1 /**************************************************************************
3 * Copyright 2007 VMware, Inc.
5 * Copyright 2008-2010 VMware, Inc. All rights reserved.
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
27 **************************************************************************/
37 #include "pipe/p_context.h"
38 #include "pipe/p_defines.h"
39 #include "pipe/p_shader_tokens.h"
40 #include "util/u_math.h"
41 #include "util/u_format.h"
42 #include "util/u_memory.h"
43 #include "util/u_inlines.h"
44 #include "sp_quad.h" /* only for #define QUAD_* tokens */
45 #include "sp_tex_sample.h"
46 #include "sp_texture.h"
47 #include "sp_tex_tile_cache.h"
50 /** Set to one to help debug texture sampling */
55 * Return fractional part of 'f'. Used for computing interpolation weights.
56 * Need to be careful with negative values.
57 * Note, if this function isn't perfect you'll sometimes see 1-pixel bands
58 * of improperly weighted linear-filtered textures.
59 * The tests/texwrap.c demo is a good test.
70 * Linear interpolation macro
73 lerp(float a
, float v0
, float v1
)
75 return v0
+ a
* (v1
- v0
);
80 * Do 2D/bilinear interpolation of float values.
81 * v00, v10, v01 and v11 are typically four texture samples in a square/box.
82 * a and b are the horizontal and vertical interpolants.
83 * It's important that this function is inlined when compiled with
84 * optimization! If we find that's not true on some systems, convert
88 lerp_2d(float a
, float b
,
89 float v00
, float v10
, float v01
, float v11
)
91 const float temp0
= lerp(a
, v00
, v10
);
92 const float temp1
= lerp(a
, v01
, v11
);
93 return lerp(b
, temp0
, temp1
);
98 * As above, but 3D interpolation of 8 values.
101 lerp_3d(float a
, float b
, float c
,
102 float v000
, float v100
, float v010
, float v110
,
103 float v001
, float v101
, float v011
, float v111
)
105 const float temp0
= lerp_2d(a
, b
, v000
, v100
, v010
, v110
);
106 const float temp1
= lerp_2d(a
, b
, v001
, v101
, v011
, v111
);
107 return lerp(c
, temp0
, temp1
);
113 * Compute coord % size for repeat wrap modes.
114 * Note that if coord is negative, coord % size doesn't give the right
115 * value. To avoid that problem we add a large multiple of the size
116 * (rather than using a conditional).
119 repeat(int coord
, unsigned size
)
121 return (coord
+ size
* 1024) % size
;
126 * Apply texture coord wrapping mode and return integer texture indexes
127 * for a vector of four texcoords (S or T or P).
128 * \param wrapMode PIPE_TEX_WRAP_x
129 * \param s the incoming texcoords
130 * \param size the texture image size
131 * \param icoord returns the integer texcoords
134 wrap_nearest_repeat(float s
, unsigned size
, int offset
, int *icoord
)
136 /* s limited to [0,1) */
137 /* i limited to [0,size-1] */
138 const int i
= util_ifloor(s
* size
);
139 *icoord
= repeat(i
+ offset
, size
);
144 wrap_nearest_clamp(float s
, unsigned size
, int offset
, int *icoord
)
146 /* s limited to [0,1] */
147 /* i limited to [0,size-1] */
155 *icoord
= util_ifloor(s
);
160 wrap_nearest_clamp_to_edge(float s
, unsigned size
, int offset
, int *icoord
)
162 /* s limited to [min,max] */
163 /* i limited to [0, size-1] */
164 const float min
= 0.5F
;
165 const float max
= (float)size
- 0.5F
;
175 *icoord
= util_ifloor(s
);
180 wrap_nearest_clamp_to_border(float s
, unsigned size
, int offset
, int *icoord
)
182 /* s limited to [min,max] */
183 /* i limited to [-1, size] */
184 const float min
= -0.5F
;
185 const float max
= size
+ 0.5F
;
194 *icoord
= util_ifloor(s
);
198 wrap_nearest_mirror_repeat(float s
, unsigned size
, int offset
, int *icoord
)
200 const float min
= 1.0F
/ (2.0F
* size
);
201 const float max
= 1.0F
- min
;
205 s
+= (float)offset
/ size
;
206 flr
= util_ifloor(s
);
215 *icoord
= util_ifloor(u
* size
);
220 wrap_nearest_mirror_clamp(float s
, unsigned size
, int offset
, int *icoord
)
222 /* s limited to [0,1] */
223 /* i limited to [0,size-1] */
224 const float u
= fabsf(s
* size
+ offset
);
230 *icoord
= util_ifloor(u
);
235 wrap_nearest_mirror_clamp_to_edge(float s
, unsigned size
, int offset
, int *icoord
)
237 /* s limited to [min,max] */
238 /* i limited to [0, size-1] */
239 const float min
= 0.5F
;
240 const float max
= (float)size
- 0.5F
;
241 const float u
= fabsf(s
* size
+ offset
);
248 *icoord
= util_ifloor(u
);
253 wrap_nearest_mirror_clamp_to_border(float s
, unsigned size
, int offset
, int *icoord
)
255 /* u limited to [-0.5, size-0.5] */
256 const float min
= -0.5F
;
257 const float max
= (float)size
+ 0.5F
;
258 const float u
= fabsf(s
* size
+ offset
);
265 *icoord
= util_ifloor(u
);
270 * Used to compute texel locations for linear sampling
271 * \param wrapMode PIPE_TEX_WRAP_x
272 * \param s the texcoord
273 * \param size the texture image size
274 * \param icoord0 returns first texture index
275 * \param icoord1 returns second texture index (usually icoord0 + 1)
276 * \param w returns blend factor/weight between texture indices
277 * \param icoord returns the computed integer texture coord
280 wrap_linear_repeat(float s
, unsigned size
, int offset
,
281 int *icoord0
, int *icoord1
, float *w
)
283 const float u
= s
* size
- 0.5F
;
284 *icoord0
= repeat(util_ifloor(u
) + offset
, size
);
285 *icoord1
= repeat(*icoord0
+ 1, size
);
291 wrap_linear_clamp(float s
, unsigned size
, int offset
,
292 int *icoord0
, int *icoord1
, float *w
)
294 const float u
= CLAMP(s
* size
+ offset
, 0.0F
, (float)size
) - 0.5f
;
296 *icoord0
= util_ifloor(u
);
297 *icoord1
= *icoord0
+ 1;
303 wrap_linear_clamp_to_edge(float s
, unsigned size
, int offset
,
304 int *icoord0
, int *icoord1
, float *w
)
306 const float u
= CLAMP(s
* size
+ offset
, 0.0F
, (float)size
) - 0.5f
;
307 *icoord0
= util_ifloor(u
);
308 *icoord1
= *icoord0
+ 1;
311 if (*icoord1
>= (int) size
)
318 wrap_linear_clamp_to_border(float s
, unsigned size
, int offset
,
319 int *icoord0
, int *icoord1
, float *w
)
321 const float min
= -0.5F
;
322 const float max
= (float)size
+ 0.5F
;
323 const float u
= CLAMP(s
* size
+ offset
, min
, max
) - 0.5f
;
324 *icoord0
= util_ifloor(u
);
325 *icoord1
= *icoord0
+ 1;
331 wrap_linear_mirror_repeat(float s
, unsigned size
, int offset
,
332 int *icoord0
, int *icoord1
, float *w
)
337 s
+= (float)offset
/ size
;
338 flr
= util_ifloor(s
);
343 *icoord0
= util_ifloor(u
);
344 *icoord1
= *icoord0
+ 1;
347 if (*icoord1
>= (int) size
)
354 wrap_linear_mirror_clamp(float s
, unsigned size
, int offset
,
355 int *icoord0
, int *icoord1
, float *w
)
357 float u
= fabsf(s
* size
+ offset
);
361 *icoord0
= util_ifloor(u
);
362 *icoord1
= *icoord0
+ 1;
368 wrap_linear_mirror_clamp_to_edge(float s
, unsigned size
, int offset
,
369 int *icoord0
, int *icoord1
, float *w
)
371 float u
= fabsf(s
* size
+ offset
);
375 *icoord0
= util_ifloor(u
);
376 *icoord1
= *icoord0
+ 1;
379 if (*icoord1
>= (int) size
)
386 wrap_linear_mirror_clamp_to_border(float s
, unsigned size
, int offset
,
387 int *icoord0
, int *icoord1
, float *w
)
389 const float min
= -0.5F
;
390 const float max
= size
+ 0.5F
;
391 const float t
= fabsf(s
* size
+ offset
);
392 const float u
= CLAMP(t
, min
, max
) - 0.5F
;
393 *icoord0
= util_ifloor(u
);
394 *icoord1
= *icoord0
+ 1;
400 * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
403 wrap_nearest_unorm_clamp(float s
, unsigned size
, int offset
, int *icoord
)
405 const int i
= util_ifloor(s
);
406 *icoord
= CLAMP(i
+ offset
, 0, (int) size
-1);
411 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
414 wrap_nearest_unorm_clamp_to_border(float s
, unsigned size
, int offset
, int *icoord
)
416 *icoord
= util_ifloor( CLAMP(s
+ offset
, -0.5F
, (float) size
+ 0.5F
) );
421 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
424 wrap_nearest_unorm_clamp_to_edge(float s
, unsigned size
, int offset
, int *icoord
)
426 *icoord
= util_ifloor( CLAMP(s
+ offset
, 0.5F
, (float) size
- 0.5F
) );
431 * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
434 wrap_linear_unorm_clamp(float s
, unsigned size
, int offset
,
435 int *icoord0
, int *icoord1
, float *w
)
437 /* Not exactly what the spec says, but it matches NVIDIA output */
438 const float u
= CLAMP(s
+ offset
- 0.5F
, 0.0f
, (float) size
- 1.0f
);
439 *icoord0
= util_ifloor(u
);
440 *icoord1
= *icoord0
+ 1;
446 * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
449 wrap_linear_unorm_clamp_to_border(float s
, unsigned size
, int offset
,
450 int *icoord0
, int *icoord1
, float *w
)
452 const float u
= CLAMP(s
+ offset
, -0.5F
, (float) size
+ 0.5F
) - 0.5F
;
453 *icoord0
= util_ifloor(u
);
454 *icoord1
= *icoord0
+ 1;
455 if (*icoord1
> (int) size
- 1)
462 * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
465 wrap_linear_unorm_clamp_to_edge(float s
, unsigned size
, int offset
,
466 int *icoord0
, int *icoord1
, float *w
)
468 const float u
= CLAMP(s
+ offset
, +0.5F
, (float) size
- 0.5F
) - 0.5F
;
469 *icoord0
= util_ifloor(u
);
470 *icoord1
= *icoord0
+ 1;
471 if (*icoord1
> (int) size
- 1)
478 * Do coordinate to array index conversion. For array textures.
481 coord_to_layer(float coord
, unsigned first_layer
, unsigned last_layer
)
483 const int c
= util_ifloor(coord
+ 0.5F
);
484 return CLAMP(c
, (int)first_layer
, (int)last_layer
);
488 compute_gradient_1d(const float s
[TGSI_QUAD_SIZE
],
489 const float t
[TGSI_QUAD_SIZE
],
490 const float p
[TGSI_QUAD_SIZE
],
491 float derivs
[3][2][TGSI_QUAD_SIZE
])
493 memset(derivs
, 0, 6 * TGSI_QUAD_SIZE
* sizeof(float));
494 derivs
[0][0][0] = s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
];
495 derivs
[0][1][0] = s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
];
499 compute_lambda_1d_explicit_gradients(const struct sp_sampler_view
*sview
,
500 const float derivs
[3][2][TGSI_QUAD_SIZE
],
503 const struct pipe_resource
*texture
= sview
->base
.texture
;
504 const float dsdx
= fabsf(derivs
[0][0][quad
]);
505 const float dsdy
= fabsf(derivs
[0][1][quad
]);
506 const float rho
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, sview
->base
.u
.tex
.first_level
);
507 return util_fast_log2(rho
);
512 * Examine the quad's texture coordinates to compute the partial
513 * derivatives w.r.t X and Y, then compute lambda (level of detail).
