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
,
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
;
359 const struct brw_tracked_state
**atoms
;
362 brw_init_caches(brw
);
366 num_atoms
= ARRAY_SIZE(gen8_atoms
);
367 } else if (brw
->gen
== 7) {
369 num_atoms
= ARRAY_SIZE(gen7_atoms
);
370 } else if (brw
->gen
== 6) {
372 num_atoms
= ARRAY_SIZE(gen6_atoms
);
375 num_atoms
= ARRAY_SIZE(gen4_atoms
);
379 brw
->num_atoms
= num_atoms
;
381 while (num_atoms
--) {
382 assert((*atoms
)->dirty
.mesa
|
383 (*atoms
)->dirty
.brw
|
384 (*atoms
)->dirty
.cache
);
385 assert((*atoms
)->emit
);
389 brw_upload_initial_gpu_state(brw
);
391 brw
->state
.dirty
.mesa
= ~0;
392 brw
->state
.dirty
.brw
= ~0ull;
394 /* ~0 is a nonsensical value which won't match anything we program, so
395 * the programming will take effect on the first time around.
397 brw
->pma_stall_bits
= ~0;
399 /* Make sure that brw->state.dirty.brw has enough bits to hold all possible
402 STATIC_ASSERT(BRW_NUM_STATE_BITS
<= 8 * sizeof(brw
->state
.dirty
.brw
));
404 ctx
->DriverFlags
.NewTransformFeedback
= BRW_NEW_TRANSFORM_FEEDBACK
;
405 ctx
->DriverFlags
.NewTransformFeedbackProg
= BRW_NEW_TRANSFORM_FEEDBACK
;
406 ctx
->DriverFlags
.NewRasterizerDiscard
= BRW_NEW_RASTERIZER_DISCARD
;
407 ctx
->DriverFlags
.NewUniformBuffer
= BRW_NEW_UNIFORM_BUFFER
;
408 ctx
->DriverFlags
.NewTextureBuffer
= BRW_NEW_TEXTURE_BUFFER
;
409 ctx
->DriverFlags
.NewAtomicBuffer
= BRW_NEW_ATOMIC_BUFFER
;
413 void brw_destroy_state( struct brw_context
*brw
)
415 brw_destroy_caches(brw
);
418 /***********************************************************************
422 check_state(const struct brw_state_flags
*a
, const struct brw_state_flags
*b
)
424 return ((a
->mesa
& b
->mesa
) |
426 (a
->cache
& b
->cache
)) != 0;
429 static void accumulate_state( struct brw_state_flags
*a
,
430 const struct brw_state_flags
*b
)
434 a
->cache
|= b
->cache
;
438 static void xor_states( struct brw_state_flags
*result
,
439 const struct brw_state_flags
*a
,
440 const struct brw_state_flags
*b
)
442 result
->mesa
= a
->mesa
^ b
->mesa
;
443 result
->brw
= a
->brw
^ b
->brw
;
444 result
->cache
= a
->cache
^ b
->cache
;
447 struct dirty_bit_map
{
453 #define DEFINE_BIT(name) {name, #name, 0}
455 static struct dirty_bit_map mesa_bits
[] = {
456 DEFINE_BIT(_NEW_MODELVIEW
),
457 DEFINE_BIT(_NEW_PROJECTION
),
458 DEFINE_BIT(_NEW_TEXTURE_MATRIX
),
459 DEFINE_BIT(_NEW_COLOR
),
460 DEFINE_BIT(_NEW_DEPTH
),
461 DEFINE_BIT(_NEW_EVAL
),
462 DEFINE_BIT(_NEW_FOG
),
463 DEFINE_BIT(_NEW_HINT
),
464 DEFINE_BIT(_NEW_LIGHT
),
465 DEFINE_BIT(_NEW_LINE
),
466 DEFINE_BIT(_NEW_PIXEL
),
467 DEFINE_BIT(_NEW_POINT
),
468 DEFINE_BIT(_NEW_POLYGON
),
469 DEFINE_BIT(_NEW_POLYGONSTIPPLE
),
470 DEFINE_BIT(_NEW_SCISSOR
),
471 DEFINE_BIT(_NEW_STENCIL
),
472 DEFINE_BIT(_NEW_TEXTURE
),
473 DEFINE_BIT(_NEW_TRANSFORM
),
474 DEFINE_BIT(_NEW_VIEWPORT
),
475 DEFINE_BIT(_NEW_ARRAY
),
476 DEFINE_BIT(_NEW_RENDERMODE
),
477 DEFINE_BIT(_NEW_BUFFERS
