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(const struct sp_sampler_view
*sp_sview
,
1865 const 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 * Clamps given lod to both lod limits and mip level limits. Clamping to the
1941 * latter limits is done so that lod is relative to the first (base) level.
1944 clamp_lod(const struct sp_sampler_view
*sp_sview
,
1945 const struct sp_sampler
*sp_samp
,
1946 const float lod
[TGSI_QUAD_SIZE
],
1947 float clamped
[TGSI_QUAD_SIZE
])
1949 const float min_lod
= sp_samp
->base
.min_lod
;
1950 const float max_lod
= sp_samp
->base
.max_lod
;
1951 const float min_level
= sp_sview
->base
.u
.tex
.first_level
;
1952 const float max_level
= sp_sview
->base
.u
.tex
.last_level
;
1955 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1958 cl
= CLAMP(cl
, min_lod
, max_lod
);
1959 cl
= CLAMP(cl
, 0, max_level
- min_level
);
1965 * Get mip level relative to base level for linear mip filter
1968 mip_rel_level_linear(const struct sp_sampler_view
*sp_sview
,
1969 const struct sp_sampler
*sp_samp
,
1970 const float lod
[TGSI_QUAD_SIZE
],
1971 float level
[TGSI_QUAD_SIZE
])
1973 clamp_lod(sp_sview
, sp_samp
, lod
, level
);
1977 mip_filter_linear(struct sp_sampler_view
*sp_sview
,
1978 struct sp_sampler
*sp_samp
,
1979 img_filter_func min_filter
,
1980 img_filter_func mag_filter
,
1981 const float s
[TGSI_QUAD_SIZE
],
1982 const float t
[TGSI_QUAD_SIZE
],
1983 const float p
[TGSI_QUAD_SIZE
],
1984 const float c0
[TGSI_QUAD_SIZE
],
1985 const float lod_in
[TGSI_QUAD_SIZE
],
1986 const struct filter_args
*filt_args
,
1987 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1989 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
1991 float lod
[TGSI_QUAD_SIZE
];
1992 struct img_filter_args args
;
1994 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
1996 args
.offset
= filt_args
->offset
;
1997 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
1998 args
.gather_comp
= get_gather_component(lod_in
);
2000 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2001 int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
2006 args
.face_id
= filt_args
->faces
[j
];
2009 args
.level
= psview
->u
.tex
.first_level
;
2010 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2012 else if (level0
>= (int) psview
->u
.tex
.last_level
) {
2013 args
.level
= psview
->u
.tex
.last_level
;
2014 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2017 float levelBlend
= frac(lod
[j
]);
2018 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2021 args
.level
= level0
;
2022 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
2023 args
.level
= level0
+1;
2024 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
2026 for (c
= 0; c
< 4; c
++) {
2027 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2033 print_sample_4(__FUNCTION__
, rgba
);
2039 * Get mip level relative to base level for nearest mip filter
2042 mip_rel_level_nearest(const struct sp_sampler_view
*sp_sview
,
2043 const struct sp_sampler
*sp_samp
,
2044 const float lod
[TGSI_QUAD_SIZE
],
2045 float level
[TGSI_QUAD_SIZE
])
2049 clamp_lod(sp_sview
, sp_samp
, lod
, level
);
2050 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++)
2051 /* TODO: It should rather be:
2052 * level[j] = ceil(level[j] + 0.5F) - 1.0F;
2054 level
[j
] = (int)(level
[j
] + 0.5F
);
2058 * Compute nearest mipmap level from texcoords.
2059 * Then sample the texture level for four elements of a quad.
2060 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
2063 mip_filter_nearest(struct sp_sampler_view
*sp_sview
,
2064 struct sp_sampler
*sp_samp
,
2065 img_filter_func min_filter
,
2066 img_filter_func mag_filter
,
2067 const float s
[TGSI_QUAD_SIZE
],
2068 const float t
[TGSI_QUAD_SIZE
],
2069 const float p
[TGSI_QUAD_SIZE
],
2070 const float c0
[TGSI_QUAD_SIZE
],
2071 const float lod_in
[TGSI_QUAD_SIZE
],
2072 const struct filter_args
*filt_args
,
2073 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2075 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2076 float lod
[TGSI_QUAD_SIZE
];
2078 struct img_filter_args args
;
2080 args
.offset
= filt_args
->offset
;
2081 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2082 args
.gather_comp
= get_gather_component(lod_in
);
2084 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2086 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2090 args
.face_id
= filt_args
->faces
[j
];
2093 args
.level
= psview
->u
.tex
.first_level
;
2094 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2096 int level
= psview
->u
.tex
.first_level
+ (int)(lod
[j
] + 0.5F
);
2097 args
.level
= MIN2(level
, (int)psview
->u
.tex
.last_level
);
2098 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2103 print_sample_4(__FUNCTION__
, rgba
);
2109 * Get mip level relative to base level for none mip filter
2112 mip_rel_level_none(const struct sp_sampler_view
*sp_sview
,
2113 const struct sp_sampler
*sp_samp
,
2114 const float lod
[TGSI_QUAD_SIZE
],
2115 float level
[TGSI_QUAD_SIZE
])
2119 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2125 mip_filter_none(struct sp_sampler_view
*sp_sview
,
2126 struct sp_sampler
*sp_samp
,
2127 img_filter_func min_filter
,
2128 img_filter_func mag_filter
,
2129 const float s
[TGSI_QUAD_SIZE
],
2130 const float t
[TGSI_QUAD_SIZE
],
2131 const float p
[TGSI_QUAD_SIZE
],
2132 const float c0
[TGSI_QUAD_SIZE
],
2133 const float lod_in
[TGSI_QUAD_SIZE
],
2134 const struct filter_args
*filt_args
,
2135 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2137 float lod
[TGSI_QUAD_SIZE
];
2139 struct img_filter_args args
;
2141 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2142 args
.offset
= filt_args
->offset
;
2143 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2145 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2147 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2151 args
.face_id
= filt_args
->faces
[j
];
2153 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2156 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2163 * Get mip level relative to base level for none mip filter
2166 mip_rel_level_none_no_filter_select(const struct sp_sampler_view
*sp_sview
,
2167 const struct sp_sampler
*sp_samp
,
2168 const float lod
[TGSI_QUAD_SIZE
],
2169 float level
[TGSI_QUAD_SIZE
])
2171 mip_rel_level_none(sp_sview
, sp_samp
, lod
, level
);
2175 mip_filter_none_no_filter_select(struct sp_sampler_view
*sp_sview
,
2176 struct sp_sampler
*sp_samp
,
2177 img_filter_func min_filter
,
2178 img_filter_func mag_filter
,
2179 const float s
[TGSI_QUAD_SIZE
],
2180 const float t
[TGSI_QUAD_SIZE
],
2181 const float p
[TGSI_QUAD_SIZE
],
2182 const float c0
[TGSI_QUAD_SIZE
],
2183 const float lod_in
[TGSI_QUAD_SIZE
],
2184 const struct filter_args
*filt_args
,
2185 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2188 struct img_filter_args args
;
2189 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2190 args
.offset
= filt_args
->offset
;
2191 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2192 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2196 args
.face_id
= filt_args
->faces
[j
];
2197 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2202 /* For anisotropic filtering */
2203 #define WEIGHT_LUT_SIZE 1024
2205 static float *weightLut
= NULL
;
2208 * Creates the look-up table used to speed-up EWA sampling
2211 create_filter_table(void)
2215 weightLut
= (float *) MALLOC(WEIGHT_LUT_SIZE
* sizeof(float));
2217 for (i
= 0; i
< WEIGHT_LUT_SIZE
; ++i
) {
2219 float r2
= (float) i
/ (float) (WEIGHT_LUT_SIZE
- 1);
2220 float weight
= (float) exp(-alpha
* r2
);
2221 weightLut
[i
] = weight
;
2228 * Elliptical weighted average (EWA) filter for producing high quality
2229 * anisotropic filtered results.
