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
,
108 static const struct brw_tracked_state
*gen6_atoms
[] =
110 &brw_vs_prog
, /* must do before state base address */
111 &brw_ff_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_wm_push_constants
, /* Before wm_state */
133 /* Surface state setup. Must come before the VS/WM unit. The binding
134 * table upload must be last.
136 &brw_vs_pull_constants
,
137 &brw_vs_ubo_surfaces
,
138 &brw_wm_pull_constants
,
139 &brw_wm_ubo_surfaces
,
140 &gen6_renderbuffer_surfaces
,
141 &brw_texture_surfaces
,
143 &brw_vs_binding_table
,
144 &gen6_gs_binding_table
,
145 &brw_wm_binding_table
,
150 &gen6_multisample_state
,
160 &gen6_binding_table_pointers
,
164 &brw_polygon_stipple
,
165 &brw_polygon_stipple_offset
,
168 &brw_aa_line_parameters
,
177 static const struct brw_tracked_state
*gen7_atoms
[] =
183 /* Command packets: */
185 /* must do before binding table pointers, cc state ptrs */
186 &brw_state_base_address
,
189 &gen7_cc_viewport_state_pointer
, /* must do after brw_cc_vp */
190 &gen7_sf_clip_viewport
,
192 &gen7_push_constant_space
,
194 &gen6_blend_state
, /* must do before cc unit */
195 &gen6_color_calc_state
, /* must do before cc unit */
196 &gen6_depth_stencil_state
, /* must do before cc unit */
198 &gen6_vs_push_constants
, /* Before vs_state */
199 &gen7_gs_push_constants
, /* Before gs_state */
200 &gen6_wm_push_constants
, /* Before wm_surfaces and constant_buffer */
202 /* Surface state setup. Must come before the VS/WM unit. The binding
203 * table upload must be last.
205 &brw_vs_pull_constants
,
206 &brw_vs_ubo_surfaces
,
207 &brw_vs_abo_surfaces
,
208 &brw_gs_pull_constants
,
209 &brw_gs_ubo_surfaces
,
210 &brw_gs_abo_surfaces
,
211 &brw_wm_pull_constants
,
212 &brw_wm_ubo_surfaces
,
213 &brw_wm_abo_surfaces
,
214 &gen6_renderbuffer_surfaces
,
215 &brw_texture_surfaces
,
216 &brw_vs_binding_table
,
217 &brw_gs_binding_table
,
218 &brw_wm_binding_table
,
223 &gen6_multisample_state
,
225 &gen7_disable_stages
,
239 &brw_polygon_stipple
,
240 &brw_polygon_stipple_offset
,
243 &brw_aa_line_parameters
,
254 static const struct brw_tracked_state
*gen8_atoms
[] =
260 /* Command packets: */
261 &gen8_state_base_address
,
264 &gen7_cc_viewport_state_pointer
, /* must do after brw_cc_vp */
265 &gen8_sf_clip_viewport
,
267 &gen7_push_constant_space
,
270 &gen6_color_calc_state
,
272 &gen6_vs_push_constants
, /* Before vs_state */
273 &gen7_gs_push_constants
, /* Before gs_state */
274 &gen6_wm_push_constants
, /* Before wm_surfaces and constant_buffer */
276 /* Surface state setup. Must come before the VS/WM unit. The binding
277 * table upload must be last.
279 &brw_vs_pull_constants
,
280 &brw_vs_ubo_surfaces
,
281 &brw_vs_abo_surfaces
,
282 &brw_gs_pull_constants
,
283 &brw_gs_ubo_surfaces
,
284 &brw_gs_abo_surfaces
,
285 &brw_wm_pull_constants
,
286 &brw_wm_ubo_surfaces
,
287 &brw_wm_abo_surfaces
,
288 &gen6_renderbuffer_surfaces
,
289 &brw_texture_surfaces
,
290 &brw_vs_binding_table
,
291 &brw_gs_binding_table
,
292 &brw_wm_binding_table
,
297 &gen8_multisample_state
,
299 &gen8_disable_stages
,
310 &gen8_wm_depth_stencil
,
317 &brw_polygon_stipple
,
318 &brw_polygon_stipple_offset
,
321 &brw_aa_line_parameters
,
334 static const struct brw_tracked_state
*gen7_compute_atoms
[] =
340 brw_upload_initial_gpu_state(struct brw_context
*brw
)
342 /* On platforms with hardware contexts, we can set our initial GPU state
343 * right away rather than doing it via state atoms. This saves a small
344 * amount of overhead on every draw call.
