1 /**************************************************************************
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
5 * Copyright 2008-2010 VMware, Inc. All rights reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
37 #include "pipe/p_context.h"
38 #include "pipe/p_defines.h"
39 #include "pipe/p_shader_tokens.h"
40 #include "util/u_math.h"
41 #include "util/u_format.h"
42 #include "util/u_memory.h"
43 #include "sp_quad.h" /* only for #define QUAD_* tokens */
44 #include "sp_tex_sample.h"
45 #include "sp_tex_tile_cache.h"
48 /** Set to one to help debug texture sampling */
53 * Return fractional part of 'f'. Used for computing interpolation weights.
54 * Need to be careful with negative values.
55 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
56 * of improperly weighted linear-filtered textures.
57 * The tests/texwrap.c demo is a good test.
68 * Linear interpolation macro
71 lerp(float a
, float v0
, float v1
)
73 return v0
+ a
* (v1
- v0
);
78 * Do 2D/bilinear interpolation of float values.
79 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
80 * a and b are the horizontal and vertical interpolants.
81 * It's important that this function is inlined when compiled with
82 * optimization! If we find that's not true on some systems, convert
86 lerp_2d(float a
, float b
,
87 float v00
, float v10
, float v01
, float v11
)
89 const float temp0
= lerp(a
, v00
, v10
);
90 const float temp1
= lerp(a
, v01
, v11
);
91 return lerp(b
, temp0
, temp1
);
96 * As above, but 3D interpolation of 8 values.
99 lerp_3d(float a
, float b
, float c
,
100 float v000
, float v100
, float v010
, float v110
,
101 float v001
, float v101
, float v011
, float v111
)
103 const float temp0
= lerp_2d(a
, b
, v000
, v100
, v010
, v110
);
104 const float temp1
= lerp_2d(a
, b
, v001
, v101
, v011
, v111
);
105 return lerp(c
, temp0
, temp1
);
111 * Compute coord % size for repeat wrap modes.
112 * Note that if coord is negative, coord % size doesn't give the right
113 * value. To avoid that problem we add a large multiple of the size
114 * (rather than using a conditional).
117 repeat(int coord
, unsigned size
)
119 return (coord
+ size
* 1024) % size
;
124 * Apply texture coord wrapping mode and return integer texture indexes
125 * for a vector of four texcoords (S or T or P).
126 * \param wrapMode PIPE_TEX_WRAP_x
127 * \param s the incoming texcoords
128 * \param size the texture image size
129 * \param icoord returns the integer texcoords
130 * \return integer texture index
133 wrap_nearest_repeat(float s
, unsigned size
, int *icoord
)
135 /* s limited to [0,1) */
136 /* i limited to [0,size-1] */
137 int i
= util_ifloor(s
* size
);
138 *icoord
= repeat(i
, size
);
143 wrap_nearest_clamp(float s
, unsigned size
, int *icoord
)
145 /* s limited to [0,1] */
146 /* i limited to [0,size-1] */
152 *icoord
= util_ifloor(s
* size
);
157 wrap_nearest_clamp_to_edge(float s
, unsigned size
, int *icoord
)
159 /* s limited to [min,max] */
160 /* i limited to [0, size-1] */
161 const float min
= 1.0F
/ (2.0F
* size
);
162 const float max
= 1.0F
- min
;
168 *icoord
= util_ifloor(s
* size
);
173 wrap_nearest_clamp_to_border(float s
, unsigned size
, int *icoord
)
175 /* s limited to [min,max] */
176 /* i limited to [-1, size] */
177 const float min
= -1.0F
/ (2.0F
* size
);
178 const float max
= 1.0F
- min
;
184 *icoord
= util_ifloor(s
* size
);
189 wrap_nearest_mirror_repeat(float s
, unsigned size
, int *icoord
)
191 const float min
= 1.0F
/ (2.0F
* size
);
192 const float max
= 1.0F
- min
;
193 const int flr
= util_ifloor(s
);
202 *icoord
= util_ifloor(u
* size
);
207 wrap_nearest_mirror_clamp(float s
, unsigned size
, int *icoord
)
209 /* s limited to [0,1] */
210 /* i limited to [0,size-1] */
211 const float u
= fabsf(s
);
217 *icoord
= util_ifloor(u
* size
);
222 wrap_nearest_mirror_clamp_to_edge(float s
, unsigned size
, int *icoord
)
224 /* s limited to [min,max] */
225 /* i limited to [0, size-1] */
226 const float min
= 1.0F
/ (2.0F
* size
);
227 const float max
= 1.0F
- min
;
228 const float u
= fabsf(s
);
234 *icoord
= util_ifloor(u
* size
);
239 wrap_nearest_mirror_clamp_to_border(float s
, unsigned size
, int *icoord
)
241 /* s limited to [min,max] */
242 /* i limited to [0, size-1] */
243 const float min
= -1.0F
/ (2.0F
* size
);
244 const float max
= 1.0F
- min
;
245 const float u
= fabsf(s
);
251 *icoord
= util_ifloor(u
* size
);
256 * Used to compute texel locations for linear sampling
257 * \param wrapMode PIPE_TEX_WRAP_x
258 * \param s the texcoord
259 * \param size the texture image size
260 * \param icoord0 returns first texture index
261 * \param icoord1 returns second texture index (usually icoord0 + 1)
262 * \param w returns blend factor/weight between texture indices
263 * \param icoord returns the computed integer texture coord
266 wrap_linear_repeat(float s
, unsigned size
,
267 int *icoord0
, int *icoord1
, float *w
)
269 float u
= s
* size
- 0.5F
;
270 *icoord0
= repeat(util_ifloor(u
), size
);
271 *icoord1
= repeat(*icoord0
+ 1, size
);
277 wrap_linear_clamp(float s
, unsigned size
,
278 int *icoord0
, int *icoord1
, float *w
)
280 float u
= CLAMP(s
, 0.0F
, 1.0F
);
282 *icoord0
= util_ifloor(u
);
283 *icoord1
= *icoord0
+ 1;
289 wrap_linear_clamp_to_edge(float s
, unsigned size
,
290 int *icoord0
, int *icoord1
, float *w
)
292 float u
= CLAMP(s
, 0.0F
, 1.0F
);
294 *icoord0
= util_ifloor(u
);
295 *icoord1
= *icoord0
+ 1;
298 if (*icoord1
>= (int) size
)
305 wrap_linear_clamp_to_border(float s
, unsigned size
,
306 int *icoord0
, int *icoord1
, float *w
)
308 const float min
= -1.0F
/ (2.0F
* size
);
309 const float max
= 1.0F
- min
;
310 float u
= CLAMP(s
, min
, max
);
312 *icoord0
= util_ifloor(u
);
313 *icoord1
= *icoord0
+ 1;
319 wrap_linear_mirror_repeat(float s
, unsigned size
,
320 int *icoord0
, int *icoord1
, float *w
)
322 const int flr
= util_ifloor(s
);
327 *icoord0
= util_ifloor(u
);
328 *icoord1
= *icoord0
+ 1;
331 if (*icoord1
>= (int) size
)
338 wrap_linear_mirror_clamp(float s
, unsigned size
,
339 int *icoord0
, int *icoord1
, float *w
)
347 *icoord0
= util_ifloor(u
);
348 *icoord1
= *icoord0
+ 1;
354 wrap_linear_mirror_clamp_to_edge(float s
, unsigned size
,
355 int *icoord0
, int *icoord1
, float *w
)
363 *icoord0
= util_ifloor(u
);
364 *icoord1
= *icoord0
+ 1;
367 if (*icoord1
>= (int) size
)
374 wrap_linear_mirror_clamp_to_border(float s
, unsigned size
,
375 int *icoord0
, int *icoord1
, float *w
)
377 const float min
= -1.0F
/ (2.0F
* size
);
378 const float max
= 1.0F
- min
;
387 *icoord0
= util_ifloor(u
);
388 *icoord1
= *icoord0
+ 1;
394 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
397 wrap_nearest_unorm_clamp(float s
, unsigned size
, int *icoord
)
399 int i
= util_ifloor(s
);
400 *icoord
= CLAMP(i
, 0, (int) size
-1);
405 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
408 wrap_nearest_unorm_clamp_to_border(float s
, unsigned size
, int *icoord
)
410 *icoord
= util_ifloor( CLAMP(s
, -0.5F
, (float) size
+ 0.5F
) );
415 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
418 wrap_nearest_unorm_clamp_to_edge(float s
, unsigned size
, int *icoord
)
420 *icoord
= util_ifloor( CLAMP(s
, 0.5F
, (float) size
- 0.5F
) );
425 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
428 wrap_linear_unorm_clamp(float s
, unsigned size
,
429 int *icoord0
, int *icoord1
, float *w
)
431 /* Not exactly what the spec says, but it matches NVIDIA output */
432 float u
= CLAMP(s
- 0.5F
, 0.0f
, (float) size
- 1.0f
);
433 *icoord0
= util_ifloor(u
);
434 *icoord1
= *icoord0
+ 1;
440 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
443 wrap_linear_unorm_clamp_to_border(float s
, unsigned size
,
444 int *icoord0
, int *icoord1
, float *w
)
446 float u
= CLAMP(s
, -0.5F
, (float) size
+ 0.5F
);
448 *icoord0
= util_ifloor(u
);
449 *icoord1
= *icoord0
+ 1;
450 if (*icoord1
> (int) size
- 1)
457 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
460 wrap_linear_unorm_clamp_to_edge(float s
, unsigned size
,
461 int *icoord0
, int *icoord1
, float *w
)
463 float u
= CLAMP(s
, +0.5F
, (float) size
- 0.5F
);
465 *icoord0
= util_ifloor(u
);
466 *icoord1
= *icoord0
+ 1;
467 if (*icoord1
> (int) size
- 1)
474 * Do coordinate to array index conversion. For array textures.
477 wrap_array_layer(float coord
, unsigned size
, int *layer
)
479 int c
= util_ifloor(coord
+ 0.5F
);
480 *layer
= CLAMP(c
, 0, size
- 1);
485 * Examine the quad's texture coordinates to compute the partial
486 * derivatives w.r.t X and Y, then compute lambda (level of detail).
