2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
29 #include "util/u_blitter.h"
30 #include "util/u_slab.h"
31 #include "util/u_suballoc.h"
32 #include "util/u_double_list.h"
33 #include "util/u_transfer.h"
34 #include "r600_llvm.h"
35 #include "r600_public.h"
36 #include "r600_resource.h"
38 #define R600_NUM_ATOMS 41
40 /* the number of CS dwords for flushing and drawing */
41 #define R600_MAX_FLUSH_CS_DWORDS 16
42 #define R600_MAX_DRAW_CS_DWORDS 34
43 #define R600_TRACE_CS_DWORDS 7
45 #define R600_MAX_USER_CONST_BUFFERS 13
46 #define R600_MAX_DRIVER_CONST_BUFFERS 3
47 #define R600_MAX_CONST_BUFFERS (R600_MAX_USER_CONST_BUFFERS + R600_MAX_DRIVER_CONST_BUFFERS)
49 /* start driver buffers after user buffers */
50 #define R600_UCP_CONST_BUFFER (R600_MAX_USER_CONST_BUFFERS)
51 #define R600_TXQ_CONST_BUFFER (R600_MAX_USER_CONST_BUFFERS + 1)
52 #define R600_BUFFER_INFO_CONST_BUFFER (R600_MAX_USER_CONST_BUFFERS + 2)
54 #define R600_MAX_CONST_BUFFER_SIZE 4096
56 #ifdef PIPE_ARCH_BIG_ENDIAN
57 #define R600_BIG_ENDIAN 1
59 #define R600_BIG_ENDIAN 0
62 #define R600_MAP_BUFFER_ALIGNMENT 64
64 #define R600_ERR(fmt, args...) \
65 fprintf(stderr, "EE %s:%d %s - "fmt, __FILE__, __LINE__, __func__, ##args)
68 #define R600_CONTEXT_INV_VERTEX_CACHE (1 << 0)
69 #define R600_CONTEXT_INV_TEX_CACHE (1 << 1)
70 #define R600_CONTEXT_INV_CONST_CACHE (1 << 2)
71 /* read-write caches */
72 #define R600_CONTEXT_STREAMOUT_FLUSH (1 << 8)
73 #define R600_CONTEXT_FLUSH_AND_INV (1 << 9)
74 #define R600_CONTEXT_FLUSH_AND_INV_CB_META (1 << 10)
75 #define R600_CONTEXT_FLUSH_AND_INV_DB_META (1 << 11)
76 #define R600_CONTEXT_FLUSH_AND_INV_DB (1 << 12)
77 #define R600_CONTEXT_FLUSH_AND_INV_CB (1 << 13)
78 /* engine synchronization */
79 #define R600_CONTEXT_PS_PARTIAL_FLUSH (1 << 16)
80 #define R600_CONTEXT_WAIT_3D_IDLE (1 << 17)
81 #define R600_CONTEXT_WAIT_CP_DMA_IDLE (1 << 18)
83 #define R600_QUERY_DRAW_CALLS (PIPE_QUERY_DRIVER_SPECIFIC + 0)
84 #define R600_QUERY_REQUESTED_VRAM (PIPE_QUERY_DRIVER_SPECIFIC + 1)
85 #define R600_QUERY_REQUESTED_GTT (PIPE_QUERY_DRIVER_SPECIFIC + 2)
86 #define R600_QUERY_BUFFER_WAIT_TIME (PIPE_QUERY_DRIVER_SPECIFIC + 3)
90 struct r600_shader_key
;
92 /* This encapsulates a state or an operation which can emitted into the GPU
93 * command stream. It's not limited to states only, it can be used for anything
94 * that wants to write commands into the CS (e.g. cache flushes). */
96 void (*emit
)(struct r600_context
*ctx
, struct r600_atom
*state
);
102 /* This is an atom containing GPU commands that never change.
