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)
67 #define R600_CONTEXT_INVAL_READ_CACHES (1 << 0)
68 #define R600_CONTEXT_STREAMOUT_FLUSH (1 << 1)
69 #define R600_CONTEXT_WAIT_3D_IDLE (1 << 2)
70 #define R600_CONTEXT_WAIT_CP_DMA_IDLE (1 << 3)
71 #define R600_CONTEXT_FLUSH_AND_INV (1 << 4)
72 #define R600_CONTEXT_FLUSH_AND_INV_CB_META (1 << 5)
73 #define R600_CONTEXT_PS_PARTIAL_FLUSH (1 << 6)
74 #define R600_CONTEXT_FLUSH_AND_INV_DB_META (1 << 7)
76 #define R600_QUERY_DRAW_CALLS (PIPE_QUERY_DRIVER_SPECIFIC + 0)
77 #define R600_QUERY_REQUESTED_VRAM (PIPE_QUERY_DRIVER_SPECIFIC + 1)
78 #define R600_QUERY_REQUESTED_GTT (PIPE_QUERY_DRIVER_SPECIFIC + 2)
79 #define R600_QUERY_BUFFER_WAIT_TIME (PIPE_QUERY_DRIVER_SPECIFIC + 3)
83 struct r600_shader_key
;
85 /* This encapsulates a state or an operation which can emitted into the GPU
86 * command stream. It's not limited to states only, it can be used for anything
87 * that wants to write commands into the CS (e.g. cache flushes). */
89 void (*emit
)(struct r600_context
*ctx
, struct r600_atom
*state
);
95 /* This is an atom containing GPU commands that never change.
96 * This is supposed to be copied directly into the CS. */
97 struct r600_command_buffer
{
104 struct r600_db_state
{
105 struct r600_atom atom
;
106 struct r600_surface
*rsurf
;
109 struct r600_db_misc_state
{
110 struct r600_atom atom
;
111 bool occlusion_query_enabled
;
112 bool flush_depthstencil_through_cb
;
113 bool flush_depthstencil_in_place
;
114 bool copy_depth
, copy_stencil
;
115 unsigned copy_sample
;
116 unsigned log_samples
;
117 unsigned db_shader_control
;
121 struct r600_cb_misc_state
{
122 struct r600_atom atom
;
123 unsigned cb_color_control
; /* this comes from blend state */
124 unsigned blend_colormask
; /* 8*4 bits for 8 RGBA colorbuffers */
126 unsigned nr_ps_color_outputs
;
131 struct r600_clip_misc_state
{
132 struct r600_atom atom
;
133 unsigned pa_cl_clip_cntl
; /* from rasterizer */
134 unsigned pa_cl_vs_out_cntl
; /* from vertex shader */
135 unsigned clip_plane_enable
; /* from rasterizer */
136 unsigned clip_dist_write
; /* from vertex shader */
139 struct r600_alphatest_state
{
140 struct r600_atom atom
;
141 unsigned sx_alpha_test_control
; /* this comes from dsa state */
142 unsigned sx_alpha_ref
; /* this comes from dsa state */
144 bool cb0_export_16bpc
; /* from set_framebuffer_state */
147 struct r600_vgt_state
{
148 struct r600_atom atom
;
149 uint32_t vgt_multi_prim_ib_reset_en
;
150 uint32_t vgt_multi_prim_ib_reset_indx
;
151 uint32_t vgt_indx_offset
;
154 struct r600_blend_color
{
155 struct r600_atom atom
;
156 struct pipe_blend_color state
;
159 struct r600_clip_state
{
160 struct r600_atom atom
;
161 struct pipe_clip_state state
;
164 struct r600_cs_shader_state
{
165 struct r600_atom atom
;
166 unsigned kernel_index
;
167 struct r600_pipe_compute
*shader
;
170 struct r600_framebuffer
{
171 struct r600_atom atom
;
172 struct pipe_framebuffer_state state
;
173 unsigned compressed_cb_mask
;
177 bool is_msaa_resolve
;
180 struct r600_sample_mask
{
181 struct r600_atom atom
;
182 uint16_t sample_mask
; /* there are only 8 bits on EG, 16 bits on Cayman */
185 struct r600_config_state
{
186 struct r600_atom atom
;
187 unsigned sq_gpr_resource_mgmt_1
;
190 struct r600_stencil_ref
197 struct r600_stencil_ref_state
{
198 struct r600_atom atom
;
199 struct r600_stencil_ref state
;
200 struct pipe_stencil_ref pipe_state
;
203 struct r600_viewport_state
{
204 struct r600_atom atom
;
205 struct pipe_viewport_state state
;
208 struct r600_pipe_fences
{
209 struct r600_resource
*bo
;
212 /* linked list of preallocated blocks */
213 struct list_head blocks
;
214 /* linked list of freed fences */
215 struct list_head pool
;
219 typedef boolean (*r600g_dma_blit_t
)(struct pipe_context
*ctx
,
220 struct pipe_resource
*dst
,
