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_sf_clip_viewport
,
195 &gen7_push_constant_space
,
197 &gen6_blend_state
, /* must do before cc unit */
198 &gen6_color_calc_state
, /* must do before cc unit */
199 &gen6_depth_stencil_state
, /* must do before cc unit */
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 */
205 /* Surface state setup. Must come before the VS/WM unit. The binding
206 * table upload must be last.
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
,
226 &gen6_multisample_state
,
228 &gen7_disable_stages
,
242 &brw_polygon_stipple
,
243 &brw_polygon_stipple_offset
,
246 &brw_aa_line_parameters
,
250 &brw_indices
, /* must come before brw_vertices */
257 static const struct brw_tracked_state
*gen8_atoms
[] =
263 /* Command packets: */
264 &gen8_state_base_address
,
267 &gen8_sf_clip_viewport
,
269 &gen7_push_constant_space
,
272 &gen6_color_calc_state
,
274 &gen6_vs_push_constants
, /* Before vs_state */
275 &gen6_gs_push_constants
, /* Before gs_state */
276 &gen6_wm_push_constants
, /* Before wm_surfaces and constant_buffer */
278 /* Surface state setup. Must come before the VS/WM unit. The binding
279 * table upload must be last.
281 &brw_vs_pull_constants
,
282 &brw_vs_ubo_surfaces
,
283 &brw_vs_abo_surfaces
,
284 &brw_gs_pull_constants
,
285 &brw_gs_ubo_surfaces
,
286 &brw_gs_abo_surfaces
,
287 &brw_wm_pull_constants
,
288 &brw_wm_ubo_surfaces
,
289 &brw_wm_abo_surfaces
,
290 &gen6_renderbuffer_surfaces
,
291 &brw_texture_surfaces
,
292 &brw_vs_binding_table
,
293 &brw_gs_binding_table
,
294 &brw_wm_binding_table
,
299 &gen8_multisample_state
,
301 &gen8_disable_stages
,
312 &gen8_wm_depth_stencil
,
319 &brw_polygon_stipple
,
320 &brw_polygon_stipple_offset
,
323 &brw_aa_line_parameters
,
338 brw_upload_initial_gpu_state(struct brw_context
*brw
)
340 /* On platforms with hardware contexts, we can set our initial GPU state
341 * right away rather than doing it via state atoms. This saves a small
342 * amount of overhead on every draw call.
347 brw_upload_invariant_state(brw
);
350 gen8_emit_3dstate_sample_pattern(brw
);
354 void brw_init_state( struct brw_context
*brw
)
356 struct gl_context
*ctx
= &brw
->ctx
;
357 const struct brw_tracked_state
**atoms
;
360 brw_init_caches(brw
);
364 num_atoms
= ARRAY_SIZE(gen8_atoms
);
365 } else if (brw
->gen
== 7) {
367 num_atoms
= ARRAY_SIZE(gen7_atoms
);
368 } else if (brw
->gen
== 6) {
370 num_atoms
= ARRAY_SIZE(gen6_atoms
);
373 num_atoms
= ARRAY_SIZE(gen4_atoms
);
377 brw
->num_atoms
= num_atoms
;
379 while (num_atoms
--) {
380 assert((*atoms
)->dirty
.mesa
|
381 (*atoms
)->dirty
.brw
|
382 (*atoms
)->dirty
.cache
);
383 assert((*atoms
)->emit
);
387 brw_upload_initial_gpu_state(brw
);
389 brw
->state
.dirty
.mesa
= ~0;
390 brw
->state
.dirty
.brw
= ~0ull;
392 /* ~0 is a nonsensical value which won't match anything we program, so
393 * the programming will take effect on the first time around.
