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.
348 intel_emit_post_sync_nonzero_flush(brw
);
350 brw_upload_invariant_state(brw
);
353 gen8_emit_3dstate_sample_pattern(brw
);
357 void brw_init_state( struct brw_context
*brw
)
359 struct gl_context
*ctx
= &brw
->ctx
;
360 const struct brw_tracked_state
**atoms
;
363 STATIC_ASSERT(ARRAY_SIZE(gen4_atoms
) <= ARRAY_SIZE(brw
->atoms
));
364 STATIC_ASSERT(ARRAY_SIZE(gen6_atoms
) <= ARRAY_SIZE(brw
->atoms
));
365 STATIC_ASSERT(ARRAY_SIZE(gen7_atoms
) <= ARRAY_SIZE(brw
->atoms
));
366 STATIC_ASSERT(ARRAY_SIZE(gen8_atoms
) <= ARRAY_SIZE(brw
->atoms
));
368 brw_init_caches(brw
);
372 num_atoms
= ARRAY_SIZE(gen8_atoms
);
373 } else if (brw
->gen
== 7) {
375 num_atoms
= ARRAY_SIZE(gen7_atoms
);
376 } else if (brw
->gen
== 6) {
378 num_atoms
= ARRAY_SIZE(gen6_atoms
);
381 num_atoms
= ARRAY_SIZE(gen4_atoms
);
384 brw
->num_atoms
= num_atoms
;
386 /* This is to work around brw_context::atoms being declared const. We want
387 * it to be const, but it needs to be initialized somehow!
389 struct brw_tracked_state
*context_atoms
=
390 (struct brw_tracked_state
*) &brw
->atoms
[0];
392 for (int i
= 0; i
< num_atoms
; i
++)
393 context_atoms
[i
] = *atoms
[i
];
395 while (num_atoms
--) {
396 assert((*atoms
)->dirty
.mesa
| (*atoms
)->dirty
.brw
);
397 assert((*atoms
)->emit
);
401 brw_upload_initial_gpu_state(brw
);
403 brw
->state
.dirty
.mesa
= ~0;
404 brw
->state
.dirty
.brw
= ~0ull;
406 /* ~0 is a nonsensical value which won't match anything we program, so
407 * the programming will take effect on the first time around.
409 brw
->pma_stall_bits
= ~0;
411 /* Make sure that brw->state.dirty.brw has enough bits to hold all possible
414 STATIC_ASSERT(BRW_NUM_STATE_BITS
<= 8 * sizeof(brw
->state
.dirty
.brw
));
416 ctx
->DriverFlags
.NewTransformFeedback
= BRW_NEW_TRANSFORM_FEEDBACK
;
417 ctx
->DriverFlags
.NewTransformFeedbackProg
= BRW_NEW_TRANSFORM_FEEDBACK
;
418 ctx
->DriverFlags
.NewRasterizerDiscard
= BRW_NEW_RASTERIZER_DISCARD
;
419 ctx
->DriverFlags
.NewUniformBuffer
= BRW_NEW_UNIFORM_BUFFER
;
420 ctx
->DriverFlags
.NewTextureBuffer
= BRW_NEW_TEXTURE_BUFFER
;
421 ctx
->DriverFlags
.NewAtomicBuffer
= BRW_NEW_ATOMIC_BUFFER
;
425 void brw_destroy_state( struct brw_context
*brw
)
427 brw_destroy_caches(brw
);
430 /***********************************************************************
434 check_state(const struct brw_state_flags
*a
, const struct brw_state_flags
*b
)
436 return ((a
->mesa
& b
->mesa
) | (a
->brw
& b
->brw
)) != 0;
439 static void accumulate_state( struct brw_state_flags
*a
,
440 const struct brw_state_flags
*b
)
447 static void xor_states( struct brw_state_flags
*result
,
448 const struct brw_state_flags
*a
,
449 const struct brw_state_flags
*b
)
451 result
->mesa
= a
->mesa
^ b
->mesa
;
452 result
->brw
= a
->brw
^ b
->brw
;
455 struct dirty_bit_map
{
461 #define DEFINE_BIT(name) {name, #name, 0}
463 static struct dirty_bit_map mesa_bits
[] = {
464 DEFINE_BIT(_NEW_MODELVIEW
),
465 DEFINE_BIT(_NEW_PROJECTION
),
466 DEFINE_BIT(_NEW_TEXTURE_MATRIX
),
467 DEFINE_BIT(_NEW_COLOR
),
468 DEFINE_BIT(_NEW_DEPTH
),
469 DEFINE_BIT(_NEW_EVAL
),
470 DEFINE_BIT(_NEW_FOG
),
471 DEFINE_BIT(_NEW_HINT
),
