i965: Silence unused parameter warnings in intel_mipmap_tree.c
[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 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 <keithw@vmware.com>
30 */
31
32
33
34 #include "brw_context.h"
35 #include "brw_state.h"
36 #include "drivers/common/meta.h"
37 #include "intel_batchbuffer.h"
38 #include "intel_buffers.h"
39 #include "brw_vs.h"
40 #include "brw_ff_gs.h"
41 #include "brw_gs.h"
42 #include "brw_wm.h"
43 #include "brw_cs.h"
44 #include "main/framebuffer.h"
45
46 static const struct brw_tracked_state *gen4_atoms[] =
47 {
48 &brw_interpolation_map,
49
50 &brw_clip_prog, /* must do before state base address */
51 &brw_sf_prog, /* must do before state base address */
52
53 /* Once all the programs are done, we know how large urb entry
54 * sizes need to be and can decide if we need to change the urb
55 * layout.
56 */
57 &brw_curbe_offsets,
58 &brw_recalculate_urb_fence,
59
60 &brw_cc_vp,
61 &brw_cc_unit,
62
63 /* Surface state setup. Must come before the VS/WM unit. The binding
64 * table upload must be last.
65 */
66 &brw_vs_pull_constants,
67 &brw_wm_pull_constants,
68 &brw_renderbuffer_surfaces,
69 &brw_texture_surfaces,
70 &brw_vs_binding_table,
71 &brw_wm_binding_table,
72
73 &brw_fs_samplers,
74 &brw_vs_samplers,
75
76 /* These set up state for brw_psp_urb_cbs */
77 &brw_wm_unit,
78 &brw_sf_vp,
79 &brw_sf_unit,
80 &brw_vs_unit, /* always required, enabled or not */
81 &brw_clip_unit,
82 &brw_gs_unit,
83
84 /* Command packets:
85 */
86 &brw_invariant_state,
87 &brw_state_base_address,
88
89 &brw_binding_table_pointers,
90 &brw_blend_constant_color,
91
92 &brw_depthbuffer,
93
94 &brw_polygon_stipple,
95 &brw_polygon_stipple_offset,
96
97 &brw_line_stipple,
98 &brw_aa_line_parameters,
99
100 &brw_psp_urb_cbs,
101
102 &brw_drawing_rect,
103 &brw_indices, /* must come before brw_vertices */
104 &brw_index_buffer,
105 &brw_vertices,
106
107 &brw_constant_buffer
108 };
109
110 static const struct brw_tracked_state *gen6_atoms[] =
111 {
112 &gen6_clip_vp,
113 &gen6_sf_vp,
114
115 /* Command packets: */
116
117 /* must do before binding table pointers, cc state ptrs */
118 &brw_state_base_address,
119
120 &brw_cc_vp,
121 &gen6_viewport_state, /* must do after *_vp stages */
122
123 &gen6_urb,
124 &gen6_blend_state, /* must do before cc unit */
125 &gen6_color_calc_state, /* must do before cc unit */
126 &gen6_depth_stencil_state, /* must do before cc unit */
127
128 &gen6_vs_push_constants, /* Before vs_state */
129 &gen6_gs_push_constants, /* Before gs_state */
130 &gen6_wm_push_constants, /* Before wm_state */
131
132 /* Surface state setup. Must come before the VS/WM unit. The binding
133 * table upload must be last.