516 compute_lambda_1d(const struct sp_sampler_view
*sview
,
517 const float s
[TGSI_QUAD_SIZE
],
518 const float t
[TGSI_QUAD_SIZE
],
519 const float p
[TGSI_QUAD_SIZE
])
521 float derivs
[3][2][TGSI_QUAD_SIZE
];
522 compute_gradient_1d(s
, t
, p
, derivs
);
523 return compute_lambda_1d_explicit_gradients(sview
, derivs
, 0);
528 compute_gradient_2d(const float s
[TGSI_QUAD_SIZE
],
529 const float t
[TGSI_QUAD_SIZE
],
530 const float p
[TGSI_QUAD_SIZE
],
531 float derivs
[3][2][TGSI_QUAD_SIZE
])
533 memset(derivs
, 0, 6 * TGSI_QUAD_SIZE
* sizeof(float));
534 derivs
[0][0][0] = s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
];
535 derivs
[0][1][0] = s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
];
536 derivs
[1][0][0] = t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
];
537 derivs
[1][1][0] = t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
];
541 compute_lambda_2d_explicit_gradients(const struct sp_sampler_view
*sview
,
542 const float derivs
[3][2][TGSI_QUAD_SIZE
],
545 const struct pipe_resource
*texture
= sview
->base
.texture
;
546 const float dsdx
= fabsf(derivs
[0][0][quad
]);
547 const float dsdy
= fabsf(derivs
[0][1][quad
]);
548 const float dtdx
= fabsf(derivs
[1][0][quad
]);
549 const float dtdy
= fabsf(derivs
[1][1][quad
]);
550 const float maxx
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, sview
->base
.u
.tex
.first_level
);
551 const float maxy
= MAX2(dtdx
, dtdy
) * u_minify(texture
->height0
, sview
->base
.u
.tex
.first_level
);
552 const float rho
= MAX2(maxx
, maxy
);
553 return util_fast_log2(rho
);
558 compute_lambda_2d(const struct sp_sampler_view
*sview
,
559 const float s
[TGSI_QUAD_SIZE
],
560 const float t
[TGSI_QUAD_SIZE
],
561 const float p
[TGSI_QUAD_SIZE
])
563 float derivs
[3][2][TGSI_QUAD_SIZE
];
564 compute_gradient_2d(s
, t
, p
, derivs
);
565 return compute_lambda_2d_explicit_gradients(sview
, derivs
, 0);
570 compute_gradient_3d(const float s
[TGSI_QUAD_SIZE
],
571 const float t
[TGSI_QUAD_SIZE
],
572 const float p
[TGSI_QUAD_SIZE
],
573 float derivs
[3][2][TGSI_QUAD_SIZE
])
575 memset(derivs
, 0, 6 * TGSI_QUAD_SIZE
* sizeof(float));
576 derivs
[0][0][0] = fabsf(s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]);
577 derivs
[0][1][0] = fabsf(s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]);
578 derivs
[1][0][0] = fabsf(t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]);
579 derivs
[1][1][0] = fabsf(t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]);
580 derivs
[2][0][0] = fabsf(p
[QUAD_BOTTOM_RIGHT
] - p
[QUAD_BOTTOM_LEFT
]);
581 derivs
[2][1][0] = fabsf(p
[QUAD_TOP_LEFT
] - p
[QUAD_BOTTOM_LEFT
]);
585 compute_lambda_3d_explicit_gradients(const struct sp_sampler_view
*sview
,
586 const float derivs
[3][2][TGSI_QUAD_SIZE
],
589 const struct pipe_resource
*texture
= sview
->base
.texture
;
590 const float dsdx
= fabsf(derivs
[0][0][quad
]);
591 const float dsdy
= fabsf(derivs
[0][1][quad
]);
592 const float dtdx
= fabsf(derivs
[1][0][quad
]);
593 const float dtdy
= fabsf(derivs
[1][1][quad
]);
594 const float dpdx
= fabsf(derivs
[2][0][quad
]);
595 const float dpdy
= fabsf(derivs
[2][1][quad
]);
596 const float maxx
= MAX2(dsdx
, dsdy
) * u_minify(texture
->width0
, sview
->base
.u
.tex
.first_level
);
597 const float maxy
= MAX2(dtdx
, dtdy
) * u_minify(texture
->height0
, sview
->base
.u
.tex
.first_level
);
598 const float maxz
= MAX2(dpdx
, dpdy
) * u_minify(texture
->depth0
, sview
->base
.u
.tex
.first_level
);
599 const float rho
= MAX3(maxx
, maxy
, maxz
);
601 return util_fast_log2(rho
);
606 compute_lambda_3d(const struct sp_sampler_view
*sview
,
607 const float s
[TGSI_QUAD_SIZE
],
608 const float t
[TGSI_QUAD_SIZE
],
609 const float p
[TGSI_QUAD_SIZE
])
611 float derivs
[3][2][TGSI_QUAD_SIZE
];
612 compute_gradient_3d(s
, t
, p
, derivs
);
613 return compute_lambda_3d_explicit_gradients(sview
, derivs
, 0);
618 compute_lambda_cube_explicit_gradients(const struct sp_sampler_view
*sview
,
619 const float derivs
[3][2][TGSI_QUAD_SIZE
],
622 const struct pipe_resource
*texture
= sview
->base
.texture
;
623 const float dsdx
= fabsf(derivs
[0][0][quad
]);
624 const float dsdy
= fabsf(derivs
[0][1][quad
]);
625 const float dtdx
= fabsf(derivs
[1][0][quad
]);
626 const float dtdy
= fabsf(derivs
[1][1][quad
]);
627 const float dpdx
= fabsf(derivs
[2][0][quad
]);
628 const float dpdy
= fabsf(derivs
[2][1][quad
]);
629 const float maxx
= MAX2(dsdx
, dsdy
);
630 const float maxy
= MAX2(dtdx
, dtdy
);
631 const float maxz
= MAX2(dpdx
, dpdy
);
632 const float rho
= MAX3(maxx
, maxy
, maxz
) * u_minify(texture
->width0
, sview
->base
.u
.tex
.first_level
) / 2.0f
;
634 return util_fast_log2(rho
);
638 compute_lambda_cube(const struct sp_sampler_view
*sview
,
639 const float s
[TGSI_QUAD_SIZE
],
640 const float t
[TGSI_QUAD_SIZE
],
641 const float p
[TGSI_QUAD_SIZE
])
643 float derivs
[3][2][TGSI_QUAD_SIZE
];
644 compute_gradient_3d(s
, t
, p
, derivs
);
645 return compute_lambda_cube_explicit_gradients(sview
, derivs
, 0);
649 * Compute lambda for a vertex texture sampler.
650 * Since there aren't derivatives to use, just return 0.
653 compute_lambda_vert(const struct sp_sampler_view
*sview
,
654 const float s
[TGSI_QUAD_SIZE
],
655 const float t
[TGSI_QUAD_SIZE
],
656 const float p
[TGSI_QUAD_SIZE
])
664 * Get a texel from a texture, using the texture tile cache.
666 * \param addr the template tex address containing cube, z, face info.
667 * \param x the x coord of texel within 2D image
668 * \param y the y coord of texel within 2D image
669 * \param rgba the quad to put the texel/color into
671 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
672 * sp_get_cached_tile_tex() function.
677 static inline const float *
678 get_texel_buffer_no_border(const struct sp_sampler_view
*sp_sview
,
679 union tex_tile_address addr
, int x
, unsigned elmsize
)
681 const struct softpipe_tex_cached_tile
*tile
;
682 addr
.bits
.x
= x
* elmsize
/ TEX_TILE_SIZE
;
683 assert(x
* elmsize
/ TEX_TILE_SIZE
== addr
.bits
.x
);
685 x
%= TEX_TILE_SIZE
/ elmsize
;
687 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
689 return &tile
->data
.color
[0][x
][0];
693 static inline const float *
694 get_texel_2d_no_border(const struct sp_sampler_view
*sp_sview
,
695 union tex_tile_address addr
, int x
, int y
)
697 const struct softpipe_tex_cached_tile
*tile
;
698 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
699 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
703 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
705 return &tile
->data
.color
[y
][x
][0];
709 static inline const float *
710 get_texel_2d(const struct sp_sampler_view
*sp_sview
,
711 const struct sp_sampler
*sp_samp
,
712 union tex_tile_address addr
, int x
, int y
)
714 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
715 const unsigned level
= addr
.bits
.level
;
717 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
718 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
719 return sp_samp
->base
.border_color
.f
;
722 return get_texel_2d_no_border( sp_sview
, addr
, x
, y
);
728 * Here's the complete logic (HOLY CRAP) for finding next face and doing the
729 * corresponding coord wrapping, implemented by get_next_face,
730 * get_next_xcoord, get_next_ycoord.
731 * Read like that (first line):
732 * If face is +x and s coord is below zero, then
733 * new face is +z, new s is max , new t is old t
734 * (max is always cube size - 1).
736 * +x s- -> +z: s = max, t = t
737 * +x s+ -> -z: s = 0, t = t
738 * +x t- -> +y: s = max, t = max-s
739 * +x t+ -> -y: s = max, t = s
741 * -x s- -> -z: s = max, t = t
742 * -x s+ -> +z: s = 0, t = t
743 * -x t- -> +y: s = 0, t = s
744 * -x t+ -> -y: s = 0, t = max-s
746 * +y s- -> -x: s = t, t = 0
747 * +y s+ -> +x: s = max-t, t = 0
748 * +y t- -> -z: s = max-s, t = 0
749 * +y t+ -> +z: s = s, t = 0
751 * -y s- -> -x: s = max-t, t = max
752 * -y s+ -> +x: s = t, t = max
753 * -y t- -> +z: s = s, t = max
754 * -y t+ -> -z: s = max-s, t = max
756 * +z s- -> -x: s = max, t = t
757 * +z s+ -> +x: s = 0, t = t
758 * +z t- -> +y: s = s, t = max
759 * +z t+ -> -y: s = s, t = 0
761 * -z s- -> +x: s = max, t = t
762 * -z s+ -> -x: s = 0, t = t
763 * -z t- -> +y: s = max-s, t = 0
764 * -z t+ -> -y: s = max-s, t = max
769 * seamless cubemap neighbour array.
770 * this array is used to find the adjacent face in each of 4 directions,
771 * left, right, up, down. (or -x, +x, -y, +y).
773 static const unsigned face_array
[PIPE_TEX_FACE_MAX
][4] = {
774 /* pos X first then neg X is Z different, Y the same */
775 /* PIPE_TEX_FACE_POS_X,*/
776 { PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
,
777 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
778 /* PIPE_TEX_FACE_NEG_X */
779 { PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
,
780 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
782 /* pos Y first then neg Y is X different, X the same */
783 /* PIPE_TEX_FACE_POS_Y */
784 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
785 PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
},
787 /* PIPE_TEX_FACE_NEG_Y */
788 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
789 PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
},
791 /* pos Z first then neg Y is X different, X the same */
792 /* PIPE_TEX_FACE_POS_Z */
793 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
794 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
796 /* PIPE_TEX_FACE_NEG_Z */
797 { PIPE_TEX_FACE_POS_X
, PIPE_TEX_FACE_NEG_X
,
798 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
}
801 static inline unsigned
802 get_next_face(unsigned face
, int idx
)
804 return face_array
[face
][idx
];
808 * return a new xcoord based on old face, old coords, cube size
809 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
812 get_next_xcoord(unsigned face
, unsigned fall_off_index
, int max
, int xc
, int yc
)
814 if ((face
== 0 && fall_off_index
!= 1) ||
815 (face
== 1 && fall_off_index
== 0) ||
816 (face
== 4 && fall_off_index
== 0) ||
817 (face
== 5 && fall_off_index
== 0)) {
820 if ((face
== 1 && fall_off_index
!= 0) ||
821 (face
== 0 && fall_off_index
== 1) ||
822 (face
== 4 && fall_off_index
== 1) ||
823 (face
== 5 && fall_off_index
== 1)) {
826 if ((face
== 4 && fall_off_index
>= 2) ||
827 (face
== 2 && fall_off_index
== 3) ||
828 (face
== 3 && fall_off_index
== 2)) {
831 if ((face
== 5 && fall_off_index
>= 2) ||
832 (face
== 2 && fall_off_index
== 2) ||
833 (face
== 3 && fall_off_index
== 3)) {
836 if ((face
== 2 && fall_off_index
== 0) ||
837 (face
== 3 && fall_off_index
== 1)) {
840 /* (face == 2 && fall_off_index == 1) ||
841 (face == 3 && fall_off_index == 0)) */
846 * return a new ycoord based on old face, old coords, cube size
847 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
850 get_next_ycoord(unsigned face
, unsigned fall_off_index
, int max
, int xc
, int yc
)
852 if ((fall_off_index
<= 1) && (face
<= 1 || face
>= 4)) {
856 (face
== 4 && fall_off_index
== 3) ||
857 (face
== 5 && fall_off_index
== 2)) {
861 (face
== 4 && fall_off_index
== 2) ||
862 (face
== 5 && fall_off_index
== 3)) {
865 if ((face
== 0 && fall_off_index
== 3) ||
866 (face
== 1 && fall_off_index
== 2)) {
869 /* (face == 0 && fall_off_index == 2) ||
870 (face == 1 && fall_off_index == 3) */
875 /* Gather a quad of adjacent texels within a tile:
878 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_view
*sp_sview
,
879 union tex_tile_address addr
,
880 unsigned x
, unsigned y
,
883 const struct softpipe_tex_cached_tile
*tile
;
885 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
886 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
890 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
892 out
[0] = &tile
->data
.color
[y
][x
][0];
893 out
[1] = &tile
->data
.color
[y
][x
+1][0];
894 out
[2] = &tile
->data
.color
[y
+1][x
][0];
895 out
[3] = &tile
->data
.color
[y
+1][x
+1][0];
899 /* Gather a quad of potentially non-adjacent texels:
902 get_texel_quad_2d_no_border(const struct sp_sampler_view
*sp_sview
,
903 union tex_tile_address addr
,
908 out
[0] = get_texel_2d_no_border( sp_sview
, addr
, x0
, y0
);
909 out
[1] = get_texel_2d_no_border( sp_sview
, addr
, x1
, y0
);
910 out
[2] = get_texel_2d_no_border( sp_sview
, addr
, x0
, y1
);
911 out
[3] = get_texel_2d_no_border( sp_sview
, addr
, x1
, y1
);
917 static inline const float *
918 get_texel_3d_no_border(const struct sp_sampler_view
*sp_sview
,
919 union tex_tile_address addr
, int x
, int y
, int z
)
921 const struct softpipe_tex_cached_tile
*tile
;
923 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
924 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
929 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
931 return &tile
->data
.color
[y
][x
][0];
935 static inline const float *
936 get_texel_3d(const struct sp_sampler_view
*sp_sview
,
937 const struct sp_sampler
*sp_samp
,
938 union tex_tile_address addr
, int x
, int y
, int z
)
940 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
941 const unsigned level
= addr
.bits
.level
;
943 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
944 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
) ||
945 z
< 0 || z
>= (int) u_minify(texture
->depth0
, level
)) {
946 return sp_samp
->base
.border_color
.f
;
949 return get_texel_3d_no_border( sp_sview
, addr
, x
, y
, z
);
954 /* Get texel pointer for 1D array texture */
955 static inline const float *
956 get_texel_1d_array(const struct sp_sampler_view
*sp_sview
,
957 const struct sp_sampler
*sp_samp
,
958 union tex_tile_address addr
, int x
, int y
)
960 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
961 const unsigned level
= addr
.bits
.level
;
963 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
)) {
964 return sp_samp
->base
.border_color
.f
;
967 return get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
972 /* Get texel pointer for 2D array texture */
973 static inline const float *
974 get_texel_2d_array(const struct sp_sampler_view
*sp_sview
,
975 const struct sp_sampler
*sp_samp
,
976 union tex_tile_address addr
, int x
, int y
, int layer
)
978 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
979 const unsigned level
= addr
.bits
.level
;
981 assert(layer
< (int) texture
->array_size
);
984 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
985 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
986 return sp_samp
->base
.border_color
.f
;
989 return get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
994 static inline const float *
995 get_texel_cube_seamless(const struct sp_sampler_view
*sp_sview
,
996 union tex_tile_address addr
, int x
, int y
,
997 float *corner
, int layer
, unsigned face
)
999 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1000 const unsigned level
= addr
.bits
.level
;
1001 int new_x
, new_y
, max_x
;
1003 max_x
= (int) u_minify(texture
->width0
, level
);
1005 assert(texture
->width0
== texture
->height0
);
1009 /* change the face */
1012 * Cheat with corners. They are difficult and I believe because we don't get
1013 * per-pixel faces we can actually have multiple corner texels per pixel,
1014 * which screws things up majorly in any case (as the per spec behavior is
1015 * to average the 3 remaining texels, which we might not have).