472 DEFINE_BIT(_NEW_LIGHT
),
473 DEFINE_BIT(_NEW_LINE
),
474 DEFINE_BIT(_NEW_PIXEL
),
475 DEFINE_BIT(_NEW_POINT
),
476 DEFINE_BIT(_NEW_POLYGON
),
477 DEFINE_BIT(_NEW_POLYGONSTIPPLE
),
478 DEFINE_BIT(_NEW_SCISSOR
),
479 DEFINE_BIT(_NEW_STENCIL
),
480 DEFINE_BIT(_NEW_TEXTURE
),
481 DEFINE_BIT(_NEW_TRANSFORM
),
482 DEFINE_BIT(_NEW_VIEWPORT
),
483 DEFINE_BIT(_NEW_ARRAY
),
484 DEFINE_BIT(_NEW_RENDERMODE
),
485 DEFINE_BIT(_NEW_BUFFERS
),
486 DEFINE_BIT(_NEW_CURRENT_ATTRIB
),
487 DEFINE_BIT(_NEW_MULTISAMPLE
),
488 DEFINE_BIT(_NEW_TRACK_MATRIX
),
489 DEFINE_BIT(_NEW_PROGRAM
),
490 DEFINE_BIT(_NEW_PROGRAM_CONSTANTS
),
491 DEFINE_BIT(_NEW_BUFFER_OBJECT
),
492 DEFINE_BIT(_NEW_FRAG_CLAMP
),
493 /* Avoid sign extension problems. */
494 {(unsigned) _NEW_VARYING_VP_INPUTS
, "_NEW_VARYING_VP_INPUTS", 0},
498 static struct dirty_bit_map brw_bits
[] = {
499 DEFINE_BIT(BRW_NEW_FS_PROG_DATA
),
500 DEFINE_BIT(BRW_NEW_BLORP_BLIT_PROG_DATA
),
501 DEFINE_BIT(BRW_NEW_SF_PROG_DATA
),
502 DEFINE_BIT(BRW_NEW_VS_PROG_DATA
),
503 DEFINE_BIT(BRW_NEW_FF_GS_PROG_DATA
),
504 DEFINE_BIT(BRW_NEW_GS_PROG_DATA
),
505 DEFINE_BIT(BRW_NEW_CLIP_PROG_DATA
),
506 DEFINE_BIT(BRW_NEW_URB_FENCE
),
507 DEFINE_BIT(BRW_NEW_FRAGMENT_PROGRAM
),
508 DEFINE_BIT(BRW_NEW_GEOMETRY_PROGRAM
),
509 DEFINE_BIT(BRW_NEW_VERTEX_PROGRAM
),
510 DEFINE_BIT(BRW_NEW_CURBE_OFFSETS
),
511 DEFINE_BIT(BRW_NEW_REDUCED_PRIMITIVE
),
512 DEFINE_BIT(BRW_NEW_PRIMITIVE
),
513 DEFINE_BIT(BRW_NEW_CONTEXT
),
514 DEFINE_BIT(BRW_NEW_PSP
),
515 DEFINE_BIT(BRW_NEW_SURFACES
),
516 DEFINE_BIT(BRW_NEW_VS_BINDING_TABLE
),
517 DEFINE_BIT(BRW_NEW_GS_BINDING_TABLE
),
518 DEFINE_BIT(BRW_NEW_PS_BINDING_TABLE
),
519 DEFINE_BIT(BRW_NEW_INDICES
),
520 DEFINE_BIT(BRW_NEW_VERTICES
),
521 DEFINE_BIT(BRW_NEW_BATCH
),
522 DEFINE_BIT(BRW_NEW_INDEX_BUFFER
),
523 DEFINE_BIT(BRW_NEW_VS_CONSTBUF
),
524 DEFINE_BIT(BRW_NEW_GS_CONSTBUF
),
525 DEFINE_BIT(BRW_NEW_PROGRAM_CACHE
),
526 DEFINE_BIT(BRW_NEW_STATE_BASE_ADDRESS
),
527 DEFINE_BIT(BRW_NEW_VUE_MAP_VS
),
528 DEFINE_BIT(BRW_NEW_VUE_MAP_GEOM_OUT
),
529 DEFINE_BIT(BRW_NEW_TRANSFORM_FEEDBACK
),
530 DEFINE_BIT(BRW_NEW_RASTERIZER_DISCARD
),
531 DEFINE_BIT(BRW_NEW_STATS_WM
),
532 DEFINE_BIT(BRW_NEW_UNIFORM_BUFFER
),
533 DEFINE_BIT(BRW_NEW_ATOMIC_BUFFER
),
534 DEFINE_BIT(BRW_NEW_META_IN_PROGRESS
),
535 DEFINE_BIT(BRW_NEW_INTERPOLATION_MAP
),
536 DEFINE_BIT(BRW_NEW_PUSH_CONSTANT_ALLOCATION
),
537 DEFINE_BIT(BRW_NEW_NUM_SAMPLES
),
538 DEFINE_BIT(BRW_NEW_TEXTURE_BUFFER
),
539 DEFINE_BIT(BRW_NEW_GEN4_UNIT_STATE
),
540 DEFINE_BIT(BRW_NEW_CC_VP
),
541 DEFINE_BIT(BRW_NEW_SF_VP
),
542 DEFINE_BIT(BRW_NEW_CLIP_VP
),
543 DEFINE_BIT(BRW_NEW_SAMPLER_STATE_TABLE
),
544 DEFINE_BIT(BRW_NEW_VS_ATTRIB_WORKAROUNDS
),
549 brw_update_dirty_count(struct dirty_bit_map
*bit_map
, uint64_t bits
)
551 for (int i
= 0; bit_map
[i
].bit
!= 0; i
++) {
552 if (bit_map
[i
].bit
& bits
)
558 brw_print_dirty_count(struct dirty_bit_map
*bit_map
)
560 for (int i
= 0; bit_map
[i
].bit
!= 0; i
++) {
561 if (bit_map
[i
].count
> 1) {
562 fprintf(stderr
, "0x%016lx: %12d (%s)\n",
563 bit_map
[i
].bit
, bit_map
[i
].count
, bit_map
[i
].name
);
568 /***********************************************************************
571 void brw_upload_state(struct brw_context
