2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **********************************************************************/
29 * Keith Whitwell <keith@tungstengraphics.com>
33 #include "brw_context.h"
34 #include "brw_state.h"
35 #include "brw_defines.h"
37 #include "main/macros.h"
41 /* Samplers aren't strictly wm state from the hardware's perspective,
42 * but that is the only situation in which we use them in this driver.
47 /* The brw (and related graphics cores) do not support GL_CLAMP. The
48 * Intel drivers for "other operating systems" implement GL_CLAMP as
49 * GL_CLAMP_TO_EDGE, so the same is done here.
51 static GLuint
translate_wrap_mode( GLenum wrap
)
55 return BRW_TEXCOORDMODE_WRAP
;
57 return BRW_TEXCOORDMODE_CLAMP
;
58 case GL_CLAMP_TO_EDGE
:
59 return BRW_TEXCOORDMODE_CLAMP
; /* conform likes it this way */
60 case GL_CLAMP_TO_BORDER
:
61 return BRW_TEXCOORDMODE_CLAMP_BORDER
;
62 case GL_MIRRORED_REPEAT
:
63 return BRW_TEXCOORDMODE_MIRROR
;
65 return BRW_TEXCOORDMODE_WRAP
;
69 static drm_intel_bo
*upload_default_color( struct brw_context
*brw
,
70 const GLfloat
*color
)
72 struct intel_context
*intel
= &brw
->intel
;
74 if (intel
->gen
>= 5) {
75 struct gen5_sampler_default_color sdc
;
77 memset(&sdc
, 0, sizeof(sdc
));
79 UNCLAMPED_FLOAT_TO_UBYTE(sdc
.ub
[0], color
[0]);
80 UNCLAMPED_FLOAT_TO_UBYTE(sdc
.ub
[1], color
[1]);
81 UNCLAMPED_FLOAT_TO_UBYTE(sdc
.ub
[2], color
[2]);
82 UNCLAMPED_FLOAT_TO_UBYTE(sdc
.ub
[3], color
[3]);
84 UNCLAMPED_FLOAT_TO_USHORT(sdc
.us
[0], color
[0]);
85 UNCLAMPED_FLOAT_TO_USHORT(sdc
.us
[1], color
[1]);
86 UNCLAMPED_FLOAT_TO_USHORT(sdc
.us
[2], color
[2]);
87 UNCLAMPED_FLOAT_TO_USHORT(sdc
.us
[3], color
[3]);
89 UNCLAMPED_FLOAT_TO_SHORT(sdc
.s
[0], color
[0]);
90 UNCLAMPED_FLOAT_TO_SHORT(sdc
.s
[1], color
[1]);
91 UNCLAMPED_FLOAT_TO_SHORT(sdc
.s
[2], color
[2]);
92 UNCLAMPED_FLOAT_TO_SHORT(sdc
.s
[3], color
[3]);
94 /* XXX: Fill in half floats */
95 /* XXX: Fill in signed bytes */
97 COPY_4V(sdc
.f
, color
);
99 return brw_cache_data(&brw
->cache
, BRW_SAMPLER_DEFAULT_COLOR
,
102 struct brw_sampler_default_color sdc
;
104 COPY_4V(sdc
.color
, color
);
106 return brw_cache_data(&brw
->cache
, BRW_SAMPLER_DEFAULT_COLOR
,
112 struct wm_sampler_key
{
115 struct wm_sampler_entry
{
117 GLenum wrap_r
, wrap_s
, wrap_t
;
118 float maxlod
, minlod
;
121 GLenum minfilter
, magfilter
;
122 GLenum comparemode
, comparefunc
;
124 /** If target is cubemap, take context setting.