222 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
223 struct pipe_resource
*src
,
225 const struct pipe_box
*src_box
);
228 #define DBG_TEX_DEPTH (1 << 0)
229 #define DBG_COMPUTE (1 << 1)
230 #define DBG_VM (1 << 2)
231 #define DBG_TRACE_CS (1 << 3)
233 #define DBG_FS (1 << 8)
234 #define DBG_VS (1 << 9)
235 #define DBG_GS (1 << 10)
236 #define DBG_PS (1 << 11)
237 #define DBG_CS (1 << 12)
239 #define DBG_NO_HYPERZ (1 << 16)
240 #define DBG_NO_LLVM (1 << 17)
241 #define DBG_NO_CP_DMA (1 << 18)
242 #define DBG_NO_ASYNC_DMA (1 << 19)
243 #define DBG_NO_DISCARD_RANGE (1 << 20)
245 #define DBG_SB (1 << 21)
246 #define DBG_SB_CS (1 << 22)
247 #define DBG_SB_DRY_RUN (1 << 23)
248 #define DBG_SB_STAT (1 << 24)
249 #define DBG_SB_DUMP (1 << 25)
250 #define DBG_SB_NO_FALLBACK (1 << 26)
251 #define DBG_SB_DISASM (1 << 27)
253 struct r600_tiling_info
{
254 unsigned num_channels
;
256 unsigned group_bytes
;
260 struct pipe_screen screen
;
261 struct radeon_winsys
*ws
;
262 unsigned debug_flags
;
264 enum chip_class chip_class
;
265 struct radeon_info info
;
269 bool has_compressed_msaa_texturing
;
270 struct r600_tiling_info tiling_info
;
271 struct r600_pipe_fences fences
;
273 /*for compute global memory binding, we allocate stuff here, instead of
275 * XXX: Not sure if this is the best place for global_pool. Also,
276 * it's not thread safe, so it won't work with multiple contexts. */
277 struct compute_memory_pool
*global_pool
;
278 struct r600_resource
*trace_bo
;
281 r600g_dma_blit_t dma_blit
;
283 /* Auxiliary context. Mainly used to initialize resources.
284 * It must be locked prior to using and flushed before unlocking. */
285 struct pipe_context
*aux_context
;
286 pipe_mutex aux_context_lock
;
289 struct r600_pipe_sampler_view
{
290 struct pipe_sampler_view base
;
291 struct r600_resource
*tex_resource
;
292 uint32_t tex_resource_words
[8];
293 bool skip_mip_address_reloc
;
296 struct r600_rasterizer_state
{
297 struct r600_command_buffer buffer
;
300 unsigned sprite_coord_enable
;
301 unsigned clip_plane_enable
;
302 unsigned pa_sc_line_stipple
;
303 unsigned pa_cl_clip_cntl
;
308 bool multisample_enable
;
311 struct r600_poly_offset_state
{
312 struct r600_atom atom
;
313 enum pipe_format zs_format
;
318 struct r600_blend_state
{
319 struct r600_command_buffer buffer
;
320 struct r600_command_buffer buffer_no_blend
;
321 unsigned cb_target_mask
;
322 unsigned cb_color_control
;
323 unsigned cb_color_control_no_blend
;
328 struct r600_dsa_state
{
329 struct r600_command_buffer buffer
;
334 unsigned sx_alpha_test_control
;
337 struct r600_pipe_shader
;
339 struct r600_pipe_shader_selector
{
340 struct r600_pipe_shader
*current
;
342 struct tgsi_token
*tokens
;
343 struct pipe_stream_output_info so
;
345 unsigned num_shaders
;
347 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
350 unsigned nr_ps_max_color_exports
;
353 struct r600_pipe_sampler_state
{
354 uint32_t tex_sampler_words
[3];
355 union pipe_color_union border_color
;
356 bool border_color_use
;
357 bool seamless_cube_map
;
360 /* needed for blitter save */
361 #define NUM_TEX_UNITS 16
363 struct r600_seamless_cube_map
{
364 struct r600_atom atom
;
368 struct r600_samplerview_state
{
369 struct r600_atom atom
;
370 struct r600_pipe_sampler_view
*views
[NUM_TEX_UNITS
];
371 uint32_t enabled_mask
;
373 uint32_t compressed_depthtex_mask
; /* which textures are depth */
374 uint32_t compressed_colortex_mask
;
375 boolean dirty_txq_constants
;
376 boolean dirty_buffer_constants
;
379 struct r600_sampler_states
{
380 struct r600_atom atom
;
381 struct r600_pipe_sampler_state
*states
[NUM_TEX_UNITS
];
382 uint32_t enabled_mask
;
384 uint32_t has_bordercolor_mask
; /* which states contain the border color */
387 struct r600_textures_info
{
388 struct r600_samplerview_state views
;
389 struct r600_sampler_states states