1016 * Hence just make sure that the 2nd coord is clamped, will simply pick the
1017 * sample which would have fallen off the x coord, but not y coord.
1018 * So the filter weight of the samples will be wrong, but at least this
1019 * ensures that only valid texels near the corner are used.
1021 if (y
< 0 || y
>= max_x
) {
1022 y
= CLAMP(y
, 0, max_x
- 1);
1024 new_x
= get_next_xcoord(face
, 0, max_x
-1, x
, y
);
1025 new_y
= get_next_ycoord(face
, 0, max_x
-1, x
, y
);
1026 face
= get_next_face(face
, 0);
1027 } else if (x
>= max_x
) {
1028 if (y
< 0 || y
>= max_x
) {
1029 y
= CLAMP(y
, 0, max_x
- 1);
1031 new_x
= get_next_xcoord(face
, 1, max_x
-1, x
, y
);
1032 new_y
= get_next_ycoord(face
, 1, max_x
-1, x
, y
);
1033 face
= get_next_face(face
, 1);
1035 new_x
= get_next_xcoord(face
, 2, max_x
-1, x
, y
);
1036 new_y
= get_next_ycoord(face
, 2, max_x
-1, x
, y
);
1037 face
= get_next_face(face
, 2);
1038 } else if (y
>= max_x
) {
1039 new_x
= get_next_xcoord(face
, 3, max_x
-1, x
, y
);
1040 new_y
= get_next_ycoord(face
, 3, max_x
-1, x
, y
);
1041 face
= get_next_face(face
, 3);
1044 return get_texel_3d_no_border(sp_sview
, addr
, new_x
, new_y
, layer
+ face
);
1048 /* Get texel pointer for cube array texture */
1049 static inline const float *
1050 get_texel_cube_array(const struct sp_sampler_view
*sp_sview
,
1051 const struct sp_sampler
*sp_samp
,
1052 union tex_tile_address addr
, int x
, int y
, int layer
)
1054 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1055 const unsigned level
= addr
.bits
.level
;
1057 assert(layer
< (int) texture
->array_size
);
1060 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
1061 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
1062 return sp_samp
->base
.border_color
.f
;
1065 return get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
1069 * Given the logbase2 of a mipmap's base level size and a mipmap level,
1070 * return the size (in texels) of that mipmap level.
1071 * For example, if level[0].width = 256 then base_pot will be 8.
1072 * If level = 2, then we'll return 64 (the width at level=2).
1073 * Return 1 if level > base_pot.
1075 static inline unsigned
1076 pot_level_size(unsigned base_pot
, unsigned level
)
1078 return (base_pot
>= level
) ? (1 << (base_pot
- level
)) : 1;
1083 print_sample(const char *function
, const float *rgba
)
1085 debug_printf("%s %g %g %g %g\n",
1087 rgba
[0], rgba
[TGSI_NUM_CHANNELS
], rgba
[2*TGSI_NUM_CHANNELS
], rgba
[3*TGSI_NUM_CHANNELS
]);
1092 print_sample_4(const char *function
, float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1094 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
1096 rgba
[0][0], rgba
[1][0], rgba
[2][0], rgba
[3][0],
1097 rgba
[0][1], rgba
[1][1], rgba
[2][1], rgba
[3][1],
1098 rgba
[0][2], rgba
[1][2], rgba
[2][2], rgba
[3][2],
1099 rgba
[0][3], rgba
[1][3], rgba
[2][3], rgba
[3][3]);
1103 /* Some image-filter fastpaths:
1106 img_filter_2d_linear_repeat_POT(const struct sp_sampler_view
*sp_sview
,
1107 const struct sp_sampler
*sp_samp
,
1108 const struct img_filter_args
*args
,
1111 const unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1112 const unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1113 const int xmax
= (xpot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
1114 const int ymax
= (ypot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
1115 union tex_tile_address addr
;
1118 const float u
= (args
->s
* xpot
- 0.5F
) + args
->offset
[0];
1119 const float v
= (args
->t
* ypot
- 0.5F
) + args
->offset
[1];
1121 const int uflr
= util_ifloor(u
);
1122 const int vflr
= util_ifloor(v
);
1124 const float xw
= u
- (float)uflr
;
1125 const float yw
= v
- (float)vflr
;
1127 const int x0
= uflr
& (xpot
- 1);
1128 const int y0
= vflr
& (ypot
- 1);
1133 addr
.bits
.level
= args
->level
;
1134 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1136 /* Can we fetch all four at once:
1138 if (x0
< xmax
&& y0
< ymax
) {
1139 get_texel_quad_2d_no_border_single_tile(sp_sview
, addr
, x0
, y0
, tx
);
1142 const unsigned x1
= (x0
+ 1) & (xpot
- 1);
1143 const unsigned y1
= (y0
+ 1) & (ypot
- 1);
1144 get_texel_quad_2d_no_border(sp_sview
, addr
, x0
, y0
, x1
, y1
, tx
);
1147 /* interpolate R, G, B, A */
1148 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++) {
1149 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1151 tx
[2][c
], tx
[3][c
]);
1155 print_sample(__FUNCTION__
, rgba
);
1161 img_filter_2d_nearest_repeat_POT(const struct sp_sampler_view
*sp_sview
,
1162 const struct sp_sampler
*sp_samp
,
1163 const struct img_filter_args
*args
,
1166 const unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1167 const unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1169 union tex_tile_address addr
;
1172 const float u
= args
->s
* xpot
+ args
->offset
[0];
1173 const float v
= args
->t
* ypot
+ args
->offset
[1];
1175 const int uflr
= util_ifloor(u
);
1176 const int vflr
= util_ifloor(v
);
1178 const int x0
= uflr
& (xpot
- 1);
1179 const int y0
= vflr
& (ypot
- 1);
1182 addr
.bits
.level
= args
->level
;
1183 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1185 out
= get_texel_2d_no_border(sp_sview
, addr
, x0
, y0
);
1186 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1187 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1190 print_sample(__FUNCTION__
, rgba
);
1196 img_filter_2d_nearest_clamp_POT(const struct sp_sampler_view
*sp_sview
,
1197 const struct sp_sampler
*sp_samp
,
1198 const struct img_filter_args
*args
,
1201 const unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1202 const unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1203 union tex_tile_address addr
;
1206 const float u
= args
->s
* xpot
+ args
->offset
[0];
1207 const float v
= args
->t
* ypot
+ args
->offset
[1];
1213 addr
.bits
.level
= args
->level
;
1214 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1216 x0
= util_ifloor(u
);
1219 else if (x0
> (int) xpot
- 1)
1222 y0
= util_ifloor(v
);
1225 else if (y0
> (int) ypot
- 1)
1228 out
= get_texel_2d_no_border(sp_sview
, addr
, x0
, y0
);
1229 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1230 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1233 print_sample(__FUNCTION__
, rgba
);
1239 img_filter_1d_nearest(const struct sp_sampler_view
*sp_sview
,
1240 const struct sp_sampler
*sp_samp
,
1241 const struct img_filter_args
*args
,
1244 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1245 const int width
= u_minify(texture
->width0
, args
->level
);
1247 union tex_tile_address addr
;
1254 addr
.bits
.level
= args
->level
;
1256 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1258 out
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x
,
1259 sp_sview
->base
.u
.tex
.first_layer
);
1260 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1261 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1264 print_sample(__FUNCTION__
, rgba
);
1270 img_filter_1d_array_nearest(const struct sp_sampler_view
*sp_sview
,
1271 const struct sp_sampler
*sp_samp
,
1272 const struct img_filter_args
*args
,
1275 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1276 const int width
= u_minify(texture
->width0
, args
->level
);
1277 const int layer
= coord_to_layer(args
->t
, sp_sview
->base
.u
.tex
.first_layer
,
1278 sp_sview
->base
.u
.tex
.last_layer
);
1280 union tex_tile_address addr
;
1287 addr
.bits
.level
= args
->level
;
1289 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1291 out
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x
, layer
);
1292 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1293 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1296 print_sample(__FUNCTION__
, rgba
);
1302 img_filter_2d_nearest(const struct sp_sampler_view
*sp_sview
,
1303 const struct sp_sampler
*sp_samp
,
1304 const struct img_filter_args
*args
,
1307 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1308 const int width
= u_minify(texture
->width0
, args
->level
);
1309 const int height
= u_minify(texture
->height0
, args
->level
);
1311 union tex_tile_address addr
;
1319 addr
.bits
.level
= args
->level
;
1320 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1322 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1323 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1325 out
= get_texel_2d(sp_sview
, sp_samp
, addr
, x
, y
);
1326 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1327 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1330 print_sample(__FUNCTION__
, rgba
);
1336 img_filter_2d_array_nearest(const struct sp_sampler_view
*sp_sview
,
1337 const struct sp_sampler
*sp_samp
,
1338 const struct img_filter_args
*args
,
1341 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1342 const int width
= u_minify(texture
->width0
, args
->level
);
1343 const int height
= u_minify(texture
->height0
, args
->level
);
1344 const int layer
= coord_to_layer(args
->p
, sp_sview
->base
.u
.tex
.first_layer
,
1345 sp_sview
->base
.u
.tex
.last_layer
);
1347 union tex_tile_address addr
;
1355 addr
.bits
.level
= args
->level
;
1357 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1358 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1360 out
= get_texel_2d_array(sp_sview
, sp_samp
, addr
, x
, y
, layer
);
1361 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1362 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1365 print_sample(__FUNCTION__
, rgba
);
1371 img_filter_cube_nearest(const struct sp_sampler_view
*sp_sview
,
1372 const struct sp_sampler
*sp_samp
,
1373 const struct img_filter_args
*args
,
1376 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1377 const int width
= u_minify(texture
->width0
, args
->level
);
1378 const int height
= u_minify(texture
->height0
, args
->level
);
1379 const int layerface
= args
->face_id
+ sp_sview
->base
.u
.tex
.first_layer
;
1381 union tex_tile_address addr
;
1389 addr
.bits
.level
= args
->level
;
1392 * If NEAREST filtering is done within a miplevel, always apply wrap
1393 * mode CLAMP_TO_EDGE.
1395 if (sp_samp
->base
.seamless_cube_map
) {
1396 wrap_nearest_clamp_to_edge(args
->s
, width
, args
->offset
[0], &x
);
1397 wrap_nearest_clamp_to_edge(args
->t
, height
, args
->offset
[1], &y
);
1399 /* Would probably make sense to ignore mode and just do edge clamp */
1400 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1401 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1404 out
= get_texel_cube_array(sp_sview
, sp_samp
, addr
, x
, y
, layerface
);
1405 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1406 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1409 print_sample(__FUNCTION__
, rgba
);
1414 img_filter_cube_array_nearest(const struct sp_sampler_view
*sp_sview
,
1415 const struct sp_sampler
*sp_samp
,
1416 const struct img_filter_args
*args
,
1419 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1420 const int width
= u_minify(texture
->width0
, args
->level
);
1421 const int height
= u_minify(texture
->height0
, args
->level
);
1422 const int layerface
=
1423 coord_to_layer(6 * args
->p
+ sp_sview
->base
.u
.tex
.first_layer
,
1424 sp_sview
->base
.u
.tex
.first_layer
,
1425 sp_sview
->base
.u
.tex
.last_layer
- 5) + args
->face_id
;
1427 union tex_tile_address addr
;
1435 addr
.bits
.level
= args
->level
;
1437 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1438 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1440 out
= get_texel_cube_array(sp_sview
, sp_samp
, addr
, x
, y
, layerface
);
1441 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1442 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1445 print_sample(__FUNCTION__
, rgba
);
1450 img_filter_3d_nearest(const struct sp_sampler_view
*sp_sview
,
1451 const struct sp_sampler
*sp_samp
,
1452 const struct img_filter_args
*args
,
1455 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1456 const int width
= u_minify(texture
->width0
, args
->level
);
1457 const int height
= u_minify(texture
->height0
, args
->level
);
1458 const int depth
= u_minify(texture
->depth0
, args
->level
);
1460 union tex_tile_address addr
;
1468 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1469 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1470 sp_samp
->nearest_texcoord_p(args
->p
, depth
, args
->offset
[2], &z
);
1473 addr
.bits
.level
= args
->level
;
1475 out
= get_texel_3d(sp_sview
, sp_samp
, addr
, x
, y
, z
);
1476 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1477 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1482 img_filter_1d_linear(const struct sp_sampler_view
*sp_sview
,
1483 const struct sp_sampler
*sp_samp
,
1484 const struct img_filter_args
*args
,
1487 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1488 const int width
= u_minify(texture
->width0
, args
->level
);
1490 float xw
; /* weights */
1491 union tex_tile_address addr
;
1492 const float *tx0
, *tx1
;
1498 addr
.bits
.level
= args
->level
;
1500 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1502 tx0
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x0
,
1503 sp_sview
->base
.u
.tex
.first_layer
);
1504 tx1
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x1
,
1505 sp_sview
->base
.u
.tex
.first_layer
);
1507 /* interpolate R, G, B, A */
1508 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1509 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1514 img_filter_1d_array_linear(const struct sp_sampler_view
*sp_sview
,
1515 const struct sp_sampler
*sp_samp
,
1516 const struct img_filter_args
*args
,
1519 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1520 const int width
= u_minify(texture
->width0
, args
->level
);
1521 const int layer
= coord_to_layer(args
->t
, sp_sview
->base
.u
.tex
.first_layer
,
1522 sp_sview
->base
.u
.tex
.last_layer
);
1524 float xw
; /* weights */
1525 union tex_tile_address addr
;
1526 const float *tx0
, *tx1
;
1532 addr
.bits
.level
= args
->level
;
1534 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1536 tx0
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x0
, layer
);
1537 tx1
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x1
, layer
);
1539 /* interpolate R, G, B, A */
1540 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1541 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1545 * Retrieve the gathered value, need to convert to the
1546 * TGSI expected interface, and take component select
1547 * and swizzling into account.
1550 get_gather_value(const struct sp_sampler_view
*sp_sview
,
1551 int chan_in
, int comp_sel
,
1558 * softpipe samples in a different order
1559 * to TGSI expects, so we need to swizzle,
1560 * the samples into the correct slots.
