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
])
663 compute_lambda_vert_explicite_gradients(UNUSED
const struct sp_sampler_view
*sview
,
664 UNUSED
const float derivs
[3][2][TGSI_QUAD_SIZE
],
671 compute_lambda_from_grad_func
672 softpipe_get_lambda_from_grad_func(const struct pipe_sampler_view
*view
,
673 enum pipe_shader_type shader
)
675 switch (view
->target
) {
677 case PIPE_TEXTURE_1D
:
678 case PIPE_TEXTURE_1D_ARRAY
:
679 return compute_lambda_1d_explicit_gradients
;
680 case PIPE_TEXTURE_2D
:
681 case PIPE_TEXTURE_2D_ARRAY
:
682 case PIPE_TEXTURE_RECT
:
683 return compute_lambda_2d_explicit_gradients
;
684 case PIPE_TEXTURE_CUBE
:
685 case PIPE_TEXTURE_CUBE_ARRAY
:
686 return compute_lambda_cube_explicit_gradients
;
687 case PIPE_TEXTURE_3D
:
688 return compute_lambda_3d_explicit_gradients
;
691 return compute_lambda_1d_explicit_gradients
;
697 * Get a texel from a texture, using the texture tile cache.
699 * \param addr the template tex address containing cube, z, face info.
700 * \param x the x coord of texel within 2D image
701 * \param y the y coord of texel within 2D image
702 * \param rgba the quad to put the texel/color into
704 * XXX maybe move this into sp_tex_tile_cache.c and merge with the
705 * sp_get_cached_tile_tex() function.
710 static inline const float *
711 get_texel_buffer_no_border(const struct sp_sampler_view
*sp_sview
,
712 union tex_tile_address addr
, int x
, unsigned elmsize
)
714 const struct softpipe_tex_cached_tile
*tile
;
715 addr
.bits
.x
= x
* elmsize
/ TEX_TILE_SIZE
;
716 assert(x
* elmsize
/ TEX_TILE_SIZE
== addr
.bits
.x
);
718 x
%= TEX_TILE_SIZE
/ elmsize
;
720 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
722 return &tile
->data
.color
[0][x
][0];
726 static inline const float *
727 get_texel_2d_no_border(const struct sp_sampler_view
*sp_sview
,
728 union tex_tile_address addr
, int x
, int y
)
730 const struct softpipe_tex_cached_tile
*tile
;
731 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
732 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
736 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
738 return &tile
->data
.color
[y
][x
][0];
742 static inline const float *
743 get_texel_2d(const struct sp_sampler_view
*sp_sview
,
744 const struct sp_sampler
*sp_samp
,
745 union tex_tile_address addr
, int x
, int y
)
747 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
748 const unsigned level
= addr
.bits
.level
;
750 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
751 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
752 return sp_samp
->base
.border_color
.f
;
755 return get_texel_2d_no_border( sp_sview
, addr
, x
, y
);
761 * Here's the complete logic (HOLY CRAP) for finding next face and doing the
762 * corresponding coord wrapping, implemented by get_next_face,
763 * get_next_xcoord, get_next_ycoord.
764 * Read like that (first line):
765 * If face is +x and s coord is below zero, then
766 * new face is +z, new s is max , new t is old t
767 * (max is always cube size - 1).
769 * +x s- -> +z: s = max, t = t
770 * +x s+ -> -z: s = 0, t = t
771 * +x t- -> +y: s = max, t = max-s
772 * +x t+ -> -y: s = max, t = s
774 * -x s- -> -z: s = max, t = t
775 * -x s+ -> +z: s = 0, t = t
776 * -x t- -> +y: s = 0, t = s
777 * -x t+ -> -y: s = 0, t = max-s
779 * +y s- -> -x: s = t, t = 0
780 * +y s+ -> +x: s = max-t, t = 0
781 * +y t- -> -z: s = max-s, t = 0
782 * +y t+ -> +z: s = s, t = 0
784 * -y s- -> -x: s = max-t, t = max
785 * -y s+ -> +x: s = t, t = max
786 * -y t- -> +z: s = s, t = max
787 * -y t+ -> -z: s = max-s, t = max
789 * +z s- -> -x: s = max, t = t
790 * +z s+ -> +x: s = 0, t = t
791 * +z t- -> +y: s = s, t = max
792 * +z t+ -> -y: s = s, t = 0
794 * -z s- -> +x: s = max, t = t
795 * -z s+ -> -x: s = 0, t = t
796 * -z t- -> +y: s = max-s, t = 0
797 * -z t+ -> -y: s = max-s, t = max
802 * seamless cubemap neighbour array.
803 * this array is used to find the adjacent face in each of 4 directions,
804 * left, right, up, down. (or -x, +x, -y, +y).
806 static const unsigned face_array
[PIPE_TEX_FACE_MAX
][4] = {
807 /* pos X first then neg X is Z different, Y the same */
808 /* PIPE_TEX_FACE_POS_X,*/
809 { PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
,
810 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
811 /* PIPE_TEX_FACE_NEG_X */
812 { PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
,
813 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
815 /* pos Y first then neg Y is X different, X the same */
816 /* PIPE_TEX_FACE_POS_Y */
817 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
818 PIPE_TEX_FACE_NEG_Z
, PIPE_TEX_FACE_POS_Z
},
820 /* PIPE_TEX_FACE_NEG_Y */
821 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
822 PIPE_TEX_FACE_POS_Z
, PIPE_TEX_FACE_NEG_Z
},
824 /* pos Z first then neg Y is X different, X the same */
825 /* PIPE_TEX_FACE_POS_Z */
826 { PIPE_TEX_FACE_NEG_X
, PIPE_TEX_FACE_POS_X
,
827 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
},
829 /* PIPE_TEX_FACE_NEG_Z */
830 { PIPE_TEX_FACE_POS_X
, PIPE_TEX_FACE_NEG_X
,
831 PIPE_TEX_FACE_POS_Y
, PIPE_TEX_FACE_NEG_Y
}
834 static inline unsigned
835 get_next_face(unsigned face
, int idx
)
837 return face_array
[face
][idx
];
841 * return a new xcoord based on old face, old coords, cube size
842 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
845 get_next_xcoord(unsigned face
, unsigned fall_off_index
, int max
, int xc
, int yc
)
847 if ((face
== 0 && fall_off_index
!= 1) ||
848 (face
== 1 && fall_off_index
== 0) ||
849 (face
== 4 && fall_off_index
== 0) ||
850 (face
== 5 && fall_off_index
== 0)) {
853 if ((face
== 1 && fall_off_index
!= 0) ||
854 (face
== 0 && fall_off_index
== 1) ||
855 (face
== 4 && fall_off_index
== 1) ||
856 (face
== 5 && fall_off_index
== 1)) {
859 if ((face
== 4 && fall_off_index
>= 2) ||
860 (face
== 2 && fall_off_index
== 3) ||
861 (face
== 3 && fall_off_index
== 2)) {
864 if ((face
== 5 && fall_off_index
>= 2) ||
865 (face
== 2 && fall_off_index
== 2) ||
866 (face
== 3 && fall_off_index
== 3)) {
869 if ((face
== 2 && fall_off_index
== 0) ||
870 (face
== 3 && fall_off_index
== 1)) {
873 /* (face == 2 && fall_off_index == 1) ||
874 (face == 3 && fall_off_index == 0)) */
879 * return a new ycoord based on old face, old coords, cube size
880 * and fall_off_index (0 for x-, 1 for x+, 2 for y-, 3 for y+)
883 get_next_ycoord(unsigned face
, unsigned fall_off_index
, int max
, int xc
, int yc
)
885 if ((fall_off_index
<= 1) && (face
<= 1 || face
>= 4)) {
889 (face
== 4 && fall_off_index
== 3) ||
890 (face
== 5 && fall_off_index
== 2)) {
894 (face
== 4 && fall_off_index
== 2) ||
895 (face
== 5 && fall_off_index
== 3)) {
898 if ((face
== 0 && fall_off_index
== 3) ||
899 (face
== 1 && fall_off_index
== 2)) {
902 /* (face == 0 && fall_off_index == 2) ||
903 (face == 1 && fall_off_index == 3) */
908 /* Gather a quad of adjacent texels within a tile:
911 get_texel_quad_2d_no_border_single_tile(const struct sp_sampler_view
*sp_sview
,
912 union tex_tile_address addr
,
913 unsigned x
, unsigned y
,
916 const struct softpipe_tex_cached_tile
*tile
;
918 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
919 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
923 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
925 out
[0] = &tile
->data
.color
[y
][x
][0];
926 out
[1] = &tile
->data
.color
[y
][x
+1][0];
927 out
[2] = &tile
->data
.color
[y
+1][x
][0];
928 out
[3] = &tile
->data
.color
[y
+1][x
+1][0];
932 /* Gather a quad of potentially non-adjacent texels:
935 get_texel_quad_2d_no_border(const struct sp_sampler_view
*sp_sview
,
936 union tex_tile_address addr
,
941 out
[0] = get_texel_2d_no_border( sp_sview
, addr
, x0
, y0
);
942 out
[1] = get_texel_2d_no_border( sp_sview
, addr
, x1
, y0
);
943 out
[2] = get_texel_2d_no_border( sp_sview
, addr
, x0
, y1
);
944 out
[3] = get_texel_2d_no_border( sp_sview
, addr
, x1
, y1
);
950 static inline const float *
951 get_texel_3d_no_border(const struct sp_sampler_view
*sp_sview
,
952 union tex_tile_address addr
, int x
, int y
, int z
)
954 const struct softpipe_tex_cached_tile
*tile
;
956 addr
.bits
.x
= x
/ TEX_TILE_SIZE
;
957 addr
.bits
.y
= y
/ TEX_TILE_SIZE
;
962 tile
= sp_get_cached_tile_tex(sp_sview
->cache
, addr
);
964 return &tile
->data
.color
[y
][x
][0];
968 static inline const float *
969 get_texel_3d(const struct sp_sampler_view
*sp_sview
,
970 const struct sp_sampler
*sp_samp
,
971 union tex_tile_address addr
, int x
, int y
, int z
)
973 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
974 const unsigned level
= addr
.bits
.level
;
976 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
977 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
) ||
978 z
< 0 || z
>= (int) u_minify(texture
->depth0
, level
)) {
979 return sp_samp
->base
.border_color
.f
;
982 return get_texel_3d_no_border( sp_sview
, addr
, x
, y
, z
);
987 /* Get texel pointer for 1D array texture */
988 static inline const float *
989 get_texel_1d_array(const struct sp_sampler_view
*sp_sview
,
990 const struct sp_sampler
*sp_samp
,
991 union tex_tile_address addr
, int x
, int y
)
993 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
994 const unsigned level
= addr
.bits
.level
;
996 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
)) {
997 return sp_samp
->base
.border_color
.f
;
1000 return get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
1005 /* Get texel pointer for 2D array texture */
1006 static inline const float *
1007 get_texel_2d_array(const struct sp_sampler_view
*sp_sview
,
1008 const struct sp_sampler
*sp_samp
,
1009 union tex_tile_address addr
, int x
, int y
, int layer
)
1011 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1012 const unsigned level
= addr
.bits
.level
;
1014 assert(layer
< (int) texture
->array_size
);
1017 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
1018 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
1019 return sp_samp
->base
.border_color
.f
;
1022 return get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
1027 static inline const float *
1028 get_texel_cube_seamless(const struct sp_sampler_view
*sp_sview
,
1029 union tex_tile_address addr
, int x
, int y
,
1030 float *corner
, int layer
, unsigned face
)
1032 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1033 const unsigned level
= addr
.bits
.level
;
1034 int new_x
, new_y
, max_x
;
1036 max_x
= (int) u_minify(texture
->width0
, level
);
1038 assert(texture
->width0
== texture
->height0
);
1042 /* change the face */
1045 * Cheat with corners. They are difficult and I believe because we don't get
1046 * per-pixel faces we can actually have multiple corner texels per pixel,
1047 * which screws things up majorly in any case (as the per spec behavior is
1048 * to average the 3 remaining texels, which we might not have).
1049 * Hence just make sure that the 2nd coord is clamped, will simply pick the
1050 * sample which would have fallen off the x coord, but not y coord.
1051 * So the filter weight of the samples will be wrong, but at least this
1052 * ensures that only valid texels near the corner are used.