395 brw
->pma_stall_bits
= ~0;
397 /* Make sure that brw->state.dirty.brw has enough bits to hold all possible
400 STATIC_ASSERT(BRW_NUM_STATE_BITS
<= 8 * sizeof(brw
->state
.dirty
.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
.NewTextureBuffer
= BRW_NEW_TEXTURE_BUFFER
;
407 ctx
->DriverFlags
.NewAtomicBuffer
= BRW_NEW_ATOMIC_BUFFER
;
411 void brw_destroy_state( struct brw_context
*brw
)
413 brw_destroy_caches(brw
);
416 /***********************************************************************
420 check_state(const struct brw_state_flags
*a
, const struct brw_state_flags
*b
)
422 return ((a
->mesa
& b
->mesa
) |
424 (a
->cache
& b
->cache
)) != 0;
427 static void accumulate_state( struct brw_state_flags
*a
,
428 const struct brw_state_flags
*b
)
432 a
->cache
|= b
->cache
;
436 static void xor_states( struct brw_state_flags
*result
,
437 const struct brw_state_flags
*a
,
438 const struct brw_state_flags
*b
)
440 result
->mesa
= a
->mesa
^ b
->mesa
;
441 result
->brw
= a
->brw
^ b
->brw
;
442 result
->cache
= a
->cache
^ b
->cache
;
445 struct dirty_bit_map
{
451 #define DEFINE_BIT(name) {name, #name, 0}
453 static struct dirty_bit_map mesa_bits
[] = {
454 DEFINE_BIT(_NEW_MODELVIEW
),
455 DEFINE_BIT(_NEW_PROJECTION
),
456 DEFINE_BIT(_NEW_TEXTURE_MATRIX
),
457 DEFINE_BIT(_NEW_COLOR
),
458 DEFINE_BIT(_NEW_DEPTH
),
459 DEFINE_BIT(_NEW_EVAL
),
460 DEFINE_BIT(_NEW_FOG
),
461 DEFINE_BIT(_NEW_HINT
),
462 DEFINE_BIT(_NEW_LIGHT
),
463 DEFINE_BIT(_NEW_LINE
),
464 DEFINE_BIT(_NEW_PIXEL
),
465 DEFINE_BIT(_NEW_POINT
),
466 DEFINE_BIT(_NEW_POLYGON
),
467 DEFINE_BIT(_NEW_POLYGONSTIPPLE
),
468 DEFINE_BIT(_NEW_SCISSOR
),
469 DEFINE_BIT(_NEW_STENCIL
),
470 DEFINE_BIT(_NEW_TEXTURE
),
471 DEFINE_BIT(_NEW_TRANSFORM
),
472 DEFINE_BIT(_NEW_VIEWPORT
),
473 DEFINE_BIT(_NEW_ARRAY
),
474 DEFINE_BIT(_NEW_RENDERMODE
),
475 DEFINE_BIT(_NEW_BUFFERS
),
476 DEFINE_BIT(_NEW_CURRENT_ATTRIB
),
477 DEFINE_BIT(_NEW_MULTISAMPLE
),
478 DEFINE_BIT(_NEW_TRACK_MATRIX
),
479 DEFINE_BIT(_NEW_PROGRAM
),
480 DEFINE_BIT(_NEW_PROGRAM_CONSTANTS
),
481 DEFINE_BIT(_NEW_BUFFER_OBJECT
),
482 DEFINE_BIT(_NEW_FRAG_CLAMP
),
483 /* Avoid sign extension problems. */
484 {(unsigned) _NEW_VARYING_VP_INPUTS
, "_NEW_VARYING_VP_INPUTS", 0},
488 static struct dirty_bit_map brw_bits
[] = {
489 DEFINE_BIT(BRW_NEW_URB_FENCE
),
490 DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM
),
491 DEFINE_BIT(BRW_NEW_GEOMETRY_PROGRAM
),
492 DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM
),
493 DEFINE_BIT(BRW_NEW_CURBE_OFFSETS
),
494 DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE
),
495 DEFINE_BIT(BRW_NEW_PRIMITIVE
),
