2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics 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 <keithw@vmware.com>
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"
40 static const struct brw_tracked_state
*gen4_atoms
[] =
42 &brw_vs_prog
, /* must do before GS prog, state base address. */
43 &brw_ff_gs_prog
, /* must do before state base address */
45 &brw_interpolation_map
,
47 &brw_clip_prog
, /* must do before state base address */
48 &brw_sf_prog
, /* must do before state base address */
49 &brw_wm_prog
, /* must do before state base address */
51 /* Once all the programs are done, we know how large urb entry
52 * sizes need to be and can decide if we need to change the urb
56 &brw_recalculate_urb_fence
,
61 /* Surface state setup. Must come before the VS/WM unit. The binding
62 * table upload must be last.
64 &brw_vs_pull_constants
,
65 &brw_wm_pull_constants
,
66 &brw_renderbuffer_surfaces
,
67 &brw_texture_surfaces
,
68 &brw_vs_binding_table
,
69 &brw_wm_binding_table
,
74 /* These set up state for brw_psp_urb_cbs */
78 &brw_vs_unit
, /* always required, enabled or not */
85 &brw_state_base_address
,
87 &brw_binding_table_pointers
,
88 &brw_blend_constant_color
,
93 &brw_polygon_stipple_offset
,
96 &brw_aa_line_parameters
,
101 &brw_indices
, /* must come before brw_vertices */
108 static const struct brw_tracked_state
*gen6_atoms
[] =
110 &brw_vs_prog
, /* must do before state base address */
111 &brw_gs_prog
, /* must do before state base address */
112 &brw_wm_prog
, /* must do before state base address */
117 /* Command packets: */
119 /* must do before binding table pointers, cc state ptrs */
120 &brw_state_base_address
,
123 &gen6_viewport_state
, /* must do after *_vp stages */
126 &gen6_blend_state
, /* must do before cc unit */
127 &gen6_color_calc_state
, /* must do before cc unit */
128 &gen6_depth_stencil_state
, /* must do before cc unit */
130 &gen6_vs_push_constants
, /* Before vs_state */
131 &gen6_gs_push_constants
, /* Before gs_state */
132 &gen6_wm_push_constants
, /* Before wm_state */
134 /* Surface state setup. Must come before the VS/WM unit. The binding
135 * table upload must be last.
137 &brw_vs_pull_constants
,
138 &brw_vs_ubo_surfaces
,
139 &brw_gs_pull_constants
,
140 &brw_gs_ubo_surfaces
,
141 &brw_wm_pull_constants
,
142 &brw_wm_ubo_surfaces
,
143 &gen6_renderbuffer_surfaces
,
144 &brw_texture_surfaces
,
146 &brw_vs_binding_table
,
147 &gen6_gs_binding_table
,
148 &brw_wm_binding_table
,
154 &gen6_multisample_state
,
164 &gen6_binding_table_pointers
,
168 &brw_polygon_stipple
,
169 &brw_polygon_stipple_offset
,
172 &brw_aa_line_parameters
,
176 &brw_indices
, /* must come before brw_vertices */
181 static const struct brw_tracked_state
*gen7_atoms
[] =
187 /* Command packets: */
189 /* must do before binding table pointers, cc state ptrs */
190 &brw_state_base_address
,
193 &gen7_cc_viewport_state_pointer
, /* must do after brw_cc_vp */
194 &gen7_sf_clip_viewport
,
196 &gen7_push_constant_space
,
198 &gen6_blend_state
, /* must do before cc unit */
199 &gen6_color_calc_state
, /* must do before cc unit */
200 &gen6_depth_stencil_state
, /* must do before cc unit */
202 &gen6_vs_push_constants
, /* Before vs_state */
203 &gen6_gs_push_constants
, /* Before gs_state */
204 &gen6_wm_push_constants
, /* Before wm_surfaces and constant_buffer */
206 /* Surface state setup. Must come before the VS/WM unit. The binding
207 * table upload must be last.
