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
,
910 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
911 unsigned xpot
= pot_level_size(samp
->xpot
, level
);
912 unsigned ypot
= pot_level_size(samp
->ypot
, level
);
913 unsigned xmax
= (xpot
- 1) & (TILE_SIZE
- 1); /* MIN2(TILE_SIZE, xpot) - 1; */
914 unsigned ymax
= (ypot
- 1) & (TILE_SIZE
- 1); /* MIN2(TILE_SIZE, ypot) - 1; */
915 union tex_tile_address addr
;
918 float u
= s
* xpot
- 0.5F
;
919 float v
= t
* ypot
- 0.5F
;
921 int uflr
= util_ifloor(u
);
922 int vflr
= util_ifloor(v
);
924 float xw
= u
- (float)uflr
;
925 float yw
= v
- (float)vflr
;
927 int x0
= uflr
& (xpot
- 1);
928 int y0
= vflr
& (ypot
- 1);
933 addr
.bits
.level
= level
;
935 /* Can we fetch all four at once:
937 if (x0
< xmax
&& y0
< ymax
) {
938 get_texel_quad_2d_no_border_single_tile(samp
, addr
, x0
, y0
, tx
);
941 unsigned x1
= (x0
+ 1) & (xpot
- 1);
942 unsigned y1
= (y0
+ 1) & (ypot
- 1);
943 get_texel_quad_2d_no_border(samp
, addr
, x0
, y0
, x1
, y1
, tx
);
946 /* interpolate R, G, B, A */
947 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++) {
948 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
954 print_sample(__FUNCTION__
, rgba
);
960 img_filter_2d_nearest_repeat_POT(struct tgsi_sampler
*tgsi_sampler
,
966 float rgba
[TGSI_QUAD_SIZE
])
968 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
969 unsigned xpot
= pot_level_size(samp
->xpot
, level
);
970 unsigned ypot
= pot_level_size(samp
->ypot
, level
);
972 union tex_tile_address addr
;
978 int uflr
= util_ifloor(u
);
979 int vflr
= util_ifloor(v
);
981 int x0
= uflr
& (xpot
- 1);
982 int y0
= vflr
& (ypot
- 1);
985 addr
.bits
.level
= level
;
987 out
= get_texel_2d_no_border(samp
, addr
, x0
, y0
);
988 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
989 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
992 print_sample(__FUNCTION__
, rgba
);
998 img_filter_2d_nearest_clamp_POT(struct tgsi_sampler
*tgsi_sampler
,
1004 float rgba
[TGSI_QUAD_SIZE
])
1006 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1007 unsigned xpot
= pot_level_size(samp
->xpot
, level
);
1008 unsigned ypot
= pot_level_size(samp
->ypot
, level
);
1009 union tex_tile_address addr
;
1019 addr
.bits
.level
= level
;
1021 x0
= util_ifloor(u
);
1024 else if (x0
> xpot
- 1)
1027 y0
= util_ifloor(v
);
1030 else if (y0
> ypot
- 1)
1033 out
= get_texel_2d_no_border(samp
, addr
, x0
, y0
);
1034 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1035 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1038 print_sample(__FUNCTION__
, rgba
);
1044 img_filter_1d_nearest(struct tgsi_sampler
*tgsi_sampler
,
1050 float rgba
[TGSI_QUAD_SIZE
])
1052 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1053 const struct pipe_resource
*texture
= samp
->view
->texture
;
1056 union tex_tile_address addr
;
1060 width
= u_minify(texture
->width0
, level
);
1065 addr
.bits
.level
= level
;
1067 samp
->nearest_texcoord_s(s
, width
, &x
);
1069 out
= get_texel_2d(samp
, addr
, x
, 0);
1070 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1071 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1074 print_sample(__FUNCTION__
, rgba
);
1080 img_filter_1d_array_nearest(struct tgsi_sampler
*tgsi_sampler
,
1088 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1089 const struct pipe_resource
*texture
= samp
->view
->texture
;
1092 union tex_tile_address addr
;
1096 width
= u_minify(texture
->width0
, level
);
1101 addr
.bits
.level
= level
;
1103 samp
->nearest_texcoord_s(s
, width
, &x
);
1104 wrap_array_layer(t
, texture
->array_size
, &layer
);
1106 out
= get_texel_1d_array(samp
, addr
, x
, layer
);
1107 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1108 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1111 print_sample(__FUNCTION__
, rgba
);
1117 img_filter_2d_nearest(struct tgsi_sampler
*tgsi_sampler
,
1125 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1126 const struct pipe_resource
*texture
= samp
->view
->texture
;
1129 union tex_tile_address addr
;
1133 width
= u_minify(texture
->width0
, level
);
1134 height
= u_minify(texture
->height0
, level
);
1140 addr
.bits
.level
= level
;
1142 samp
->nearest_texcoord_s(s
, width
, &x
);
1143 samp
->nearest_texcoord_t(t
, height
, &y
);
1145 out
= get_texel_2d(samp
, addr
, x
, y
);
1146 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1147 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1150 print_sample(__FUNCTION__
, rgba
);
1156 img_filter_2d_array_nearest(struct tgsi_sampler
*tgsi_sampler
,
1164 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1165 const struct pipe_resource
*texture
= samp
->view
->texture
;
1168 union tex_tile_address addr
;
1172 width
= u_minify(texture
->width0
, level
);
1173 height
= u_minify(texture
->height0
, level
);
1179 addr
.bits
.level
= level
;
1181 samp
->nearest_texcoord_s(s
, width
, &x
);
1182 samp
->nearest_texcoord_t(t
, height
, &y
);
1183 wrap_array_layer(p
, texture
->array_size
, &layer
);
1185 out
= get_texel_2d_array(samp
, addr
, x
, y
, layer
);
1186 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1187 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1190 print_sample(__FUNCTION__
, rgba
);
1195 static INLINE
union tex_tile_address
1196 face(union tex_tile_address addr
, unsigned face
)
1198 addr
.bits
.face
= face
;
1204 img_filter_cube_nearest(struct tgsi_sampler
*tgsi_sampler
,
1212 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1213 const struct pipe_resource
*texture
= samp
->view
->texture
;
1216 union tex_tile_address addr
;
1220 width
= u_minify(texture
->width0
, level
);
1221 height
= u_minify(texture
->height0
, level
);
1227 addr
.bits
.level
= level
;
1230 * If NEAREST filtering is done within a miplevel, always apply wrap
1231 * mode CLAMP_TO_EDGE.