103 * This is supposed to be copied directly into the CS. */
104 struct r600_command_buffer
{
111 struct r600_db_state
{
112 struct r600_atom atom
;
113 struct r600_surface
*rsurf
;
116 struct r600_db_misc_state
{
117 struct r600_atom atom
;
118 bool occlusion_query_enabled
;
119 bool flush_depthstencil_through_cb
;
120 bool flush_depthstencil_in_place
;
121 bool copy_depth
, copy_stencil
;
122 unsigned copy_sample
;
123 unsigned log_samples
;
124 unsigned db_shader_control
;
128 struct r600_cb_misc_state
{
129 struct r600_atom atom
;
130 unsigned cb_color_control
; /* this comes from blend state */
131 unsigned blend_colormask
; /* 8*4 bits for 8 RGBA colorbuffers */
133 unsigned nr_ps_color_outputs
;
138 struct r600_clip_misc_state
{
139 struct r600_atom atom
;
140 unsigned pa_cl_clip_cntl
; /* from rasterizer */
141 unsigned pa_cl_vs_out_cntl
; /* from vertex shader */
142 unsigned clip_plane_enable
; /* from rasterizer */
143 unsigned clip_dist_write
; /* from vertex shader */
146 struct r600_alphatest_state
{
147 struct r600_atom atom
;
148 unsigned sx_alpha_test_control
; /* this comes from dsa state */
149 unsigned sx_alpha_ref
; /* this comes from dsa state */
151 bool cb0_export_16bpc
; /* from set_framebuffer_state */
154 struct r600_vgt_state
{
155 struct r600_atom atom
;
156 uint32_t vgt_multi_prim_ib_reset_en
;
157 uint32_t vgt_multi_prim_ib_reset_indx
;
158 uint32_t vgt_indx_offset
;
161 struct r600_blend_color
{
162 struct r600_atom atom
;
163 struct pipe_blend_color state
;
166 struct r600_clip_state
{
167 struct r600_atom atom
;
168 struct pipe_clip_state state
;
171 struct r600_cs_shader_state
{
172 struct r600_atom atom
;
173 unsigned kernel_index
;
174 struct r600_pipe_compute
*shader
;
177 struct r600_framebuffer
{
178 struct r600_atom atom
;
179 struct pipe_framebuffer_state state
;
180 unsigned compressed_cb_mask
;
184 bool is_msaa_resolve
;
187 struct r600_sample_mask
{
188 struct r600_atom atom
;
189 uint16_t sample_mask
; /* there are only 8 bits on EG, 16 bits on Cayman */
192 struct r600_config_state
{
193 struct r600_atom atom
;
194 unsigned sq_gpr_resource_mgmt_1
;
197 struct r600_stencil_ref
204 struct r600_stencil_ref_state
{
205 struct r600_atom atom
;
206 struct r600_stencil_ref state
;
207 struct pipe_stencil_ref pipe_state
;
210 struct r600_viewport_state
{
211 struct r600_atom atom
;
212 struct pipe_viewport_state state
;
215 struct r600_pipe_fences
{
216 struct r600_resource
*bo
;
219 /* linked list of preallocated blocks */
220 struct list_head blocks
;
221 /* linked list of freed fences */
222 struct list_head pool
;
226 typedef boolean (*r600g_dma_blit_t
)(struct pipe_context
*ctx
,
227 struct pipe_resource
*dst
,
229 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
230 struct pipe_resource
*src
,
232 const struct pipe_box
*src_box
);
235 #define DBG_TEX_DEPTH (1 << 0)
236 #define DBG_COMPUTE (1 << 1)
237 #define DBG_VM (1 << 2)
238 #define DBG_TRACE_CS (1 << 3)
240 #define DBG_FS (1 << 8)
241 #define DBG_VS (1 << 9)
242 #define DBG_GS (1 << 10)
243 #define DBG_PS (1 << 11)
244 #define DBG_CS (1 << 12)
246 #define DBG_NO_HYPERZ (1 << 16)
247 #define DBG_NO_LLVM (1 << 17)
248 #define DBG_NO_CP_DMA (1 << 18)
249 #define DBG_NO_ASYNC_DMA (1 << 19)
250 #define DBG_NO_DISCARD_RANGE (1 << 20)
252 #define DBG_SB (1 << 21)
253 #define DBG_SB_CS (1 << 22)
254 #define DBG_SB_DRY_RUN (1 << 23)
255 #define DBG_SB_STAT (1 << 24)
256 #define DBG_SB_DUMP (1 << 25)
257 #define DBG_SB_NO_FALLBACK (1 << 26)
258 #define DBG_SB_DISASM (1 << 27)
259 #define DBG_SB_SAFEMATH (1 << 28)
261 struct r600_tiling_info
{
262 unsigned num_channels
;
264 unsigned group_bytes
;
268 struct pipe_screen screen
;
269 struct radeon_winsys
*ws
;
270 unsigned debug_flags
;
272 enum chip_class chip_class
;
273 struct radeon_info info
;
277 bool has_compressed_msaa_texturing
;
278 struct r600_tiling_info tiling_info
;
279 struct r600_pipe_fences fences
;
281 /*for compute global memory binding, we allocate stuff here, instead of
283 * XXX: Not sure if this is the best place for global_pool. Also,
284 * it's not thread safe, so it won't work with multiple contexts. */
285 struct compute_memory_pool
*global_pool
;
286 struct r600_resource
*trace_bo
;
289 r600g_dma_blit_t dma_blit
;
291 /* Auxiliary context. Mainly used to initialize resources.