209 &brw_vs_pull_constants
,
210 &brw_vs_ubo_surfaces
,
211 &brw_vs_abo_surfaces
,
212 &brw_gs_pull_constants
,
213 &brw_gs_ubo_surfaces
,
214 &brw_gs_abo_surfaces
,
215 &brw_wm_pull_constants
,
216 &brw_wm_ubo_surfaces
,
217 &brw_wm_abo_surfaces
,
218 &gen6_renderbuffer_surfaces
,
219 &brw_texture_surfaces
,
220 &brw_vs_binding_table
,
221 &brw_gs_binding_table
,
222 &brw_wm_binding_table
,
227 &gen6_multisample_state
,
229 &gen7_disable_stages
,
243 &brw_polygon_stipple
,
244 &brw_polygon_stipple_offset
,
247 &brw_aa_line_parameters
,
251 &brw_indices
, /* must come before brw_vertices */
258 static const struct brw_tracked_state
*gen8_atoms
[] =
264 /* Command packets: */
265 &gen8_state_base_address
,
268 &gen7_cc_viewport_state_pointer
, /* must do after brw_cc_vp */
269 &gen8_sf_clip_viewport
,
271 &gen7_push_constant_space
,
274 &gen6_color_calc_state
,
276 &gen6_vs_push_constants
, /* Before vs_state */
277 &gen6_gs_push_constants
, /* Before gs_state */
278 &gen6_wm_push_constants
, /* Before wm_surfaces and constant_buffer */
280 /* Surface state setup. Must come before the VS/WM unit. The binding
281 * table upload must be last.
283 &brw_vs_pull_constants
,
284 &brw_vs_ubo_surfaces
,
285 &brw_vs_abo_surfaces
,
286 &brw_gs_pull_constants
,
287 &brw_gs_ubo_surfaces
,
288 &brw_gs_abo_surfaces
,
289 &brw_wm_pull_constants
,
290 &brw_wm_ubo_surfaces
,
291 &brw_wm_abo_surfaces
,
292 &gen6_renderbuffer_surfaces
,
293 &brw_texture_surfaces
,
294 &brw_vs_binding_table
,
295 &brw_gs_binding_table
,
296 &brw_wm_binding_table
,
301 &gen8_multisample_state
,
303 &gen8_disable_stages
,
314 &gen8_wm_depth_stencil
,
321 &brw_polygon_stipple
,
322 &brw_polygon_stipple_offset
,
325 &brw_aa_line_parameters
,
339 brw_upload_initial_gpu_state(struct brw_context
*brw
)
341 /* On platforms with hardware contexts, we can set our initial GPU state
342 * right away rather than doing it via state atoms. This saves a small
343 * amount of overhead on every draw call.
348 brw_upload_invariant_state(brw
);
351 gen8_emit_3dstate_sample_pattern(brw
);
355 void brw_init_state( struct brw_context
*brw
)
357 struct gl_context
*ctx
= &brw
->ctx
;
358 const struct brw_tracked_state
**atoms
;
361 brw_init_caches(brw
);
365 num_atoms
= ARRAY_SIZE(gen8_atoms
);
366 } else if (brw
->gen
== 7) {
368 num_atoms
= ARRAY_SIZE(gen7_atoms
);
369 } else if (brw
->gen
== 6) {
371 num_atoms
= ARRAY_SIZE(gen6_atoms
);
374 num_atoms
= ARRAY_SIZE(gen4_atoms
);
378 brw
->num_atoms
= num_atoms
;
380 while (num_atoms
--) {
381 assert((*atoms
)->dirty
.mesa
|
382 (*atoms
)->dirty
.brw
|
383 (*atoms
)->dirty
.cache
);
384 assert((*atoms
)->emit
);
388 brw_upload_initial_gpu_state(brw
);
390 brw
->state
.dirty
.mesa
= ~0;
391 brw
->state