134 */
135 &brw_vs_pull_constants,
136 &brw_vs_ubo_surfaces,
137 &brw_gs_pull_constants,
138 &brw_gs_ubo_surfaces,
139 &brw_wm_pull_constants,
140 &brw_wm_ubo_surfaces,
141 &gen6_renderbuffer_surfaces,
142 &brw_texture_surfaces,
143 &gen6_sol_surface,
144 &brw_vs_binding_table,
145 &gen6_gs_binding_table,
146 &brw_wm_binding_table,
147
148 &brw_fs_samplers,
149 &brw_vs_samplers,
150 &brw_gs_samplers,
151 &gen6_sampler_state,
152 &gen6_multisample_state,
153
154 &gen6_vs_state,
155 &gen6_gs_state,
156 &gen6_clip_state,
157 &gen6_sf_state,
158 &gen6_wm_state,
159
160 &gen6_scissor_state,
161
162 &gen6_binding_table_pointers,
163
164 &brw_depthbuffer,
165
166 &brw_polygon_stipple,
167 &brw_polygon_stipple_offset,
168
169 &brw_line_stipple,
170 &brw_aa_line_parameters,
171
172 &brw_drawing_rect,
173
174 &brw_indices, /* must come before brw_vertices */
175 &brw_index_buffer,
176 &brw_vertices,
177 };
178
179 static const struct brw_tracked_state *gen7_render_atoms[] =
180 {
181 /* Command packets: */
182
183 /* must do before binding table pointers, cc state ptrs */
184 &brw_state_base_address,
185
186 &brw_cc_vp,
187 &gen7_sf_clip_viewport,
188
189 &gen7_push_constant_space,
190 &gen7_urb,
191 &gen6_blend_state, /* must do before cc unit */
192 &gen6_color_calc_state, /* must do before cc unit */
193 &gen6_depth_stencil_state, /* must do before cc unit */
194
195 &gen7_hw_binding_tables, /* Enable hw-generated binding tables for Haswell */
196
197 &brw_vs_image_surfaces, /* Before vs push/pull constants and binding table */
198 &brw_gs_image_surfaces, /* Before gs push/pull constants and binding table */
199 &brw_wm_image_surfaces, /* Before wm push/pull constants and binding table */
200
201 &gen6_vs_push_constants, /* Before vs_state */
202 &gen6_gs_push_constants, /* Before gs_state */
203 &gen6_wm_push_constants, /* Before wm_surfaces and constant_buffer */
204
205 /* Surface state setup. Must come before the VS/WM unit. The binding
206 * table upload must be last.
207 */
208 &brw_vs_pull_constants,
209 &brw_vs_ubo_surfaces,
210 &brw_vs_abo_surfaces,
211 &brw_gs_pull_constants,
212 &brw_gs_ubo_surfaces,
213 &brw_gs_abo_surfaces,
214 &brw_wm_pull_constants,
215 &brw_wm_ubo_surfaces,
216 &brw_wm_abo_surfaces,
217 &gen6_renderbuffer_surfaces,
218 &brw_texture_surfaces,
219 &brw_vs_binding_table,
220 &brw_gs_binding_table,
221 &brw_wm_binding_table,
222
223 &brw_fs_samplers,
224 &brw_vs_samplers,
225 &brw_gs_samplers,
226 &gen6_multisample_state,
227
228 &gen7_disable_stages,
229 &gen7_vs_state,
230 &gen7_gs_state,
231 &gen7_sol_state,
232 &gen7_clip_state,
233 &gen7_sbe_state,
234 &gen7_sf_state,
235 &gen7_wm_state,
236 &gen7_ps_state,
237
238 &gen6_scissor_state,
239
240 &gen7_depthbuffer,
241
242 &brw_polygon_stipple,
243 &brw_polygon_stipple_offset,
244
245 &brw_line_stipple,
246 &brw_aa_line_parameters,
247
248 &brw_drawing_rect,
249
250 &brw_indices, /* must come before brw_vertices */
251 &brw_index_buffer,
252 &brw_vertices,
253
254 &haswell_cut_index,
255 };
256
257 static const struct brw_tracked_state *gen7_compute_atoms[] =
258 {
259 &brw_state_base_address,
260 &brw_cs_image_surfaces,
261 &gen7_cs_push_constants,
262 &brw_cs_abo_surfaces,
263 &brw_texture_surfaces,
264 &brw_cs_state,
265 };
266
267 static const struct brw_tracked_state *gen8_render_atoms[] =
268 {
269 /* Command packets: */
270 &gen8_state_base_address,
271
272 &brw_cc_vp,
273 &gen8_sf_clip_viewport,
274
275 &gen7_push_constant_space,
276 &gen7_urb,
277 &gen8_blend_state,
278 &gen6_color_calc_state,
279
280 &gen7_hw_binding_tables, /* Enable hw-generated binding tables for Broadwell */
281
282 &brw_vs_image_surfaces, /* Before vs push/pull constants and binding table */
283 &brw_gs_image_surfaces, /* Before gs push/pull constants and binding table */
284 &brw_wm_image_surfaces, /* Before wm push/pull constants and binding table */
285
286 &gen6_vs_push_constants, /* Before vs_state */
287 &gen6_gs_push_constants, /* Before gs_state */
288 &gen6_wm_push_constants, /* Before wm_surfaces and constant_buffer */
289
290 /* Surface state setup. Must come before the VS/WM unit. The binding
291 * table upload must be last.