1580 /* pick which component to use for the swizzle */
1583 swizzle
= sp_sview
->base
.swizzle_r
;
1586 swizzle
= sp_sview
->base
.swizzle_g
;
1589 swizzle
= sp_sview
->base
.swizzle_b
;
1592 swizzle
= sp_sview
->base
.swizzle_a
;
1599 /* get correct result using the channel and swizzle */
1601 case PIPE_SWIZZLE_0
:
1603 case PIPE_SWIZZLE_1
:
1606 return tx
[chan
][swizzle
];
1612 img_filter_2d_linear(const struct sp_sampler_view
*sp_sview
,
1613 const struct sp_sampler
*sp_samp
,
1614 const struct img_filter_args
*args
,
1617 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1618 const int width
= u_minify(texture
->width0
, args
->level
);
1619 const int height
= u_minify(texture
->height0
, args
->level
);
1621 float xw
, yw
; /* weights */
1622 union tex_tile_address addr
;
1630 addr
.bits
.level
= args
->level
;
1631 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1633 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1634 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1636 tx
[0] = get_texel_2d(sp_sview
, sp_samp
, addr
, x0
, y0
);
1637 tx
[1] = get_texel_2d(sp_sview
, sp_samp
, addr
, x1
, y0
);
1638 tx
[2] = get_texel_2d(sp_sview
, sp_samp
, addr
, x0
, y1
);
1639 tx
[3] = get_texel_2d(sp_sview
, sp_samp
, addr
, x1
, y1
);
1641 if (args
->gather_only
) {
1642 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1643 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1647 /* interpolate R, G, B, A */
1648 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1649 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1651 tx
[2][c
], tx
[3][c
]);
1657 img_filter_2d_array_linear(const struct sp_sampler_view
*sp_sview
,
1658 const struct sp_sampler
*sp_samp
,
1659 const struct img_filter_args
*args
,
1662 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1663 const int width
= u_minify(texture
->width0
, args
->level
);
1664 const int height
= u_minify(texture
->height0
, args
->level
);
1665 const int layer
= coord_to_layer(args
->p
, sp_sview
->base
.u
.tex
.first_layer
,
1666 sp_sview
->base
.u
.tex
.last_layer
);
1668 float xw
, yw
; /* weights */
1669 union tex_tile_address addr
;
1677 addr
.bits
.level
= args
->level
;
1679 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1680 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1682 tx
[0] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
);
1683 tx
[1] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
);
1684 tx
[2] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
);
1685 tx
[3] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
);
1687 if (args
->gather_only
) {
1688 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1689 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1693 /* interpolate R, G, B, A */
1694 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1695 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1697 tx
[2][c
], tx
[3][c
]);
1703 img_filter_cube_linear(const struct sp_sampler_view
*sp_sview
,
1704 const struct sp_sampler
*sp_samp
,
1705 const struct img_filter_args
*args
,
1708 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1709 const int width
= u_minify(texture
->width0
, args
->level
);
1710 const int height
= u_minify(texture
->height0
, args
->level
);
1711 const int layer
= sp_sview
->base
.u
.tex
.first_layer
;
1713 float xw
, yw
; /* weights */
1714 union tex_tile_address addr
;
1716 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
],
1717 corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1724 addr
.bits
.level
= args
->level
;
1727 * For seamless if LINEAR filtering is done within a miplevel,
1728 * always apply wrap mode CLAMP_TO_BORDER.
1730 if (sp_samp
->base
.seamless_cube_map
) {
1731 /* Note this is a bit overkill, actual clamping is not required */
1732 wrap_linear_clamp_to_border(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1733 wrap_linear_clamp_to_border(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1735 /* Would probably make sense to ignore mode and just do edge clamp */
1736 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1737 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1740 if (sp_samp
->base
.seamless_cube_map
) {
1741 tx
[0] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y0
, corner0
, layer
, args
->face_id
);
1742 tx
[1] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y0
, corner1
, layer
, args
->face_id
);
1743 tx
[2] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y1
, corner2
, layer
, args
->face_id
);
1744 tx
[3] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y1
, corner3
, layer
, args
->face_id
);
1746 tx
[0] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
+ args
->face_id
);
1747 tx
[1] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
+ args
->face_id
);
1748 tx
[2] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
+ args
->face_id
);
1749 tx
[3] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
+ args
->face_id
);
1752 if (args
->gather_only
) {
1753 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1754 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1758 /* interpolate R, G, B, A */
1759 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1760 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1762 tx
[2][c
], tx
[3][c
]);
1768 img_filter_cube_array_linear(const struct sp_sampler_view
*sp_sview
,
1769 const struct sp_sampler
*sp_samp
,
1770 const struct img_filter_args
*args
,
1773 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1774 const int width
= u_minify(texture
->width0
, args
->level
);
1775 const int height
= u_minify(texture
->height0
, args
->level
);
1777 coord_to_layer(6 * args
->p
+ sp_sview
->base
.u
.tex
.first_layer
,
1778 sp_sview
->base
.u
.tex
.first_layer
,
1779 sp_sview
->base
.u
.tex
.last_layer
- 5);
1781 float xw
, yw
; /* weights */
1782 union tex_tile_address addr
;
1784 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
],
1785 corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1792 addr
.bits
.level
= args
->level
;
1795 * For seamless if LINEAR filtering is done within a miplevel,
1796 * always apply wrap mode CLAMP_TO_BORDER.
1798 if (sp_samp
->base
.seamless_cube_map
) {
1799 /* Note this is a bit overkill, actual clamping is not required */
1800 wrap_linear_clamp_to_border(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1801 wrap_linear_clamp_to_border(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1803 /* Would probably make sense to ignore mode and just do edge clamp */
1804 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1805 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1808 if (sp_samp
->base
.seamless_cube_map
) {
1809 tx
[0] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y0
, corner0
, layer
, args
->face_id
);
1810 tx
[1] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y0
, corner1
, layer
, args
->face_id
);
1811 tx
[2] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y1
, corner2
, layer
, args
->face_id
);
1812 tx
[3] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y1
, corner3
, layer
, args
->face_id
);
1814 tx
[0] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
+ args
->face_id
);
1815 tx
[1] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
+ args
->face_id
);
1816 tx
[2] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
+ args
->face_id
);
1817 tx
[3] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
+ args
->face_id
);
1820 if (args
->gather_only
) {
1821 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1822 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1826 /* interpolate R, G, B, A */
1827 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1828 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1830 tx
[2][c
], tx
[3][c
]);
1835 img_filter_3d_linear(const struct sp_sampler_view
*sp_sview
,
1836 const struct sp_sampler
*sp_samp
,
1837 const struct img_filter_args
*args
,
1840 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1841 const int width
= u_minify(texture
->width0
, args
->level
);
1842 const int height
= u_minify(texture
->height0
, args
->level
);
1843 const int depth
= u_minify(texture
->depth0
, args
->level
);
1844 int x0
, x1
, y0
, y1
, z0
, z1
;
1845 float xw
, yw
, zw
; /* interpolation weights */
1846 union tex_tile_address addr
;
1847 const float *tx00
, *tx01
, *tx02
, *tx03
, *tx10
, *tx11
, *tx12
, *tx13
;
1851 addr
.bits
.level
= args
->level
;
1857 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1858 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1859 sp_samp
->linear_texcoord_p(args
->p
, depth
, args
->offset
[2], &z0
, &z1
, &zw
);
1861 tx00
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y0
, z0
);
1862 tx01
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y0
, z0
);
1863 tx02
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y1
, z0
);
1864 tx03
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y1
, z0
);
1866 tx10
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y0
, z1
);
1867 tx11
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y0
, z1
);
1868 tx12
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y1
, z1
);
1869 tx13
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y1
, z1
);
1871 /* interpolate R, G, B, A */
1872 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1873 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_3d(xw
, yw
, zw
,
1881 /* Calculate level of detail for every fragment,
1882 * with lambda already computed.
1883 * Note that lambda has already been biased by global LOD bias.
1884 * \param biased_lambda per-quad lambda.
1885 * \param lod_in per-fragment lod_bias or explicit_lod.
1886 * \param lod returns the per-fragment lod.
1889 compute_lod(const struct pipe_sampler_state
*sampler
,
1890 enum tgsi_sampler_control control
,
1891 const float biased_lambda
,
1892 const float lod_in
[TGSI_QUAD_SIZE
],
1893 float lod
[TGSI_QUAD_SIZE
])
1895 const float min_lod
= sampler
->min_lod
;
1896 const float max_lod
= sampler
->max_lod
;
1900 case TGSI_SAMPLER_LOD_NONE
:
1901 case TGSI_SAMPLER_LOD_ZERO
:
1903 case TGSI_SAMPLER_DERIVS_EXPLICIT
:
1904 lod
[0] = lod
[1] = lod
[2] = lod
[3] = CLAMP(biased_lambda
, min_lod
, max_lod
);
1906 case TGSI_SAMPLER_LOD_BIAS
:
1907 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1908 lod
[i
] = biased_lambda
+ lod_in
[i
];
1909 lod
[i
] = CLAMP(lod
[i
], min_lod
, max_lod
);
1912 case TGSI_SAMPLER_LOD_EXPLICIT
:
1913 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1914 lod
[i
] = CLAMP(lod_in
[i
], min_lod
, max_lod
);
1919 lod
[0] = lod
[1] = lod
[2] = lod
[3] = 0.0f
;
1924 /* Calculate level of detail for every fragment. The computed value is not
1925 * clamped to lod_min and lod_max.
1926 * \param lod_in per-fragment lod_bias or explicit_lod.
1927 * \param lod results per-fragment lod.
1930 compute_lambda_lod_unclamped(const struct sp_sampler_view
*sp_sview
,
1931 const struct sp_sampler
*sp_samp
,
1932 const float s
[TGSI_QUAD_SIZE
],
1933 const float t
[TGSI_QUAD_SIZE
],
1934 const float p
[TGSI_QUAD_SIZE
],
1935 const float lod_in
[TGSI_QUAD_SIZE
],
1936 enum tgsi_sampler_control control
,
1937 float lod
[TGSI_QUAD_SIZE
])
1939 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
1940 const float lod_bias
= sampler
->lod_bias
;
1945 case TGSI_SAMPLER_LOD_NONE
:
1947 case TGSI_SAMPLER_DERIVS_EXPLICIT
:
1948 lambda
= sp_sview
->compute_lambda(sp_sview
, s
, t
, p
) + lod_bias
;
1949 lod
[0] = lod
[1] = lod
[2] = lod
[3] = lambda
;
1951 case TGSI_SAMPLER_LOD_BIAS
:
1952 lambda
= sp_sview
->compute_lambda(sp_sview
, s
, t
, p
) + lod_bias
;
1953 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1954 lod
[i
] = lambda
+ lod_in
[i
];
1957 case TGSI_SAMPLER_LOD_EXPLICIT
:
1958 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1959 lod
[i
] = lod_in
[i
] + lod_bias
;
1962 case TGSI_SAMPLER_LOD_ZERO
:
1963 case TGSI_SAMPLER_GATHER
:
1964 lod
[0] = lod
[1] = lod
[2] = lod
[3] = lod_bias
;
1968 lod
[0] = lod
[1] = lod
[2] = lod
[3] = 0.0f
;
1972 /* Calculate level of detail for every fragment.
1973 * \param lod_in per-fragment lod_bias or explicit_lod.
1974 * \param lod results per-fragment lod.
1977 compute_lambda_lod(const struct sp_sampler_view
*sp_sview
,
1978 const struct sp_sampler
*sp_samp
,
1979 const float s
[TGSI_QUAD_SIZE
],
1980 const float t
[TGSI_QUAD_SIZE
],
1981 const float p
[TGSI_QUAD_SIZE
],
1982 const float lod_in
[TGSI_QUAD_SIZE
],
1983 enum tgsi_sampler_control control
,
1984 float lod
[TGSI_QUAD_SIZE
])
1986 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
1987 const float min_lod
= sampler
->min_lod
;
1988 const float max_lod
= sampler
->max_lod
;
1991 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
1992 s
, t
, p
, lod_in
, control
, lod
);
1993 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1994 lod
[i
] = CLAMP(lod
[i
], min_lod
, max_lod
);
1998 static inline unsigned
1999 get_gather_component(const float lod_in
[TGSI_QUAD_SIZE
])
2001 /* gather component is stored in lod_in slot as unsigned */
2002 return (*(unsigned int *)lod_in
) & 0x3;
2006 * Clamps given lod to both lod limits and mip level limits. Clamping to the
2007 * latter limits is done so that lod is relative to the first (base) level.
2010 clamp_lod(const struct sp_sampler_view
*sp_sview
,
2011 const struct sp_sampler
*sp_samp
,
2012 const float lod
[TGSI_QUAD_SIZE
],
2013 float clamped
[TGSI_QUAD_SIZE
])
2015 const float min_lod
= sp_samp
->base
.min_lod
;
2016 const float max_lod
= sp_samp
->base
.max_lod
;
2017 const float min_level
= sp_sview
->base
.u
.tex
.first_level
;
2018 const float max_level
= sp_sview
->base
.u
.tex
.last_level
;
2021 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2024 cl
= CLAMP(cl
, min_lod
, max_lod
);
2025 cl
= CLAMP(cl
, 0, max_level
- min_level
);
2031 * Get mip level relative to base level for linear mip filter
2034 mip_rel_level_linear(const struct sp_sampler_view
*sp_sview
,
2035 const struct sp_sampler
*sp_samp
,
2036 const float lod
[TGSI_QUAD_SIZE
],
2037 float level
[TGSI_QUAD_SIZE
])
2039 clamp_lod(sp_sview
, sp_samp
, lod
, level
);
2043 mip_filter_linear(const struct sp_sampler_view
*sp_sview
,
2044 const struct sp_sampler
*sp_samp
,
2045 img_filter_func min_filter
,
2046 img_filter_func mag_filter
,
2047 const float s
[TGSI_QUAD_SIZE
],
2048 const float t
[TGSI_QUAD_SIZE
],
2049 const float p
[TGSI_QUAD_SIZE
],
2050 const float c0
[TGSI_QUAD_SIZE
],
2051 const float lod_in
[TGSI_QUAD_SIZE
],
2052 const struct filter_args
*filt_args
,
2053 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2055 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2057 float lod
[TGSI_QUAD_SIZE
];
2058 struct img_filter_args args
;
2060 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2062 args
.offset
= filt_args
->offset
;
2063 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2064 args
.gather_comp
= get_gather_component(lod_in
);
2066 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2067 const int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
2072 args
.face_id
= filt_args
->faces
[j
];
2074 if (lod
[j
] <= 0.0 && !args
.gather_only
) {
2075 args
.level
= psview
->u
.tex
.first_level
;
2076 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2078 else if (level0
>= (int) psview
->u
.tex
.last_level
) {
2079 args
.level
= psview
->u
.tex
.last_level
;
2080 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2083 float levelBlend
= frac(lod
[j
]);
2084 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2087 args
.level
= level0
;
2088 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
2089 args
.level
= level0
+1;
2090 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
2092 for (c
= 0; c
< 4; c
++) {
2093 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2099 print_sample_4(__FUNCTION__
, rgba
);
2105 * Get mip level relative to base level for nearest mip filter
2108 mip_rel_level_nearest(const struct sp_sampler_view
*sp_sview
,
2109 const struct sp_sampler
*sp_samp
,
2110 const float lod
[TGSI_QUAD_SIZE
],
2111 float level
[TGSI_QUAD_SIZE
])
2115 clamp_lod(sp_sview
, sp_samp
, lod
, level
);
2116 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++)
2117 /* TODO: It should rather be:
2118 * level[j] = ceil(level[j] + 0.5F) - 1.0F;
2120 level
[j
] = (int)(level
[j
] + 0.5F
);
2124 * Compute nearest mipmap level from texcoords.
