i965: Remove caching of surface state objects.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_state_upload.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
34 #include "brw_context.h"
35 #include "brw_state.h"
36 #include "intel_batchbuffer.h"
37 #include "intel_buffers.h"
38 #include "intel_chipset.h"
39
40 /* This is used to initialize brw->state.atoms[]. We could use this
41 * list directly except for a single atom, brw_constant_buffer, which
42 * has a .dirty value which changes according to the parameters of the
43 * current fragment and vertex programs, and so cannot be a static
44 * value.
45 */
46 static const struct brw_tracked_state *gen4_atoms[] =
47 {
48 &brw_check_fallback,
49
50 &brw_wm_input_sizes,
51 &brw_vs_prog,
52 &brw_gs_prog,
53 &brw_clip_prog,
54 &brw_sf_prog,
55 &brw_wm_prog,
56
57 /* Once all the programs are done, we know how large urb entry
58 * sizes need to be and can decide if we need to change the urb
59 * layout.
60 */
61 &brw_curbe_offsets,
62 &brw_recalculate_urb_fence,
63
64 &brw_cc_unit,
65
66 &brw_vs_constants, /* Before vs_surfaces and constant_buffer */
67 &brw_wm_constants, /* Before wm_surfaces and constant_buffer */
68
69 &brw_vs_surfaces, /* must do before unit */
70 &brw_wm_constant_surface, /* must do before wm surfaces/bind bo */
71 &brw_wm_surfaces, /* must do before samplers and unit */
72 &brw_wm_binding_table,
73 &brw_wm_samplers,
74
75 &brw_wm_unit,
76 &brw_sf_vp,
77 &brw_sf_unit,
78 &brw_vs_unit, /* always required, enabled or not */
79 &brw_clip_unit,
80 &brw_gs_unit,
81
82 /* Command packets:
83 */
84 &brw_invarient_state,
85 &brw_state_base_address,
86
87 &brw_binding_table_pointers,
88 &brw_blend_constant_color,
89
90 &brw_depthbuffer,
91
92 &brw_polygon_stipple,
93 &brw_polygon_stipple_offset,
94
95 &brw_line_stipple,
96 &brw_aa_line_parameters,
97
98 &brw_psp_urb_cbs,
99
100 &brw_drawing_rect,
101 &brw_indices,
102 &brw_index_buffer,
103 &brw_vertices,
104
105 &brw_constant_buffer
106 };
107
108 const struct brw_tracked_state *gen6_atoms[] =
109 {
110 &brw_check_fallback,
111
112 &brw_wm_input_sizes,
113 &brw_vs_prog,
114 &brw_gs_prog,
115 &brw_wm_prog,
116
117 &gen6_clip_vp,
118 &gen6_sf_vp,
119 &gen6_cc_vp,
120
121 /* Command packets: */
122 &brw_invarient_state,
123
124 &gen6_viewport_state, /* must do after *_vp stages */
125
126 &gen6_urb,
127 &gen6_blend_state, /* must do before cc unit */
128 &gen6_color_calc_state, /* must do before cc unit */
129 &gen6_depth_stencil_state, /* must do before cc unit */
130 &gen6_cc_state_pointers,
131
132 &brw_vs_surfaces, /* must do before unit */
133 &brw_wm_constant_surface, /* must do before wm surfaces/bind bo */
134 &brw_wm_surfaces, /* must do before samplers and unit */
135
136 &brw_wm_samplers,
137 &gen6_sampler_state,
138
139 &gen6_vs_state,
140 &gen6_gs_state,
141 &gen6_clip_state,
142 &gen6_sf_state,
143 &gen6_wm_state,
144
145 &gen6_scissor_state,
146
147 &brw_state_base_address,
148
149 &gen6_binding_table_pointers,
150
151 &brw_depthbuffer,
152
153 &brw_polygon_stipple,
154 &brw_polygon_stipple_offset,
155
156 &brw_line_stipple,
157 &brw_aa_line_parameters,
158
159 &brw_drawing_rect,
160
161 &brw_indices,
162 &brw_index_buffer,