.dirty
.brw
= ~0;
393 /* Make sure that brw->state.dirty.brw has enough bits to hold all possible
396 STATIC_ASSERT(BRW_NUM_STATE_BITS
<= 8 * sizeof(brw
->state
.dirty
.brw
));
398 ctx
->DriverFlags
.NewTransformFeedback
= BRW_NEW_TRANSFORM_FEEDBACK
;
399 ctx
->DriverFlags
.NewTransformFeedbackProg
= BRW_NEW_TRANSFORM_FEEDBACK
;
400 ctx
->DriverFlags
.NewRasterizerDiscard
= BRW_NEW_RASTERIZER_DISCARD
;
401 ctx
->DriverFlags
.NewUniformBuffer
= BRW_NEW_UNIFORM_BUFFER
;
402 ctx
->DriverFlags
.NewAtomicBuffer
= BRW_NEW_ATOMIC_BUFFER
;
406 void brw_destroy_state( struct brw_context
*brw
)
408 brw_destroy_caches(brw
);
411 /***********************************************************************
415 check_state(const struct brw_state_flags
*a
, const struct brw_state_flags
*b
)
417 return ((a
->mesa
& b
->mesa
) |
419 (a
->cache
& b
->cache
)) != 0;
422 static void accumulate_state( struct brw_state_flags
*a
,
423 const struct brw_state_flags
*b
)
427 a
->cache
|= b
->cache
;
431 static void xor_states( struct brw_state_flags
*result
,
432 const struct brw_state_flags
*a
,
433 const struct brw_state_flags
*b
)
435 result
->mesa
= a
->mesa
^ b
->mesa
;
436 result
->brw
= a
->brw
^ b
->brw
;
437 result
->cache
= a
->cache
^ b
->cache
;
440 struct dirty_bit_map
{
446 #define DEFINE_BIT(name) {name, #name, 0}
448 static struct dirty_bit_map mesa_bits
[] = {
449 DEFINE_BIT(_NEW_MODELVIEW
),
450 DEFINE_BIT(_NEW_PROJECTION
),
451 DEFINE_BIT(_NEW_TEXTURE_MATRIX
),
452 DEFINE_BIT(_NEW_COLOR
),
453 DEFINE_BIT(_NEW_DEPTH
),
454 DEFINE_BIT(_NEW_EVAL
),
455 DEFINE_BIT(_NEW_FOG
),
456 DEFINE_BIT(_NEW_HINT
),
457 DEFINE_BIT(_NEW_LIGHT
),
458 DEFINE_BIT(_NEW_LINE
),
459 DEFINE_BIT(_NEW_PIXEL
),
460 DEFINE_BIT(_NEW_POINT
),
461 DEFINE_BIT(_NEW_POLYGON
),
462 DEFINE_BIT(_NEW_POLYGONSTIPPLE
),
463 DEFINE_BIT(_NEW_SCISSOR
),
464 DEFINE_BIT(_NEW_STENCIL
),
465 DEFINE_BIT(_NEW_TEXTURE
),
466 DEFINE_BIT(_NEW_TRANSFORM
),
467 DEFINE_BIT(_NEW_VIEWPORT
),
468 DEFINE_BIT(_NEW_ARRAY
),
469 DEFINE_BIT(_NEW_RENDERMODE
),
470 DEFINE_BIT(_NEW_BUFFERS
),
471 DEFINE_BIT(_NEW_CURRENT_ATTRIB
),
472 DEFINE_BIT(_NEW_MULTISAMPLE
),
473 DEFINE_BIT(_NEW_TRACK_MATRIX
),
474 DEFINE_BIT(_NEW_PROGRAM
),
475 DEFINE_BIT(_NEW_PROGRAM_CONSTANTS
),
476 DEFINE_BIT(_NEW_BUFFER_OBJECT
),
477 DEFINE_BIT(_NEW_FRAG_CLAMP
),
478 DEFINE_BIT(_NEW_VARYING_VP_INPUTS
),
482 static struct dirty_bit_map brw_bits
[] = {
483 DEFINE_BIT(BRW_NEW_URB_FENCE
),
484 DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM
),
485 DEFINE_BIT(BRW_NEW_GEOMETRY_PROGRAM
),
486 DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM
),
487 DEFINE_BIT(BRW_NEW_CURBE_OFFSETS
),
488 DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE
),
489 DEFINE_BIT(BRW_NEW_PRIMITIVE
),