292 * It must be locked prior to using and flushed before unlocking. */
293 struct pipe_context
*aux_context
;
294 pipe_mutex aux_context_lock
;
297 struct r600_pipe_sampler_view
{
298 struct pipe_sampler_view base
;
299 struct r600_resource
*tex_resource
;
300 uint32_t tex_resource_words
[8];
301 bool skip_mip_address_reloc
;
304 struct r600_rasterizer_state
{
305 struct r600_command_buffer buffer
;
308 unsigned sprite_coord_enable
;
309 unsigned clip_plane_enable
;
310 unsigned pa_sc_line_stipple
;
311 unsigned pa_cl_clip_cntl
;
316 bool multisample_enable
;
319 struct r600_poly_offset_state
{
320 struct r600_atom atom
;
321 enum pipe_format zs_format
;
326 struct r600_blend_state
{
327 struct r600_command_buffer buffer
;
328 struct r600_command_buffer buffer_no_blend
;
329 unsigned cb_target_mask
;
330 unsigned cb_color_control
;
331 unsigned cb_color_control_no_blend
;
336 struct r600_dsa_state
{
337 struct r600_command_buffer buffer
;
342 unsigned sx_alpha_test_control
;
345 struct r600_pipe_shader
;
347 struct r600_pipe_shader_selector
{
348 struct r600_pipe_shader
*current
;
350 struct tgsi_token
*tokens
;
351 struct pipe_stream_output_info so
;
353 unsigned num_shaders
;
355 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
358 unsigned nr_ps_max_color_exports
;
361 struct r600_pipe_sampler_state
{
362 uint32_t tex_sampler_words
[3];
363 union pipe_color_union border_color
;
364 bool border_color_use
;
365 bool seamless_cube_map
;
368 /* needed for blitter save */
369 #define NUM_TEX_UNITS 16
371 struct r600_seamless_cube_map
{
372 struct r600_atom atom
;
376 struct r600_samplerview_state
{
377 struct r600_atom atom
;
378 struct r600_pipe_sampler_view
*views
[NUM_TEX_UNITS
];
379 uint32_t enabled_mask
;
381 uint32_t compressed_depthtex_mask
; /* which textures are depth */
382 uint32_t compressed_colortex_mask
;
383 boolean dirty_txq_constants
;
384 boolean dirty_buffer_constants
;
387 struct r600_sampler_states
{
388 struct r600_atom atom
;
389 struct r600_pipe_sampler_state
*states
[NUM_TEX_UNITS
];
390 uint32_t enabled_mask
;
392 uint32_t has_bordercolor_mask
; /* which states contain the border color */
395 struct r600_textures_info
{
396 struct r600_samplerview_state views
;
397 struct r600_sampler_states states
;
398 bool is_array_sampler
[NUM_TEX_UNITS
];
400 /* cube array txq workaround */
401 uint32_t *txq_constants
;
402 /* buffer related workarounds */
403 uint32_t *buffer_constants
;
407 struct pipe_reference reference
;
408 unsigned index
; /* in the shared bo */
409 struct r600_resource
*sleep_bo
;
410 struct list_head head
;
413 #define FENCE_BLOCK_SIZE 16
415 struct r600_fence_block
{
416 struct r600_fence fences
[FENCE_BLOCK_SIZE
];
417 struct list_head head
;
420 #define R600_CONSTANT_ARRAY_SIZE 256
421 #define R600_RESOURCE_ARRAY_SIZE 160
423 struct r600_constbuf_state
425 struct r600_atom atom
;
426 struct pipe_constant_buffer cb
[PIPE_MAX_CONSTANT_BUFFERS
];
427 uint32_t enabled_mask
;
431 struct r600_vertexbuf_state
433 struct r600_atom atom
;
434 struct pipe_vertex_buffer vb
[PIPE_MAX_ATTRIBS
];
435 uint32_t enabled_mask
; /* non-NULL buffers */
439 /* CSO (constant state object, in other words, immutable state). */
440 struct r600_cso_state
442 struct r600_atom atom
;
443 void *cso
; /* e.g. r600_blend_state */
444 struct r600_command_buffer
*cb
;
447 struct r600_scissor_state
449 struct r600_atom atom
;
450 struct pipe_scissor_state scissor
;
451 bool enable
; /* r6xx only */
454 struct r600_fetch_shader
{
455 struct r600_resource
*buffer
;
459 struct r600_shader_state
{
460 struct r600_atom atom
;
461 struct r600_pipe_shader_selector
*shader
;
464 struct r600_query_buffer
{
465 /* The buffer where query results are stored. */
466 struct r600_resource
*buf
;
467 /* Offset of the next free result after current query data */
468 unsigned results_end
;
469 /* If a query buffer is full, a new buffer is created and the old one
470 * is put in here. When we calculate the result, we sum up the samples
471 * from all buffers. */
472 struct r600_query_buffer
*previous
;
476 /* The query buffer and how many results are in it. */
477 struct r600_query_buffer buffer
;
478 /* The type of query */
480 /* Size of the result in memory for both begin_query and end_query,
481 * this can be one or two numbers, or it could even be a size of a structure. */
482 unsigned result_size
;
483 /* The number of dwords for begin_query or end_query. */
485 /* linked list of queries */
486 struct list_head list
;
487 /* for custom non-GPU queries */
488 uint64_t begin_result
;
492 struct r600_so_target
{
493 struct pipe_stream_output_target b
;
495 /* The buffer where BUFFER_FILLED_SIZE is stored. */
496 struct r600_resource
*buf_filled_size
;
497 unsigned buf_filled_size_offset
;
499 unsigned stride_in_dw
;
503 struct r600_streamout
{
504 struct r600_atom begin_atom
;
506 unsigned num_dw_for_end
;
508 unsigned enabled_mask
;
509 unsigned num_targets
;
510 struct r600_so_target
*targets
[PIPE_MAX_SO_BUFFERS
];
512 unsigned append_bitmask
;
517 struct radeon_winsys_cs
*cs
;
519 void (*flush
)(void *ctx
, unsigned flags
);
523 struct r600_ring gfx
;
524 struct r600_ring dma
;
527 struct r600_context
{
528 struct pipe_context context
;
529 struct r600_screen
*screen
;
530 struct radeon_winsys
*ws
;
531 struct r600_rings rings
;
532 struct blitter_context
*blitter
;
533 struct u_upload_mgr
*uploader
;
534 struct u_suballocator
*allocator_so_filled_size
;
535 struct u_suballocator
*allocator_fetch_shader
;
536 struct util_slab_mempool pool_transfers
;
537 unsigned initial_gfx_cs_size
;
540 enum radeon_family family
;
541 enum chip_class chip_class
;
542 boolean has_vertex_cache
;
543 boolean keep_tiling_flags
;
544 unsigned default_ps_gprs
, default_vs_gprs
;
545 unsigned r6xx_num_clause_temp_gprs
;
546 unsigned backend_mask
;
547 unsigned max_db
; /* for OQ */
549 /* current unaccounted memory usage */
553 /* Miscellaneous state objects. */
554 void *custom_dsa_flush
;
555 void *custom_blend_resolve
;
556 void *custom_blend_decompress
;
557 /* With rasterizer discard, there doesn't have to be a pixel shader.