349 brw_upload_invariant_state(brw
);
352 gen8_emit_3dstate_sample_pattern(brw
);
356 void brw_init_state( struct brw_context
*brw
)
358 struct gl_context
*ctx
= &brw
->ctx
;
361 brw_init_caches(brw
);
363 memset(brw
->atoms
, 0, sizeof(brw
->atoms
));
364 memset(brw
->num_atoms
, 0, sizeof(brw
->num_atoms
));
367 brw
->atoms
[BRW_PIPELINE_3D
] = gen8_atoms
;
368 brw
->num_atoms
[BRW_PIPELINE_3D
] = ARRAY_SIZE(gen8_atoms
);
369 } else if (brw
->gen
== 7) {
370 brw
->atoms
[BRW_PIPELINE_3D
] = gen7_atoms
;
371 brw
->num_atoms
[BRW_PIPELINE_3D
] = ARRAY_SIZE(gen7_atoms
);
372 brw
->atoms
[BRW_PIPELINE_COMPUTE
] = gen7_compute_atoms
;
373 brw
->num_atoms
[BRW_PIPELINE_COMPUTE
] = ARRAY_SIZE(gen7_compute_atoms
);
374 } else if (brw
->gen
== 6) {
375 brw
->atoms
[BRW_PIPELINE_3D
] = gen6_atoms
;
376 brw
->num_atoms
[BRW_PIPELINE_3D
] = ARRAY_SIZE(gen6_atoms
);
378 brw
->atoms
[BRW_PIPELINE_3D
] = gen4_atoms
;
379 brw
->num_atoms
[BRW_PIPELINE_3D
] = ARRAY_SIZE(gen4_atoms
);
382 for (i
= 0; i
< BRW_NUM_PIPELINES
; i
++) {
383 for (j
= 0; j
< brw
->num_atoms
[i
]; j
++) {
384 assert(brw
->atoms
[i
][j
]->dirty
.mesa
|
385 brw
->atoms
[i
][j
]->dirty
.brw
|
386 brw
->atoms
[i
][j
]->dirty
.cache
);
387 assert(brw
->atoms
[i
][j
]->emit
);
391 brw_upload_initial_gpu_state(brw
);
394 SET_DIRTY64_ALL(brw
);
396 /* Make sure that brw->state.dirty.brw has enough bits to hold all possible
399 STATIC_ASSERT(BRW_NUM_STATE_BITS
<=
400 8 * sizeof(brw
->state
.pipeline_dirty
[0].brw
));
402 ctx
->DriverFlags
.NewTransformFeedback
= BRW_NEW_TRANSFORM_FEEDBACK
;
403 ctx
->DriverFlags
.NewTransformFeedbackProg
= BRW_NEW_TRANSFORM_FEEDBACK
;
404 ctx
->DriverFlags
.NewRasterizerDiscard
= BRW_NEW_RASTERIZER_DISCARD
;
405 ctx
->DriverFlags
.NewUniformBuffer
= BRW_NEW_UNIFORM_BUFFER
;
406 ctx
->DriverFlags
.NewAtomicBuffer
= BRW_NEW_ATOMIC_BUFFER
;
410 void brw_destroy_state( struct brw_context
*brw
)
412 brw_destroy_caches(brw
);
415 /***********************************************************************
419 check_state(const struct brw_state_flags
*a
, const struct brw_state_flags
*b
)
421 return ((a
->mesa
& b
->mesa
) |
423 (a
->cache
& b
->cache
)) != 0;
426 static void accumulate_state( struct brw_state_flags
*a
,
427 const struct brw_state_flags
*b
)
431 a
->cache
|= b
->cache
;
435 static void xor_states( struct brw_state_flags
*result
,
436 const struct brw_state_flags
*a
,
437 const struct brw_state_flags
*b
)