489 compute_lambda_1d(const struct sp_sampler_variant
*samp
,
490 const float s
[TGSI_QUAD_SIZE
],
491 const float t
[TGSI_QUAD_SIZE
],
492 const float p
[TGSI_QUAD_SIZE
])
494 const struct pipe_resource
*texture
= samp
->view
->texture
;
495 float dsdx
= fabsf(s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]);
496 float dsdy
= fabsf(s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]);
497 float rho
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, samp
->view
->u
.tex
.first_level
);
499 return util_fast_log2(rho
);
504 compute_lambda_2d(const struct sp_sampler_variant
*samp
,
505 const float s
[TGSI_QUAD_SIZE
],
506 const float t
[TGSI_QUAD_SIZE
],
507 const float p
[TGSI_QUAD_SIZE
])
509 const struct pipe_resource
*texture
= samp
->view
->texture
;
510 float dsdx
= fabsf(s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]);
511 float dsdy
= fabsf(s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]);
512 float dtdx
= fabsf(t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]);
513 float dtdy
= fabsf(t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]);
514 float maxx
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, samp
->view
->u
.tex
.first_level
);
515 float maxy
= MAX2(dtdx
, dtdy
) * u_minify(texture
->height0
, samp
->view
->u
.tex
.first_level
);
516 float rho
= MAX2(maxx
, maxy
);
518 return util_fast_log2(rho
);
523 compute_lambda_3d(const struct sp_sampler_variant
*samp
,
524 const float s
[TGSI_QUAD_SIZE
],
525 const float t
[TGSI_QUAD_SIZE
],
526 const float p
[TGSI_QUAD_SIZE
])
528 const struct pipe_resource
*texture
= samp
->view
->texture
;
529 float dsdx
= fabsf(s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]);
530 float dsdy
= fabsf(s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]);
531 float dtdx
= fabsf(t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]);
532 float dtdy
= fabsf(t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]);
533 float dpdx
= fabsf(p
[QUAD_BOTTOM_RIGHT
] - p
[QUAD_BOTTOM_LEFT
]);
534 float dpdy
= fabsf(p
[QUAD_TOP_LEFT
] - p
[QUAD_BOTTOM_LEFT
]);
535 float maxx
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, samp
->view
->u
.tex
.first_level
);
536 float maxy
= MAX2(dtdx
, dtdy
) * u_minify(texture
->height0
, samp
->view
->u
.tex
.first_level
);
537 float maxz
= MAX2(dpdx
, dpdy
) * u_minify(texture
->depth0
, samp
->view
->u
.tex
.first_level
);
540 rho
= MAX2(maxx
, maxy
);
541 rho
= MAX2(rho
, maxz
);
543 return util_fast_log2(rho
);
548 * Compute lambda for a vertex texture sampler.
549 * Since there aren't derivatives to use, just return 0.
552 compute_lambda_vert(const struct sp_sampler_variant
*samp
,
553 const float s
[TGSI_QUAD_SIZE
],
554 const float t
[TGSI_QUAD_SIZE
],
555 const float p
[TGSI_QUAD_SIZE
])
563 * Get a texel from a texture, using the texture tile cache.
565 * \param addr the template tex address containing cube, z, face info.
566 * \param x the x coord of texel within 2D image
567 * \param y the y coord of texel within 2D image
568 * \param rgba the quad to put the texel/color into
570 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
571 * sp_get_cached_tile_tex() function.
577 static INLINE
const float *
578 get_texel_2d_no_border(const struct sp_sampler_variant
*samp
,
579 union tex_tile_address addr
, int x
, int y
)
581 const struct softpipe_tex_cached_tile
*tile
;
583 addr
.bits
.x
= x
/ TILE_SIZE
;
584 addr
.bits
.y
= y
/ TILE_SIZE
;
588 tile
= sp_get_cached_tile_tex(samp
->cache
, addr
);
590 return &tile
->data
.color
[y
][x
][0];
594 static INLINE
const float *
595 get_texel_2d(const struct sp_sampler_variant
*samp
,
596 union tex_tile_address addr
, int x
, int y
)
598 const struct pipe_resource
*texture
= samp
->view
->texture
;
599 unsigned level
= addr
.bits
.level
;
601 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
602 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
603 return samp
->sampler
->border_color
.f
;
606 return get_texel_2d_no_border( samp
, addr
, x
, y
);
611 * seamless cubemap neighbour array.
612 * this array is used to find the adjacent face in each of 4 directions,
613 * left, right, up, down. (or -x, +x, -y, +y).
615 static const unsigned face_array
[PIPE_TEX_FACE_MAX
][4] = {
616 /* pos X first then neg X is Z different, Y the same */
617 /* PIPE_TEX_FACE_POS_X,*/
618 { PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
,
619 PIPE_TEX_FACE_NEG_Y
, PIPE_TEX_FACE_POS_Y
},
620 /* PIPE_TEX_FACE_NEG_X */
621 { PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
,
622 PIPE_TEX_FACE_NEG_Y
, PIPE_TEX_FACE_POS_Y
},
624 /* pos Y first then neg Y is X different, X the same */
625 /* PIPE_TEX_FACE_POS_Y */
626 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
627 PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
},
629 /* PIPE_TEX_FACE_NEG_Y */
630 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
631 PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
},
633 /* pos Z first then neg Y is X different, X the same */
634 /* PIPE_TEX_FACE_POS_Z */
635 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
636 PIPE_TEX_FACE_NEG_Y
, PIPE_TEX_FACE_POS_Y
},
638 /* PIPE_TEX_FACE_NEG_Z */
639 { PIPE_TEX_FACE_POS_X
, PIPE_TEX_FACE_NEG_X
,
640 PIPE_TEX_FACE_NEG_Y
, PIPE_TEX_FACE_POS_Y
}
643 static INLINE
unsigned
644 get_next_face(unsigned face
, int x
, int y
)
648 if (x
== 0 && y
== 0)
659 return face_array
[face
][idx
];
662 static INLINE
const float *
663 get_texel_cube_seamless(const struct sp_sampler_variant
*samp
,
664 union tex_tile_address addr
, int x
, int y
,
667 const struct pipe_resource
*texture
= samp
->view
->texture
;
668 unsigned level
= addr
.bits
.level
;
669 unsigned face
= addr
.bits
.face
;
674 max_x
= (int) u_minify(texture
->width0
, level
);
675 max_y
= (int) u_minify(texture
->height0
, level
);
679 /* the corner case */
680 if ((x
< 0 || x
>= max_x
) &&
681 (y
< 0 || y
>= max_y
)) {
682 const float *c1
, *c2
, *c3
;
683 int fx
= x
< 0 ? 0 : max_x
- 1;
684 int fy
= y
< 0 ? 0 : max_y
- 1;
685 c1
= get_texel_2d_no_border( samp
, addr
, fx
, fy
);
686 addr
.bits
.face
= get_next_face(face
, (x
< 0) ? -1 : 1, 0);
687 c2
= get_texel_2d_no_border( samp
, addr
, (x
< 0) ? max_x
- 1 : 0, fy
);
688 addr
.bits
.face
= get_next_face(face
, 0, (y
< 0) ? -1 : 1);
689 c3
= get_texel_2d_no_border( samp
, addr
, fx
, (y
< 0) ? max_y
- 1 : 0);
690 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
691 corner
[c
] = CLAMP((c1
[c
] + c2
[c
] + c3
[c
]), 0.0F
, 1.0F
) / 3;
695 /* change the face */
698 face
= get_next_face(face
, -1, 0);
699 } else if (x
>= max_x
) {
701 face
= get_next_face(face
, 1, 0);
704 face
= get_next_face(face
, 0, -1);
705 } else if (y
>= max_y
) {
707 face
= get_next_face(face
, 0, 1);
710 addr
.bits
.face
= face
;
711 return get_texel_2d_no_border( samp
, addr
, new_x
, new_y
);
714 /* Gather a quad of adjacent texels within a tile:
717 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_variant
*samp
,
718 union tex_tile_address addr
,
719 unsigned x
, unsigned y
,
722 const struct softpipe_tex_cached_tile
*tile
;
724 addr
.bits
.x
= x
/ TILE_SIZE
;
725 addr
.bits
.y
= y
/ TILE_SIZE
;
729 tile
= sp_get_cached_tile_tex(samp
->cache
, addr
);
731 out
[0] = &tile
->data
.color
[y
][x
][0];
732 out
[1] = &tile
->data
.color
[y
][x
+1][0];
733 out
[2] = &tile
->data
.color
[y
+1][x
][0];
734 out
[3] = &tile
->data
.color
[y
+1][x
+1][0];
738 /* Gather a quad of potentially non-adjacent texels:
741 get_texel_quad_2d_no_border(const struct sp_sampler_variant
*samp
,
742 union tex_tile_address addr
,
747 out
[0] = get_texel_2d_no_border( samp
, addr
, x0
, y0
);
748 out
[1] = get_texel_2d_no_border( samp
, addr
, x1
, y0
);
749 out
[2] = get_texel_2d_no_border( samp
, addr
, x0
, y1
);
750 out
[3] = get_texel_2d_no_border( samp
, addr
, x1
, y1
);
753 /* Can involve a lot of unnecessary checks for border color:
756 get_texel_quad_2d(const struct sp_sampler_variant
*samp
,
757 union tex_tile_address addr
,
762 out
[0] = get_texel_2d( samp
, addr
, x0
, y0
);
763 out
[1] = get_texel_2d( samp
, addr
, x1
, y0
);
764 out
[3] = get_texel_2d( samp
, addr
, x1
, y1
);
765 out
[2] = get_texel_2d( samp
, addr
, x0
, y1
);
772 static INLINE
const float *
773 get_texel_3d_no_border(const struct sp_sampler_variant
*samp
,
774 union tex_tile_address addr
, int x
, int y
, int z
)
776 const struct softpipe_tex_cached_tile
*tile
;
778 addr
.bits
.x
= x
/ TILE_SIZE
;
779 addr
.bits
.y
= y
/ TILE_SIZE
;
784 tile
= sp_get_cached_tile_tex(samp
->cache
, addr
);
786 return &tile
->data
.color
[y
][x
][0];
790 static INLINE
const float *
791 get_texel_3d(const struct sp_sampler_variant
*samp
,
792 union tex_tile_address addr
, int x
, int y
, int z
)
794 const struct pipe_resource
*texture
= samp
->view
->texture
;
795 unsigned level
= addr
.bits
.level
;
797 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
798 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
) ||
799 z
< 0 || z
>= (int) u_minify(texture
->depth0
, level
)) {
800 return samp
->sampler
->border_color
.f
;
803 return get_texel_3d_no_border( samp
, addr
, x
, y
, z
);
808 /* Get texel pointer for 1D array texture */
809 static INLINE
const float *
810 get_texel_1d_array(const struct sp_sampler_variant
*samp
,
811 union tex_tile_address addr
, int x
, int y
)
813 const struct pipe_resource
*texture
= samp
->view
->texture
;
814 unsigned level
= addr
.bits
.level
;
816 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
)) {
817 return samp
->sampler
->border_color
.f
;
820 return get_texel_2d_no_border(samp
, addr
, x
, y
);
825 /* Get texel pointer for 2D array texture */
826 static INLINE
const float *
827 get_texel_2d_array(const struct sp_sampler_variant
*samp
,
828 union tex_tile_address addr
, int x
, int y
, int layer
)
830 const struct pipe_resource
*texture
= samp
->view
->texture
;
831 unsigned level
= addr
.bits
.level
;
833 assert(layer
< (int) texture
->array_size
);
836 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
837 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
838 return samp
->sampler
->border_color
.f
;
841 return get_texel_3d_no_border(samp
, addr
, x
, y
, layer
);
846 /* Get texel pointer for cube array texture */
847 static INLINE
const float *
848 get_texel_cube_array(const struct sp_sampler_variant
*samp
,
849 union tex_tile_address addr
, int x
, int y
, int layer
)
851 const struct pipe_resource
*texture
= samp
->view
->texture
;
852 unsigned level
= addr
.bits
.level
;
854 assert(layer
< (int) texture
->array_size
);
857 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
858 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
859 return samp
->sampler
->border_color
.f
;
862 return get_texel_3d_no_border(samp
, addr
, x
, y
, layer
);
866 * Given the logbase2 of a mipmap's base level size and a mipmap level,
867 * return the size (in texels) of that mipmap level.