2230 * Based on the Higher Quality Elliptical Weighted Average Filter
2231 * published by Paul S. Heckbert in his Master's Thesis
2232 * "Fundamentals of Texture Mapping and Image Warping" (1989)
2235 img_filter_2d_ewa(struct sp_sampler_view
*sp_sview
,
2236 struct sp_sampler
*sp_samp
,
2237 img_filter_func min_filter
,
2238 img_filter_func mag_filter
,
2239 const float s
[TGSI_QUAD_SIZE
],
2240 const float t
[TGSI_QUAD_SIZE
],
2241 const float p
[TGSI_QUAD_SIZE
],
2242 const float faces
[TGSI_QUAD_SIZE
],
2244 const float dudx
, const float dvdx
,
2245 const float dudy
, const float dvdy
,
2246 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2248 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2250 // ??? Won't the image filters blow up if level is negative?
2251 unsigned level0
= level
> 0 ? level
: 0;
2252 float scaling
= 1.0f
/ (1 << level0
);
2253 int width
= u_minify(texture
->width0
, level0
);
2254 int height
= u_minify(texture
->height0
, level0
);
2255 struct img_filter_args args
;
2256 float ux
= dudx
* scaling
;
2257 float vx
= dvdx
* scaling
;
2258 float uy
= dudy
* scaling
;
2259 float vy
= dvdy
* scaling
;
2261 /* compute ellipse coefficients to bound the region:
2262 * A*x*x + B*x*y + C*y*y = F.
2264 float A
= vx
*vx
+vy
*vy
+1;
2265 float B
= -2*(ux
*vx
+uy
*vy
);
2266 float C
= ux
*ux
+uy
*uy
+1;
2267 float F
= A
*C
-B
*B
/4.0f
;
2269 /* check if it is an ellipse */
2270 /* assert(F > 0.0); */
2272 /* Compute the ellipse's (u,v) bounding box in texture space */
2273 float d
= -B
*B
+4.0f
*C
*A
;
2274 float box_u
= 2.0f
/ d
* sqrtf(d
*C
*F
); /* box_u -> half of bbox with */
2275 float box_v
= 2.0f
/ d
* sqrtf(A
*d
*F
); /* box_v -> half of bbox height */
2277 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2278 float s_buffer
[TGSI_QUAD_SIZE
];
2279 float t_buffer
[TGSI_QUAD_SIZE
];
2280 float weight_buffer
[TGSI_QUAD_SIZE
];
2281 unsigned buffer_next
;
2283 float den
; /* = 0.0F; */
2285 float U
; /* = u0 - tex_u; */
2288 /* Scale ellipse formula to directly index the Filter Lookup Table.
2289 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2291 double formScale
= (double) (WEIGHT_LUT_SIZE
- 1) / F
;
2295 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2297 /* For each quad, the du and dx values are the same and so the ellipse is
2298 * also the same. Note that texel/image access can only be performed using
2299 * a quad, i.e. it is not possible to get the pixel value for a single
2300 * tex coord. In order to have a better performance, the access is buffered
2301 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2302 * full, then the pixel values are read from the image.
2307 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2308 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2309 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2310 * value, q, is less than F, we're inside the ellipse
2312 float tex_u
= -0.5F
+ s
[j
] * texture
->width0
* scaling
;
2313 float tex_v
= -0.5F
+ t
[j
] * texture
->height0
* scaling
;
2315 int u0
= (int) floorf(tex_u
- box_u
);
2316 int u1
= (int) ceilf(tex_u
+ box_u
);
2317 int v0
= (int) floorf(tex_v
- box_v
);
2318 int v1
= (int) ceilf(tex_v
+ box_v
);
2320 float num
[4] = {0.0F
, 0.0F
, 0.0F
, 0.0F
};
2323 args
.face_id
= faces
[j
];
2326 for (v
= v0
; v
<= v1
; ++v
) {
2327 float V
= v
- tex_v
;
2328 float dq
= A
* (2 * U
+ 1) + B
* V
;
2329 float q
= (C
* V
+ B
* U
) * V
+ A
* U
* U
;
2332 for (u
= u0
; u
<= u1
; ++u
) {
2333 /* Note that the ellipse has been pre-scaled so F =
2334 * WEIGHT_LUT_SIZE - 1
2336 if (q
< WEIGHT_LUT_SIZE
) {
2337 /* as a LUT is used, q must never be negative;
2338 * should not happen, though
2340 const int qClamped
= q
>= 0.0F
? q
: 0;
2341 float weight
= weightLut
[qClamped
];
2343 weight_buffer
[buffer_next
] = weight
;
2344 s_buffer
[buffer_next
] = u
/ ((float) width
);
2345 t_buffer
[buffer_next
] = v
/ ((float) height
);
2348 if (buffer_next
== TGSI_QUAD_SIZE
) {
2349 /* 4 texel coords are in the buffer -> read it now */
2351 /* it is assumed that samp->min_img_filter is set to
2352 * img_filter_2d_nearest or one of the
2353 * accelerated img_filter_2d_nearest_XXX functions.
2355 for (jj
= 0; jj
< buffer_next
; jj
++) {
2356 args
.s
= s_buffer
[jj
];
2357 args
.t
= t_buffer
[jj
];
2359 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2360 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2361 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2362 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2363 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2376 /* if the tex coord buffer contains unread values, we will read
2379 if (buffer_next
> 0) {
2381 /* it is assumed that samp->min_img_filter is set to
2382 * img_filter_2d_nearest or one of the
2383 * accelerated img_filter_2d_nearest_XXX functions.
2385 for (jj
= 0; jj
< buffer_next
; jj
++) {
2386 args
.s
= s_buffer
[jj
];
2387 args
.t
= t_buffer
[jj
];
2389 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2390 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2391 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2392 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2393 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2398 /* Reaching this place would mean that no pixels intersected
2399 * the ellipse. This should never happen because the filter
2400 * we use always intersects at least one pixel.
2407 /* not enough pixels in resampling, resort to direct interpolation */
2411 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][j
]);
2413 num
[0] = rgba_temp
[0][j
];
2414 num
[1] = rgba_temp
[1][j
];
2415 num
[2] = rgba_temp
[2][j
];
2416 num
[3] = rgba_temp
[3][j
];
2419 rgba
[0][j
] = num
[0] / den
;
2420 rgba
[1][j
] = num
[1] / den
;
2421 rgba
[2][j
] = num
[2] / den
;
2422 rgba
[3][j
] = num
[3] / den
;
2428 * Get mip level relative to base level for linear mip filter
2431 mip_rel_level_linear_aniso(const struct sp_sampler_view
*sp_sview
,
2432 const struct sp_sampler
*sp_samp
,
2433 const float lod
[TGSI_QUAD_SIZE
],
2434 float level
[TGSI_QUAD_SIZE
])
2436 mip_rel_level_linear(sp_sview
, sp_samp
, lod
, level
);
2440 * Sample 2D texture using an anisotropic filter.