1054 if (y
< 0 || y
>= max_x
) {
1055 y
= CLAMP(y
, 0, max_x
- 1);
1057 new_x
= get_next_xcoord(face
, 0, max_x
-1, x
, y
);
1058 new_y
= get_next_ycoord(face
, 0, max_x
-1, x
, y
);
1059 face
= get_next_face(face
, 0);
1060 } else if (x
>= max_x
) {
1061 if (y
< 0 || y
>= max_x
) {
1062 y
= CLAMP(y
, 0, max_x
- 1);
1064 new_x
= get_next_xcoord(face
, 1, max_x
-1, x
, y
);
1065 new_y
= get_next_ycoord(face
, 1, max_x
-1, x
, y
);
1066 face
= get_next_face(face
, 1);
1068 new_x
= get_next_xcoord(face
, 2, max_x
-1, x
, y
);
1069 new_y
= get_next_ycoord(face
, 2, max_x
-1, x
, y
);
1070 face
= get_next_face(face
, 2);
1071 } else if (y
>= max_x
) {
1072 new_x
= get_next_xcoord(face
, 3, max_x
-1, x
, y
);
1073 new_y
= get_next_ycoord(face
, 3, max_x
-1, x
, y
);
1074 face
= get_next_face(face
, 3);
1077 return get_texel_3d_no_border(sp_sview
, addr
, new_x
, new_y
, layer
+ face
);
1081 /* Get texel pointer for cube array texture */
1082 static inline const float *
1083 get_texel_cube_array(const struct sp_sampler_view
*sp_sview
,
1084 const struct sp_sampler
*sp_samp
,
1085 union tex_tile_address addr
, int x
, int y
, int layer
)
1087 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1088 const unsigned level
= addr
.bits
.level
;
1090 assert(layer
< (int) texture
->array_size
);
1093 if (x
< 0 || x
>= (int) u_minify(texture
->width0
, level
) ||
1094 y
< 0 || y
>= (int) u_minify(texture
->height0
, level
)) {
1095 return sp_samp
->base
.border_color
.f
;
1098 return get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
1102 * Given the logbase2 of a mipmap's base level size and a mipmap level,
1103 * return the size (in texels) of that mipmap level.
1104 * For example, if level[0].width = 256 then base_pot will be 8.
1105 * If level = 2, then we'll return 64 (the width at level=2).
1106 * Return 1 if level > base_pot.
1108 static inline unsigned
1109 pot_level_size(unsigned base_pot
, unsigned level
)
1111 return (base_pot
>= level
) ? (1 << (base_pot
- level
)) : 1;
1116 print_sample(const char *function
, const float *rgba
)
1118 debug_printf("%s %g %g %g %g\n",
1120 rgba
[0], rgba
[TGSI_NUM_CHANNELS
], rgba
[2*TGSI_NUM_CHANNELS
], rgba
[3*TGSI_NUM_CHANNELS
]);
1125 print_sample_4(const char *function
, float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
1127 debug_printf("%s %g %g %g %g, %g %g %g %g, %g %g %g %g, %g %g %g %g\n",
1129 rgba
[0][0], rgba
[1][0], rgba
[2][0], rgba
[3][0],
1130 rgba
[0][1], rgba
[1][1], rgba
[2][1], rgba
[3][1],
1131 rgba
[0][2], rgba
[1][2], rgba
[2][2], rgba
[3][2],
1132 rgba
[0][3], rgba
[1][3], rgba
[2][3], rgba
[3][3]);
1136 /* Some image-filter fastpaths:
1139 img_filter_2d_linear_repeat_POT(const struct sp_sampler_view
*sp_sview
,
1140 const struct sp_sampler
*sp_samp
,
1141 const struct img_filter_args
*args
,
1144 const unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1145 const unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1146 const int xmax
= (xpot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, xpot) - 1; */
1147 const int ymax
= (ypot
- 1) & (TEX_TILE_SIZE
- 1); /* MIN2(TEX_TILE_SIZE, ypot) - 1; */
1148 union tex_tile_address addr
;
1151 const float u
= (args
->s
* xpot
- 0.5F
) + args
->offset
[0];
1152 const float v
= (args
->t
* ypot
- 0.5F
) + args
->offset
[1];
1154 const int uflr
= util_ifloor(u
);
1155 const int vflr
= util_ifloor(v
);
1157 const float xw
= u
- (float)uflr
;
1158 const float yw
= v
- (float)vflr
;
1160 const int x0
= uflr
& (xpot
- 1);
1161 const int y0
= vflr
& (ypot
- 1);
1166 addr
.bits
.level
= args
->level
;
1167 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1169 /* Can we fetch all four at once:
1171 if (x0
< xmax
&& y0
< ymax
) {
1172 get_texel_quad_2d_no_border_single_tile(sp_sview
, addr
, x0
, y0
, tx
);
1175 const unsigned x1
= (x0
+ 1) & (xpot
- 1);
1176 const unsigned y1
= (y0
+ 1) & (ypot
- 1);
1177 get_texel_quad_2d_no_border(sp_sview
, addr
, x0
, y0
, x1
, y1
, tx
);
1180 /* interpolate R, G, B, A */
1181 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++) {
1182 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1184 tx
[2][c
], tx
[3][c
]);
1188 print_sample(__FUNCTION__
, rgba
);
1194 img_filter_2d_nearest_repeat_POT(const struct sp_sampler_view
*sp_sview
,
1195 const struct sp_sampler
*sp_samp
,
1196 const struct img_filter_args
*args
,
1199 const unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1200 const unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1202 union tex_tile_address addr
;
1205 const float u
= args
->s
* xpot
+ args
->offset
[0];
1206 const float v
= args
->t
* ypot
+ args
->offset
[1];
1208 const int uflr
= util_ifloor(u
);
1209 const int vflr
= util_ifloor(v
);
1211 const int x0
= uflr
& (xpot
- 1);
1212 const int y0
= vflr
& (ypot
- 1);
1215 addr
.bits
.level
= args
->level
;
1216 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1218 out
= get_texel_2d_no_border(sp_sview
, addr
, x0
, y0
);
1219 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1220 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1223 print_sample(__FUNCTION__
, rgba
);
1229 img_filter_2d_nearest_clamp_POT(const struct sp_sampler_view
*sp_sview
,
1230 const struct sp_sampler
*sp_samp
,
1231 const struct img_filter_args
*args
,
1234 const unsigned xpot
= pot_level_size(sp_sview
->xpot
, args
->level
);
1235 const unsigned ypot
= pot_level_size(sp_sview
->ypot
, args
->level
);
1236 union tex_tile_address addr
;
1239 const float u
= args
->s
* xpot
+ args
->offset
[0];
1240 const float v
= args
->t
* ypot
+ args
->offset
[1];
1246 addr
.bits
.level
= args
->level
;
1247 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1249 x0
= util_ifloor(u
);
1252 else if (x0
> (int) xpot
- 1)
1255 y0
= util_ifloor(v
);
1258 else if (y0
> (int) ypot
- 1)
1261 out
= get_texel_2d_no_border(sp_sview
, addr
, x0
, y0
);
1262 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1263 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1266 print_sample(__FUNCTION__
, rgba
);
1272 img_filter_1d_nearest(const struct sp_sampler_view
*sp_sview
,
1273 const struct sp_sampler
*sp_samp
,
1274 const struct img_filter_args
*args
,
1277 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1278 const int width
= u_minify(texture
->width0
, args
->level
);
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
,
1292 sp_sview
->base
.u
.tex
.first_layer
);
1293 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1294 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1297 print_sample(__FUNCTION__
, rgba
);
1303 img_filter_1d_array_nearest(const struct sp_sampler_view
*sp_sview
,
1304 const struct sp_sampler
*sp_samp
,
1305 const struct img_filter_args
*args
,
1308 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1309 const int width
= u_minify(texture
->width0
, args
->level
);
1310 const int layer
= coord_to_layer(args
->t
, sp_sview
->base
.u
.tex
.first_layer
,
1311 sp_sview
->base
.u
.tex
.last_layer
);
1313 union tex_tile_address addr
;
1320 addr
.bits
.level
= args
->level
;
1322 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1324 out
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x
, layer
);
1325 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1326 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1329 print_sample(__FUNCTION__
, rgba
);
1335 img_filter_2d_nearest(const struct sp_sampler_view
*sp_sview
,
1336 const struct sp_sampler
*sp_samp
,
1337 const struct img_filter_args
*args
,
1340 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1341 const int width
= u_minify(texture
->width0
, args
->level
);
1342 const int height
= u_minify(texture
->height0
, args
->level
);
1344 union tex_tile_address addr
;
1352 addr
.bits
.level
= args
->level
;
1353 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1355 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1356 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1358 out
= get_texel_2d(sp_sview
, sp_samp
, addr
, x
, y
);
1359 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1360 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1363 print_sample(__FUNCTION__
, rgba
);
1369 img_filter_2d_array_nearest(const struct sp_sampler_view
*sp_sview
,
1370 const struct sp_sampler
*sp_samp
,
1371 const struct img_filter_args
*args
,
1374 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1375 const int width
= u_minify(texture
->width0
, args
->level
);
1376 const int height
= u_minify(texture
->height0
, args
->level
);
1377 const int layer
= coord_to_layer(args
->p
, sp_sview
->base
.u
.tex
.first_layer
,
1378 sp_sview
->base
.u
.tex
.last_layer
);
1380 union tex_tile_address addr
;
1388 addr
.bits
.level
= args
->level
;
1390 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1391 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1393 out
= get_texel_2d_array(sp_sview
, sp_samp
, addr
, x
, y
, layer
);
1394 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1395 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1398 print_sample(__FUNCTION__
, rgba
);
1404 img_filter_cube_nearest(const struct sp_sampler_view
*sp_sview
,
1405 const struct sp_sampler
*sp_samp
,
1406 const struct img_filter_args
*args
,
1409 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1410 const int width
= u_minify(texture
->width0
, args
->level
);
1411 const int height
= u_minify(texture
->height0
, args
->level
);
1412 const int layerface
= args
->face_id
+ sp_sview
->base
.u
.tex
.first_layer
;
1414 union tex_tile_address addr
;
1422 addr
.bits
.level
= args
->level
;
1425 * If NEAREST filtering is done within a miplevel, always apply wrap
1426 * mode CLAMP_TO_EDGE.
1428 if (sp_samp
->base
.seamless_cube_map
) {
1429 wrap_nearest_clamp_to_edge(args
->s
, width
, args
->offset
[0], &x
);
1430 wrap_nearest_clamp_to_edge(args
->t
, height
, args
->offset
[1], &y
);
1432 /* Would probably make sense to ignore mode and just do edge clamp */
1433 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1434 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1437 out
= get_texel_cube_array(sp_sview
, sp_samp
, addr
, x
, y
, layerface
);
1438 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1439 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1442 print_sample(__FUNCTION__
, rgba
);
1447 img_filter_cube_array_nearest(const struct sp_sampler_view
*sp_sview
,
1448 const struct sp_sampler
*sp_samp
,
1449 const struct img_filter_args
*args
,
1452 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1453 const int width
= u_minify(texture
->width0
, args
->level
);
1454 const int height
= u_minify(texture
->height0
, args
->level
);
1455 const int layerface
=
1456 coord_to_layer(6 * args
->p
+ sp_sview
->base
.u
.tex
.first_layer
,
1457 sp_sview
->base
.u
.tex
.first_layer
,
1458 sp_sview
->base
.u
.tex
.last_layer
- 5) + args
->face_id
;
1460 union tex_tile_address addr
;
1468 addr
.bits
.level
= args
->level
;
1470 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1471 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1473 out
= get_texel_cube_array(sp_sview
, sp_samp
, addr
, x
, y
, layerface
);
1474 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1475 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1478 print_sample(__FUNCTION__
, rgba
);
1483 img_filter_3d_nearest(const struct sp_sampler_view
*sp_sview
,
1484 const struct sp_sampler
*sp_samp
,
1485 const struct img_filter_args
*args
,
1488 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1489 const int width
= u_minify(texture
->width0
, args
->level
);
1490 const int height
= u_minify(texture
->height0
, args
->level
);
1491 const int depth
= u_minify(texture
->depth0
, args
->level
);
1493 union tex_tile_address addr
;
1501 sp_samp
->nearest_texcoord_s(args
->s
, width
, args
->offset
[0], &x
);
1502 sp_samp
->nearest_texcoord_t(args
->t
, height
, args
->offset
[1], &y
);
1503 sp_samp
->nearest_texcoord_p(args
->p
, depth
, args
->offset
[2], &z
);
1506 addr
.bits
.level
= args
->level
;
1508 out
= get_texel_3d(sp_sview
, sp_samp
, addr
, x
, y
, z
);
1509 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1510 rgba
[TGSI_NUM_CHANNELS
*c
] = out
[c
];
1515 img_filter_1d_linear(const struct sp_sampler_view
*sp_sview
,
1516 const struct sp_sampler
*sp_samp
,
1517 const struct img_filter_args
*args
,
1520 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1521 const int width
= u_minify(texture
->width0
, args
->level
);
1523 float xw
; /* weights */
1524 union tex_tile_address addr
;
1525 const float *tx0
, *tx1
;
1531 addr
.bits
.level
= args
->level
;
1533 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1535 tx0
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x0
,
1536 sp_sview
->base
.u
.tex
.first_layer
);
1537 tx1
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x1
,
1538 sp_sview
->base
.u
.tex
.first_layer
);
1540 /* interpolate R, G, B, A */
1541 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1542 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1547 img_filter_1d_array_linear(const struct sp_sampler_view
*sp_sview
,
1548 const struct sp_sampler
*sp_samp
,
1549 const struct img_filter_args
*args
,
1552 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1553 const int width
= u_minify(texture
->width0
, args
->level
);
1554 const int layer
= coord_to_layer(args
->t
, sp_sview
->base
.u
.tex
.first_layer
,
1555 sp_sview
->base
.u
.tex
.last_layer
);
1557 float xw
; /* weights */
1558 union tex_tile_address addr
;
1559 const float *tx0
, *tx1
;
1565 addr
.bits
.level
= args
->level
;
1567 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1569 tx0
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x0
, layer
);
1570 tx1
= get_texel_1d_array(sp_sview
, sp_samp
, addr
, x1
, layer
);
1572 /* interpolate R, G, B, A */
1573 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1574 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp(xw
, tx0
[c
], tx1
[c
]);
1578 * Retrieve the gathered value, need to convert to the
1579 * TGSI expected interface, and take component select
1580 * and swizzling into account.