2125 * Then sample the texture level for four elements of a quad.
2126 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
2129 mip_filter_nearest(const struct sp_sampler_view
*sp_sview
,
2130 const struct sp_sampler
*sp_samp
,
2131 img_filter_func min_filter
,
2132 img_filter_func mag_filter
,
2133 const float s
[TGSI_QUAD_SIZE
],
2134 const float t
[TGSI_QUAD_SIZE
],
2135 const float p
[TGSI_QUAD_SIZE
],
2136 const float c0
[TGSI_QUAD_SIZE
],
2137 const float lod_in
[TGSI_QUAD_SIZE
],
2138 const struct filter_args
*filt_args
,
2139 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2141 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2142 float lod
[TGSI_QUAD_SIZE
];
2144 struct img_filter_args args
;
2146 args
.offset
= filt_args
->offset
;
2147 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2148 args
.gather_comp
= get_gather_component(lod_in
);
2150 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2152 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2156 args
.face_id
= filt_args
->faces
[j
];
2158 if (lod
[j
] <= 0.0 && !args
.gather_only
) {
2159 args
.level
= psview
->u
.tex
.first_level
;
2160 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2162 const int level
= psview
->u
.tex
.first_level
+ (int)(lod
[j
] + 0.5F
);
2163 args
.level
= MIN2(level
, (int)psview
->u
.tex
.last_level
);
2164 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2169 print_sample_4(__FUNCTION__
, rgba
);
2175 * Get mip level relative to base level for none mip filter
2178 mip_rel_level_none(const struct sp_sampler_view
*sp_sview
,
2179 const struct sp_sampler
*sp_samp
,
2180 const float lod
[TGSI_QUAD_SIZE
],
2181 float level
[TGSI_QUAD_SIZE
])
2185 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2191 mip_filter_none(const struct sp_sampler_view
*sp_sview
,
2192 const struct sp_sampler
*sp_samp
,
2193 img_filter_func min_filter
,
2194 img_filter_func mag_filter
,
2195 const float s
[TGSI_QUAD_SIZE
],
2196 const float t
[TGSI_QUAD_SIZE
],
2197 const float p
[TGSI_QUAD_SIZE
],
2198 const float c0
[TGSI_QUAD_SIZE
],
2199 const float lod_in
[TGSI_QUAD_SIZE
],
2200 const struct filter_args
*filt_args
,
2201 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2203 float lod
[TGSI_QUAD_SIZE
];
2205 struct img_filter_args args
;
2207 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2208 args
.offset
= filt_args
->offset
;
2209 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2211 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2213 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2217 args
.face_id
= filt_args
->faces
[j
];
2218 if (lod
[j
] <= 0.0f
&& !args
.gather_only
) {
2219 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2222 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2229 * Get mip level relative to base level for none mip filter
2232 mip_rel_level_none_no_filter_select(const struct sp_sampler_view
*sp_sview
,
2233 const struct sp_sampler
*sp_samp
,
2234 const float lod
[TGSI_QUAD_SIZE
],
2235 float level
[TGSI_QUAD_SIZE
])
2237 mip_rel_level_none(sp_sview
, sp_samp
, lod
, level
);
2241 mip_filter_none_no_filter_select(const struct sp_sampler_view
*sp_sview
,
2242 const struct sp_sampler
*sp_samp
,
2243 img_filter_func min_filter
,
2244 img_filter_func mag_filter
,
2245 const float s
[TGSI_QUAD_SIZE
],
2246 const float t
[TGSI_QUAD_SIZE
],
2247 const float p
[TGSI_QUAD_SIZE
],
2248 const float c0
[TGSI_QUAD_SIZE
],
2249 const float lod_in
[TGSI_QUAD_SIZE
],
2250 const struct filter_args
*filt_args
,
2251 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2254 struct img_filter_args args
;
2255 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2256 args
.offset
= filt_args
->offset
;
2257 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2258 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2262 args
.face_id
= filt_args
->faces
[j
];
2263 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2268 /* For anisotropic filtering */
2269 #define WEIGHT_LUT_SIZE 1024
2271 static const float *weightLut
= NULL
;
2274 * Creates the look-up table used to speed-up EWA sampling
2277 create_filter_table(void)
2281 float *lut
= (float *) MALLOC(WEIGHT_LUT_SIZE
* sizeof(float));
2283 for (i
= 0; i
< WEIGHT_LUT_SIZE
; ++i
) {
2284 const float alpha
= 2;
2285 const float r2
= (float) i
/ (float) (WEIGHT_LUT_SIZE
- 1);
2286 const float weight
= (float) exp(-alpha
* r2
);
2295 * Elliptical weighted average (EWA) filter for producing high quality
2296 * anisotropic filtered results.
2297 * Based on the Higher Quality Elliptical Weighted Average Filter
2298 * published by Paul S. Heckbert in his Master's Thesis
2299 * "Fundamentals of Texture Mapping and Image Warping" (1989)
2302 img_filter_2d_ewa(const struct sp_sampler_view
*sp_sview
,
2303 const struct sp_sampler
*sp_samp
,
2304 img_filter_func min_filter
,
2305 img_filter_func mag_filter
,
2306 const float s
[TGSI_QUAD_SIZE
],
2307 const float t
[TGSI_QUAD_SIZE
],
2308 const float p
[TGSI_QUAD_SIZE
],
2309 const uint faces
[TGSI_QUAD_SIZE
],
2310 const int8_t *offset
,
2312 const float dudx
, const float dvdx
,
2313 const float dudy
, const float dvdy
,
2314 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2316 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2318 // ??? Won't the image filters blow up if level is negative?
2319 const unsigned level0
= level
> 0 ? level
: 0;
2320 const float scaling
= 1.0f
/ (1 << level0
);
2321 const int width
= u_minify(texture
->width0
, level0
);
2322 const int height
= u_minify(texture
->height0
, level0
);
2323 struct img_filter_args args
;
2324 const float ux
= dudx
* scaling
;
2325 const float vx
= dvdx
* scaling
;
2326 const float uy
= dudy
* scaling
;
2327 const float vy
= dvdy
* scaling
;
2329 /* compute ellipse coefficients to bound the region:
2330 * A*x*x + B*x*y + C*y*y = F.
2332 float A
= vx
*vx
+vy
*vy
+1;
2333 float B
= -2*(ux
*vx
+uy
*vy
);
2334 float C
= ux
*ux
+uy
*uy
+1;
2335 float F
= A
*C
-B
*B
/4.0f
;
2337 /* check if it is an ellipse */
2338 /* assert(F > 0.0); */
2340 /* Compute the ellipse's (u,v) bounding box in texture space */
2341 const float d
= -B
*B
+4.0f
*C
*A
;
2342 const float box_u
= 2.0f
/ d
* sqrtf(d
*C
*F
); /* box_u -> half of bbox with */
2343 const float box_v
= 2.0f
/ d
* sqrtf(A
*d
*F
); /* box_v -> half of bbox height */
2345 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2346 float s_buffer
[TGSI_QUAD_SIZE
];
2347 float t_buffer
[TGSI_QUAD_SIZE
];
2348 float weight_buffer
[TGSI_QUAD_SIZE
];
2351 /* For each quad, the du and dx values are the same and so the ellipse is
2352 * also the same. Note that texel/image access can only be performed using
2353 * a quad, i.e. it is not possible to get the pixel value for a single
2354 * tex coord. In order to have a better performance, the access is buffered
2355 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2356 * full, then the pixel values are read from the image.
2358 const float ddq
= 2 * A
;
2360 /* Scale ellipse formula to directly index the Filter Lookup Table.
2361 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2363 const double formScale
= (double) (WEIGHT_LUT_SIZE
- 1) / F
;
2367 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2370 args
.offset
= offset
;
2372 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2373 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2374 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2375 * value, q, is less than F, we're inside the ellipse
2377 const float tex_u
= -0.5F
+ s
[j
] * texture
->width0
* scaling
;
2378 const float tex_v
= -0.5F
+ t
[j
] * texture
->height0
* scaling
;
2380 const int u0
= (int) floorf(tex_u
- box_u
);
2381 const int u1
= (int) ceilf(tex_u
+ box_u
);
2382 const int v0
= (int) floorf(tex_v
- box_v
);
2383 const int v1
= (int) ceilf(tex_v
+ box_v
);
2384 const float U
= u0
- tex_u
;
2386 float num
[4] = {0.0F
, 0.0F
, 0.0F
, 0.0F
};
2387 unsigned buffer_next
= 0;
2390 args
.face_id
= faces
[j
];
2392 for (v
= v0
; v
<= v1
; ++v
) {
2393 const float V
= v
- tex_v
;
2394 float dq
= A
* (2 * U
+ 1) + B
* V
;
2395 float q
= (C
* V
+ B
* U
) * V
+ A
* U
* U
;
2398 for (u
= u0
; u
<= u1
; ++u
) {
2399 /* Note that the ellipse has been pre-scaled so F =
2400 * WEIGHT_LUT_SIZE - 1
2402 if (q
< WEIGHT_LUT_SIZE
) {
2403 /* as a LUT is used, q must never be negative;
2404 * should not happen, though
2406 const int qClamped
= q
>= 0.0F
? q
: 0;
2407 const float weight
= weightLut
[qClamped
];
2409 weight_buffer
[buffer_next
] = weight
;
2410 s_buffer
[buffer_next
] = u
/ ((float) width
);
2411 t_buffer
[buffer_next
] = v
/ ((float) height
);
2414 if (buffer_next
== TGSI_QUAD_SIZE
) {
2415 /* 4 texel coords are in the buffer -> read it now */
2417 /* it is assumed that samp->min_img_filter is set to
2418 * img_filter_2d_nearest or one of the
2419 * accelerated img_filter_2d_nearest_XXX functions.
2421 for (jj
= 0; jj
< buffer_next
; jj
++) {
2422 args
.s
= s_buffer
[jj
];
2423 args
.t
= t_buffer
[jj
];
2425 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2426 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2427 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2428 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2429 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2442 /* if the tex coord buffer contains unread values, we will read
2445 if (buffer_next
> 0) {
2447 /* it is assumed that samp->min_img_filter is set to
2448 * img_filter_2d_nearest or one of the
2449 * accelerated img_filter_2d_nearest_XXX functions.
2451 for (jj
= 0; jj
< buffer_next
; jj
++) {
2452 args
.s
= s_buffer
[jj
];
2453 args
.t
= t_buffer
[jj
];
2455 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2456 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2457 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2458 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2459 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2464 /* Reaching this place would mean that no pixels intersected
2465 * the ellipse. This should never happen because the filter
2466 * we use always intersects at least one pixel.
2473 /* not enough pixels in resampling, resort to direct interpolation */
2477 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][j
]);
2479 num
[0] = rgba_temp
[0][j
];
2480 num
[1] = rgba_temp
[1][j
];
2481 num
[2] = rgba_temp
[2][j
];
2482 num
[3] = rgba_temp
[3][j
];
2485 rgba
[0][j
] = num
[0] / den
;
2486 rgba
[1][j
] = num
[1] / den
;
2487 rgba
[2][j
] = num
[2] / den
;
2488 rgba
[3][j
] = num
[3] / den
;
2494 * Get mip level relative to base level for linear mip filter
2497 mip_rel_level_linear_aniso(const struct sp_sampler_view
*sp_sview
,
2498 const struct sp_sampler
*sp_samp
,
2499 const float lod
[TGSI_QUAD_SIZE
],
2500 float level
[TGSI_QUAD_SIZE
])
2502 mip_rel_level_linear(sp_sview
, sp_samp
, lod
, level
);
2506 * Sample 2D texture using an anisotropic filter.
2509 mip_filter_linear_aniso(const struct sp_sampler_view
*sp_sview
,
2510 const struct sp_sampler
*sp_samp
,
2511 img_filter_func min_filter
,
2512 img_filter_func mag_filter
,
2513 const float s
[TGSI_QUAD_SIZE
],
2514 const float t
[TGSI_QUAD_SIZE
],
2515 const float p
[TGSI_QUAD_SIZE
],
2516 const float c0
[TGSI_QUAD_SIZE
],
2517 const float lod_in
[TGSI_QUAD_SIZE
],
2518 const struct filter_args
*filt_args
,
2519 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2521 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2522 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2525 float lod
[TGSI_QUAD_SIZE
];
2527 const float s_to_u
= u_minify(texture
->width0
, psview
->u
.tex
.first_level
);
2528 const float t_to_v
= u_minify(texture
->height0
, psview
->u
.tex
.first_level
);
2529 const float dudx
= (s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2530 const float dudy
= (s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2531 const float dvdx
= (t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2532 const float dvdy
= (t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2533 struct img_filter_args args
;
2535 args
.offset
= filt_args
->offset
;
2537 if (filt_args
->control
== TGSI_SAMPLER_LOD_BIAS
||
2538 filt_args
->control
== TGSI_SAMPLER_LOD_NONE
||
2540 filt_args
->control
== TGSI_SAMPLER_DERIVS_EXPLICIT
) {
2541 /* note: instead of working with Px and Py, we will use the
2542 * squared length instead, to avoid sqrt.