),
478 DEFINE_BIT(_NEW_CURRENT_ATTRIB
),
479 DEFINE_BIT(_NEW_MULTISAMPLE
),
480 DEFINE_BIT(_NEW_TRACK_MATRIX
),
481 DEFINE_BIT(_NEW_PROGRAM
),
482 DEFINE_BIT(_NEW_PROGRAM_CONSTANTS
),
483 DEFINE_BIT(_NEW_BUFFER_OBJECT
),
484 DEFINE_BIT(_NEW_FRAG_CLAMP
),
485 /* Avoid sign extension problems. */
486 {(unsigned) _NEW_VARYING_VP_INPUTS
, "_NEW_VARYING_VP_INPUTS", 0},
490 static struct dirty_bit_map brw_bits
[] = {
491 DEFINE_BIT(BRW_NEW_URB_FENCE
),
492 DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM
),
493 DEFINE_BIT(BRW_NEW_GEOMETRY_PROGRAM
),
494 DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM
),
495 DEFINE_BIT(BRW_NEW_CURBE_OFFSETS
),
496 DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE
),
497 DEFINE_BIT(BRW_NEW_PRIMITIVE
),
498 DEFINE_BIT(BRW_NEW_CONTEXT
),
499 DEFINE_BIT(BRW_NEW_PSP
),
500 DEFINE_BIT(BRW_NEW_SURFACES
),
501 DEFINE_BIT(BRW_NEW_VS_BINDING_TABLE
),
502 DEFINE_BIT(BRW_NEW_GS_BINDING_TABLE
),
503 DEFINE_BIT(BRW_NEW_PS_BINDING_TABLE
),
504 DEFINE_BIT(BRW_NEW_INDICES
),
505 DEFINE_BIT(BRW_NEW_VERTICES
),
506 DEFINE_BIT(BRW_NEW_BATCH
),
507 DEFINE_BIT(BRW_NEW_INDEX_BUFFER
),
508 DEFINE_BIT(BRW_NEW_VS_CONSTBUF
),
509 DEFINE_BIT(BRW_NEW_GS_CONSTBUF
),
510 DEFINE_BIT(BRW_NEW_PROGRAM_CACHE
),
511 DEFINE_BIT(BRW_NEW_STATE_BASE_ADDRESS
),
512 DEFINE_BIT(BRW_NEW_VUE_MAP_VS
),
513 DEFINE_BIT(BRW_NEW_VUE_MAP_GEOM_OUT
),
514 DEFINE_BIT(BRW_NEW_TRANSFORM_FEEDBACK
),
515 DEFINE_BIT(BRW_NEW_RASTERIZER_DISCARD
),
516 DEFINE_BIT(BRW_NEW_STATS_WM
),
517 DEFINE_BIT(BRW_NEW_UNIFORM_BUFFER
),
518 DEFINE_BIT(BRW_NEW_ATOMIC_BUFFER
),
519 DEFINE_BIT(BRW_NEW_META_IN_PROGRESS
),
520 DEFINE_BIT(BRW_NEW_INTERPOLATION_MAP
),
521 DEFINE_BIT(BRW_NEW_PUSH_CONSTANT_ALLOCATION
),
522 DEFINE_BIT(BRW_NEW_NUM_SAMPLES
),
523 DEFINE_BIT(BRW_NEW_TEXTURE_BUFFER
),
524 DEFINE_BIT(BRW_NEW_GEN4_UNIT_STATE
),
528 static struct dirty_bit_map cache_bits
[] = {
529 DEFINE_BIT(CACHE_NEW_CC_VP
),
530 DEFINE_BIT(CACHE_NEW_WM_PROG
),
531 DEFINE_BIT(CACHE_NEW_BLORP_BLIT_PROG
),
532 DEFINE_BIT(CACHE_NEW_SAMPLER
),
533 DEFINE_BIT(CACHE_NEW_SF_PROG
),
534 DEFINE_BIT(CACHE_NEW_SF_VP
),
535 DEFINE_BIT(CACHE_NEW_VS_PROG
),
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_PROG
),
545 brw_update_dirty_count(struct dirty_bit_map
*bit_map
, uint64_t bits
)
547 for (int i
= 0; bit_map
[i
].bit
!= 0; i
++) {
548 if (bit_map
[i
].bit
& bits
)
554 brw_print_dirty_count(struct dirty_bit_map
*bit_map
)
556 for (int i
= 0; bit_map
[i
].bit
!= 0; i
++) {
557 fprintf(stderr
, "0x%016lx: %12d (%s)\n",
558 bit_map
[i
].bit
, bit_map
[i
].count
, bit_map
[i
].name
);
562 /***********************************************************************
565 void brw_upload_state(struct brw_context
*brw
)
567 struct gl_context
*ctx
= &brw
->ctx
;
568 struct brw_state_flags
*state
= &brw
->state
.dirty
;
570 static int dirty_count
= 0;
572 state
->mesa
|= brw
->NewGLState
;
575 state
->brw
|= ctx
->NewDriverState
;
576 ctx
->NewDriverState
= 0;