*brw
)
573 struct gl_context
*ctx
= &brw
->ctx
;
574 struct brw_state_flags
*state
= &brw
->state
.dirty
;
576 static int dirty_count
= 0;
578 state
->mesa
|= brw
->NewGLState
;
581 state
->brw
|= ctx
->NewDriverState
;
582 ctx
->NewDriverState
= 0;
585 /* Always re-emit all state. */
590 if (brw
->fragment_program
!= ctx
->FragmentProgram
._Current
) {
591 brw
->fragment_program
= ctx
->FragmentProgram
._Current
;
592 brw
->state
.dirty
.brw
|= BRW_NEW_FRAGMENT_PROGRAM
;
595 if (brw
->geometry_program
!= ctx
->GeometryProgram
._Current
) {
596 brw
->geometry_program
= ctx
->GeometryProgram
._Current
;
597 brw
->state
.dirty
.brw
|= BRW_NEW_GEOMETRY_PROGRAM
;
600 if (brw
->vertex_program
!= ctx
->VertexProgram
._Current
) {
601 brw
->vertex_program
= ctx
->VertexProgram
._Current
;
602 brw
->state
.dirty
.brw
|= BRW_NEW_VERTEX_PROGRAM
;
605 if (brw
->meta_in_progress
!= _mesa_meta_in_progress(ctx
)) {
606 brw
->meta_in_progress
= _mesa_meta_in_progress(ctx
);
607 brw
->state
.dirty
.brw
|= BRW_NEW_META_IN_PROGRESS
;
610 if (brw
->num_samples
!= ctx
->DrawBuffer
->Visual
.samples
) {
611 brw
->num_samples
= ctx
->DrawBuffer
->Visual
.samples
;
612 brw
->state
.dirty
.brw
|= BRW_NEW_NUM_SAMPLES
;
615 if ((state
->mesa
| state
->brw
) == 0)
618 /* Emit Sandybridge workaround flushes on every primitive, for safety. */
620 intel_emit_post_sync_nonzero_flush(brw
);
622 if (unlikely(INTEL_DEBUG
)) {
623 /* Debug version which enforces various sanity checks on the
624 * state flags which are generated and checked to help ensure
625 * state atoms are ordered correctly in the list.
627 struct brw_state_flags examined
, prev
;
628 memset(&examined
, 0, sizeof(examined
));
631 for (i
= 0; i
< brw
->num_atoms
; i
++) {
632 const struct brw_tracked_state
*atom
= &brw
->atoms
[i
];
633 struct brw_state_flags generated
;
635 if (check_state(state
, &atom
->dirty
)) {
639 accumulate_state(&examined
, &atom
->dirty
);
641 /* generated = (prev ^ state)
642 * if (examined & generated)
645 xor_states(&generated
, &prev
, state
);
646 assert(!check_state(&examined
, &generated
));
651 for (i
= 0; i
< brw
->num_atoms
; i
++) {
652 const struct brw_tracked_state
*atom
= &brw
->atoms
[i
];
654 if (check_state(state
, &atom
->dirty
)) {
660 if (unlikely(INTEL_DEBUG
& DEBUG_STATE
)) {
661 STATIC_ASSERT(ARRAY_SIZE(brw_bits
) == BRW_NUM_STATE_BITS
+ 1);
663 brw_update_dirty_count(mesa_bits
, state
->mesa
);
664 brw_update_dirty_count(brw_bits
, state
->brw
);
665 if (dirty_count
++ % 1000 == 0) {
666 brw_print_dirty_count(mesa_bits
);
667 brw_print_dirty_count(brw_bits
);
668 fprintf(stderr
, "\n");
675 * Clear dirty bits to account for the fact that the state emitted by
676 * brw_upload_state() has been committed to the hardware. This is a separate
677 * call from brw_upload_state() because it's possible that after the call to
678 * brw_upload_state(), we will discover that we've run out of aperture space,
679 * and need to rewind the batch buffer to the state it had before the
680 * brw_upload_state() call.
683 brw_clear_dirty_bits(struct brw_context
*brw
)
685 struct brw_state_flags
*state
= &brw
->state
.dirty
;
686 memset(state
, 0, sizeof(*state
));