1233 if (samp
->sampler
->seamless_cube_map
) {
1234 wrap_nearest_clamp_to_edge(s
, width
, &x
);
1235 wrap_nearest_clamp_to_edge(t
, height
, &y
);
1237 samp
->nearest_texcoord_s(s
, width
, &x
);
1238 samp
->nearest_texcoord_t(t
, height
, &y
);
1241 out
= get_texel_2d(samp
, face(addr
, face_id
), x
, y
);
1242 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1243 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1246 print_sample(__FUNCTION__
, rgba
);
1251 img_filter_cube_array_nearest(struct tgsi_sampler
*tgsi_sampler
,
1259 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1260 const struct pipe_resource
*texture
= samp
->view
->texture
;
1263 union tex_tile_address addr
;
1267 width
= u_minify(texture
->width0
, level
);
1268 height
= u_minify(texture
->height0
, level
);
1274 addr
.bits
.level
= level
;
1276 samp
->nearest_texcoord_s(s
, width
, &x
);
1277 samp
->nearest_texcoord_t(t
, height
, &y
);
1278 wrap_array_layer(p
, texture
->array_size
, &layer
);
1280 out
= get_texel_cube_array(samp
, addr
, x
, y
, layer
* 6 + face_id
);
1281 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1282 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1285 print_sample(__FUNCTION__
, rgba
);
1290 img_filter_3d_nearest(struct tgsi_sampler
*tgsi_sampler
,
1298 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1299 const struct pipe_resource
*texture
= samp
->view
->texture
;
1300 int width
, height
, depth
;
1302 union tex_tile_address addr
;
1306 width
= u_minify(texture
->width0
, level
);
1307 height
= u_minify(texture
->height0
, level
);
1308 depth
= u_minify(texture
->depth0
, level
);
1314 samp
->nearest_texcoord_s(s
, width
, &x
);
1315 samp
->nearest_texcoord_t(t
, height
, &y
);
1316 samp
->nearest_texcoord_p(p
, depth
, &z
);
1319 addr
.bits
.level
= level
;
1321 out
= get_texel_3d(samp
, addr
, x
, y
, z
);
1322 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1323 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1328 img_filter_1d_linear(struct tgsi_sampler
*tgsi_sampler
,
1336 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1337 const struct pipe_resource
*texture
= samp
->view
->texture
;
1340 float xw
; /* weights */
1341 union tex_tile_address addr
;
1342 const float *tx0
, *tx1
;
1345 width
= u_minify(texture
->width0
, level
);
1350 addr
.bits
.level
= level
;
1352 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1354 tx0
= get_texel_2d(samp
, addr
, x0
, 0);
1355 tx1
= get_texel_2d(samp
, addr
, x1
, 0);
1357 /* interpolate R, G, B, A */
1358 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1359 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1364 img_filter_1d_array_linear(struct tgsi_sampler
*tgsi_sampler
,
1372 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1373 const struct pipe_resource
*texture
= samp
->view
->texture
;
1376 float xw
; /* weights */
1377 union tex_tile_address addr
;
1378 const float *tx0
, *tx1
;
1381 width
= u_minify(texture
->width0
, level
);
1386 addr
.bits
.level
= level
;
1388 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1389 wrap_array_layer(t
, texture
->array_size
, &layer
);
1391 tx0
= get_texel_1d_array(samp
, addr
, x0
, layer
);
1392 tx1
= get_texel_1d_array(samp
, addr
, x1
, layer
);
1394 /* interpolate R, G, B, A */
1395 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1396 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1401 img_filter_2d_linear(struct tgsi_sampler
*tgsi_sampler
,
1409 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1410 const struct pipe_resource
*texture
= samp
->view
->texture
;
1413 float xw
, yw
; /* weights */
1414 union tex_tile_address addr
;
1415 const float *tx0
, *tx1
, *tx2
, *tx3
;
1418 width
= u_minify(texture
->width0
, level
);
1419 height
= u_minify(texture
->height0
, level
);
1425 addr
.bits
.level
= level
;
1427 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1428 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1430 tx0
= get_texel_2d(samp
, addr
, x0
, y0
);
1431 tx1
= get_texel_2d(samp
, addr
, x1
, y0
);
1432 tx2
= get_texel_2d(samp
, addr
, x0
, y1
);
1433 tx3
= get_texel_2d(samp
, addr
, x1
, y1
);
1435 /* interpolate R, G, B, A */
1436 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1437 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1444 img_filter_2d_array_linear(struct tgsi_sampler
*tgsi_sampler
,
1452 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1453 const struct pipe_resource
*texture
= samp
->view
->texture
;
1455 int x0
, y0
, x1
, y1
, layer
;
1456 float xw
, yw
; /* weights */
1457 union tex_tile_address addr
;
1458 const float *tx0
, *tx1
, *tx2
, *tx3
;
1461 width
= u_minify(texture
->width0
, level
);
1462 height
= u_minify(texture
->height0
, level
);
1468 addr
.bits
.level
= level
;
1470 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1471 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1472 wrap_array_layer(p
, texture
->array_size
, &layer
);
1474 tx0
= get_texel_2d_array(samp
, addr
, x0
, y0
, layer
);
1475 tx1
= get_texel_2d_array(samp
, addr
, x1
, y0
, layer
);
1476 tx2
= get_texel_2d_array(samp
, addr
, x0
, y1
, layer
);
1477 tx3
= get_texel_2d_array(samp
, addr
, x1
, y1
, layer
);
1479 /* interpolate R, G, B, A */
1480 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1481 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1488 img_filter_cube_linear(struct tgsi_sampler
*tgsi_sampler
,
1496 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1497 const struct pipe_resource
*texture
= samp
->view
->texture
;
1500 float xw
, yw
; /* weights */
1501 union tex_tile_address addr
, addrj
;
1502 const float *tx0
, *tx1
, *tx2
, *tx3
;
1503 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
], corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1506 width
= u_minify(texture
->width0
, level
);
1507 height
= u_minify(texture
->height0
, level
);
1513 addr
.bits
.level
= level
;
1516 * For seamless if LINEAR filtering is done within a miplevel,
1517 * always apply wrap mode CLAMP_TO_BORDER.
1519 if (samp
->sampler
->seamless_cube_map
) {
1520 wrap_linear_clamp_to_border(s
, width
, &x0
, &x1
, &xw
);
1521 wrap_linear_clamp_to_border(t
, height
, &y0
, &y1
, &yw
);
1523 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1524 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1527 addrj
= face(addr
, face_id
);
1529 if (samp
->sampler
->seamless_cube_map
) {
1530 tx0
= get_texel_cube_seamless(samp
, addrj
, x0
, y0
, corner0
);
1531 tx1
= get_texel_cube_seamless(samp
, addrj
, x1
, y0
, corner1
);
1532 tx2
= get_texel_cube_seamless(samp
, addrj
, x0
, y1
, corner2
);
1533 tx3
= get_texel_cube_seamless(samp
, addrj
, x1
, y1
, corner3
);
1535 tx0
= get_texel_2d(samp
, addrj
, x0
, y0
);
1536 tx1
= get_texel_2d(samp
, addrj
, x1
, y0
);
1537 tx2
= get_texel_2d(samp
, addrj
, x0
, y1
);
1538 tx3
= get_texel_2d(samp
, addrj
, x1
, y1
);
1540 /* interpolate R, G, B, A */
1541 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1542 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1549 img_filter_cube_array_linear(struct tgsi_sampler
*tgsi_sampler
,
1557 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1558 const struct pipe_resource
*texture
= samp
->view
->texture
;
1560 int x0
, y0
, x1
, y1
, layer
;
1561 float xw
, yw
; /* weights */
1562 union tex_tile_address addr
;
1563 const float *tx0
, *tx1
, *tx2
, *tx3
;
1566 width
= u_minify(texture
->width0
, level
);
1567 height
= u_minify(texture
->height0
, level
);
1573 addr
.bits
.level
= level
;
1575 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1576 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1577 wrap_array_layer(p
, texture
->array_size
, &layer
);
1579 tx0
= get_texel_cube_array(samp
, addr
, x0
, y0
, layer
* 6 + face_id
);
1580 tx1
= get_texel_cube_array(samp
, addr
, x1
, y0
, layer
* 6 + face_id
);
1581 tx2
= get_texel_cube_array(samp
, addr
, x0
, y1
, layer
* 6 + face_id
);
1582 tx3
= get_texel_cube_array(samp
, addr
, x1
, y1
, layer
* 6 + face_id
);
1584 /* interpolate R, G, B, A */
1585 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1586 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1592 img_filter_3d_linear(struct tgsi_sampler
*tgsi_sampler
,
1600 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1601 const struct pipe_resource
*texture
= samp
->view
->texture
;
1602 int width
, height
, depth
;
1603 int x0
, x1
, y0
, y1
, z0
, z1
;
1604 float xw
, yw
, zw
; /* interpolation weights */
1605 union tex_tile_address addr
;
1606 const float *tx00
, *tx01
, *tx02
, *tx03
, *tx10
, *tx11
, *tx12
, *tx13
;
1609 width
= u_minify(texture
->width0
, level
);
1610 height
= u_minify(texture
->height0
, level
);
1611 depth
= u_minify(texture
->depth0
, level
);
1614 addr
.bits
.level
= level
;
1620 samp
->linear_texcoord_s(s
, width
, &x0
, &x1
, &xw
);
1621 samp
->linear_texcoord_t(t
, height
, &y0
, &y1
, &yw
);
1622 samp
->linear_texcoord_p(p
, depth
, &z0
, &z1
, &zw
);
1625 tx00
= get_texel_3d(samp
, addr
, x0
, y0
, z0
);
1626 tx01
= get_texel_3d(samp
, addr
, x1
, y0
, z0
);
1627 tx02
= get_texel_3d(samp
, addr
, x0
, y1
, z0
);
1628 tx03
= get_texel_3d(samp
, addr
, x1
, y1
, z0
);
1630 tx10
= get_texel_3d(samp
, addr
, x0
, y0
, z1
);
1631 tx11
= get_texel_3d(samp
, addr
, x1
, y0
, z1
);
1632 tx12
= get_texel_3d(samp
, addr
, x0
, y1
, z1
);
1633 tx13
= get_texel_3d(samp
, addr
, x1
, y1
, z1
);
1635 /* interpolate R, G, B, A */
1636 for (c
= 0; c
< TGSI_QUAD_SIZE
; c
++)
1637 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_3d(xw
, yw
, zw
,
1645 /* Calculate level of detail for every fragment.