;
390 bool is_array_sampler
[NUM_TEX_UNITS
];
392 /* cube array txq workaround */
393 uint32_t *txq_constants
;
394 /* buffer related workarounds */
395 uint32_t *buffer_constants
;
399 struct pipe_reference reference
;
400 unsigned index
; /* in the shared bo */
401 struct r600_resource
*sleep_bo
;
402 struct list_head head
;
405 #define FENCE_BLOCK_SIZE 16
407 struct r600_fence_block
{
408 struct r600_fence fences
[FENCE_BLOCK_SIZE
];
409 struct list_head head
;
412 #define R600_CONSTANT_ARRAY_SIZE 256
413 #define R600_RESOURCE_ARRAY_SIZE 160
415 struct r600_constbuf_state
417 struct r600_atom atom
;
418 struct pipe_constant_buffer cb
[PIPE_MAX_CONSTANT_BUFFERS
];
419 uint32_t enabled_mask
;
423 struct r600_vertexbuf_state
425 struct r600_atom atom
;
426 struct pipe_vertex_buffer vb
[PIPE_MAX_ATTRIBS
];
427 uint32_t enabled_mask
; /* non-NULL buffers */
431 /* CSO (constant state object, in other words, immutable state). */
432 struct r600_cso_state
434 struct r600_atom atom
;
435 void *cso
; /* e.g. r600_blend_state */
436 struct r600_command_buffer
*cb
;
439 struct r600_scissor_state
441 struct r600_atom atom
;
442 struct pipe_scissor_state scissor
;
443 bool enable
; /* r6xx only */
446 struct r600_fetch_shader
{
447 struct r600_resource
*buffer
;
451 struct r600_shader_state
{
452 struct r600_atom atom
;
453 struct r600_pipe_shader_selector
*shader
;
456 struct r600_query_buffer
{
457 /* The buffer where query results are stored. */
458 struct r600_resource
*buf
;
459 /* Offset of the next free result after current query data */
460 unsigned results_end
;
461 /* If a query buffer is full, a new buffer is created and the old one
462 * is put in here. When we calculate the result, we sum up the samples
463 * from all buffers. */
464 struct r600_query_buffer
*previous
;
468 /* The query buffer and how many results are in it. */
469 struct r600_query_buffer buffer
;
470 /* The type of query */
472 /* Size of the result in memory for both begin_query and end_query,
473 * this can be one or two numbers, or it could even be a size of a structure. */
474 unsigned result_size
;
475 /* The number of dwords for begin_query or end_query. */
477 /* linked list of queries */
478 struct list_head list
;
479 /* for custom non-GPU queries */
480 uint64_t begin_result
;
484 struct r600_so_target
{
485 struct pipe_stream_output_target b
;
487 /* The buffer where BUFFER_FILLED_SIZE is stored. */
488 struct r600_resource
*buf_filled_size
;
489 unsigned buf_filled_size_offset
;
491 unsigned stride_in_dw
;
495 struct r600_streamout
{
496 struct r600_atom begin_atom
;
498 unsigned num_dw_for_end
;
500 unsigned enabled_mask
;
501 unsigned num_targets
;
502 struct r600_so_target
*targets
[PIPE_MAX_SO_BUFFERS
];
504 unsigned append_bitmask
;
509 struct radeon_winsys_cs
*cs
;
511 void (*flush
)(void *ctx
, unsigned flags
);
515 struct r600_ring gfx
;
516 struct r600_ring dma
;
519 struct r600_context
{
520 struct pipe_context context
;
521 struct r600_screen
*screen
;
522 struct radeon_winsys
*ws
;
523 struct r600_rings rings
;
524 struct blitter_context
*blitter
;
525 struct u_upload_mgr
*uploader
;
526 struct u_suballocator
*allocator_so_filled_size
;
527 struct u_suballocator
*allocator_fetch_shader
;
528 struct util_slab_mempool pool_transfers
;
531 enum radeon_family family
;
532 enum chip_class chip_class
;
533 boolean has_vertex_cache
;
534 boolean keep_tiling_flags
;
535 unsigned default_ps_gprs
, default_vs_gprs
;
536 unsigned r6xx_num_clause_temp_gprs
;
537 unsigned backend_mask
;
538 unsigned max_db
; /* for OQ */
540 /* current unaccounted memory usage */
544 /* Miscellaneous state objects. */
545 void *custom_dsa_flush
;
546 void *custom_blend_resolve
;
547 void *custom_blend_decompress
;
548 /* With rasterizer discard, there doesn't have to be a pixel shader.