126 GLboolean seamless_cube_map
;
127 } sampler
[BRW_MAX_TEX_UNIT
];
131 * Sets the sampler state for a single unit based off of the sampler key
134 static void brw_update_sampler_state(struct brw_context
*brw
,
135 struct wm_sampler_entry
*key
,
136 drm_intel_bo
*sdc_bo
,
137 struct brw_sampler_state
*sampler
)
139 struct intel_context
*intel
= &brw
->intel
;
141 memset(sampler
, 0, sizeof(*sampler
));
143 switch (key
->minfilter
) {
145 sampler
->ss0
.min_filter
= BRW_MAPFILTER_NEAREST
;
146 sampler
->ss0
.mip_filter
= BRW_MIPFILTER_NONE
;
149 sampler
->ss0
.min_filter
= BRW_MAPFILTER_LINEAR
;
150 sampler
->ss0
.mip_filter
= BRW_MIPFILTER_NONE
;
152 case GL_NEAREST_MIPMAP_NEAREST
:
153 sampler
->ss0
.min_filter
= BRW_MAPFILTER_NEAREST
;
154 sampler
->ss0
.mip_filter
= BRW_MIPFILTER_NEAREST
;
156 case GL_LINEAR_MIPMAP_NEAREST
:
157 sampler
->ss0
.min_filter
= BRW_MAPFILTER_LINEAR
;
158 sampler
->ss0
.mip_filter
= BRW_MIPFILTER_NEAREST
;
160 case GL_NEAREST_MIPMAP_LINEAR
:
161 sampler
->ss0
.min_filter
= BRW_MAPFILTER_NEAREST
;
162 sampler
->ss0
.mip_filter
= BRW_MIPFILTER_LINEAR
;
164 case GL_LINEAR_MIPMAP_LINEAR
:
165 sampler
->ss0
.min_filter
= BRW_MAPFILTER_LINEAR
;
166 sampler
->ss0
.mip_filter
= BRW_MIPFILTER_LINEAR
;
174 if (key
->max_aniso
> 1.0) {
175 sampler
->ss0
.min_filter
= BRW_MAPFILTER_ANISOTROPIC
;
176 sampler
->ss0
.mag_filter
= BRW_MAPFILTER_ANISOTROPIC
;
178 if (key
->max_aniso
> 2.0) {
179 sampler
->ss3
.max_aniso
= MIN2((key
->max_aniso
- 2) / 2,
184 switch (key
->magfilter
) {
186 sampler
->ss0
.mag_filter
= BRW_MAPFILTER_NEAREST
;
189 sampler
->ss0
.mag_filter
= BRW_MAPFILTER_LINEAR
;
196 sampler
->ss1
.r_wrap_mode
= translate_wrap_mode(key
->wrap_r
);
197 sampler
->ss1
.s_wrap_mode
= translate_wrap_mode(key
->wrap_s
);
198 sampler
->ss1
.t_wrap_mode
= translate_wrap_mode(key
->wrap_t
);
200 if (intel
->gen
>= 6 &&
201 sampler
->ss0
.min_filter
!= sampler
->ss0
.mag_filter
)
202 sampler
->ss0
.min_mag_neq
= 1;
204 /* Cube-maps on 965 and later must use the same wrap mode for all 3
205 * coordinate dimensions. Futher, only CUBE and CLAMP are valid.
207 if (key
->tex_target
== GL_TEXTURE_CUBE_MAP
) {
208 if (key
->seamless_cube_map
&&
209 (key
->minfilter
!= GL_NEAREST
|| key
->magfilter
!= GL_NEAREST
)) {
210 sampler
->ss1
.r_wrap_mode
= BRW_TEXCOORDMODE_CUBE
;
211 sampler
->ss1
.s_wrap_mode
= BRW_TEXCOORDMODE_CUBE
;
212 sampler
->ss1
.t_wrap_mode
= BRW_TEXCOORDMODE_CUBE
;
214 sampler
->ss1
.r_wrap_mode
= BRW_TEXCOORDMODE_CLAMP
;
215 sampler
->ss1
.s_wrap_mode
= BRW_TEXCOORDMODE_CLAMP
;
216 sampler
->ss1
.t_wrap_mode
= BRW_TEXCOORDMODE_CLAMP
;
218 } else if (key
->tex_target
== GL_TEXTURE_1D
) {
219 /* There's a bug in 1D texture sampling - it actually pays
220 * attention to the wrap_t value, though it should not.