1583 get_gather_value(const struct sp_sampler_view
*sp_sview
,
1584 int chan_in
, int comp_sel
,
1591 * softpipe samples in a different order
1592 * to TGSI expects, so we need to swizzle,
1593 * the samples into the correct slots.
1613 /* pick which component to use for the swizzle */
1616 swizzle
= sp_sview
->base
.swizzle_r
;
1619 swizzle
= sp_sview
->base
.swizzle_g
;
1622 swizzle
= sp_sview
->base
.swizzle_b
;
1625 swizzle
= sp_sview
->base
.swizzle_a
;
1632 /* get correct result using the channel and swizzle */
1634 case PIPE_SWIZZLE_0
:
1636 case PIPE_SWIZZLE_1
:
1639 return tx
[chan
][swizzle
];
1645 img_filter_2d_linear(const struct sp_sampler_view
*sp_sview
,
1646 const struct sp_sampler
*sp_samp
,
1647 const struct img_filter_args
*args
,
1650 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1651 const int width
= u_minify(texture
->width0
, args
->level
);
1652 const int height
= u_minify(texture
->height0
, args
->level
);
1654 float xw
, yw
; /* weights */
1655 union tex_tile_address addr
;
1663 addr
.bits
.level
= args
->level
;
1664 addr
.bits
.z
= sp_sview
->base
.u
.tex
.first_layer
;
1666 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1667 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1669 tx
[0] = get_texel_2d(sp_sview
, sp_samp
, addr
, x0
, y0
);
1670 tx
[1] = get_texel_2d(sp_sview
, sp_samp
, addr
, x1
, y0
);
1671 tx
[2] = get_texel_2d(sp_sview
, sp_samp
, addr
, x0
, y1
);
1672 tx
[3] = get_texel_2d(sp_sview
, sp_samp
, addr
, x1
, y1
);
1674 if (args
->gather_only
) {
1675 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1676 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1680 /* interpolate R, G, B, A */
1681 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1682 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1684 tx
[2][c
], tx
[3][c
]);
1690 img_filter_2d_array_linear(const struct sp_sampler_view
*sp_sview
,
1691 const struct sp_sampler
*sp_samp
,
1692 const struct img_filter_args
*args
,
1695 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1696 const int width
= u_minify(texture
->width0
, args
->level
);
1697 const int height
= u_minify(texture
->height0
, args
->level
);
1698 const int layer
= coord_to_layer(args
->p
, sp_sview
->base
.u
.tex
.first_layer
,
1699 sp_sview
->base
.u
.tex
.last_layer
);
1701 float xw
, yw
; /* weights */
1702 union tex_tile_address addr
;
1710 addr
.bits
.level
= args
->level
;
1712 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1713 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1715 tx
[0] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
);
1716 tx
[1] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
);
1717 tx
[2] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
);
1718 tx
[3] = get_texel_2d_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
);
1720 if (args
->gather_only
) {
1721 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1722 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1726 /* interpolate R, G, B, A */
1727 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1728 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1730 tx
[2][c
], tx
[3][c
]);
1736 img_filter_cube_linear(const struct sp_sampler_view
*sp_sview
,
1737 const struct sp_sampler
*sp_samp
,
1738 const struct img_filter_args
*args
,
1741 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1742 const int width
= u_minify(texture
->width0
, args
->level
);
1743 const int height
= u_minify(texture
->height0
, args
->level
);
1744 const int layer
= sp_sview
->base
.u
.tex
.first_layer
;
1746 float xw
, yw
; /* weights */
1747 union tex_tile_address addr
;
1749 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
],
1750 corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1757 addr
.bits
.level
= args
->level
;
1760 * For seamless if LINEAR filtering is done within a miplevel,
1761 * always apply wrap mode CLAMP_TO_BORDER.
1763 if (sp_samp
->base
.seamless_cube_map
) {
1764 /* Note this is a bit overkill, actual clamping is not required */
1765 wrap_linear_clamp_to_border(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1766 wrap_linear_clamp_to_border(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1768 /* Would probably make sense to ignore mode and just do edge clamp */
1769 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1770 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1773 if (sp_samp
->base
.seamless_cube_map
) {
1774 tx
[0] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y0
, corner0
, layer
, args
->face_id
);
1775 tx
[1] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y0
, corner1
, layer
, args
->face_id
);
1776 tx
[2] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y1
, corner2
, layer
, args
->face_id
);
1777 tx
[3] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y1
, corner3
, layer
, args
->face_id
);
1779 tx
[0] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
+ args
->face_id
);
1780 tx
[1] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
+ args
->face_id
);
1781 tx
[2] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
+ args
->face_id
);
1782 tx
[3] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
+ args
->face_id
);
1785 if (args
->gather_only
) {
1786 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1787 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1791 /* interpolate R, G, B, A */
1792 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1793 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1795 tx
[2][c
], tx
[3][c
]);
1801 img_filter_cube_array_linear(const struct sp_sampler_view
*sp_sview
,
1802 const struct sp_sampler
*sp_samp
,
1803 const struct img_filter_args
*args
,
1806 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1807 const int width
= u_minify(texture
->width0
, args
->level
);
1808 const int height
= u_minify(texture
->height0
, args
->level
);
1810 coord_to_layer(6 * args
->p
+ sp_sview
->base
.u
.tex
.first_layer
,
1811 sp_sview
->base
.u
.tex
.first_layer
,
1812 sp_sview
->base
.u
.tex
.last_layer
- 5);
1814 float xw
, yw
; /* weights */
1815 union tex_tile_address addr
;
1817 float corner0
[TGSI_QUAD_SIZE
], corner1
[TGSI_QUAD_SIZE
],
1818 corner2
[TGSI_QUAD_SIZE
], corner3
[TGSI_QUAD_SIZE
];
1825 addr
.bits
.level
= args
->level
;
1828 * For seamless if LINEAR filtering is done within a miplevel,
1829 * always apply wrap mode CLAMP_TO_BORDER.
1831 if (sp_samp
->base
.seamless_cube_map
) {
1832 /* Note this is a bit overkill, actual clamping is not required */
1833 wrap_linear_clamp_to_border(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1834 wrap_linear_clamp_to_border(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1836 /* Would probably make sense to ignore mode and just do edge clamp */
1837 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1838 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1841 if (sp_samp
->base
.seamless_cube_map
) {
1842 tx
[0] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y0
, corner0
, layer
, args
->face_id
);
1843 tx
[1] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y0
, corner1
, layer
, args
->face_id
);
1844 tx
[2] = get_texel_cube_seamless(sp_sview
, addr
, x0
, y1
, corner2
, layer
, args
->face_id
);
1845 tx
[3] = get_texel_cube_seamless(sp_sview
, addr
, x1
, y1
, corner3
, layer
, args
->face_id
);
1847 tx
[0] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y0
, layer
+ args
->face_id
);
1848 tx
[1] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y0
, layer
+ args
->face_id
);
1849 tx
[2] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x0
, y1
, layer
+ args
->face_id
);
1850 tx
[3] = get_texel_cube_array(sp_sview
, sp_samp
, addr
, x1
, y1
, layer
+ args
->face_id
);
1853 if (args
->gather_only
) {
1854 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1855 rgba
[TGSI_NUM_CHANNELS
*c
] = get_gather_value(sp_sview
, c
,
1859 /* interpolate R, G, B, A */
1860 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1861 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_2d(xw
, yw
,
1863 tx
[2][c
], tx
[3][c
]);
1868 img_filter_3d_linear(const struct sp_sampler_view
*sp_sview
,
1869 const struct sp_sampler
*sp_samp
,
1870 const struct img_filter_args
*args
,
1873 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
1874 const int width
= u_minify(texture
->width0
, args
->level
);
1875 const int height
= u_minify(texture
->height0
, args
->level
);
1876 const int depth
= u_minify(texture
->depth0
, args
->level
);
1877 int x0
, x1
, y0
, y1
, z0
, z1
;
1878 float xw
, yw
, zw
; /* interpolation weights */
1879 union tex_tile_address addr
;
1880 const float *tx00
, *tx01
, *tx02
, *tx03
, *tx10
, *tx11
, *tx12
, *tx13
;
1884 addr
.bits
.level
= args
->level
;
1890 sp_samp
->linear_texcoord_s(args
->s
, width
, args
->offset
[0], &x0
, &x1
, &xw
);
1891 sp_samp
->linear_texcoord_t(args
->t
, height
, args
->offset
[1], &y0
, &y1
, &yw
);
1892 sp_samp
->linear_texcoord_p(args
->p
, depth
, args
->offset
[2], &z0
, &z1
, &zw
);
1894 tx00
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y0
, z0
);
1895 tx01
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y0
, z0
);
1896 tx02
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y1
, z0
);
1897 tx03
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y1
, z0
);
1899 tx10
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y0
, z1
);
1900 tx11
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y0
, z1
);
1901 tx12
= get_texel_3d(sp_sview
, sp_samp
, addr
, x0
, y1
, z1
);
1902 tx13
= get_texel_3d(sp_sview
, sp_samp
, addr
, x1
, y1
, z1
);
1904 /* interpolate R, G, B, A */
1905 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
1906 rgba
[TGSI_NUM_CHANNELS
*c
] = lerp_3d(xw
, yw
, zw
,
1914 /* Calculate level of detail for every fragment,
1915 * with lambda already computed.
1916 * Note that lambda has already been biased by global LOD bias.
1917 * \param biased_lambda per-quad lambda.
1918 * \param lod_in per-fragment lod_bias or explicit_lod.
1919 * \param lod returns the per-fragment lod.
1922 compute_lod(const struct pipe_sampler_state
*sampler
,
1923 enum tgsi_sampler_control control
,
1924 const float biased_lambda
,
1925 const float lod_in
[TGSI_QUAD_SIZE
],
1926 float lod
[TGSI_QUAD_SIZE
])
1928 const float min_lod
= sampler
->min_lod
;
1929 const float max_lod
= sampler
->max_lod
;
1933 case TGSI_SAMPLER_LOD_NONE
:
1934 case TGSI_SAMPLER_LOD_ZERO
:
1936 case TGSI_SAMPLER_DERIVS_EXPLICIT
:
1937 lod
[0] = lod
[1] = lod
[2] = lod
[3] = CLAMP(biased_lambda
, min_lod
, max_lod
);
1939 case TGSI_SAMPLER_LOD_BIAS
:
1940 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1941 lod
[i
] = biased_lambda
+ lod_in
[i
];
1942 lod
[i
] = CLAMP(lod
[i
], min_lod
, max_lod
);
1945 case TGSI_SAMPLER_LOD_EXPLICIT
:
1946 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1947 lod
[i
] = CLAMP(lod_in
[i
], min_lod
, max_lod
);
1952 lod
[0] = lod
[1] = lod
[2] = lod
[3] = 0.0f
;
1957 /* Calculate level of detail for every fragment. The computed value is not
1958 * clamped to lod_min and lod_max.
1959 * \param lod_in per-fragment lod_bias or explicit_lod.
1960 * \param lod results per-fragment lod.
1963 compute_lambda_lod_unclamped(const struct sp_sampler_view
*sp_sview
,
1964 const struct sp_sampler
*sp_samp
,
1965 const float s
[TGSI_QUAD_SIZE
],
1966 const float t
[TGSI_QUAD_SIZE
],
1967 const float p
[TGSI_QUAD_SIZE
],
1968 const float lod_in
[TGSI_QUAD_SIZE
],
1969 enum tgsi_sampler_control control
,
1970 float lod
[TGSI_QUAD_SIZE
])
1972 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
1973 const float lod_bias
= sampler
->lod_bias
;
1978 case TGSI_SAMPLER_LOD_NONE
:
1980 case TGSI_SAMPLER_DERIVS_EXPLICIT
:
1981 lambda
= sp_sview
->compute_lambda(sp_sview
, s
, t
, p
) + lod_bias
;
1982 lod
[0] = lod
[1] = lod
[2] = lod
[3] = lambda
;
1984 case TGSI_SAMPLER_LOD_BIAS
:
1985 lambda
= sp_sview
->compute_lambda(sp_sview
, s
, t
, p
) + lod_bias
;
1986 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1987 lod
[i
] = lambda
+ lod_in
[i
];
1990 case TGSI_SAMPLER_LOD_EXPLICIT
:
1991 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
1992 lod
[i
] = lod_in
[i
] + lod_bias
;
1995 case TGSI_SAMPLER_LOD_ZERO
:
1996 case TGSI_SAMPLER_GATHER
:
1997 lod
[0] = lod
[1] = lod
[2] = lod
[3] = lod_bias
;
2001 lod
[0] = lod
[1] = lod
[2] = lod
[3] = 0.0f
;
2005 /* Calculate level of detail for every fragment.
