
101 lerp_3d(float a
, float b
, float c
,
102 float v000
, float v100
, float v010
, float v110
,
103 float v001
, float v101
, float v011
, float v111
)
105 const float temp0
= lerp_2d(a
, b
, v000
, v100
, v010
, v110
);
106 const float temp1
= lerp_2d(a
, b
, v001
, v101
, v011
, v111
);
107 return lerp(c
, temp0
, temp1
);
113 * Compute coord % size for repeat wrap modes.
114 * Note that if coord is negative, coord % size doesn't give the right
115 * value. To avoid that problem we add a large multiple of the size
116 * (rather than using a conditional).
119 repeat(int coord
, unsigned size
)
121 return (coord
+ size
* 1024) % size
;
126 * Apply texture coord wrapping mode and return integer texture indexes
127 * for a vector of four texcoords (S or T or P).
128 * \param wrapMode PIPE_TEX_WRAP_x
129 * \param s the incoming texcoords
130 * \param size the texture image size
131 * \param icoord returns the integer texcoords
134 wrap_nearest_repeat(float s
, unsigned size
, int offset
, int *icoord
)
136 /* s limited to [0,1) */
137 /* i limited to [0,size-1] */
138 int i
= util_ifloor(s
* size
);
139 *icoord
= repeat(i
+ offset
, size
);
144 wrap_nearest_clamp(float s
, unsigned size
, int offset
, int *icoord
)
146 /* s limited to [0,1] */
147 /* i limited to [0,size-1] */
155 *icoord
= util_ifloor(s
);
160 wrap_nearest_clamp_to_edge(float s
, unsigned size
, int offset
, int *icoord
)
162 /* s limited to [min,max] */
163 /* i limited to [0, size-1] */
164 const float min
= 0.5F
;
165 const float max
= (float)size
- 0.5F
;
175 *icoord
= util_ifloor(s
);
180 wrap_nearest_clamp_to_border(float s
, unsigned size
, int offset
, int *icoord
)
182 /* s limited to [min,max] */
183 /* i limited to [-1, size] */
184 const float min
= -0.5F
;
185 const float max
= size
+ 0.5F
;
194 *icoord
= util_ifloor(s
);
198 wrap_nearest_mirror_repeat(float s
, unsigned size
, int offset
, int *icoord
)
200 const float min
= 1.0F
/ (2.0F
* size
);
201 const float max
= 1.0F
- min
;
205 s
+= (float)offset
/ size
;
206 flr
= util_ifloor(s
);
215 *icoord
= util_ifloor(u
* size
);
220 wrap_nearest_mirror_clamp(float s
, unsigned size
, int offset
, int *icoord
)
222 /* s limited to [0,1] */
223 /* i limited to [0,size-1] */
224 const float u
= fabsf(s
* size
+ offset
);
230 *icoord
= util_ifloor(u
);
235 wrap_nearest_mirror_clamp_to_edge(float s
, unsigned size
, int offset
, int *icoord
)
237 /* s limited to [min,max] */
238 /* i limited to [0, size-1] */
239 const float min
= 0.5F
;
240 const float max
= (float)size
- 0.5F
;
241 const float u
= fabsf(s
* size
+ offset
);
248 *icoord
= util_ifloor(u
);
253 wrap_nearest_mirror_clamp_to_border(float s
, unsigned size
, int offset
, int *icoord
)
255 /* u limited to [-0.5, size-0.5] */
256 const float min
= -0.5F
;
257 const float max
= (float)size
+ 0.5F
;
258 const float u
= fabsf(s
* size
+ offset
);
265 *icoord
= util_ifloor(u
);
270 * Used to compute texel locations for linear sampling
271 * \param wrapMode PIPE_TEX_WRAP_x
272 * \param s the texcoord
273 * \param size the texture image size
274 * \param icoord0 returns first texture index
275 * \param icoord1 returns second texture index (usually icoord0 + 1)
276 * \param w returns blend factor/weight between texture indices
277 * \param icoord returns the computed integer texture coord
280 wrap_linear_repeat(float s
, unsigned size
, int offset
,
281 int *icoord0
, int *icoord1
, float *w
)
283 float u
= s
* size
- 0.5F
;
284 *icoord0
= repeat(util_ifloor(u
) + offset
, size
);
285 *icoord1
= repeat(*icoord0
+ 1, size
);
291 wrap_linear_clamp(float s
, unsigned size
, int offset
,
292 int *icoord0
, int *icoord1
, float *w
)
294 float u
= CLAMP(s
* size
+ offset
, 0.0F
, (float)size
);
297 *icoord0
= util_ifloor(u
);
298 *icoord1
= *icoord0
+ 1;
304 wrap_linear_clamp_to_edge(float s
, unsigned size
, int offset
,
305 int *icoord0
, int *icoord1
, float *w
)
307 float u
= CLAMP(s
* size
+ offset
, 0.0F
, (float)size
);
309 *icoord0
= util_ifloor(u
);
310 *icoord1
= *icoord0
+ 1;
313 if (*icoord1
>= (int) size
)
320 wrap_linear_clamp_to_border(float s
, unsigned size
, int offset
,
321 int *icoord0
, int *icoord1
, float *w
)
323 const float min
= -0.5F
;
324 const float max
= (float)size
+ 0.5F
;
325 float u
= CLAMP(s
* size
+ offset
, min
, max
);
327 *icoord0
= util_ifloor(u
);
328 *icoord1
= *icoord0
+ 1;
334 wrap_linear_mirror_repeat(float s
, unsigned size
, int offset
,
335 int *icoord0
, int *icoord1
, float *w
)
340 s
+= (float)offset
/ size
;
341 flr
= util_ifloor(s
);
346 *icoord0
= util_ifloor(u
);
347 *icoord1
= *icoord0
+ 1;
350 if (*icoord1
>= (int) size
)
357 wrap_linear_mirror_clamp(float s
, unsigned size
, int offset
,
358 int *icoord0
, int *icoord1
, float *w
)
360 float u
= fabsf(s
* size
+ offset
);
364 *icoord0
= util_ifloor(u
);
365 *icoord1
= *icoord0
+ 1;
371 wrap_linear_mirror_clamp_to_edge(float s
, unsigned size
, int offset
,
372 int *icoord0
, int *icoord1
, float *w
)
374 float u
= fabsf(s
* size
+ offset
);
378 *icoord0
= util_ifloor(u
);
379 *icoord1
= *icoord0
+ 1;
382 if (*icoord1
>= (int) size
)
389 wrap_linear_mirror_clamp_to_border(float s
, unsigned size
, int offset
,
390 int *icoord0
, int *icoord1
, float *w
)
392 const float min
= -0.5F
;
393 const float max
= size
+ 0.5F
;
394 float u
= fabsf(s
* size
+ offset
);
400 *icoord0
= util_ifloor(u
);
401 *icoord1
= *icoord0
+ 1;
407 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
410 wrap_nearest_unorm_clamp(float s
, unsigned size
, int offset
, int *icoord
)
412 int i
= util_ifloor(s
);
413 *icoord
= CLAMP(i
+ offset
, 0, (int) size
-1);
418 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
421 wrap_nearest_unorm_clamp_to_border(float s
, unsigned size
, int offset
, int *icoord
)
423 *icoord
= util_ifloor( CLAMP(s
+ offset
, -0.5F
, (float) size
+ 0.5F
) );
428 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
431 wrap_nearest_unorm_clamp_to_edge(float s
, unsigned size
, int offset
, int *icoord
)
433 *icoord
= util_ifloor( CLAMP(s
+ offset
, 0.5F
, (float) size
- 0.5F
) );
438 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
441 wrap_linear_unorm_clamp(float s
, unsigned size
, int offset
,
442 int *icoord0
, int *icoord1
, float *w
)
444 /* Not exactly what the spec says, but it matches NVIDIA output */
445 float u
= CLAMP(s
+ offset
- 0.5F
, 0.0f
, (float) size
- 1.0f
);
446 *icoord0
= util_ifloor(u
);
447 *icoord1
= *icoord0
+ 1;
453 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
456 wrap_linear_unorm_clamp_to_border(float s
, unsigned size
, int offset
,
457 int *icoord0
, int *icoord1
, float *w
)
459 float u
= CLAMP(s
+ offset
, -0.5F
, (float) size
+ 0.5F
);
461 *icoord0
= util_ifloor(u
);
462 *icoord1
= *icoord0
+ 1;
463 if (*icoord1
> (int) size
- 1)
470 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
473 wrap_linear_unorm_clamp_to_edge(float s
, unsigned size
, int offset
,
474 int *icoord0
, int *icoord1
, float *w
)
476 float u
= CLAMP(s
+ offset
, +0.5F
, (float) size
- 0.5F
);
478 *icoord0
= util_ifloor(u
);
479 *icoord1
= *icoord0
+ 1;
480 if (*icoord1
> (int) size
- 1)
487 * Do coordinate to array index conversion. For array textures.
490 coord_to_layer(float coord
, unsigned first_layer
, unsigned last_layer
)
492 int c
= util_ifloor(coord
+ 0.5F
);
493 return CLAMP(c
, (int)first_layer
, (int)last_layer
);
498 * Examine the quad's texture coordinates to compute the partial
499 * derivatives w.r.t X and Y, then compute lambda (level of detail).
502 compute_lambda_1d(const struct sp_sampler_view
*sview
,
503 const float s
[TGSI_QUAD_SIZE
],
504 const float t
[TGSI_QUAD_SIZE
],
505 const float p
[TGSI_QUAD_SIZE
])
507 const struct pipe_resource
*texture
= sview
->base
.texture
;
508 float dsdx
= fabsf(s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]);
509 float dsdy
= fabsf(s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]);
510 float rho
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, sview
->base
.u
.tex
.first_level
);
512 return util_fast_log2(rho
);
517 compute_lambda_2d(const struct sp_sampler_view
*sview
,
518 const float s
[TGSI_QUAD_SIZE
],
519 const float t
[TGSI_QUAD_SIZE
],
520 const float p
[TGSI_QUAD_SIZE
])
522 const struct pipe_resource
*texture
= sview
->base
.texture
;
523 float dsdx
= fabsf(s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]);
524 float dsdy
= fabsf(s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]);
525 float dtdx
= fabsf(t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]);
526 float dtdy
= fabsf(t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]);
527 float maxx
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, sview
->base
.u
.tex
.first_level
);
528 float maxy
= MAX2(dtdx
, dtdy
) * u_minify(texture
->height0
, sview
->base
.u
.tex
.first_level
);
529 float rho
= MAX2(maxx
, maxy
);
531 return util_fast_log2(rho
);
536 compute_lambda_3d(const struct sp_sampler_view
*sview
,
537 const float s
[TGSI_QUAD_SIZE
],
538 const float t
[TGSI_QUAD_SIZE
],
539 const float p
[TGSI_QUAD_SIZE
])
541 const struct pipe_resource
*texture
= sview
->base
.texture
;
542 float dsdx
= fabsf(s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]);
543 float dsdy
= fabsf(s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]);
544 float dtdx
= fabsf(t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]);
545 float dtdy
= fabsf(t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]);
546 float dpdx
= fabsf(p
[QUAD_BOTTOM_RIGHT
] - p
[QUAD_BOTTOM_LEFT
]);
547 float dpdy
= fabsf(p
[QUAD_TOP_LEFT
] - p
[QUAD_BOTTOM_LEFT
]);
548 float maxx
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, sview
->base
.u
.tex
.first_level
);
549 float maxy
= MAX2(dtdx
, dtdy
) * u_minify(texture
->height0
, sview
->base
.u
.tex
.first_level
);
550 float maxz
= MAX2(dpdx
, dpdy
) * u_minify(texture
->depth0
, sview
->base
.u
.tex
.first_level
);
553 rho
= MAX2(maxx
, maxy
);
554 rho
= MAX2(rho
, maxz
);
556 return util_fast_log2(rho
);
561 * Compute lambda for a vertex texture sampler.
562 * Since there aren't derivatives to use, just return 0.
565 compute_lambda_vert(const struct sp_sampler_view
*sview
,
566 const float s
[TGSI_QUAD_SIZE
],
567 const float t
[TGSI_QUAD_SIZE
],
568 const float p
[TGSI_QUAD_SIZE
])
576 * Get a texel from a texture, using the texture tile cache.
578 * \param addr the template tex address containing cube, z, face info.