558 * In that case, we bind this one: */
559 void *dummy_pixel_shader
;
560 /* These dummy CMASK and FMASK buffers are used to get around the R6xx hardware
561 * bug where valid CMASK and FMASK are required to be present to avoid
562 * a hardlock in certain operations but aren't actually used
563 * for anything useful. */
564 struct r600_resource
*dummy_fmask
;
565 struct r600_resource
*dummy_cmask
;
567 /* State binding slots are here. */
568 struct r600_atom
*atoms
[R600_NUM_ATOMS
];
569 /* States for CS initialization. */
570 struct r600_command_buffer start_cs_cmd
; /* invariant state mostly */
571 /** Compute specific registers initializations. The start_cs_cmd atom
572 * must be emitted before start_compute_cs_cmd. */
573 struct r600_command_buffer start_compute_cs_cmd
;
574 /* Register states. */
575 struct r600_alphatest_state alphatest_state
;
576 struct r600_cso_state blend_state
;
577 struct r600_blend_color blend_color
;
578 struct r600_cb_misc_state cb_misc_state
;
579 struct r600_clip_misc_state clip_misc_state
;
580 struct r600_clip_state clip_state
;
581 struct r600_db_misc_state db_misc_state
;
582 struct r600_db_state db_state
;
583 struct r600_cso_state dsa_state
;
584 struct r600_framebuffer framebuffer
;
585 struct r600_poly_offset_state poly_offset_state
;
586 struct r600_cso_state rasterizer_state
;
587 struct r600_sample_mask sample_mask
;
588 struct r600_scissor_state scissor
;
589 struct r600_seamless_cube_map seamless_cube_map
;
590 struct r600_config_state config_state
;
591 struct r600_stencil_ref_state stencil_ref
;
592 struct r600_vgt_state vgt_state
;
593 struct r600_viewport_state viewport
;
594 /* Shaders and shader resources. */
595 struct r600_cso_state vertex_fetch_shader
;
596 struct r600_shader_state vertex_shader
;
597 struct r600_shader_state pixel_shader
;
598 struct r600_cs_shader_state cs_shader_state
;
599 struct r600_constbuf_state constbuf_state
[PIPE_SHADER_TYPES
];
600 struct r600_textures_info samplers
[PIPE_SHADER_TYPES
];
601 /** Vertex buffers for fetch shaders */
602 struct r600_vertexbuf_state vertex_buffer_state
;
603 /** Vertex buffers for compute shaders */
604 struct r600_vertexbuf_state cs_vertex_buffer_state
;
605 struct r600_streamout streamout
;
607 /* Additional context states. */
609 unsigned compute_cb_target_mask
;
610 struct r600_pipe_shader_selector
*ps_shader
;
611 struct r600_pipe_shader_selector
*vs_shader
;
612 struct r600_rasterizer_state
*rasterizer
;
614 bool force_blend_disable
;
615 boolean dual_src_blend
;
619 struct pipe_index_buffer index_buffer
;
621 /* Last draw state (-1 = unset). */
622 int last_primitive_type
; /* Last primitive type used in draw_vbo. */
623 int last_start_instance
;
626 /* The list of active queries. Only one query of each type can be active. */
627 int num_occlusion_queries
;
628 int num_pipelinestat_queries
;
629 /* Keep track of non-timer queries, because they should be suspended
630 * during context flushing.