2006 * \param lod_in per-fragment lod_bias or explicit_lod.
2007 * \param lod results per-fragment lod.
2010 compute_lambda_lod(const struct sp_sampler_view
*sp_sview
,
2011 const struct sp_sampler
*sp_samp
,
2012 const float s
[TGSI_QUAD_SIZE
],
2013 const float t
[TGSI_QUAD_SIZE
],
2014 const float p
[TGSI_QUAD_SIZE
],
2015 const float lod_in
[TGSI_QUAD_SIZE
],
2016 enum tgsi_sampler_control control
,
2017 float lod
[TGSI_QUAD_SIZE
])
2019 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
2020 const float min_lod
= sampler
->min_lod
;
2021 const float max_lod
= sampler
->max_lod
;
2024 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
2025 s
, t
, p
, lod_in
, control
, lod
);
2026 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2027 lod
[i
] = CLAMP(lod
[i
], min_lod
, max_lod
);
2031 static inline unsigned
2032 get_gather_component(const float lod_in
[TGSI_QUAD_SIZE
])
2034 /* gather component is stored in lod_in slot as unsigned */
2035 return (*(unsigned int *)lod_in
) & 0x3;
2039 * Clamps given lod to both lod limits and mip level limits. Clamping to the
2040 * latter limits is done so that lod is relative to the first (base) level.
2043 clamp_lod(const struct sp_sampler_view
*sp_sview
,
2044 const struct sp_sampler
*sp_samp
,
2045 const float lod
[TGSI_QUAD_SIZE
],
2046 float clamped
[TGSI_QUAD_SIZE
])
2048 const float min_lod
= sp_samp
->base
.min_lod
;
2049 const float max_lod
= sp_samp
->base
.max_lod
;
2050 const float min_level
= sp_sview
->base
.u
.tex
.first_level
;
2051 const float max_level
= sp_sview
->base
.u
.tex
.last_level
;
2054 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
2057 cl
= CLAMP(cl
, min_lod
, max_lod
);
2058 cl
= CLAMP(cl
, 0, max_level
- min_level
);
2064 * Get mip level relative to base level for linear mip filter
2067 mip_rel_level_linear(const struct sp_sampler_view
*sp_sview
,
2068 const struct sp_sampler
*sp_samp
,
2069 const float lod
[TGSI_QUAD_SIZE
],
2070 float level
[TGSI_QUAD_SIZE
])
2072 clamp_lod(sp_sview
, sp_samp
, lod
, level
);
2076 mip_filter_linear(const struct sp_sampler_view
*sp_sview
,
2077 const struct sp_sampler
*sp_samp
,
2078 img_filter_func min_filter
,
2079 img_filter_func mag_filter
,
2080 const float s
[TGSI_QUAD_SIZE
],
2081 const float t
[TGSI_QUAD_SIZE
],
2082 const float p
[TGSI_QUAD_SIZE
],
2083 const float c0
[TGSI_QUAD_SIZE
],
2084 const float lod_in
[TGSI_QUAD_SIZE
],
2085 const struct filter_args
*filt_args
,
2086 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2088 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2090 float lod
[TGSI_QUAD_SIZE
];
2091 struct img_filter_args args
;
2093 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2095 args
.offset
= filt_args
->offset
;
2096 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2097 args
.gather_comp
= get_gather_component(lod_in
);
2099 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2100 const int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
2105 args
.face_id
= filt_args
->faces
[j
];
2107 if (lod
[j
] <= 0.0 && !args
.gather_only
) {
2108 args
.level
= psview
->u
.tex
.first_level
;
2109 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2111 else if (level0
>= (int) psview
->u
.tex
.last_level
) {
2112 args
.level
= psview
->u
.tex
.last_level
;
2113 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2116 float levelBlend
= frac(lod
[j
]);
2117 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2120 args
.level
= level0
;
2121 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
2122 args
.level
= level0
+1;
2123 min_filter(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
2125 for (c
= 0; c
< 4; c
++) {
2126 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2132 print_sample_4(__FUNCTION__
, rgba
);
2138 * Get mip level relative to base level for nearest mip filter
2141 mip_rel_level_nearest(const struct sp_sampler_view
*sp_sview
,
2142 const struct sp_sampler
*sp_samp
,
2143 const float lod
[TGSI_QUAD_SIZE
],
2144 float level
[TGSI_QUAD_SIZE
])
2148 clamp_lod(sp_sview
, sp_samp
, lod
, level
);
2149 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++)
2150 /* TODO: It should rather be:
2151 * level[j] = ceil(level[j] + 0.5F) - 1.0F;
2153 level
[j
] = (int)(level
[j
] + 0.5F
);
2157 * Compute nearest mipmap level from texcoords.
2158 * Then sample the texture level for four elements of a quad.
2159 * \param c0 the LOD bias factors, or absolute LODs (depending on control)
2162 mip_filter_nearest(const struct sp_sampler_view
*sp_sview
,
2163 const struct sp_sampler
*sp_samp
,
2164 img_filter_func min_filter
,
2165 img_filter_func mag_filter
,
2166 const float s
[TGSI_QUAD_SIZE
],
2167 const float t
[TGSI_QUAD_SIZE
],
2168 const float p
[TGSI_QUAD_SIZE
],
2169 const float c0
[TGSI_QUAD_SIZE
],
2170 const float lod_in
[TGSI_QUAD_SIZE
],
2171 const struct filter_args
*filt_args
,
2172 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2174 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2175 float lod
[TGSI_QUAD_SIZE
];
2177 struct img_filter_args args
;
2179 args
.offset
= filt_args
->offset
;
2180 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2181 args
.gather_comp
= get_gather_component(lod_in
);
2183 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2185 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2189 args
.face_id
= filt_args
->faces
[j
];
2191 if (lod
[j
] <= 0.0 && !args
.gather_only
) {
2192 args
.level
= psview
->u
.tex
.first_level
;
2193 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2195 const int level
= psview
->u
.tex
.first_level
+ (int)(lod
[j
] + 0.5F
);
2196 args
.level
= MIN2(level
, (int)psview
->u
.tex
.last_level
);
2197 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2202 print_sample_4(__FUNCTION__
, rgba
);
2208 * Get mip level relative to base level for none mip filter
2211 mip_rel_level_none(const struct sp_sampler_view
*sp_sview
,
2212 const struct sp_sampler
*sp_samp
,
2213 const float lod
[TGSI_QUAD_SIZE
],
2214 float level
[TGSI_QUAD_SIZE
])
2218 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2224 mip_filter_none(const struct sp_sampler_view
*sp_sview
,
2225 const struct sp_sampler
*sp_samp
,
2226 img_filter_func min_filter
,
2227 img_filter_func mag_filter
,
2228 const float s
[TGSI_QUAD_SIZE
],
2229 const float t
[TGSI_QUAD_SIZE
],
2230 const float p
[TGSI_QUAD_SIZE
],
2231 const float c0
[TGSI_QUAD_SIZE
],
2232 const float lod_in
[TGSI_QUAD_SIZE
],
2233 const struct filter_args
*filt_args
,
2234 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2236 float lod
[TGSI_QUAD_SIZE
];
2238 struct img_filter_args args
;
2240 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2241 args
.offset
= filt_args
->offset
;
2242 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2244 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2246 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2250 args
.face_id
= filt_args
->faces
[j
];
2251 if (lod
[j
] <= 0.0f
&& !args
.gather_only
) {
2252 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2255 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2262 * Get mip level relative to base level for none mip filter
2265 mip_rel_level_none_no_filter_select(const struct sp_sampler_view
*sp_sview
,
2266 const struct sp_sampler
*sp_samp
,
2267 const float lod
[TGSI_QUAD_SIZE
],
2268 float level
[TGSI_QUAD_SIZE
])
2270 mip_rel_level_none(sp_sview
, sp_samp
, lod
, level
);
2274 mip_filter_none_no_filter_select(const struct sp_sampler_view
*sp_sview
,
2275 const struct sp_sampler
*sp_samp
,
2276 img_filter_func min_filter
,
2277 img_filter_func mag_filter
,
2278 const float s
[TGSI_QUAD_SIZE
],
2279 const float t
[TGSI_QUAD_SIZE
],
2280 const float p
[TGSI_QUAD_SIZE
],
2281 const float c0
[TGSI_QUAD_SIZE
],
2282 const float lod_in
[TGSI_QUAD_SIZE
],
2283 const struct filter_args
*filt_args
,
2284 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2287 struct img_filter_args args
;
2288 args
.level
= sp_sview
->base
.u
.tex
.first_level
;
2289 args
.offset
= filt_args
->offset
;
2290 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2291 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2295 args
.face_id
= filt_args
->faces
[j
];
2296 mag_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2301 /* For anisotropic filtering */
2302 #define WEIGHT_LUT_SIZE 1024
2304 static const float *weightLut
= NULL
;
2307 * Creates the look-up table used to speed-up EWA sampling
2310 create_filter_table(void)
2314 float *lut
= (float *) MALLOC(WEIGHT_LUT_SIZE
* sizeof(float));
2316 for (i
= 0; i
< WEIGHT_LUT_SIZE
; ++i
) {
2317 const float alpha
= 2;
2318 const float r2
= (float) i
/ (float) (WEIGHT_LUT_SIZE
- 1);
2319 const float weight
= (float) exp(-alpha
* r2
);
2328 * Elliptical weighted average (EWA) filter for producing high quality
2329 * anisotropic filtered results.
2330 * Based on the Higher Quality Elliptical Weighted Average Filter
2331 * published by Paul S. Heckbert in his Master's Thesis
2332 * "Fundamentals of Texture Mapping and Image Warping" (1989)
2335 img_filter_2d_ewa(const struct sp_sampler_view
*sp_sview
,
2336 const struct sp_sampler
*sp_samp
,
2337 img_filter_func min_filter
,
2338 img_filter_func mag_filter
,
2339 const float s
[TGSI_QUAD_SIZE
],
2340 const float t
[TGSI_QUAD_SIZE
],
2341 const float p
[TGSI_QUAD_SIZE
],
2342 const uint faces
[TGSI_QUAD_SIZE
],
2343 const int8_t *offset
,
2345 const float dudx
, const float dvdx
,
2346 const float dudy
, const float dvdy
,
2347 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2349 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2351 // ??? Won't the image filters blow up if level is negative?
2352 const unsigned level0
= level
> 0 ? level
: 0;
2353 const float scaling
= 1.0f
/ (1 << level0
);
2354 const int width
= u_minify(texture
->width0
, level0
);
2355 const int height
= u_minify(texture
->height0
, level0
);
2356 struct img_filter_args args
;
2357 const float ux
= dudx
* scaling
;
2358 const float vx
= dvdx
* scaling
;
2359 const float uy
= dudy
* scaling
;
2360 const float vy
= dvdy
* scaling
;
2362 /* compute ellipse coefficients to bound the region:
2363 * A*x*x + B*x*y + C*y*y = F.
2365 float A
= vx
*vx
+vy
*vy
+1;
2366 float B
= -2*(ux
*vx
+uy
*vy
);
2367 float C
= ux
*ux
+uy
*uy
+1;
2368 float F
= A
*C
-B
*B
/4.0f
;
2370 /* check if it is an ellipse */
2371 /* assert(F > 0.0); */
2373 /* Compute the ellipse's (u,v) bounding box in texture space */
2374 const float d
= -B
*B
+4.0f
*C
*A
;
2375 const float box_u
= 2.0f
/ d
* sqrtf(d
*C
*F
); /* box_u -> half of bbox with */
2376 const float box_v
= 2.0f
/ d
* sqrtf(A
*d
*F
); /* box_v -> half of bbox height */
2378 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2379 float s_buffer
[TGSI_QUAD_SIZE
];
2380 float t_buffer
[TGSI_QUAD_SIZE
];
2381 float weight_buffer
[TGSI_QUAD_SIZE
];
2384 /* For each quad, the du and dx values are the same and so the ellipse is
2385 * also the same. Note that texel/image access can only be performed using
2386 * a quad, i.e. it is not possible to get the pixel value for a single
2387 * tex coord. In order to have a better performance, the access is buffered
2388 * using the s_buffer/t_buffer and weight_buffer. Only when the buffer is
2389 * full, then the pixel values are read from the image.
2391 const float ddq
= 2 * A
;
2393 /* Scale ellipse formula to directly index the Filter Lookup Table.