868 * For example, if level[0].width = 256 then base_pot will be 8.
869 * If level = 2, then we'll return 64 (the width at level=2).
870 * Return 1 if level > base_pot.
872 static INLINE
unsigned
873 pot_level_size(unsigned base_pot
, unsigned level
)
875 return (base_pot
>= level
) ? (1 << (base_pot
- level
)) : 1;
880 print_sample(const char *function
, const float *rgba
)
882 debug_printf("%s %g %g %g %g\n",
884 rgba
[0], rgba
[TGSI_NUM_CHANNELS
], rgba
[2*TGSI_NUM_CHANNELS
], rgba
[3*TGSI_NUM_CHANNELS
]);
889 print_sample_4(const char *function
, float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
891 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
893 rgba
[0][0], rgba
[1][0], rgba
[2][0], rgba
[3][0],
894 rgba
[0][1], rgba
[1][1], rgba
[2][1], rgba
[3][1],
895 rgba
[0][2], rgba
[1][2], rgba
[2][2], rgba
[3][2],
896 rgba
[0][3], rgba
[1][3], rgba
[2][3], rgba
[3][3]);
899 /* Some image-filter fastpaths:
902 img_filter_2d_linear_repeat_POT(struct tgsi_sampler
*tgsi_sampler
,
908 enum tgsi_sampler_control control
,
911 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
912 unsigned xpot
= pot_level_size(samp
->xpot
, level
);
913 unsigned ypot
= pot_level_size(samp
->ypot
, level
);
914 unsigned xmax
= (xpot
- 1) & (TILE_SIZE
- 1); /* MIN2(TILE_SIZE, xpot) - 1; */
915 unsigned ymax
= (ypot
- 1) & (TILE_SIZE
- 1); /* MIN2(TILE_SIZE, ypot) - 1; */
916 union tex_tile_address addr
;
919 float u
= s
* xpot
- 0.5F
;
920 float v
= t
* ypot
- 0.5F
;
922 int uflr
= util_ifloor(u
);
923 int vflr
= util_ifloor(v
);
925 float xw
= u
- (float)uflr
;
926 float yw
= v
- (float)vflr
;
928 int x0
= uflr
& (xpot
- 1);
929 int y0
= vflr
& (ypot
- 1);
934 addr
.bits
.level
= level
;
936 /* Can we fetch all four at once:
938 if (x0
< xmax
&& y0
< ymax
) {
939 get_texel_quad_2d_no_border_single_tile(samp
, addr
, x0
, y0
, tx
);
942 unsigned x1
= (x0
+ 1) & (xpot
- 1);
943 unsigned y1
= (y0
+ 1) & (ypot
- 1);
944 get_texel_quad_2d_no_border(samp
, addr
, x0
, y0
, x1
, y1
, tx
);
947 /* interpolate R, G, B, A */
948 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++) {
949 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
955 print_sample(__FUNCTION__
, rgba
);
961 img_filter_2d_nearest_repeat_POT(struct tgsi_sampler
*tgsi_sampler
,
967 enum tgsi_sampler_control control
,
968 float rgba
[TGSI_QUAD_SIZE
])
970 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
971 unsigned xpot
= pot_level_size(samp
->xpot
, level
);
972 unsigned ypot
= pot_level_size(samp
->ypot
, level
);
974 union tex_tile_address addr
;
980 int uflr
= util_ifloor(u
);
981 int vflr
= util_ifloor(v
);
983 int x0
= uflr
& (xpot
- 1);
984 int y0
= vflr
& (ypot
- 1);
987 addr
.bits
.level
= level
;
989 out
= get_texel_2d_no_border(samp
, addr
, x0
, y0
);
990 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
991 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
994 print_sample(__FUNCTION__
, rgba
);
1000 img_filter_2d_nearest_clamp_POT(struct tgsi_sampler
*tgsi_sampler
,
1006 enum tgsi_sampler_control control
,
1007 float rgba
[TGSI_QUAD_SIZE
])
1009 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1010 unsigned xpot
= pot_level_size(samp
->xpot
, level
);
1011 unsigned ypot
= pot_level_size(samp
->ypot
, level
);
1012 union tex_tile_address addr
;
1022 addr
.bits
.level
= level
;
1024 x0
= util_ifloor(u
);
1027 else if (x0
> xpot
- 1)
1030 y0
= util_ifloor(v
);
1033 else if (y0
> ypot
- 1)
1036 out
= get_texel_2d_no_border(samp
, addr
, x0
, y0
);
1037 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1038 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1041 print_sample(__FUNCTION__
, rgba
);
1047 img_filter_1d_nearest(struct tgsi_sampler
*tgsi_sampler
,
1053 enum tgsi_sampler_control control
,
1054 float rgba
[TGSI_QUAD_SIZE
])
1056 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1057 const struct pipe_resource
*texture
= samp
->view
->texture
;
1060 union tex_tile_address addr
;
1064 width
= u_minify(texture
->width0
, level
);
1069 addr
.bits
.level
= level
;
1071 samp
->nearest_texcoord_s(s
, width
, &x
);
1073 out
= get_texel_2d(samp
, addr
, x
, 0);
1074 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1075 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1078 print_sample(__FUNCTION__
, rgba
);
1084 img_filter_1d_array_nearest(struct tgsi_sampler
*tgsi_sampler
,
1090 enum tgsi_sampler_control control
,
1093 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1094 const struct pipe_resource
*texture
= samp
->view
->texture
;
1097 union tex_tile_address addr
;
1101 width
= u_minify(texture
->width0
, level
);
1106 addr
.bits
.level
= level
;
1108 samp
->nearest_texcoord_s(s
, width
, &x
);
1109 wrap_array_layer(t
, texture
->array_size
, &layer
);
1111 out
= get_texel_1d_array(samp
, addr
, x
, layer
);
1112 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1113 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1116 print_sample(__FUNCTION__
, rgba
);
1122 img_filter_2d_nearest(struct tgsi_sampler
*tgsi_sampler
,
1128 enum tgsi_sampler_control control
,
1131 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1132 const struct pipe_resource
*texture
= samp
->view
->texture
;
1135 union tex_tile_address addr
;
1139 width
= u_minify(texture
->width0
, level
);
1140 height
= u_minify(texture
->height0
, level
);
1146 addr
.bits
.level
= level
;
1148 samp
->nearest_texcoord_s(s
, width
, &x
);
1149 samp
->nearest_texcoord_t(t
, height
, &y
);
1151 out
= get_texel_2d(samp
, addr
, x
, y
);
1152 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1153 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1156 print_sample(__FUNCTION__
, rgba
);
1162 img_filter_2d_array_nearest(struct tgsi_sampler
*tgsi_sampler
,
1168 enum tgsi_sampler_control control
,
1171 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1172 const struct pipe_resource
*texture
= samp
->view
->texture
;
1175 union tex_tile_address addr
;
1179 width
= u_minify(texture
->width0
, level
);
1180 height
= u_minify(texture
->height0
, level
);
1186 addr
.bits
.level
= level
;
1188 samp
->nearest_texcoord_s(s
, width
, &x
);
1189 samp
->nearest_texcoord_t(t
, height
, &y
);
1190 wrap_array_layer(p
, texture
->array_size
, &layer
);
1192 out
= get_texel_2d_array(samp
, addr
, x
, y
, layer
);
1193 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1194 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1197 print_sample(__FUNCTION__
, rgba
);
1202 static INLINE
union tex_tile_address
1203 face(union tex_tile_address addr
, unsigned face
)
1205 addr
.bits
.face
= face
;
1211 img_filter_cube_nearest(struct tgsi_sampler
*tgsi_sampler
,
1217 enum tgsi_sampler_control control
,
1220 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1221 const struct pipe_resource
*texture
= samp
->view
->texture
;
1224 union tex_tile_address addr
;
1227 float corner0
[TGSI_QUAD_SIZE
];
1229 width
= u_minify(texture
->width0
, level
);
1230 height
= u_minify(texture
->height0
, level
);
1236 addr
.bits
.level
= level
;
1239 * If NEAREST filtering is done within a miplevel, always apply wrap
1240 * mode CLAMP_TO_EDGE.