579 * \param x the x coord of texel within 2D image
580 * \param y the y coord of texel within 2D image
581 * \param rgba the quad to put the texel/color into
583 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
584 * sp_get_cached_tile_tex() function.
590 static inline const float *
591 get_texel_2d_no_border(const struct sp_sampler_view
*sp_sview
,
592 union tex_tile_address addr
, int x
, int y
)
594 const struct softpipe_tex_cached_tile
*tile
;
595 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
596 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
600 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
602 return &tile
->data
.color
[y
][x
][0];
606 static inline const float *
607 get_texel_2d(const struct sp_sampler_view
*sp_sview
,
608 const struct sp_sampler
*sp_samp
,
609 union tex_tile_address addr
, int x
, int y
)
611 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
612 unsigned level
= addr
.bits
.level
;
614 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
615 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
616 return sp_samp
->base
.border_color
.f
;
619 return get_texel_2d_no_border( sp_sview
, addr
, x
, y
);
625 * Here's the complete logic (HOLY CRAP) for finding next face and doing the
626 * corresponding coord wrapping, implemented by get_next_face,
627 * get_next_xcoord, get_next_ycoord.
628 * Read like that (first line):
629 * If face is +x and s coord is below zero, then
630 * new face is +z, new s is max , new t is old t
631 * (max is always cube size - 1).
633 * +x s- -> +z: s = max, t = t
634 * +x s+ -> -z: s = 0, t = t
635 * +x t- -> +y: s = max, t = max-s
636 * +x t+ -> -y: s = max, t = s
638 * -x s- -> -z: s = max, t = t
639 * -x s+ -> +z: s = 0, t = t
640 * -x t- -> +y: s = 0, t = s
641 * -x t+ -> -y: s = 0, t = max-s
643 * +y s- -> -x: s = t, t = 0
644 * +y s+ -> +x: s = max-t, t = 0
645 * +y t- -> -z: s = max-s, t = 0
646 * +y t+ -> +z: s = s, t = 0
648 * -y s- -> -x: s = max-t, t = max
649 * -y s+ -> +x: s = t, t = max
650 * -y t- -> +z: s = s, t = max
651 * -y t+ -> -z: s = max-s, t = max
653 * +z s- -> -x: s = max, t = t
654 * +z s+ -> +x: s = 0, t = t
655 * +z t- -> +y: s = s, t = max
656 * +z t+ -> -y: s = s, t = 0
658 * -z s- -> +x: s = max, t = t
659 * -z s+ -> -x: s = 0, t = t
660 * -z t- -> +y: s = max-s, t = 0
661 * -z t+ -> -y: s = max-s, t = max
666 * seamless cubemap neighbour array.
667 * this array is used to find the adjacent face in each of 4 directions,
668 * left, right, up, down. (or -x, +x, -y, +y).
670 static const unsigned face_array
[PIPE_TEX_FACE_MAX
][4] = {
671 /* pos X first then neg X is Z different, Y the same */
672 /* PIPE_TEX_FACE_POS_X,*/
673 { PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
,
674 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
675 /* PIPE_TEX_FACE_NEG_X */
676 { PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
,
677 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
679 /* pos Y first then neg Y is X different, X the same */
680 /* PIPE_TEX_FACE_POS_Y */
681 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
682 PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
},
684 /* PIPE_TEX_FACE_NEG_Y */
685 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
686 PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
},
688 /* pos Z first then neg Y is X different, X the same */
689 /* PIPE_TEX_FACE_POS_Z */
690 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
691 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
693 /* PIPE_TEX_FACE_NEG_Z */
694 { PIPE_TEX_FACE_POS_X
, PIPE_TEX_FACE_NEG_X
,
695 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
}
698 static inline unsigned
699 get_next_face(unsigned face
, int idx
)
701 return face_array
[face
][idx
];
705 * return a new xcoord based on old face, old coords, cube size
706 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
709 get_next_xcoord(unsigned face
, unsigned fall_off_index
, int max
, int xc
, int yc
)
711 if ((face
== 0 && fall_off_index
!= 1) ||
712 (face
== 1 && fall_off_index
== 0) ||
713 (face
== 4 && fall_off_index
== 0) ||
714 (face
== 5 && fall_off_index
== 0)) {
717 if ((face
== 1 && fall_off_index
!= 0) ||
718 (face
== 0 && fall_off_index
== 1) ||
719 (face
== 4 && fall_off_index
== 1) ||
720 (face
== 5 && fall_off_index
== 1)) {
723 if ((face
== 4 && fall_off_index
>= 2) ||
724 (face
== 2 && fall_off_index
== 3) ||
725 (face
== 3 && fall_off_index
== 2)) {
728 if ((face
== 5 && fall_off_index
>= 2) ||
729 (face
== 2 && fall_off_index
== 2) ||
730 (face
== 3 && fall_off_index
== 3)) {
733 if ((face
== 2 && fall_off_index
== 0) ||
734 (face
== 3 && fall_off_index
== 1)) {
737 /* (face == 2 && fall_off_index == 1) ||
738 (face == 3 && fall_off_index == 0)) */
743 * return a new ycoord based on old face, old coords, cube size
744 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
747 get_next_ycoord(unsigned face
, unsigned fall_off_index
, int max
, int xc
, int yc
)
749 if ((fall_off_index
<= 1) && (face
<= 1 || face
>= 4)) {
753 (face
== 4 && fall_off_index
== 3) ||
754 (face
== 5 && fall_off_index
== 2)) {
758 (face
== 4 && fall_off_index
== 2) ||
759 (face
== 5 && fall_off_index
== 3)) {
762 if ((face
== 0 && fall_off_index
== 3) ||
763 (face
== 1 && fall_off_index
== 2)) {
766 /* (face == 0 && fall_off_index == 2) ||
767 (face == 1 && fall_off_index == 3) */
772 /* Gather a quad of adjacent texels within a tile:
775 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_view
*sp_sview
,
776 union tex_tile_address addr
,
777 unsigned x
, unsigned y
,
780 const struct softpipe_tex_cached_tile
*tile
;
782 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
783 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
787 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
789 out
[0] = &tile
->data
.color
[y
][x
][0];
790 out
[1] = &tile
->data
.color
[y
][x
+1][0];
791 out
[2] = &tile
->data
.color
[y
+1][x
][0];
792 out
[3] = &tile
->data
.color
[y
+1][x
+1][0];
796 /* Gather a quad of potentially non-adjacent texels:
799 get_texel_quad_2d_no_border(const struct sp_sampler_view
*sp_sview
,
800 union tex_tile_address addr
,
805 out
[0] = get_texel_2d_no_border( sp_sview
, addr
, x0
, y0
);
806 out
[1] = get_texel_2d_no_border( sp_sview
, addr
, x1
, y0
);
807 out
[2] = get_texel_2d_no_border( sp_sview
, addr
, x0
, y1
);
808 out
[3] = get_texel_2d_no_border( sp_sview
, addr
, x1
, y1
);
811 /* Can involve a lot of unnecessary checks for border color:
814 get_texel_quad_2d(const struct sp_sampler_view
*sp_sview
,
815 const struct sp_sampler
*sp_samp
,
816 union tex_tile_address addr
,
821 out
[0] = get_texel_2d( sp_sview
, sp_samp
, addr
, x0
, y0
);
822 out
[1] = get_texel_2d( sp_sview
, sp_samp
, addr
, x1
, y0
);
823 out
[3] = get_texel_2d( sp_sview
, sp_samp
, addr
, x1
, y1
);
824 out
[2] = get_texel_2d( sp_sview
, sp_samp
, addr
, x0
, y1
);
831 static inline const float *
832 get_texel_3d_no_border(const struct sp_sampler_view
*sp_sview
,
833 union tex_tile_address addr
, int x
, int y
, int z
)
835 const struct softpipe_tex_cached_tile
*tile
;
837 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
838 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
843 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
845 return &tile
->data
.color
[y
][x
][0];
849 static inline const float *
850 get_texel_3d(const struct sp_sampler_view
*sp_sview
,
851 const struct sp_sampler
*sp_samp
,
852 union tex_tile_address addr
, int x
, int y
, int z
)
854 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
855 unsigned level
= addr
.bits
.level
;
857 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
858 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
) ||
859 z
< 0 || z
>= (int) u_minify(texture
->depth0
, level
)) {
860 return sp_samp
->base
.border_color
.f
;
863 return get_texel_3d_no_border( sp_sview
, addr
, x
, y
, z
);
868 /* Get texel pointer for 1D array texture */
869 static inline const float *
870 get_texel_1d_array(const struct sp_sampler_view
*sp_sview
,
871 const struct sp_sampler
*sp_samp
,
872 union tex_tile_address addr
, int x
, int y
)
874 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
875 unsigned level
= addr
.bits
.level
;
877 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
)) {
878 return sp_samp
->base
.border_color
.f
;
881 return get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
886 /* Get texel pointer for 2D array texture */
887 static inline const float *
888 get_texel_2d_array(const struct sp_sampler_view
*sp_sview
,
889 const struct sp_sampler
*sp_samp
,
890 union tex_tile_address addr
, int x
, int y
, int layer
)
892 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
893 unsigned level
= addr
.bits
.level
;
895 assert(layer
< (int) texture
->array_size
);
898 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
899 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
900 return sp_samp
->base
.border_color
.f
;
903 return get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
908 static inline const float *
909 get_texel_cube_seamless(const struct sp_sampler_view
*sp_sview
,
910 union tex_tile_address addr
, int x
, int y
,
911 float *corner
, int layer
, unsigned face
)
913 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
914 unsigned level
= addr
.bits
.level
;
915 int new_x
, new_y
, max_x
;
917 max_x
= (int) u_minify(texture
->width0
, level
);
919 assert(texture
->width0
== texture
->height0
);
923 /* change the face */
926 * Cheat with corners. They are difficult and I believe because we don't get
927 * per-pixel faces we can actually have multiple corner texels per pixel,
928 * which screws things up majorly in any case (as the per spec behavior is
929 * to average the 3 remaining texels, which we might not have).
930 * Hence just make sure that the 2nd coord is clamped, will simply pick the
931 * sample which would have fallen off the x coord, but not y coord.
932 * So the filter weight of the samples will be wrong, but at least this
933 * ensures that only valid texels near the corner are used.
935 if (y
< 0 || y
>= max_x
) {
936 y
= CLAMP(y
, 0, max_x
- 1);
938 new_x
= get_next_xcoord(face
, 0, max_x
-1, x
, y
);
939 new_y
= get_next_ycoord(face
, 0, max_x
-1, x
, y
);
940 face
= get_next_face(face
, 0);
941 } else if (x
>= max_x
) {
942 if (y
< 0 || y
>= max_x
) {
943 y
= CLAMP(y
, 0, max_x
- 1);
945 new_x
= get_next_xcoord(face
, 1, max_x
-1, x
, y
);
946 new_y
= get_next_ycoord(face
, 1, max_x
-1, x
, y
);
947 face
= get_next_face(face
, 1);
949 new_x
= get_next_xcoord(face
, 2, max_x
-1, x
, y
);
950 new_y
= get_next_ycoord(face
, 2, max_x
-1, x
, y
);
951 face
= get_next_face(face
, 2);
952 } else if (y
>= max_x
) {
953 new_x
= get_next_xcoord(face
, 3, max_x
-1, x
, y
);
954 new_y
= get_next_ycoord(face
, 3, max_x
-1, x
, y
);
955 face
= get_next_face(face
, 3);
958 return get_texel_3d_no_border(sp_sview
, addr
, new_x
, new_y
, layer
+ face
);
962 /* Get texel pointer for cube array texture */
963 static inline const float *
964 get_texel_cube_array(const struct sp_sampler_view
*sp_sview
,
965 const struct sp_sampler
*sp_samp
,
966 union tex_tile_address addr
, int x
, int y
, int layer
)
968 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
969 unsigned level
= addr
.bits
.level
;
971 assert(layer
< (int) texture
->array_size
);
974 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
975 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
976 return sp_samp
->base
.border_color
.f
;
979 return get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
983 * Given the logbase2 of a mipmap's base level size and a mipmap level,
984 * return the size (in texels) of that mipmap level.
985 * For example, if level[0].width = 256 then base_pot will be 8.
986 * If level = 2, then we'll return 64 (the width at level=2).
987 * Return 1 if level > base_pot.