1646 * Note that lambda has already been biased by global LOD bias.
1649 compute_lod(const struct pipe_sampler_state
*sampler
,
1650 const float biased_lambda
,
1651 const float lodbias
[TGSI_QUAD_SIZE
],
1652 float lod
[TGSI_QUAD_SIZE
])
1656 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1657 lod
[i
] = biased_lambda
+ lodbias
[i
];
1658 lod
[i
] = CLAMP(lod
[i
], sampler
->min_lod
, sampler
->max_lod
);
1664 mip_filter_linear(struct tgsi_sampler
*tgsi_sampler
,
1665 const float s
[TGSI_QUAD_SIZE
],
1666 const float t
[TGSI_QUAD_SIZE
],
1667 const float p
[TGSI_QUAD_SIZE
],
1668 const float c0
[TGSI_QUAD_SIZE
],
1669 const float c1
[TGSI_QUAD_SIZE
],
1670 enum tgsi_sampler_control control
,
1671 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1673 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1674 const struct pipe_resource
*texture
= samp
->view
->texture
;
1676 float lod
[TGSI_QUAD_SIZE
];
1678 if (control
== tgsi_sampler_lod_bias
) {
1679 float lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
1680 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1681 compute_lod(samp
->sampler
, lambda
, c1
, lod
);
1683 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
1685 assert(control
== tgsi_sampler_lod_explicit
);
1687 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1688 memcpy(lod
, c1
, sizeof(lod
));
1690 memcpy(lod
, c0
, sizeof(lod
));
1694 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1695 int level0
= samp
->view
->u
.tex
.first_level
+ (int)lod
[j
];
1698 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
,
1699 samp
->faces
[j
], &rgba
[0][j
]);
1701 else if (level0
>= texture
->last_level
)
1702 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], texture
->last_level
,
1703 samp
->faces
[j
], &rgba
[0][j
]);
1706 float levelBlend
= frac(lod
[j
]);
1707 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1710 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
,
1711 samp
->faces
[j
], &rgbax
[0][0]);
1712 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
+1,
1713 samp
->faces
[j
], &rgbax
[0][1]);
1715 for (c
= 0; c
< 4; c
++) {
1716 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
1722 print_sample_4(__FUNCTION__
, rgba
);
1728 * Compute nearest mipmap level from texcoords.
1729 * Then sample the texture level for four elements of a quad.
1730 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
1733 mip_filter_nearest(struct tgsi_sampler
*tgsi_sampler
,
1734 const float s
[TGSI_QUAD_SIZE
],
1735 const float t
[TGSI_QUAD_SIZE
],
1736 const float p
[TGSI_QUAD_SIZE
],
1737 const float c0
[TGSI_QUAD_SIZE
],
1738 const float c1
[TGSI_QUAD_SIZE
],
1739 enum tgsi_sampler_control control
,
1740 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1742 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1743 const struct pipe_resource
*texture
= samp
->view
->texture
;
1744 float lod
[TGSI_QUAD_SIZE
];
1747 if (control
== tgsi_sampler_lod_bias
) {
1748 float lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
1749 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1750 compute_lod(samp
->sampler
, lambda
, c1
, lod
);
1752 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
1754 assert(control
== tgsi_sampler_lod_explicit
);
1756 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1757 memcpy(lod
, c1
, sizeof(lod
));
1759 memcpy(lod
, c0
, sizeof(lod
));
1762 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1764 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
,
1765 samp
->faces
[j
], &rgba
[0][j
]);
1767 float level
= samp
->view
->u
.tex
.first_level
+ (int)(lod
[j
] + 0.5F
) ;
1768 level
= MIN2(level
, (int)texture
->last_level
);
1769 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level
, samp
->faces
[j
],
1775 print_sample_4(__FUNCTION__
, rgba
);
1781 mip_filter_none(struct tgsi_sampler
*tgsi_sampler
,
1782 const float s
[TGSI_QUAD_SIZE
],
1783 const float t
[TGSI_QUAD_SIZE
],
1784 const float p
[TGSI_QUAD_SIZE
],
1785 const float c0
[TGSI_QUAD_SIZE
],
1786 const float c1
[TGSI_QUAD_SIZE
],
1787 enum tgsi_sampler_control control
,
1788 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1790 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1791 float lod
[TGSI_QUAD_SIZE
];
1794 if (control
== tgsi_sampler_lod_bias
) {
1795 float lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
1796 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1797 compute_lod(samp
->sampler
, lambda
, c1
, lod
);
1799 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
1801 assert(control
== tgsi_sampler_lod_explicit
);
1803 if (samp
->key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
)
1804 memcpy(lod
, c1
, sizeof(lod
));
1806 memcpy(lod
, c0
, sizeof(lod
));
1809 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1811 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
,
1812 samp
->faces
[j
], &rgba
[0][j
]);
1815 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
,
1816 samp
->faces
[j
], &rgba
[0][j
]);
1823 mip_filter_none_no_filter_select(struct tgsi_sampler
*tgsi_sampler
,
1824 const float s
[TGSI_QUAD_SIZE
],
1825 const float t
[TGSI_QUAD_SIZE
],
1826 const float p
[TGSI_QUAD_SIZE
],
1827 const float c0
[TGSI_QUAD_SIZE
],
1828 const float c1
[TGSI_QUAD_SIZE
],
1829 enum tgsi_sampler_control control
,
1830 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1832 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1835 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++)
1836 samp
->mag_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], samp
->view
->u
.tex
.first_level
,
1837 samp
->faces
[j
], &rgba
[0][j
]);
1841 /* For anisotropic filtering */
1842 #define WEIGHT_LUT_SIZE 1024
1844 static float *weightLut
= NULL
;
1847 * Creates the look-up table used to speed-up EWA sampling
1850 create_filter_table(void)
1854 weightLut
= (float *) MALLOC(WEIGHT_LUT_SIZE
* sizeof(float));
1856 for (i
= 0; i
< WEIGHT_LUT_SIZE
; ++i
) {
1858 float r2
= (float) i
/ (float) (WEIGHT_LUT_SIZE
- 1);
1859 float weight
= (float) exp(-alpha
* r2
);
1860 weightLut
[i
] = weight
;
1867 * Elliptical weighted average (EWA) filter for producing high quality
1868 * anisotropic filtered results.