490 DEFINE_BIT(BRW_NEW_CONTEXT
),
491 DEFINE_BIT(BRW_NEW_PSP
),
492 DEFINE_BIT(BRW_NEW_SURFACES
),
493 DEFINE_BIT(BRW_NEW_VS_BINDING_TABLE
),
494 DEFINE_BIT(BRW_NEW_GS_BINDING_TABLE
),
495 DEFINE_BIT(BRW_NEW_PS_BINDING_TABLE
),
496 DEFINE_BIT(BRW_NEW_INDICES
),
497 DEFINE_BIT(BRW_NEW_VERTICES
),
498 DEFINE_BIT(BRW_NEW_BATCH
),
499 DEFINE_BIT(BRW_NEW_INDEX_BUFFER
),
500 DEFINE_BIT(BRW_NEW_VS_CONSTBUF
),
501 DEFINE_BIT(BRW_NEW_GS_CONSTBUF
),
502 DEFINE_BIT(BRW_NEW_PROGRAM_CACHE
),
503 DEFINE_BIT(BRW_NEW_STATE_BASE_ADDRESS
),
504 DEFINE_BIT(BRW_NEW_VUE_MAP_VS
),
505 DEFINE_BIT(BRW_NEW_VUE_MAP_GEOM_OUT
),
506 DEFINE_BIT(BRW_NEW_TRANSFORM_FEEDBACK
),
507 DEFINE_BIT(BRW_NEW_RASTERIZER_DISCARD
),
508 DEFINE_BIT(BRW_NEW_STATS_WM
),
509 DEFINE_BIT(BRW_NEW_UNIFORM_BUFFER
),
510 DEFINE_BIT(BRW_NEW_ATOMIC_BUFFER
),
511 DEFINE_BIT(BRW_NEW_META_IN_PROGRESS
),
512 DEFINE_BIT(BRW_NEW_INTERPOLATION_MAP
),
513 DEFINE_BIT(BRW_NEW_PUSH_CONSTANT_ALLOCATION
),
514 DEFINE_BIT(BRW_NEW_NUM_SAMPLES
),
518 static struct dirty_bit_map cache_bits
[] = {
519 DEFINE_BIT(CACHE_NEW_CC_VP
),
520 DEFINE_BIT(CACHE_NEW_CC_UNIT
),
521 DEFINE_BIT(CACHE_NEW_WM_PROG
),
522 DEFINE_BIT(CACHE_NEW_BLORP_BLIT_PROG
),
523 DEFINE_BIT(CACHE_NEW_BLORP_CONST_COLOR_PROG
),
524 DEFINE_BIT(CACHE_NEW_SAMPLER
),
525 DEFINE_BIT(CACHE_NEW_WM_UNIT
),
526 DEFINE_BIT(CACHE_NEW_SF_PROG
),
527 DEFINE_BIT(CACHE_NEW_SF_VP
),
528 DEFINE_BIT(CACHE_NEW_SF_UNIT
),
529 DEFINE_BIT(CACHE_NEW_VS_UNIT
),
530 DEFINE_BIT(CACHE_NEW_VS_PROG
),
531 DEFINE_BIT(CACHE_NEW_FF_GS_UNIT
),
532 DEFINE_BIT(CACHE_NEW_FF_GS_PROG
),
533 DEFINE_BIT(CACHE_NEW_GS_PROG
),
534 DEFINE_BIT(CACHE_NEW_CLIP_VP
),
535 DEFINE_BIT(CACHE_NEW_CLIP_UNIT
),
536 DEFINE_BIT(CACHE_NEW_CLIP_PROG
),
542 brw_update_dirty_count(struct dirty_bit_map
*bit_map
, int32_t bits
)
546 for (i
= 0; i
< 32; i
++) {
547 if (bit_map
[i
].bit
== 0)
550 if (bit_map
[i
].bit
& bits
)
556 brw_print_dirty_count(struct dirty_bit_map
*bit_map
)
560 for (i
= 0; i
< 32; i
++) {
561 if (bit_map
[i
].bit
== 0)
564 fprintf(stderr
, "0x%08x: %12d (%s)\n",
565 bit_map
[i
].bit
, bit_map
[i
].count
, bit_map
[i
].name
);
569 /***********************************************************************
572 void brw_upload_state(struct brw_context
*brw
)
574 struct gl_context
*ctx
= &brw
->ctx
;
575 struct brw_state_flags
*state
= &brw
->state
.dirty
;
577 static int dirty_count
= 0;
579 state
->mesa
|= brw
->NewGLState
;
582 state
->brw
|= ctx
->NewDriverState
;
583 ctx
->NewDriverState
= 0;
586 /* Always re-emit all state. */
592 if (brw
->fragment_program
!= ctx
->FragmentProgram
._Current
) {
593 brw
->fragment_program
= ctx
->FragmentProgram
._Current
;
594 brw
->state
.dirty
.brw
|= BRW_NEW_FRAGMENT_PROGRAM
;
597 if (brw