292 */
293 &brw_vs_pull_constants,
294 &brw_vs_ubo_surfaces,
295 &brw_vs_abo_surfaces,
296 &brw_gs_pull_constants,
297 &brw_gs_ubo_surfaces,
298 &brw_gs_abo_surfaces,
299 &brw_wm_pull_constants,
300 &brw_wm_ubo_surfaces,
301 &brw_wm_abo_surfaces,
302 &gen6_renderbuffer_surfaces,
303 &brw_texture_surfaces,
304 &brw_vs_binding_table,
305 &brw_gs_binding_table,
306 &brw_wm_binding_table,
307
308 &brw_fs_samplers,
309 &brw_vs_samplers,
310 &brw_gs_samplers,
311 &gen8_multisample_state,
312
313 &gen8_disable_stages,
314 &gen8_vs_state,
315 &gen8_gs_state,
316 &gen8_sol_state,
317 &gen6_clip_state,
318 &gen8_raster_state,
319 &gen8_sbe_state,
320 &gen8_sf_state,
321 &gen8_ps_blend,
322 &gen8_ps_extra,
323 &gen8_ps_state,
324 &gen8_wm_depth_stencil,
325 &gen8_wm_state,
326
327 &gen6_scissor_state,
328
329 &gen7_depthbuffer,
330
331 &brw_polygon_stipple,
332 &brw_polygon_stipple_offset,
333
334 &brw_line_stipple,
335 &brw_aa_line_parameters,
336
337 &brw_drawing_rect,
338
339 &gen8_vf_topology,
340
341 &brw_indices,
342 &gen8_index_buffer,
343 &gen8_vertices,
344
345 &haswell_cut_index,
346 &gen8_pma_fix,
347 };
348
349 static const struct brw_tracked_state *gen8_compute_atoms[] =
350 {
351 &gen8_state_base_address,
352 &brw_cs_image_surfaces,
353 &gen7_cs_push_constants,
354 &brw_cs_abo_surfaces,
355 &brw_texture_surfaces,
356 &brw_cs_state,
357 };
358
359 static void
360 brw_upload_initial_gpu_state(struct brw_context *brw)
361 {
362 /* On platforms with hardware contexts, we can set our initial GPU state
363 * right away rather than doing it via state atoms. This saves a small
364 * amount of overhead on every draw call.
365 */
366 if (!brw->hw_ctx)
367 return;
368
369 if (brw->gen == 6)
370 brw_emit_post_sync_nonzero_flush(brw);
371
372 brw_upload_invariant_state(brw);
373
374 /* Recommended optimization for Victim Cache eviction in pixel backend. */
375 if (brw->gen >= 9) {
376 BEGIN_BATCH(3);
377 OUT_BATCH(MI_LOAD_REGISTER_IMM | (3 - 2));
378 OUT_BATCH(GEN7_CACHE_MODE_1);
379 OUT_BATCH((GEN9_PARTIAL_RESOLVE_DISABLE_IN_VC << 16) |
380 GEN9_PARTIAL_RESOLVE_DISABLE_IN_VC);
381 ADVANCE_BATCH();
382 }
383
384 if (brw->gen >= 8) {
385 gen8_emit_3dstate_sample_pattern(brw);
386 }
387 }
388
389 static inline const struct brw_tracked_state *
390 brw_get_pipeline_atoms(struct brw_context *brw,
391 enum brw_pipeline pipeline)
392 {
393 switch (pipeline) {
394 case BRW_RENDER_PIPELINE:
395 return brw->render_atoms;
396 case BRW_COMPUTE_PIPELINE:
397 return brw->compute_atoms;
398 default:
399 STATIC_ASSERT(BRW_NUM_PIPELINES == 2);
400 unreachable("Unsupported pipeline");
401 return NULL;
402 }
403 }
404
405 static void
406 brw_copy_pipeline_atoms(struct brw_context *brw,
407 enum brw_pipeline pipeline,
408 const struct brw_tracked_state **atoms,
409 int num_atoms)
410 {
411 /* This is to work around brw_context::atoms being declared const. We want
412 * it to be const, but it needs to be initialized somehow!