1242 if (samp
->sampler
->seamless_cube_map
) {
1243 wrap_nearest_clamp_to_edge(s
, width
, &x
);
1244 wrap_nearest_clamp_to_edge(t
, height
, &y
);
1246 samp
->nearest_texcoord_s(s
, width
, &x
);
1247 samp
->nearest_texcoord_t(t
, height
, &y
);
1250 out
= get_texel_2d(samp
, face(addr
, face_id
), x
, y
);
1251 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1252 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1255 print_sample(__FUNCTION__
, rgba
);
1260 img_filter_cube_array_nearest(struct tgsi_sampler
*tgsi_sampler
,
1266 enum tgsi_sampler_control control
,
1269 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1270 const struct pipe_resource
*texture
= samp
->view
->texture
;
1273 union tex_tile_address addr
;
1277 width
= u_minify(texture
->width0
, level
);
1278 height
= u_minify(texture
->height0
, level
);
1284 addr
.bits
.level
= level
;
1286 samp
->nearest_texcoord_s(s
, width
, &x
);
1287 samp
->nearest_texcoord_t(t
, height
, &y
);
1288 wrap_array_layer(p
, texture
->array_size
, &layer
);
1290 out
= get_texel_cube_array(samp
, addr
, x
, y
, layer
* 6 + face_id
);
1291 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1292 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1295 print_sample(__FUNCTION__
, rgba
);
1300 img_filter_3d_nearest(struct tgsi_sampler
*tgsi_sampler
,
1306 enum tgsi_sampler_control control
,
1309 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1310 const struct pipe_resource
*texture
= samp
->view
->texture
;
1311 int width
, height
, depth
;
1313 union tex_tile_address addr
;
1317 width
= u_minify(texture
->width0
, level
);
1318 height
= u_minify(texture
->height0
, level
);
1319 depth
= u_minify(texture
->depth0
, level
);
1325 samp
->nearest_texcoord_s(s
, width
, &x
);
1326 samp
->nearest_texcoord_t(t
, height
, &y
);
1327 samp
->nearest_texcoord_p(p
, depth
, &z
);
1330 addr
.bits
.level
= level
;
1332 out
= get_texel_3d(samp
, addr
, x
, y
, z
);
1333 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1334 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1339 img_filter_1d_linear(struct tgsi_sampler
*tgsi_sampler
,
1345 enum tgsi_sampler_control control
,
1348 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1349 const struct pipe_resource
*texture
= samp
->view
->texture
;
1352 float xw
; /* weights */
1353 union tex_tile_address addr
;
1354 const float *tx0
, *tx1
;
1357 width
= u_minify(texture
->width0
, level
);
1362 addr
.bits
.level
= level
;
1364 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1366 tx0
= get_texel_2d(samp
, addr
, x0
, 0);
1367 tx1
= get_texel_2d(samp
, addr
, x1
, 0);
1369 /* interpolate R, G, B, A */
1370 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1371 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1376 img_filter_1d_array_linear(struct tgsi_sampler
*tgsi_sampler
,
1382 enum tgsi_sampler_control control
,
1385 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1386 const struct pipe_resource
*texture
= samp
->view
->texture
;
1389 float xw
; /* weights */
1390 union tex_tile_address addr
;
1391 const float *tx0
, *tx1
;
1394 width
= u_minify(texture
->width0
, level
);
1399 addr
.bits
.level
= level
;
1401 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1402 wrap_array_layer(t
, texture
->array_size
, &layer
);
1404 tx0
= get_texel_1d_array(samp
, addr
, x0
, layer
);
1405 tx1
= get_texel_1d_array(samp
, addr
, x1
, layer
);
1407 /* interpolate R, G, B, A */
1408 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1409 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1414 img_filter_2d_linear(struct tgsi_sampler
*tgsi_sampler
,
1420 enum tgsi_sampler_control control
,
1423 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1424 const struct pipe_resource
*texture
= samp
->view
->texture
;
1427 float xw
, yw
; /* weights */
1428 union tex_tile_address addr
;
1429 const float *tx0
, *tx1
, *tx2
, *tx3
;
1432 width
= u_minify(texture
->width0
, level
);
1433 height
= u_minify(texture
->height0
, level
);
1439 addr
.bits
.level
= level
;
1441 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1442 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1444 tx0
= get_texel_2d(samp
, addr
, x0
, y0
);
1445 tx1
= get_texel_2d(samp
, addr
, x1
, y0
);
1446 tx2
= get_texel_2d(samp
, addr
, x0
, y1
);
1447 tx3
= get_texel_2d(samp
, addr
, x1
, y1
);
1449 /* interpolate R, G, B, A */
1450 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1451 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1458 img_filter_2d_array_linear(struct tgsi_sampler
*tgsi_sampler
,
1464 enum tgsi_sampler_control control
,
1467 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1468 const struct pipe_resource
*texture
= samp
->view
->texture
;
1470 int x0
, y0
, x1
, y1
, layer
;
1471 float xw
, yw
; /* weights */
1472 union tex_tile_address addr
;
1473 const float *tx0
, *tx1
, *tx2
, *tx3
;
1476 width
= u_minify(texture
->width0
, level
);
1477 height
= u_minify(texture
->height0
, level
);
1483 addr
.bits
.level
= level
;
1485 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1486 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1487 wrap_array_layer(p
, texture
->array_size
, &layer
);
1489 tx0
= get_texel_2d_array(samp
, addr
, x0
, y0
, layer
);
1490 tx1
= get_texel_2d_array(samp
, addr
, x1
, y0
, layer
);
1491 tx2
= get_texel_2d_array(samp
, addr
, x0
, y1
, layer
);
1492 tx3
= get_texel_2d_array(samp
, addr
, x1
, y1
, layer
);
1494 /* interpolate R, G, B, A */
1495 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1496 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1503 img_filter_cube_linear(struct tgsi_sampler
*tgsi_sampler
,
1509 enum tgsi_sampler_control control
,
1512 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1513 const struct pipe_resource
*texture
= samp
->view
->texture
;
1516 float xw
, yw
; /* weights */
1517 union tex_tile_address addr
, addrj
;
1518 const float *tx0
, *tx1
, *tx2
, *tx3
;
1519 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
], corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1522 width
= u_minify(texture
->width0
, level
);
1523 height
= u_minify(texture
->height0
, level
);
1529 addr
.bits
.level
= level
;
1532 * For seamless if LINEAR filtering is done within a miplevel,
1533 * always apply wrap mode CLAMP_TO_BORDER.
1535 if (samp
->sampler
->seamless_cube_map
) {
1536 wrap_linear_clamp_to_border(s
, width
, &x0
, &x1
, &xw
);
1537 wrap_linear_clamp_to_border(t
, height
, &y0
, &y1
, &yw
);
1539 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1540 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1543 addrj
= face(addr
, face_id
);
1545 if (samp
->sampler
->seamless_cube_map
) {
1546 tx0
= get_texel_cube_seamless(samp
, addrj
, x0
, y0
, corner0
);
1547 tx1
= get_texel_cube_seamless(samp
, addrj
, x1
, y0
, corner1
);
1548 tx2
= get_texel_cube_seamless(samp
, addrj
, x0
, y1
, corner2
);
1549 tx3
= get_texel_cube_seamless(samp
, addrj
, x1
, y1
, corner3
);
1551 tx0
= get_texel_2d(samp
, addrj
, x0
, y0
);
1552 tx1
= get_texel_2d(samp
, addrj
, x1
, y0
);
1553 tx2
= get_texel_2d(samp
, addrj
, x0
, y1
);
1554 tx3
= get_texel_2d(samp
, addrj
, x1
, y1
);
1556 /* interpolate R, G, B, A */
1557 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1558 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1565 img_filter_cube_array_linear(struct tgsi_sampler
*tgsi_sampler
,
1571 enum tgsi_sampler_control control
,
1574 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1575 const struct pipe_resource
*texture
= samp
->view
->texture
;
1577 int x0
, y0
, x1
, y1
, layer
;
1578 float xw
, yw
; /* weights */
1579 union tex_tile_address addr
;
1580 const float *tx0
, *tx1
, *tx2
, *tx3
;
1583 width
= u_minify(texture
->width0
, level
);
1584 height
= u_minify(texture
->height0
, level
);
1590 addr
.bits
.level
= level
;
1592 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1593 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1594 wrap_array_layer(p
, texture
->array_size
, &layer
);
1596 tx0
= get_texel_cube_array(samp
, addr
, x0
, y0
, layer
* 6 + face_id
);
1597 tx1
= get_texel_cube_array(samp
, addr
, x1
, y0
, layer
* 6 + face_id
);
1598 tx2
= get_texel_cube_array(samp
, addr
, x0
, y1
, layer
* 6 + face_id
);
1599 tx3
= get_texel_cube_array(samp
, addr
, x1
, y1
, layer
* 6 + face_id
);
1601 /* interpolate R, G, B, A */
1602 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1603 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1609 img_filter_3d_linear(struct tgsi_sampler
*tgsi_sampler
,
1615 enum tgsi_sampler_control control
,
1618 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1619 const struct pipe_resource
*texture
= samp
->view
->texture
;
1620 int width
, height
, depth
;
1621 int x0
, x1
, y0
, y1
, z0
, z1
;
1622 float xw
, yw
, zw
; /* interpolation weights */
1623 union tex_tile_address addr
;
1624 const float *tx00
, *tx01
, *tx02
, *tx03
, *tx10
, *tx11
, *tx12
, *tx13
;
1627 width
= u_minify(texture
->width0
, level
);
1628 height
= u_minify(texture
->height0
, level
);
1629 depth
= u_minify(texture
->depth0
, level
);
1632 addr
.bits
.level
= level
;
1638 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1639 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1640 samp
->linear_texcoord_p(p
, depth
, &z0
, &z1
, &zw
);
1643 tx00
= get_texel_3d(samp
, addr
, x0
, y0
, z0
);
1644 tx01
= get_texel_3d(samp
, addr
, x1
, y0
, z0
);
1645 tx02
= get_texel_3d(samp
, addr
, x0
, y1
, z0
);
1646 tx03
= get_texel_3d(samp
, addr
, x1
, y1
, z0
);
1648 tx10
= get_texel_3d(samp
, addr
, x0
, y0
, z1
);
1649 tx11
= get_texel_3d(samp
, addr
, x1
, y0
, z1
);
1650 tx12
= get_texel_3d(samp
, addr
, x0
, y1
, z1
);
1651 tx13
= get_texel_3d(samp
, addr
, x1
, y1
, z1
);
1653 /* interpolate R, G, B, A */
1654 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1655 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_3d(xw
, yw
, zw
,
1663 /* Calculate level of detail for every fragment.
1664 * Note that lambda has already been biased by global LOD bias.
1667 compute_lod(const struct pipe_sampler_state
*sampler
,
1668 const float biased_lambda
,
1669 const float lodbias
[TGSI_QUAD_SIZE
],
1670 float lod
[TGSI_QUAD_SIZE
])
1674 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1675 lod
[i
] = biased_lambda
+ lodbias
[i
];
1676 lod
[i
] = CLAMP(lod
[i
], sampler
->min_lod
, sampler
->max_lod
);
1682 mip_filter_linear(struct tgsi_sampler
*tgsi_sampler
,
1683 const float s
[TGSI_QUAD_SIZE
],
1684 const float t
[TGSI_QUAD_SIZE
],
1685 const float p
[TGSI_QUAD_SIZE
],
1686 const float c0
[TGSI_QUAD_SIZE
],
1687 const float c1
[TGSI_QUAD_SIZE
],
1688 enum tgsi_sampler_control control
,
1689 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1691 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1692 const struct pipe_resource
*texture
= samp
->view
->texture
;
1694 float lod
[TGSI_QUAD_SIZE
];
1696 if (control
== tgsi_sampler_lod_bias
) {
1697 float lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
1698 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1699 compute_lod(samp
->sampler
, lambda
, c1
, lod
);
1701 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
1703 assert(control
== tgsi_sampler_lod_explicit
);
1705 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1706 memcpy(lod
, c1
, sizeof(lod
));
1708 memcpy(lod
, c0
, sizeof(lod
));
1712 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1713 int level0
= samp
->view
->u
.tex
.first_level
+ (int)lod
[j
];
1716 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
1718 else if (level0
>= texture
->last_level
)
1719 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], texture
->last_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
1722 float levelBlend
= frac(lod
[j
]);
1723 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1726 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgbax
[0][0]);
1727 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
+1, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgbax
[0][1]);
1729 for (c
= 0; c
< 4; c
++) {
1730 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
1736 print_sample_4(__FUNCTION__
, rgba
);
1742 * Compute nearest mipmap level from texcoords.