989 static inline unsigned
990 pot_level_size(unsigned base_pot
, unsigned level
)
992 return (base_pot
>= level
) ? (1 << (base_pot
- level
)) : 1;
997 print_sample(const char *function
, const float *rgba
)
999 debug_printf("%s %g %g %g %g\n",
1001 rgba
[0], rgba
[TGSI_NUM_CHANNELS
], rgba
[2*TGSI_NUM_CHANNELS
], rgba
[3*TGSI_NUM_CHANNELS
]);
1006 print_sample_4(const char *function
, float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1008 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
1010 rgba
[0][0], rgba
[1][0], rgba
[2][0], rgba
[3][0],
1011 rgba
[0][1], rgba
[1][1], rgba
[2][1], rgba
[3][1],
1012 rgba
[0][2], rgba
[1][2], rgba
[2][2], rgba
[3][2],
1013 rgba
[0][3], rgba
[1][3], rgba
[2][3], rgba
[3][3]);
1017 /* Some image-filter fastpaths:
1020 img_filter_2d_linear_repeat_POT(struct sp_sampler_view
*sp_sview
,
1021 struct sp_sampler
*sp_samp
,
1022 const struct img_filter_args
*args
,
1025 unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1026 unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1027 int xmax
= (xpot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
1028 int ymax
= (ypot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
1029 union tex_tile_address addr
;
1032 float u
= (args
->s
* xpot
- 0.5F
) + args
->offset
[0];
1033 float v
= (args
->t
* ypot
- 0.5F
) + args
->offset
[1];
1035 int uflr
= util_ifloor(u
);
1036 int vflr
= util_ifloor(v
);
1038 float xw
= u
- (float)uflr
;
1039 float yw
= v
- (float)vflr
;
1041 int x0
= uflr
& (xpot
- 1);
1042 int y0
= vflr
& (ypot
- 1);
1047 addr
.bits
.level
= args
->level
;
1049 /* Can we fetch all four at once:
1051 if (x0
< xmax
&& y0
< ymax
) {
1052 get_texel_quad_2d_no_border_single_tile(sp_sview
, addr
, x0
, y0
, tx
);
1055 unsigned x1
= (x0
+ 1) & (xpot
- 1);
1056 unsigned y1
= (y0
+ 1) & (ypot
- 1);
1057 get_texel_quad_2d_no_border(sp_sview
, addr
, x0
, y0
, x1
, y1
, tx
);
1060 /* interpolate R, G, B, A */
1061 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++) {
1062 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1064 tx
[2][c
], tx
[3][c
]);
1068 print_sample(__FUNCTION__
, rgba
);
1074 img_filter_2d_nearest_repeat_POT(struct sp_sampler_view
*sp_sview
,
1075 struct sp_sampler
*sp_samp
,
1076 const struct img_filter_args
*args
,
1077 float rgba
[TGSI_QUAD_SIZE
])
1079 unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1080 unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1082 union tex_tile_address addr
;
1085 float u
= args
->s
* xpot
+ args
->offset
[0];
1086 float v
= args
->t
* ypot
+ args
->offset
[1];
1088 int uflr
= util_ifloor(u
);
1089 int vflr
= util_ifloor(v
);
1091 int x0
= uflr
& (xpot
- 1);
1092 int y0
= vflr
& (ypot
- 1);
1095 addr
.bits
.level
= args
->level
;
1097 out
= get_texel_2d_no_border(sp_sview
, addr
, x0
, y0
);
1098 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1099 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1102 print_sample(__FUNCTION__
, rgba
);
1108 img_filter_2d_nearest_clamp_POT(struct sp_sampler_view
*sp_sview
,
1109 struct sp_sampler
*sp_samp
,
1110 const struct img_filter_args
*args
,
1111 float rgba
[TGSI_QUAD_SIZE
])
1113 unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1114 unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1115 union tex_tile_address addr
;
1118 float u
= args
->s
* xpot
+ args
->offset
[0];
1119 float v
= args
->t
* ypot
+ args
->offset
[1];
1125 addr
.bits
.level
= args
->level
;
1127 x0
= util_ifloor(u
);
1130 else if (x0
> (int) xpot
- 1)
1133 y0
= util_ifloor(v
);
1136 else if (y0
> (int) ypot
- 1)
1139 out
= get_texel_2d_no_border(sp_sview
, addr
, x0
, y0
);
1140 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1141 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1144 print_sample(__FUNCTION__
, rgba
);
1150 img_filter_1d_nearest(struct sp_sampler_view
*sp_sview
,
1151 struct sp_sampler
*sp_samp
,
1152 const struct img_filter_args
*args
,
1153 float rgba
[TGSI_QUAD_SIZE
])
1155 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1158 union tex_tile_address addr
;
1162 width
= u_minify(texture
->width0
, args
->level
);
1167 addr
.bits
.level
= args
->level
;
1169 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1171 out
= get_texel_2d(sp_sview
, sp_samp
, addr
, x
, 0);
1172 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1173 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1176 print_sample(__FUNCTION__
, rgba
);
1182 img_filter_1d_array_nearest(struct sp_sampler_view
*sp_sview
,
1183 struct sp_sampler
*sp_samp
,
1184 const struct img_filter_args
*args
,
1187 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1190 union tex_tile_address addr
;
1194 width
= u_minify(texture
->width0
, args
->level
);
1199 addr
.bits
.level
= args
->level
;
1201 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1202 layer
= coord_to_layer(args
->t
, sp_sview
->base
.u
.tex
.first_layer
,
1203 sp_sview
->base
.u
.tex
.last_layer
);
1205 out
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x
, layer
);
1206 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1207 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1210 print_sample(__FUNCTION__
, rgba
);
1216 img_filter_2d_nearest(struct sp_sampler_view
*sp_sview
,
1217 struct sp_sampler
*sp_samp
,
1218 const struct img_filter_args
*args
,
1221 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1224 union tex_tile_address addr
;
1228 width
= u_minify(texture
->width0
, args
->level
);
1229 height
= u_minify(texture
->height0
, args
->level
);
1235 addr
.bits
.level
= args
->level
;
1237 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1238 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1240 out
= get_texel_2d(sp_sview
, sp_samp
, addr
, x
, y
);
1241 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1242 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1245 print_sample(__FUNCTION__
, rgba
);
1251 img_filter_2d_array_nearest(struct sp_sampler_view
*sp_sview
,
1252 struct sp_sampler
*sp_samp
,
1253 const struct img_filter_args
*args
,
1256 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1259 union tex_tile_address addr
;
1263 width
= u_minify(texture
->width0
, args
->level
);
1264 height
= u_minify(texture
->height0
, args
->level
);
1270 addr
.bits
.level
= args
->level
;
1272 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1273 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1274 layer
= coord_to_layer(args
->p
, sp_sview
->base
.u
.tex
.first_layer
,
1275 sp_sview
->base
.u
.tex
.last_layer
);
1277 out
= get_texel_2d_array(sp_sview
, sp_samp
, addr
, x
, y
, layer
);
1278 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1279 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1282 print_sample(__FUNCTION__
, rgba
);
1288 img_filter_cube_nearest(struct sp_sampler_view
*sp_sview
,
1289 struct sp_sampler
*sp_samp
,
1290 const struct img_filter_args
*args
,
1293 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1295 int x
, y
, layerface
;
1296 union tex_tile_address addr
;
1300 width
= u_minify(texture
->width0
, args
->level
);
1301 height
= u_minify(texture
->height0
, args
->level
);
1307 addr
.bits
.level
= args
->level
;
1310 * If NEAREST filtering is done within a miplevel, always apply wrap
1311 * mode CLAMP_TO_EDGE.
1313 if (sp_samp
->base
.seamless_cube_map
) {
1314 wrap_nearest_clamp_to_edge(args
->s
, width
, args
->offset
[0], &x
);
1315 wrap_nearest_clamp_to_edge(args
->t
, height
, args
->offset
[1], &y
);
1317 /* Would probably make sense to ignore mode and just do edge clamp */
1318 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1319 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1322 layerface
= args
->face_id
+ sp_sview
->base
.u
.tex
.first_layer
;
1323 out
= get_texel_cube_array(sp_sview
, sp_samp
, addr
, x
, y
, layerface
);
1324 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1325 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1328 print_sample(__FUNCTION__
, rgba
);
1333 img_filter_cube_array_nearest(struct sp_sampler_view
*sp_sview
,
1334 struct sp_sampler
*sp_samp
,
1335 const struct img_filter_args
*args
,
1338 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1340 int x
, y
, layerface
;
1341 union tex_tile_address addr
;
1345 width
= u_minify(texture
->width0
, args
->level
);
1346 height
= u_minify(texture
->height0
, args
->level
);
1352 addr
.bits
.level
= args
->level
;
1354 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1355 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1356 layerface
= coord_to_layer(6 * args
->p
+ sp_sview
->base
.u
.tex
.first_layer
,
1357 sp_sview
->base
.u
.tex
.first_layer
,
1358 sp_sview
->base
.u
.tex
.last_layer
- 5) + args
->face_id
;
1360 out
= get_texel_cube_array(sp_sview
, sp_samp
, addr
, x
, y
, layerface
);
1361 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1362 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1365 print_sample(__FUNCTION__
, rgba
);
1370 img_filter_3d_nearest(struct sp_sampler_view
*sp_sview
,
1371 struct sp_sampler
*sp_samp
,
1372 const struct img_filter_args
*args
,
1375 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1376 int width
, height
, depth
;
1378 union tex_tile_address addr
;
1382 width
= u_minify(texture
->width0
, args
->level
);
1383 height
= u_minify(texture
->height0
, args
->level
);
1384 depth
= u_minify(texture
->depth0
, args
->level
);
1390 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1391 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1392 sp_samp
->nearest_texcoord_p(args
->p
, depth
, args
->offset
[2], &z
);
1395 addr
.bits
.level
= args
->level
;
1397 out
= get_texel_3d(sp_sview
, sp_samp
, addr
, x
, y
, z
);
1398 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1399 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1404 img_filter_1d_linear(struct sp_sampler_view
*sp_sview
,
1405 struct sp_sampler
*sp_samp
,
1406 const struct img_filter_args
*args
,
1409 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1412 float xw
; /* weights */
1413 union tex_tile_address addr
;
1414 const float *tx0
, *tx1
;
1417 width
= u_minify(texture
->width0
, args
->level
);
1422 addr
.bits
.level
= args
->level
;
1424 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1426 tx0
= get_texel_2d(sp_sview
, sp_samp
, addr
, x0
, 0);
1427 tx1
= get_texel_2d(sp_sview
, sp_samp
, addr
, x1
, 0);
1429 /* interpolate R, G, B, A */
1430 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1431 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1436 img_filter_1d_array_linear(struct sp_sampler_view
*sp_sview
,
1437 struct sp_sampler
*sp_samp
,
1438 const struct img_filter_args
*args
,
1441 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1444 float xw
; /* weights */
1445 union tex_tile_address addr
;
1446 const float *tx0
, *tx1
;
1449 width
= u_minify(texture
->width0
, args
->level
);
1454 addr
.bits
.level
= args
->level
;
1456 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1457 layer
= coord_to_layer(args
->t
, sp_sview
->base
.u
.tex
.first_layer
,
1458 sp_sview
->base
.u
.tex
.last_layer
);
1460 tx0
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x0
, layer
);
1461 tx1
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x1
, layer
);
1463 /* interpolate R, G, B, A */
1464 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1465 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1469 * Retrieve the gathered value, need to convert to the
1470 * TGSI expected interface, and take component select
1471 * and swizzling into account.
1474 get_gather_value(const struct sp_sampler_view
*sp_sview
,
1475 int chan_in
, int comp_sel
,
1482 * softpipe samples in a different order
1483 * to TGSI expects, so we need to swizzle,
1484 * the samples into the correct slots.