2443 mip_filter_linear_aniso(struct sp_sampler_view
*sp_sview
,
2444 struct sp_sampler
*sp_samp
,
2445 img_filter_func min_filter
,
2446 img_filter_func mag_filter
,
2447 const float s
[TGSI_QUAD_SIZE
],
2448 const float t
[TGSI_QUAD_SIZE
],
2449 const float p
[TGSI_QUAD_SIZE
],
2450 const float c0
[TGSI_QUAD_SIZE
],
2451 const float lod_in
[TGSI_QUAD_SIZE
],
2452 const struct filter_args
*filt_args
,
2453 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2455 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2456 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2459 float lod
[TGSI_QUAD_SIZE
];
2461 float s_to_u
= u_minify(texture
->width0
, psview
->u
.tex
.first_level
);
2462 float t_to_v
= u_minify(texture
->height0
, psview
->u
.tex
.first_level
);
2463 float dudx
= (s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2464 float dudy
= (s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2465 float dvdx
= (t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2466 float dvdy
= (t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2467 struct img_filter_args args
;
2469 if (filt_args
->control
== TGSI_SAMPLER_LOD_BIAS
||
2470 filt_args
->control
== TGSI_SAMPLER_LOD_NONE
||
2472 filt_args
->control
== TGSI_SAMPLER_DERIVS_EXPLICIT
) {
2473 /* note: instead of working with Px and Py, we will use the
2474 * squared length instead, to avoid sqrt.
2476 float Px2
= dudx
* dudx
+ dvdx
* dvdx
;
2477 float Py2
= dudy
* dudy
+ dvdy
* dvdy
;
2482 const float maxEccentricity
= sp_samp
->base
.max_anisotropy
* sp_samp
->base
.max_anisotropy
;
2493 /* if the eccentricity of the ellipse is too big, scale up the shorter
2494 * of the two vectors to limit the maximum amount of work per pixel
2497 if (e
> maxEccentricity
) {
2498 /* float s=e / maxEccentricity;
2502 Pmin2
= Pmax2
/ maxEccentricity
;
2505 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2506 * this since 0.5*log(x) = log(sqrt(x))
2508 lambda
= 0.5F
* util_fast_log2(Pmin2
) + sp_samp
->base
.lod_bias
;
2509 compute_lod(&sp_samp
->base
, filt_args
->control
, lambda
, lod_in
, lod
);
2512 assert(filt_args
->control
== TGSI_SAMPLER_LOD_EXPLICIT
||
2513 filt_args
->control
== TGSI_SAMPLER_LOD_ZERO
);
2514 compute_lod(&sp_samp
->base
, filt_args
->control
, sp_samp
->base
.lod_bias
, lod_in
, lod
);
2517 /* XXX: Take into account all lod values.
2520 level0
= psview
->u
.tex
.first_level
+ (int)lambda
;
2522 /* If the ellipse covers the whole image, we can
2523 * simply return the average of the whole image.
2525 if (level0
>= (int) psview
->u
.tex
.last_level
) {
2527 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2531 args
.level
= psview
->u
.tex
.last_level
;
2532 args
.face_id
= filt_args
->faces
[j
];
2533 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2537 /* don't bother interpolating between multiple LODs; it doesn't
2538 * seem to be worth the extra running time.
2540 img_filter_2d_ewa(sp_sview
, sp_samp
, min_filter
, mag_filter
,
2541 s
, t
, p
, filt_args
->faces
, level0
,
2542 dudx
, dvdx
, dudy
, dvdy
, rgba
);
2546 print_sample_4(__FUNCTION__
, rgba
);
2551 * Get mip level relative to base level for linear mip filter
2554 mip_rel_level_linear_2d_linear_repeat_POT(
2555 const struct sp_sampler_view
*sp_sview
,
2556 const struct sp_sampler
*sp_samp
,
2557 const float lod
[TGSI_QUAD_SIZE
],
2558 float level
[TGSI_QUAD_SIZE
])
2560 mip_rel_level_linear(sp_sview
, sp_samp
, lod
, level
);
2564 * Specialized version of mip_filter_linear with hard-wired calls to
2565 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2568 mip_filter_linear_2d_linear_repeat_POT(
2569 struct sp_sampler_view
*sp_sview
,
2570 struct sp_sampler
*sp_samp
,
2571 img_filter_func min_filter
,
2572 img_filter_func mag_filter
,
2573 const float s
[TGSI_QUAD_SIZE
],
2574 const float t
[TGSI_QUAD_SIZE
],
2575 const float p
[TGSI_QUAD_SIZE
],
2576 const float c0
[TGSI_QUAD_SIZE
],
2577 const float lod_in
[TGSI_QUAD_SIZE
],
2578 const struct filter_args
*filt_args
,
2579 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2581 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2583 float lod
[TGSI_QUAD_SIZE
];
2585 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2587 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2588 int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
2589 struct img_filter_args args
;
2590 /* Catches both negative and large values of level0:
2595 args
.face_id
= filt_args
->faces
[j
];
2596 args
.offset
= filt_args
->offset
;
2597 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2598 if ((unsigned)level0
>= psview
->u
.tex
.last_level
) {
2600 args
.level
= psview
->u
.tex
.first_level
;
2602 args
.level
= psview
->u
.tex
.last_level
;
2603 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
,
2608 float levelBlend
= frac(lod
[j
]);
2609 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2612 args
.level
= level0
;
2613 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
2614 args
.level
= level0
+1;
2615 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
2617 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
2618 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2623 print_sample_4(__FUNCTION__
, rgba
);
2627 static const struct sp_filter_funcs funcs_linear
= {
2628 mip_rel_level_linear
,
2632 static const struct sp_filter_funcs funcs_nearest
= {
2633 mip_rel_level_nearest
,
2637 static const struct sp_filter_funcs funcs_none
= {
2642 static const struct sp_filter_funcs funcs_none_no_filter_select
= {
2643 mip_rel_level_none_no_filter_select
,
2644 mip_filter_none_no_filter_select
2647 static const struct sp_filter_funcs funcs_linear_aniso
= {
2648 mip_rel_level_linear_aniso
,
2649 mip_filter_linear_aniso
2652 static const struct sp_filter_funcs funcs_linear_2d_linear_repeat_POT
= {
2653 mip_rel_level_linear_2d_linear_repeat_POT
,
2654 mip_filter_linear_2d_linear_repeat_POT
2658 * Do shadow/depth comparisons.
2661 sample_compare(struct sp_sampler_view
*sp_sview
,
2662 struct sp_sampler
*sp_samp
,
2663 const float s
[TGSI_QUAD_SIZE
],
2664 const float t
[TGSI_QUAD_SIZE
],
2665 const float p
[TGSI_QUAD_SIZE
],
2666 const float c0
[TGSI_QUAD_SIZE
],
2667 const float c1
[TGSI_QUAD_SIZE
],
2668 enum tgsi_sampler_control control
,
2669 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2671 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
2673 int k
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2675 const struct util_format_description
*format_desc
;
2677 bool is_gather
= (control
== TGSI_SAMPLER_GATHER
);
2680 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2681 * for 2D Array texture we need to use the 'c0' (aka Q).