1743 * Then sample the texture level for four elements of a quad.
1744 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1747 mip_filter_nearest(struct tgsi_sampler
*tgsi_sampler
,
1748 const float s
[TGSI_QUAD_SIZE
],
1749 const float t
[TGSI_QUAD_SIZE
],
1750 const float p
[TGSI_QUAD_SIZE
],
1751 const float c0
[TGSI_QUAD_SIZE
],
1752 const float c1
[TGSI_QUAD_SIZE
],
1753 enum tgsi_sampler_control control
,
1754 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1756 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1757 const struct pipe_resource
*texture
= samp
->view
->texture
;
1758 float lod
[TGSI_QUAD_SIZE
];
1761 if (control
== tgsi_sampler_lod_bias
) {
1762 float lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
1763 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1764 compute_lod(samp
->sampler
, lambda
, c1
, lod
);
1766 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
1768 assert(control
== tgsi_sampler_lod_explicit
);
1770 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1771 memcpy(lod
, c1
, sizeof(lod
));
1773 memcpy(lod
, c0
, sizeof(lod
));
1776 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1778 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
1780 float level
= samp
->view
->u
.tex
.first_level
+ (int)(lod
[j
] + 0.5F
) ;
1781 level
= MIN2(level
, (int)texture
->last_level
);
1782 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
1787 print_sample_4(__FUNCTION__
, rgba
);
1793 mip_filter_none(struct tgsi_sampler
*tgsi_sampler
,
1794 const float s
[TGSI_QUAD_SIZE
],
1795 const float t
[TGSI_QUAD_SIZE
],
1796 const float p
[TGSI_QUAD_SIZE
],
1797 const float c0
[TGSI_QUAD_SIZE
],
1798 const float c1
[TGSI_QUAD_SIZE
],
1799 enum tgsi_sampler_control control
,
1800 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1802 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1803 float lod
[TGSI_QUAD_SIZE
];
1806 if (control
== tgsi_sampler_lod_bias
) {
1807 float lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
1808 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1809 compute_lod(samp
->sampler
, lambda
, c1
, lod
);
1811 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
1813 assert(control
== tgsi_sampler_lod_explicit
);
1815 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1816 memcpy(lod
, c1
, sizeof(lod
));
1818 memcpy(lod
, c0
, sizeof(lod
));
1821 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1823 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
1826 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
1833 mip_filter_none_no_filter_select(struct tgsi_sampler
*tgsi_sampler
,
1834 const float s
[TGSI_QUAD_SIZE
],
1835 const float t
[TGSI_QUAD_SIZE
],
1836 const float p
[TGSI_QUAD_SIZE
],
1837 const float c0
[TGSI_QUAD_SIZE
],
1838 const float c1
[TGSI_QUAD_SIZE
],
1839 enum tgsi_sampler_control control
,
1840 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1842 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1845 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++)
1846 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
1850 /* For anisotropic filtering */
1851 #define WEIGHT_LUT_SIZE 1024
1853 static float *weightLut
= NULL
;
1856 * Creates the look-up table used to speed-up EWA sampling
1859 create_filter_table(void)
1863 weightLut
= (float *) MALLOC(WEIGHT_LUT_SIZE
* sizeof(float));
1865 for (i
= 0; i
< WEIGHT_LUT_SIZE
; ++i
) {
1867 float r2
= (float) i
/ (float) (WEIGHT_LUT_SIZE
- 1);
1868 float weight
= (float) exp(-alpha
* r2
);
1869 weightLut
[i
] = weight
;
1876 * Elliptical weighted average (EWA) filter for producing high quality
1877 * anisotropic filtered results.
1878 * Based on the Higher Quality Elliptical Weighted Average Filter
1879 * published by Paul S. Heckbert in his Master's Thesis
1880 * "Fundamentals of Texture Mapping and Image Warping" (1989)
1883 img_filter_2d_ewa(struct tgsi_sampler
*tgsi_sampler
,
1884 const float s
[TGSI_QUAD_SIZE
],
1885 const float t
[TGSI_QUAD_SIZE
],
1886 const float p
[TGSI_QUAD_SIZE
],
1888 enum tgsi_sampler_control control
,
1889 const float dudx
, const float dvdx
,
1890 const float dudy
, const float dvdy
,
1891 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1893 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1894 const struct pipe_resource
*texture
= samp
->view
->texture
;
1896 // ??? Won't the image filters blow up if level is negative?
1897 unsigned level0
= level
> 0 ? level
: 0;
1898 float scaling
= 1.0 / (1 << level0
);
1899 int width
= u_minify(texture
->width0
, level0
);
1900 int height
= u_minify(texture
->height0
, level0
);
1902 float ux
= dudx
* scaling
;
1903 float vx
= dvdx
* scaling
;
1904 float uy
= dudy
* scaling
;
1905 float vy
= dvdy
* scaling
;
1907 /* compute ellipse coefficients to bound the region:
1908 * A*x*x + B*x*y + C*y*y = F.
1910 float A
= vx
*vx
+vy
*vy
+1;
1911 float B
= -2*(ux
*vx
+uy
*vy
);
1912 float C
= ux
*ux
+uy
*uy
+1;
1913 float F
= A
*C
-B
*B
/4.0;
1915 /* check if it is an ellipse */
1916 /* ASSERT(F > 0.0); */
1918 /* Compute the ellipse's (u,v) bounding box in texture space */
1919 float d
= -B
*B
+4.0*C
*A
;
1920 float box_u
= 2.0 / d
* sqrt(d
*C
*F
); /* box_u -> half of bbox with */
1921 float box_v
= 2.0 / d
* sqrt(A
*d
*F
); /* box_v -> half of bbox height */
1923 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1924 float s_buffer
[TGSI_QUAD_SIZE
];
1925 float t_buffer
[TGSI_QUAD_SIZE
];
1926 float weight_buffer
[TGSI_QUAD_SIZE
];
1927 unsigned buffer_next
;
1929 float den
; /* = 0.0F; */
1931 float U
; /* = u0 - tex_u; */
1934 /* Scale ellipse formula to directly index the Filter Lookup Table.
1935 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
1937 double formScale
= (double) (WEIGHT_LUT_SIZE
- 1) / F
;
1941 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
1943 /* For each quad, the du and dx values are the same and so the ellipse is
1944 * also the same. Note that texel/image access can only be performed using
1945 * a quad, i.e. it is not possible to get the pixel value for a single
1946 * tex coord. In order to have a better performance, the access is buffered
1947 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
1948 * full, then the pixel values are read from the image.
1952 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1953 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
1954 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
1955 * value, q, is less than F, we're inside the ellipse
1957 float tex_u
= -0.5F
+ s
[j
] * texture
->width0
* scaling
;
1958 float tex_v
= -0.5F
+ t
[j
] * texture
->height0
* scaling
;
1960 int u0
= (int) floorf(tex_u
- box_u
);
1961 int u1
= (int) ceilf(tex_u
+ box_u
);
1962 int v0
= (int) floorf(tex_v
- box_v
);
1963 int v1
= (int) ceilf(tex_v
+ box_v
);
1965 float num
[4] = {0.0F
, 0.0F
, 0.0F
, 0.0F
};
1969 for (v
= v0
; v
<= v1
; ++v
) {
1970 float V
= v
- tex_v
;
1971 float dq
= A
* (2 * U
+ 1) + B
* V
;
1972 float q
= (C
* V
+ B
* U
) * V
+ A
* U
* U
;
1975 for (u
= u0
; u
<= u1
; ++u
) {
1976 /* Note that the ellipse has been pre-scaled so F =
1977 * WEIGHT_LUT_SIZE - 1
1979 if (q
< WEIGHT_LUT_SIZE
) {
1980 /* as a LUT is used, q must never be negative;
1981 * should not happen, though
1983 const int qClamped
= q
>= 0.0F
? q
: 0;
1984 float weight
= weightLut
[qClamped
];
1986 weight_buffer
[buffer_next
] = weight
;
1987 s_buffer
[buffer_next
] = u
/ ((float) width
);
1988 t_buffer
[buffer_next
] = v
/ ((float) height
);
1991 if (buffer_next
== TGSI_QUAD_SIZE
) {
1992 /* 4 texel coords are in the buffer -> read it now */
1994 /* it is assumed that samp->min_img_filter is set to
1995 * img_filter_2d_nearest or one of the
1996 * accelerated img_filter_2d_nearest_XXX functions.
1998 for (jj
= 0; jj
< buffer_next
; jj
++) {
1999 samp
->min_img_filter(tgsi_sampler
, s_buffer
[jj
], t_buffer
[jj
], p
[jj
], level
, samp
->faces
[j
],
2000 tgsi_sampler_lod_bias
, &rgba_temp
[0][jj
]);
2001 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2002 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2003 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2004 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2017 /* if the tex coord buffer contains unread values, we will read
2020 if (buffer_next
> 0) {
2022 /* it is assumed that samp->min_img_filter is set to
2023 * img_filter_2d_nearest or one of the
2024 * accelerated img_filter_2d_nearest_XXX functions.
2026 for (jj
= 0; jj
< buffer_next
; jj
++) {
2027 samp
->min_img_filter(tgsi_sampler
, s_buffer
[jj
], t_buffer
[jj
], p
[jj
], level
, samp
->faces
[j
],
2028 tgsi_sampler_lod_bias
, &rgba_temp
[0][jj
]);
2029 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2030 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2031 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2032 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2037 /* Reaching this place would mean that no pixels intersected
2038 * the ellipse. This should never happen because the filter
2039 * we use always intersects at least one pixel.
2046 /* not enough pixels in resampling, resort to direct interpolation */
2047 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level
, samp
->faces
[j
],
2048 tgsi_sampler_lod_bias
, &rgba_temp
[0][j
]);
2050 num
[0] = rgba_temp
[0][j
];
2051 num
[1] = rgba_temp
[1][j
];
2052 num
[2] = rgba_temp
[2][j
];
2053 num
[3] = rgba_temp
[3][j
];
2056 rgba
[0][j
] = num
[0] / den
;
2057 rgba
[1][j
] = num
[1] / den
;
2058 rgba
[2][j
] = num
[2] / den
;
2059 rgba
[3][j
] = num
[3] / den
;
2065 * Sample 2D texture using an anisotropic filter.