1504 /* pick which component to use for the swizzle */
1507 swizzle
= sp_sview
->base
.swizzle_r
;
1510 swizzle
= sp_sview
->base
.swizzle_g
;
1513 swizzle
= sp_sview
->base
.swizzle_b
;
1516 swizzle
= sp_sview
->base
.swizzle_a
;
1523 /* get correct result using the channel and swizzle */
1525 case PIPE_SWIZZLE_ZERO
:
1527 case PIPE_SWIZZLE_ONE
:
1530 return tx
[chan
][swizzle
];
1536 img_filter_2d_linear(struct sp_sampler_view
*sp_sview
,
1537 struct sp_sampler
*sp_samp
,
1538 const struct img_filter_args
*args
,
1541 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1544 float xw
, yw
; /* weights */
1545 union tex_tile_address addr
;
1549 width
= u_minify(texture
->width0
, args
->level
);
1550 height
= u_minify(texture
->height0
, args
->level
);
1556 addr
.bits
.level
= args
->level
;
1558 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1559 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1561 tx
[0] = get_texel_2d(sp_sview
, sp_samp
, addr
, x0
, y0
);
1562 tx
[1] = get_texel_2d(sp_sview
, sp_samp
, addr
, x1
, y0
);
1563 tx
[2] = get_texel_2d(sp_sview
, sp_samp
, addr
, x0
, y1
);
1564 tx
[3] = get_texel_2d(sp_sview
, sp_samp
, addr
, x1
, y1
);
1566 if (args
->gather_only
) {
1567 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1568 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1572 /* interpolate R, G, B, A */
1573 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1574 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1576 tx
[2][c
], tx
[3][c
]);
1582 img_filter_2d_array_linear(struct sp_sampler_view
*sp_sview
,
1583 struct sp_sampler
*sp_samp
,
1584 const struct img_filter_args
*args
,
1587 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1589 int x0
, y0
, x1
, y1
, layer
;
1590 float xw
, yw
; /* weights */
1591 union tex_tile_address addr
;
1595 width
= u_minify(texture
->width0
, args
->level
);
1596 height
= u_minify(texture
->height0
, args
->level
);
1602 addr
.bits
.level
= args
->level
;
1604 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1605 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1606 layer
= coord_to_layer(args
->p
, sp_sview
->base
.u
.tex
.first_layer
,
1607 sp_sview
->base
.u
.tex
.last_layer
);
1609 tx
[0] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
);
1610 tx
[1] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
);
1611 tx
[2] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
);
1612 tx
[3] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
);
1614 if (args
->gather_only
) {
1615 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1616 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1620 /* interpolate R, G, B, A */
1621 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1622 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1624 tx
[2][c
], tx
[3][c
]);
1630 img_filter_cube_linear(struct sp_sampler_view
*sp_sview
,
1631 struct sp_sampler
*sp_samp
,
1632 const struct img_filter_args
*args
,
1635 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1637 int x0
, y0
, x1
, y1
, layer
;
1638 float xw
, yw
; /* weights */
1639 union tex_tile_address addr
;
1641 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
],
1642 corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1645 width
= u_minify(texture
->width0
, args
->level
);
1646 height
= u_minify(texture
->height0
, args
->level
);
1652 addr
.bits
.level
= args
->level
;
1655 * For seamless if LINEAR filtering is done within a miplevel,
1656 * always apply wrap mode CLAMP_TO_BORDER.
1658 if (sp_samp
->base
.seamless_cube_map
) {
1659 /* Note this is a bit overkill, actual clamping is not required */
1660 wrap_linear_clamp_to_border(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1661 wrap_linear_clamp_to_border(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1663 /* Would probably make sense to ignore mode and just do edge clamp */
1664 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1665 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1668 layer
= sp_sview
->base
.u
.tex
.first_layer
;
1670 if (sp_samp
->base
.seamless_cube_map
) {
1671 tx
[0] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y0
, corner0
, layer
, args
->face_id
);
1672 tx
[1] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y0
, corner1
, layer
, args
->face_id
);
1673 tx
[2] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y1
, corner2
, layer
, args
->face_id
);
1674 tx
[3] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y1
, corner3
, layer
, args
->face_id
);
1676 tx
[0] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
+ args
->face_id
);
1677 tx
[1] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
+ args
->face_id
);
1678 tx
[2] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
+ args
->face_id
);
1679 tx
[3] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
+ args
->face_id
);
1682 if (args
->gather_only
) {
1683 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1684 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1688 /* interpolate R, G, B, A */
1689 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1690 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1692 tx
[2][c
], tx
[3][c
]);
1698 img_filter_cube_array_linear(struct sp_sampler_view
*sp_sview
,
1699 struct sp_sampler
*sp_samp
,
1700 const struct img_filter_args
*args
,
1703 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1705 int x0
, y0
, x1
, y1
, layer
;
1706 float xw
, yw
; /* weights */
1707 union tex_tile_address addr
;
1709 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
],
1710 corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1713 width
= u_minify(texture
->width0
, args
->level
);
1714 height
= u_minify(texture
->height0
, args
->level
);
1720 addr
.bits
.level
= args
->level
;
1723 * For seamless if LINEAR filtering is done within a miplevel,
1724 * always apply wrap mode CLAMP_TO_BORDER.
1726 if (sp_samp
->base
.seamless_cube_map
) {
1727 /* Note this is a bit overkill, actual clamping is not required */
1728 wrap_linear_clamp_to_border(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1729 wrap_linear_clamp_to_border(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1731 /* Would probably make sense to ignore mode and just do edge clamp */
1732 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1733 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1736 layer
= coord_to_layer(6 * args
->p
+ sp_sview
->base
.u
.tex
.first_layer
,
1737 sp_sview
->base
.u
.tex
.first_layer
,
1738 sp_sview
->base
.u
.tex
.last_layer
- 5);
1740 if (sp_samp
->base
.seamless_cube_map
) {
1741 tx
[0] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y0
, corner0
, layer
, args
->face_id
);
1742 tx
[1] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y0
, corner1
, layer
, args
->face_id
);
1743 tx
[2] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y1
, corner2
, layer
, args
->face_id
);
1744 tx
[3] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y1
, corner3
, layer
, args
->face_id
);
1746 tx
[0] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
+ args
->face_id
);
1747 tx
[1] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
+ args
->face_id
);
1748 tx
[2] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
+ args
->face_id
);
1749 tx
[3] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
+ args
->face_id
);
1752 if (args
->gather_only
) {
1753 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1754 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1758 /* interpolate R, G, B, A */
1759 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1760 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1762 tx
[2][c
], tx
[3][c
]);
1767 img_filter_3d_linear(struct sp_sampler_view
*sp_sview
,
1768 struct sp_sampler
*sp_samp
,
1769 const struct img_filter_args
*args
,
1772 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1773 int width
, height
, depth
;
1774 int x0
, x1
, y0
, y1
, z0
, z1
;
1775 float xw
, yw
, zw
; /* interpolation weights */
1776 union tex_tile_address addr
;
1777 const float *tx00
, *tx01
, *tx02
, *tx03
, *tx10
, *tx11
, *tx12
, *tx13
;
1780 width
= u_minify(texture
->width0
, args
->level
);
1781 height
= u_minify(texture
->height0
, args
->level
);
1782 depth
= u_minify(texture
->depth0
, args
->level
);
1785 addr
.bits
.level
= args
->level
;
1791 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1792 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1793 sp_samp
->linear_texcoord_p(args
->p
, depth
, args
->offset
[2], &z0
, &z1
, &zw
);
1795 tx00
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y0
, z0
);
1796 tx01
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y0
, z0
);
1797 tx02
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y1
, z0
);
1798 tx03
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y1
, z0
);
1800 tx10
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y0
, z1
);
1801 tx11
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y0
, z1
);
1802 tx12
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y1
, z1
);
1803 tx13
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y1
, z1
);
1805 /* interpolate R, G, B, A */
1806 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1807 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_3d(xw
, yw
, zw
,
1815 /* Calculate level of detail for every fragment,
1816 * with lambda already computed.
1817 * Note that lambda has already been biased by global LOD bias.
1818 * \param biased_lambda per-quad lambda.
1819 * \param lod_in per-fragment lod_bias or explicit_lod.
1820 * \param lod returns the per-fragment lod.
1823 compute_lod(const struct pipe_sampler_state
*sampler
,
1824 enum tgsi_sampler_control control
,
1825 const float biased_lambda
,
1826 const float lod_in
[TGSI_QUAD_SIZE
],
1827 float lod
[TGSI_QUAD_SIZE
])
1829 float min_lod
= sampler
->min_lod
;
1830 float max_lod
= sampler
->max_lod
;
1834 case tgsi_sampler_lod_none
:
1835 case tgsi_sampler_lod_zero
:
1837 case tgsi_sampler_derivs_explicit
:
1838 lod
[0] = lod
[1] = lod
[2] = lod
[3] = CLAMP(biased_lambda
, min_lod
, max_lod
);
1840 case tgsi_sampler_lod_bias
:
1841 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1842 lod
[i
] = biased_lambda
+ lod_in
[i
];
1843 lod
[i
] = CLAMP(lod
[i
], min_lod
, max_lod
);
1846 case tgsi_sampler_lod_explicit
:
1847 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1848 lod
[i
] = CLAMP(lod_in
[i
], min_lod
, max_lod
);
1853 lod
[0] = lod
[1] = lod
[2] = lod
[3] = 0.0f
;
1858 /* Calculate level of detail for every fragment. The computed value is not
1859 * clamped to lod_min and lod_max.
1860 * \param lod_in per-fragment lod_bias or explicit_lod.
1861 * \param lod results per-fragment lod.
1864 compute_lambda_lod_unclamped(struct sp_sampler_view
*sp_sview
,
1865 struct sp_sampler
*sp_samp
,
1866 const float s
[TGSI_QUAD_SIZE
],
1867 const float t
[TGSI_QUAD_SIZE
],
1868 const float p
[TGSI_QUAD_SIZE
],
1869 const float lod_in
[TGSI_QUAD_SIZE
],
1870 enum tgsi_sampler_control control
,
1871 float lod
[TGSI_QUAD_SIZE
])
1873 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
1874 const float lod_bias
= sampler
->lod_bias
;
1879 case tgsi_sampler_lod_none
:
1881 case tgsi_sampler_derivs_explicit
:
1882 lambda
= sp_sview
->compute_lambda(sp_sview
, s
, t
, p
) + lod_bias
;
1883 lod
[0] = lod
[1] = lod
[2] = lod
[3] = lambda
;
1885 case tgsi_sampler_lod_bias
:
1886 lambda
= sp_sview
->compute_lambda(sp_sview
, s
, t
, p
) + lod_bias
;
1887 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1888 lod
[i
] = lambda
+ lod_in
[i
];
1891 case tgsi_sampler_lod_explicit
:
1892 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1893 lod
[i
] = lod_in
[i
] + lod_bias
;
1896 case tgsi_sampler_lod_zero
:
1897 case tgsi_sampler_gather
:
1898 lod
[0] = lod
[1] = lod
[2] = lod
[3] = lod_bias
;
1902 lod
[0] = lod
[1] = lod
[2] = lod
[3] = 0.0f
;
1906 /* Calculate level of detail for every fragment.
1907 * \param lod_in per-fragment lod_bias or explicit_lod.
1908 * \param lod results per-fragment lod.
1911 compute_lambda_lod(struct sp_sampler_view
*sp_sview
,
1912 struct sp_sampler
*sp_samp
,
1913 const float s
[TGSI_QUAD_SIZE
],
1914 const float t
[TGSI_QUAD_SIZE
],
1915 const float p
[TGSI_QUAD_SIZE
],
1916 const float lod_in
[TGSI_QUAD_SIZE
],
1917 enum tgsi_sampler_control control
,
1918 float lod
[TGSI_QUAD_SIZE
])
1920 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
1921 const float min_lod
= sampler
->min_lod
;
1922 const float max_lod
= sampler
->max_lod
;
1925 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
1926 s
, t
, p
, lod_in
, control
, lod
);
1927 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1928 lod
[i
] = CLAMP(lod
[i
], min_lod
, max_lod
);
1932 static inline unsigned
1933 get_gather_component(const float lod_in
[TGSI_QUAD_SIZE
])
1935 /* gather component is stored in lod_in slot as unsigned */
1936 return (*(unsigned int *)lod_in
) & 0x3;
1940 mip_filter_linear(struct sp_sampler_view
*sp_sview
,
1941 struct sp_sampler
*sp_samp
,
1942 img_filter_func min_filter
,
1943 img_filter_func mag_filter
,
1944 const float s
[TGSI_QUAD_SIZE
],
1945 const float t
[TGSI_QUAD_SIZE
],
1946 const float p
[TGSI_QUAD_SIZE
],
1947 const float c0
[TGSI_QUAD_SIZE
],
1948 const float lod_in
[TGSI_QUAD_SIZE
],
1949 const struct filter_args
*filt_args
,
1950 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1952 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
1954 float lod
[TGSI_QUAD_SIZE
];
1955 struct img_filter_args args
;
1957 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
1959 args
.offset
= filt_args
->offset
;
1960 args
.gather_only
= filt_args
->control
== tgsi_sampler_gather
;
1961 args
.gather_comp
= get_gather_component(lod_in
);
1963 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1964 int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
1969 args
.face_id
= sp_sview
->faces
[j
];
1972 args
.level
= psview
->u
.tex
.first_level
;
1973 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
1975 else if (level0
>= (int) psview
->u
.tex
.last_level
) {
1976 args
.level
= psview
->u
.tex
.last_level
;
1977 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
1980 float levelBlend
= frac(lod
[j
]);
1981 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1984 args
.level
= level0
;
1985 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
1986 args
.level
= level0
+1;
1987 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
1989 for (c
= 0; c
< 4; c
++) {
1990 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
1996 print_sample_4(__FUNCTION__
, rgba
);
2002 * Compute nearest mipmap level from texcoords.