1869 * Based on the Higher Quality Elliptical Weighted Average Filter
1870 * published by Paul S. Heckbert in his Master's Thesis
1871 * "Fundamentals of Texture Mapping and Image Warping" (1989)
1874 img_filter_2d_ewa(struct tgsi_sampler
*tgsi_sampler
,
1875 const float s
[TGSI_QUAD_SIZE
],
1876 const float t
[TGSI_QUAD_SIZE
],
1877 const float p
[TGSI_QUAD_SIZE
],
1879 const float dudx
, const float dvdx
,
1880 const float dudy
, const float dvdy
,
1881 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1883 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
1884 const struct pipe_resource
*texture
= samp
->view
->texture
;
1886 // ??? Won't the image filters blow up if level is negative?
1887 unsigned level0
= level
> 0 ? level
: 0;
1888 float scaling
= 1.0 / (1 << level0
);
1889 int width
= u_minify(texture
->width0
, level0
);
1890 int height
= u_minify(texture
->height0
, level0
);
1892 float ux
= dudx
* scaling
;
1893 float vx
= dvdx
* scaling
;
1894 float uy
= dudy
* scaling
;
1895 float vy
= dvdy
* scaling
;
1897 /* compute ellipse coefficients to bound the region:
1898 * A*x*x + B*x*y + C*y*y = F.
1900 float A
= vx
*vx
+vy
*vy
+1;
1901 float B
= -2*(ux
*vx
+uy
*vy
);
1902 float C
= ux
*ux
+uy
*uy
+1;
1903 float F
= A
*C
-B
*B
/4.0;
1905 /* check if it is an ellipse */
1906 /* ASSERT(F > 0.0); */
1908 /* Compute the ellipse's (u,v) bounding box in texture space */
1909 float d
= -B
*B
+4.0*C
*A
;
1910 float box_u
= 2.0 / d
* sqrt(d
*C
*F
); /* box_u -> half of bbox with */
1911 float box_v
= 2.0 / d
* sqrt(A
*d
*F
); /* box_v -> half of bbox height */
1913 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
1914 float s_buffer
[TGSI_QUAD_SIZE
];
1915 float t_buffer
[TGSI_QUAD_SIZE
];
1916 float weight_buffer
[TGSI_QUAD_SIZE
];
1917 unsigned buffer_next
;
1919 float den
; /* = 0.0F; */
1921 float U
; /* = u0 - tex_u; */
1924 /* Scale ellipse formula to directly index the Filter Lookup Table.
1925 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
1927 double formScale
= (double) (WEIGHT_LUT_SIZE
- 1) / F
;
1931 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
1933 /* For each quad, the du and dx values are the same and so the ellipse is
1934 * also the same. Note that texel/image access can only be performed using
1935 * a quad, i.e. it is not possible to get the pixel value for a single
1936 * tex coord. In order to have a better performance, the access is buffered
1937 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
1938 * full, then the pixel values are read from the image.
1942 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
1943 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
1944 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
1945 * value, q, is less than F, we're inside the ellipse
1947 float tex_u
= -0.5F
+ s
[j
] * texture
->width0
* scaling
;
1948 float tex_v
= -0.5F
+ t
[j
] * texture
->height0
* scaling
;
1950 int u0
= (int) floorf(tex_u
- box_u
);
1951 int u1
= (int) ceilf(tex_u
+ box_u
);
1952 int v0
= (int) floorf(tex_v
- box_v
);
1953 int v1
= (int) ceilf(tex_v
+ box_v
);
1955 float num
[4] = {0.0F
, 0.0F
, 0.0F
, 0.0F
};
1959 for (v
= v0
; v
<= v1
; ++v
) {
1960 float V
= v
- tex_v
;
1961 float dq
= A
* (2 * U
+ 1) + B
* V
;
1962 float q
= (C
* V
+ B
* U
) * V
+ A
* U
* U
;
1965 for (u
= u0
; u
<= u1
; ++u
) {
1966 /* Note that the ellipse has been pre-scaled so F =
1967 * WEIGHT_LUT_SIZE - 1
1969 if (q
< WEIGHT_LUT_SIZE
) {
1970 /* as a LUT is used, q must never be negative;
1971 * should not happen, though
1973 const int qClamped
= q
>= 0.0F
? q
: 0;
1974 float weight
= weightLut
[qClamped
];
1976 weight_buffer
[buffer_next
] = weight
;
1977 s_buffer
[buffer_next
] = u
/ ((float) width
);
1978 t_buffer
[buffer_next
] = v
/ ((float) height
);
1981 if (buffer_next
== TGSI_QUAD_SIZE
) {
1982 /* 4 texel coords are in the buffer -> read it now */
1984 /* it is assumed that samp->min_img_filter is set to
1985 * img_filter_2d_nearest or one of the
1986 * accelerated img_filter_2d_nearest_XXX functions.
1988 for (jj
= 0; jj
< buffer_next
; jj
++) {
1989 samp
->min_img_filter(tgsi_sampler
, s_buffer
[jj
], t_buffer
[jj
], p
[jj
],
1990 level
, samp
->faces
[j
], &rgba_temp
[0][jj
]);
1991 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
1992 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
1993 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
1994 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2007 /* if the tex coord buffer contains unread values, we will read
2010 if (buffer_next
> 0) {
2012 /* it is assumed that samp->min_img_filter is set to
2013 * img_filter_2d_nearest or one of the
2014 * accelerated img_filter_2d_nearest_XXX functions.
2016 for (jj
= 0; jj
< buffer_next
; jj
++) {
2017 samp
->min_img_filter(tgsi_sampler
, s_buffer
[jj
], t_buffer
[jj
], p
[jj
], level
,
2018 samp
->faces
[j
], &rgba_temp
[0][jj
]);
2019 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2020 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2021 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2022 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2027 /* Reaching this place would mean that no pixels intersected
2028 * the ellipse. This should never happen because the filter
2029 * we use always intersects at least one pixel.
2036 /* not enough pixels in resampling, resort to direct interpolation */
2037 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level
, samp
->faces
[j
],
2040 num
[0] = rgba_temp
[0][j
];
2041 num
[1] = rgba_temp
[1][j
];
2042 num
[2] = rgba_temp
[2][j
];
2043 num
[3] = rgba_temp
[3][j
];
2046 rgba
[0][j
] = num
[0] / den
;
2047 rgba
[1][j
] = num
[1] / den
;
2048 rgba
[2][j
] = num
[2] / den
;
2049 rgba
[3][j
] = num
[3] / den
;
2055 * Sample 2D texture using an anisotropic filter.
2058 mip_filter_linear_aniso(struct tgsi_sampler
*tgsi_sampler
,
2059 const float s
[TGSI_QUAD_SIZE
],
2060 const float t
[TGSI_QUAD_SIZE
],
2061 const float p
[TGSI_QUAD_SIZE
],
2062 const float c0
[TGSI_QUAD_SIZE
],
2063 const float c1
[TGSI_QUAD_SIZE
],
2064 enum tgsi_sampler_control control
,
2065 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2067 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2068 const struct pipe_resource
*texture
= samp
->view
->texture
;
2071 float lod
[TGSI_QUAD_SIZE
];
2073 float s_to_u
= u_minify(texture
->width0
, samp
->view
->u
.tex
.first_level
);
2074 float t_to_v
= u_minify(texture
->height0
, samp
->view
->u
.tex
.first_level
);
2075 float dudx
= (s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2076 float dudy
= (s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2077 float dvdx
= (t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2078 float dvdy
= (t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2080 if (control
== tgsi_sampler_lod_bias
) {
2081 /* note: instead of working with Px and Py, we will use the
2082 * squared length instead, to avoid sqrt.
