Merge branch 'mesa_7_5_branch'
[mesa.git] / src / mesa / drivers / dri / i965 / brw_wm_sampler_state.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
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:
13
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.
17
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.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32
33 #include "brw_context.h"
34 #include "brw_state.h"
35 #include "brw_defines.h"
36
37 #include "main/macros.h"
38
39
40
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.
43 */
44
45
46
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.
50 */
51 static GLuint translate_wrap_mode( GLenum wrap )
52 {
53 switch( wrap ) {
54 case GL_REPEAT:
55 return BRW_TEXCOORDMODE_WRAP;
56 case GL_CLAMP:
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;
64 default:
65 return BRW_TEXCOORDMODE_WRAP;
66 }
67 }
68
69
70 static GLuint U_FIXED(GLfloat value, GLuint frac_bits)
71 {
72 value *= (1<<frac_bits);
73 return value < 0 ? 0 : value;
74 }
75
76 static GLint S_FIXED(GLfloat value, GLuint frac_bits)
77 {
78 return value * (1<<frac_bits);
79 }
80
81
82 static dri_bo *upload_default_color( struct brw_context *brw,
83 const GLfloat *color )
84 {
85 struct brw_sampler_default_color sdc;
86
87 COPY_4V(sdc.color, color);
88
89 return brw_cache_data( &brw->cache, BRW_SAMPLER_DEFAULT_COLOR, &sdc,
90 NULL, 0 );
91 }
92
93
94 struct wm_sampler_key {
95 int sampler_count;
96
97 struct wm_sampler_entry {
98 GLenum tex_target;
99 GLenum wrap_r, wrap_s, wrap_t;
100 float maxlod, minlod;
101 float lod_bias;
102 float max_aniso;
103 GLenum minfilter, magfilter;
104 GLenum comparemode, comparefunc;
105 dri_bo *sdc_bo;
106 } sampler[BRW_MAX_TEX_UNIT];
107 };
108
109 /**
110 * Sets the sampler state for a single unit based off of the sampler key
111 * entry.
112 */
113 static void brw_update_sampler_state(struct wm_sampler_entry *key,
114 dri_bo *sdc_bo,
115 struct brw_sampler_state *sampler)
116 {
117 _mesa_memset(sampler, 0, sizeof(*sampler));
118
119 switch (key->minfilter) {
120 case GL_NEAREST:
121 sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
122 sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
123 break;
124 case GL_LINEAR:
125 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
126 sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
127 break;
128 case GL_NEAREST_MIPMAP_NEAREST:
129 sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
130 sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
131 break;
132 case GL_LINEAR_MIPMAP_NEAREST:
133 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
134 sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
135 break;
136 case GL_NEAREST_MIPMAP_LINEAR:
137 sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
138 sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
139 break;
140 case GL_LINEAR_MIPMAP_LINEAR:
141 sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
142 sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
143 break;
144 default:
145 break;
146 }
147
148 /* Set Anisotropy:
149 */
150 if (key->max_aniso > 1.0) {
151 sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
152 sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
153
154 if (key->max_aniso > 2.0) {
155 sampler->ss3.max_aniso = MIN2((key->max_aniso - 2) / 2,
156 BRW_ANISORATIO_16);
157 }
158 }
159 else {
160 switch (key->magfilter) {
161 case GL_NEAREST:
162 sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
163 break;
164 case GL_LINEAR:
165 sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
166 break;
167 default:
168 break;
169 }
170 }
171
172 if (key->tex_target == GL_TEXTURE_CUBE_MAP &&
173 (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) {
174 /* If we're using anything but nearest sampling for a cube map, we
175 * need to set this wrap mode to avoid GPU lock-ups.
176 */
177 sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
178 sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
179 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
180 }
181 else if (key->tex_target == GL_TEXTURE_1D) {
182 /* There's a bug in 1D texture sampling - it actually pays
183 * attention to the wrap_t value, though it should not.
184 * Override the wrap_t value here to GL_REPEAT to keep
185 * any nonexistent border pixels from floating in.
186 */
187 sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r);
188 sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s);
189 sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP;
190 }
191 else {
192 sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r);
193 sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s);
194 sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t);
195 }
196
197 /* Set shadow function:
198 */
199 if (key->comparemode == GL_COMPARE_R_TO_TEXTURE_ARB) {
200 /* Shadowing is "enabled" by emitting a particular sampler
201 * message (sample_c). So need to recompile WM program when
202 * shadow comparison is enabled on each/any texture unit.