2003 * Then sample the texture level for four elements of a quad.
2004 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
2007 mip_filter_nearest(struct sp_sampler_view
*sp_sview
,
2008 struct sp_sampler
*sp_samp
,
2009 img_filter_func min_filter
,
2010 img_filter_func mag_filter
,
2011 const float s
[TGSI_QUAD_SIZE
],
2012 const float t
[TGSI_QUAD_SIZE
],
2013 const float p
[TGSI_QUAD_SIZE
],
2014 const float c0
[TGSI_QUAD_SIZE
],
2015 const float lod_in
[TGSI_QUAD_SIZE
],
2016 const struct filter_args
*filt_args
,
2017 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2019 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2020 float lod
[TGSI_QUAD_SIZE
];
2022 struct img_filter_args args
;
2024 args
.offset
= filt_args
->offset
;
2025 args
.gather_only
= filt_args
->control
== tgsi_sampler_gather
;
2026 args
.gather_comp
= get_gather_component(lod_in
);
2028 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2030 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2034 args
.face_id
= sp_sview
->faces
[j
];
2037 args
.level
= psview
->u
.tex
.first_level
;
2038 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2040 int level
= psview
->u
.tex
.first_level
+ (int)(lod
[j
] + 0.5F
);
2041 args
.level
= MIN2(level
, (int)psview
->u
.tex
.last_level
);
2042 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2047 print_sample_4(__FUNCTION__
, rgba
);
2053 mip_filter_none(struct sp_sampler_view
*sp_sview
,
2054 struct sp_sampler
*sp_samp
,
2055 img_filter_func min_filter
,
2056 img_filter_func mag_filter
,
2057 const float s
[TGSI_QUAD_SIZE
],
2058 const float t
[TGSI_QUAD_SIZE
],
2059 const float p
[TGSI_QUAD_SIZE
],
2060 const float c0
[TGSI_QUAD_SIZE
],
2061 const float lod_in
[TGSI_QUAD_SIZE
],
2062 const struct filter_args
*filt_args
,
2063 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2065 float lod
[TGSI_QUAD_SIZE
];
2067 struct img_filter_args args
;
2069 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2070 args
.offset
= filt_args
->offset
;
2071 args
.gather_only
= filt_args
->control
== tgsi_sampler_gather
;
2073 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2075 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2079 args
.face_id
= sp_sview
->faces
[j
];
2081 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2084 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2091 mip_filter_none_no_filter_select(struct sp_sampler_view
*sp_sview
,
2092 struct sp_sampler
*sp_samp
,
2093 img_filter_func min_filter
,
2094 img_filter_func mag_filter
,
2095 const float s
[TGSI_QUAD_SIZE
],
2096 const float t
[TGSI_QUAD_SIZE
],
2097 const float p
[TGSI_QUAD_SIZE
],
2098 const float c0
[TGSI_QUAD_SIZE
],
2099 const float lod_in
[TGSI_QUAD_SIZE
],
2100 const struct filter_args
*filt_args
,
2101 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2104 struct img_filter_args args
;
2105 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2106 args
.offset
= filt_args
->offset
;
2107 args
.gather_only
= filt_args
->control
== tgsi_sampler_gather
;
2108 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2112 args
.face_id
= sp_sview
->faces
[j
];
2113 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2118 /* For anisotropic filtering */
2119 #define WEIGHT_LUT_SIZE 1024
2121 static float *weightLut
= NULL
;
2124 * Creates the look-up table used to speed-up EWA sampling
2127 create_filter_table(void)
2131 weightLut
= (float *) MALLOC(WEIGHT_LUT_SIZE
* sizeof(float));
2133 for (i
= 0; i
< WEIGHT_LUT_SIZE
; ++i
) {
2135 float r2
= (float) i
/ (float) (WEIGHT_LUT_SIZE
- 1);
2136 float weight
= (float) exp(-alpha
* r2
);
2137 weightLut
[i
] = weight
;
2144 * Elliptical weighted average (EWA) filter for producing high quality
2145 * anisotropic filtered results.
2146 * Based on the Higher Quality Elliptical Weighted Average Filter
2147 * published by Paul S. Heckbert in his Master's Thesis
2148 * "Fundamentals of Texture Mapping and Image Warping" (1989)
2151 img_filter_2d_ewa(struct sp_sampler_view
*sp_sview
,
2152 struct sp_sampler
*sp_samp
,
2153 img_filter_func min_filter
,
2154 img_filter_func mag_filter
,
2155 const float s
[TGSI_QUAD_SIZE
],
2156 const float t
[TGSI_QUAD_SIZE
],
2157 const float p
[TGSI_QUAD_SIZE
],
2159 const float dudx
, const float dvdx
,
2160 const float dudy
, const float dvdy
,
2161 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2163 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2165 // ??? Won't the image filters blow up if level is negative?
2166 unsigned level0
= level
> 0 ? level
: 0;
2167 float scaling
= 1.0f
/ (1 << level0
);
2168 int width
= u_minify(texture
->width0
, level0
);
2169 int height
= u_minify(texture
->height0
, level0
);
2170 struct img_filter_args args
;
2171 float ux
= dudx
* scaling
;
2172 float vx
= dvdx
* scaling
;
2173 float uy
= dudy
* scaling
;
2174 float vy
= dvdy
* scaling
;
2176 /* compute ellipse coefficients to bound the region:
2177 * A*x*x + B*x*y + C*y*y = F.
2179 float A
= vx
*vx
+vy
*vy
+1;
2180 float B
= -2*(ux
*vx
+uy
*vy
);
2181 float C
= ux
*ux
+uy
*uy
+1;
2182 float F
= A
*C
-B
*B
/4.0f
;
2184 /* check if it is an ellipse */
2185 /* assert(F > 0.0); */
2187 /* Compute the ellipse's (u,v) bounding box in texture space */
2188 float d
= -B
*B
+4.0f
*C
*A
;
2189 float box_u
= 2.0f
/ d
* sqrtf(d
*C
*F
); /* box_u -> half of bbox with */
2190 float box_v
= 2.0f
/ d
* sqrtf(A
*d
*F
); /* box_v -> half of bbox height */
2192 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2193 float s_buffer
[TGSI_QUAD_SIZE
];
2194 float t_buffer
[TGSI_QUAD_SIZE
];
2195 float weight_buffer
[TGSI_QUAD_SIZE
];
2196 unsigned buffer_next
;
2198 float den
; /* = 0.0F; */
2200 float U
; /* = u0 - tex_u; */
2203 /* Scale ellipse formula to directly index the Filter Lookup Table.
2204 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2206 double formScale
= (double) (WEIGHT_LUT_SIZE
- 1) / F
;
2210 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2212 /* For each quad, the du and dx values are the same and so the ellipse is
2213 * also the same. Note that texel/image access can only be performed using
2214 * a quad, i.e. it is not possible to get the pixel value for a single
2215 * tex coord. In order to have a better performance, the access is buffered
2216 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2217 * full, then the pixel values are read from the image.
2222 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2223 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2224 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2225 * value, q, is less than F, we're inside the ellipse
2227 float tex_u
= -0.5F
+ s
[j
] * texture
->width0
* scaling
;
2228 float tex_v
= -0.5F
+ t
[j
] * texture
->height0
* scaling
;
2230 int u0
= (int) floorf(tex_u
- box_u
);
2231 int u1
= (int) ceilf(tex_u
+ box_u
);
2232 int v0
= (int) floorf(tex_v
- box_v
);
2233 int v1
= (int) ceilf(tex_v
+ box_v
);
2235 float num
[4] = {0.0F
, 0.0F
, 0.0F
, 0.0F
};
2238 args
.face_id
= sp_sview
->faces
[j
];
2241 for (v
= v0
; v
<= v1
; ++v
) {
2242 float V
= v
- tex_v
;
2243 float dq
= A
* (2 * U
+ 1) + B
* V
;
2244 float q
= (C
* V
+ B
* U
) * V
+ A
* U
* U
;
2247 for (u
= u0
; u
<= u1
; ++u
) {
2248 /* Note that the ellipse has been pre-scaled so F =
2249 * WEIGHT_LUT_SIZE - 1
2251 if (q
< WEIGHT_LUT_SIZE
) {
2252 /* as a LUT is used, q must never be negative;
2253 * should not happen, though
2255 const int qClamped
= q
>= 0.0F
? q
: 0;
2256 float weight
= weightLut
[qClamped
];
2258 weight_buffer
[buffer_next
] = weight
;
2259 s_buffer
[buffer_next
] = u
/ ((float) width
);
2260 t_buffer
[buffer_next
] = v
/ ((float) height
);
2263 if (buffer_next
== TGSI_QUAD_SIZE
) {
2264 /* 4 texel coords are in the buffer -> read it now */
2266 /* it is assumed that samp->min_img_filter is set to
2267 * img_filter_2d_nearest or one of the
2268 * accelerated img_filter_2d_nearest_XXX functions.
2270 for (jj
= 0; jj
< buffer_next
; jj
++) {
2271 args
.s
= s_buffer
[jj
];
2272 args
.t
= t_buffer
[jj
];
2274 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2275 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2276 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2277 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2278 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2291 /* if the tex coord buffer contains unread values, we will read
2294 if (buffer_next
> 0) {
2296 /* it is assumed that samp->min_img_filter is set to
2297 * img_filter_2d_nearest or one of the
2298 * accelerated img_filter_2d_nearest_XXX functions.
2300 for (jj
= 0; jj
< buffer_next
; jj
++) {
2301 args
.s
= s_buffer
[jj
];
2302 args
.t
= t_buffer
[jj
];
2304 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2305 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2306 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2307 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2308 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2313 /* Reaching this place would mean that no pixels intersected
2314 * the ellipse. This should never happen because the filter
2315 * we use always intersects at least one pixel.
2322 /* not enough pixels in resampling, resort to direct interpolation */
2326 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][j
]);
2328 num
[0] = rgba_temp
[0][j
];
2329 num
[1] = rgba_temp
[1][j
];
2330 num
[2] = rgba_temp
[2][j
];
2331 num
[3] = rgba_temp
[3][j
];
2334 rgba
[0][j
] = num
[0] / den
;
2335 rgba
[1][j
] = num
[1] / den
;
2336 rgba
[2][j
] = num
[2] / den
;
2337 rgba
[3][j
] = num
[3] / den
;
2343 * Sample 2D texture using an anisotropic filter.
2346 mip_filter_linear_aniso(struct sp_sampler_view
*sp_sview
,
2347 struct sp_sampler
*sp_samp
,
2348 img_filter_func min_filter
,
2349 img_filter_func mag_filter
,
2350 const float s
[TGSI_QUAD_SIZE
],
2351 const float t
[TGSI_QUAD_SIZE
],
2352 const float p
[TGSI_QUAD_SIZE
],
2353 const float c0
[TGSI_QUAD_SIZE
],
2354 const float lod_in
[TGSI_QUAD_SIZE
],
2355 const struct filter_args
*filt_args
,
2356 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2358 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2359 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2362 float lod
[TGSI_QUAD_SIZE
];
2364 float s_to_u
= u_minify(texture
->width0
, psview
->u
.tex
.first_level
);
2365 float t_to_v
= u_minify(texture
->height0
, psview
->u
.tex
.first_level
);
2366 float dudx
= (s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2367 float dudy
= (s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2368 float dvdx
= (t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2369 float dvdy
= (t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2370 struct img_filter_args args
;
2372 if (filt_args
->control
== tgsi_sampler_lod_bias
||
2373 filt_args
->control
== tgsi_sampler_lod_none
||
2375 filt_args
->control
== tgsi_sampler_derivs_explicit
) {
2376 /* note: instead of working with Px and Py, we will use the
2377 * squared length instead, to avoid sqrt.