221 * Override the wrap_t value here to GL_REPEAT to keep
222 * any nonexistent border pixels from floating in.
224 sampler
->ss1
.t_wrap_mode
= BRW_TEXCOORDMODE_WRAP
;
228 /* Set shadow function:
230 if (key
->comparemode
== GL_COMPARE_R_TO_TEXTURE_ARB
) {
231 /* Shadowing is "enabled" by emitting a particular sampler
232 * message (sample_c). So need to recompile WM program when
233 * shadow comparison is enabled on each/any texture unit.
235 sampler
->ss0
.shadow_function
=
236 intel_translate_shadow_compare_func(key
->comparefunc
);
241 sampler
->ss0
.lod_bias
= S_FIXED(CLAMP(key
->lod_bias
, -16, 15), 6);
243 sampler
->ss0
.lod_preclamp
= 1; /* OpenGL mode */
244 sampler
->ss0
.default_color_mode
= 0; /* OpenGL/DX10 mode */
246 /* Set BaseMipLevel, MaxLOD, MinLOD:
248 * XXX: I don't think that using firstLevel, lastLevel works,
249 * because we always setup the surface state as if firstLevel ==
250 * level zero. Probably have to subtract firstLevel from each of
253 sampler
->ss0
.base_level
= U_FIXED(0, 1);
255 sampler
->ss1
.max_lod
= U_FIXED(CLAMP(key
->maxlod
, 0, 13), 6);
256 sampler
->ss1
.min_lod
= U_FIXED(CLAMP(key
->minlod
, 0, 13), 6);
258 sampler
->ss2
.default_color_pointer
= sdc_bo
->offset
>> 5; /* reloc */
262 /** Sets up the cache key for sampler state for all texture units */
264 brw_wm_sampler_populate_key(struct brw_context
*brw
,
265 struct wm_sampler_key
*key
)
267 struct gl_context
*ctx
= &brw
->intel
.ctx
;
269 char *last_entry_end
= ((char*)&key
->sampler_count
) +
270 sizeof(key
->sampler_count
);
272 key
->sampler_count
= 0;
274 for (unit
= 0; unit
< BRW_MAX_TEX_UNIT
; unit
++) {
275 if (ctx
->Texture
.Unit
[unit
]._ReallyEnabled
) {
276 struct wm_sampler_entry
*entry
= &key
->sampler
[unit
];
277 struct gl_texture_unit
*texUnit
= &ctx
->Texture
.Unit
[unit
];
278 struct gl_texture_object
*texObj
= texUnit
->_Current
;
279 struct intel_texture_object
*intelObj
= intel_texture_object(texObj
);
280 struct gl_texture_image
*firstImage
=
281 texObj
->Image
[0][intelObj
->firstLevel
];
283 memset(last_entry_end
, 0,
284 (char*)entry
- last_entry_end
+ sizeof(*entry
));
285 last_entry_end
= ((char*)entry
) + sizeof(*entry
);
287 entry
->tex_target
= texObj
->Target
;
289 entry
->seamless_cube_map
= (texObj
->Target
== GL_TEXTURE_CUBE_MAP
)
290 ? ctx
->Texture
.CubeMapSeamless
: GL_FALSE
;
292 entry
->wrap_r
= texObj
->WrapR
;
293 entry
->wrap_s
= texObj
->WrapS
;
294 entry
->wrap_t
= texObj
->WrapT
;
296 entry
->maxlod
= texObj
->MaxLod
;
297 entry
->minlod
= texObj
->MinLod
;
298 entry
->lod_bias
= texUnit
->LodBias
+ texObj
->LodBias
;
299 entry
->max_aniso
= texObj
->MaxAnisotropy
;
300 entry
->minfilter
= texObj
->MinFilter
;
301 entry
->magfilter
= texObj
->MagFilter
;
302 entry
->comparemode
= texObj
->CompareMode
;
303 entry
->comparefunc
= texObj
->CompareFunc
;
305 drm_intel_bo_unreference(brw
->wm
.sdc_bo
[unit
]);
306 if (firstImage
->_BaseFormat
== GL_DEPTH_COMPONENT
) {
307 float bordercolor
[4] = {
308 texObj
->BorderColor
.f
[0],
309 texObj
->BorderColor
.f
[0],
310 texObj
->BorderColor
.f
[0],
311 texObj
->BorderColor
.f
[0]
313 /* GL specs that border color for depth textures is taken from the
314 * R channel, while the hardware uses A. Spam R into all the
315 * channels for safety.