496 DEFINE_BIT(BRW_NEW_CONTEXT
),
497 DEFINE_BIT(BRW_NEW_PSP
),
498 DEFINE_BIT(BRW_NEW_SURFACES
),
499 DEFINE_BIT(BRW_NEW_VS_BINDING_TABLE
),
500 DEFINE_BIT(BRW_NEW_GS_BINDING_TABLE
),
501 DEFINE_BIT(BRW_NEW_PS_BINDING_TABLE
),
502 DEFINE_BIT(BRW_NEW_INDICES
),
503 DEFINE_BIT(BRW_NEW_VERTICES
),
504 DEFINE_BIT(BRW_NEW_BATCH
),
505 DEFINE_BIT(BRW_NEW_INDEX_BUFFER
),
506 DEFINE_BIT(BRW_NEW_VS_CONSTBUF
),
507 DEFINE_BIT(BRW_NEW_GS_CONSTBUF
),
508 DEFINE_BIT(BRW_NEW_PROGRAM_CACHE
),
509 DEFINE_BIT(BRW_NEW_STATE_BASE_ADDRESS
),
510 DEFINE_BIT(BRW_NEW_VUE_MAP_VS
),
511 DEFINE_BIT(BRW_NEW_VUE_MAP_GEOM_OUT
),
512 DEFINE_BIT(BRW_NEW_TRANSFORM_FEEDBACK
),
513 DEFINE_BIT(BRW_NEW_RASTERIZER_DISCARD
),
514 DEFINE_BIT(BRW_NEW_STATS_WM
),
515 DEFINE_BIT(BRW_NEW_UNIFORM_BUFFER
),
516 DEFINE_BIT(BRW_NEW_ATOMIC_BUFFER
),
517 DEFINE_BIT(BRW_NEW_META_IN_PROGRESS
),
518 DEFINE_BIT(BRW_NEW_INTERPOLATION_MAP
),
519 DEFINE_BIT(BRW_NEW_PUSH_CONSTANT_ALLOCATION
),
520 DEFINE_BIT(BRW_NEW_NUM_SAMPLES
),
521 DEFINE_BIT(BRW_NEW_TEXTURE_BUFFER
),
522 DEFINE_BIT(BRW_NEW_GEN4_UNIT_STATE
),
523 DEFINE_BIT(BRW_NEW_CC_VP
),
524 DEFINE_BIT(BRW_NEW_SF_VP
),
525 DEFINE_BIT(BRW_NEW_CLIP_VP
),
526 DEFINE_BIT(BRW_NEW_SAMPLER_STATE_TABLE
),
530 static struct dirty_bit_map cache_bits
[] = {
531 DEFINE_BIT(CACHE_NEW_WM_PROG
),
532 DEFINE_BIT(CACHE_NEW_BLORP_BLIT_PROG
),
533 DEFINE_BIT(CACHE_NEW_SF_PROG
),
534 DEFINE_BIT(CACHE_NEW_VS_PROG
),
535 DEFINE_BIT(CACHE_NEW_FF_GS_PROG
),
536 DEFINE_BIT(CACHE_NEW_GS_PROG
),
537 DEFINE_BIT(CACHE_NEW_CLIP_PROG
),
543 brw_update_dirty_count(struct dirty_bit_map
*bit_map
, uint64_t bits
)
545 for (int i
= 0; bit_map
[i
].bit
!= 0; i
++) {
546 if (bit_map
[i
].bit
& bits
)
552 brw_print_dirty_count(struct dirty_bit_map
*bit_map
)
554 for (int i
= 0; bit_map
[i
].bit
!= 0; i
++) {
555 fprintf(stderr
, "0x%016lx: %12d (%s)\n",
556 bit_map
[i
].bit
, bit_map
[i
].count
, bit_map
[i
].name
);
560 /***********************************************************************
563 void brw_upload_state(struct brw_context
*brw
)
565 struct gl_context
*ctx
= &brw
->ctx
;
566 struct brw_state_flags
*state
= &brw
->state
.dirty
;
568 static int dirty_count
= 0;
570 state
->mesa
|= brw
->NewGLState
;
573 state
->brw
|= ctx
->NewDriverState
;
574 ctx
->NewDriverState
= 0;
577 /* Always re-emit all state. */
583 if (brw
->fragment_program
!= ctx
->FragmentProgram
._Current
) {
584 brw
->fragment_program
= ctx
->FragmentProgram
._Current
;
585 brw
->state
.dirty
.brw
|= BRW_NEW_FRAGMENT_PROGRAM
;
588 if (brw
->geometry_program
!= ctx
->GeometryProgram
._Current
) {
589 brw
->geometry_program
= ctx
->GeometryProgram
._Current