2084 float Px2
= dudx
* dudx
+ dvdx
* dvdx
;
2085 float Py2
= dudy
* dudy
+ dvdy
* dvdy
;
2090 const float maxEccentricity
= samp
->sampler
->max_anisotropy
* samp
->sampler
->max_anisotropy
;
2101 /* if the eccentricity of the ellipse is too big, scale up the shorter
2102 * of the two vectors to limit the maximum amount of work per pixel
2105 if (e
> maxEccentricity
) {
2106 /* float s=e / maxEccentricity;
2110 Pmin2
= Pmax2
/ maxEccentricity
;
2113 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2114 * this since 0.5*log(x) = log(sqrt(x))
2116 lambda
= 0.5F
* util_fast_log2(Pmin2
) + samp
->sampler
->lod_bias
;
2117 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
2120 assert(control
== tgsi_sampler_lod_explicit
);
2122 memcpy(lod
, c0
, sizeof(lod
));
2125 /* XXX: Take into account all lod values.
2128 level0
= samp
->view
->u
.tex
.first_level
+ (int)lambda
;
2130 /* If the ellipse covers the whole image, we can
2131 * simply return the average of the whole image.
2133 if (level0
>= (int) texture
->last_level
) {
2135 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++)
2136 samp
->min_img_filter(tgsi_sampler
, s
[j
], t
[j
], p
[j
], texture
->last_level
,
2137 samp
->faces
[j
], &rgba
[0][j
]);
2140 /* don't bother interpolating between multiple LODs; it doesn't
2141 * seem to be worth the extra running time.
2143 img_filter_2d_ewa(tgsi_sampler
, s
, t
, p
, level0
,
2144 dudx
, dvdx
, dudy
, dvdy
, rgba
);
2148 print_sample_4(__FUNCTION__
, rgba
);
2154 * Specialized version of mip_filter_linear with hard-wired calls to
2155 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2158 mip_filter_linear_2d_linear_repeat_POT(
2159 struct tgsi_sampler
*tgsi_sampler
,
2160 const float s
[TGSI_QUAD_SIZE
],
2161 const float t
[TGSI_QUAD_SIZE
],
2162 const float p
[TGSI_QUAD_SIZE
],
2163 const float c0
[TGSI_QUAD_SIZE
],
2164 const float c1
[TGSI_QUAD_SIZE
],
2165 enum tgsi_sampler_control control
,
2166 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2168 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2169 const struct pipe_resource
*texture
= samp
->view
->texture
;
2172 float lod
[TGSI_QUAD_SIZE
];
2174 if (control
== tgsi_sampler_lod_bias
) {
2175 lambda
= samp
->compute_lambda(samp
, s
, t
, p
) + samp
->sampler
->lod_bias
;
2176 compute_lod(samp
->sampler
, lambda
, c0
, lod
);
2178 assert(control
== tgsi_sampler_lod_explicit
);
2180 memcpy(lod
, c0
, sizeof(lod
));
2183 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2184 int level0
= samp
->view
->u
.tex
.first_level
+ (int)lod
[j
];
2186 /* Catches both negative and large values of level0:
2188 if ((unsigned)level0
>= texture
->last_level
) {
2190 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
],
2191 samp
->view
->u
.tex
.first_level
,
2192 samp
->faces
[j
], &rgba
[0][j
]);
2194 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
],
2195 samp
->view
->texture
->last_level
,
2196 samp
->faces
[j
], &rgba
[0][j
]);
2200 float levelBlend
= frac(lod
[j
]);
2201 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2204 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
,
2205 samp
->faces
[j
], &rgbax
[0][0]);
2206 img_filter_2d_linear_repeat_POT(tgsi_sampler
, s
[j
], t
[j
], p
[j
], level0
+1,
2207 samp
->faces
[j
], &rgbax
[0][1]);
2209 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
2210 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2215 print_sample_4(__FUNCTION__
, rgba
);
2221 * Do shadow/depth comparisons.
2224 sample_compare(struct tgsi_sampler
*tgsi_sampler
,
2225 const float s
[TGSI_QUAD_SIZE
],
2226 const float t
[TGSI_QUAD_SIZE
],
2227 const float p
[TGSI_QUAD_SIZE
],
2228 const float c0
[TGSI_QUAD_SIZE
],
2229 const float c1
[TGSI_QUAD_SIZE
],
2230 enum tgsi_sampler_control control
,
2231 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2233 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2234 const struct pipe_sampler_state
*sampler
= samp
->sampler
;
2235 int j
, k0
, k1
, k2
, k3
;
2237 float pc0
, pc1
, pc2
, pc3
;
2239 samp
->mip_filter(tgsi_sampler
, s
, t
, p
, c0
, c1
, control
, rgba
);
2242 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2243 * for 2D Array texture we need to use the 'c0' (aka Q).
2244 * When we sampled the depth texture, the depth value was put into all
2245 * RGBA channels. We look at the red channel here.
2248 if (samp
->view
->texture
->target
== PIPE_TEXTURE_2D_ARRAY
||
2249 samp
->view
->texture
->target
== PIPE_TEXTURE_CUBE
) {
2250 pc0
= CLAMP(c0
[0], 0.0F
, 1.0F
);
2251 pc1
= CLAMP(c0
[1], 0.0F
, 1.0F
);
2252 pc2
= CLAMP(c0
[2], 0.0F
, 1.0F
);
2253 pc3
= CLAMP(c0
[3], 0.0F
, 1.0F
);
2254 } else if (samp
->view
->texture
->target
== PIPE_TEXTURE_CUBE_ARRAY
) {
2255 pc0
= CLAMP(c1
[0], 0.0F
, 1.0F
);
2256 pc1
= CLAMP(c1
[1], 0.0F
, 1.0F
);
2257 pc2
= CLAMP(c1
[2], 0.0F
, 1.0F
);
2258 pc3
= CLAMP(c1
[3], 0.0F
, 1.0F
);
2260 pc0
= CLAMP(p
[0], 0.0F
, 1.0F
);
2261 pc1
= CLAMP(p
[1], 0.0F
, 1.0F
);
2262 pc2
= CLAMP(p
[2], 0.0F
, 1.0F
);
2263 pc3
= CLAMP(p
[3], 0.0F
, 1.0F
);
2265 /* compare four texcoords vs. four texture samples */
2266 switch (sampler
->compare_func
) {
2267 case PIPE_FUNC_LESS
:
2268 k0
= pc0
< rgba
[0][0];
2269 k1
= pc1
< rgba
[0][1];
2270 k2
= pc2
< rgba
[0][2];
2271 k3
= pc3
< rgba
[0][3];
2273 case PIPE_FUNC_LEQUAL
:
2274 k0
= pc0
<= rgba
[0][0];
2275 k1
= pc1
<= rgba
[0][1];
2276 k2
= pc2
<= rgba
[0][2];
2277 k3
= pc3
<= rgba
[0][3];
2279 case PIPE_FUNC_GREATER
:
2280 k0
= pc0
> rgba
[0][0];
2281 k1
= pc1
> rgba
[0][1];
2282 k2
= pc2
> rgba
[0][2];
2283 k3
= pc3
> rgba
[0][3];
2285 case PIPE_FUNC_GEQUAL
:
2286 k0
= pc0
>= rgba
[0][0];
2287 k1
= pc1
>= rgba
[0][1];
2288 k2
= pc2
>= rgba
[0][2];
2289 k3
= pc3
>= rgba
[0][3];
2291 case PIPE_FUNC_EQUAL
:
2292 k0
= pc0
== rgba
[0][0];
2293 k1
= pc1
== rgba
[0][1];
2294 k2
= pc2
== rgba
[0][2];
2295 k3
= pc3
== rgba
[0][3];
2297 case PIPE_FUNC_NOTEQUAL
:
2298 k0
= pc0
!= rgba
[0][0];
2299 k1
= pc1
!= rgba
[0][1];
2300 k2
= pc2
!= rgba
[0][2];
2301 k3
= pc3
!= rgba
[0][3];
2303 case PIPE_FUNC_ALWAYS
:
2304 k0
= k1
= k2
= k3
= 1;
2306 case PIPE_FUNC_NEVER
:
2307 k0
= k1
= k2
= k3
= 0;
2310 k0
= k1
= k2
= k3
= 0;
2315 if (sampler
->mag_img_filter
== PIPE_TEX_FILTER_LINEAR
) {
2316 /* convert four pass/fail values to an intensity in [0,1] */
2317 val
= 0.25F
* (k0
+ k1
+ k2
+ k3
);
2319 /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
2320 for (j
= 0; j
< 4; j
++) {
2321 rgba
[0][j
] = rgba
[1][j
] = rgba
[2][j
] = val
;
2325 for (j
= 0; j
< 4; j
++) {
2336 * Use 3D texcoords to choose a cube face, then sample the 2D cube faces.