163 &brw_vertices,
164 };
165
166 void brw_init_state( struct brw_context *brw )
167 {
168 brw_init_caches(brw);
169 }
170
171
172 void brw_destroy_state( struct brw_context *brw )
173 {
174 brw_destroy_caches(brw);
175 brw_destroy_batch_cache(brw);
176 }
177
178 /***********************************************************************
179 */
180
181 static GLboolean check_state( const struct brw_state_flags *a,
182 const struct brw_state_flags *b )
183 {
184 return ((a->mesa & b->mesa) ||
185 (a->brw & b->brw) ||
186 (a->cache & b->cache));
187 }
188
189 static void accumulate_state( struct brw_state_flags *a,
190 const struct brw_state_flags *b )
191 {
192 a->mesa |= b->mesa;
193 a->brw |= b->brw;
194 a->cache |= b->cache;
195 }
196
197
198 static void xor_states( struct brw_state_flags *result,
199 const struct brw_state_flags *a,
200 const struct brw_state_flags *b )
201 {
202 result->mesa = a->mesa ^ b->mesa;
203 result->brw = a->brw ^ b->brw;
204 result->cache = a->cache ^ b->cache;
205 }
206
207 void
208 brw_clear_validated_bos(struct brw_context *brw)
209 {
210 int i;
211
212 /* Clear the last round of validated bos */
213 for (i = 0; i < brw->state.validated_bo_count; i++) {
214 drm_intel_bo_unreference(brw->state.validated_bos[i]);
215 brw->state.validated_bos[i] = NULL;
216 }
217 brw->state.validated_bo_count = 0;
218 }
219
220 struct dirty_bit_map {
221 uint32_t bit;
222 char *name;
223 uint32_t count;
224 };
225
226 #define DEFINE_BIT(name) {name, #name, 0}
227
228 static struct dirty_bit_map mesa_bits[] = {
229 DEFINE_BIT(_NEW_MODELVIEW),
230 DEFINE_BIT(_NEW_PROJECTION),
231 DEFINE_BIT(_NEW_TEXTURE_MATRIX),
232 DEFINE_BIT(_NEW_COLOR_MATRIX),
233 DEFINE_BIT(_NEW_ACCUM),
234 DEFINE_BIT(_NEW_COLOR),
235 DEFINE_BIT(_NEW_DEPTH),
236 DEFINE_BIT(_NEW_EVAL),
237 DEFINE_BIT(_NEW_FOG),
238 DEFINE_BIT(_NEW_HINT),
239 DEFINE_BIT(_NEW_LIGHT),
240 DEFINE_BIT(_NEW_LINE),
241 DEFINE_BIT(_NEW_PIXEL),
242 DEFINE_BIT(_NEW_POINT),
243 DEFINE_BIT(_NEW_POLYGON),
244 DEFINE_BIT(_NEW_POLYGONSTIPPLE),
245 DEFINE_BIT(_NEW_SCISSOR),
246 DEFINE_BIT(_NEW_STENCIL),
247 DEFINE_BIT(_NEW_TEXTURE),
248 DEFINE_BIT(_NEW_TRANSFORM),
249 DEFINE_BIT(_NEW_VIEWPORT),
250 DEFINE_BIT(_NEW_PACKUNPACK),
251 DEFINE_BIT(_NEW_ARRAY),
252 DEFINE_BIT(_NEW_RENDERMODE),
253 DEFINE_BIT(_NEW_BUFFERS),
254 DEFINE_BIT(_NEW_MULTISAMPLE),
255 DEFINE_BIT(_NEW_TRACK_MATRIX),
256 DEFINE_BIT(_NEW_PROGRAM),
257 DEFINE_BIT(_NEW_PROGRAM_CONSTANTS),
258 {0, 0, 0}
259 };
260
261 static struct dirty_bit_map brw_bits[] = {
262 DEFINE_BIT(BRW_NEW_URB_FENCE),
263 DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM),
264 DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM),
265 DEFINE_BIT(BRW_NEW_INPUT_DIMENSIONS),
266 DEFINE_BIT(BRW_NEW_CURBE_OFFSETS),
267 DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE),
268 DEFINE_BIT(BRW_NEW_PRIMITIVE),
269 DEFINE_BIT(BRW_NEW_CONTEXT),
270 DEFINE_BIT(BRW_NEW_WM_INPUT_DIMENSIONS),
271 DEFINE_BIT(BRW_NEW_PSP),
272 DEFINE_BIT(BRW_NEW_WM_SURFACES),
273 DEFINE_BIT(BRW_NEW_BINDING_TABLE),
274 DEFINE_BIT(BRW_NEW_INDICES),
275 DEFINE_BIT(BRW_NEW_INDEX_BUFFER),
276 DEFINE_BIT(BRW_NEW_VERTICES),
277 DEFINE_BIT(BRW_NEW_BATCH),
278 DEFINE_BIT(BRW_NEW_DEPTH_BUFFER),
279 {0, 0, 0}