2394 * i.e. scale so that F = WEIGHT_LUT_SIZE-1
2396 const double formScale
= (double) (WEIGHT_LUT_SIZE
- 1) / F
;
2400 /* F *= formScale; */ /* no need to scale F as we don't use it below here */
2403 args
.offset
= offset
;
2405 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2406 /* Heckbert MS thesis, p. 59; scan over the bounding box of the ellipse
2407 * and incrementally update the value of Ax^2+Bxy*Cy^2; when this
2408 * value, q, is less than F, we're inside the ellipse
2410 const float tex_u
= -0.5F
+ s
[j
] * texture
->width0
* scaling
;
2411 const float tex_v
= -0.5F
+ t
[j
] * texture
->height0
* scaling
;
2413 const int u0
= (int) floorf(tex_u
- box_u
);
2414 const int u1
= (int) ceilf(tex_u
+ box_u
);
2415 const int v0
= (int) floorf(tex_v
- box_v
);
2416 const int v1
= (int) ceilf(tex_v
+ box_v
);
2417 const float U
= u0
- tex_u
;
2419 float num
[4] = {0.0F
, 0.0F
, 0.0F
, 0.0F
};
2420 unsigned buffer_next
= 0;
2423 args
.face_id
= faces
[j
];
2425 for (v
= v0
; v
<= v1
; ++v
) {
2426 const float V
= v
- tex_v
;
2427 float dq
= A
* (2 * U
+ 1) + B
* V
;
2428 float q
= (C
* V
+ B
* U
) * V
+ A
* U
* U
;
2431 for (u
= u0
; u
<= u1
; ++u
) {
2432 /* Note that the ellipse has been pre-scaled so F =
2433 * WEIGHT_LUT_SIZE - 1
2435 if (q
< WEIGHT_LUT_SIZE
) {
2436 /* as a LUT is used, q must never be negative;
2437 * should not happen, though
2439 const int qClamped
= q
>= 0.0F
? q
: 0;
2440 const float weight
= weightLut
[qClamped
];
2442 weight_buffer
[buffer_next
] = weight
;
2443 s_buffer
[buffer_next
] = u
/ ((float) width
);
2444 t_buffer
[buffer_next
] = v
/ ((float) height
);
2447 if (buffer_next
== TGSI_QUAD_SIZE
) {
2448 /* 4 texel coords are in the buffer -> read it now */
2450 /* it is assumed that samp->min_img_filter is set to
2451 * img_filter_2d_nearest or one of the
2452 * accelerated img_filter_2d_nearest_XXX functions.
2454 for (jj
= 0; jj
< buffer_next
; jj
++) {
2455 args
.s
= s_buffer
[jj
];
2456 args
.t
= t_buffer
[jj
];
2458 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2459 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2460 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2461 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2462 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2475 /* if the tex coord buffer contains unread values, we will read
2478 if (buffer_next
> 0) {
2480 /* it is assumed that samp->min_img_filter is set to
2481 * img_filter_2d_nearest or one of the
2482 * accelerated img_filter_2d_nearest_XXX functions.
2484 for (jj
= 0; jj
< buffer_next
; jj
++) {
2485 args
.s
= s_buffer
[jj
];
2486 args
.t
= t_buffer
[jj
];
2488 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][jj
]);
2489 num
[0] += weight_buffer
[jj
] * rgba_temp
[0][jj
];
2490 num
[1] += weight_buffer
[jj
] * rgba_temp
[1][jj
];
2491 num
[2] += weight_buffer
[jj
] * rgba_temp
[2][jj
];
2492 num
[3] += weight_buffer
[jj
] * rgba_temp
[3][jj
];
2497 /* Reaching this place would mean that no pixels intersected
2498 * the ellipse. This should never happen because the filter
2499 * we use always intersects at least one pixel.
2506 /* not enough pixels in resampling, resort to direct interpolation */
2510 min_filter(sp_sview
, sp_samp
, &args
, &rgba_temp
[0][j
]);
2512 num
[0] = rgba_temp
[0][j
];
2513 num
[1] = rgba_temp
[1][j
];
2514 num
[2] = rgba_temp
[2][j
];
2515 num
[3] = rgba_temp
[3][j
];
2518 rgba
[0][j
] = num
[0] / den
;
2519 rgba
[1][j
] = num
[1] / den
;
2520 rgba
[2][j
] = num
[2] / den
;
2521 rgba
[3][j
] = num
[3] / den
;
2527 * Get mip level relative to base level for linear mip filter
2530 mip_rel_level_linear_aniso(const struct sp_sampler_view
*sp_sview
,
2531 const struct sp_sampler
*sp_samp
,
2532 const float lod
[TGSI_QUAD_SIZE
],
2533 float level
[TGSI_QUAD_SIZE
])
2535 mip_rel_level_linear(sp_sview
, sp_samp
, lod
, level
);
2539 * Sample 2D texture using an anisotropic filter.
2542 mip_filter_linear_aniso(const struct sp_sampler_view
*sp_sview
,
2543 const struct sp_sampler
*sp_samp
,
2544 img_filter_func min_filter
,
2545 img_filter_func mag_filter
,
2546 const float s
[TGSI_QUAD_SIZE
],
2547 const float t
[TGSI_QUAD_SIZE
],
2548 const float p
[TGSI_QUAD_SIZE
],
2549 const float c0
[TGSI_QUAD_SIZE
],
2550 const float lod_in
[TGSI_QUAD_SIZE
],
2551 const struct filter_args
*filt_args
,
2552 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2554 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
2555 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2558 float lod
[TGSI_QUAD_SIZE
];
2560 const float s_to_u
= u_minify(texture
->width0
, psview
->u
.tex
.first_level
);
2561 const float t_to_v
= u_minify(texture
->height0
, psview
->u
.tex
.first_level
);
2562 const float dudx
= (s
[QUAD_BOTTOM_RIGHT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2563 const float dudy
= (s
[QUAD_TOP_LEFT
] - s
[QUAD_BOTTOM_LEFT
]) * s_to_u
;
2564 const float dvdx
= (t
[QUAD_BOTTOM_RIGHT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2565 const float dvdy
= (t
[QUAD_TOP_LEFT
] - t
[QUAD_BOTTOM_LEFT
]) * t_to_v
;
2566 struct img_filter_args args
;
2568 args
.offset
= filt_args
->offset
;
2570 if (filt_args
->control
== TGSI_SAMPLER_LOD_BIAS
||
2571 filt_args
->control
== TGSI_SAMPLER_LOD_NONE
||
2573 filt_args
->control
== TGSI_SAMPLER_DERIVS_EXPLICIT
) {
2574 /* note: instead of working with Px and Py, we will use the
2575 * squared length instead, to avoid sqrt.
2577 const float Px2
= dudx
* dudx
+ dvdx
* dvdx
;
2578 const float Py2
= dudy
* dudy
+ dvdy
* dvdy
;
2583 const float maxEccentricity
= sp_samp
->base
.max_anisotropy
* sp_samp
->base
.max_anisotropy
;
2594 /* if the eccentricity of the ellipse is too big, scale up the shorter
2595 * of the two vectors to limit the maximum amount of work per pixel
2598 if (e
> maxEccentricity
) {
2599 /* float s=e / maxEccentricity;
2603 Pmin2
= Pmax2
/ maxEccentricity
;
2606 /* note: we need to have Pmin=sqrt(Pmin2) here, but we can avoid
2607 * this since 0.5*log(x) = log(sqrt(x))
2609 lambda
= 0.5F
* util_fast_log2(Pmin2
) + sp_samp
->base
.lod_bias
;
2610 compute_lod(&sp_samp
->base
, filt_args
->control
, lambda
, lod_in
, lod
);
2613 assert(filt_args
->control
== TGSI_SAMPLER_LOD_EXPLICIT
||
2614 filt_args
->control
== TGSI_SAMPLER_LOD_ZERO
);
2615 compute_lod(&sp_samp
->base
, filt_args
->control
, sp_samp
->base
.lod_bias
, lod_in
, lod
);
2618 /* XXX: Take into account all lod values.
2621 level0
= psview
->u
.tex
.first_level
+ (int)lambda
;
2623 /* If the ellipse covers the whole image, we can
2624 * simply return the average of the whole image.
2626 if (level0
>= (int) psview
->u
.tex
.last_level
) {
2628 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2632 args
.level
= psview
->u
.tex
.last_level
;
2633 args
.face_id
= filt_args
->faces
[j
];
2635 * XXX: we overwrote any linear filter with nearest, so this
2636 * isn't right (albeit if last level is 1x1 and no border it
2637 * will work just the same).
2639 min_filter(sp_sview
, sp_samp
, &args
, &rgba
[0][j
]);
2643 /* don't bother interpolating between multiple LODs; it doesn't
2644 * seem to be worth the extra running time.
2646 img_filter_2d_ewa(sp_sview
, sp_samp
, min_filter
, mag_filter
,
2647 s
, t
, p
, filt_args
->faces
, filt_args
->offset
,
2648 level0
, dudx
, dvdx
, dudy
, dvdy
, rgba
);
2652 print_sample_4(__FUNCTION__
, rgba
);
2657 * Get mip level relative to base level for linear mip filter
2660 mip_rel_level_linear_2d_linear_repeat_POT(
2661 const struct sp_sampler_view
*sp_sview
,
2662 const struct sp_sampler
*sp_samp
,
2663 const float lod
[TGSI_QUAD_SIZE
],
2664 float level
[TGSI_QUAD_SIZE
])
2666 mip_rel_level_linear(sp_sview
, sp_samp
, lod
, level
);
2670 * Specialized version of mip_filter_linear with hard-wired calls to
2671 * 2d lambda calculation and 2d_linear_repeat_POT img filters.
2674 mip_filter_linear_2d_linear_repeat_POT(
2675 const struct sp_sampler_view
*sp_sview
,
2676 const struct sp_sampler
*sp_samp
,
2677 img_filter_func min_filter
,
2678 img_filter_func mag_filter
,
2679 const float s
[TGSI_QUAD_SIZE
],
2680 const float t
[TGSI_QUAD_SIZE
],
2681 const float p
[TGSI_QUAD_SIZE
],
2682 const float c0
[TGSI_QUAD_SIZE
],
2683 const float lod_in
[TGSI_QUAD_SIZE
],
2684 const struct filter_args
*filt_args
,
2685 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2687 const struct pipe_sampler_view
*psview
= &sp_sview
->base
;
2689 float lod
[TGSI_QUAD_SIZE
];
2691 compute_lambda_lod(sp_sview
, sp_samp
, s
, t
, p
, lod_in
, filt_args
->control
, lod
);
2693 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2694 const int level0
= psview
->u
.tex
.first_level
+ (int)lod
[j
];
2695 struct img_filter_args args
;
2696 /* Catches both negative and large values of level0:
2701 args
.face_id
= filt_args
->faces
[j
];
2702 args
.offset
= filt_args
->offset
;
2703 args
.gather_only
= filt_args
->control
== TGSI_SAMPLER_GATHER
;
2704 if ((unsigned)level0
>= psview
->u
.tex
.last_level
) {
2706 args
.level
= psview
->u
.tex
.first_level
;
2708 args
.level
= psview
->u
.tex
.last_level
;
2709 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
,
2714 const float levelBlend
= frac(lod
[j
]);
2715 float rgbax
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2718 args
.level
= level0
;
2719 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][0]);
2720 args
.level
= level0
+1;
2721 img_filter_2d_linear_repeat_POT(sp_sview
, sp_samp
, &args
, &rgbax
[0][1]);
2723 for (c
= 0; c
< TGSI_NUM_CHANNELS
; c
++)
2724 rgba
[c
][j
] = lerp(levelBlend
, rgbax
[c
][0], rgbax
[c
][1]);
2729 print_sample_4(__FUNCTION__
, rgba
);
2733 static const struct sp_filter_funcs funcs_linear
= {
2734 mip_rel_level_linear
,
2738 static const struct sp_filter_funcs funcs_nearest
= {
2739 mip_rel_level_nearest
,
2743 static const struct sp_filter_funcs funcs_none
= {
2748 static const struct sp_filter_funcs funcs_none_no_filter_select
= {
2749 mip_rel_level_none_no_filter_select
,
2750 mip_filter_none_no_filter_select
2753 static const struct sp_filter_funcs funcs_linear_aniso
= {
2754 mip_rel_level_linear_aniso
,
2755 mip_filter_linear_aniso
2758 static const struct sp_filter_funcs funcs_linear_2d_linear_repeat_POT
= {
2759 mip_rel_level_linear_2d_linear_repeat_POT
,
2760 mip_filter_linear_2d_linear_repeat_POT
2764 * Do shadow/depth comparisons.
2767 sample_compare(const struct sp_sampler_view
*sp_sview
,
2768 const struct sp_sampler
*sp_samp
,
2769 const float s
[TGSI_QUAD_SIZE
],
2770 const float t
[TGSI_QUAD_SIZE
],
2771 const float p
[TGSI_QUAD_SIZE
],
2772 const float c0
[TGSI_QUAD_SIZE
],
2773 const float c1
[TGSI_QUAD_SIZE
],
2774 enum tgsi_sampler_control control
,
2775 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2777 const struct pipe_sampler_state
*sampler
= &sp_samp
->base
;
2779 int k
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
2781 const struct util_format_description
*format_desc
=
2782 util_format_description(sp_sview
->base
.format
);
2783 /* not entirely sure we couldn't end up with non-valid swizzle here */
2784 const unsigned chan_type
=
2785 format_desc
->swizzle
[0] <= PIPE_SWIZZLE_W
?