631 * The timer queries (TIME_ELAPSED) shouldn't be suspended. */
632 struct list_head active_nontimer_queries
;
633 unsigned num_cs_dw_nontimer_queries_suspend
;
634 /* If queries have been suspended. */
635 bool nontimer_queries_suspended
;
636 unsigned num_draw_calls
;
638 /* Render condition. */
639 struct pipe_query
*current_render_cond
;
640 unsigned current_render_cond_mode
;
641 boolean current_render_cond_cond
;
642 boolean predicate_drawing
;
645 struct r600_isa
*isa
;
648 static INLINE
void r600_emit_command_buffer(struct radeon_winsys_cs
*cs
,
649 struct r600_command_buffer
*cb
)
651 assert(cs
->cdw
+ cb
->num_dw
<= RADEON_MAX_CMDBUF_DWORDS
);
652 memcpy(cs
->buf
+ cs
->cdw
, cb
->buf
, 4 * cb
->num_dw
);
653 cs
->cdw
+= cb
->num_dw
;
656 void r600_trace_emit(struct r600_context
*rctx
);
658 static INLINE
void r600_emit_atom(struct r600_context
*rctx
, struct r600_atom
*atom
)
660 atom
->emit(rctx
, atom
);
662 if (rctx
->screen
->trace_bo
) {
663 r600_trace_emit(rctx
);
667 static INLINE
void r600_set_cso_state(struct r600_cso_state
*state
, void *cso
)
670 state
->atom
.dirty
= cso
!= NULL
;
673 static INLINE
void r600_set_cso_state_with_cb(struct r600_cso_state
*state
, void *cso
,
674 struct r600_command_buffer
*cb
)
677 state
->atom
.num_dw
= cb
->num_dw
;
678 r600_set_cso_state(state
, cso
);
681 /* compute_memory_pool.c */
682 struct compute_memory_pool
;
683 void compute_memory_pool_delete(struct compute_memory_pool
* pool
);
684 struct compute_memory_pool
* compute_memory_pool_new(
685 struct r600_screen
*rscreen
);
687 /* evergreen_state.c */
688 struct pipe_sampler_view
*
689 evergreen_create_sampler_view_custom(struct pipe_context
*ctx
,
690 struct pipe_resource
*texture
,
691 const struct pipe_sampler_view
*state
,
692 unsigned width0
, unsigned height0
);
693 void evergreen_init_common_regs(struct r600_command_buffer
*cb
,
694 enum chip_class ctx_chip_class
,
695 enum radeon_family ctx_family
,
697 void cayman_init_common_regs(struct r600_command_buffer
*cb
,
698 enum chip_class ctx_chip_class
,
699 enum radeon_family ctx_family
,
702 void evergreen_init_state_functions(struct r600_context
*rctx
);
703 void evergreen_init_atom_start_cs(struct r600_context
*rctx
);
704 void evergreen_update_ps_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
705 void evergreen_update_vs_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
706 void *evergreen_create_db_flush_dsa(struct r600_context
*rctx
);
707 void *evergreen_create_resolve_blend(struct r600_context
*rctx
);
708 void *evergreen_create_decompress_blend(struct r600_context
*rctx
);
709 boolean
evergreen_is_format_supported(struct pipe_screen
*screen
,
710 enum pipe_format format
,
711 enum pipe_texture_target target
,
712 unsigned sample_count
,
714 void evergreen_init_color_surface(struct r600_context
*rctx
,
715 struct r600_surface
*surf
);
716 void evergreen_init_color_surface_rat(struct r600_context
*rctx
,
717 struct r600_surface
*surf
);
718 void evergreen_update_db_shader_control(struct r600_context
* rctx
);
721 void r600_copy_buffer(struct pipe_context
*ctx
, struct pipe_resource
*dst
, unsigned dstx
,
722 struct pipe_resource
*src
, const struct pipe_box
*src_box
);
723 void r600_screen_clear_buffer(struct r600_screen
*rscreen
, struct pipe_resource
*dst
,
724 unsigned offset
, unsigned size
, unsigned char value
);
725 void r600_init_blit_functions(struct r600_context
*rctx
);
726 void r600_blit_decompress_depth(struct pipe_context
*ctx
,
727 struct r600_texture
*texture
,
728 struct r600_texture
*staging
,
729 unsigned first_level
, unsigned last_level
,
730 unsigned first_layer
, unsigned last_layer
,
731 unsigned first_sample
, unsigned last_sample
);
732 void r600_decompress_depth_textures(struct r600_context
*rctx
,
733 struct r600_samplerview_state
*textures
);
734 void r600_decompress_color_textures(struct r600_context
*rctx
,
735 struct r600_samplerview_state
*textures
);
738 bool r600_init_resource(struct r600_screen
*rscreen
,
739 struct r600_resource
*res
,
740 unsigned size
, unsigned alignment
,
741 bool use_reusable_pool
, unsigned usage
);
742 struct pipe_resource
*r600_buffer_create(struct pipe_screen
*screen
,
743 const struct pipe_resource
*templ
,
747 boolean
r600_rings_is_buffer_referenced(struct r600_context
*ctx
,
748 struct radeon_winsys_cs_handle
*buf
,
749 enum radeon_bo_usage usage
);
750 void *r600_buffer_mmap_sync_with_rings(struct r600_context
*ctx
,
751 struct r600_resource
*resource
,
753 const char * r600_llvm_gpu_string(enum radeon_family family
);
757 void r600_init_query_functions(struct r600_context
*rctx
);
758 void r600_suspend_nontimer_queries(struct r600_context
*ctx
);
759 void r600_resume_nontimer_queries(struct r600_context
*ctx
);
761 /* r600_resource.c */
762 void r600_init_context_resource_functions(struct r600_context
*r600
);
765 int r600_pipe_shader_create(struct pipe_context
*ctx
,
766 struct r600_pipe_shader
*shader
,
767 struct r600_shader_key key
);
769 void r600_pipe_shader_destroy(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
772 struct pipe_sampler_view
*
773 r600_create_sampler_view_custom(struct pipe_context
*ctx
,
774 struct pipe_resource