2337 * Put face info into the sampler faces[] array.
2340 sample_cube(struct tgsi_sampler
*tgsi_sampler
,
2341 const float s
[TGSI_QUAD_SIZE
],
2342 const float t
[TGSI_QUAD_SIZE
],
2343 const float p
[TGSI_QUAD_SIZE
],
2344 const float c0
[TGSI_QUAD_SIZE
],
2345 const float c1
[TGSI_QUAD_SIZE
],
2346 enum tgsi_sampler_control control
,
2347 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2349 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2351 float ssss
[4], tttt
[4];
2353 /* Not actually used, but the intermediate steps that do the
2354 * dereferencing don't know it.
2356 static float pppp
[4] = { 0, 0, 0, 0 };
2364 direction target sc tc ma
2365 ---------- ------------------------------- --- --- ---
2366 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
2367 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
2368 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
2369 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
2370 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
2371 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
2374 /* Choose the cube face and compute new s/t coords for the 2D face.
2376 * Use the same cube face for all four pixels in the quad.
2378 * This isn't ideal, but if we want to use a different cube face
2379 * per pixel in the quad, we'd have to also compute the per-face
2380 * LOD here too. That's because the four post-face-selection
2381 * texcoords are no longer related to each other (they're
2382 * per-face!) so we can't use subtraction to compute the partial
2383 * deriviates to compute the LOD. Doing so (near cube edges
2384 * anyway) gives us pretty much random values.
2387 /* use the average of the four pixel's texcoords to choose the face */
2388 const float rx
= 0.25F
* (s
[0] + s
[1] + s
[2] + s
[3]);
2389 const float ry
= 0.25F
* (t
[0] + t
[1] + t
[2] + t
[3]);
2390 const float rz
= 0.25F
* (p
[0] + p
[1] + p
[2] + p
[3]);
2391 const float arx
= fabsf(rx
), ary
= fabsf(ry
), arz
= fabsf(rz
);
2393 if (arx
>= ary
&& arx
>= arz
) {
2394 float sign
= (rx
>= 0.0F
) ? 1.0F
: -1.0F
;
2395 uint face
= (rx
>= 0.0F
) ? PIPE_TEX_FACE_POS_X
: PIPE_TEX_FACE_NEG_X
;
2396 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2397 const float ima
= -0.5F
/ fabsf(s
[j
]);
2398 ssss
[j
] = sign
* p
[j
] * ima
+ 0.5F
;
2399 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
2400 samp
->faces
[j
] = face
;
2403 else if (ary
>= arx
&& ary
>= arz
) {
2404 float sign
= (ry
>= 0.0F
) ? 1.0F
: -1.0F
;
2405 uint face
= (ry
>= 0.0F
) ? PIPE_TEX_FACE_POS_Y
: PIPE_TEX_FACE_NEG_Y
;
2406 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2407 const float ima
= -0.5F
/ fabsf(t
[j
]);
2408 ssss
[j
] = -s
[j
] * ima
+ 0.5F
;
2409 tttt
[j
] = sign
* -p
[j
] * ima
+ 0.5F
;
2410 samp
->faces
[j
] = face
;
2414 float sign
= (rz
>= 0.0F
) ? 1.0F
: -1.0F
;
2415 uint face
= (rz
>= 0.0F
) ? PIPE_TEX_FACE_POS_Z
: PIPE_TEX_FACE_NEG_Z
;
2416 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2417 const float ima
= -0.5F
/ fabsf(p
[j
]);
2418 ssss
[j
] = sign
* -s
[j
] * ima
+ 0.5F
;
2419 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
2420 samp
->faces
[j
] = face
;
2425 /* In our little pipeline, the compare stage is next. If compare
2426 * is not active, this will point somewhere deeper into the
2427 * pipeline, eg. to mip_filter or even img_filter.
2429 samp
->compare(tgsi_sampler
, ssss
, tttt
, pppp
, c0
, c1
, control
, rgba
);
2434 do_swizzling(const struct sp_sampler_variant
*samp
,
2435 float in
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
2436 float out
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2439 const unsigned swizzle_r
= samp
->key
.bits
.swizzle_r
;
2440 const unsigned swizzle_g
= samp
->key
.bits
.swizzle_g
;
2441 const unsigned swizzle_b
= samp
->key
.bits
.swizzle_b
;
2442 const unsigned swizzle_a
= samp
->key
.bits
.swizzle_a
;
2444 switch (swizzle_r
) {
2445 case PIPE_SWIZZLE_ZERO
:
2446 for (j
= 0; j
< 4; j
++)
2449 case PIPE_SWIZZLE_ONE
:
2450 for (j
= 0; j
< 4; j
++)
2454 assert(swizzle_r
< 4);
2455 for (j
= 0; j
< 4; j
++)
2456 out
[0][j
] = in
[swizzle_r
][j
];
2459 switch (swizzle_g
) {
2460 case PIPE_SWIZZLE_ZERO
:
2461 for (j
= 0; j
< 4; j
++)
2464 case PIPE_SWIZZLE_ONE
:
2465 for (j
= 0; j
< 4; j
++)
2469 assert(swizzle_g
< 4);
2470 for (j
= 0; j
< 4; j
++)
2471 out
[1][j
] = in
[swizzle_g
][j
];
2474 switch (swizzle_b
) {
2475 case PIPE_SWIZZLE_ZERO
:
2476 for (j
= 0; j
< 4; j
++)
2479 case PIPE_SWIZZLE_ONE
:
2480 for (j
= 0; j
< 4; j
++)
2484 assert(swizzle_b
< 4);
2485 for (j
= 0; j
< 4; j
++)
2486 out
[2][j
] = in
[swizzle_b
][j
];
2489 switch (swizzle_a
) {
2490 case PIPE_SWIZZLE_ZERO
:
2491 for (j
= 0; j
< 4; j
++)
2494 case PIPE_SWIZZLE_ONE
:
2495 for (j
= 0; j
< 4; j
++)
2499 assert(swizzle_a
< 4);
2500 for (j
= 0; j
< 4; j
++)
2501 out
[3][j
] = in
[swizzle_a
][j
];
2507 sample_swizzle(struct tgsi_sampler
*tgsi_sampler
,
2508 const float s
[TGSI_QUAD_SIZE
],
2509 const float t
[TGSI_QUAD_SIZE
],
2510 const float p
[TGSI_QUAD_SIZE
],
2511 const float c0
[TGSI_QUAD_SIZE
],
2512 const float c1
[TGSI_QUAD_SIZE
],
2513 enum tgsi_sampler_control control
,
2514 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2516 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2517 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2519 samp
->sample_target(tgsi_sampler
, s
, t
, p
, c0
, c1
, control
, rgba_temp
);
2521 do_swizzling(samp
, rgba_temp
, rgba
);
2525 static wrap_nearest_func
2526 get_nearest_unorm_wrap(unsigned mode
)
2529 case PIPE_TEX_WRAP_CLAMP
:
2530 return wrap_nearest_unorm_clamp
;
2531 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2532 return wrap_nearest_unorm_clamp_to_edge
;
2533 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2534 return wrap_nearest_unorm_clamp_to_border
;
2537 return wrap_nearest_unorm_clamp
;
2542 static wrap_nearest_func
2543 get_nearest_wrap(unsigned mode
)
2546 case PIPE_TEX_WRAP_REPEAT
:
2547 return wrap_nearest_repeat
;
2548 case PIPE_TEX_WRAP_CLAMP
:
2549 return wrap_nearest_clamp
;
2550 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2551 return wrap_nearest_clamp_to_edge
;
2552 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2553 return wrap_nearest_clamp_to_border
;
2554 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2555 return wrap_nearest_mirror_repeat
;
2556 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2557 return wrap_nearest_mirror_clamp
;
2558 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2559 return wrap_nearest_mirror_clamp_to_edge
;
2560 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2561 return wrap_nearest_mirror_clamp_to_border
;
2564 return wrap_nearest_repeat
;
2569 static wrap_linear_func
2570 get_linear_unorm_wrap(unsigned mode
)
2573 case PIPE_TEX_WRAP_CLAMP
:
2574 return wrap_linear_unorm_clamp
;
2575 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2576 return wrap_linear_unorm_clamp_to_edge
;
2577 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2578 return wrap_linear_unorm_clamp_to_border
;
2581 return wrap_linear_unorm_clamp
;
2586 static wrap_linear_func
2587 get_linear_wrap(unsigned mode
)
2590 case PIPE_TEX_WRAP_REPEAT
:
2591 return wrap_linear_repeat
;
2592 case PIPE_TEX_WRAP_CLAMP
:
2593 return wrap_linear_clamp
;
2594 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2595 return wrap_linear_clamp_to_edge
;
2596 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2597 return wrap_linear_clamp_to_border
;
2598 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2599 return wrap_linear_mirror_repeat
;
2600 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2601 return wrap_linear_mirror_clamp
;
2602 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2603 return wrap_linear_mirror_clamp_to_edge
;
2604 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2605 return wrap_linear_mirror_clamp_to_border
;
2608 return wrap_linear_repeat
;
2614 * Is swizzling needed for the given state key?