280 };
281
282 static struct dirty_bit_map cache_bits[] = {
283 DEFINE_BIT(CACHE_NEW_BLEND_STATE),
284 DEFINE_BIT(CACHE_NEW_CC_VP),
285 DEFINE_BIT(CACHE_NEW_CC_UNIT),
286 DEFINE_BIT(CACHE_NEW_WM_PROG),
287 DEFINE_BIT(CACHE_NEW_SAMPLER_DEFAULT_COLOR),
288 DEFINE_BIT(CACHE_NEW_SAMPLER),
289 DEFINE_BIT(CACHE_NEW_WM_UNIT),
290 DEFINE_BIT(CACHE_NEW_SF_PROG),
291 DEFINE_BIT(CACHE_NEW_SF_VP),
292 DEFINE_BIT(CACHE_NEW_SF_UNIT),
293 DEFINE_BIT(CACHE_NEW_VS_UNIT),
294 DEFINE_BIT(CACHE_NEW_VS_PROG),
295 DEFINE_BIT(CACHE_NEW_GS_UNIT),
296 DEFINE_BIT(CACHE_NEW_GS_PROG),
297 DEFINE_BIT(CACHE_NEW_CLIP_VP),
298 DEFINE_BIT(CACHE_NEW_CLIP_UNIT),
299 DEFINE_BIT(CACHE_NEW_CLIP_PROG),
300 {0, 0, 0}
301 };
302
303
304 static void
305 brw_update_dirty_count(struct dirty_bit_map *bit_map, int32_t bits)
306 {
307 int i;
308
309 for (i = 0; i < 32; i++) {
310 if (bit_map[i].bit == 0)
311 return;
312
313 if (bit_map[i].bit & bits)
314 bit_map[i].count++;
315 }
316 }
317
318 static void
319 brw_print_dirty_count(struct dirty_bit_map *bit_map, int32_t bits)
320 {
321 int i;
322
323 for (i = 0; i < 32; i++) {
324 if (bit_map[i].bit == 0)
325 return;
326
327 fprintf(stderr, "0x%08x: %12d (%s)\n",
328 bit_map[i].bit, bit_map[i].count, bit_map[i].name);
329 }
330 }
331
332 /***********************************************************************
333 * Emit all state:
334 */
335 void brw_validate_state( struct brw_context *brw )
336 {
337 GLcontext *ctx = &brw->intel.ctx;
338 struct intel_context *intel = &brw->intel;
339 struct brw_state_flags *state = &brw->state.dirty;
340 GLuint i;
341 const struct brw_tracked_state **atoms;
342 int num_atoms;
343
344 brw_clear_validated_bos(brw);
345
346 state->mesa |= brw->intel.NewGLState;
347 brw->intel.NewGLState = 0;
348
349 brw_add_validated_bo(brw, intel->batch->buf);
350
351 if (IS_GEN6(intel->intelScreen->deviceID)) {
352 atoms = gen6_atoms;
353 num_atoms = ARRAY_SIZE(gen6_atoms);
354 } else {
355 atoms = gen4_atoms;
356 num_atoms = ARRAY_SIZE(gen4_atoms);
357 }
358
359 if (brw->emit_state_always) {
360 state->mesa |= ~0;
361 state->brw |= ~0;
362 state->cache |= ~0;
363 }
364
365 if (brw->fragment_program != ctx->FragmentProgram._Current) {
366 brw->fragment_program = ctx->FragmentProgram._Current;
367 brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM;
368 }
369
370 if (brw->vertex_program != ctx->VertexProgram._Current) {
371 brw->vertex_program = ctx->VertexProgram._Current;
372 brw->state.dirty.brw |= BRW_NEW_VERTEX_PROGRAM;
373 }
374
375 if (state->mesa == 0 &&
376 state->cache == 0 &&
377 state->brw == 0)
378 return;
379
380 if (brw->state.dirty.brw & BRW_NEW_CONTEXT)
381 brw_clear_batch_cache(brw);
382
383 brw->intel.Fallback = GL_FALSE; /* boolean, not bitfield */
384
385 /* do prepare stage for all atoms */
386 for (i = 0; i < num_atoms; i++) {
387 const struct brw_tracked_state *atom = atoms[i];
388
389 if (brw->intel.Fallback)
390 break;
391
392 if (check_state(state, &atom->dirty)) {
393 if (atom->prepare) {
394 atom->prepare(brw);
395 }
396 }
397 }
398
399 intel_check_front_buffer_rendering(intel);
400
401 /* Make sure that the textures which are referenced by the current
402 * brw fragment program are actually present/valid.