2068 mip_filter_linear_aniso(struct tgsi_sampler
*tgsi_sampler
,
2069 const float s
[TGSI_QUAD_SIZE
],
2070 const float t
[TGSI_QUAD_SIZE
],
2071 const float p
[TGSI_QUAD_SIZE
],
2072 const float c0
[TGSI_QUAD_SIZE
],
2073 const float c1
[TGSI_QUAD_SIZE
],
2074 enum tgsi_sampler_control control
,
2075 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2077 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2078 const struct pipe_resource
*texture
= samp
->view
->texture
;
2081 float lod
[TGSI_QUAD_SIZE
];
2083 float s_to_u
= u_minify(texture
->width0
, samp
->view
->u
.tex
.first_level
);
2084 float t_to_v
= u_minify(texture
->height0
, samp
->view
->u
.tex
.first_level
);
2085 float dudx
= (s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2086 float dudy
= (s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2087 float dvdx
= (t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2088 float dvdy
= (t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2090 if (control
== tgsi_sampler_lod_bias
) {
2091 /* note: instead of working with Px and Py, we will use the
2092 * squared length instead, to avoid sqrt.
2094 float Px2
= dudx
* dudx
+ dvdx
* dvdx
;
2095 float Py2
= dudy
* dudy
+ dvdy
* dvdy
;
2100 const float maxEccentricity
= samp
->sampler
->max_anisotropy
* samp
->sampler
->max_anisotropy
;
2111 /* if the eccentricity of the ellipse is too big, scale up the shorter
2112 * of the two vectors to limit the maximum amount of work per pixel
2115 if (e
> maxEccentricity
) {
2116 /* float s=e / maxEccentricity;
2120 Pmin2
= Pmax2
/ maxEccentricity
;
2123 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2124 * this since 0.5*log(x) = log(sqrt(x))
2126 lambda
= 0.5F
* util_fast_log2(Pmin2
) + samp
->sampler
->lod_bias
;
2127 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
2130 assert(control
== tgsi_sampler_lod_explicit
);
2132 memcpy(lod
, c0
, sizeof(lod
));
2135 /* XXX: Take into account all lod values.
2138 level0
= samp
->view
->u
.tex
.first_level
+ (int)lambda
;
2140 /* If the ellipse covers the whole image, we can
2141 * simply return the average of the whole image.
2143 if (level0
>= (int) texture
->last_level
) {
2145 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++)
2146 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], texture
->last_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
2149 /* don't bother interpolating between multiple LODs; it doesn't
2150 * seem to be worth the extra running time.
2152 img_filter_2d_ewa(tgsi_sampler
, s
, t
, p
, level0
, tgsi_sampler_lod_bias
,
2153 dudx
, dvdx
, dudy
, dvdy
, rgba
);
2157 print_sample_4(__FUNCTION__
, rgba
);
2163 * Specialized version of mip_filter_linear with hard-wired calls to
2164 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2167 mip_filter_linear_2d_linear_repeat_POT(
2168 struct tgsi_sampler
*tgsi_sampler
,
2169 const float s
[TGSI_QUAD_SIZE
],
2170 const float t
[TGSI_QUAD_SIZE
],
2171 const float p
[TGSI_QUAD_SIZE
],
2172 const float c0
[TGSI_QUAD_SIZE
],
2173 const float c1
[TGSI_QUAD_SIZE
],
2174 enum tgsi_sampler_control control
,
2175 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2177 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2178 const struct pipe_resource
*texture
= samp
->view
->texture
;
2181 float lod
[TGSI_QUAD_SIZE
];
2183 if (control
== tgsi_sampler_lod_bias
) {
2184 lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
2185 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
2187 assert(control
== tgsi_sampler_lod_explicit
);
2189 memcpy(lod
, c0
, sizeof(lod
));
2192 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2193 int level0
= samp
->view
->u
.tex
.first_level
+ (int)lod
[j
];
2195 /* Catches both negative and large values of level0:
2197 if ((unsigned)level0
>= texture
->last_level
) {
2199 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
2201 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->texture
->last_level
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgba
[0][j
]);
2205 float levelBlend
= frac(lod
[j
]);
2206 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2209 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgbax
[0][0]);
2210 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
+1, samp
->faces
[j
], tgsi_sampler_lod_bias
, &rgbax
[0][1]);
2212 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
2213 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2218 print_sample_4(__FUNCTION__
, rgba
);
2224 * Do shadow/depth comparisons.
2227 sample_compare(struct tgsi_sampler
*tgsi_sampler
,
2228 const float s
[TGSI_QUAD_SIZE
],
2229 const float t
[TGSI_QUAD_SIZE
],
2230 const float p
[TGSI_QUAD_SIZE
],
2231 const float c0
[TGSI_QUAD_SIZE
],
2232 const float c1
[TGSI_QUAD_SIZE
],
2233 enum tgsi_sampler_control control
,
2234 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2236 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2237 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
2238 int j
, k0
, k1
, k2
, k3
;
2240 float pc0
, pc1
, pc2
, pc3
;
2242 samp
->mip_filter(tgsi_sampler
, s
, t
, p
, c0
, c1
, control
, rgba
);
2245 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2246 * for 2D Array texture we need to use the 'c0' (aka Q).
2247 * When we sampled the depth texture, the depth value was put into all
2248 * RGBA channels. We look at the red channel here.
2251 if (samp
->view
->texture
->target
== PIPE_TEXTURE_2D_ARRAY
||
2252 samp
->view
->texture
->target
== PIPE_TEXTURE_CUBE
) {
2253 pc0
= CLAMP(c0
[0], 0.0F
, 1.0F
);
2254 pc1
= CLAMP(c0
[1], 0.0F
, 1.0F
);
2255 pc2
= CLAMP(c0
[2], 0.0F
, 1.0F
);
2256 pc3
= CLAMP(c0
[3], 0.0F
, 1.0F
);
2257 } else if (samp
->view
->texture
->target
== PIPE_TEXTURE_CUBE_ARRAY
) {
2258 pc0
= CLAMP(c1
[0], 0.0F
, 1.0F
);
2259 pc1
= CLAMP(c1
[1], 0.0F
, 1.0F
);
2260 pc2
= CLAMP(c1
[2], 0.0F
, 1.0F
);
2261 pc3
= CLAMP(c1
[3], 0.0F
, 1.0F
);
2263 pc0
= CLAMP(p
[0], 0.0F
, 1.0F
);
2264 pc1
= CLAMP(p
[1], 0.0F
, 1.0F
);
2265 pc2
= CLAMP(p
[2], 0.0F
, 1.0F
);
2266 pc3
= CLAMP(p
[3], 0.0F
, 1.0F
);
2268 /* compare four texcoords vs. four texture samples */
2269 switch (sampler
->compare_func
) {
2270 case PIPE_FUNC_LESS
:
2271 k0
= pc0
< rgba
[0][0];
2272 k1
= pc1
< rgba
[0][1];
2273 k2
= pc2
< rgba
[0][2];
2274 k3
= pc3
< rgba
[0][3];
2276 case PIPE_FUNC_LEQUAL
:
2277 k0
= pc0
<= rgba
[0][0];
2278 k1
= pc1
<= rgba
[0][1];
2279 k2
= pc2
<= rgba
[0][2];
2280 k3
= pc3
<= rgba
[0][3];
2282 case PIPE_FUNC_GREATER
:
2283 k0
= pc0
> rgba
[0][0];
2284 k1
= pc1
> rgba
[0][1];
2285 k2
= pc2
> rgba
[0][2];
2286 k3
= pc3
> rgba
[0][3];
2288 case PIPE_FUNC_GEQUAL
:
2289 k0
= pc0
>= rgba
[0][0];
2290 k1
= pc1
>= rgba
[0][1];
2291 k2
= pc2
>= rgba
[0][2];
2292 k3
= pc3
>= rgba
[0][3];
2294 case PIPE_FUNC_EQUAL
:
2295 k0
= pc0
== rgba
[0][0];
2296 k1
= pc1
== rgba
[0][1];
2297 k2
= pc2
== rgba
[0][2];
2298 k3
= pc3
== rgba
[0][3];
2300 case PIPE_FUNC_NOTEQUAL
:
2301 k0
= pc0
!= rgba
[0][0];
2302 k1
= pc1
!= rgba
[0][1];
2303 k2
= pc2
!= rgba
[0][2];
2304 k3
= pc3
!= rgba
[0][3];
2306 case PIPE_FUNC_ALWAYS
:
2307 k0
= k1
= k2
= k3
= 1;
2309 case PIPE_FUNC_NEVER
:
2310 k0
= k1
= k2
= k3
= 0;
2313 k0
= k1
= k2
= k3
= 0;
2318 if (sampler
->mag_img_filter
== PIPE_TEX_FILTER_LINEAR
) {
2319 /* convert four pass/fail values to an intensity in [0,1] */
2320 val
= 0.25F
* (k0
+ k1
+ k2
+ k3
);
2322 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
2323 for (j
= 0; j
< 4; j
++) {
2324 rgba
[0][j
] = rgba
[1][j
] = rgba
[2][j
] = val
;
2328 for (j
= 0; j
< 4; j
++) {
2339 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2340 * Put face info into the sampler faces[] array.
2343 sample_cube(struct tgsi_sampler
*tgsi_sampler
,
2344 const float s
[TGSI_QUAD_SIZE
],
2345 const float t
[TGSI_QUAD_SIZE
],
2346 const float p
[TGSI_QUAD_SIZE
],
2347 const float c0
[TGSI_QUAD_SIZE
],
2348 const float c1
[TGSI_QUAD_SIZE
],
2349 enum tgsi_sampler_control control
,
2350 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2352 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2354 float ssss
[4], tttt
[4];
2356 /* Not actually used, but the intermediate steps that do the
2357 * dereferencing don't know it.
2359 static float pppp
[4] = { 0, 0, 0, 0 };
2367 direction target sc tc ma
2368 ---------- ------------------------------- --- --- ---
2369 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2370 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2371 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2372 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2373 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2374 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2377 /* Choose the cube face and compute new s/t coords for the 2D face.
2379 * Use the same cube face for all four pixels in the quad.
2381 * This isn't ideal, but if we want to use a different cube face
2382 * per pixel in the quad, we'd have to also compute the per-face
2383 * LOD here too. That's because the four post-face-selection
2384 * texcoords are no longer related to each other (they're
2385 * per-face!) so we can't use subtraction to compute the partial
2386 * deriviates to compute the LOD. Doing so (near cube edges
2387 * anyway) gives us pretty much random values.