2682 * When we sampled the depth texture, the depth value was put into all
2683 * RGBA channels. We look at the red channel here.
2686 if (sp_sview
->base
.target
== PIPE_TEXTURE_2D_ARRAY
||
2687 sp_sview
->base
.target
== PIPE_TEXTURE_CUBE
) {
2692 } else if (sp_sview
->base
.target
== PIPE_TEXTURE_CUBE_ARRAY
) {
2704 format_desc
= util_format_description(sp_sview
->base
.format
);
2705 /* not entirely sure we couldn't end up with non-valid swizzle here */
2706 chan_type
= format_desc
->swizzle
[0] <= UTIL_FORMAT_SWIZZLE_W
?
2707 format_desc
->channel
[format_desc
->swizzle
[0]].type
:
2708 UTIL_FORMAT_TYPE_FLOAT
;
2709 if (chan_type
!= UTIL_FORMAT_TYPE_FLOAT
) {
2711 * clamping is a result of conversion to texture format, hence
2712 * doesn't happen with floats. Technically also should do comparison
2713 * in texture format (quantization!).
2715 pc
[0] = CLAMP(pc
[0], 0.0F
, 1.0F
);
2716 pc
[1] = CLAMP(pc
[1], 0.0F
, 1.0F
);
2717 pc
[2] = CLAMP(pc
[2], 0.0F
, 1.0F
);
2718 pc
[3] = CLAMP(pc
[3], 0.0F
, 1.0F
);
2721 for (v
= 0; v
< (is_gather
? TGSI_NUM_CHANNELS
: 1); v
++) {
2722 /* compare four texcoords vs. four texture samples */
2723 switch (sampler
->compare_func
) {
2724 case PIPE_FUNC_LESS
:
2725 k
[v
][0] = pc
[0] < rgba
[v
][0];
2726 k
[v
][1] = pc
[1] < rgba
[v
][1];
2727 k
[v
][2] = pc
[2] < rgba
[v
][2];
2728 k
[v
][3] = pc
[3] < rgba
[v
][3];
2730 case PIPE_FUNC_LEQUAL
:
2731 k
[v
][0] = pc
[0] <= rgba
[v
][0];
2732 k
[v
][1] = pc
[1] <= rgba
[v
][1];
2733 k
[v
][2] = pc
[2] <= rgba
[v
][2];
2734 k
[v
][3] = pc
[3] <= rgba
[v
][3];
2736 case PIPE_FUNC_GREATER
:
2737 k
[v
][0] = pc
[0] > rgba
[v
][0];
2738 k
[v
][1] = pc
[1] > rgba
[v
][1];
2739 k
[v
][2] = pc
[2] > rgba
[v
][2];
2740 k
[v
][3] = pc
[3] > rgba
[v
][3];
2742 case PIPE_FUNC_GEQUAL
:
2743 k
[v
][0] = pc
[0] >= rgba
[v
][0];
2744 k
[v
][1] = pc
[1] >= rgba
[v
][1];
2745 k
[v
][2] = pc
[2] >= rgba
[v
][2];
2746 k
[v
][3] = pc
[3] >= rgba
[v
][3];
2748 case PIPE_FUNC_EQUAL
:
2749 k
[v
][0] = pc
[0] == rgba
[v
][0];
2750 k
[v
][1] = pc
[1] == rgba
[v
][1];
2751 k
[v
][2] = pc
[2] == rgba
[v
][2];
2752 k
[v
][3] = pc
[3] == rgba
[v
][3];
2754 case PIPE_FUNC_NOTEQUAL
:
2755 k
[v
][0] = pc
[0] != rgba
[v
][0];
2756 k
[v
][1] = pc
[1] != rgba
[v
][1];
2757 k
[v
][2] = pc
[2] != rgba
[v
][2];
2758 k
[v
][3] = pc
[3] != rgba
[v
][3];
2760 case PIPE_FUNC_ALWAYS
:
2761 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 1;
2763 case PIPE_FUNC_NEVER
:
2764 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2767 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2774 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2775 for (v
= 0; v
< TGSI_NUM_CHANNELS
; v
++) {
2776 rgba
[v
][j
] = k
[v
][j
];
2780 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2781 rgba
[0][j
] = k
[0][j
];
2782 rgba
[1][j
] = k
[0][j
];
2783 rgba
[2][j
] = k
[0][j
];
2790 do_swizzling(const struct pipe_sampler_view
*sview
,
2791 float in
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
2792 float out
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2795 const unsigned swizzle_r
= sview
->swizzle_r
;
2796 const unsigned swizzle_g
= sview
->swizzle_g
;
2797 const unsigned swizzle_b
= sview
->swizzle_b
;
2798 const unsigned swizzle_a
= sview
->swizzle_a
;
2800 switch (swizzle_r
) {
2801 case PIPE_SWIZZLE_ZERO
:
2802 for (j
= 0; j
< 4; j
++)
2805 case PIPE_SWIZZLE_ONE
:
2806 for (j
= 0; j
< 4; j
++)
2810 assert(swizzle_r
< 4);
2811 for (j
= 0; j
< 4; j
++)
2812 out
[0][j
] = in
[swizzle_r
][j
];
2815 switch (swizzle_g
) {
2816 case PIPE_SWIZZLE_ZERO
:
2817 for (j
= 0; j
< 4; j
++)
2820 case PIPE_SWIZZLE_ONE
:
2821 for (j
= 0; j
< 4; j
++)
2825 assert(swizzle_g
< 4);
2826 for (j
= 0; j
< 4; j
++)
2827 out
[1][j
] = in
[swizzle_g
][j
];
2830 switch (swizzle_b
) {
2831 case PIPE_SWIZZLE_ZERO
:
2832 for (j
= 0; j
< 4; j
++)
2835 case PIPE_SWIZZLE_ONE
:
2836 for (j
= 0; j
< 4; j
++)
2840 assert(swizzle_b
< 4);
2841 for (j
= 0; j
< 4; j
++)
2842 out
[2][j
] = in
[swizzle_b
][j
];
2845 switch (swizzle_a
) {
2846 case PIPE_SWIZZLE_ZERO
:
2847 for (j
= 0; j
< 4; j
++)
2850 case PIPE_SWIZZLE_ONE
:
2851 for (j
= 0; j
< 4; j
++)
2855 assert(swizzle_a
< 4);
2856 for (j
= 0; j
< 4; j
++)
2857 out
[3][j
] = in
[swizzle_a
][j
];
2862 static wrap_nearest_func
2863 get_nearest_unorm_wrap(unsigned mode
)
2866 case PIPE_TEX_WRAP_CLAMP
:
2867 return wrap_nearest_unorm_clamp
;
2868 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2869 return wrap_nearest_unorm_clamp_to_edge
;
2870 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2871 return wrap_nearest_unorm_clamp_to_border
;
2873 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
2874 return wrap_nearest_unorm_clamp
;
2879 static wrap_nearest_func
2880 get_nearest_wrap(unsigned mode
)
2883 case PIPE_TEX_WRAP_REPEAT
:
2884 return wrap_nearest_repeat
;
2885 case PIPE_TEX_WRAP_CLAMP
:
2886 return wrap_nearest_clamp
;
2887 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2888 return wrap_nearest_clamp_to_edge
;
2889 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2890 return wrap_nearest_clamp_to_border
;
2891 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2892 return wrap_nearest_mirror_repeat
;
2893 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2894 return wrap_nearest_mirror_clamp
;
2895 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2896 return wrap_nearest_mirror_clamp_to_edge
;
2897 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2898 return wrap_nearest_mirror_clamp_to_border
;
2901 return wrap_nearest_repeat
;
2906 static wrap_linear_func
2907 get_linear_unorm_wrap(unsigned mode
)
2910 case PIPE_TEX_WRAP_CLAMP
:
2911 return wrap_linear_unorm_clamp
;
2912 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2913 return wrap_linear_unorm_clamp_to_edge
;
2914 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2915 return wrap_linear_unorm_clamp_to_border
;
2917 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
2918 return wrap_linear_unorm_clamp
;
2923 static wrap_linear_func
2924 get_linear_wrap(unsigned mode
)
2927 case PIPE_TEX_WRAP_REPEAT
:
2928 return wrap_linear_repeat
;
2929 case PIPE_TEX_WRAP_CLAMP
:
2930 return wrap_linear_clamp
;
2931 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2932 return wrap_linear_clamp_to_edge
;
2933 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2934 return wrap_linear_clamp_to_border
;
2935 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2936 return wrap_linear_mirror_repeat
;
2937 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2938 return wrap_linear_mirror_clamp
;
2939 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2940 return wrap_linear_mirror_clamp_to_edge
;
2941 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2942 return wrap_linear_mirror_clamp_to_border
;
2945 return wrap_linear_repeat
;
2951 * Is swizzling needed for the given state key?