403 * If this fails, we can experience GPU lock-ups.
404 */
405 {
406 const struct brw_fragment_program *fp;
407 fp = brw_fragment_program_const(brw->fragment_program);
408 if (fp) {
409 assert((fp->tex_units_used & ctx->Texture._EnabledUnits)
410 == fp->tex_units_used);
411 }
412 }
413 }
414
415
416 void brw_upload_state(struct brw_context *brw)
417 {
418 struct intel_context *intel = &brw->intel;
419 struct brw_state_flags *state = &brw->state.dirty;
420 int i;
421 static int dirty_count = 0;
422 const struct brw_tracked_state **atoms;
423 int num_atoms;
424
425 if (IS_GEN6(intel->intelScreen->deviceID)) {
426 atoms = gen6_atoms;
427 num_atoms = ARRAY_SIZE(gen6_atoms);
428 } else {
429 atoms = gen4_atoms;
430 num_atoms = ARRAY_SIZE(gen4_atoms);
431 }
432
433 brw_clear_validated_bos(brw);
434
435 if (INTEL_DEBUG) {
436 /* Debug version which enforces various sanity checks on the
437 * state flags which are generated and checked to help ensure
438 * state atoms are ordered correctly in the list.
439 */
440 struct brw_state_flags examined, prev;
441 memset(&examined, 0, sizeof(examined));
442 prev = *state;
443
444 for (i = 0; i < num_atoms; i++) {
445 const struct brw_tracked_state *atom = atoms[i];
446 struct brw_state_flags generated;
447
448 assert(atom->dirty.mesa ||
449 atom->dirty.brw ||
450 atom->dirty.cache);
451
452 if (brw->intel.Fallback)
453 break;
454
455 if (check_state(state, &atom->dirty)) {
456 if (atom->emit) {
457 atom->emit( brw );
458 }
459 }
460
461 accumulate_state(&examined, &atom->dirty);
462
463 /* generated = (prev ^ state)
464 * if (examined & generated)
465 * fail;
466 */
467 xor_states(&generated, &prev, state);
468 assert(!check_state(&examined, &generated));
469 prev = *state;
470 }
471 }
472 else {
473 for (i = 0; i < num_atoms; i++) {
474 const struct brw_tracked_state *atom = atoms[i];
475
476 if (brw->intel.Fallback)
477 break;
478
479 if (check_state(state, &atom->dirty)) {
480 if (atom->emit) {
481 atom->emit( brw );
482 }
483 }
484 }
485 }
486
487 if (INTEL_DEBUG & DEBUG_STATE) {
488 brw_update_dirty_count(mesa_bits, state->mesa);
489 brw_update_dirty_count(brw_bits, state->brw);
490 brw_update_dirty_count(cache_bits, state->cache);
491 if (dirty_count++ % 1000 == 0) {
492 brw_print_dirty_count(mesa_bits, state->mesa);
493 brw_print_dirty_count(brw_bits, state->brw);
494 brw_print_dirty_count(cache_bits, state->cache);
495 fprintf(stderr, "\n");
496 }
497 }
498
499 if (!brw->intel.Fallback)
500 memset(state, 0, sizeof(*state));
501 }