*texture
,
775 const struct pipe_sampler_view
*state
,
776 unsigned width_first_level
, unsigned height_first_level
);
777 void r600_init_state_functions(struct r600_context
*rctx
);
778 void r600_init_atom_start_cs(struct r600_context
*rctx
);
779 void r600_update_ps_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
780 void r600_update_vs_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
781 void *r600_create_db_flush_dsa(struct r600_context
*rctx
);
782 void *r600_create_resolve_blend(struct r600_context
*rctx
);
783 void *r700_create_resolve_blend(struct r600_context
*rctx
);
784 void *r600_create_decompress_blend(struct r600_context
*rctx
);
785 bool r600_adjust_gprs(struct r600_context
*rctx
);
786 boolean
r600_is_format_supported(struct pipe_screen
*screen
,
787 enum pipe_format format
,
788 enum pipe_texture_target target
,
789 unsigned sample_count
,
791 void r600_update_db_shader_control(struct r600_context
* rctx
);
794 void r600_init_screen_texture_functions(struct pipe_screen
*screen
);
795 void r600_init_surface_functions(struct r600_context
*r600
);
796 uint32_t r600_translate_texformat(struct pipe_screen
*screen
, enum pipe_format format
,
797 const unsigned char *swizzle_view
,
798 uint32_t *word4_p
, uint32_t *yuv_format_p
);
799 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
800 struct pipe_resource
*texture
,
801 const struct pipe_surface
*templ
,
802 unsigned width
, unsigned height
);
804 unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format
,
805 const unsigned char *swizzle_view
,
808 /* r600_hw_context.c */
809 void r600_get_backend_mask(struct r600_context
*ctx
);
810 void r600_context_flush(struct r600_context
*ctx
, unsigned flags
);
811 void r600_begin_new_cs(struct r600_context
*ctx
);
812 void r600_context_emit_fence(struct r600_context
*ctx
, struct r600_resource
*fence
,
813 unsigned offset
, unsigned value
);
814 void r600_flush_emit(struct r600_context
*ctx
);
815 void r600_need_cs_space(struct r600_context
*ctx
, unsigned num_dw
, boolean count_draw_in
);
816 void r600_need_dma_space(struct r600_context
*ctx
, unsigned num_dw
);
817 void r600_cp_dma_copy_buffer(struct r600_context
*rctx
,
818 struct pipe_resource
*dst
, uint64_t dst_offset
,
819 struct pipe_resource
*src
, uint64_t src_offset
,
821 void evergreen_cp_dma_clear_buffer(struct r600_context
*rctx
,
822 struct pipe_resource
*dst
, uint64_t offset
,
823 unsigned size
, uint32_t clear_value
);
824 void r600_dma_copy(struct r600_context
*rctx
,
825 struct pipe_resource
*dst
,
826 struct pipe_resource
*src
,
830 boolean
r600_dma_blit(struct pipe_context
*ctx
,
831 struct pipe_resource
*dst
,
833 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
834 struct pipe_resource
*src
,
836 const struct pipe_box
*src_box
);
837 void r600_emit_streamout_begin(struct r600_context
*ctx
, struct r600_atom
*atom
);
838 void r600_emit_streamout_end(struct r600_context
*ctx
);
839 void r600_flag_resource_cache_flush(struct r600_context
*rctx
,
840 struct pipe_resource
*res
);
843 * evergreen_hw_context.c
845 void evergreen_flush_vgt_streamout(struct r600_context
*ctx
);
846 void evergreen_set_streamout_enable(struct r600_context
*ctx
, unsigned buffer_enable_bit
);
847 void evergreen_dma_copy(struct r600_context
*rctx
,
848 struct pipe_resource
*dst
,
849 struct pipe_resource
*src
,
853 boolean
evergreen_dma_blit(struct pipe_context
*ctx
,
854 struct pipe_resource
*dst
,
856 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
857 struct pipe_resource
*src
,
859 const struct pipe_box
*src_box
);
861 /* r600_state_common.c */
862 void r600_init_common_state_functions(struct r600_context
*rctx
);
863 void r600_emit_cso_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
864 void r600_emit_alphatest_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
865 void r600_emit_blend_color(struct r600_context
*rctx
, struct r600_atom
*atom
);
866 void r600_emit_vgt_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
867 void r600_emit_clip_misc_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
868 void r600_emit_stencil_ref(struct r600_context
*rctx
, struct r600_atom
*atom
);
869 void r600_emit_viewport_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
870 void r600_emit_shader(struct r600_context
*rctx
, struct r600_atom
*a
);
871 void r600_init_atom(struct r600_context
*rctx
, struct r600_atom
*atom
, unsigned id
,
872 void (*emit
)(struct r600_context
*ctx
, struct r600_atom
*state
),
874 void r600_vertex_buffers_dirty(struct r600_context
*rctx
);
875 void r600_sampler_views_dirty(struct r600_context
*rctx
,
876 struct r600_samplerview_state
*state
);
877 void r600_sampler_states_dirty(struct r600_context
*rctx
,
878 struct r600_sampler_states
*state
);
879 void r600_constant_buffers_dirty(struct r600_context
*rctx
, struct r600_constbuf_state
*state
);
880 void r600_streamout_buffers_dirty(struct r600_context
*rctx
);
881 void r600_draw_rectangle(struct blitter_context
*blitter
,
882 int x1
, int y1
, int x2
, int y2
, float depth
,
883 enum blitter_attrib_type type
, const union pipe_color_union
*attrib
);
884 uint32_t r600_translate_stencil_op(int s_op
);