413 */
414 struct brw_tracked_state *context_atoms =
415 (struct brw_tracked_state *) brw_get_pipeline_atoms(brw, pipeline);
416
417 for (int i = 0; i < num_atoms; i++) {
418 context_atoms[i] = *atoms[i];
419 assert(context_atoms[i].dirty.mesa | context_atoms[i].dirty.brw);
420 assert(context_atoms[i].emit);
421 }
422
423 brw->num_atoms[pipeline] = num_atoms;
424 }
425
426 void brw_init_state( struct brw_context *brw )
427 {
428 struct gl_context *ctx = &brw->ctx;
429
430 /* Force the first brw_select_pipeline to emit pipeline select */
431 brw->last_pipeline = BRW_NUM_PIPELINES;
432
433 STATIC_ASSERT(ARRAY_SIZE(gen4_atoms) <= ARRAY_SIZE(brw->render_atoms));
434 STATIC_ASSERT(ARRAY_SIZE(gen6_atoms) <= ARRAY_SIZE(brw->render_atoms));
435 STATIC_ASSERT(ARRAY_SIZE(gen7_render_atoms) <=
436 ARRAY_SIZE(brw->render_atoms));
437 STATIC_ASSERT(ARRAY_SIZE(gen8_render_atoms) <=
438 ARRAY_SIZE(brw->render_atoms));
439 STATIC_ASSERT(ARRAY_SIZE(gen7_compute_atoms) <=
440 ARRAY_SIZE(brw->compute_atoms));
441 STATIC_ASSERT(ARRAY_SIZE(gen8_compute_atoms) <=
442 ARRAY_SIZE(brw->compute_atoms));
443
444 brw_init_caches(brw);
445
446 if (brw->gen >= 8) {
447 brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
448 gen8_render_atoms,
449 ARRAY_SIZE(gen8_render_atoms));
450 brw_copy_pipeline_atoms(brw, BRW_COMPUTE_PIPELINE,
451 gen8_compute_atoms,
452 ARRAY_SIZE(gen8_compute_atoms));
453 } else if (brw->gen == 7) {
454 brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
455 gen7_render_atoms,
456 ARRAY_SIZE(gen7_render_atoms));
457 brw_copy_pipeline_atoms(brw, BRW_COMPUTE_PIPELINE,
458 gen7_compute_atoms,
459 ARRAY_SIZE(gen7_compute_atoms));
460 } else if (brw->gen == 6) {
461 brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
462 gen6_atoms, ARRAY_SIZE(gen6_atoms));
463 } else {
464 brw_copy_pipeline_atoms(brw, BRW_RENDER_PIPELINE,
465 gen4_atoms, ARRAY_SIZE(gen4_atoms));
466 }
467
468 brw_upload_initial_gpu_state(brw);
469
470 brw->NewGLState = ~0;
471 brw->ctx.NewDriverState = ~0ull;
472
473 /* ~0 is a nonsensical value which won't match anything we program, so
474 * the programming will take effect on the first time around.
475 */
476 brw->pma_stall_bits = ~0;
477
478 /* Make sure that brw->ctx.NewDriverState has enough bits to hold all possible
479 * dirty flags.