2954 any_swizzle(const struct pipe_sampler_view
*view
)
2956 return (view
->swizzle_r
!= PIPE_SWIZZLE_RED
||
2957 view
->swizzle_g
!= PIPE_SWIZZLE_GREEN
||
2958 view
->swizzle_b
!= PIPE_SWIZZLE_BLUE
||
2959 view
->swizzle_a
!= PIPE_SWIZZLE_ALPHA
);
2963 static img_filter_func
2964 get_img_filter(const struct sp_sampler_view
*sp_sview
,
2965 const struct pipe_sampler_state
*sampler
,
2966 unsigned filter
, bool gather
)
2968 switch (sp_sview
->base
.target
) {
2970 case PIPE_TEXTURE_1D
:
2971 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2972 return img_filter_1d_nearest
;
2974 return img_filter_1d_linear
;
2976 case PIPE_TEXTURE_1D_ARRAY
:
2977 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2978 return img_filter_1d_array_nearest
;
2980 return img_filter_1d_array_linear
;
2982 case PIPE_TEXTURE_2D
:
2983 case PIPE_TEXTURE_RECT
:
2984 /* Try for fast path:
2986 if (!gather
&& sp_sview
->pot2d
&&
2987 sampler
->wrap_s
== sampler
->wrap_t
&&
2988 sampler
->normalized_coords
)
2990 switch (sampler
->wrap_s
) {
2991 case PIPE_TEX_WRAP_REPEAT
:
2993 case PIPE_TEX_FILTER_NEAREST
:
2994 return img_filter_2d_nearest_repeat_POT
;
2995 case PIPE_TEX_FILTER_LINEAR
:
2996 return img_filter_2d_linear_repeat_POT
;
3001 case PIPE_TEX_WRAP_CLAMP
:
3003 case PIPE_TEX_FILTER_NEAREST
:
3004 return img_filter_2d_nearest_clamp_POT
;
3010 /* Otherwise use default versions:
3012 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3013 return img_filter_2d_nearest
;
3015 return img_filter_2d_linear
;
3017 case PIPE_TEXTURE_2D_ARRAY
:
3018 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3019 return img_filter_2d_array_nearest
;
3021 return img_filter_2d_array_linear
;
3023 case PIPE_TEXTURE_CUBE
:
3024 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3025 return img_filter_cube_nearest
;
3027 return img_filter_cube_linear
;
3029 case PIPE_TEXTURE_CUBE_ARRAY
:
3030 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3031 return img_filter_cube_array_nearest
;
3033 return img_filter_cube_array_linear
;
3035 case PIPE_TEXTURE_3D
:
3036 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3037 return img_filter_3d_nearest
;
3039 return img_filter_3d_linear
;
3043 return img_filter_1d_nearest
;
3048 * Get mip filter funcs, and optionally both img min filter and img mag
3049 * filter. Note that both img filter function pointers must be either non-NULL
3053 get_filters(const struct sp_sampler_view
*sp_sview
,
3054 const struct sp_sampler
*sp_samp
,
3055 const enum tgsi_sampler_control control
,
3056 const struct sp_filter_funcs
**funcs
,
3057 img_filter_func
*min
,
3058 img_filter_func
*mag
)
3061 if (control
== TGSI_SAMPLER_GATHER
) {
3062 *funcs
= &funcs_nearest
;
3064 *min
= get_img_filter(sp_sview
, &sp_samp
->base
,
3065 PIPE_TEX_FILTER_LINEAR
, true);
3067 } else if (sp_sview
->pot2d
& sp_samp
->min_mag_equal_repeat_linear
) {
3068 *funcs
= &funcs_linear_2d_linear_repeat_POT
;
3070 *funcs
= sp_samp
->filter_funcs
;
3073 *min
= get_img_filter(sp_sview
, &sp_samp
->base
,
3074 sp_samp
->min_img_filter
, false);
3075 if (sp_samp
->min_mag_equal
) {
3078 *mag
= get_img_filter(sp_sview
, &sp_samp
->base
,
3079 sp_samp
->base
.mag_img_filter
, false);
3086 sample_mip(struct sp_sampler_view
*sp_sview
,
3087 struct sp_sampler
*sp_samp
,
3088 const float s
[TGSI_QUAD_SIZE
],
3089 const float t
[TGSI_QUAD_SIZE
],
3090 const float p
[TGSI_QUAD_SIZE
],
3091 const float c0
[TGSI_QUAD_SIZE
],
3092 const float lod
[TGSI_QUAD_SIZE
],
3093 const struct filter_args
*filt_args
,
3094 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3096 const struct sp_filter_funcs
*funcs
= NULL
;
3097 img_filter_func min_img_filter
= NULL
;
3098 img_filter_func mag_img_filter
= NULL
;
3100 get_filters(sp_sview
, sp_samp
, filt_args
->control
,
3101 &funcs
, &min_img_filter
, &mag_img_filter
);
3103 funcs
->filter(sp_sview
, sp_samp
, min_img_filter
, mag_img_filter
,
3104 s
, t
, p
, c0
, lod
, filt_args
, rgba
);
3106 if (sp_samp
->base
.compare_mode
!= PIPE_TEX_COMPARE_NONE
) {
3107 sample_compare(sp_sview
, sp_samp
, s
, t
, p
, c0
,
3108 lod
, filt_args
->control
, rgba
);
3111 if (sp_sview
->need_swizzle
&& filt_args
->control
!= TGSI_SAMPLER_GATHER
) {
3112 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3113 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
3114 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
3121 * This function uses cube texture coordinates to choose a face of a cube and
3122 * computes the 2D cube face coordinates. Puts face info into the sampler
3126 convert_cube(const struct sp_sampler_view
*sp_sview
,
3127 const struct sp_sampler
*sp_samp
,
3128 const float s
[TGSI_QUAD_SIZE
],
3129 const float t
[TGSI_QUAD_SIZE
],
3130 const float p
[TGSI_QUAD_SIZE
],
3131 const float c0
[TGSI_QUAD_SIZE
],
3132 float ssss
[TGSI_QUAD_SIZE
],
3133 float tttt
[TGSI_QUAD_SIZE
],
3134 float pppp
[TGSI_QUAD_SIZE
],
3135 float faces
[TGSI_QUAD_SIZE
])
3145 direction target sc tc ma
3146 ---------- ------------------------------- --- --- ---
3147 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
3148 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
3149 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
3150 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
3151 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
3152 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
3155 /* Choose the cube face and compute new s/t coords for the 2D face.