;
590 brw
->state
.dirty
.brw
|= BRW_NEW_GEOMETRY_PROGRAM
;
593 if (brw
->vertex_program
!= ctx
->VertexProgram
._Current
) {
594 brw
->vertex_program
= ctx
->VertexProgram
._Current
;
595 brw
->state
.dirty
.brw
|= BRW_NEW_VERTEX_PROGRAM
;
598 if (brw
->meta_in_progress
!= _mesa_meta_in_progress(ctx
)) {
599 brw
->meta_in_progress
= _mesa_meta_in_progress(ctx
);
600 brw
->state
.dirty
.brw
|= BRW_NEW_META_IN_PROGRESS
;
603 if (brw
->num_samples
!= ctx
->DrawBuffer
->Visual
.samples
) {
604 brw
->num_samples
= ctx
->DrawBuffer
->Visual
.samples
;
605 brw
->state
.dirty
.brw
|= BRW_NEW_NUM_SAMPLES
;
608 if ((state
->mesa
| state
->cache
| state
->brw
) == 0)
611 if (unlikely(INTEL_DEBUG
)) {
612 /* Debug version which enforces various sanity checks on the
613 * state flags which are generated and checked to help ensure
614 * state atoms are ordered correctly in the list.
616 struct brw_state_flags examined
, prev
;
617 memset(&examined
, 0, sizeof(examined
));
620 for (i
= 0; i
< brw
->num_atoms
; i
++) {
621 const struct brw_tracked_state
*atom
= brw
->atoms
[i
];
622 struct brw_state_flags generated
;
624 if (check_state(state
, &atom
->dirty
)) {
628 accumulate_state(&examined
, &atom
->dirty
);
630 /* generated = (prev ^ state)
631 * if (examined & generated)
634 xor_states(&generated
, &prev
, state
);
635 assert(!check_state(&examined
, &generated
));
640 for (i
= 0; i
< brw
->num_atoms
; i
++) {
641 const struct brw_tracked_state
*atom
= brw
->atoms
[i
];
643 if (check_state(state
, &atom
->dirty
)) {
649 if (unlikely(INTEL_DEBUG
& DEBUG_STATE
)) {
650 STATIC_ASSERT(ARRAY_SIZE(brw_bits
) == BRW_NUM_STATE_BITS
+ 1);
651 STATIC_ASSERT(ARRAY_SIZE(cache_bits
) == BRW_MAX_CACHE
+ 1);
653 brw_update_dirty_count(mesa_bits
, state
->mesa
);
654 brw_update_dirty_count(brw_bits
, state
->brw
);
655 brw_update_dirty_count(cache_bits
, state
->cache
);
656 if (dirty_count
++ % 1000 == 0) {
657 brw_print_dirty_count(mesa_bits
);
658 brw_print_dirty_count(brw_bits
);
659 brw_print_dirty_count(cache_bits
);
660 fprintf(stderr
, "\n");
667 * Clear dirty bits to account for the fact that the state emitted by
668 * brw_upload_state() has been committed to the hardware. This is a separate
669 * call from brw_upload_state() because it's possible that after the call to
670 * brw_upload_state(), we will discover that we've run out of aperture space,
671 * and need to rewind the batch buffer to the state it had before the
672 * brw_upload_state() call.
675 brw_clear_dirty_bits(struct brw_context
*brw
)
677 struct brw_state_flags
*state
= &brw
->state
.dirty
;
678 memset(state
, 0, sizeof(*state
));