480 */
481 STATIC_ASSERT(BRW_NUM_STATE_BITS <= 8 * sizeof(brw->ctx.NewDriverState));
482
483 ctx->DriverFlags.NewTransformFeedback = BRW_NEW_TRANSFORM_FEEDBACK;
484 ctx->DriverFlags.NewTransformFeedbackProg = BRW_NEW_TRANSFORM_FEEDBACK;
485 ctx->DriverFlags.NewRasterizerDiscard = BRW_NEW_RASTERIZER_DISCARD;
486 ctx->DriverFlags.NewUniformBuffer = BRW_NEW_UNIFORM_BUFFER;
487 ctx->DriverFlags.NewTextureBuffer = BRW_NEW_TEXTURE_BUFFER;
488 ctx->DriverFlags.NewAtomicBuffer = BRW_NEW_ATOMIC_BUFFER;
489 ctx->DriverFlags.NewImageUnits = BRW_NEW_IMAGE_UNITS;
490 }
491
492
493 void brw_destroy_state( struct brw_context *brw )
494 {
495 brw_destroy_caches(brw);
496 }
497
498 /***********************************************************************
499 */
500
501 static bool
502 check_state(const struct brw_state_flags *a, const struct brw_state_flags *b)
503 {
504 return ((a->mesa & b->mesa) | (a->brw & b->brw)) != 0;
505 }
506
507 static void accumulate_state( struct brw_state_flags *a,
508 const struct brw_state_flags *b )
509 {
510 a->mesa |= b->mesa;
511 a->brw |= b->brw;
512 }
513
514
515 static void xor_states( struct brw_state_flags *result,
516 const struct brw_state_flags *a,
517 const struct brw_state_flags *b )
518 {
519 result->mesa = a->mesa ^ b->mesa;
520 result->brw = a->brw ^ b->brw;
521 }
522
523 struct dirty_bit_map {
524 uint64_t bit;
525 char *name;
526 uint32_t count;
527 };
528
529 #define DEFINE_BIT(name) {name, #name, 0}
530
531 static struct dirty_bit_map mesa_bits[] = {
532 DEFINE_BIT(_NEW_MODELVIEW),
533 DEFINE_BIT(_NEW_PROJECTION),
534 DEFINE_BIT(_NEW_TEXTURE_MATRIX),
535 DEFINE_BIT(_NEW_COLOR),
536 DEFINE_BIT(_NEW_DEPTH),
537 DEFINE_BIT(_NEW_EVAL),
538 DEFINE_BIT(_NEW_FOG),
539 DEFINE_BIT(_NEW_HINT),
540 DEFINE_BIT(_NEW_LIGHT),
541 DEFINE_BIT(_NEW_LINE),
542 DEFINE_BIT(_NEW_PIXEL),
543 DEFINE_BIT(_NEW_POINT),
544 DEFINE_BIT(_NEW_POLYGON),
545 DEFINE_BIT(_NEW_POLYGONSTIPPLE),
546 DEFINE_BIT(_NEW_SCISSOR),
547 DEFINE_BIT(_NEW_STENCIL),
548 DEFINE_BIT(_NEW_TEXTURE),
549 DEFINE_BIT(_NEW_TRANSFORM),
550 DEFINE_BIT(_NEW_VIEWPORT),
551 DEFINE_BIT(_NEW_ARRAY),
552 DEFINE_BIT(_NEW_RENDERMODE),
553 DEFINE_BIT(_NEW_BUFFERS),
554 DEFINE_BIT(_NEW_CURRENT_ATTRIB),
555 DEFINE_BIT(_NEW_MULTISAMPLE),
556 DEFINE_BIT(_NEW_TRACK_MATRIX),
557 DEFINE_BIT(_NEW_PROGRAM),
558 DEFINE_BIT(_NEW_PROGRAM_CONSTANTS),
559 DEFINE_BIT(_NEW_BUFFER_OBJECT),
560 DEFINE_BIT(_NEW_FRAG_CLAMP),
561 /* Avoid sign extension problems. */
562 {(unsigned) _NEW_VARYING_VP_INPUTS, "_NEW_VARYING_VP_INPUTS", 0},
563 {0, 0, 0}
564 };
565
566 static struct dirty_bit_map brw_bits[] = {
567 DEFINE_BIT(BRW_NEW_FS_PROG_DATA),
568 DEFINE_BIT(BRW_NEW_BLORP_BLIT_PROG_DATA),
569 DEFINE_BIT(BRW_NEW_SF_PROG_DATA),
570 DEFINE_BIT(BRW_NEW_VS_PROG_DATA),