2617 any_swizzle(union sp_sampler_key key
)
2619 return (key
.bits
.swizzle_r
!= PIPE_SWIZZLE_RED
||
2620 key
.bits
.swizzle_g
!= PIPE_SWIZZLE_GREEN
||
2621 key
.bits
.swizzle_b
!= PIPE_SWIZZLE_BLUE
||
2622 key
.bits
.swizzle_a
!= PIPE_SWIZZLE_ALPHA
);
2626 static compute_lambda_func
2627 get_lambda_func(const union sp_sampler_key key
)
2629 if (key
.bits
.processor
== TGSI_PROCESSOR_VERTEX
)
2630 return compute_lambda_vert
;
2632 switch (key
.bits
.target
) {
2634 case PIPE_TEXTURE_1D
:
2635 case PIPE_TEXTURE_1D_ARRAY
:
2636 return compute_lambda_1d
;
2637 case PIPE_TEXTURE_2D
:
2638 case PIPE_TEXTURE_2D_ARRAY
:
2639 case PIPE_TEXTURE_RECT
:
2640 case PIPE_TEXTURE_CUBE
:
2641 case PIPE_TEXTURE_CUBE_ARRAY
:
2642 return compute_lambda_2d
;
2643 case PIPE_TEXTURE_3D
:
2644 return compute_lambda_3d
;
2647 return compute_lambda_1d
;
2652 static img_filter_func
2653 get_img_filter(const union sp_sampler_key key
,
2655 const struct pipe_sampler_state
*sampler
)
2657 switch (key
.bits
.target
) {
2659 case PIPE_TEXTURE_1D
:
2660 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2661 return img_filter_1d_nearest
;
2663 return img_filter_1d_linear
;
2665 case PIPE_TEXTURE_1D_ARRAY
:
2666 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2667 return img_filter_1d_array_nearest
;
2669 return img_filter_1d_array_linear
;
2671 case PIPE_TEXTURE_2D
:
2672 case PIPE_TEXTURE_RECT
:
2673 /* Try for fast path:
2675 if (key
.bits
.is_pot
&&
2676 sampler
->wrap_s
== sampler
->wrap_t
&&
2677 sampler
->normalized_coords
)
2679 switch (sampler
->wrap_s
) {
2680 case PIPE_TEX_WRAP_REPEAT
:
2682 case PIPE_TEX_FILTER_NEAREST
:
2683 return img_filter_2d_nearest_repeat_POT
;
2684 case PIPE_TEX_FILTER_LINEAR
:
2685 return img_filter_2d_linear_repeat_POT
;
2690 case PIPE_TEX_WRAP_CLAMP
:
2692 case PIPE_TEX_FILTER_NEAREST
:
2693 return img_filter_2d_nearest_clamp_POT
;
2699 /* Otherwise use default versions:
2701 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2702 return img_filter_2d_nearest
;
2704 return img_filter_2d_linear
;
2706 case PIPE_TEXTURE_2D_ARRAY
:
2707 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2708 return img_filter_2d_array_nearest
;
2710 return img_filter_2d_array_linear
;
2712 case PIPE_TEXTURE_CUBE
:
2713 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2714 return img_filter_cube_nearest
;
2716 return img_filter_cube_linear
;
2718 case PIPE_TEXTURE_CUBE_ARRAY
:
2719 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2720 return img_filter_cube_array_nearest
;
2722 return img_filter_cube_array_linear
;
2724 case PIPE_TEXTURE_3D
:
2725 if (filter
== PIPE_TEX_FILTER_NEAREST
)
2726 return img_filter_3d_nearest
;
2728 return img_filter_3d_linear
;
2732 return img_filter_1d_nearest
;
2738 * Bind the given texture object and texture cache to the sampler variant.
2741 sp_sampler_variant_bind_view( struct sp_sampler_variant
*samp
,
2742 struct softpipe_tex_tile_cache
*tex_cache
,
2743 const struct pipe_sampler_view
*view
)
2745 const struct pipe_resource
*texture
= view
->texture
;
2748 samp
->cache
= tex_cache
;
2749 samp
->xpot
= util_logbase2( texture
->width0
);
2750 samp
->ypot
= util_logbase2( texture
->height0
);
2755 sp_sampler_variant_destroy( struct sp_sampler_variant
*samp
)
2762 sample_get_dims(struct tgsi_sampler
*tgsi_sampler
, int level
,
2765 struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2766 const struct pipe_sampler_view
*view
= samp
->view
;
2767 const struct pipe_resource
*texture
= view
->texture
;
2769 /* undefined according to EXT_gpu_program */
2770 level
+= view
->u
.tex
.first_level
;
2771 if (level
> view
->u
.tex
.last_level
)
2774 dims
[0] = u_minify(texture
->width0
, level
);
2776 switch(texture
->target
) {
2777 case PIPE_TEXTURE_1D_ARRAY
:
2778 dims
[1] = texture
->array_size
;
2780 case PIPE_TEXTURE_1D
:
2782 case PIPE_TEXTURE_2D_ARRAY
:
2783 dims
[2] = texture
->array_size
;
2785 case PIPE_TEXTURE_2D
:
2786 case PIPE_TEXTURE_CUBE
:
2787 case PIPE_TEXTURE_RECT
:
2788 dims
[1] = u_minify(texture
->height0
, level
);
2790 case PIPE_TEXTURE_3D
:
2791 dims
[1] = u_minify(texture
->height0
, level
);
2792 dims
[2] = u_minify(texture
->depth0
, level
);
2794 case PIPE_TEXTURE_CUBE_ARRAY
:
2795 dims
[1] = u_minify(texture
->height0
, level
);
2796 dims
[2] = texture
->array_size
/ 6;
2799 dims
[0] /= util_format_get_blocksize(view
->format
);
2802 assert(!"unexpected texture target in sample_get_dims()");
2808 * This function is only used for getting unfiltered texels via the
2809 * TXF opcode. The GL spec says that out-of-bounds texel fetches
2810 * produce undefined results. Instead of crashing, lets just clamp
2811 * coords to the texture image size.