2544 const float Px2
= dudx
* dudx
+ dvdx
* dvdx
;
2545 const float Py2
= dudy
* dudy
+ dvdy
* dvdy
;
2550 const float maxEccentricity
= sp_samp
->base
.max_anisotropy
* sp_samp
->base
.max_anisotropy
;
2561 /* if the eccentricity of the ellipse is too big, scale up the shorter
2562 * of the two vectors to limit the maximum amount of work per pixel
2565 if (e
> maxEccentricity
) {
2566 /* float s=e / maxEccentricity;
2570 Pmin2
= Pmax2
/ maxEccentricity
;
2573 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2574 * this since 0.5*log(x) = log(sqrt(x))
2576 lambda
= 0.5F
* util_fast_log2(Pmin2
) + sp_samp
->base
.lod_bias
;
2577 compute_lod(&sp_samp
->base
, filt_args
->control
, lambda
, lod_in
, lod
);
2580 assert(filt_args
->control
== TGSI_SAMPLER_LOD_EXPLICIT
||
2581 filt_args
->control
== TGSI_SAMPLER_LOD_ZERO
);
2582 compute_lod(&sp_samp
->base
, filt_args
->control
, sp_samp
->base
.lod_bias
, lod_in
, lod
);
2585 /* XXX: Take into account all lod values.
2588 level0
= psview
->u
.tex
.first_level
+ (int)lambda
;
2590 /* If the ellipse covers the whole image, we can
2591 * simply return the average of the whole image.
2593 if (level0
>= (int) psview
->u
.tex
.last_level
) {
2595 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2599 args
.level
= psview
->u
.tex
.last_level
;
2600 args
.face_id
= filt_args
->faces
[j
];
2602 * XXX: we overwrote any linear filter with nearest, so this
2603 * isn't right (albeit if last level is 1x1 and no border it
2604 * will work just the same).
2606 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2610 /* don't bother interpolating between multiple LODs; it doesn't
2611 * seem to be worth the extra running time.
2613 img_filter_2d_ewa(sp_sview
, sp_samp
, min_filter
, mag_filter
,
2614 s
, t
, p
, filt_args
->faces
, filt_args
->offset
,
2615 level0
, dudx
, dvdx
, dudy
, dvdy
, rgba
);
2619 print_sample_4(__FUNCTION__
, rgba
);
2624 * Get mip level relative to base level for linear mip filter
2627 mip_rel_level_linear_2d_linear_repeat_POT(
2628 const struct sp_sampler_view
*sp_sview
,
2629 const struct sp_sampler
*sp_samp
,
2630 const float lod
[TGSI_QUAD_SIZE
],
2631 float level
[TGSI_QUAD_SIZE
])
2633 mip_rel_level_linear(sp_sview
, sp_samp
, lod
, level
);
2637 * Specialized version of mip_filter_linear with hard-wired calls to
2638 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2641 mip_filter_linear_2d_linear_repeat_POT(
2642 const struct sp_sampler_view
*sp_sview
,
2643 const struct sp_sampler
*sp_samp
,
2644 img_filter_func min_filter
,
2645 img_filter_func mag_filter
,
2646 const float s
[TGSI_QUAD_SIZE
],
2647 const float t
[TGSI_QUAD_SIZE
],
2648 const float p
[TGSI_QUAD_SIZE
],
2649 const float c0
[TGSI_QUAD_SIZE
],
2650 const float lod_in
[TGSI_QUAD_SIZE
],
2651 const struct filter_args
*filt_args
,
2652 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2654 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2656 float lod
[TGSI_QUAD_SIZE
];
2658 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2660 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2661 const int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
2662 struct img_filter_args args
;
2663 /* Catches both negative and large values of level0:
2668 args
.face_id
= filt_args
->faces
[j
];
2669 args
.offset
= filt_args
->offset
;
2670 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2671 if ((unsigned)level0
>= psview
->u
.tex
.last_level
) {
2673 args
.level
= psview
->u
.tex
.first_level
;
2675 args
.level
= psview
->u
.tex
.last_level
;
2676 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
,
2681 const float levelBlend
= frac(lod
[j
]);
2682 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2685 args
.level
= level0
;
2686 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
2687 args
.level
= level0
+1;
2688 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
2690 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
2691 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2696 print_sample_4(__FUNCTION__
, rgba
);
2700 static const struct sp_filter_funcs funcs_linear
= {
2701 mip_rel_level_linear
,
2705 static const struct sp_filter_funcs funcs_nearest
= {
2706 mip_rel_level_nearest
,
2710 static const struct sp_filter_funcs funcs_none
= {
2715 static const struct sp_filter_funcs funcs_none_no_filter_select
= {
2716 mip_rel_level_none_no_filter_select
,
2717 mip_filter_none_no_filter_select
2720 static const struct sp_filter_funcs funcs_linear_aniso
= {
2721 mip_rel_level_linear_aniso
,
2722 mip_filter_linear_aniso
2725 static const struct sp_filter_funcs funcs_linear_2d_linear_repeat_POT
= {
2726 mip_rel_level_linear_2d_linear_repeat_POT
,
2727 mip_filter_linear_2d_linear_repeat_POT
2731 * Do shadow/depth comparisons.
2734 sample_compare(const struct sp_sampler_view
*sp_sview
,
2735 const struct sp_sampler
*sp_samp
,
2736 const float s
[TGSI_QUAD_SIZE
],
2737 const float t
[TGSI_QUAD_SIZE
],
2738 const float p
[TGSI_QUAD_SIZE
],
2739 const float c0
[TGSI_QUAD_SIZE
],
2740 const float c1
[TGSI_QUAD_SIZE
],
2741 enum tgsi_sampler_control control
,
2742 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2744 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
2746 int k
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2748 const struct util_format_description
*format_desc
=
2749 util_format_description(sp_sview
->base
.format
);
2750 /* not entirely sure we couldn't end up with non-valid swizzle here */
2751 const unsigned chan_type
=
2752 format_desc
->swizzle
[0] <= PIPE_SWIZZLE_W
?
2753 format_desc
->channel
[format_desc
->swizzle
[0]].type
:
2754 UTIL_FORMAT_TYPE_FLOAT
;
2755 const bool is_gather
= (control
== TGSI_SAMPLER_GATHER
);
2758 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2759 * for 2D Array texture we need to use the 'c0' (aka Q).
2760 * When we sampled the depth texture, the depth value was put into all
2761 * RGBA channels. We look at the red channel here.
2764 if (sp_sview
->base
.target
== PIPE_TEXTURE_2D_ARRAY
||
2765 sp_sview
->base
.target
== PIPE_TEXTURE_CUBE
) {
2770 } else if (sp_sview
->base
.target
== PIPE_TEXTURE_CUBE_ARRAY
) {
2782 if (chan_type
!= UTIL_FORMAT_TYPE_FLOAT
) {
2784 * clamping is a result of conversion to texture format, hence
2785 * doesn't happen with floats. Technically also should do comparison
2786 * in texture format (quantization!).
2788 pc
[0] = CLAMP(pc
[0], 0.0F
, 1.0F
);
2789 pc
[1] = CLAMP(pc
[1], 0.0F
, 1.0F
);
2790 pc
[2] = CLAMP(pc
[2], 0.0F
, 1.0F
);
2791 pc
[3] = CLAMP(pc
[3], 0.0F
, 1.0F
);
2794 for (v
= 0; v
< (is_gather
? TGSI_NUM_CHANNELS
: 1); v
++) {
2795 /* compare four texcoords vs. four texture samples */
2796 switch (sampler
->compare_func
) {
2797 case PIPE_FUNC_LESS
:
2798 k
[v
][0] = pc
[0] < rgba
[v
][0];
2799 k
[v
][1] = pc
[1] < rgba
[v
][1];
2800 k
[v
][2] = pc
[2] < rgba
[v
][2];
2801 k
[v
][3] = pc
[3] < rgba
[v
][3];
2803 case PIPE_FUNC_LEQUAL
:
2804 k
[v
][0] = pc
[0] <= rgba
[v
][0];
2805 k
[v
][1] = pc
[1] <= rgba
[v
][1];
2806 k
[v
][2] = pc
[2] <= rgba
[v
][2];
2807 k
[v
][3] = pc
[3] <= rgba
[v
][3];
2809 case PIPE_FUNC_GREATER
:
2810 k
[v
][0] = pc
[0] > rgba
[v
][0];
2811 k
[v
][1] = pc
[1] > rgba
[v
][1];
2812 k
[v
][2] = pc
[2] > rgba
[v
][2];
2813 k
[v
][3] = pc
[3] > rgba
[v
][3];
2815 case PIPE_FUNC_GEQUAL
:
2816 k
[v
][0] = pc
[0] >= rgba
[v
][0];
2817 k
[v
][1] = pc
[1] >= rgba
[v
][1];
2818 k
[v
][2] = pc
[2] >= rgba
[v
][2];
2819 k
[v
][3] = pc
[3] >= rgba
[v
][3];
2821 case PIPE_FUNC_EQUAL
:
2822 k
[v
][0] = pc
[0] == rgba
[v
][0];
2823 k
[v
][1] = pc
[1] == rgba
[v
][1];
2824 k
[v
][2] = pc
[2] == rgba
[v
][2];
2825 k
[v
][3] = pc
[3] == rgba
[v
][3];
2827 case PIPE_FUNC_NOTEQUAL
:
2828 k
[v
][0] = pc
[0] != rgba
[v
][0];
2829 k
[v
][1] = pc
[1] != rgba
[v
][1];
2830 k
[v
][2] = pc
[2] != rgba
[v
][2];
2831 k
[v
][3] = pc
[3] != rgba
[v
][3];
2833 case PIPE_FUNC_ALWAYS
:
2834 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 1;
2836 case PIPE_FUNC_NEVER
:
2837 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2840 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2847 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2848 for (v
= 0; v
< TGSI_NUM_CHANNELS
; v
++) {
2849 rgba
[v
][j
] = k
[v
][j
];
2853 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2854 rgba
[0][j
] = k
[0][j
];
2855 rgba
[1][j
] = k
[0][j
];
2856 rgba
[2][j
] = k
[0][j
];
2863 do_swizzling(const struct pipe_sampler_view
*sview
,
2864 float in
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
2865 float out
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2868 const unsigned swizzle_r
= sview
->swizzle_r
;
2869 const unsigned swizzle_g
= sview
->swizzle_g
;
2870 const unsigned swizzle_b
= sview
->swizzle_b
;
2871 const unsigned swizzle_a
= sview
->swizzle_a
;
2872 float oneval
= util_format_is_pure_integer(sview
->format
) ? uif(1) : 1.0f
;
2874 switch (swizzle_r
) {
2875 case PIPE_SWIZZLE_0
:
2876 for (j
= 0; j
< 4; j
++)
2879 case PIPE_SWIZZLE_1
:
2880 for (j
= 0; j
< 4; j
++)
2884 assert(swizzle_r
< 4);
2885 for (j
= 0; j
< 4; j
++)
2886 out
[0][j
] = in
[swizzle_r
][j
];
2889 switch (swizzle_g
) {
2890 case PIPE_SWIZZLE_0
:
2891 for (j
= 0; j
< 4; j
++)
2894 case PIPE_SWIZZLE_1
:
2895 for (j
= 0; j
< 4; j
++)
2899 assert(swizzle_g
< 4);
2900 for (j
= 0; j
< 4; j
++)
2901 out
[1][j
] = in
[swizzle_g
][j
];
2904 switch (swizzle_b
) {
2905 case PIPE_SWIZZLE_0
:
2906 for (j
= 0; j
< 4; j
++)
2909 case PIPE_SWIZZLE_1
:
2910 for (j
= 0; j
< 4; j
++)
2914 assert(swizzle_b
< 4);
2915 for (j
= 0; j
< 4; j
++)
2916 out
[2][j
] = in
[swizzle_b
][j
];
2919 switch (swizzle_a
) {
2920 case PIPE_SWIZZLE_0
:
2921 for (j
= 0; j
< 4; j
++)
2924 case PIPE_SWIZZLE_1
:
2925 for (j
= 0; j
< 4; j
++)
2929 assert(swizzle_a
< 4);
2930 for (j
= 0; j
< 4; j
++)
2931 out
[3][j
] = in
[swizzle_a
][j
];
2936 static wrap_nearest_func
2937 get_nearest_unorm_wrap(unsigned mode
)
2940 case PIPE_TEX_WRAP_CLAMP
:
2941 return wrap_nearest_unorm_clamp
;
2942 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2943 return wrap_nearest_unorm_clamp_to_edge
;
2944 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2945 return wrap_nearest_unorm_clamp_to_border
;
2947 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
2948 return wrap_nearest_unorm_clamp
;
2953 static wrap_nearest_func
2954 get_nearest_wrap(unsigned mode
)
2957 case PIPE_TEX_WRAP_REPEAT
:
2958 return wrap_nearest_repeat
;
2959 case PIPE_TEX_WRAP_CLAMP
:
2960 return wrap_nearest_clamp
;
2961 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2962 return wrap_nearest_clamp_to_edge
;
2963 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2964 return wrap_nearest_clamp_to_border
;
2965 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2966 return wrap_nearest_mirror_repeat
;
2967 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
2968 return wrap_nearest_mirror_clamp
;
2969 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
2970 return wrap_nearest_mirror_clamp_to_edge
;
2971 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
2972 return wrap_nearest_mirror_clamp_to_border
;
2975 return wrap_nearest_repeat
;
2980 static wrap_linear_func
2981 get_linear_unorm_wrap(unsigned mode
)
2984 case PIPE_TEX_WRAP_CLAMP
:
2985 return wrap_linear_unorm_clamp
;
2986 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2987 return wrap_linear_unorm_clamp_to_edge
;
2988 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2989 return wrap_linear_unorm_clamp_to_border
;
2991 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
2992 return wrap_linear_unorm_clamp
;
2997 static wrap_linear_func
2998 get_linear_wrap(unsigned mode
)
3001 case PIPE_TEX_WRAP_REPEAT
:
3002 return wrap_linear_repeat
;
3003 case PIPE_TEX_WRAP_CLAMP
:
3004 return wrap_linear_clamp
;
3005 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
3006 return wrap_linear_clamp_to_edge
;
3007 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
3008 return wrap_linear_clamp_to_border
;
3009 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
3010 return wrap_linear_mirror_repeat
;
3011 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
3012 return wrap_linear_mirror_clamp
;
3013 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
3014 return wrap_linear_mirror_clamp_to_edge
;
3015 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
3016 return wrap_linear_mirror_clamp_to_border
;
3019 return wrap_linear_repeat
;
3025 * Is swizzling needed for the given state key?