571 DEFINE_BIT(BRW_NEW_FF_GS_PROG_DATA),
572 DEFINE_BIT(BRW_NEW_GS_PROG_DATA),
573 DEFINE_BIT(BRW_NEW_CLIP_PROG_DATA),
574 DEFINE_BIT(BRW_NEW_CS_PROG_DATA),
575 DEFINE_BIT(BRW_NEW_URB_FENCE),
576 DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM),
577 DEFINE_BIT(BRW_NEW_GEOMETRY_PROGRAM),
578 DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM),
579 DEFINE_BIT(BRW_NEW_CURBE_OFFSETS),
580 DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE),
581 DEFINE_BIT(BRW_NEW_PRIMITIVE),
582 DEFINE_BIT(BRW_NEW_CONTEXT),
583 DEFINE_BIT(BRW_NEW_PSP),
584 DEFINE_BIT(BRW_NEW_SURFACES),
585 DEFINE_BIT(BRW_NEW_VS_BINDING_TABLE),
586 DEFINE_BIT(BRW_NEW_GS_BINDING_TABLE),
587 DEFINE_BIT(BRW_NEW_PS_BINDING_TABLE),
588 DEFINE_BIT(BRW_NEW_INDICES),
589 DEFINE_BIT(BRW_NEW_VERTICES),
590 DEFINE_BIT(BRW_NEW_BATCH),
591 DEFINE_BIT(BRW_NEW_INDEX_BUFFER),
592 DEFINE_BIT(BRW_NEW_VS_CONSTBUF),
593 DEFINE_BIT(BRW_NEW_GS_CONSTBUF),
594 DEFINE_BIT(BRW_NEW_PROGRAM_CACHE),
595 DEFINE_BIT(BRW_NEW_STATE_BASE_ADDRESS),
596 DEFINE_BIT(BRW_NEW_VUE_MAP_VS),
597 DEFINE_BIT(BRW_NEW_VUE_MAP_GEOM_OUT),
598 DEFINE_BIT(BRW_NEW_TRANSFORM_FEEDBACK),
599 DEFINE_BIT(BRW_NEW_RASTERIZER_DISCARD),
600 DEFINE_BIT(BRW_NEW_STATS_WM),
601 DEFINE_BIT(BRW_NEW_UNIFORM_BUFFER),
602 DEFINE_BIT(BRW_NEW_ATOMIC_BUFFER),
603 DEFINE_BIT(BRW_NEW_IMAGE_UNITS),
604 DEFINE_BIT(BRW_NEW_META_IN_PROGRESS),
605 DEFINE_BIT(BRW_NEW_INTERPOLATION_MAP),
606 DEFINE_BIT(BRW_NEW_PUSH_CONSTANT_ALLOCATION),
607 DEFINE_BIT(BRW_NEW_NUM_SAMPLES),
608 DEFINE_BIT(BRW_NEW_TEXTURE_BUFFER),
609 DEFINE_BIT(BRW_NEW_GEN4_UNIT_STATE),
610 DEFINE_BIT(BRW_NEW_CC_VP),
611 DEFINE_BIT(BRW_NEW_SF_VP),
612 DEFINE_BIT(BRW_NEW_CLIP_VP),
613 DEFINE_BIT(BRW_NEW_SAMPLER_STATE_TABLE),
614 DEFINE_BIT(BRW_NEW_VS_ATTRIB_WORKAROUNDS),
615 DEFINE_BIT(BRW_NEW_COMPUTE_PROGRAM),
616 {0, 0, 0}
617 };
618
619 static void
620 brw_update_dirty_count(struct dirty_bit_map *bit_map, uint64_t bits)
621 {
622 for (int i = 0; bit_map[i].bit != 0; i++) {
623 if (bit_map[i].bit & bits)
624 bit_map[i].count++;
625 }
626 }
627
628 static void
629 brw_print_dirty_count(struct dirty_bit_map *bit_map)
630 {
631 for (int i = 0; bit_map[i].bit != 0; i++) {
632 if (bit_map[i].count > 1) {
633 fprintf(stderr, "0x%016lx: %12d (%s)\n",
634 bit_map[i].bit, bit_map[i].count, bit_map[i].name);
635 }
636 }
637 }
638
639 static inline void
640 brw_upload_programs(struct brw_context *brw,
641 enum brw_pipeline pipeline)
642 {
643 if (pipeline == BRW_RENDER_PIPELINE) {
644 brw_upload_vs_prog(brw);
645
646 if (brw->gen < 6)
647 brw_upload_ff_gs_prog(brw);
648 else
649 brw_upload_gs_prog(brw);
650
651 brw_upload_wm_prog(brw);
652 } else if (pipeline == BRW_COMPUTE_PIPELINE) {
653 brw_upload_cs_prog(brw);
654 }
655 }
656
657 static inline void
658 merge_ctx_state(struct brw_context *brw,