2390 /* use the average of the four pixel's texcoords to choose the face */
2391 const float rx
= 0.25F
* (s
[0] + s
[1] + s
[2] + s
[3]);
2392 const float ry
= 0.25F
* (t
[0] + t
[1] + t
[2] + t
[3]);
2393 const float rz
= 0.25F
* (p
[0] + p
[1] + p
[2] + p
[3]);
2394 const float arx
= fabsf(rx
), ary
= fabsf(ry
), arz
= fabsf(rz
);
2396 if (arx
>= ary
&& arx
>= arz
) {
2397 float sign
= (rx
>= 0.0F
) ? 1.0F
: -1.0F
;
2398 uint face
= (rx
>= 0.0F
) ? PIPE_TEX_FACE_POS_X
: PIPE_TEX_FACE_NEG_X
;
2399 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2400 const float ima
= -0.5F
/ fabsf(s
[j
]);
2401 ssss
[j
] = sign
* p
[j
] * ima
+ 0.5F
;
2402 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
2403 samp
->faces
[j
] = face
;
2406 else if (ary
>= arx
&& ary
>= arz
) {
2407 float sign
= (ry
>= 0.0F
) ? 1.0F
: -1.0F
;
2408 uint face
= (ry
>= 0.0F
) ? PIPE_TEX_FACE_POS_Y
: PIPE_TEX_FACE_NEG_Y
;
2409 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2410 const float ima
= -0.5F
/ fabsf(t
[j
]);
2411 ssss
[j
] = -s
[j
] * ima
+ 0.5F
;
2412 tttt
[j
] = sign
* -p
[j
] * ima
+ 0.5F
;
2413 samp
->faces
[j
] = face
;
2417 float sign
= (rz
>= 0.0F
) ? 1.0F
: -1.0F
;
2418 uint face
= (rz
>= 0.0F
) ? PIPE_TEX_FACE_POS_Z
: PIPE_TEX_FACE_NEG_Z
;
2419 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2420 const float ima
= -0.5F
/ fabsf(p
[j
]);
2421 ssss
[j
] = sign
* -s
[j
] * ima
+ 0.5F
;
2422 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
2423 samp
->faces
[j
] = face
;
2428 /* In our little pipeline, the compare stage is next. If compare
2429 * is not active, this will point somewhere deeper into the
2430 * pipeline, eg. to mip_filter or even img_filter.
2432 samp
->compare(tgsi_sampler
, ssss
, tttt
, pppp
, c0
, c1
, control
, rgba
);
2437 do_swizzling(const struct sp_sampler_variant
*samp
,
2438 float in
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
2439 float out
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2442 const unsigned swizzle_r
= samp
->key
.bits
.swizzle_r
;
2443 const unsigned swizzle_g
= samp
->key
.bits
.swizzle_g
;
2444 const unsigned swizzle_b
= samp
->key
.bits
.swizzle_b
;
2445 const unsigned swizzle_a
= samp
->key
.bits
.swizzle_a
;
2447 switch (swizzle_r
) {
2448 case PIPE_SWIZZLE_ZERO
:
2449 for (j
= 0; j
< 4; j
++)
2452 case PIPE_SWIZZLE_ONE
:
2453 for (j
= 0; j
< 4; j
++)
2457 assert(swizzle_r
< 4);
2458 for (j
= 0; j
< 4; j
++)
2459 out
[0][j
] = in
[swizzle_r
][j
];
2462 switch (swizzle_g
) {
2463 case PIPE_SWIZZLE_ZERO
:
2464 for (j
= 0; j
< 4; j
++)
2467 case PIPE_SWIZZLE_ONE
:
2468 for (j
= 0; j
< 4; j
++)
2472 assert(swizzle_g
< 4);
2473 for (j
= 0; j
< 4; j
++)
2474 out
[1][j
] = in
[swizzle_g
][j
];
2477 switch (swizzle_b
) {
2478 case PIPE_SWIZZLE_ZERO
:
2479 for (j
= 0; j
< 4; j
++)
2482 case PIPE_SWIZZLE_ONE
:
2483 for (j
= 0; j
< 4; j
++)
2487 assert(swizzle_b
< 4);
2488 for (j
= 0; j
< 4; j
++)
2489 out
[2][j
] = in
[swizzle_b
][j
];
2492 switch (swizzle_a
) {
2493 case PIPE_SWIZZLE_ZERO
:
2494 for (j
= 0; j
< 4; j
++)
2497 case PIPE_SWIZZLE_ONE
:
2498 for (j
= 0; j
< 4; j
++)
2502 assert(swizzle_a
< 4);
2503 for (j
= 0; j
< 4; j
++)
2504 out
[3][j
] = in
[swizzle_a
][j
];
2510 sample_swizzle(struct tgsi_sampler
*tgsi_sampler
,
2511 const float s
[TGSI_QUAD_SIZE
],
2512 const float t
[TGSI_QUAD_SIZE
],
2513 const float p
[TGSI_QUAD_SIZE
],
2514 const float c0
[TGSI_QUAD_SIZE
],
2515 const float c1
[TGSI_QUAD_SIZE
],
2516 enum tgsi_sampler_control control
,
2517 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2519 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2520 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2522 samp
->sample_target(tgsi_sampler
, s
, t
, p
, c0
, c1
, control
, rgba_temp
);
2524 do_swizzling(samp
, rgba_temp
, rgba
);
2528 static wrap_nearest_func
2529 get_nearest_unorm_wrap(unsigned mode
)
2532 case PIPE_TEX_WRAP_CLAMP
:
2533 return wrap_nearest_unorm_clamp
;
2534 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2535 return wrap_nearest_unorm_clamp_to_edge
;
2536 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2537 return wrap_nearest_unorm_clamp_to_border
;
2540 return wrap_nearest_unorm_clamp
;
2545 static wrap_nearest_func
2546 get_nearest_wrap(unsigned mode
)
2549 case PIPE_TEX_WRAP_REPEAT
:
2550 return wrap_nearest_repeat
;
2551 case PIPE_TEX_WRAP_CLAMP
:
2552 return wrap_nearest_clamp
;
2553 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2554 return wrap_nearest_clamp_to_edge
;
2555 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2556 return wrap_nearest_clamp_to_border
;
2557 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2558 return wrap_nearest_mirror_repeat
;
2559 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2560 return wrap_nearest_mirror_clamp
;
2561 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2562 return wrap_nearest_mirror_clamp_to_edge
;
2563 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2564 return wrap_nearest_mirror_clamp_to_border
;
2567 return wrap_nearest_repeat
;
2572 static wrap_linear_func
2573 get_linear_unorm_wrap(unsigned mode
)
2576 case PIPE_TEX_WRAP_CLAMP
:
2577 return wrap_linear_unorm_clamp
;
2578 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2579 return wrap_linear_unorm_clamp_to_edge
;
2580 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2581 return wrap_linear_unorm_clamp_to_border
;
2584 return wrap_linear_unorm_clamp
;
2589 static wrap_linear_func
2590 get_linear_wrap(unsigned mode
)
2593 case PIPE_TEX_WRAP_REPEAT
:
2594 return wrap_linear_repeat
;
2595 case PIPE_TEX_WRAP_CLAMP
:
2596 return wrap_linear_clamp
;
2597 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2598 return wrap_linear_clamp_to_edge
;
2599 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2600 return wrap_linear_clamp_to_border
;
2601 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2602 return wrap_linear_mirror_repeat
;
2603 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2604 return wrap_linear_mirror_clamp
;
2605 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2606 return wrap_linear_mirror_clamp_to_edge
;
2607 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2608 return wrap_linear_mirror_clamp_to_border
;
2611 return wrap_linear_repeat
;
2617 * Is swizzling needed for the given state key?
2620 any_swizzle(union sp_sampler_key key
)
2622 return (key
.bits
.swizzle_r
!= PIPE_SWIZZLE_RED
||
2623 key
.bits
.swizzle_g
!= PIPE_SWIZZLE_GREEN
||
2624 key
.bits
.swizzle_b
!= PIPE_SWIZZLE_BLUE
||
2625 key
.bits
.swizzle_a
!= PIPE_SWIZZLE_ALPHA
);
2629 static compute_lambda_func
2630 get_lambda_func(const union sp_sampler_key key
)
2632 if (key
.bits
.processor
== TGSI_PROCESSOR_VERTEX
)
2633 return compute_lambda_vert
;
2635 switch (key
.bits
.target
) {
2637 case PIPE_TEXTURE_1D
:
2638 case PIPE_TEXTURE_1D_ARRAY
:
2639 return compute_lambda_1d
;
2640 case PIPE_TEXTURE_2D
:
2641 case PIPE_TEXTURE_2D_ARRAY
:
2642 case PIPE_TEXTURE_RECT
:
2643 case PIPE_TEXTURE_CUBE
:
2644 case PIPE_TEXTURE_CUBE_ARRAY
:
2645 return compute_lambda_2d
;
2646 case PIPE_TEXTURE_3D
:
2647 return compute_lambda_3d
;
2650 return compute_lambda_1d
;
2655 static img_filter_func
2656 get_img_filter(const union sp_sampler_key key
,
2658 const struct pipe_sampler_state
*sampler
)
2660 switch (key
.bits
.target
) {
2662 case PIPE_TEXTURE_1D
:
2663 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2664 return img_filter_1d_nearest
;
2666 return img_filter_1d_linear
;
2668 case PIPE_TEXTURE_1D_ARRAY
:
2669 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2670 return img_filter_1d_array_nearest
;
2672 return img_filter_1d_array_linear
;
2674 case PIPE_TEXTURE_2D
:
2675 case PIPE_TEXTURE_RECT
:
2676 /* Try for fast path:
2678 if (key
.bits
.is_pot
&&
2679 sampler
->wrap_s
== sampler
->wrap_t
&&
2680 sampler
->normalized_coords
)
2682 switch (sampler
->wrap_s
) {
2683 case PIPE_TEX_WRAP_REPEAT
:
2685 case PIPE_TEX_FILTER_NEAREST
:
2686 return img_filter_2d_nearest_repeat_POT
;
2687 case PIPE_TEX_FILTER_LINEAR
:
2688 return img_filter_2d_linear_repeat_POT
;
2693 case PIPE_TEX_WRAP_CLAMP
:
2695 case PIPE_TEX_FILTER_NEAREST
:
2696 return img_filter_2d_nearest_clamp_POT
;
2702 /* Otherwise use default versions:
2704 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2705 return img_filter_2d_nearest
;
2707 return img_filter_2d_linear
;
2709 case PIPE_TEXTURE_2D_ARRAY
:
2710 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2711 return img_filter_2d_array_nearest
;
2713 return img_filter_2d_array_linear
;
2715 case PIPE_TEXTURE_CUBE
:
2716 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2717 return img_filter_cube_nearest
;
2719 return img_filter_cube_linear
;
2721 case PIPE_TEXTURE_CUBE_ARRAY
:
2722 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2723 return img_filter_cube_array_nearest
;
2725 return img_filter_cube_array_linear
;
2727 case PIPE_TEXTURE_3D
:
2728 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2729 return img_filter_3d_nearest
;
2731 return img_filter_3d_linear
;
2735 return img_filter_1d_nearest
;
2741 * Bind the given texture object and texture cache to the sampler variant.