2786 format_desc
->channel
[format_desc
->swizzle
[0]].type
:
2787 UTIL_FORMAT_TYPE_FLOAT
;
2788 const bool is_gather
= (control
== TGSI_SAMPLER_GATHER
);
2791 * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
2792 * for 2D Array texture we need to use the 'c0' (aka Q).
2793 * When we sampled the depth texture, the depth value was put into all
2794 * RGBA channels. We look at the red channel here.
2797 if (sp_sview
->base
.target
== PIPE_TEXTURE_2D_ARRAY
||
2798 sp_sview
->base
.target
== PIPE_TEXTURE_CUBE
) {
2803 } else if (sp_sview
->base
.target
== PIPE_TEXTURE_CUBE_ARRAY
) {
2815 if (chan_type
!= UTIL_FORMAT_TYPE_FLOAT
) {
2817 * clamping is a result of conversion to texture format, hence
2818 * doesn't happen with floats. Technically also should do comparison
2819 * in texture format (quantization!).
2821 pc
[0] = CLAMP(pc
[0], 0.0F
, 1.0F
);
2822 pc
[1] = CLAMP(pc
[1], 0.0F
, 1.0F
);
2823 pc
[2] = CLAMP(pc
[2], 0.0F
, 1.0F
);
2824 pc
[3] = CLAMP(pc
[3], 0.0F
, 1.0F
);
2827 for (v
= 0; v
< (is_gather
? TGSI_NUM_CHANNELS
: 1); v
++) {
2828 /* compare four texcoords vs. four texture samples */
2829 switch (sampler
->compare_func
) {
2830 case PIPE_FUNC_LESS
:
2831 k
[v
][0] = pc
[0] < rgba
[v
][0];
2832 k
[v
][1] = pc
[1] < rgba
[v
][1];
2833 k
[v
][2] = pc
[2] < rgba
[v
][2];
2834 k
[v
][3] = pc
[3] < rgba
[v
][3];
2836 case PIPE_FUNC_LEQUAL
:
2837 k
[v
][0] = pc
[0] <= rgba
[v
][0];
2838 k
[v
][1] = pc
[1] <= rgba
[v
][1];
2839 k
[v
][2] = pc
[2] <= rgba
[v
][2];
2840 k
[v
][3] = pc
[3] <= rgba
[v
][3];
2842 case PIPE_FUNC_GREATER
:
2843 k
[v
][0] = pc
[0] > rgba
[v
][0];
2844 k
[v
][1] = pc
[1] > rgba
[v
][1];
2845 k
[v
][2] = pc
[2] > rgba
[v
][2];
2846 k
[v
][3] = pc
[3] > rgba
[v
][3];
2848 case PIPE_FUNC_GEQUAL
:
2849 k
[v
][0] = pc
[0] >= rgba
[v
][0];
2850 k
[v
][1] = pc
[1] >= rgba
[v
][1];
2851 k
[v
][2] = pc
[2] >= rgba
[v
][2];
2852 k
[v
][3] = pc
[3] >= rgba
[v
][3];
2854 case PIPE_FUNC_EQUAL
:
2855 k
[v
][0] = pc
[0] == rgba
[v
][0];
2856 k
[v
][1] = pc
[1] == rgba
[v
][1];
2857 k
[v
][2] = pc
[2] == rgba
[v
][2];
2858 k
[v
][3] = pc
[3] == rgba
[v
][3];
2860 case PIPE_FUNC_NOTEQUAL
:
2861 k
[v
][0] = pc
[0] != rgba
[v
][0];
2862 k
[v
][1] = pc
[1] != rgba
[v
][1];
2863 k
[v
][2] = pc
[2] != rgba
[v
][2];
2864 k
[v
][3] = pc
[3] != rgba
[v
][3];
2866 case PIPE_FUNC_ALWAYS
:
2867 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 1;
2869 case PIPE_FUNC_NEVER
:
2870 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2873 k
[v
][0] = k
[v
][1] = k
[v
][2] = k
[v
][3] = 0;
2880 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2881 for (v
= 0; v
< TGSI_NUM_CHANNELS
; v
++) {
2882 rgba
[v
][j
] = k
[v
][j
];
2886 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
2887 rgba
[0][j
] = k
[0][j
];
2888 rgba
[1][j
] = k
[0][j
];
2889 rgba
[2][j
] = k
[0][j
];
2896 do_swizzling(const struct pipe_sampler_view
*sview
,
2897 float in
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
],
2898 float out
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
2901 const unsigned swizzle_r
= sview
->swizzle_r
;
2902 const unsigned swizzle_g
= sview
->swizzle_g
;
2903 const unsigned swizzle_b
= sview
->swizzle_b
;
2904 const unsigned swizzle_a
= sview
->swizzle_a
;
2905 float oneval
= util_format_is_pure_integer(sview
->format
) ? uif(1) : 1.0f
;
2907 switch (swizzle_r
) {
2908 case PIPE_SWIZZLE_0
:
2909 for (j
= 0; j
< 4; j
++)
2912 case PIPE_SWIZZLE_1
:
2913 for (j
= 0; j
< 4; j
++)
2917 assert(swizzle_r
< 4);
2918 for (j
= 0; j
< 4; j
++)
2919 out
[0][j
] = in
[swizzle_r
][j
];
2922 switch (swizzle_g
) {
2923 case PIPE_SWIZZLE_0
:
2924 for (j
= 0; j
< 4; j
++)
2927 case PIPE_SWIZZLE_1
:
2928 for (j
= 0; j
< 4; j
++)
2932 assert(swizzle_g
< 4);
2933 for (j
= 0; j
< 4; j
++)
2934 out
[1][j
] = in
[swizzle_g
][j
];
2937 switch (swizzle_b
) {
2938 case PIPE_SWIZZLE_0
:
2939 for (j
= 0; j
< 4; j
++)
2942 case PIPE_SWIZZLE_1
:
2943 for (j
= 0; j
< 4; j
++)
2947 assert(swizzle_b
< 4);
2948 for (j
= 0; j
< 4; j
++)
2949 out
[2][j
] = in
[swizzle_b
][j
];
2952 switch (swizzle_a
) {
2953 case PIPE_SWIZZLE_0
:
2954 for (j
= 0; j
< 4; j
++)
2957 case PIPE_SWIZZLE_1
:
2958 for (j
= 0; j
< 4; j
++)
2962 assert(swizzle_a
< 4);
2963 for (j
= 0; j
< 4; j
++)
2964 out
[3][j
] = in
[swizzle_a
][j
];
2969 static wrap_nearest_func
2970 get_nearest_unorm_wrap(unsigned mode
)
2973 case PIPE_TEX_WRAP_CLAMP
:
2974 return wrap_nearest_unorm_clamp
;
2975 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2976 return wrap_nearest_unorm_clamp_to_edge
;
2977 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2978 return wrap_nearest_unorm_clamp_to_border
;
2980 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
2981 return wrap_nearest_unorm_clamp
;
2986 static wrap_nearest_func
2987 get_nearest_wrap(unsigned mode
)
2990 case PIPE_TEX_WRAP_REPEAT
:
2991 return wrap_nearest_repeat
;
2992 case PIPE_TEX_WRAP_CLAMP
:
2993 return wrap_nearest_clamp
;
2994 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
2995 return wrap_nearest_clamp_to_edge
;
2996 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
2997 return wrap_nearest_clamp_to_border
;
2998 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
2999 return wrap_nearest_mirror_repeat
;
3000 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
3001 return wrap_nearest_mirror_clamp
;
3002 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
3003 return wrap_nearest_mirror_clamp_to_edge
;
3004 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
3005 return wrap_nearest_mirror_clamp_to_border
;
3008 return wrap_nearest_repeat
;
3013 static wrap_linear_func
3014 get_linear_unorm_wrap(unsigned mode
)
3017 case PIPE_TEX_WRAP_CLAMP
:
3018 return wrap_linear_unorm_clamp
;
3019 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
3020 return wrap_linear_unorm_clamp_to_edge
;
3021 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
3022 return wrap_linear_unorm_clamp_to_border
;
3024 debug_printf("illegal wrap mode %d with non-normalized coords\n", mode
);
3025 return wrap_linear_unorm_clamp
;
3030 static wrap_linear_func
3031 get_linear_wrap(unsigned mode
)
3034 case PIPE_TEX_WRAP_REPEAT
:
3035 return wrap_linear_repeat
;
3036 case PIPE_TEX_WRAP_CLAMP
:
3037 return wrap_linear_clamp
;
3038 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
3039 return wrap_linear_clamp_to_edge
;
3040 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
3041 return wrap_linear_clamp_to_border
;
3042 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
3043 return wrap_linear_mirror_repeat
;
3044 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
3045 return wrap_linear_mirror_clamp
;
3046 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
3047 return wrap_linear_mirror_clamp_to_edge
;
3048 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
3049 return wrap_linear_mirror_clamp_to_border
;
3052 return wrap_linear_repeat
;
3058 * Is swizzling needed for the given state key?
3061 any_swizzle(const struct pipe_sampler_view
*view
)
3063 return (view
->swizzle_r
!= PIPE_SWIZZLE_X
||
3064 view
->swizzle_g
!= PIPE_SWIZZLE_Y
||
3065 view
->swizzle_b
!= PIPE_SWIZZLE_Z
||
3066 view
->swizzle_a
!= PIPE_SWIZZLE_W
);
3070 static img_filter_func
3071 get_img_filter(const struct sp_sampler_view
*sp_sview
,
3072 const struct pipe_sampler_state
*sampler
,
3073 unsigned filter
, bool gather
)
3075 switch (sp_sview
->base
.target
) {
3077 case PIPE_TEXTURE_1D
:
3078 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3079 return img_filter_1d_nearest
;
3081 return img_filter_1d_linear
;
3083 case PIPE_TEXTURE_1D_ARRAY
:
3084 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3085 return img_filter_1d_array_nearest
;
3087 return img_filter_1d_array_linear
;
3089 case PIPE_TEXTURE_2D
:
3090 case PIPE_TEXTURE_RECT
:
3091 /* Try for fast path:
3093 if (!gather
&& sp_sview
->pot2d
&&
3094 sampler
->wrap_s
== sampler
->wrap_t
&&
3095 sampler
->normalized_coords
)
3097 switch (sampler
->wrap_s
) {
3098 case PIPE_TEX_WRAP_REPEAT
:
3100 case PIPE_TEX_FILTER_NEAREST
:
3101 return img_filter_2d_nearest_repeat_POT
;
3102 case PIPE_TEX_FILTER_LINEAR
:
3103 return img_filter_2d_linear_repeat_POT
;
3108 case PIPE_TEX_WRAP_CLAMP
:
3110 case PIPE_TEX_FILTER_NEAREST
:
3111 return img_filter_2d_nearest_clamp_POT
;
3117 /* Otherwise use default versions:
3119 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3120 return img_filter_2d_nearest
;
3122 return img_filter_2d_linear
;
3124 case PIPE_TEXTURE_2D_ARRAY
:
3125 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3126 return img_filter_2d_array_nearest
;
3128 return img_filter_2d_array_linear
;
3130 case PIPE_TEXTURE_CUBE
:
3131 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3132 return img_filter_cube_nearest
;
3134 return img_filter_cube_linear
;
3136 case PIPE_TEXTURE_CUBE_ARRAY
:
3137 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3138 return img_filter_cube_array_nearest
;
3140 return img_filter_cube_array_linear
;
3142 case PIPE_TEXTURE_3D
:
3143 if (filter
== PIPE_TEX_FILTER_NEAREST
)
3144 return img_filter_3d_nearest
;
3146 return img_filter_3d_linear
;
3150 return img_filter_1d_nearest
;
3155 * Get mip filter funcs, and optionally both img min filter and img mag
3156 * filter. Note that both img filter function pointers must be either non-NULL
3160 get_filters(const struct sp_sampler_view
*sp_sview
,
3161 const struct sp_sampler
*sp_samp
,
3162 const enum tgsi_sampler_control control
,
3163 const struct sp_filter_funcs
**funcs
,
3164 img_filter_func
*min
,
3165 img_filter_func
*mag
)
3168 if (control
== TGSI_SAMPLER_GATHER
) {
3169 *funcs
= &funcs_nearest
;
3171 *min
= get_img_filter(sp_sview
, &sp_samp
->base
,
3172 PIPE_TEX_FILTER_LINEAR
, true);
3174 } else if (sp_sview
->pot2d
& sp_samp
->min_mag_equal_repeat_linear
) {
3175 *funcs
= &funcs_linear_2d_linear_repeat_POT
;
3177 *funcs
= sp_samp
->filter_funcs
;
3180 *min
= get_img_filter(sp_sview
, &sp_samp
->base
,
3181 sp_samp
->min_img_filter
, false);
3182 if (sp_samp
->min_mag_equal
) {
3185 *mag
= get_img_filter(sp_sview
, &sp_samp
->base
,
3186 sp_samp
->base
.mag_img_filter
, false);
3193 sample_mip(const struct sp_sampler_view
*sp_sview
,
3194 const struct sp_sampler
*sp_samp
,
3195 const float s
[TGSI_QUAD_SIZE
],
3196 const float t
[TGSI_QUAD_SIZE
],
3197 const float p
[TGSI_QUAD_SIZE
],
3198 const float c0
[TGSI_QUAD_SIZE
],
3199 const float lod
[TGSI_QUAD_SIZE
],
3200 const struct filter_args
*filt_args
,
3201 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3203 const struct sp_filter_funcs
*funcs
= NULL
;
3204 img_filter_func min_img_filter
= NULL
;
3205 img_filter_func mag_img_filter
= NULL
;
3207 get_filters(sp_sview
, sp_samp
, filt_args
->control
,
3208 &funcs
, &min_img_filter
, &mag_img_filter
);
3210 funcs
->filter(sp_sview
, sp_samp
, min_img_filter
, mag_img_filter
,
3211 s
, t
, p
, c0
, lod
, filt_args
, rgba
);
3213 if (sp_samp
->base
.compare_mode
!= PIPE_TEX_COMPARE_NONE
) {
3214 sample_compare(sp_sview
, sp_samp
, s
, t
, p
, c0
,
3215 lod
, filt_args
->control
, rgba
);
3218 if (sp_sview
->need_swizzle
&& filt_args
->control
!= TGSI_SAMPLER_GATHER
) {
3219 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3220 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
3221 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
3228 * This function uses cube texture coordinates to choose a face of a cube and
3229 * computes the 2D cube face coordinates. Puts face info into the sampler
3233 convert_cube(const struct sp_sampler_view
*sp_sview
,
3234 const struct sp_sampler
*sp_samp
,
3235 const float s
[TGSI_QUAD_SIZE
],
3236 const float t
[TGSI_QUAD_SIZE
],
3237 const float p
[TGSI_QUAD_SIZE
],
3238 const float c0
[TGSI_QUAD_SIZE
],
3239 float ssss
[TGSI_QUAD_SIZE
],
3240 float tttt
[TGSI_QUAD_SIZE
],
3241 float pppp
[TGSI_QUAD_SIZE
],
3242 uint faces
[TGSI_QUAD_SIZE
])
3252 direction target sc tc ma
3253 ---------- ------------------------------- --- --- ---
3254 +rx TEXTURE_CUBE_MAP_POSITIVE_X_EXT -rz -ry rx
3255 -rx TEXTURE_CUBE_MAP_NEGATIVE_X_EXT +rz -ry rx
3256 +ry TEXTURE_CUBE_MAP_POSITIVE_Y_EXT +rx +rz ry
3257 -ry TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT +rx -rz ry
3258 +rz TEXTURE_CUBE_MAP_POSITIVE_Z_EXT +rx -ry rz
3259 -rz TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT -rx -ry rz
3262 /* Choose the cube face and compute new s/t coords for the 2D face.