439 result
->mesa
= a
->mesa
^ b
->mesa
;
440 result
->brw
= a
->brw
^ b
->brw
;
441 result
->cache
= a
->cache
^ b
->cache
;
444 struct dirty_bit_map
{
450 #define DEFINE_BIT(name) {name, #name, 0}
452 static struct dirty_bit_map mesa_bits
[] = {
453 DEFINE_BIT(_NEW_MODELVIEW
),
454 DEFINE_BIT(_NEW_PROJECTION
),
455 DEFINE_BIT(_NEW_TEXTURE_MATRIX
),
456 DEFINE_BIT(_NEW_COLOR
),
457 DEFINE_BIT(_NEW_DEPTH
),
458 DEFINE_BIT(_NEW_EVAL
),
459 DEFINE_BIT(_NEW_FOG
),
460 DEFINE_BIT(_NEW_HINT
),
461 DEFINE_BIT(_NEW_LIGHT
),
462 DEFINE_BIT(_NEW_LINE
),
463 DEFINE_BIT(_NEW_PIXEL
),
464 DEFINE_BIT(_NEW_POINT
),
465 DEFINE_BIT(_NEW_POLYGON
),
466 DEFINE_BIT(_NEW_POLYGONSTIPPLE
),
467 DEFINE_BIT(_NEW_SCISSOR
),
468 DEFINE_BIT(_NEW_STENCIL
),
469 DEFINE_BIT(_NEW_TEXTURE
),
470 DEFINE_BIT(_NEW_TRANSFORM
),
471 DEFINE_BIT(_NEW_VIEWPORT
),
472 DEFINE_BIT(_NEW_ARRAY
),
473 DEFINE_BIT(_NEW_RENDERMODE
),
474 DEFINE_BIT(_NEW_BUFFERS
),
475 DEFINE_BIT(_NEW_CURRENT_ATTRIB
),
476 DEFINE_BIT(_NEW_MULTISAMPLE
),
477 DEFINE_BIT(_NEW_TRACK_MATRIX
),
478 DEFINE_BIT(_NEW_PROGRAM
),
479 DEFINE_BIT(_NEW_PROGRAM_CONSTANTS
),
480 DEFINE_BIT(_NEW_BUFFER_OBJECT
),
481 DEFINE_BIT(_NEW_FRAG_CLAMP
),
482 DEFINE_BIT(_NEW_VARYING_VP_INPUTS
),
486 static struct dirty_bit_map brw_bits
[] = {
487 DEFINE_BIT(BRW_NEW_URB_FENCE
),
488 DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM
),
489 DEFINE_BIT(BRW_NEW_GEOMETRY_PROGRAM
),
490 DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM
),
491 DEFINE_BIT(BRW_NEW_CURBE_OFFSETS
),
492 DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE
),
493 DEFINE_BIT(BRW_NEW_PRIMITIVE
),
494 DEFINE_BIT(BRW_NEW_CONTEXT
),
495 DEFINE_BIT(BRW_NEW_PSP
),
496 DEFINE_BIT(BRW_NEW_SURFACES
),
497 DEFINE_BIT(BRW_NEW_VS_BINDING_TABLE
),
498 DEFINE_BIT(BRW_NEW_GS_BINDING_TABLE
),
499 DEFINE_BIT(BRW_NEW_PS_BINDING_TABLE
),
500 DEFINE_BIT(BRW_NEW_INDICES
),
501 DEFINE_BIT(BRW_NEW_VERTICES
),
502 DEFINE_BIT(BRW_NEW_BATCH
),
503 DEFINE_BIT(BRW_NEW_INDEX_BUFFER
),
504 DEFINE_BIT(BRW_NEW_VS_CONSTBUF
),
505 DEFINE_BIT(BRW_NEW_GS_CONSTBUF
),
506 DEFINE_BIT(BRW_NEW_PROGRAM_CACHE
),
507 DEFINE_BIT(BRW_NEW_STATE_BASE_ADDRESS
),
508 DEFINE_BIT(BRW_NEW_VUE_MAP_VS
),
509 DEFINE_BIT(BRW_NEW_VUE_MAP_GEOM_OUT
),
510 DEFINE_BIT(BRW_NEW_TRANSFORM_FEEDBACK
),
511 DEFINE_BIT(BRW_NEW_RASTERIZER_DISCARD
),
512 DEFINE_BIT(BRW_NEW_STATS_WM
),
513 DEFINE_BIT(BRW_NEW_UNIFORM_BUFFER
),
514 DEFINE_BIT(BRW_NEW_ATOMIC_BUFFER
),