2744 sp_sampler_variant_bind_view( struct sp_sampler_variant
*samp
,
2745 struct softpipe_tex_tile_cache
*tex_cache
,
2746 const struct pipe_sampler_view
*view
)
2748 const struct pipe_resource
*texture
= view
->texture
;
2751 samp
->cache
= tex_cache
;
2752 samp
->xpot
= util_logbase2( texture
->width0
);
2753 samp
->ypot
= util_logbase2( texture
->height0
);
2758 sp_sampler_variant_destroy( struct sp_sampler_variant
*samp
)
2765 sample_get_dims(struct tgsi_sampler
*tgsi_sampler
, int level
,
2768 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2769 const struct pipe_sampler_view
*view
= samp
->view
;
2770 const struct pipe_resource
*texture
= view
->texture
;
2772 /* undefined according to EXT_gpu_program */
2773 level
+= view
->u
.tex
.first_level
;
2774 if (level
> view
->u
.tex
.last_level
)
2777 dims
[0] = u_minify(texture
->width0
, level
);
2779 switch(texture
->target
) {
2780 case PIPE_TEXTURE_1D_ARRAY
:
2781 dims
[1] = texture
->array_size
;
2783 case PIPE_TEXTURE_1D
:
2785 case PIPE_TEXTURE_2D_ARRAY
:
2786 dims
[2] = texture
->array_size
;
2788 case PIPE_TEXTURE_2D
:
2789 case PIPE_TEXTURE_CUBE
:
2790 case PIPE_TEXTURE_RECT
:
2791 dims
[1] = u_minify(texture
->height0
, level
);
2793 case PIPE_TEXTURE_3D
:
2794 dims
[1] = u_minify(texture
->height0
, level
);
2795 dims
[2] = u_minify(texture
->depth0
, level
);
2797 case PIPE_TEXTURE_CUBE_ARRAY
:
2798 dims
[1] = u_minify(texture
->height0
, level
);
2799 dims
[2] = texture
->array_size
/ 6;
2802 dims
[0] /= util_format_get_blocksize(view
->format
);
2805 assert(!"unexpected texture target in sample_get_dims()");
2811 * This function is only used for getting unfiltered texels via the
2812 * TXF opcode. The GL spec says that out-of-bounds texel fetches
2813 * produce undefined results. Instead of crashing, lets just clamp
2814 * coords to the texture image size.
2817 sample_get_texels(struct tgsi_sampler
*tgsi_sampler
,
2818 const int v_i
[TGSI_QUAD_SIZE
],
2819 const int v_j
[TGSI_QUAD_SIZE
],
2820 const int v_k
[TGSI_QUAD_SIZE
],
2821 const int lod
[TGSI_QUAD_SIZE
],
2822 const int8_t offset
[3],
2823 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2825 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2826 union tex_tile_address addr
;
2827 const struct pipe_resource
*texture
= samp
->view
->texture
;
2830 const bool need_swizzle
= any_swizzle(samp
->key
);
2831 int width
, height
, depth
, layers
;
2834 /* TODO write a better test for LOD */
2835 addr
.bits
.level
= lod
[0];
2837 width
= u_minify(texture
->width0
, addr
.bits
.level
);
2838 height
= u_minify(texture
->height0
, addr
.bits
.level
);
2839 depth
= u_minify(texture
->depth0
, addr
.bits
.level
);
2840 layers
= texture
->array_size
;
2842 switch(texture
->target
) {
2844 case PIPE_TEXTURE_1D
:
2845 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2846 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2847 tx
= get_texel_2d(samp
, addr
, x
, 0);
2848 for (c
= 0; c
< 4; c
++) {
2853 case PIPE_TEXTURE_1D_ARRAY
:
2854 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2855 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2856 int y
= CLAMP(v_j
[j
], 0, layers
- 1);
2857 tx
= get_texel_1d_array(samp
, addr
, x
, y
);
2858 for (c
= 0; c
< 4; c
++) {
2863 case PIPE_TEXTURE_2D
:
2864 case PIPE_TEXTURE_RECT
:
2865 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2866 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2867 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
2868 tx
= get_texel_2d(samp
, addr
, x
, y
);
2869 for (c
= 0; c
< 4; c
++) {
2874 case PIPE_TEXTURE_2D_ARRAY
:
2875 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2876 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2877 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
2878 int layer
= CLAMP(v_k
[j
], 0, layers
- 1);
2879 tx
= get_texel_2d_array(samp
, addr
, x
, y
, layer
);
2880 for (c
= 0; c
< 4; c
++) {
2885 case PIPE_TEXTURE_3D
:
2886 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2887 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2888 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
2889 int z
= CLAMP(v_k
[j
] + offset
[2], 0, depth
- 1);
2891 tx
= get_texel_3d(samp
, addr
, x
, y
, z
);
2892 for (c
= 0; c
< 4; c
++) {
2897 case PIPE_TEXTURE_CUBE
: /* TXF can't work on CUBE according to spec */
2899 assert(!"Unknown or CUBE texture type in TXF processing\n");
2904 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2905 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
2906 do_swizzling(samp
, rgba_temp
, rgba
);
2912 * Create a sampler variant for a given set of non-orthogonal state.
2914 struct sp_sampler_variant
*
2915 sp_create_sampler_variant( const struct pipe_sampler_state
*sampler
,
2916 const union sp_sampler_key key
)
2918 struct sp_sampler_variant
*samp
= CALLOC_STRUCT(sp_sampler_variant
);
2922 samp
->sampler
= sampler
;
2925 /* Note that (for instance) linear_texcoord_s and
2926 * nearest_texcoord_s may be active at the same time, if the
2927 * sampler min_img_filter differs from its mag_img_filter.
2929 if (sampler
->normalized_coords
) {
2930 samp
->linear_texcoord_s
= get_linear_wrap( sampler
->wrap_s
);
2931 samp
->linear_texcoord_t
= get_linear_wrap( sampler
->wrap_t
);
2932 samp
->linear_texcoord_p
= get_linear_wrap( sampler
->wrap_r
);
2934 samp
->nearest_texcoord_s
= get_nearest_wrap( sampler
->wrap_s
);
2935 samp
->nearest_texcoord_t
= get_nearest_wrap( sampler
->wrap_t
);
2936 samp
->nearest_texcoord_p
= get_nearest_wrap( sampler
->wrap_r
);
2939 samp
->linear_texcoord_s
= get_linear_unorm_wrap( sampler
->wrap_s
);
2940 samp
->linear_texcoord_t
= get_linear_unorm_wrap( sampler
->wrap_t
);
2941 samp
->linear_texcoord_p
= get_linear_unorm_wrap( sampler
->wrap_r
);
2943 samp
->nearest_texcoord_s
= get_nearest_unorm_wrap( sampler
->wrap_s
);
2944 samp
->nearest_texcoord_t
= get_nearest_unorm_wrap( sampler
->wrap_t
);
2945 samp
->nearest_texcoord_p
= get_nearest_unorm_wrap( sampler
->wrap_r
);
2948 samp
->compute_lambda
= get_lambda_func( key
);
2950 samp
->min_img_filter
= get_img_filter(key
, sampler
->min_img_filter
, sampler
);
2951 samp
->mag_img_filter
= get_img_filter(key
, sampler
->mag_img_filter
, sampler
);
2953 switch (sampler
->min_mip_filter
) {
2954 case PIPE_TEX_MIPFILTER_NONE
:
2955 if (sampler
->min_img_filter
== sampler
->mag_img_filter
)
2956 samp
->mip_filter
= mip_filter_none_no_filter_select
;
2958 samp
->mip_filter
= mip_filter_none
;
2961 case PIPE_TEX_MIPFILTER_NEAREST
:
2962 samp
->mip_filter
= mip_filter_nearest
;
2965 case PIPE_TEX_MIPFILTER_LINEAR
:
2966 if (key
.bits
.is_pot
&&
2967 sampler
->min_img_filter
== sampler
->mag_img_filter
&&
2968 sampler
->normalized_coords
&&
2969 sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
&&
2970 sampler
->wrap_t
== PIPE_TEX_WRAP_REPEAT
&&
2971 sampler
->min_img_filter
== PIPE_TEX_FILTER_LINEAR
) {
2972 samp
->mip_filter
= mip_filter_linear_2d_linear_repeat_POT
;
2975 samp
->mip_filter
= mip_filter_linear
;
2978 /* Anisotropic filtering extension. */
2979 if (sampler
->max_anisotropy
> 1) {
2980 samp
->mip_filter
= mip_filter_linear_aniso
;
2982 /* Override min_img_filter:
2983 * min_img_filter needs to be set to NEAREST since we need to access
2984 * each texture pixel as it is and weight it later; using linear
2985 * filters will have incorrect results.
2986 * By setting the filter to NEAREST here, we can avoid calling the
2987 * generic img_filter_2d_nearest in the anisotropic filter function,
2988 * making it possible to use one of the accelerated implementations
2990 samp
->min_img_filter
= get_img_filter(key
, PIPE_TEX_FILTER_NEAREST
, sampler
);
2992 /* on first access create the lookup table containing the filter weights. */
2994 create_filter_table();
3001 if (sampler
->compare_mode
!= PIPE_TEX_COMPARE_NONE
) {
3002 samp
->compare
= sample_compare
;
3005 /* Skip compare operation by promoting the mip_filter function
3008 samp
->compare
= samp
->mip_filter
;
3011 if (key
.bits
.target
== PIPE_TEXTURE_CUBE
|| key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
) {
3012 samp
->sample_target
= sample_cube
;
3020 /* Skip cube face determination by promoting the compare
3023 samp
->sample_target
= samp
->compare
;
3026 if (any_swizzle(key
)) {
3027 samp
->base
.get_samples
= sample_swizzle
;
3030 samp
->base
.get_samples
= samp
->sample_target
;
3033 samp
->base
.get_dims
= sample_get_dims
;
3034 samp
->base
.get_texel
= sample_get_texels
;