3264 * Use the same cube face for all four pixels in the quad.
3266 * This isn't ideal, but if we want to use a different cube face
3267 * per pixel in the quad, we'd have to also compute the per-face
3268 * LOD here too. That's because the four post-face-selection
3269 * texcoords are no longer related to each other (they're
3270 * per-face!) so we can't use subtraction to compute the partial
3271 * deriviates to compute the LOD. Doing so (near cube edges
3272 * anyway) gives us pretty much random values.
3275 /* use the average of the four pixel's texcoords to choose the face */
3276 const float rx
= 0.25F
* (s
[0] + s
[1] + s
[2] + s
[3]);
3277 const float ry
= 0.25F
* (t
[0] + t
[1] + t
[2] + t
[3]);
3278 const float rz
= 0.25F
* (p
[0] + p
[1] + p
[2] + p
[3]);
3279 const float arx
= fabsf(rx
), ary
= fabsf(ry
), arz
= fabsf(rz
);
3281 if (arx
>= ary
&& arx
>= arz
) {
3282 const float sign
= (rx
>= 0.0F
) ? 1.0F
: -1.0F
;
3283 const uint face
= (rx
>= 0.0F
) ?
3284 PIPE_TEX_FACE_POS_X
: PIPE_TEX_FACE_NEG_X
;
3285 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3286 const float ima
= -0.5F
/ fabsf(s
[j
]);
3287 ssss
[j
] = sign
* p
[j
] * ima
+ 0.5F
;
3288 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3292 else if (ary
>= arx
&& ary
>= arz
) {
3293 const float sign
= (ry
>= 0.0F
) ? 1.0F
: -1.0F
;
3294 const uint face
= (ry
>= 0.0F
) ?
3295 PIPE_TEX_FACE_POS_Y
: PIPE_TEX_FACE_NEG_Y
;
3296 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3297 const float ima
= -0.5F
/ fabsf(t
[j
]);
3298 ssss
[j
] = -s
[j
] * ima
+ 0.5F
;
3299 tttt
[j
] = sign
* -p
[j
] * ima
+ 0.5F
;
3304 const float sign
= (rz
>= 0.0F
) ? 1.0F
: -1.0F
;
3305 const uint face
= (rz
>= 0.0F
) ?
3306 PIPE_TEX_FACE_POS_Z
: PIPE_TEX_FACE_NEG_Z
;
3307 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3308 const float ima
= -0.5F
/ fabsf(p
[j
]);
3309 ssss
[j
] = sign
* -s
[j
] * ima
+ 0.5F
;
3310 tttt
[j
] = t
[j
] * ima
+ 0.5F
;
3319 sp_get_dims(const struct sp_sampler_view
*sp_sview
,
3323 const struct pipe_sampler_view
*view
= &sp_sview
->base
;
3324 const struct pipe_resource
*texture
= view
->texture
;
3326 if (view
->target
== PIPE_BUFFER
) {
3327 dims
[0] = view
->u
.buf
.size
/ util_format_get_blocksize(view
->format
);
3328 /* the other values are undefined, but let's avoid potential valgrind
3331 dims
[1] = dims
[2] = dims
[3] = 0;
3335 /* undefined according to EXT_gpu_program */
3336 level
+= view
->u
.tex
.first_level
;
3337 if (level
> view
->u
.tex
.last_level
)
3340 dims
[3] = view
->u
.tex
.last_level
- view
->u
.tex
.first_level
+ 1;
3341 dims
[0] = u_minify(texture
->width0
, level
);
3343 switch (view
->target
) {
3344 case PIPE_TEXTURE_1D_ARRAY
:
3345 dims
[1] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3347 case PIPE_TEXTURE_1D
:
3349 case PIPE_TEXTURE_2D_ARRAY
:
3350 dims
[2] = view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1;
3352 case PIPE_TEXTURE_2D
:
3353 case PIPE_TEXTURE_CUBE
:
3354 case PIPE_TEXTURE_RECT
:
3355 dims
[1] = u_minify(texture
->height0
, level
);
3357 case PIPE_TEXTURE_3D
:
3358 dims
[1] = u_minify(texture
->height0
, level
);
3359 dims
[2] = u_minify(texture
->depth0
, level
);
3361 case PIPE_TEXTURE_CUBE_ARRAY
:
3362 dims
[1] = u_minify(texture
->height0
, level
);
3363 dims
[2] = (view
->u
.tex
.last_layer
- view
->u
.tex
.first_layer
+ 1) / 6;
3366 assert(!"unexpected texture target in sp_get_dims()");
3372 * This function is only used for getting unfiltered texels via the
3373 * TXF opcode. The GL spec says that out-of-bounds texel fetches
3374 * produce undefined results. Instead of crashing, lets just clamp
3375 * coords to the texture image size.
3378 sp_get_texels(const struct sp_sampler_view
*sp_sview
,
3379 const int v_i
[TGSI_QUAD_SIZE
],
3380 const int v_j
[TGSI_QUAD_SIZE
],
3381 const int v_k
[TGSI_QUAD_SIZE
],
3382 const int lod
[TGSI_QUAD_SIZE
],
3383 const int8_t offset
[3],
3384 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3386 union tex_tile_address addr
;
3387 const struct pipe_resource
*texture
= sp_sview
->base
.texture
;
3390 /* TODO write a better test for LOD */
3391 const unsigned level
=
3392 sp_sview
->base
.target
== PIPE_BUFFER
? 0 :
3393 CLAMP(lod
[0] + sp_sview
->base
.u
.tex
.first_level
,
3394 sp_sview
->base
.u
.tex
.first_level
,
3395 sp_sview
->base
.u
.tex
.last_level
);
3396 const int width
= u_minify(texture
->width0
, level
);
3397 const int height
= u_minify(texture
->height0
, level
);
3398 const int depth
= u_minify(texture
->depth0
, level
);
3399 unsigned elem_size
, first_element
, last_element
;
3402 addr
.bits
.level
= level
;
3404 switch (sp_sview
->base
.target
) {
3406 elem_size
= util_format_get_blocksize(sp_sview
->base
.format
);
3407 first_element
= sp_sview
->base
.u
.buf
.offset
/ elem_size
;
3408 last_element
= (sp_sview
->base
.u
.buf
.offset
+
3409 sp_sview
->base
.u
.buf
.size
) / elem_size
- 1;
3410 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3411 const int x
= CLAMP(v_i
[j
] + offset
[0] +
3415 tx
= get_texel_buffer_no_border(sp_sview
, addr
, x
, elem_size
);
3416 for (c
= 0; c
< 4; c
++) {
3421 case PIPE_TEXTURE_1D
:
3422 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3423 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3424 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
,
3425 sp_sview
->base
.u
.tex
.first_layer
);
3426 for (c
= 0; c
< 4; c
++) {
3431 case PIPE_TEXTURE_1D_ARRAY
:
3432 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3433 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3434 const int y
= CLAMP(v_j
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3435 sp_sview
->base
.u
.tex
.last_layer
);
3436 tx
= get_texel_2d_no_border(sp_sview
, addr
, x
, y
);
3437 for (c
= 0; c
< 4; c
++) {
3442 case PIPE_TEXTURE_2D
:
3443 case PIPE_TEXTURE_RECT
:
3444 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3445 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3446 const int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3447 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
,
3448 sp_sview
->base
.u
.tex
.first_layer
);
3449 for (c
= 0; c
< 4; c
++) {
3454 case PIPE_TEXTURE_2D_ARRAY
:
3455 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3456 const int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3457 const int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3458 const int layer
= CLAMP(v_k
[j
], sp_sview
->base
.u
.tex
.first_layer
,
3459 sp_sview
->base
.u
.tex
.last_layer
);
3460 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, layer
);
3461 for (c
= 0; c
< 4; c
++) {
3466 case PIPE_TEXTURE_3D
:
3467 for (j
= 0; j
< TGSI_QUAD_SIZE
; j
++) {
3468 int x
= CLAMP(v_i
[j
] + offset
[0], 0, width
- 1);
3469 int y
= CLAMP(v_j
[j
] + offset
[1], 0, height
- 1);
3470 int z
= CLAMP(v_k
[j
] + offset
[2], 0, depth
- 1);
3471 tx
= get_texel_3d_no_border(sp_sview
, addr
, x
, y
, z
);
3472 for (c
= 0; c
< 4; c
++) {
3477 case PIPE_TEXTURE_CUBE
: /* TXF can't work on CUBE according to spec */
3478 case PIPE_TEXTURE_CUBE_ARRAY
:
3480 assert(!"Unknown or CUBE texture type in TXF processing\n");
3484 if (sp_sview
->need_swizzle
) {
3485 float rgba_temp
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
];
3486 memcpy(rgba_temp
, rgba
, sizeof(rgba_temp
));
3487 do_swizzling(&sp_sview
->base
, rgba_temp
, rgba
);
3493 softpipe_create_sampler_state(struct pipe_context
*pipe
,
3494 const struct pipe_sampler_state
*sampler
)
3496 struct sp_sampler
*samp
= CALLOC_STRUCT(sp_sampler
);
3498 samp
->base
= *sampler
;
3500 /* Note that (for instance) linear_texcoord_s and
3501 * nearest_texcoord_s may be active at the same time, if the
3502 * sampler min_img_filter differs from its mag_img_filter.