3157 * Use the same cube face for all four pixels in the quad.
3159 * This isn't ideal, but if we want to use a different cube face
3160 * per pixel in the quad, we'd have to also compute the per-face
3161 * LOD here too. That's because the four post-face-selection
3162 * texcoords are no longer related to each other (they're
3163 * per-face!) so we can't use subtraction to compute the partial
3164 * deriviates to compute the LOD. Doing so (near cube edges
3165 * anyway) gives us pretty much random values.
3168 /* use the average of the four pixel's texcoords to choose the face */
3169 const float rx
= 0.25F
* (s
[0] + s
[1] + s
[2] + s
[3]);
3170 const float ry
= 0.25F
* (t
[0] + t
[1] + t
[2] + t
[3]);
3171 const float rz
= 0.25F
* (p
[0] + p
[1] + p
[2] + p
[3]);
3172 const float arx
= fabsf(rx
), ary
= fabsf(ry
), arz
= fabsf(rz
);
3174 if (arx
>= ary
&& arx
>= arz
) {
3175 const float sign
= (rx
>= 0.0F
) ? 1.0F
: -1.0F
;
3176 const uint face
= (rx
>= 0.0F
) ?
3177 PIPE_TEX_FACE_POS_X
: PIPE_TEX_FACE_NEG_X
;
3178 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3179 const float ima
= -0.5F
/ fabsf(s
[j
]);
3180 ssss
[j
] = sign
* p
[j
] * ima
+ 0.5F
;
3181 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3185 else if (ary
>= arx
&& ary
>= arz
) {
3186 const float sign
= (ry
>= 0.0F
) ? 1.0F
: -1.0F
;
3187 const uint face
= (ry
>= 0.0F
) ?
3188 PIPE_TEX_FACE_POS_Y
: PIPE_TEX_FACE_NEG_Y
;
3189 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3190 const float ima
= -0.5F
/ fabsf(t
[j
]);
3191 ssss
[j
] = -s
[j
] * ima
+ 0.5F
;
3192 tttt
[j
] = sign
* -p
[j
] * ima
+ 0.5F
;
3197 const float sign
= (rz
>= 0.0F
) ? 1.0F
: -1.0F
;
3198 const uint face
= (rz
>= 0.0F
) ?
3199 PIPE_TEX_FACE_POS_Z
: PIPE_TEX_FACE_NEG_Z
;
3200 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3201 const float ima
= -0.5F
/ fabsf(p
[j
]);
3202 ssss
[j
] = sign
* -s
[j
] * ima
+ 0.5F
;
3203 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3212 sp_get_dims(struct sp_sampler_view
*sp_sview
, int level
,
3215 const struct pipe_sampler_view
*view
= &sp_sview
->base
;
3216 const struct pipe_resource
*texture
= view
->texture
;
3218 if (view
->target
== PIPE_BUFFER
) {
3219 dims
[0] = (view
->u
.buf
.last_element
- view
->u
.buf
.first_element
) + 1;
3220 /* the other values are undefined, but let's avoid potential valgrind
3223 dims
[1] = dims
[2] = dims
[3] = 0;
3227 /* undefined according to EXT_gpu_program */
3228 level
+= view
->u
.tex
.first_level
;
3229 if (level
> view
->u
.tex
.last_level
)
3232 dims
[3] = view
->u
.tex
.last_level
- view
->u
.tex
.first_level
+ 1;
3233 dims
[0] = u_minify(texture
->width0
, level
);
3235 switch (view
->target
) {
3236 case PIPE_TEXTURE_1D_ARRAY
:
3237 dims
[1] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3239 case PIPE_TEXTURE_1D
:
3241 case PIPE_TEXTURE_2D_ARRAY
:
3242 dims
[2] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3244 case PIPE_TEXTURE_2D
:
3245 case PIPE_TEXTURE_CUBE
:
3246 case PIPE_TEXTURE_RECT
:
3247 dims
[1] = u_minify(texture
->height0
, level
);
3249 case PIPE_TEXTURE_3D
:
3250 dims
[1] = u_minify(texture
->height0
, level
);
3251 dims
[2] = u_minify(texture
->depth0
, level
);
3253 case PIPE_TEXTURE_CUBE_ARRAY
:
3254 dims
[1] = u_minify(texture
->height0
, level
);
3255 dims
[2] = (view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1) / 6;
3258 assert(!"unexpected texture target in sp_get_dims()");
3264 * This function is only used for getting unfiltered texels via the
3265 * TXF opcode. The GL spec says that out-of-bounds texel fetches
3266 * produce undefined results. Instead of crashing, lets just clamp
3267 * coords to the texture image size.
3270 sp_get_texels(struct sp_sampler_view
*sp_sview
,
3271 const int v_i
[TGSI_QUAD_SIZE
],
3272 const int v_j
[TGSI_QUAD_SIZE
],
3273 const int v_k
[TGSI_QUAD_SIZE
],
3274 const int lod
[TGSI_QUAD_SIZE
],
3275 const int8_t offset
[3],
3276 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3278 union tex_tile_address addr
;
3279 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
3282 int width
, height
, depth
;
3285 /* TODO write a better test for LOD */
3286 addr
.bits
.level
= sp_sview
->base
.target
== PIPE_BUFFER
? 0 :
3287 CLAMP(lod
[0] + sp_sview
->base
.u
.tex
.first_level
,
3288 sp_sview
->base
.u
.tex
.first_level
,
3289 sp_sview
->base
.u
.tex
.last_level
);
3291 width
= u_minify(texture
->width0
, addr
.bits
.level
);
3292 height
= u_minify(texture
->height0
, addr
.bits
.level
);
3293 depth
= u_minify(texture
->depth0
, addr
.bits
.level
);
3295 switch (sp_sview
->base
.target
) {
3297 case PIPE_TEXTURE_1D
:
3298 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3299 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3300 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, 0);
3301 for (c
= 0; c
< 4; c
++) {
3306 case PIPE_TEXTURE_1D_ARRAY
:
3307 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3308 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3309 int y
= CLAMP(v_j
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3310 sp_sview
->base
.u
.tex
.last_layer
);
3311 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
3312 for (c
= 0; c
< 4; c
++) {
3317 case PIPE_TEXTURE_2D
:
3318 case PIPE_TEXTURE_RECT
:
3319 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3320 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3321 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3322 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
3323 for (c
= 0; c
< 4; c
++) {
3328 case PIPE_TEXTURE_2D_ARRAY
:
3329 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3330 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3331 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3332 int layer
= CLAMP(v_k
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3333 sp_sview
->base
.u
.tex
.last_layer
);
3334 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
3335 for (c
= 0; c
< 4; c
++) {
3340 case PIPE_TEXTURE_3D
:
3341 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3342 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3343 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3344 int z
= CLAMP(v_k
[j
] + offset
[2], 0, depth
- 1);
3345 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, z
);
3346 for (c
= 0; c
< 4; c
++) {
3351 case PIPE_TEXTURE_CUBE
: /* TXF can't work on CUBE according to spec */
3353 assert(!"Unknown or CUBE texture type in TXF processing\n");
3357 if (sp_sview
->need_swizzle
) {
3358 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3359 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
3360 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
3366 softpipe_create_sampler_state(struct pipe_context
*pipe
,
3367 const struct pipe_sampler_state
*sampler
)
3369 struct sp_sampler
*samp
= CALLOC_STRUCT(sp_sampler
);
3371 samp
->base
= *sampler
;
3373 /* Note that (for instance) linear_texcoord_s and
3374 * nearest_texcoord_s may be active at the same time, if the
3375 * sampler min_img_filter differs from its mag_img_filter.