885 uint32_t r600_translate_fill(uint32_t func
);
886 unsigned r600_tex_wrap(unsigned wrap
);
887 unsigned r600_tex_filter(unsigned filter
);
888 unsigned r600_tex_mipfilter(unsigned filter
);
889 unsigned r600_tex_compare(unsigned compare
);
890 bool sampler_state_needs_border_color(const struct pipe_sampler_state
*state
);
893 struct pipe_video_codec
*r600_uvd_create_decoder(struct pipe_context
*context
,
894 const struct pipe_video_codec
*decoder
);
896 struct pipe_video_buffer
*r600_video_buffer_create(struct pipe_context
*pipe
,
897 const struct pipe_video_buffer
*tmpl
);
899 int r600_uvd_get_video_param(struct pipe_screen
*screen
,
900 enum pipe_video_profile profile
,
901 enum pipe_video_entrypoint entrypoint
,
902 enum pipe_video_cap param
);
905 * Helpers for building command buffers
908 #define PKT3_SET_CONFIG_REG 0x68
909 #define PKT3_SET_CONTEXT_REG 0x69
910 #define PKT3_SET_CTL_CONST 0x6F
911 #define PKT3_SET_LOOP_CONST 0x6C
913 #define R600_CONFIG_REG_OFFSET 0x08000
914 #define R600_CONTEXT_REG_OFFSET 0x28000
915 #define R600_CTL_CONST_OFFSET 0x3CFF0
916 #define R600_LOOP_CONST_OFFSET 0X0003E200
917 #define EG_LOOP_CONST_OFFSET 0x0003A200
919 #define PKT_TYPE_S(x) (((x) & 0x3) << 30)
920 #define PKT_COUNT_S(x) (((x) & 0x3FFF) << 16)
921 #define PKT3_IT_OPCODE_S(x) (((x) & 0xFF) << 8)
922 #define PKT3_PREDICATE(x) (((x) >> 0) & 0x1)
923 #define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT_COUNT_S(count) | PKT3_IT_OPCODE_S(op) | PKT3_PREDICATE(predicate))
925 #define RADEON_CP_PACKET3_COMPUTE_MODE 0x00000002
927 /*Evergreen Compute packet3*/
928 #define PKT3C(op, count, predicate) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count) | PKT3_PREDICATE(predicate) | RADEON_CP_PACKET3_COMPUTE_MODE)
930 static INLINE
void r600_store_value(struct r600_command_buffer
*cb
, unsigned value
)
932 cb
->buf
[cb
->num_dw
++] = value
;
935 static INLINE
void r600_store_array(struct r600_command_buffer
*cb
, unsigned num
, unsigned *ptr
)
937 assert(cb
->num_dw
+num
<= cb
->max_num_dw
);
938 memcpy(&cb
->buf
[cb
->num_dw
], ptr
, num
* sizeof(ptr
[0]));
942 static INLINE
void r600_store_config_reg_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
944 assert(reg
< R600_CONTEXT_REG_OFFSET
);
945 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
946 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_CONFIG_REG
, num
, 0);
947 cb
->buf
[cb
->num_dw
++] = (reg
- R600_CONFIG_REG_OFFSET
) >> 2;
951 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
954 static INLINE
void r600_store_context_reg_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
956 assert(reg
>= R600_CONTEXT_REG_OFFSET
&& reg
< R600_CTL_CONST_OFFSET
);
957 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
958 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_CONTEXT_REG
, num
, 0) | cb
->pkt_flags
;
959 cb
->buf
[cb
->num_dw
++] = (reg
- R600_CONTEXT_REG_OFFSET
) >> 2;
963 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
966 static INLINE
void r600_store_ctl_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
968 assert(reg
>= R600_CTL_CONST_OFFSET
);
969 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
970 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_CTL_CONST
, num
, 0) | cb
->pkt_flags
;
971 cb
->buf
[cb
->num_dw
++] = (reg
- R600_CTL_CONST_OFFSET
) >> 2;
974 static INLINE
void r600_store_loop_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
976 assert(reg
>= R600_LOOP_CONST_OFFSET
);
977 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
978 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_LOOP_CONST
, num
, 0);
979 cb
->buf
[cb
->num_dw
++] = (reg
- R600_LOOP_CONST_OFFSET
) >> 2;
983 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
986 static INLINE
void eg_store_loop_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
988 assert(reg
>= EG_LOOP_CONST_OFFSET
);
989 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
990 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_LOOP_CONST
, num
, 0) | cb
->pkt_flags
;
991 cb
->buf
[cb
->num_dw
++] = (reg
- EG_LOOP_CONST_OFFSET
) >> 2;
994 static INLINE
void r600_store_config_reg(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
996 r600_store_config_reg_seq(cb
, reg
, 1);
997 r600_store_value(cb
, value
);
1000 static INLINE
void r600_store_context_reg(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
1002 r600_store_context_reg_seq(cb
, reg
, 1);
1003 r600_store_value(cb
, value
);
1006 static INLINE
void r600_store_ctl_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
1008 r600_store_ctl_const_seq(cb
, reg
, 1);
1009 r600_store_value(cb
, value
);
1012 static INLINE
void r600_store_loop_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
1014 r600_store_loop_const_seq(cb
, reg
, 1);
1015 r600_store_value(cb
, value
);
1018 static INLINE
void eg_store_loop_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
1020 eg_store_loop_const_seq(cb
, reg
, 1);
1021 r600_store_value(cb
, value
);
1024 void r600_init_command_buffer(struct r600_command_buffer
*cb
, unsigned num_dw
);
1025 void r600_release_command_buffer(struct r600_command_buffer
*cb
);
1028 * Helpers for emitting state into a command stream directly.