659 struct brw_state_flags *state)
660 {
661 state->mesa |= brw->NewGLState;
662 state->brw |= brw->ctx.NewDriverState;
663 }
664
665 static inline void
666 check_and_emit_atom(struct brw_context *brw,
667 struct brw_state_flags *state,
668 const struct brw_tracked_state *atom)
669 {
670 if (check_state(state, &atom->dirty)) {
671 atom->emit(brw);
672 merge_ctx_state(brw, state);
673 }
674 }
675
676 static inline void
677 brw_upload_pipeline_state(struct brw_context *brw,
678 enum brw_pipeline pipeline)
679 {
680 struct gl_context *ctx = &brw->ctx;
681 int i;
682 static int dirty_count = 0;
683 struct brw_state_flags state = brw->state.pipelines[pipeline];
684 unsigned int fb_samples = _mesa_geometric_samples(ctx->DrawBuffer);
685
686 brw_select_pipeline(brw, pipeline);
687
688 if (0) {
689 /* Always re-emit all state. */
690 brw->NewGLState = ~0;
691 ctx->NewDriverState = ~0ull;
692 }
693
694 if (pipeline == BRW_RENDER_PIPELINE) {
695 if (brw->fragment_program != ctx->FragmentProgram._Current) {
696 brw->fragment_program = ctx->FragmentProgram._Current;
697 brw->ctx.NewDriverState |= BRW_NEW_FRAGMENT_PROGRAM;
698 }
699
700 if (brw->geometry_program != ctx->GeometryProgram._Current) {
701 brw->geometry_program = ctx->GeometryProgram._Current;
702 brw->ctx.NewDriverState |= BRW_NEW_GEOMETRY_PROGRAM;
703 }
704
705 if (brw->vertex_program != ctx->VertexProgram._Current) {
706 brw->vertex_program = ctx->VertexProgram._Current;
707 brw->ctx.NewDriverState |= BRW_NEW_VERTEX_PROGRAM;
708 }
709 }
710
711 if (brw->compute_program != ctx->ComputeProgram._Current) {
712 brw->compute_program = ctx->ComputeProgram._Current;
713 brw->ctx.NewDriverState |= BRW_NEW_COMPUTE_PROGRAM;
714 }
715
716 if (brw->meta_in_progress != _mesa_meta_in_progress(ctx)) {
717 brw->meta_in_progress = _mesa_meta_in_progress(ctx);
718 brw->ctx.NewDriverState |= BRW_NEW_META_IN_PROGRESS;
719 }
720
721 if (brw->num_samples != fb_samples) {
722 brw->num_samples = fb_samples;
723 brw->ctx.NewDriverState |= BRW_NEW_NUM_SAMPLES;
724 }
725
726 /* Exit early if no state is flagged as dirty */
727 merge_ctx_state(brw, &state);
728 if ((state.mesa | state.brw) == 0)
729 return;
730
731 /* Emit Sandybridge workaround flushes on every primitive, for safety. */
732 if (brw->gen == 6)
733 brw_emit_post_sync_nonzero_flush(brw);
734
735 brw_upload_programs(brw, pipeline);
736 merge_ctx_state(brw, &state);
737
738 const struct brw_tracked_state *atoms =
739 brw_get_pipeline_atoms(brw, pipeline);
740 const int num_atoms = brw->num_atoms[pipeline];
741
742 if (unlikely(INTEL_DEBUG)) {
743 /* Debug version which enforces various sanity checks on the
744 * state flags which are generated and checked to help ensure
745 * state atoms are ordered correctly in the list.