515 DEFINE_BIT(BRW_NEW_META_IN_PROGRESS
),
516 DEFINE_BIT(BRW_NEW_INTERPOLATION_MAP
),
517 DEFINE_BIT(BRW_NEW_PUSH_CONSTANT_ALLOCATION
),
518 DEFINE_BIT(BRW_NEW_NUM_SAMPLES
),
522 static struct dirty_bit_map cache_bits
[] = {
523 DEFINE_BIT(CACHE_NEW_CC_VP
),
524 DEFINE_BIT(CACHE_NEW_CC_UNIT
),
525 DEFINE_BIT(CACHE_NEW_WM_PROG
),
526 DEFINE_BIT(CACHE_NEW_BLORP_BLIT_PROG
),
527 DEFINE_BIT(CACHE_NEW_BLORP_CONST_COLOR_PROG
),
528 DEFINE_BIT(CACHE_NEW_SAMPLER
),
529 DEFINE_BIT(CACHE_NEW_WM_UNIT
),
530 DEFINE_BIT(CACHE_NEW_SF_PROG
),
531 DEFINE_BIT(CACHE_NEW_SF_VP
),
532 DEFINE_BIT(CACHE_NEW_SF_UNIT
),
533 DEFINE_BIT(CACHE_NEW_VS_UNIT
),
534 DEFINE_BIT(CACHE_NEW_VS_PROG
),
535 DEFINE_BIT(CACHE_NEW_FF_GS_UNIT
),
536 DEFINE_BIT(CACHE_NEW_FF_GS_PROG
),
537 DEFINE_BIT(CACHE_NEW_GS_PROG
),
538 DEFINE_BIT(CACHE_NEW_CLIP_VP
),
539 DEFINE_BIT(CACHE_NEW_CLIP_UNIT
),
540 DEFINE_BIT(CACHE_NEW_CLIP_PROG
),
546 brw_update_dirty_count(struct dirty_bit_map
*bit_map
, int32_t bits
)
550 for (i
= 0; i
< 32; i
++) {
551 if (bit_map
[i
].bit
== 0)
554 if (bit_map
[i
].bit
& bits
)
560 brw_print_dirty_count(struct dirty_bit_map
*bit_map
)
564 for (i
= 0; i
< 32; i
++) {
565 if (bit_map
[i
].bit
== 0)
568 fprintf(stderr
, "0x%08x: %12d (%s)\n",
569 bit_map
[i
].bit
, bit_map
[i
].count
, bit_map
[i
].name
);
573 /***********************************************************************
576 void brw_upload_state(struct brw_context
*brw
, brw_pipeline pipeline
)
578 struct gl_context
*ctx
= &brw
->ctx
;
579 struct brw_state_flags
*state
= &brw
->state
.pipeline_dirty
[pipeline
];
581 static int dirty_count
= 0;
583 assert(0 <= pipeline
&& pipeline
< BRW_NUM_PIPELINES
);
584 brw
->state
.current_pipeline
= pipeline
;
586 SET_DIRTY_BIT(mesa
, brw
->NewGLState
);
589 SET_DIRTY_BIT(brw
, ctx
->NewDriverState
);
590 ctx
->NewDriverState
= 0;
593 /* Always re-emit all state. */
599 if (brw
->fragment_program
!= ctx
->FragmentProgram
._Current
) {
600 brw
->fragment_program
= ctx
->FragmentProgram
._Current
;
601 SET_DIRTY_BIT(brw
, BRW_NEW_FRAGMENT_PROGRAM
);
604 if (brw
->geometry_program
!= ctx
->GeometryProgram
._Current
) {
605 brw
->geometry_program
= ctx
->GeometryProgram
._Current
;
606 SET_DIRTY_BIT(brw
, BRW_NEW_GEOMETRY_PROGRAM
);
609 if (brw
->vertex_program
!= ctx
->VertexProgram
._Current
) {
610 brw
->vertex_program
= ctx
->VertexProgram
._Current
;
611 SET_DIRTY_BIT(brw
, BRW_NEW_VERTEX_PROGRAM
);
614 if (brw
->meta_in_progress