3377 if (sampler
->normalized_coords
) {
3378 samp
->linear_texcoord_s
= get_linear_wrap( sampler
->wrap_s
);
3379 samp
->linear_texcoord_t
= get_linear_wrap( sampler
->wrap_t
);
3380 samp
->linear_texcoord_p
= get_linear_wrap( sampler
->wrap_r
);
3382 samp
->nearest_texcoord_s
= get_nearest_wrap( sampler
->wrap_s
);
3383 samp
->nearest_texcoord_t
= get_nearest_wrap( sampler
->wrap_t
);
3384 samp
->nearest_texcoord_p
= get_nearest_wrap( sampler
->wrap_r
);
3387 samp
->linear_texcoord_s
= get_linear_unorm_wrap( sampler
->wrap_s
);
3388 samp
->linear_texcoord_t
= get_linear_unorm_wrap( sampler
->wrap_t
);
3389 samp
->linear_texcoord_p
= get_linear_unorm_wrap( sampler
->wrap_r
);
3391 samp
->nearest_texcoord_s
= get_nearest_unorm_wrap( sampler
->wrap_s
);
3392 samp
->nearest_texcoord_t
= get_nearest_unorm_wrap( sampler
->wrap_t
);
3393 samp
->nearest_texcoord_p
= get_nearest_unorm_wrap( sampler
->wrap_r
);
3396 samp
->min_img_filter
= sampler
->min_img_filter
;
3398 switch (sampler
->min_mip_filter
) {
3399 case PIPE_TEX_MIPFILTER_NONE
:
3400 if (sampler
->min_img_filter
== sampler
->mag_img_filter
)
3401 samp
->filter_funcs
= &funcs_none_no_filter_select
;
3403 samp
->filter_funcs
= &funcs_none
;
3406 case PIPE_TEX_MIPFILTER_NEAREST
:
3407 samp
->filter_funcs
= &funcs_nearest
;
3410 case PIPE_TEX_MIPFILTER_LINEAR
:
3411 if (sampler
->min_img_filter
== sampler
->mag_img_filter
&&
3412 sampler
->normalized_coords
&&
3413 sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
&&
3414 sampler
->wrap_t
== PIPE_TEX_WRAP_REPEAT
&&
3415 sampler
->min_img_filter
== PIPE_TEX_FILTER_LINEAR
&&
3416 sampler
->max_anisotropy
<= 1) {
3417 samp
->min_mag_equal_repeat_linear
= TRUE
;
3419 samp
->filter_funcs
= &funcs_linear
;
3421 /* Anisotropic filtering extension. */
3422 if (sampler
->max_anisotropy
> 1) {
3423 samp
->filter_funcs
= &funcs_linear_aniso
;
3425 /* Override min_img_filter:
3426 * min_img_filter needs to be set to NEAREST since we need to access
3427 * each texture pixel as it is and weight it later; using linear
3428 * filters will have incorrect results.
3429 * By setting the filter to NEAREST here, we can avoid calling the
3430 * generic img_filter_2d_nearest in the anisotropic filter function,
3431 * making it possible to use one of the accelerated implementations
3433 samp
->min_img_filter
= PIPE_TEX_FILTER_NEAREST
;
3435 /* on first access create the lookup table containing the filter weights. */
3437 create_filter_table();
3442 if (samp
->min_img_filter
== sampler
->mag_img_filter
) {
3443 samp
->min_mag_equal
= TRUE
;
3446 return (void *)samp
;
3451 softpipe_get_lambda_func(const struct pipe_sampler_view
*view
, unsigned shader
)
3453 if (shader
!= PIPE_SHADER_FRAGMENT
)
3454 return compute_lambda_vert
;
3456 switch (view
->target
) {
3458 case PIPE_TEXTURE_1D
:
3459 case PIPE_TEXTURE_1D_ARRAY
:
3460 return compute_lambda_1d
;
3461 case PIPE_TEXTURE_2D
:
3462 case PIPE_TEXTURE_2D_ARRAY
:
3463 case PIPE_TEXTURE_RECT
:
3464 case PIPE_TEXTURE_CUBE
:
3465 case PIPE_TEXTURE_CUBE_ARRAY
:
3466 return compute_lambda_2d
;
3467 case PIPE_TEXTURE_3D
:
3468 return compute_lambda_3d
;
3471 return compute_lambda_1d
;
3476 struct pipe_sampler_view
*
3477 softpipe_create_sampler_view(struct pipe_context
*pipe
,
3478 struct pipe_resource
*resource
,
3479 const struct pipe_sampler_view
*templ
)
3481 struct sp_sampler_view
*sview
= CALLOC_STRUCT(sp_sampler_view
);
3482 struct softpipe_resource
*spr
= (struct softpipe_resource
*)resource
;
3485 struct pipe_sampler_view
*view
= &sview
->base
;
3487 view
->reference
.count
= 1;
3488 view
->texture
= NULL
;
3489 pipe_resource_reference(&view
->texture
, resource
);
3490 view
->context
= pipe
;
3494 * This is possibly too lenient, but the primary reason is just
3495 * to catch state trackers which forget to initialize this, so
3496 * it only catches clearly impossible view targets.