746 */
747 struct brw_state_flags examined, prev;
748 memset(&examined, 0, sizeof(examined));
749 prev = state;
750
751 for (i = 0; i < num_atoms; i++) {
752 const struct brw_tracked_state *atom = &atoms[i];
753 struct brw_state_flags generated;
754
755 check_and_emit_atom(brw, &state, atom);
756
757 accumulate_state(&examined, &atom->dirty);
758
759 /* generated = (prev ^ state)
760 * if (examined & generated)
761 * fail;
762 */
763 xor_states(&generated, &prev, &state);
764 assert(!check_state(&examined, &generated));
765 prev = state;
766 }
767 }
768 else {
769 for (i = 0; i < num_atoms; i++) {
770 const struct brw_tracked_state *atom = &atoms[i];
771
772 check_and_emit_atom(brw, &state, atom);
773 }
774 }
775
776 if (unlikely(INTEL_DEBUG & DEBUG_STATE)) {
777 STATIC_ASSERT(ARRAY_SIZE(brw_bits) == BRW_NUM_STATE_BITS + 1);
778
779 brw_update_dirty_count(mesa_bits, state.mesa);
780 brw_update_dirty_count(brw_bits, state.brw);
781 if (dirty_count++ % 1000 == 0) {
782 brw_print_dirty_count(mesa_bits);
783 brw_print_dirty_count(brw_bits);
784 fprintf(stderr, "\n");
785 }
786 }
787 }
788
789 /***********************************************************************
790 * Emit all state:
791 */
792 void brw_upload_render_state(struct brw_context *brw)
793 {
794 brw_upload_pipeline_state(brw, BRW_RENDER_PIPELINE);
795 }
796
797 static inline void
798 brw_pipeline_state_finished(struct brw_context *brw,
799 enum brw_pipeline pipeline)
800 {
801 /* Save all dirty state into the other pipelines */
802 for (unsigned i = 0; i < BRW_NUM_PIPELINES; i++) {
803 if (i != pipeline) {
804 brw->state.pipelines[i].mesa |= brw->NewGLState;
805 brw->state.pipelines[i].brw |= brw->ctx.NewDriverState;
806 } else {
807 memset(&brw->state.pipelines[i], 0, sizeof(struct brw_state_flags));
808 }
809 }
810
811 brw->NewGLState = 0;
812 brw->ctx.NewDriverState = 0ull;
813 }
814
815 /**
816 * Clear dirty bits to account for the fact that the state emitted by
817 * brw_upload_render_state() has been committed to the hardware. This is a
818 * separate call from brw_upload_render_state() because it's possible that
819 * after the call to brw_upload_render_state(), we will discover that we've
820 * run out of aperture space, and need to rewind the batch buffer to the state
821 * it had before the brw_upload_render_state() call.
822 */
823 void
824 brw_render_state_finished(struct brw_context *brw)
825 {
826 brw_pipeline_state_finished(brw, BRW_RENDER_PIPELINE);
827 }
828
829 void
830 brw_upload_compute_state(struct brw_context *brw)
831 {
832 brw_upload_pipeline_state(brw, BRW_COMPUTE_PIPELINE);
833 }
834
835 void
836 brw_compute_state_finished(struct brw_context *brw)
837 {
838 brw_pipeline_state_finished(brw, BRW_COMPUTE_PIPELINE);
839 }