!= _mesa_meta_in_progress(ctx
)) {
615 brw
->meta_in_progress
= _mesa_meta_in_progress(ctx
);
616 SET_DIRTY_BIT(brw
, BRW_NEW_META_IN_PROGRESS
);
619 if (brw
->num_samples
!= ctx
->DrawBuffer
->Visual
.samples
) {
620 brw
->num_samples
= ctx
->DrawBuffer
->Visual
.samples
;
621 SET_DIRTY_BIT(brw
, BRW_NEW_NUM_SAMPLES
);
624 if ((state
->mesa
| state
->cache
| state
->brw
) == 0)
627 if (unlikely(INTEL_DEBUG
)) {
628 /* Debug version which enforces various sanity checks on the
629 * state flags which are generated and checked to help ensure
630 * state atoms are ordered correctly in the list.
632 struct brw_state_flags examined
, prev
;
633 memset(&examined
, 0, sizeof(examined
));
636 for (i
= 0; i
< brw
->num_atoms
[pipeline
]; i
++) {
637 const struct brw_tracked_state
*atom
= brw
->atoms
[pipeline
][i
];
638 struct brw_state_flags generated
;
640 if (check_state(state
, &atom
->dirty
)) {
644 accumulate_state(&examined
, &atom
->dirty
);
646 /* generated = (prev ^ state)
647 * if (examined & generated)
650 xor_states(&generated
, &prev
, state
);
651 assert(!check_state(&examined
, &generated
));
656 for (i
= 0; i
< brw
->num_atoms
[pipeline
]; i
++) {
657 const struct brw_tracked_state
*atom
= brw
->atoms
[pipeline
][i
];
659 if (check_state(state
, &atom
->dirty
)) {
665 if (unlikely(INTEL_DEBUG
& DEBUG_STATE
)) {
666 STATIC_ASSERT(ARRAY_SIZE(brw_bits
) == BRW_NUM_STATE_BITS
+ 1);
667 STATIC_ASSERT(ARRAY_SIZE(cache_bits
) == BRW_MAX_CACHE
+ 1);
669 brw_update_dirty_count(mesa_bits
, state
->mesa
);
670 brw_update_dirty_count(brw_bits
, state
->brw
);
671 brw_update_dirty_count(cache_bits
, state
->cache
);
672 if (dirty_count
++ % 1000 == 0) {
673 brw_print_dirty_count(mesa_bits
);
674 brw_print_dirty_count(brw_bits
);
675 brw_print_dirty_count(cache_bits
);
676 fprintf(stderr
, "\n");
683 * Clear dirty bits to account for the fact that the state emitted by
684 * brw_upload_state() has been committed to the hardware. This is a separate
685 * call from brw_upload_state() because it's possible that after the call to
686 * brw_upload_state(), we will discover that we've run out of aperture space,
687 * and need to rewind the batch buffer to the state it had before the
688 * brw_upload_state() call.
691 brw_clear_dirty_bits(struct brw_context
*brw
, brw_pipeline pipeline
)
693 struct brw_state_flags
*state
= &brw
->state
.pipeline_dirty
[pipeline
];
694 memset(state
, 0, sizeof(*state
));