2814 sample_get_texels(struct tgsi_sampler
*tgsi_sampler
,
2815 const int v_i
[TGSI_QUAD_SIZE
],
2816 const int v_j
[TGSI_QUAD_SIZE
],
2817 const int v_k
[TGSI_QUAD_SIZE
],
2818 const int lod
[TGSI_QUAD_SIZE
],
2819 const int8_t offset
[3],
2820 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2822 const struct sp_sampler_variant
*samp
= sp_sampler_variant(tgsi_sampler
);
2823 union tex_tile_address addr
;
2824 const struct pipe_resource
*texture
= samp
->view
->texture
;
2827 const bool need_swizzle
= any_swizzle(samp
->key
);
2828 int width
, height
, depth
, layers
;
2831 /* TODO write a better test for LOD */
2832 addr
.bits
.level
= lod
[0];
2834 width
= u_minify(texture
->width0
, addr
.bits
.level
);
2835 height
= u_minify(texture
->height0
, addr
.bits
.level
);
2836 depth
= u_minify(texture
->depth0
, addr
.bits
.level
);
2837 layers
= texture
->array_size
;
2839 switch(texture
->target
) {
2841 case PIPE_TEXTURE_1D
:
2842 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2843 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2844 tx
= get_texel_2d(samp
, addr
, x
, 0);
2845 for (c
= 0; c
< 4; c
++) {
2850 case PIPE_TEXTURE_1D_ARRAY
:
2851 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2852 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2853 int y
= CLAMP(v_j
[j
], 0, layers
- 1);
2854 tx
= get_texel_1d_array(samp
, addr
, x
, y
);
2855 for (c
= 0; c
< 4; c
++) {
2860 case PIPE_TEXTURE_2D
:
2861 case PIPE_TEXTURE_RECT
:
2862 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2863 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2864 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
2865 tx
= get_texel_2d(samp
, addr
, x
, y
);
2866 for (c
= 0; c
< 4; c
++) {
2871 case PIPE_TEXTURE_2D_ARRAY
:
2872 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2873 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2874 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
2875 int layer
= CLAMP(v_k
[j
], 0, layers
- 1);
2876 tx
= get_texel_2d_array(samp
, addr
, x
, y
, layer
);
2877 for (c
= 0; c
< 4; c
++) {
2882 case PIPE_TEXTURE_3D
:
2883 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2884 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
2885 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
2886 int z
= CLAMP(v_k
[j
] + offset
[2], 0, depth
- 1);
2888 tx
= get_texel_3d(samp
, addr
, x
, y
, z
);
2889 for (c
= 0; c
< 4; c
++) {
2894 case PIPE_TEXTURE_CUBE
: /* TXF can't work on CUBE according to spec */
2896 assert(!"Unknown or CUBE texture type in TXF processing\n");
2901 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2902 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
2903 do_swizzling(samp
, rgba_temp
, rgba
);
2909 * Create a sampler variant for a given set of non-orthogonal state.
2911 struct sp_sampler_variant
*
2912 sp_create_sampler_variant( const struct pipe_sampler_state
*sampler
,
2913 const union sp_sampler_key key
)
2915 struct sp_sampler_variant
*samp
= CALLOC_STRUCT(sp_sampler_variant
);
2919 samp
->sampler
= sampler
;
2922 /* Note that (for instance) linear_texcoord_s and
2923 * nearest_texcoord_s may be active at the same time, if the
2924 * sampler min_img_filter differs from its mag_img_filter.
2926 if (sampler
->normalized_coords
) {
2927 samp
->linear_texcoord_s
= get_linear_wrap( sampler
->wrap_s
);
2928 samp
->linear_texcoord_t
= get_linear_wrap( sampler
->wrap_t
);
2929 samp
->linear_texcoord_p
= get_linear_wrap( sampler
->wrap_r
);
2931 samp
->nearest_texcoord_s
= get_nearest_wrap( sampler
->wrap_s
);
2932 samp
->nearest_texcoord_t
= get_nearest_wrap( sampler
->wrap_t
);
2933 samp
->nearest_texcoord_p
= get_nearest_wrap( sampler
->wrap_r
);
2936 samp
->linear_texcoord_s
= get_linear_unorm_wrap( sampler
->wrap_s
);
2937 samp
->linear_texcoord_t
= get_linear_unorm_wrap( sampler
->wrap_t
);
2938 samp
->linear_texcoord_p
= get_linear_unorm_wrap( sampler
->wrap_r
);
2940 samp
->nearest_texcoord_s
= get_nearest_unorm_wrap( sampler
->wrap_s
);
2941 samp
->nearest_texcoord_t
= get_nearest_unorm_wrap( sampler
->wrap_t
);
2942 samp
->nearest_texcoord_p
= get_nearest_unorm_wrap( sampler
->wrap_r
);
2945 samp
->compute_lambda
= get_lambda_func( key
);
2947 samp
->min_img_filter
= get_img_filter(key
, sampler
->min_img_filter
, sampler
);
2948 samp
->mag_img_filter
= get_img_filter(key
, sampler
->mag_img_filter
, sampler
);
2950 switch (sampler
->min_mip_filter
) {
2951 case PIPE_TEX_MIPFILTER_NONE
:
2952 if (sampler
->min_img_filter
== sampler
->mag_img_filter
)
2953 samp
->mip_filter
= mip_filter_none_no_filter_select
;
2955 samp
->mip_filter
= mip_filter_none
;
2958 case PIPE_TEX_MIPFILTER_NEAREST
:
2959 samp
->mip_filter
= mip_filter_nearest
;
2962 case PIPE_TEX_MIPFILTER_LINEAR
:
2963 if (key
.bits
.is_pot
&&
2964 key
.bits
.target
== PIPE_TEXTURE_2D
&&
2965 sampler
->min_img_filter
== sampler
->mag_img_filter
&&
2966 sampler
->normalized_coords
&&
2967 sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
&&
2968 sampler
->wrap_t
== PIPE_TEX_WRAP_REPEAT
&&
2969 sampler
->min_img_filter
== PIPE_TEX_FILTER_LINEAR
) {
2970 samp
->mip_filter
= mip_filter_linear_2d_linear_repeat_POT
;
2973 samp
->mip_filter
= mip_filter_linear
;
2976 /* Anisotropic filtering extension. */
2977 if (sampler
->max_anisotropy
> 1) {
2978 samp
->mip_filter
= mip_filter_linear_aniso
;
2980 /* Override min_img_filter:
2981 * min_img_filter needs to be set to NEAREST since we need to access
2982 * each texture pixel as it is and weight it later; using linear
2983 * filters will have incorrect results.
2984 * By setting the filter to NEAREST here, we can avoid calling the
2985 * generic img_filter_2d_nearest in the anisotropic filter function,
2986 * making it possible to use one of the accelerated implementations
2988 samp
->min_img_filter
= get_img_filter(key
, PIPE_TEX_FILTER_NEAREST
, sampler
);
2990 /* on first access create the lookup table containing the filter weights. */
2992 create_filter_table();
2999 if (sampler
->compare_mode
!= PIPE_TEX_COMPARE_NONE
) {
3000 samp
->compare
= sample_compare
;
3003 /* Skip compare operation by promoting the mip_filter function
3006 samp
->compare
= samp
->mip_filter
;
3009 if (key
.bits
.target
== PIPE_TEXTURE_CUBE
|| key
.bits
.target
== PIPE_TEXTURE_CUBE_ARRAY
) {
3010 samp
->sample_target
= sample_cube
;
3018 /* Skip cube face determination by promoting the compare
3021 samp
->sample_target
= samp
->compare
;
3024 if (any_swizzle(key
)) {
3025 samp
->base
.get_samples
= sample_swizzle
;
3028 samp
->base
.get_samples
= samp
->sample_target
;
3031 samp
->base
.get_dims
= sample_get_dims
;
3032 samp
->base
.get_texel
= sample_get_texels
;