2379 float Px2
= dudx
* dudx
+ dvdx
* dvdx
;
2380 float Py2
= dudy
* dudy
+ dvdy
* dvdy
;
2385 const float maxEccentricity
= sp_samp
->base
.max_anisotropy
* sp_samp
->base
.max_anisotropy
;
2396 /* if the eccentricity of the ellipse is too big, scale up the shorter
2397 * of the two vectors to limit the maximum amount of work per pixel
2400 if (e
> maxEccentricity
) {
2401 /* float s=e / maxEccentricity;
2405 Pmin2
= Pmax2
/ maxEccentricity
;
2408 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2409 * this since 0.5*log(x) = log(sqrt(x))
2411 lambda
= 0.5F
* util_fast_log2(Pmin2
) + sp_samp
->base
.lod_bias
;
2412 compute_lod(&sp_samp
->base
, filt_args
->control
, lambda
, lod_in
, lod
);
2415 assert(filt_args
->control
== tgsi_sampler_lod_explicit
||
2416 filt_args
->control
== tgsi_sampler_lod_zero
);
2417 compute_lod(&sp_samp
->base
, filt_args
->control
, sp_samp
->base
.lod_bias
, lod_in
, lod
);
2420 /* XXX: Take into account all lod values.
2423 level0
= psview
->u
.tex
.first_level
+ (int)lambda
;
2425 /* If the ellipse covers the whole image, we can
2426 * simply return the average of the whole image.
2428 if (level0
>= (int) psview
->u
.tex
.last_level
) {
2430 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2434 args
.level
= psview
->u
.tex
.last_level
;
2435 args
.face_id
= sp_sview
->faces
[j
];
2436 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2440 /* don't bother interpolating between multiple LODs; it doesn't
2441 * seem to be worth the extra running time.
2443 img_filter_2d_ewa(sp_sview
, sp_samp
, min_filter
, mag_filter
,
2445 dudx
, dvdx
, dudy
, dvdy
, rgba
);
2449 print_sample_4(__FUNCTION__
, rgba
);
2455 * Specialized version of mip_filter_linear with hard-wired calls to
2456 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2459 mip_filter_linear_2d_linear_repeat_POT(
2460 struct sp_sampler_view
*sp_sview
,
2461 struct sp_sampler
*sp_samp
,
2462 img_filter_func min_filter
,
2463 img_filter_func mag_filter
,
2464 const float s
[TGSI_QUAD_SIZE
],
2465 const float t
[TGSI_QUAD_SIZE
],
2466 const float p
[TGSI_QUAD_SIZE
],
2467 const float c0
[TGSI_QUAD_SIZE
],
2468 const float lod_in
[TGSI_QUAD_SIZE
],
2469 const struct filter_args
*filt_args
,
2470 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2472 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2474 float lod
[TGSI_QUAD_SIZE
];
2476 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2478 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2479 int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
2480 struct img_filter_args args
;
2481 /* Catches both negative and large values of level0:
2486 args
.face_id
= sp_sview
->faces
[j
];
2487 args
.offset
= filt_args
->offset
;
2488 args
.gather_only
= filt_args
->control
== tgsi_sampler_gather
;
2489 if ((unsigned)level0
>= psview
->u
.tex
.last_level
) {
2491 args
.level
= psview
->u
.tex
.first_level
;
2493 args
.level
= psview
->u
.tex
.last_level
;
2494 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
,
2499 float levelBlend
= frac(lod
[j
]);
2500 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2503 args
.level
= level0
;
2504 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
2505 args
.level
= level0
+1;
2506 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
2508 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
2509 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2514 print_sample_4(__FUNCTION__
, rgba
);
2520 * Do shadow/depth comparisons.
2523 sample_compare(struct sp_sampler_view
*sp_sview
,
2524 struct sp_sampler
*sp_samp
,
2525 const float s
[TGSI_QUAD_SIZE
],
2526 const float t
[TGSI_QUAD_SIZE
],
2527 const float p
[TGSI_QUAD_SIZE
],
2528 const float c0
[TGSI_QUAD_SIZE
],
2529 const float c1
[TGSI_QUAD_SIZE
],
2530 enum tgsi_sampler_control control
,
2531 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2533 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
2535 int k
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2537 const struct util_format_description
*format_desc
;
2539 bool is_gather
= (control
== tgsi_sampler_gather
);
2542 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2543 * for 2D Array texture we need to use the 'c0' (aka Q).
2544 * When we sampled the depth texture, the depth value was put into all
2545 * RGBA channels. We look at the red channel here.
2548 if (sp_sview
->base
.target
== PIPE_TEXTURE_2D_ARRAY
||
2549 sp_sview
->base
.target
== PIPE_TEXTURE_CUBE
) {
2554 } else if (sp_sview
->base
.target
== PIPE_TEXTURE_CUBE_ARRAY
) {
2566 format_desc
= util_format_description(sp_sview
->base
.format
);
2567 /* not entirely sure we couldn't end up with non-valid swizzle here */
2568 chan_type
= format_desc
->swizzle
[0] <= UTIL_FORMAT_SWIZZLE_W
?
2569 format_desc
->channel
[format_desc
->swizzle
[0]].type
:
2570 UTIL_FORMAT_TYPE_FLOAT
;
2571 if (chan_type
!= UTIL_FORMAT_TYPE_FLOAT
) {
2573 * clamping is a result of conversion to texture format, hence
2574 * doesn't happen with floats. Technically also should do comparison
2575 * in texture format (quantization!).
2577 pc
[0] = CLAMP(pc
[0], 0.0F
, 1.0F
);
2578 pc
[1] = CLAMP(pc
[1], 0.0F
, 1.0F
);
2579 pc
[2] = CLAMP(pc
[2], 0.0F
, 1.0F
);
2580 pc
[3] = CLAMP(pc
[3], 0.0F
, 1.0F
);
2583 for (v
= 0; v
< (is_gather
? TGSI_NUM_CHANNELS
: 1); v
++) {
2584 /* compare four texcoords vs. four texture samples */
2585 switch (sampler
->compare_func
) {
2586 case PIPE_FUNC_LESS
:
2587 k
[v
][0] = pc
[0] < rgba
[v
][0];
2588 k
[v
][1] = pc
[1] < rgba
[v
][1];
2589 k
[v
][2] = pc
[2] < rgba
[v
][2];
2590 k
[v
][3] = pc
[3] < rgba
[v
][3];
2592 case PIPE_FUNC_LEQUAL
:
2593 k
[v
][0] = pc
[0] <= rgba
[v
][0];
2594 k
[v
][1] = pc
[1] <= rgba
[v
][1];
2595 k
[v
][2] = pc
[2] <= rgba
[v
][2];
2596 k
[v
][3] = pc
[3] <= rgba
[v
][3];
2598 case PIPE_FUNC_GREATER
:
2599 k
[v
][0] = pc
[0] > rgba
[v
][0];
2600 k
[v
][1] = pc
[1] > rgba
[v
][1];
2601 k
[v
][2] = pc
[2] > rgba
[v
][2];
2602 k
[v
][3] = pc
[3] > rgba
[v
][3];
2604 case PIPE_FUNC_GEQUAL
:
2605 k
[v
][0] = pc
[0] >= rgba
[v
][0];
2606 k
[v
][1] = pc
[1] >= rgba
[v
][1];
2607 k
[v
][2] = pc
[2] >= rgba
[v
][2];
2608 k
[v
][3] = pc
[3] >= rgba
[v
][3];
2610 case PIPE_FUNC_EQUAL
:
2611 k
[v
][0] = pc
[0] == rgba
[v
][0];
2612 k
[v
][1] = pc
[1] == rgba
[v
][1];
2613 k
[v
][2] = pc
[2] == rgba
[v
][2];
2614 k
[v
][3] = pc
[3] == rgba
[v
][3];
2616 case PIPE_FUNC_NOTEQUAL
:
2617 k
[v
][0] = pc
[0] != rgba
[v
][0];
2618 k
[v
][1] = pc
[1] != rgba
[v
][1];
2619 k
[v
][2] = pc
[2] != rgba
[v
][2];
2620 k
[v
][3] = pc
[3] != rgba
[v
][3];
2622 case PIPE_FUNC_ALWAYS
:
2623 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 1;
2625 case PIPE_FUNC_NEVER
:
2626 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2629 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2636 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2637 for (v
= 0; v
< TGSI_NUM_CHANNELS
; v
++) {
2638 rgba
[v
][j
] = k
[v
][j
];
2642 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2643 rgba
[0][j
] = k
[0][j
];
2644 rgba
[1][j
] = k
[0][j
];
2645 rgba
[2][j
] = k
[0][j
];
2652 do_swizzling(const struct pipe_sampler_view
*sview
,
2653 float in
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
2654 float out
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2657 const unsigned swizzle_r
= sview
->swizzle_r
;
2658 const unsigned swizzle_g
= sview
->swizzle_g
;
2659 const unsigned swizzle_b
= sview
->swizzle_b
;
2660 const unsigned swizzle_a
= sview
->swizzle_a
;
2662 switch (swizzle_r
) {
2663 case PIPE_SWIZZLE_ZERO
:
2664 for (j
= 0; j
< 4; j
++)
2667 case PIPE_SWIZZLE_ONE
:
2668 for (j
= 0; j
< 4; j
++)
2672 assert(swizzle_r
< 4);
2673 for (j
= 0; j
< 4; j
++)
2674 out
[0][j
] = in
[swizzle_r
][j
];
2677 switch (swizzle_g
) {
2678 case PIPE_SWIZZLE_ZERO
:
2679 for (j
= 0; j
< 4; j
++)
2682 case PIPE_SWIZZLE_ONE
:
2683 for (j
= 0; j
< 4; j
++)
2687 assert(swizzle_g
< 4);
2688 for (j
= 0; j
< 4; j
++)
2689 out
[1][j
] = in
[swizzle_g
][j
];
2692 switch (swizzle_b
) {
2693 case PIPE_SWIZZLE_ZERO
:
2694 for (j
= 0; j
< 4; j
++)
2697 case PIPE_SWIZZLE_ONE
:
2698 for (j
= 0; j
< 4; j
++)
2702 assert(swizzle_b
< 4);
2703 for (j
= 0; j
< 4; j
++)
2704 out
[2][j
] = in
[swizzle_b
][j
];
2707 switch (swizzle_a
) {
2708 case PIPE_SWIZZLE_ZERO
:
2709 for (j
= 0; j
< 4; j
++)
2712 case PIPE_SWIZZLE_ONE
:
2713 for (j
= 0; j
< 4; j
++)
2717 assert(swizzle_a
< 4);
2718 for (j
= 0; j
< 4; j
++)
2719 out
[3][j
] = in
[swizzle_a
][j
];
2724 static wrap_nearest_func
2725 get_nearest_unorm_wrap(unsigned mode
)
2728 case PIPE_TEX_WRAP_CLAMP
:
2729 return wrap_nearest_unorm_clamp
;
2730 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2731 return wrap_nearest_unorm_clamp_to_edge
;
2732 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2733 return wrap_nearest_unorm_clamp_to_border
;
2735 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
2736 return wrap_nearest_unorm_clamp
;
2741 static wrap_nearest_func
2742 get_nearest_wrap(unsigned mode
)
2745 case PIPE_TEX_WRAP_REPEAT
:
2746 return wrap_nearest_repeat
;
2747 case PIPE_TEX_WRAP_CLAMP
:
2748 return wrap_nearest_clamp
;
2749 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2750 return wrap_nearest_clamp_to_edge
;
2751 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2752 return wrap_nearest_clamp_to_border
;
2753 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2754 return wrap_nearest_mirror_repeat
;
2755 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2756 return wrap_nearest_mirror_clamp
;
2757 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2758 return wrap_nearest_mirror_clamp_to_edge
;
2759 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2760 return wrap_nearest_mirror_clamp_to_border
;
2763 return wrap_nearest_repeat
;
2768 static wrap_linear_func
2769 get_linear_unorm_wrap(unsigned mode
)
2772 case PIPE_TEX_WRAP_CLAMP
:
2773 return wrap_linear_unorm_clamp
;
2774 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2775 return wrap_linear_unorm_clamp_to_edge
;
2776 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2777 return wrap_linear_unorm_clamp_to_border
;
2779 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
2780 return wrap_linear_unorm_clamp
;
2785 static wrap_linear_func
2786 get_linear_wrap(unsigned mode
)
2789 case PIPE_TEX_WRAP_REPEAT
:
2790 return wrap_linear_repeat
;
2791 case PIPE_TEX_WRAP_CLAMP
:
2792 return wrap_linear_clamp
;
2793 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2794 return wrap_linear_clamp_to_edge
;
2795 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2796 return wrap_linear_clamp_to_border
;
2797 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2798 return wrap_linear_mirror_repeat
;
2799 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2800 return wrap_linear_mirror_clamp
;
2801 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2802 return wrap_linear_mirror_clamp_to_edge
;
2803 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2804 return wrap_linear_mirror_clamp_to_border
;
2807 return wrap_linear_repeat
;
2813 * Is swizzling needed for the given state key?