3498 if (view
->target
!= resource
->target
) {
3499 if (view
->target
== PIPE_TEXTURE_1D
)
3500 assert(resource
->target
== PIPE_TEXTURE_1D_ARRAY
);
3501 else if (view
->target
== PIPE_TEXTURE_1D_ARRAY
)
3502 assert(resource
->target
== PIPE_TEXTURE_1D
);
3503 else if (view
->target
== PIPE_TEXTURE_2D
)
3504 assert(resource
->target
== PIPE_TEXTURE_2D_ARRAY
||
3505 resource
->target
== PIPE_TEXTURE_CUBE
||
3506 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3507 else if (view
->target
== PIPE_TEXTURE_2D_ARRAY
)
3508 assert(resource
->target
== PIPE_TEXTURE_2D
||
3509 resource
->target
== PIPE_TEXTURE_CUBE
||
3510 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3511 else if (view
->target
== PIPE_TEXTURE_CUBE
)
3512 assert(resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
||
3513 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3514 else if (view
->target
== PIPE_TEXTURE_CUBE_ARRAY
)
3515 assert(resource
->target
== PIPE_TEXTURE_CUBE
||
3516 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3522 if (any_swizzle(view
)) {
3523 sview
->need_swizzle
= TRUE
;
3526 sview
->need_cube_convert
= (view
->target
== PIPE_TEXTURE_CUBE
||
3527 view
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3528 sview
->pot2d
= spr
->pot
&&
3529 (view
->target
== PIPE_TEXTURE_2D
||
3530 view
->target
== PIPE_TEXTURE_RECT
);
3532 sview
->xpot
= util_logbase2( resource
->width0
);
3533 sview
->ypot
= util_logbase2( resource
->height0
);
3536 return (struct pipe_sampler_view
*) sview
;
3541 sp_tgsi_get_dims(struct tgsi_sampler
*tgsi_sampler
,
3542 const unsigned sview_index
,
3543 int level
, int dims
[4])
3545 struct sp_tgsi_sampler
*sp_samp
= (struct sp_tgsi_sampler
*)tgsi_sampler
;
3547 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3548 /* always have a view here but texture is NULL if no sampler view was set. */
3549 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3550 dims
[0] = dims
[1] = dims
[2] = dims
[3] = 0;
3553 sp_get_dims(&sp_samp
->sp_sview
[sview_index
], level
, dims
);
3558 sp_tgsi_get_samples(struct tgsi_sampler
*tgsi_sampler
,
3559 const unsigned sview_index
,
3560 const unsigned sampler_index
,
3561 const float s
[TGSI_QUAD_SIZE
],
3562 const float t
[TGSI_QUAD_SIZE
],
3563 const float p
[TGSI_QUAD_SIZE
],
3564 const float c0
[TGSI_QUAD_SIZE
],
3565 const float lod
[TGSI_QUAD_SIZE
],
3566 float derivs
[3][2][TGSI_QUAD_SIZE
],
3567 const int8_t offset
[3],
3568 enum tgsi_sampler_control control
,
3569 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3571 struct sp_tgsi_sampler
*sp_tgsi_samp
= (struct sp_tgsi_sampler
*)tgsi_sampler
;
3572 struct sp_sampler_view
*sp_sview
;
3573 struct sp_sampler
*sp_samp
;
3574 struct filter_args filt_args
;
3576 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3577 assert(sampler_index
< PIPE_MAX_SAMPLERS
);
3578 assert(sp_tgsi_samp
->sp_sampler
[sampler_index
]);
3580 sp_sview
= &sp_tgsi_samp
->sp_sview
[sview_index
];
3581 sp_samp
= sp_tgsi_samp
->sp_sampler
[sampler_index
];
3582 /* always have a view here but texture is NULL if no sampler view was set. */
3583 if (!sp_sview
->base
.texture
) {
3585 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3586 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3593 filt_args
.control
= control
;
3594 filt_args
.offset
= offset
;
3596 if (sp_sview
->need_cube_convert
) {
3597 float cs
[TGSI_QUAD_SIZE
];
3598 float ct
[TGSI_QUAD_SIZE
];
3599 float cp
[TGSI_QUAD_SIZE
];
3600 float faces
[TGSI_QUAD_SIZE
];
3602 convert_cube(sp_sview
, sp_samp
, s
, t
, p
, c0
, cs
, ct
, cp
, faces
);
3604 filt_args
.faces
= faces
;
3605 sample_mip(sp_sview
, sp_samp
, cs
, ct
, cp
, c0
, lod
, &filt_args
, rgba
);
3607 static const float zero_faces
[TGSI_QUAD_SIZE
] = {0.0f
, 0.0f
, 0.0f
, 0.0f
};
3609 filt_args
.faces
= zero_faces
;
3610 sample_mip(sp_sview
, sp_samp
, s
, t
, p
, c0
, lod
, &filt_args
, rgba
);
3615 sp_tgsi_query_lod(const struct tgsi_sampler
*tgsi_sampler
,
3616 const unsigned sview_index
,
3617 const unsigned sampler_index
,
3618 const float s
[TGSI_QUAD_SIZE
],
3619 const float t
[TGSI_QUAD_SIZE
],
3620 const float p
[TGSI_QUAD_SIZE
],
3621 const float c0
[TGSI_QUAD_SIZE
],
3622 const enum tgsi_sampler_control control
,
3623 float mipmap
[TGSI_QUAD_SIZE
],
3624 float lod
[TGSI_QUAD_SIZE
])
3626 static const float lod_in
[TGSI_QUAD_SIZE
] = { 0.0, 0.0, 0.0, 0.0 };
3628 const struct sp_tgsi_sampler
*sp_tgsi_samp
=
3629 (const struct sp_tgsi_sampler
*)tgsi_sampler
;
3630 const struct sp_sampler_view
*sp_sview
;
3631 const struct sp_sampler
*sp_samp
;
3632 const struct sp_filter_funcs
*funcs
;
3635 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3636 assert(sampler_index
< PIPE_MAX_SAMPLERS
);
3637 assert(sp_tgsi_samp
->sp_sampler
[sampler_index
]);
3639 sp_sview
= &sp_tgsi_samp
->sp_sview
[sview_index
];
3640 sp_samp
= sp_tgsi_samp
->sp_sampler
[sampler_index
];
3641 /* always have a view here but texture is NULL if no sampler view was
3643 if (!sp_sview
->base
.texture
) {
3644 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3651 if (sp_sview
->need_cube_convert
) {
3652 float cs
[TGSI_QUAD_SIZE
];
3653 float ct
[TGSI_QUAD_SIZE
];
3654 float cp
[TGSI_QUAD_SIZE
];
3655 float unused_faces
[TGSI_QUAD_SIZE
];
3657 convert_cube(sp_sview
, sp_samp
, s
, t
, p
, c0
, cs
, ct
, cp
, unused_faces
);
3658 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
3659 cs
, ct
, cp
, lod_in
, control
, lod
);
3661 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
3662 s
, t
, p
, lod_in
, control
, lod
);
3665 get_filters(sp_sview
, sp_samp
, control
, &funcs
, NULL
, NULL
);
3666 funcs
->relative_level(sp_sview
, sp_samp
, lod
, mipmap
);
3670 sp_tgsi_get_texel(struct tgsi_sampler
*tgsi_sampler
,
3671 const unsigned sview_index
,
3672 const int i
[TGSI_QUAD_SIZE
],
3673 const int j
[TGSI_QUAD_SIZE
], const int k
[TGSI_QUAD_SIZE
],
3674 const int lod
[TGSI_QUAD_SIZE
], const int8_t offset
[3],
3675 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3677 struct sp_tgsi_sampler
*sp_samp
= (struct sp_tgsi_sampler
*)tgsi_sampler
;
3679 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3680 /* always have a view here but texture is NULL if no sampler view was set. */
3681 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3683 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3684 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3690 sp_get_texels(&sp_samp
->sp_sview
[sview_index
], i
, j
, k
, lod
, offset
, rgba
);
3694 struct sp_tgsi_sampler
*
3695 sp_create_tgsi_sampler(void)
3697 struct sp_tgsi_sampler
*samp
= CALLOC_STRUCT(sp_tgsi_sampler
);
3701 samp
->base
.get_dims
= sp_tgsi_get_dims
;
3702 samp
->base
.get_samples
= sp_tgsi_get_samples
;
3703 samp
->base
.get_texel
= sp_tgsi_get_texel
;
3704 samp
->base
.query_lod
= sp_tgsi_query_lod
;