3028 any_swizzle(const struct pipe_sampler_view
*view
)
3030 return (view
->swizzle_r
!= PIPE_SWIZZLE_X
||
3031 view
->swizzle_g
!= PIPE_SWIZZLE_Y
||
3032 view
->swizzle_b
!= PIPE_SWIZZLE_Z
||
3033 view
->swizzle_a
!= PIPE_SWIZZLE_W
);
3037 static img_filter_func
3038 get_img_filter(const struct sp_sampler_view
*sp_sview
,
3039 const struct pipe_sampler_state
*sampler
,
3040 unsigned filter
, bool gather
)
3042 switch (sp_sview
->base
.target
) {
3044 case PIPE_TEXTURE_1D
:
3045 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3046 return img_filter_1d_nearest
;
3048 return img_filter_1d_linear
;
3050 case PIPE_TEXTURE_1D_ARRAY
:
3051 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3052 return img_filter_1d_array_nearest
;
3054 return img_filter_1d_array_linear
;
3056 case PIPE_TEXTURE_2D
:
3057 case PIPE_TEXTURE_RECT
:
3058 /* Try for fast path:
3060 if (!gather
&& sp_sview
->pot2d
&&
3061 sampler
->wrap_s
== sampler
->wrap_t
&&
3062 sampler
->normalized_coords
)
3064 switch (sampler
->wrap_s
) {
3065 case PIPE_TEX_WRAP_REPEAT
:
3067 case PIPE_TEX_FILTER_NEAREST
:
3068 return img_filter_2d_nearest_repeat_POT
;
3069 case PIPE_TEX_FILTER_LINEAR
:
3070 return img_filter_2d_linear_repeat_POT
;
3075 case PIPE_TEX_WRAP_CLAMP
:
3077 case PIPE_TEX_FILTER_NEAREST
:
3078 return img_filter_2d_nearest_clamp_POT
;
3084 /* Otherwise use default versions:
3086 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3087 return img_filter_2d_nearest
;
3089 return img_filter_2d_linear
;
3091 case PIPE_TEXTURE_2D_ARRAY
:
3092 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3093 return img_filter_2d_array_nearest
;
3095 return img_filter_2d_array_linear
;
3097 case PIPE_TEXTURE_CUBE
:
3098 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3099 return img_filter_cube_nearest
;
3101 return img_filter_cube_linear
;
3103 case PIPE_TEXTURE_CUBE_ARRAY
:
3104 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3105 return img_filter_cube_array_nearest
;
3107 return img_filter_cube_array_linear
;
3109 case PIPE_TEXTURE_3D
:
3110 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3111 return img_filter_3d_nearest
;
3113 return img_filter_3d_linear
;
3117 return img_filter_1d_nearest
;
3122 * Get mip filter funcs, and optionally both img min filter and img mag
3123 * filter. Note that both img filter function pointers must be either non-NULL
3127 get_filters(const struct sp_sampler_view
*sp_sview
,
3128 const struct sp_sampler
*sp_samp
,
3129 const enum tgsi_sampler_control control
,
3130 const struct sp_filter_funcs
**funcs
,
3131 img_filter_func
*min
,
3132 img_filter_func
*mag
)
3135 if (control
== TGSI_SAMPLER_GATHER
) {
3136 *funcs
= &funcs_nearest
;
3138 *min
= get_img_filter(sp_sview
, &sp_samp
->base
,
3139 PIPE_TEX_FILTER_LINEAR
, true);
3141 } else if (sp_sview
->pot2d
& sp_samp
->min_mag_equal_repeat_linear
) {
3142 *funcs
= &funcs_linear_2d_linear_repeat_POT
;
3144 *funcs
= sp_samp
->filter_funcs
;
3147 *min
= get_img_filter(sp_sview
, &sp_samp
->base
,
3148 sp_samp
->min_img_filter
, false);
3149 if (sp_samp
->min_mag_equal
) {
3152 *mag
= get_img_filter(sp_sview
, &sp_samp
->base
,
3153 sp_samp
->base
.mag_img_filter
, false);
3160 sample_mip(const struct sp_sampler_view
*sp_sview
,
3161 const struct sp_sampler
*sp_samp
,
3162 const float s
[TGSI_QUAD_SIZE
],
3163 const float t
[TGSI_QUAD_SIZE
],
3164 const float p
[TGSI_QUAD_SIZE
],
3165 const float c0
[TGSI_QUAD_SIZE
],
3166 const float lod
[TGSI_QUAD_SIZE
],
3167 const struct filter_args
*filt_args
,
3168 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3170 const struct sp_filter_funcs
*funcs
= NULL
;
3171 img_filter_func min_img_filter
= NULL
;
3172 img_filter_func mag_img_filter
= NULL
;
3174 get_filters(sp_sview
, sp_samp
, filt_args
->control
,
3175 &funcs
, &min_img_filter
, &mag_img_filter
);
3177 funcs
->filter(sp_sview
, sp_samp
, min_img_filter
, mag_img_filter
,
3178 s
, t
, p
, c0
, lod
, filt_args
, rgba
);
3180 if (sp_samp
->base
.compare_mode
!= PIPE_TEX_COMPARE_NONE
) {
3181 sample_compare(sp_sview
, sp_samp
, s
, t
, p
, c0
,
3182 lod
, filt_args
->control
, rgba
);
3185 if (sp_sview
->need_swizzle
&& filt_args
->control
!= TGSI_SAMPLER_GATHER
) {
3186 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3187 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
3188 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
3195 * This function uses cube texture coordinates to choose a face of a cube and
3196 * computes the 2D cube face coordinates. Puts face info into the sampler
3200 convert_cube(const struct sp_sampler_view
*sp_sview
,
3201 const struct sp_sampler
*sp_samp
,
3202 const float s
[TGSI_QUAD_SIZE
],
3203 const float t
[TGSI_QUAD_SIZE
],
3204 const float p
[TGSI_QUAD_SIZE
],
3205 const float c0
[TGSI_QUAD_SIZE
],
3206 float ssss
[TGSI_QUAD_SIZE
],
3207 float tttt
[TGSI_QUAD_SIZE
],
3208 float pppp
[TGSI_QUAD_SIZE
],
3209 uint faces
[TGSI_QUAD_SIZE
])
3219 direction target sc tc ma
3220 ---------- ------------------------------- --- --- ---
3221 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
3222 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
3223 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
3224 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
3225 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
3226 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
3229 /* Choose the cube face and compute new s/t coords for the 2D face.
3231 * Use the same cube face for all four pixels in the quad.
3233 * This isn't ideal, but if we want to use a different cube face
3234 * per pixel in the quad, we'd have to also compute the per-face
3235 * LOD here too. That's because the four post-face-selection
3236 * texcoords are no longer related to each other (they're
3237 * per-face!) so we can't use subtraction to compute the partial
3238 * deriviates to compute the LOD. Doing so (near cube edges
3239 * anyway) gives us pretty much random values.
3242 /* use the average of the four pixel's texcoords to choose the face */
3243 const float rx
= 0.25F
* (s
[0] + s
[1] + s
[2] + s
[3]);
3244 const float ry
= 0.25F
* (t
[0] + t
[1] + t
[2] + t
[3]);
3245 const float rz
= 0.25F
* (p
[0] + p
[1] + p
[2] + p
[3]);
3246 const float arx
= fabsf(rx
), ary
= fabsf(ry
), arz
= fabsf(rz
);
3248 if (arx
>= ary
&& arx
>= arz
) {
3249 const float sign
= (rx
>= 0.0F
) ? 1.0F
: -1.0F
;
3250 const uint face
= (rx
>= 0.0F
) ?
3251 PIPE_TEX_FACE_POS_X
: PIPE_TEX_FACE_NEG_X
;
3252 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3253 const float ima
= -0.5F
/ fabsf(s
[j
]);
3254 ssss
[j
] = sign
* p
[j
] * ima
+ 0.5F
;
3255 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3259 else if (ary
>= arx
&& ary
>= arz
) {
3260 const float sign
= (ry
>= 0.0F
) ? 1.0F
: -1.0F
;
3261 const uint face
= (ry
>= 0.0F
) ?
3262 PIPE_TEX_FACE_POS_Y
: PIPE_TEX_FACE_NEG_Y
;
3263 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3264 const float ima
= -0.5F
/ fabsf(t
[j
]);
3265 ssss
[j
] = -s
[j
] * ima
+ 0.5F
;
3266 tttt
[j
] = sign
* -p
[j
] * ima
+ 0.5F
;
3271 const float sign
= (rz
>= 0.0F
) ? 1.0F
: -1.0F
;
3272 const uint face
= (rz
>= 0.0F
) ?
3273 PIPE_TEX_FACE_POS_Z
: PIPE_TEX_FACE_NEG_Z
;
3274 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3275 const float ima
= -0.5F
/ fabsf(p
[j
]);
3276 ssss
[j
] = sign
* -s
[j
] * ima
+ 0.5F
;
3277 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3286 sp_get_dims(const struct sp_sampler_view
*sp_sview
,
3290 const struct pipe_sampler_view
*view
= &sp_sview
->base
;
3291 const struct pipe_resource
*texture
= view
->texture
;
3293 if (view
->target
== PIPE_BUFFER
) {
3294 dims
[0] = view
->u
.buf
.size
/ util_format_get_blocksize(view
->format
);
3295 /* the other values are undefined, but let's avoid potential valgrind
3298 dims
[1] = dims
[2] = dims
[3] = 0;
3302 /* undefined according to EXT_gpu_program */
3303 level
+= view
->u
.tex
.first_level
;
3304 if (level
> view
->u
.tex
.last_level
)
3307 dims
[3] = view
->u
.tex
.last_level
- view
->u
.tex
.first_level
+ 1;
3308 dims
[0] = u_minify(texture
->width0
, level
);
3310 switch (view
->target
) {
3311 case PIPE_TEXTURE_1D_ARRAY
:
3312 dims
[1] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3314 case PIPE_TEXTURE_1D
:
3316 case PIPE_TEXTURE_2D_ARRAY
:
3317 dims
[2] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3319 case PIPE_TEXTURE_2D
:
3320 case PIPE_TEXTURE_CUBE
:
3321 case PIPE_TEXTURE_RECT
:
3322 dims
[1] = u_minify(texture
->height0
, level
);
3324 case PIPE_TEXTURE_3D
:
3325 dims
[1] = u_minify(texture
->height0
, level
);
3326 dims
[2] = u_minify(texture
->depth0
, level
);
3328 case PIPE_TEXTURE_CUBE_ARRAY
:
3329 dims
[1] = u_minify(texture
->height0
, level
);
3330 dims
[2] = (view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1) / 6;
3333 assert(!"unexpected texture target in sp_get_dims()");
3339 * This function is only used for getting unfiltered texels via the
3340 * TXF opcode. The GL spec says that out-of-bounds texel fetches
3341 * produce undefined results. Instead of crashing, lets just clamp
3342 * coords to the texture image size.
3345 sp_get_texels(const struct sp_sampler_view
*sp_sview
,
3346 const int v_i
[TGSI_QUAD_SIZE
],
3347 const int v_j
[TGSI_QUAD_SIZE
],
3348 const int v_k
[TGSI_QUAD_SIZE
],
3349 const int lod
[TGSI_QUAD_SIZE
],
3350 const int8_t offset
[3],
3351 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3353 union tex_tile_address addr
;
3354 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
3357 /* TODO write a better test for LOD */
3358 const unsigned level
=
3359 sp_sview
->base
.target
== PIPE_BUFFER
? 0 :
3360 CLAMP(lod
[0] + sp_sview
->base
.u
.tex
.first_level
,
3361 sp_sview
->base
.u
.tex
.first_level
,
3362 sp_sview
->base
.u
.tex
.last_level
);
3363 const int width
= u_minify(texture
->width0
, level
);
3364 const int height
= u_minify(texture
->height0
, level
);
3365 const int depth
= u_minify(texture
->depth0
, level
);
3366 unsigned elem_size
, first_element
, last_element
;
3369 addr
.bits
.level
= level
;
3371 switch (sp_sview
->base
.target
) {
3373 elem_size
= util_format_get_blocksize(sp_sview
->base
.format
);
3374 first_element
= sp_sview
->base
.u
.buf
.offset
/ elem_size
;
3375 last_element
= (sp_sview
->base
.u
.buf
.offset
+
3376 sp_sview
->base
.u
.buf
.size
) / elem_size
- 1;
3377 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3378 const int x
= CLAMP(v_i
[j
] + offset
[0] +
3382 tx
= get_texel_buffer_no_border(sp_sview
, addr
, x
, elem_size
);
3383 for (c
= 0; c
< 4; c
++) {
3388 case PIPE_TEXTURE_1D
:
3389 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3390 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3391 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
,
3392 sp_sview
->base
.u
.tex
.first_layer
);
3393 for (c
= 0; c
< 4; c
++) {
3398 case PIPE_TEXTURE_1D_ARRAY
:
3399 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3400 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3401 const int y
= CLAMP(v_j
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3402 sp_sview
->base
.u
.tex
.last_layer
);
3403 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
3404 for (c
= 0; c
< 4; c
++) {
3409 case PIPE_TEXTURE_2D
:
3410 case PIPE_TEXTURE_RECT
:
3411 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3412 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3413 const int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3414 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
,
3415 sp_sview
->base
.u
.tex
.first_layer
);
3416 for (c
= 0; c
< 4; c
++) {
3421 case PIPE_TEXTURE_2D_ARRAY
:
3422 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3423 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3424 const int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3425 const int layer
= CLAMP(v_k
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3426 sp_sview
->base
.u
.tex
.last_layer
);
3427 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
3428 for (c
= 0; c
< 4; c
++) {
3433 case PIPE_TEXTURE_3D
:
3434 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3435 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3436 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3437 int z
= CLAMP(v_k
[j
] + offset
[2], 0, depth
- 1);
3438 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, z
);
3439 for (c
= 0; c
< 4; c
++) {
3444 case PIPE_TEXTURE_CUBE
: /* TXF can't work on CUBE according to spec */
3445 case PIPE_TEXTURE_CUBE_ARRAY
:
3447 assert(!"Unknown or CUBE texture type in TXF processing\n");
3451 if (sp_sview
->need_swizzle
) {
3452 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3453 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
3454 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
3460 softpipe_create_sampler_state(struct pipe_context
*pipe
,
3461 const struct pipe_sampler_state
*sampler
)
3463 struct sp_sampler
*samp
= CALLOC_STRUCT(sp_sampler
);
3465 samp
->base
= *sampler
;
3467 /* Note that (for instance) linear_texcoord_s and
3468 * nearest_texcoord_s may be active at the same time, if the
3469 * sampler min_img_filter differs from its mag_img_filter.