2816 any_swizzle(const struct pipe_sampler_view
*view
)
2818 return (view
->swizzle_r
!= PIPE_SWIZZLE_RED
||
2819 view
->swizzle_g
!= PIPE_SWIZZLE_GREEN
||
2820 view
->swizzle_b
!= PIPE_SWIZZLE_BLUE
||
2821 view
->swizzle_a
!= PIPE_SWIZZLE_ALPHA
);
2825 static img_filter_func
2826 get_img_filter(const struct sp_sampler_view
*sp_sview
,
2827 const struct pipe_sampler_state
*sampler
,
2828 unsigned filter
, bool gather
)
2830 switch (sp_sview
->base
.target
) {
2832 case PIPE_TEXTURE_1D
:
2833 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2834 return img_filter_1d_nearest
;
2836 return img_filter_1d_linear
;
2838 case PIPE_TEXTURE_1D_ARRAY
:
2839 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2840 return img_filter_1d_array_nearest
;
2842 return img_filter_1d_array_linear
;
2844 case PIPE_TEXTURE_2D
:
2845 case PIPE_TEXTURE_RECT
:
2846 /* Try for fast path:
2848 if (!gather
&& sp_sview
->pot2d
&&
2849 sampler
->wrap_s
== sampler
->wrap_t
&&
2850 sampler
->normalized_coords
)
2852 switch (sampler
->wrap_s
) {
2853 case PIPE_TEX_WRAP_REPEAT
:
2855 case PIPE_TEX_FILTER_NEAREST
:
2856 return img_filter_2d_nearest_repeat_POT
;
2857 case PIPE_TEX_FILTER_LINEAR
:
2858 return img_filter_2d_linear_repeat_POT
;
2863 case PIPE_TEX_WRAP_CLAMP
:
2865 case PIPE_TEX_FILTER_NEAREST
:
2866 return img_filter_2d_nearest_clamp_POT
;
2872 /* Otherwise use default versions:
2874 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2875 return img_filter_2d_nearest
;
2877 return img_filter_2d_linear
;
2879 case PIPE_TEXTURE_2D_ARRAY
:
2880 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2881 return img_filter_2d_array_nearest
;
2883 return img_filter_2d_array_linear
;
2885 case PIPE_TEXTURE_CUBE
:
2886 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2887 return img_filter_cube_nearest
;
2889 return img_filter_cube_linear
;
2891 case PIPE_TEXTURE_CUBE_ARRAY
:
2892 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2893 return img_filter_cube_array_nearest
;
2895 return img_filter_cube_array_linear
;
2897 case PIPE_TEXTURE_3D
:
2898 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2899 return img_filter_3d_nearest
;
2901 return img_filter_3d_linear
;
2905 return img_filter_1d_nearest
;
2911 sample_mip(struct sp_sampler_view
*sp_sview
,
2912 struct sp_sampler
*sp_samp
,
2913 const float s
[TGSI_QUAD_SIZE
],
2914 const float t
[TGSI_QUAD_SIZE
],
2915 const float p
[TGSI_QUAD_SIZE
],
2916 const float c0
[TGSI_QUAD_SIZE
],
2917 const float lod
[TGSI_QUAD_SIZE
],
2918 const struct filter_args
*filt_args
,
2919 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2921 mip_filter_func mip_filter
;
2922 img_filter_func min_img_filter
= NULL
;
2923 img_filter_func mag_img_filter
= NULL
;
2925 if (filt_args
->control
== tgsi_sampler_gather
) {
2926 mip_filter
= mip_filter_nearest
;
2927 min_img_filter
= get_img_filter(sp_sview
, &sp_samp
->base
, PIPE_TEX_FILTER_LINEAR
, true);
2928 } else if (sp_sview
->pot2d
& sp_samp
->min_mag_equal_repeat_linear
) {
2929 mip_filter
= mip_filter_linear_2d_linear_repeat_POT
;
2932 mip_filter
= sp_samp
->mip_filter
;
2933 min_img_filter
= get_img_filter(sp_sview
, &sp_samp
->base
, sp_samp
->min_img_filter
, false);
2934 if (sp_samp
->min_mag_equal
) {
2935 mag_img_filter
= min_img_filter
;
2938 mag_img_filter
= get_img_filter(sp_sview
, &sp_samp
->base
, sp_samp
->base
.mag_img_filter
, false);
2942 mip_filter(sp_sview
, sp_samp
, min_img_filter
, mag_img_filter
,
2943 s
, t
, p
, c0
, lod
, filt_args
, rgba
);
2945 if (sp_samp
->base
.compare_mode
!= PIPE_TEX_COMPARE_NONE
) {
2946 sample_compare(sp_sview
, sp_samp
, s
, t
, p
, c0
, lod
, filt_args
->control
, rgba
);
2949 if (sp_sview
->need_swizzle
&& filt_args
->control
!= tgsi_sampler_gather
) {
2950 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2951 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
2952 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
2959 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2960 * Put face info into the sampler faces[] array.
2963 sample_cube(struct sp_sampler_view
*sp_sview
,
2964 struct sp_sampler
*sp_samp
,
2965 const float s
[TGSI_QUAD_SIZE
],
2966 const float t
[TGSI_QUAD_SIZE
],
2967 const float p
[TGSI_QUAD_SIZE
],
2968 const float c0
[TGSI_QUAD_SIZE
],
2969 const float c1
[TGSI_QUAD_SIZE
],
2970 const struct filter_args
*filt_args
,
2971 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2974 float ssss
[4], tttt
[4];
2976 /* Not actually used, but the intermediate steps that do the
2977 * dereferencing don't know it.
2979 static float pppp
[4] = { 0, 0, 0, 0 };
2987 direction target sc tc ma
2988 ---------- ------------------------------- --- --- ---
2989 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2990 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2991 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2992 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2993 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2994 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2997 /* Choose the cube face and compute new s/t coords for the 2D face.
2999 * Use the same cube face for all four pixels in the quad.
3001 * This isn't ideal, but if we want to use a different cube face
3002 * per pixel in the quad, we'd have to also compute the per-face
3003 * LOD here too. That's because the four post-face-selection
3004 * texcoords are no longer related to each other (they're
3005 * per-face!) so we can't use subtraction to compute the partial
3006 * deriviates to compute the LOD. Doing so (near cube edges
3007 * anyway) gives us pretty much random values.
3010 /* use the average of the four pixel's texcoords to choose the face */
3011 const float rx
= 0.25F
* (s
[0] + s
[1] + s
[2] + s
[3]);
3012 const float ry
= 0.25F
* (t
[0] + t
[1] + t
[2] + t
[3]);
3013 const float rz
= 0.25F
* (p
[0] + p
[1] + p
[2] + p
[3]);
3014 const float arx
= fabsf(rx
), ary
= fabsf(ry
), arz
= fabsf(rz
);
3016 if (arx
>= ary
&& arx
>= arz
) {
3017 float sign
= (rx
>= 0.0F
) ? 1.0F
: -1.0F
;
3018 uint face
= (rx
>= 0.0F
) ? PIPE_TEX_FACE_POS_X
: PIPE_TEX_FACE_NEG_X
;
3019 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3020 const float ima
= -0.5F
/ fabsf(s
[j
]);
3021 ssss
[j
] = sign
* p
[j
] * ima
+ 0.5F
;
3022 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3023 sp_sview
->faces
[j
] = face
;
3026 else if (ary
>= arx
&& ary
>= arz
) {
3027 float sign
= (ry
>= 0.0F
) ? 1.0F
: -1.0F
;
3028 uint face
= (ry
>= 0.0F
) ? PIPE_TEX_FACE_POS_Y
: PIPE_TEX_FACE_NEG_Y
;
3029 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3030 const float ima
= -0.5F
/ fabsf(t
[j
]);
3031 ssss
[j
] = -s
[j
] * ima
+ 0.5F
;
3032 tttt
[j
] = sign
* -p
[j
] * ima
+ 0.5F
;
3033 sp_sview
->faces
[j
] = face
;
3037 float sign
= (rz
>= 0.0F
) ? 1.0F
: -1.0F
;
3038 uint face
= (rz
>= 0.0F
) ? PIPE_TEX_FACE_POS_Z
: PIPE_TEX_FACE_NEG_Z
;
3039 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3040 const float ima
= -0.5F
/ fabsf(p
[j
]);
3041 ssss
[j
] = sign
* -s
[j
] * ima
+ 0.5F
;
3042 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3043 sp_sview
->faces
[j
] = face
;
3048 sample_mip(sp_sview
, sp_samp
, ssss
, tttt
, pppp
, c0
, c1
, filt_args
, rgba
);
3053 sp_get_dims(struct sp_sampler_view
*sp_sview
, int level
,
3056 const struct pipe_sampler_view
*view
= &sp_sview
->base
;
3057 const struct pipe_resource
*texture
= view
->texture
;
3059 if (view
->target
== PIPE_BUFFER
) {
3060 dims
[0] = (view
->u
.buf
.last_element
- view
->u
.buf
.first_element
) + 1;
3061 /* the other values are undefined, but let's avoid potential valgrind
3064 dims
[1] = dims
[2] = dims
[3] = 0;
3068 /* undefined according to EXT_gpu_program */
3069 level
+= view
->u
.tex
.first_level
;
3070 if (level
> view
->u
.tex
.last_level
)
3073 dims
[3] = view
->u
.tex
.last_level
- view
->u
.tex
.first_level
+ 1;
3074 dims
[0] = u_minify(texture
->width0
, level
);
3076 switch (view
->target
) {
3077 case PIPE_TEXTURE_1D_ARRAY
:
3078 dims
[1] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3080 case PIPE_TEXTURE_1D
:
3082 case PIPE_TEXTURE_2D_ARRAY
:
3083 dims
[2] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3085 case PIPE_TEXTURE_2D
:
3086 case PIPE_TEXTURE_CUBE
:
3087 case PIPE_TEXTURE_RECT
:
3088 dims
[1] = u_minify(texture
->height0
, level
);
3090 case PIPE_TEXTURE_3D
:
3091 dims
[1] = u_minify(texture
->height0
, level
);
3092 dims
[2] = u_minify(texture
->depth0
, level
);
3094 case PIPE_TEXTURE_CUBE_ARRAY
:
3095 dims
[1] = u_minify(texture
->height0
, level
);
3096 dims
[2] = (view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1) / 6;
3099 assert(!"unexpected texture target in sp_get_dims()");
3105 * This function is only used for getting unfiltered texels via the
3106 * TXF opcode. The GL spec says that out-of-bounds texel fetches
3107 * produce undefined results. Instead of crashing, lets just clamp
3108 * coords to the texture image size.