->geometry_program
!= ctx
->GeometryProgram
._Current
) {
598 brw
->geometry_program
= ctx
->GeometryProgram
._Current
;
599 brw
->state
.dirty
.brw
|= BRW_NEW_GEOMETRY_PROGRAM
;
602 if (brw
->vertex_program
!= ctx
->VertexProgram
._Current
) {
603 brw
->vertex_program
= ctx
->VertexProgram
._Current
;
604 brw
->state
.dirty
.brw
|= BRW_NEW_VERTEX_PROGRAM
;
607 if (brw
->meta_in_progress
!= _mesa_meta_in_progress(ctx
)) {
608 brw
->meta_in_progress
= _mesa_meta_in_progress(ctx
);
609 brw
->state
.dirty
.brw
|= BRW_NEW_META_IN_PROGRESS
;
612 if (brw
->num_samples
!= ctx
->DrawBuffer
->Visual
.samples
) {
613 brw
->num_samples
= ctx
->DrawBuffer
->Visual
.samples
;
614 brw
->state
.dirty
.brw
|= BRW_NEW_NUM_SAMPLES
;
617 if ((state
->mesa
| state
->cache
| state
->brw
) == 0)
620 if (unlikely(INTEL_DEBUG
)) {
621 /* Debug version which enforces various sanity checks on the
622 * state flags which are generated and checked to help ensure
623 * state atoms are ordered correctly in the list.
625 struct brw_state_flags examined
, prev
;
626 memset(&examined
, 0, sizeof(examined
));
629 for (i
= 0; i
< brw
->num_atoms
; i
++) {
630 const struct brw_tracked_state
*atom
= brw
->atoms
[i
];
631 struct brw_state_flags generated
;
633 if (check_state(state
, &atom
->dirty
)) {
637 accumulate_state(&examined
, &atom
->dirty
);
639 /* generated = (prev ^ state)
640 * if (examined & generated)
643 xor_states(&generated
, &prev
, state
);
644 assert(!check_state(&examined
, &generated
));
649 for (i
= 0; i
< brw
->num_atoms
; i
++) {
650 const struct brw_tracked_state
*atom
= brw
->atoms
[i
];
652 if (check_state(state
, &atom
->dirty
)) {
658 if (unlikely(INTEL_DEBUG
& DEBUG_STATE
)) {
659 STATIC_ASSERT(ARRAY_SIZE(brw_bits
) == BRW_NUM_STATE_BITS
+ 1);
660 STATIC_ASSERT(ARRAY_SIZE(cache_bits
) == BRW_MAX_CACHE
+ 1);
662 brw_update_dirty_count(mesa_bits
, state
->mesa
);
663 brw_update_dirty_count(brw_bits
, state
->brw
);
664 brw_update_dirty_count(cache_bits
, state
->cache
);
665 if (dirty_count
++ % 1000 == 0) {
666 brw_print_dirty_count(mesa_bits
);
667 brw_print_dirty_count(brw_bits
);
668 brw_print_dirty_count(cache_bits
);
669 fprintf(stderr
, "\n");
676 * Clear dirty bits to account for the fact that the state emitted by
677 * brw_upload_state() has been committed to the hardware. This is a separate
678 * call from brw_upload_state() because it's possible that after the call to
679 * brw_upload_state(), we will discover that we've run out of aperture space,
680 * and need to rewind the batch buffer to the state it had before the
681 * brw_upload_state() call.
684 brw_clear_dirty_bits(struct brw_context
*brw
)
686 struct brw_state_flags
*state
= &brw
->state
.dirty
;
687 memset(state
, 0, sizeof(*state
));