3471 if (sampler
->normalized_coords
) {
3472 samp
->linear_texcoord_s
= get_linear_wrap( sampler
->wrap_s
);
3473 samp
->linear_texcoord_t
= get_linear_wrap( sampler
->wrap_t
);
3474 samp
->linear_texcoord_p
= get_linear_wrap( sampler
->wrap_r
);
3476 samp
->nearest_texcoord_s
= get_nearest_wrap( sampler
->wrap_s
);
3477 samp
->nearest_texcoord_t
= get_nearest_wrap( sampler
->wrap_t
);
3478 samp
->nearest_texcoord_p
= get_nearest_wrap( sampler
->wrap_r
);
3481 samp
->linear_texcoord_s
= get_linear_unorm_wrap( sampler
->wrap_s
);
3482 samp
->linear_texcoord_t
= get_linear_unorm_wrap( sampler
->wrap_t
);
3483 samp
->linear_texcoord_p
= get_linear_unorm_wrap( sampler
->wrap_r
);
3485 samp
->nearest_texcoord_s
= get_nearest_unorm_wrap( sampler
->wrap_s
);
3486 samp
->nearest_texcoord_t
= get_nearest_unorm_wrap( sampler
->wrap_t
);
3487 samp
->nearest_texcoord_p
= get_nearest_unorm_wrap( sampler
->wrap_r
);
3490 samp
->min_img_filter
= sampler
->min_img_filter
;
3492 switch (sampler
->min_mip_filter
) {
3493 case PIPE_TEX_MIPFILTER_NONE
:
3494 if (sampler
->min_img_filter
== sampler
->mag_img_filter
)
3495 samp
->filter_funcs
= &funcs_none_no_filter_select
;
3497 samp
->filter_funcs
= &funcs_none
;
3500 case PIPE_TEX_MIPFILTER_NEAREST
:
3501 samp
->filter_funcs
= &funcs_nearest
;
3504 case PIPE_TEX_MIPFILTER_LINEAR
:
3505 if (sampler
->min_img_filter
== sampler
->mag_img_filter
&&
3506 sampler
->normalized_coords
&&
3507 sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
&&
3508 sampler
->wrap_t
== PIPE_TEX_WRAP_REPEAT
&&
3509 sampler
->min_img_filter
== PIPE_TEX_FILTER_LINEAR
&&
3510 sampler
->max_anisotropy
<= 1) {
3511 samp
->min_mag_equal_repeat_linear
= TRUE
;
3513 samp
->filter_funcs
= &funcs_linear
;
3515 /* Anisotropic filtering extension. */
3516 if (sampler
->max_anisotropy
> 1) {
3517 samp
->filter_funcs
= &funcs_linear_aniso
;
3519 /* Override min_img_filter:
3520 * min_img_filter needs to be set to NEAREST since we need to access
3521 * each texture pixel as it is and weight it later; using linear
3522 * filters will have incorrect results.
3523 * By setting the filter to NEAREST here, we can avoid calling the
3524 * generic img_filter_2d_nearest in the anisotropic filter function,
3525 * making it possible to use one of the accelerated implementations
3527 samp
->min_img_filter
= PIPE_TEX_FILTER_NEAREST
;
3529 /* on first access create the lookup table containing the filter weights. */
3531 create_filter_table();
3536 if (samp
->min_img_filter
== sampler
->mag_img_filter
) {
3537 samp
->min_mag_equal
= TRUE
;
3540 return (void *)samp
;
3545 softpipe_get_lambda_func(const struct pipe_sampler_view
*view
,
3546 enum pipe_shader_type shader
)
3548 if (shader
!= PIPE_SHADER_FRAGMENT
)
3549 return compute_lambda_vert
;
3551 switch (view
->target
) {
3553 case PIPE_TEXTURE_1D
:
3554 case PIPE_TEXTURE_1D_ARRAY
:
3555 return compute_lambda_1d
;
3556 case PIPE_TEXTURE_2D
:
3557 case PIPE_TEXTURE_2D_ARRAY
:
3558 case PIPE_TEXTURE_RECT
:
3559 case PIPE_TEXTURE_CUBE
:
3560 case PIPE_TEXTURE_CUBE_ARRAY
:
3561 return compute_lambda_2d
;
3562 case PIPE_TEXTURE_3D
:
3563 return compute_lambda_3d
;
3566 return compute_lambda_1d
;
3571 struct pipe_sampler_view
*
3572 softpipe_create_sampler_view(struct pipe_context
*pipe
,
3573 struct pipe_resource
*resource
,
3574 const struct pipe_sampler_view
*templ
)
3576 struct sp_sampler_view
*sview
= CALLOC_STRUCT(sp_sampler_view
);
3577 const struct softpipe_resource
*spr
= (struct softpipe_resource
*)resource
;
3580 struct pipe_sampler_view
*view
= &sview
->base
;
3582 view
->reference
.count
= 1;
3583 view
->texture
= NULL
;
3584 pipe_resource_reference(&view
->texture
, resource
);
3585 view
->context
= pipe
;
3589 * This is possibly too lenient, but the primary reason is just
3590 * to catch state trackers which forget to initialize this, so
3591 * it only catches clearly impossible view targets.
3593 if (view
->target
!= resource
->target
) {
3594 if (view
->target
== PIPE_TEXTURE_1D
)
3595 assert(resource
->target
== PIPE_TEXTURE_1D_ARRAY
);
3596 else if (view
->target
== PIPE_TEXTURE_1D_ARRAY
)
3597 assert(resource
->target
== PIPE_TEXTURE_1D
);
3598 else if (view
->target
== PIPE_TEXTURE_2D
)
3599 assert(resource
->target
== PIPE_TEXTURE_2D_ARRAY
||
3600 resource
->target
== PIPE_TEXTURE_CUBE
||
3601 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3602 else if (view
->target
== PIPE_TEXTURE_2D_ARRAY
)
3603 assert(resource
->target
== PIPE_TEXTURE_2D
||
3604 resource
->target
== PIPE_TEXTURE_CUBE
||
3605 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3606 else if (view
->target
== PIPE_TEXTURE_CUBE
)
3607 assert(resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
||
3608 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3609 else if (view
->target
== PIPE_TEXTURE_CUBE_ARRAY
)
3610 assert(resource
->target
== PIPE_TEXTURE_CUBE
||
3611 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3617 if (any_swizzle(view
)) {
3618 sview
->need_swizzle
= TRUE
;
3621 sview
->need_cube_convert
= (view
->target
== PIPE_TEXTURE_CUBE
||
3622 view
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3623 sview
->pot2d
= spr
->pot
&&
3624 (view
->target
== PIPE_TEXTURE_2D
||
3625 view
->target
== PIPE_TEXTURE_RECT
);
3627 sview
->xpot
= util_logbase2( resource
->width0
);
3628 sview
->ypot
= util_logbase2( resource
->height0
);
3631 return (struct pipe_sampler_view
*) sview
;
3635 static inline const struct sp_tgsi_sampler
*
3636 sp_tgsi_sampler_cast_c(const struct tgsi_sampler
*sampler
)
3638 return (const struct sp_tgsi_sampler
*)sampler
;
3643 sp_tgsi_get_dims(struct tgsi_sampler
*tgsi_sampler
,
3644 const unsigned sview_index
,
3645 int level
, int dims
[4])
3647 const struct sp_tgsi_sampler
*sp_samp
=
3648 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3650 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3651 /* always have a view here but texture is NULL if no sampler view was set. */
3652 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3653 dims
[0] = dims
[1] = dims
[2] = dims
[3] = 0;
3656 sp_get_dims(&sp_samp
->sp_sview
[sview_index
], level
, dims
);
3661 sp_tgsi_get_samples(struct tgsi_sampler
*tgsi_sampler
,
3662 const unsigned sview_index
,
3663 const unsigned sampler_index
,
3664 const float s
[TGSI_QUAD_SIZE
],
3665 const float t
[TGSI_QUAD_SIZE
],
3666 const float p
[TGSI_QUAD_SIZE
],
3667 const float c0
[TGSI_QUAD_SIZE
],
3668 const float lod
[TGSI_QUAD_SIZE
],
3669 float derivs
[3][2][TGSI_QUAD_SIZE
],
3670 const int8_t offset
[3],
3671 enum tgsi_sampler_control control
,
3672 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3674 const struct sp_tgsi_sampler
*sp_tgsi_samp
=
3675 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3676 const struct sp_sampler_view
*sp_sview
;
3677 const struct sp_sampler
*sp_samp
;
3678 struct filter_args filt_args
;
3680 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3681 assert(sampler_index
< PIPE_MAX_SAMPLERS
);
3682 assert(sp_tgsi_samp
->sp_sampler
[sampler_index
]);
3684 sp_sview
= &sp_tgsi_samp
->sp_sview
[sview_index
];
3685 sp_samp
= sp_tgsi_samp
->sp_sampler
[sampler_index
];
3686 /* always have a view here but texture is NULL if no sampler view was set. */
3687 if (!sp_sview
->base
.texture
) {
3689 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3690 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3697 filt_args
.control
= control
;
3698 filt_args
.offset
= offset
;
3700 if (sp_sview
->need_cube_convert
) {
3701 float cs
[TGSI_QUAD_SIZE
];
3702 float ct
[TGSI_QUAD_SIZE
];
3703 float cp
[TGSI_QUAD_SIZE
];
3704 uint faces
[TGSI_QUAD_SIZE
];
3706 convert_cube(sp_sview
, sp_samp
, s
, t
, p
, c0
, cs
, ct
, cp
, faces
);
3708 filt_args
.faces
= faces
;
3709 sample_mip(sp_sview
, sp_samp
, cs
, ct
, cp
, c0
, lod
, &filt_args
, rgba
);
3711 static const uint zero_faces
[TGSI_QUAD_SIZE
] = {0, 0, 0, 0};
3713 filt_args
.faces
= zero_faces
;
3714 sample_mip(sp_sview
, sp_samp
, s
, t
, p
, c0
, lod
, &filt_args
, rgba
);
3719 sp_tgsi_query_lod(const struct tgsi_sampler
*tgsi_sampler
,
3720 const unsigned sview_index
,
3721 const unsigned sampler_index
,
3722 const float s
[TGSI_QUAD_SIZE
],
3723 const float t
[TGSI_QUAD_SIZE
],
3724 const float p
[TGSI_QUAD_SIZE
],
3725 const float c0
[TGSI_QUAD_SIZE
],
3726 const enum tgsi_sampler_control control
,
3727 float mipmap
[TGSI_QUAD_SIZE
],
3728 float lod
[TGSI_QUAD_SIZE
])
3730 static const float lod_in
[TGSI_QUAD_SIZE
] = { 0.0, 0.0, 0.0, 0.0 };
3732 const struct sp_tgsi_sampler
*sp_tgsi_samp
=
3733 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3734 const struct sp_sampler_view
*sp_sview
;
3735 const struct sp_sampler
*sp_samp
;
3736 const struct sp_filter_funcs
*funcs
;
3739 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3740 assert(sampler_index
< PIPE_MAX_SAMPLERS
);
3741 assert(sp_tgsi_samp
->sp_sampler
[sampler_index
]);
3743 sp_sview
= &sp_tgsi_samp
->sp_sview
[sview_index
];
3744 sp_samp
= sp_tgsi_samp
->sp_sampler
[sampler_index
];
3745 /* always have a view here but texture is NULL if no sampler view was
3747 if (!sp_sview
->base
.texture
) {
3748 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3755 if (sp_sview
->need_cube_convert
) {
3756 float cs
[TGSI_QUAD_SIZE
];
3757 float ct
[TGSI_QUAD_SIZE
];
3758 float cp
[TGSI_QUAD_SIZE
];
3759 uint unused_faces
[TGSI_QUAD_SIZE
];
3761 convert_cube(sp_sview
, sp_samp
, s
, t
, p
, c0
, cs
, ct
, cp
, unused_faces
);
3762 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
3763 cs
, ct
, cp
, lod_in
, control
, lod
);
3765 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
3766 s
, t
, p
, lod_in
, control
, lod
);
3769 get_filters(sp_sview
, sp_samp
, control
, &funcs
, NULL
, NULL
);
3770 funcs
->relative_level(sp_sview
, sp_samp
, lod
, mipmap
);
3774 sp_tgsi_get_texel(struct tgsi_sampler
*tgsi_sampler
,
3775 const unsigned sview_index
,
3776 const int i
[TGSI_QUAD_SIZE
],
3777 const int j
[TGSI_QUAD_SIZE
], const int k
[TGSI_QUAD_SIZE
],
3778 const int lod
[TGSI_QUAD_SIZE
], const int8_t offset
[3],
3779 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3781 const struct sp_tgsi_sampler
*sp_samp
=
3782 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3784 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3785 /* always have a view here but texture is NULL if no sampler view was set. */
3786 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3788 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3789 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3795 sp_get_texels(&sp_samp
->sp_sview
[sview_index
], i
, j
, k
, lod
, offset
, rgba
);
3799 struct sp_tgsi_sampler
*
3800 sp_create_tgsi_sampler(void)
3802 struct sp_tgsi_sampler
*samp
= CALLOC_STRUCT(sp_tgsi_sampler
);
3806 samp
->base
.get_dims
= sp_tgsi_get_dims
;
3807 samp
->base
.get_samples
= sp_tgsi_get_samples
;
3808 samp
->base
.get_texel
= sp_tgsi_get_texel
;
3809 samp
->base
.query_lod
= sp_tgsi_query_lod
;