317 brw
->wm
.sdc_bo
[unit
] = upload_default_color(brw
, bordercolor
);
319 brw
->wm
.sdc_bo
[unit
] = upload_default_color(brw
,
320 texObj
->BorderColor
.f
);
322 key
->sampler_count
= unit
+ 1;
325 struct wm_sampler_entry
*entry
= &key
->sampler
[key
->sampler_count
];
326 memset(last_entry_end
, 0, (char*)entry
- last_entry_end
);
329 /* All samplers must be uploaded in a single contiguous array, which
330 * complicates various things. However, this is still too confusing -
331 * FIXME: simplify all the different new texture state flags.
333 static void upload_wm_samplers( struct brw_context
*brw
)
335 struct gl_context
*ctx
= &brw
->intel
.ctx
;
336 struct wm_sampler_key key
;
337 int i
, sampler_key_size
;
339 brw_wm_sampler_populate_key(brw
, &key
);
341 if (brw
->wm
.sampler_count
!= key
.sampler_count
) {
342 brw
->wm
.sampler_count
= key
.sampler_count
;
343 brw
->state
.dirty
.cache
|= CACHE_NEW_SAMPLER
;
346 drm_intel_bo_unreference(brw
->wm
.sampler_bo
);
347 brw
->wm
.sampler_bo
= NULL
;
348 if (brw
->wm
.sampler_count
== 0)
351 /* Only include the populated portion of the key in the search. */
352 sampler_key_size
= offsetof(struct wm_sampler_key
,
353 sampler
[key
.sampler_count
]);
354 brw
->wm
.sampler_bo
= brw_search_cache(&brw
->cache
, BRW_SAMPLER
,
355 &key
, sampler_key_size
,
356 brw
->wm
.sdc_bo
, key
.sampler_count
,
359 /* If we didnt find it in the cache, compute the state and put it in the
362 if (brw
->wm
.sampler_bo
== NULL
) {
363 struct brw_sampler_state sampler
[BRW_MAX_TEX_UNIT
];
365 memset(sampler
, 0, sizeof(sampler
));
366 for (i
= 0; i
< key
.sampler_count
; i
++) {
367 if (brw
->wm
.sdc_bo
[i
] == NULL
)
370 brw_update_sampler_state(brw
, &key
.sampler
[i
], brw
->wm
.sdc_bo
[i
],
374 brw
->wm
.sampler_bo
= brw_upload_cache(&brw
->cache
, BRW_SAMPLER
,
375 &key
, sampler_key_size
,
376 brw
->wm
.sdc_bo
, key
.sampler_count
,
377 &sampler
, sizeof(sampler
));
379 /* Emit SDC relocations */
380 for (i
= 0; i
< BRW_MAX_TEX_UNIT
; i
++) {
381 if (!ctx
->Texture
.Unit
[i
]._ReallyEnabled
)
384 drm_intel_bo_emit_reloc(brw
->wm
.sampler_bo
,
385 i
* sizeof(struct brw_sampler_state
) +
386 offsetof(struct brw_sampler_state
, ss2
),
387 brw
->wm
.sdc_bo
[i
], 0,
388 I915_GEM_DOMAIN_SAMPLER
, 0);
393 const struct brw_tracked_state brw_wm_samplers
= {
395 .mesa
= _NEW_TEXTURE
,
399 .prepare
= upload_wm_samplers
,