579 /* Always re-emit all state. */
585 if (brw
->fragment_program
!= ctx
->FragmentProgram
._Current
) {
586 brw
->fragment_program
= ctx
->FragmentProgram
._Current
;
587 brw
->state
.dirty
.brw
|= BRW_NEW_FRAGMENT_PROGRAM
;
590 if (brw
->geometry_program
!= ctx
->GeometryProgram
._Current
) {
591 brw
->geometry_program
= ctx
->GeometryProgram
._Current
;
592 brw
->state
.dirty
.brw
|= BRW_NEW_GEOMETRY_PROGRAM
;
595 if (brw
->vertex_program
!= ctx
->VertexProgram
._Current
) {
596 brw
->vertex_program
= ctx
->VertexProgram
._Current
;
597 brw
->state
.dirty
.brw
|= BRW_NEW_VERTEX_PROGRAM
;
600 if (brw
->meta_in_progress
!= _mesa_meta_in_progress(ctx
)) {
601 brw
->meta_in_progress
= _mesa_meta_in_progress(ctx
);
602 brw
->state
.dirty
.brw
|= BRW_NEW_META_IN_PROGRESS
;
605 if (brw
->num_samples
!= ctx
->DrawBuffer
->Visual
.samples
) {
606 brw
->num_samples
= ctx
->DrawBuffer
->Visual
.samples
;
607 brw
->state
.dirty
.brw
|= BRW_NEW_NUM_SAMPLES
;
610 if ((state
->mesa
| state
->cache
| state
->brw
) == 0)
613 if (unlikely(INTEL_DEBUG
)) {
614 /* Debug version which enforces various sanity checks on the
615 * state flags which are generated and checked to help ensure
616 * state atoms are ordered correctly in the list.
618 struct brw_state_flags examined
, prev
;
619 memset(&examined
, 0, sizeof(examined
));
622 for (i
= 0; i
< brw
->num_atoms
; i
++) {
623 const struct brw_tracked_state
*atom
= brw
->atoms
[i
];
624 struct brw_state_flags generated
;
626 if (check_state(state
, &atom
->dirty
)) {
630 accumulate_state(&examined
, &atom
->dirty
);
632 /* generated = (prev ^ state)
633 * if (examined & generated)
636 xor_states(&generated
, &prev
, state
);
637 assert(!check_state(&examined
, &generated
));
642 for (i
= 0; i
< brw
->num_atoms
; i
++) {
643 const struct brw_tracked_state
*atom
= brw
->atoms
[i
];
645 if (check_state(state
, &atom
->dirty
)) {
651 if (unlikely(INTEL_DEBUG
& DEBUG_STATE
)) {
652 STATIC_ASSERT(ARRAY_SIZE(brw_bits
) == BRW_NUM_STATE_BITS
+ 1);
653 STATIC_ASSERT(ARRAY_SIZE(cache_bits
) == BRW_MAX_CACHE
+ 1);
655 brw_update_dirty_count(mesa_bits
, state
->mesa
);
656 brw_update_dirty_count(brw_bits
, state
->brw
);
657 brw_update_dirty_count(cache_bits
, state
->cache
);
658 if (dirty_count
++ % 1000 == 0) {
659 brw_print_dirty_count(mesa_bits
);
660 brw_print_dirty_count(brw_bits
);
661 brw_print_dirty_count(cache_bits
);
662 fprintf(stderr
, "\n");
669 * Clear dirty bits to account for the fact that the state emitted by
670 * brw_upload_state() has been committed to the hardware. This is a separate
671 * call from brw_upload_state() because it's possible that after the call to
672 * brw_upload_state(), we will discover that we've run out of aperture space,
673 * and need to rewind the batch buffer to the state it had before the
674 * brw_upload_state() call.
677 brw_clear_dirty_bits(struct brw_context
*brw
)
679 struct brw_state_flags
*state
= &brw
->state
.dirty
;
680 memset(state
, 0, sizeof(*state
));