1030 static INLINE
unsigned r600_context_bo_reloc(struct r600_context
*ctx
,
1031 struct r600_ring
*ring
,
1032 struct r600_resource
*rbo
,
1033 enum radeon_bo_usage usage
)
1036 /* make sure that all previous ring use are flushed so everything
1037 * look serialized from driver pov
1039 if (!ring
->flushing
) {
1040 if (ring
== &ctx
->rings
.gfx
) {
1041 if (ctx
->rings
.dma
.cs
) {
1042 /* flush dma ring */
1043 ctx
->rings
.dma
.flush(ctx
, RADEON_FLUSH_ASYNC
);
1046 /* flush gfx ring */
1047 ctx
->rings
.gfx
.flush(ctx
, RADEON_FLUSH_ASYNC
);
1050 return ctx
->ws
->cs_add_reloc(ring
->cs
, rbo
->cs_buf
, usage
, rbo
->domains
) * 4;
1053 static INLINE
void r600_write_value(struct radeon_winsys_cs
*cs
, unsigned value
)
1055 cs
->buf
[cs
->cdw
++] = value
;
1058 static INLINE
void r600_write_array(struct radeon_winsys_cs
*cs
, unsigned num
, unsigned *ptr
)
1060 assert(cs
->cdw
+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1061 memcpy(&cs
->buf
[cs
->cdw
], ptr
, num
* sizeof(ptr
[0]));
1065 static INLINE
void r600_write_config_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1067 assert(reg
< R600_CONTEXT_REG_OFFSET
);
1068 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1069 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CONFIG_REG
, num
, 0);
1070 cs
->buf
[cs
->cdw
++] = (reg
- R600_CONFIG_REG_OFFSET
) >> 2;
1073 static INLINE
void r600_write_context_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1075 assert(reg
>= R600_CONTEXT_REG_OFFSET
&& reg
< R600_CTL_CONST_OFFSET
);
1076 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1077 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CONTEXT_REG
, num
, 0);
1078 cs
->buf
[cs
->cdw
++] = (reg
- R600_CONTEXT_REG_OFFSET
) >> 2;
1081 static INLINE
void r600_write_compute_context_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1083 r600_write_context_reg_seq(cs
, reg
, num
);
1084 /* Set the compute bit on the packet header */
1085 cs
->buf
[cs
->cdw
- 2] |= RADEON_CP_PACKET3_COMPUTE_MODE
;
1088 static INLINE
void r600_write_ctl_const_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1090 assert(reg
>= R600_CTL_CONST_OFFSET
);
1091 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1092 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CTL_CONST
, num
, 0);
1093 cs
->buf
[cs
->cdw
++] = (reg
- R600_CTL_CONST_OFFSET
) >> 2;
1096 static INLINE
void r600_write_config_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1098 r600_write_config_reg_seq(cs
, reg
, 1);
1099 r600_write_value(cs
, value
);
1102 static INLINE
void r600_write_context_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1104 r600_write_context_reg_seq(cs
, reg
, 1);
1105 r600_write_value(cs
, value
);
1108 static INLINE
void r600_write_compute_context_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1110 r600_write_compute_context_reg_seq(cs
, reg
, 1);
1111 r600_write_value(cs
, value
);
1114 static INLINE
void r600_write_context_reg_flag(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
, unsigned flag
)
1116 if (flag
& RADEON_CP_PACKET3_COMPUTE_MODE
) {
1117 r600_write_compute_context_reg(cs
, reg
, value
);
1119 r600_write_context_reg(cs
, reg
, value
);
1123 static INLINE
void r600_write_ctl_const(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1125 r600_write_ctl_const_seq(cs
, reg
, 1);
1126 r600_write_value(cs
, value
);
1132 static INLINE
uint32_t S_FIXED(float value
, uint32_t frac_bits
)
1134 return value
* (1 << frac_bits
);
1136 #define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
1138 static inline unsigned r600_tex_aniso_filter(unsigned filter
)
1140 if (filter
<= 1) return 0;
1141 if (filter
<= 2) return 1;
1142 if (filter
<= 4) return 2;
1143 if (filter
<= 8) return 3;
1144 /* else */ return 4;
1147 /* 12.4 fixed-point */
1148 static INLINE
unsigned r600_pack_float_12p4(float x
)
1151 x
>= 4096 ? 0xffff : x
* 16;
1154 static INLINE
uint64_t r600_resource_va(struct pipe_screen
*screen
, struct pipe_resource
*resource
)
1156 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
1157 struct r600_resource
*rresource
= (struct r600_resource
*)resource
;
1159 return rscreen
->ws
->buffer_get_virtual_address(rresource
->cs_buf
);
1162 static INLINE
void r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
)
1164 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
1165 struct r600_resource
*rr
= (struct r600_resource
*)r
;
1172 * The idea is to compute a gross estimate of memory requirement of
1173 * each draw call. After each draw call, memory will be precisely
1174 * accounted. So the uncertainty is only on the current draw call.
1175 * In practice this gave very good estimate (+/- 10% of the target
1178 if (rr
->domains
& RADEON_DOMAIN_GTT
) {
1179 rctx
->gtt
+= rr
->buf
->size
;
1181 if (rr
->domains
& RADEON_DOMAIN_VRAM
) {
1182 rctx
->vram
+= rr
->buf
->size
;