203 */
204 sampler->ss0.shadow_function =
205 intel_translate_shadow_compare_func(key->comparefunc);
206 }
207
208 /* Set LOD bias:
209 */
210 sampler->ss0.lod_bias = S_FIXED(CLAMP(key->lod_bias, -16, 15), 6);
211
212 sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
213 sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
214
215 /* Set BaseMipLevel, MaxLOD, MinLOD:
216 *
217 * XXX: I don't think that using firstLevel, lastLevel works,
218 * because we always setup the surface state as if firstLevel ==
219 * level zero. Probably have to subtract firstLevel from each of
220 * these:
221 */
222 sampler->ss0.base_level = U_FIXED(0, 1);
223
224 sampler->ss1.max_lod = U_FIXED(MIN2(MAX2(key->maxlod, 0), 13), 6);
225 sampler->ss1.min_lod = U_FIXED(MIN2(MAX2(key->minlod, 0), 13), 6);
226
227 sampler->ss2.default_color_pointer = sdc_bo->offset >> 5; /* reloc */
228 }
229
230
231 /** Sets up the cache key for sampler state for all texture units */
232 static void
233 brw_wm_sampler_populate_key(struct brw_context *brw,
234 struct wm_sampler_key *key)
235 {
236 GLcontext *ctx = &brw->intel.ctx;
237 int unit;
238
239 memset(key, 0, sizeof(*key));
240
241 for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
242 if (ctx->Texture.Unit[unit]._ReallyEnabled) {
243 struct wm_sampler_entry *entry = &key->sampler[unit];
244 struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
245 struct gl_texture_object *texObj = texUnit->_Current;
246 struct intel_texture_object *intelObj = intel_texture_object(texObj);
247 struct gl_texture_image *firstImage =
248 texObj->Image[0][intelObj->firstLevel];
249
250 entry->tex_target = texObj->Target;
251
252 entry->wrap_r = texObj->WrapR;
253 entry->wrap_s = texObj->WrapS;
254 entry->wrap_t = texObj->WrapT;
255
256 entry->maxlod = texObj->MaxLod;
257 entry->minlod = texObj->MinLod;
258 entry->lod_bias = texUnit->LodBias + texObj->LodBias;
259 entry->max_aniso = texObj->MaxAnisotropy;
260 entry->minfilter = texObj->MinFilter;
261 entry->magfilter = texObj->MagFilter;
262 entry->comparemode = texObj->CompareMode;
263 entry->comparefunc = texObj->CompareFunc;
264
265 dri_bo_unreference(brw->wm.sdc_bo[unit]);
266 if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
267 float bordercolor[4] = {
268 texObj->BorderColor[0],
269 texObj->BorderColor[0],
270 texObj->BorderColor[0],
271 texObj->BorderColor[0]
272 };
273 /* GL specs that border color for depth textures is taken from the
274 * R channel, while the hardware uses A. Spam R into all the
275 * channels for safety.
276 */
277 brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor);
278 } else {
279 brw->wm.sdc_bo[unit] = upload_default_color(brw,
280 texObj->BorderColor);
281 }
282 key->sampler_count = unit + 1;
283 }
284 }
285 }
286
287 /* All samplers must be uploaded in a single contiguous array, which
288 * complicates various things. However, this is still too confusing -
289 * FIXME: simplify all the different new texture state flags.
290 */
291 static void upload_wm_samplers( struct brw_context *brw )
292 {
293 GLcontext *ctx = &brw->intel.ctx;
294 struct wm_sampler_key key;
295 int i;
296
297 brw_wm_sampler_populate_key(brw, &key);
298
299 if (brw->wm.sampler_count != key.sampler_count) {
300 brw->wm.sampler_count = key.sampler_count;
301 brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
302 }
303
304 dri_bo_unreference(brw->wm.sampler_bo);
305 brw->wm.sampler_bo = NULL;
306 if (brw->wm.sampler_count == 0)
307 return;
308
309 brw->wm.sampler_bo = brw_search_cache(&brw->cache, BRW_SAMPLER,
310 &key, sizeof(key),
311 brw->wm.sdc_bo, key.sampler_count,
312 NULL);
313
314 /* If we didnt find it in the cache, compute the state and put it in the
315 * cache.
316 */
317 if (brw->wm.sampler_bo == NULL) {
318 struct brw_sampler_state sampler[BRW_MAX_TEX_UNIT];
319
320 memset(sampler, 0, sizeof(sampler));
321 for (i = 0; i < key.sampler_count; i++) {
322 if (brw->wm.sdc_bo[i] == NULL)
323 continue;
324
325 brw_update_sampler_state(&key.sampler[i], brw->wm.sdc_bo[i],
326 &sampler[i]);
327 }
328
329 brw->wm.sampler_bo = brw_upload_cache(&brw->cache, BRW_SAMPLER,
330 &key, sizeof(key),
331 brw->wm.sdc_bo, key.sampler_count,
332 &sampler, sizeof(sampler),
333 NULL, NULL);
334
335 /* Emit SDC relocations */
336 for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
337 if (!ctx->Texture.Unit[i]._ReallyEnabled)
338 continue;
339
340 dri_bo_emit_reloc(brw->wm.sampler_bo,
341 I915_GEM_DOMAIN_SAMPLER, 0,
342 0,
343 i * sizeof(struct brw_sampler_state) +
344 offsetof(struct brw_sampler_state, ss2),
345 brw->wm.sdc_bo[i]);
346 }
347 }
348 }
349
350 const struct brw_tracked_state brw_wm_samplers = {
351 .dirty = {
352 .mesa = _NEW_TEXTURE,
353 .brw = 0,
354 .cache = 0
355 },
356 .prepare = upload_wm_samplers,
357 };
358
359