3504 if (sampler
->normalized_coords
) {
3505 samp
->linear_texcoord_s
= get_linear_wrap( sampler
->wrap_s
);
3506 samp
->linear_texcoord_t
= get_linear_wrap( sampler
->wrap_t
);
3507 samp
->linear_texcoord_p
= get_linear_wrap( sampler
->wrap_r
);
3509 samp
->nearest_texcoord_s
= get_nearest_wrap( sampler
->wrap_s
);
3510 samp
->nearest_texcoord_t
= get_nearest_wrap( sampler
->wrap_t
);
3511 samp
->nearest_texcoord_p
= get_nearest_wrap( sampler
->wrap_r
);
3514 samp
->linear_texcoord_s
= get_linear_unorm_wrap( sampler
->wrap_s
);
3515 samp
->linear_texcoord_t
= get_linear_unorm_wrap( sampler
->wrap_t
);
3516 samp
->linear_texcoord_p
= get_linear_unorm_wrap( sampler
->wrap_r
);
3518 samp
->nearest_texcoord_s
= get_nearest_unorm_wrap( sampler
->wrap_s
);
3519 samp
->nearest_texcoord_t
= get_nearest_unorm_wrap( sampler
->wrap_t
);
3520 samp
->nearest_texcoord_p
= get_nearest_unorm_wrap( sampler
->wrap_r
);
3523 samp
->min_img_filter
= sampler
->min_img_filter
;
3525 switch (sampler
->min_mip_filter
) {
3526 case PIPE_TEX_MIPFILTER_NONE
:
3527 if (sampler
->min_img_filter
== sampler
->mag_img_filter
)
3528 samp
->filter_funcs
= &funcs_none_no_filter_select
;
3530 samp
->filter_funcs
= &funcs_none
;
3533 case PIPE_TEX_MIPFILTER_NEAREST
:
3534 samp
->filter_funcs
= &funcs_nearest
;
3537 case PIPE_TEX_MIPFILTER_LINEAR
:
3538 if (sampler
->min_img_filter
== sampler
->mag_img_filter
&&
3539 sampler
->normalized_coords
&&
3540 sampler
->wrap_s
== PIPE_TEX_WRAP_REPEAT
&&
3541 sampler
->wrap_t
== PIPE_TEX_WRAP_REPEAT
&&
3542 sampler
->min_img_filter
== PIPE_TEX_FILTER_LINEAR
&&
3543 sampler
->max_anisotropy
<= 1) {
3544 samp
->min_mag_equal_repeat_linear
= TRUE
;
3546 samp
->filter_funcs
= &funcs_linear
;
3548 /* Anisotropic filtering extension. */
3549 if (sampler
->max_anisotropy
> 1) {
3550 samp
->filter_funcs
= &funcs_linear_aniso
;
3552 /* Override min_img_filter:
3553 * min_img_filter needs to be set to NEAREST since we need to access
3554 * each texture pixel as it is and weight it later; using linear
3555 * filters will have incorrect results.
3556 * By setting the filter to NEAREST here, we can avoid calling the
3557 * generic img_filter_2d_nearest in the anisotropic filter function,
3558 * making it possible to use one of the accelerated implementations
3560 samp
->min_img_filter
= PIPE_TEX_FILTER_NEAREST
;
3562 /* on first access create the lookup table containing the filter weights. */
3564 create_filter_table();
3569 if (samp
->min_img_filter
== sampler
->mag_img_filter
) {
3570 samp
->min_mag_equal
= TRUE
;
3573 return (void *)samp
;
3578 softpipe_get_lambda_func(const struct pipe_sampler_view
*view
,
3579 enum pipe_shader_type shader
)
3581 if (shader
!= PIPE_SHADER_FRAGMENT
)
3582 return compute_lambda_vert
;
3584 switch (view
->target
) {
3586 case PIPE_TEXTURE_1D
:
3587 case PIPE_TEXTURE_1D_ARRAY
:
3588 return compute_lambda_1d
;
3589 case PIPE_TEXTURE_2D
:
3590 case PIPE_TEXTURE_2D_ARRAY
:
3591 case PIPE_TEXTURE_RECT
:
3592 case PIPE_TEXTURE_CUBE
:
3593 case PIPE_TEXTURE_CUBE_ARRAY
:
3594 return compute_lambda_2d
;
3595 case PIPE_TEXTURE_3D
:
3596 return compute_lambda_3d
;
3599 return compute_lambda_1d
;
3604 struct pipe_sampler_view
*
3605 softpipe_create_sampler_view(struct pipe_context
*pipe
,
3606 struct pipe_resource
*resource
,
3607 const struct pipe_sampler_view
*templ
)
3609 struct sp_sampler_view
*sview
= CALLOC_STRUCT(sp_sampler_view
);
3610 const struct softpipe_resource
*spr
= (struct softpipe_resource
*)resource
;
3613 struct pipe_sampler_view
*view
= &sview
->base
;
3615 view
->reference
.count
= 1;
3616 view
->texture
= NULL
;
3617 pipe_resource_reference(&view
->texture
, resource
);
3618 view
->context
= pipe
;
3622 * This is possibly too lenient, but the primary reason is just
3623 * to catch state trackers which forget to initialize this, so
3624 * it only catches clearly impossible view targets.
3626 if (view
->target
!= resource
->target
) {
3627 if (view
->target
== PIPE_TEXTURE_1D
)
3628 assert(resource
->target
== PIPE_TEXTURE_1D_ARRAY
);
3629 else if (view
->target
== PIPE_TEXTURE_1D_ARRAY
)
3630 assert(resource
->target
== PIPE_TEXTURE_1D
);
3631 else if (view
->target
== PIPE_TEXTURE_2D
)
3632 assert(resource
->target
== PIPE_TEXTURE_2D_ARRAY
||
3633 resource
->target
== PIPE_TEXTURE_CUBE
||
3634 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3635 else if (view
->target
== PIPE_TEXTURE_2D_ARRAY
)
3636 assert(resource
->target
== PIPE_TEXTURE_2D
||
3637 resource
->target
== PIPE_TEXTURE_CUBE
||
3638 resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3639 else if (view
->target
== PIPE_TEXTURE_CUBE
)
3640 assert(resource
->target
== PIPE_TEXTURE_CUBE_ARRAY
||
3641 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3642 else if (view
->target
== PIPE_TEXTURE_CUBE_ARRAY
)
3643 assert(resource
->target
== PIPE_TEXTURE_CUBE
||
3644 resource
->target
== PIPE_TEXTURE_2D_ARRAY
);
3650 if (any_swizzle(view
)) {
3651 sview
->need_swizzle
= TRUE
;
3654 sview
->need_cube_convert
= (view
->target
== PIPE_TEXTURE_CUBE
||
3655 view
->target
== PIPE_TEXTURE_CUBE_ARRAY
);
3656 sview
->pot2d
= spr
->pot
&&
3657 (view
->target
== PIPE_TEXTURE_2D
||
3658 view
->target
== PIPE_TEXTURE_RECT
);
3660 sview
->xpot
= util_logbase2( resource
->width0
);
3661 sview
->ypot
= util_logbase2( resource
->height0
);
3664 return (struct pipe_sampler_view
*) sview
;
3668 static inline const struct sp_tgsi_sampler
*
3669 sp_tgsi_sampler_cast_c(const struct tgsi_sampler
*sampler
)
3671 return (const struct sp_tgsi_sampler
*)sampler
;
3676 sp_tgsi_get_dims(struct tgsi_sampler
*tgsi_sampler
,
3677 const unsigned sview_index
,
3678 int level
, int dims
[4])
3680 const struct sp_tgsi_sampler
*sp_samp
=
3681 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3683 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3684 /* always have a view here but texture is NULL if no sampler view was set. */
3685 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3686 dims
[0] = dims
[1] = dims
[2] = dims
[3] = 0;
3689 sp_get_dims(&sp_samp
->sp_sview
[sview_index
], level
, dims
);
3694 sp_tgsi_get_samples(struct tgsi_sampler
*tgsi_sampler
,
3695 const unsigned sview_index
,
3696 const unsigned sampler_index
,
3697 const float s
[TGSI_QUAD_SIZE
],
3698 const float t
[TGSI_QUAD_SIZE
],
3699 const float p
[TGSI_QUAD_SIZE
],
3700 const float c0
[TGSI_QUAD_SIZE
],
3701 const float lod
[TGSI_QUAD_SIZE
],
3702 float derivs
[3][2][TGSI_QUAD_SIZE
],
3703 const int8_t offset
[3],
3704 enum tgsi_sampler_control control
,
3705 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3707 const struct sp_tgsi_sampler
*sp_tgsi_samp
=
3708 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3709 const struct sp_sampler_view
*sp_sview
;
3710 const struct sp_sampler
*sp_samp
;
3711 struct filter_args filt_args
;
3713 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3714 assert(sampler_index
< PIPE_MAX_SAMPLERS
);
3715 assert(sp_tgsi_samp
->sp_sampler
[sampler_index
]);
3717 sp_sview
= &sp_tgsi_samp
->sp_sview
[sview_index
];
3718 sp_samp
= sp_tgsi_samp
->sp_sampler
[sampler_index
];
3719 /* always have a view here but texture is NULL if no sampler view was set. */
3720 if (!sp_sview
->base
.texture
) {
3722 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3723 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3730 filt_args
.control
= control
;
3731 filt_args
.offset
= offset
;
3733 if (sp_sview
->need_cube_convert
) {
3734 float cs
[TGSI_QUAD_SIZE
];
3735 float ct
[TGSI_QUAD_SIZE
];
3736 float cp
[TGSI_QUAD_SIZE
];
3737 uint faces
[TGSI_QUAD_SIZE
];
3739 convert_cube(sp_sview
, sp_samp
, s
, t
, p
, c0
, cs
, ct
, cp
, faces
);
3741 filt_args
.faces
= faces
;
3742 sample_mip(sp_sview
, sp_samp
, cs
, ct
, cp
, c0
, lod
, &filt_args
, rgba
);
3744 static const uint zero_faces
[TGSI_QUAD_SIZE
] = {0, 0, 0, 0};
3746 filt_args
.faces
= zero_faces
;
3747 sample_mip(sp_sview
, sp_samp
, s
, t
, p
, c0
, lod
, &filt_args
, rgba
);
3752 sp_tgsi_query_lod(const struct tgsi_sampler
*tgsi_sampler
,
3753 const unsigned sview_index
,
3754 const unsigned sampler_index
,
3755 const float s
[TGSI_QUAD_SIZE
],
3756 const float t
[TGSI_QUAD_SIZE
],
3757 const float p
[TGSI_QUAD_SIZE
],
3758 const float c0
[TGSI_QUAD_SIZE
],
3759 const enum tgsi_sampler_control control
,
3760 float mipmap
[TGSI_QUAD_SIZE
],
3761 float lod
[TGSI_QUAD_SIZE
])
3763 static const float lod_in
[TGSI_QUAD_SIZE
] = { 0.0, 0.0, 0.0, 0.0 };
3765 const struct sp_tgsi_sampler
*sp_tgsi_samp
=
3766 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3767 const struct sp_sampler_view
*sp_sview
;
3768 const struct sp_sampler
*sp_samp
;
3769 const struct sp_filter_funcs
*funcs
;
3772 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3773 assert(sampler_index
< PIPE_MAX_SAMPLERS
);
3774 assert(sp_tgsi_samp
->sp_sampler
[sampler_index
]);
3776 sp_sview
= &sp_tgsi_samp
->sp_sview
[sview_index
];
3777 sp_samp
= sp_tgsi_samp
->sp_sampler
[sampler_index
];
3778 /* always have a view here but texture is NULL if no sampler view was
3780 if (!sp_sview
->base
.texture
) {
3781 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3788 if (sp_sview
->need_cube_convert
) {
3789 float cs
[TGSI_QUAD_SIZE
];
3790 float ct
[TGSI_QUAD_SIZE
];
3791 float cp
[TGSI_QUAD_SIZE
];
3792 uint unused_faces
[TGSI_QUAD_SIZE
];
3794 convert_cube(sp_sview
, sp_samp
, s
, t
, p
, c0
, cs
, ct
, cp
, unused_faces
);
3795 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
3796 cs
, ct
, cp
, lod_in
, control
, lod
);
3798 compute_lambda_lod_unclamped(sp_sview
, sp_samp
,
3799 s
, t
, p
, lod_in
, control
, lod
);
3802 get_filters(sp_sview
, sp_samp
, control
, &funcs
, NULL
, NULL
);
3803 funcs
->relative_level(sp_sview
, sp_samp
, lod
, mipmap
);
3807 sp_tgsi_get_texel(struct tgsi_sampler
*tgsi_sampler
,
3808 const unsigned sview_index
,
3809 const int i
[TGSI_QUAD_SIZE
],
3810 const int j
[TGSI_QUAD_SIZE
], const int k
[TGSI_QUAD_SIZE
],
3811 const int lod
[TGSI_QUAD_SIZE
], const int8_t offset
[3],
3812 float rgba
[TGSI_NUM_CHANNELS
][TGSI_QUAD_SIZE
])
3814 const struct sp_tgsi_sampler
*sp_samp
=
3815 sp_tgsi_sampler_cast_c(tgsi_sampler
);
3817 assert(sview_index
< PIPE_MAX_SHADER_SAMPLER_VIEWS
);
3818 /* always have a view here but texture is NULL if no sampler view was set. */
3819 if (!sp_samp
->sp_sview
[sview_index
].base
.texture
) {
3821 for (j
= 0; j
< TGSI_NUM_CHANNELS
; j
++) {
3822 for (i
= 0; i
< TGSI_QUAD_SIZE
; i
++) {
3828 sp_get_texels(&sp_samp
->sp_sview
[sview_index
], i
, j
, k
, lod
, offset
, rgba
);
3832 struct sp_tgsi_sampler
*
3833 sp_create_tgsi_sampler(void)
3835 struct sp_tgsi_sampler
*samp
= CALLOC_STRUCT(sp_tgsi_sampler
);
3839 samp
->base
.get_dims
= sp_tgsi_get_dims
;
3840 samp
->base
.get_samples
= sp_tgsi_get_samples
;
3841 samp
->base
.get_texel
= sp_tgsi_get_texel
;
3842 samp
->base
.query_lod
= sp_tgsi_query_lod
;