3111 sp_get_texels(struct sp_sampler_view
*sp_sview
,
3112 const int v_i
[TGSI_QUAD_SIZE
],
3113 const int v_j
[TGSI_QUAD_SIZE
],
3114 const int v_k
[TGSI_QUAD_SIZE
],
3115 const int lod
[TGSI_QUAD_SIZE
],
3116 const int8_t offset
[3],
3117 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3119 union tex_tile_address addr
;
3120 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
3123 int width
, height
, depth
;
3126 /* TODO write a better test for LOD */
3127 addr
.bits
.level
= sp_sview
->base
.target
== PIPE_BUFFER
? 0 :
3128 CLAMP(lod
[0] + sp_sview
->base
.u
.tex
.first_level
,
3129 sp_sview
->base
.u
.tex
.first_level
,
3130 sp_sview
->base
.u
.tex
.last_level
);
3132 width
= u_minify(texture
->width0
, addr
.bits
.level
);
3133 height
= u_minify(texture
->height0
, addr
.bits
.level
);
3134 depth
= u_minify(texture
->depth0
, addr
.bits
.level
);
3136 switch (sp_sview
->base
.target
) {
3138 case PIPE_TEXTURE_1D
:
3139 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3140 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3141 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, 0);
3142 for (c
= 0; c
< 4; c
++) {
3147 case PIPE_TEXTURE_1D_ARRAY
:
3148 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3149 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3150 int y
= CLAMP(v_j
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3151 sp_sview
->base
.u
.tex
.last_layer
);
3152 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
3153 for (c
= 0; c
< 4; c
++) {
3158 case PIPE_TEXTURE_2D
:
3159 case PIPE_TEXTURE_RECT
:
3160 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3161 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3162 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3163 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
3164 for (c
= 0; c
< 4; c
++) {
3169 case PIPE_TEXTURE_2D_ARRAY
:
3170 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3171 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3172 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3173 int layer
= CLAMP(v_k
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3174 sp_sview
->base
.u
.tex
.last_layer
);
3175 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
3176 for (c
= 0; c
< 4; c
++) {
3181 case PIPE_TEXTURE_3D
:
3182 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3183 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3184 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3185 int z
= CLAMP(v_k
[j
] + offset
[2], 0, depth
- 1);
3186 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, z
);
3187 for (c
= 0; c
< 4; c
++) {
3192 case PIPE_TEXTURE_CUBE
: /* TXF can't work on CUBE according to spec */
3194 assert(!"Unknown or CUBE texture type in TXF processing\n");
3198 if (sp_sview
->need_swizzle
) {
3199 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3200 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
3201 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
3207 softpipe_create_sampler_state(struct pipe_context
*pipe
,
3208 const struct pipe_sampler_state
*sampler
)
3210 struct sp_sampler
*samp
= CALLOC_STRUCT(sp_sampler
);
3212 samp
->base
= *sampler
;
3214 /* Note that (for instance) linear_texcoord_s and
3215 * nearest_texcoord_s may be active at the same time, if the
3216 * sampler min_img_filter differs from its mag_img_filter.
3218 if (sampler
->normalized_coords
) {
3219 samp
->linear_texcoord_s
= get_linear_wrap( sampler
->wrap_s
);
3220 samp
->linear_texcoord_t
= get_linear_wrap( sampler
->wrap_t
);
3221 samp
->linear_texcoord_p
= get_linear_wrap( sampler
->wrap_r
);
3223 samp
->nearest_texcoord_s
= get_nearest_wrap( sampler
->wrap_s
);
3224 samp
->nearest_texcoord_t
= get_nearest_wrap( sampler
->wrap_t
);
3225 samp
->nearest_texcoord_p
= get_nearest_wrap( sampler
->wrap_r
);
3228 samp
->linear_texcoord_s
= get_linear_unorm_wrap( sampler
->wrap_s
);
3229 samp
->linear_texcoord_t
= get_linear_unorm_wrap( sampler
->wrap_t
);
3230 samp
->linear_texcoord_p
= get_linear_unorm_wrap( sampler
->wrap_r
);
3232 samp
->nearest_texcoord_s
= get_nearest_unorm_wrap( sampler
->wrap_s
);
3233 samp
->nearest_texcoord_t
= get_nearest_unorm_wrap( sampler
->wrap_t
);
3234 samp
->nearest_texcoord_p
= get_nearest_unorm_wrap( sampler
->wrap_r
);
3237 samp
->min_img_filter
= sampler
->min_img_filter
;
3239 switch (sampler
->min_mip_filter
) {
3240 case PIPE_TEX_MIPFILTER_NONE
:
3241 if (sampler
->min_img_filter
== sampler
->mag_img_filter
)
3242 samp
->mip_filter
= mip_filter_none_no_filter_select
;
3244 samp
->mip_filter
= mip_filter_none
;
3247 case PIPE_TEX_MIPFILTER_NEAREST
:
3248 samp
->mip_filter
= mip_filter_nearest
;
3251 case PIPE_TEX_MIPFILTER_LINEAR
:
3252 if (sampler
->min_img_filter
== sampler
->mag_img_filter
&&
3253 sampler
->normalized_coords
&&
3254 sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
&&
3255 sampler
->wrap_t
== PIPE_TEX_WRAP_REPEAT
&&
3256 sampler
->min_img_filter
== PIPE_TEX_FILTER_LINEAR
&&
3257 sampler
->max_anisotropy
<= 1) {
3258 samp
->min_mag_equal_repeat_linear
= TRUE
;
3260 samp
->mip_filter
= mip_filter_linear
;
3262 /* Anisotropic filtering extension. */
3263 if (sampler
->max_anisotropy
> 1) {
3264 samp
->mip_filter
= mip_filter_linear_aniso
;
3266 /* Override min_img_filter:
3267 * min_img_filter needs to be set to NEAREST since we need to access
3268 * each texture pixel as it is and weight it later; using linear
3269 * filters will have incorrect results.
3270 * By setting the filter to NEAREST here, we can avoid calling the
3271 * generic img_filter_2d_nearest in the anisotropic filter function,
3272 * making it possible to use one of the accelerated implementations
3274 samp
->min_img_filter
= PIPE_TEX_FILTER_NEAREST
;
3276 /* on first access create the lookup table containing the filter weights. */
3278 create_filter_table();
3283 if (samp
->min_img_filter
== sampler
->mag_img_filter
) {
3284 samp
->min_mag_equal
= TRUE
;
3287 return (void *)samp
;
3292 softpipe_get_lambda_func(const struct pipe_sampler_view
*view
, unsigned shader
)
3294 if (shader
!= PIPE_SHADER_FRAGMENT
)
3295 return compute_lambda_vert
;
3297 switch (view
->target
) {
3299 case PIPE_TEXTURE_1D
:
3300 case PIPE_TEXTURE_1D_ARRAY
:
3301 return compute_lambda_1d
;
3302 case PIPE_TEXTURE_2D
:
3303 case PIPE_TEXTURE_2D_ARRAY
:
3304 case PIPE_TEXTURE_RECT
:
3305 case PIPE_TEXTURE_CUBE
:
3306 case PIPE_TEXTURE_CUBE_ARRAY
:
3307 return compute_lambda_2d
;
3308 case PIPE_TEXTURE_3D
:
3309 return compute_lambda_3d
;
3312 return compute_lambda_1d
;
3317 struct pipe_sampler_view
*
3318 softpipe_create_sampler_view(struct pipe_context
*pipe
,
3319 struct pipe_resource
*resource
,
3320 const struct pipe_sampler_view
*templ
)
3322 struct sp_sampler_view
*sview
= CALLOC_STRUCT(sp_sampler_view
);
3323 struct softpipe_resource
*spr
= (struct softpipe_resource
*)resource
;
3326 struct pipe_sampler_view
*view
= &sview
->base
;
3328 view
->reference
.count
= 1;
3329 view
->texture
= NULL
;
3330 pipe_resource_reference(&view
->texture
, resource
);
3331 view
->context
= pipe
;
3335 * This is possibly too lenient, but the primary reason is just
3336 * to catch state trackers which forget to initialize this, so
3337 * it only catches clearly impossible view targets.
3339 if (view
->target
!= resource
->target
) {
3340 if (view
->target
== PIPE_TEXTURE_1D
)
3341 assert(resource
->target
== PIPE_TEXTURE_1D_ARRAY
);
3342 else if (view
->target
== PIPE_TEXTURE_1D_ARRAY
)
3343 assert(resource
->target
== PIPE_TEXTURE_1D
);
3344 else if (view
->target
== PIPE_TEXTURE_2D
)
3345 assert(resource
->target
== PIPE_TEXTURE_2D_ARRAY
||
3346 resource
->target
== PIPE_TEXTURE_CUBE
||
3347 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3348 else if (view
->target
== PIPE_TEXTURE_2D_ARRAY
)
3349 assert(resource
->target
== PIPE_TEXTURE_2D
||
3350 resource
->target
== PIPE_TEXTURE_CUBE
||
3351 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3352 else if (view
->target
== PIPE_TEXTURE_CUBE
)
3353 assert(resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
||
3354 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3355 else if (view
->target
== PIPE_TEXTURE_CUBE_ARRAY
)
3356 assert(resource
->target
== PIPE_TEXTURE_CUBE
||
3357 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3363 if (any_swizzle(view
)) {
3364 sview
->need_swizzle
= TRUE
;
3367 if (view
->target
== PIPE_TEXTURE_CUBE
||
3368 view
->target
== PIPE_TEXTURE_CUBE_ARRAY
)
3369 sview
->get_samples
= sample_cube
;
3371 sview
->get_samples
= sample_mip
;
3373 sview
->pot2d
= spr
->pot
&&
3374 (view
->target
== PIPE_TEXTURE_2D
||
3375 view
->target
== PIPE_TEXTURE_RECT
);
3377 sview
->xpot
= util_logbase2( resource
->width0
);
3378 sview
->ypot
= util_logbase2( resource
->height0
);
3381 return (struct pipe_sampler_view
*) sview
;
3386 sp_tgsi_get_dims(struct tgsi_sampler
*tgsi_sampler
,
3387 const unsigned sview_index
,
3388 int level
, int dims
[4])
3390 struct sp_tgsi_sampler
*sp_samp
= (struct sp_tgsi_sampler
*)tgsi_sampler
;
3392 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3393 /* always have a view here but texture is NULL if no sampler view was set. */
3394 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3395 dims
[0] = dims
[1] = dims
[2] = dims
[3] = 0;
3398 sp_get_dims(&sp_samp
->sp_sview
[sview_index
], level
, dims
);
3403 sp_tgsi_get_samples(struct tgsi_sampler
*tgsi_sampler
,
3404 const unsigned sview_index
,
3405 const unsigned sampler_index
,
3406 const float s
[TGSI_QUAD_SIZE
],
3407 const float t
[TGSI_QUAD_SIZE
],
3408 const float p
[TGSI_QUAD_SIZE
],
3409 const float c0
[TGSI_QUAD_SIZE
],
3410 const float lod
[TGSI_QUAD_SIZE
],
3411 float derivs
[3][2][TGSI_QUAD_SIZE
],
3412 const int8_t offset
[3],
3413 enum tgsi_sampler_control control
,
3414 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3416 struct sp_tgsi_sampler
*sp_samp
= (struct sp_tgsi_sampler
*)tgsi_sampler
;
3417 struct filter_args filt_args
;
3418 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3419 assert(sampler_index
< PIPE_MAX_SAMPLERS
);
3420 assert(sp_samp
->sp_sampler
[sampler_index
]);
3421 /* always have a view here but texture is NULL if no sampler view was set. */
3422 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3424 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3425 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3432 filt_args
.control
= control
;
3433 filt_args
.offset
= offset
;
3434 sp_samp
->sp_sview
[sview_index
].get_samples(&sp_samp
->sp_sview
[sview_index
],
3435 sp_samp
->sp_sampler
[sampler_index
],
3436 s
, t
, p
, c0
, lod
, &filt_args
, rgba
);
3441 sp_tgsi_get_texel(struct tgsi_sampler
*tgsi_sampler
,
3442 const unsigned sview_index
,
3443 const int i
[TGSI_QUAD_SIZE
],
3444 const int j
[TGSI_QUAD_SIZE
], const int k
[TGSI_QUAD_SIZE
],
3445 const int lod
[TGSI_QUAD_SIZE
], const int8_t offset
[3],
3446 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3448 struct sp_tgsi_sampler
*sp_samp
= (struct sp_tgsi_sampler
*)tgsi_sampler
;
3450 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3451 /* always have a view here but texture is NULL if no sampler view was set. */
3452 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3454 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3455 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3461 sp_get_texels(&sp_samp
->sp_sview
[sview_index
], i
, j
, k
, lod
, offset
, rgba
);
3465 struct sp_tgsi_sampler
*
3466 sp_create_tgsi_sampler(void)
3468 struct sp_tgsi_sampler
*samp
= CALLOC_STRUCT(sp_tgsi_sampler
);
3472 samp
->base
.get_dims
= sp_tgsi_get_dims
;
3473 samp
->base
.get_samples
= sp_tgsi_get_samples
;
3474 samp
->base
.get_texel
= sp_tgsi_get_texel
;