549 * In that case, we bind this one: */
550 void *dummy_pixel_shader
;
551 /* These dummy CMASK and FMASK buffers are used to get around the R6xx hardware
552 * bug where valid CMASK and FMASK are required to be present to avoid
553 * a hardlock in certain operations but aren't actually used
554 * for anything useful. */
555 struct r600_resource
*dummy_fmask
;
556 struct r600_resource
*dummy_cmask
;
558 /* State binding slots are here. */
559 struct r600_atom
*atoms
[R600_NUM_ATOMS
];
560 /* States for CS initialization. */
561 struct r600_command_buffer start_cs_cmd
; /* invariant state mostly */
562 /** Compute specific registers initializations. The start_cs_cmd atom
563 * must be emitted before start_compute_cs_cmd. */
564 struct r600_command_buffer start_compute_cs_cmd
;
565 /* Register states. */
566 struct r600_alphatest_state alphatest_state
;
567 struct r600_cso_state blend_state
;
568 struct r600_blend_color blend_color
;
569 struct r600_cb_misc_state cb_misc_state
;
570 struct r600_clip_misc_state clip_misc_state
;
571 struct r600_clip_state clip_state
;
572 struct r600_db_misc_state db_misc_state
;
573 struct r600_db_state db_state
;
574 struct r600_cso_state dsa_state
;
575 struct r600_framebuffer framebuffer
;
576 struct r600_poly_offset_state poly_offset_state
;
577 struct r600_cso_state rasterizer_state
;
578 struct r600_sample_mask sample_mask
;
579 struct r600_scissor_state scissor
;
580 struct r600_seamless_cube_map seamless_cube_map
;
581 struct r600_config_state config_state
;
582 struct r600_stencil_ref_state stencil_ref
;
583 struct r600_vgt_state vgt_state
;
584 struct r600_viewport_state viewport
;
585 /* Shaders and shader resources. */
586 struct r600_cso_state vertex_fetch_shader
;
587 struct r600_shader_state vertex_shader
;
588 struct r600_shader_state pixel_shader
;
589 struct r600_cs_shader_state cs_shader_state
;
590 struct r600_constbuf_state constbuf_state
[PIPE_SHADER_TYPES
];
591 struct r600_textures_info samplers
[PIPE_SHADER_TYPES
];
592 /** Vertex buffers for fetch shaders */
593 struct r600_vertexbuf_state vertex_buffer_state
;
594 /** Vertex buffers for compute shaders */
595 struct r600_vertexbuf_state cs_vertex_buffer_state
;
596 struct r600_streamout streamout
;
598 /* Additional context states. */
600 unsigned compute_cb_target_mask
;
601 struct r600_pipe_shader_selector
*ps_shader
;
602 struct r600_pipe_shader_selector
*vs_shader
;
603 struct r600_rasterizer_state
*rasterizer
;
605 bool force_blend_disable
;
606 boolean dual_src_blend
;
610 struct pipe_index_buffer index_buffer
;
612 /* Last draw state (-1 = unset). */
613 int last_primitive_type
; /* Last primitive type used in draw_vbo. */
614 int last_start_instance
;
617 /* The list of active queries. Only one query of each type can be active. */
618 int num_occlusion_queries
;
619 int num_pipelinestat_queries
;
620 /* Keep track of non-timer queries, because they should be suspended
621 * during context flushing.
622 * The timer queries (TIME_ELAPSED) shouldn't be suspended. */
623 struct list_head active_nontimer_queries
;
624 unsigned num_cs_dw_nontimer_queries_suspend
;
625 /* If queries have been suspended. */
626 bool nontimer_queries_suspended
;
627 unsigned num_draw_calls
;
629 /* Render condition. */
630 struct pipe_query
*current_render_cond
;
631 unsigned current_render_cond_mode
;
632 boolean current_render_cond_cond
;
633 boolean predicate_drawing
;
636 struct r600_isa
*isa
;
639 static INLINE
void r600_emit_command_buffer(struct radeon_winsys_cs
*cs
,
640 struct r600_command_buffer
*cb
)
642 assert(cs
->cdw
+ cb
->num_dw
<= RADEON_MAX_CMDBUF_DWORDS
);
643 memcpy(cs
->buf
+ cs
->cdw
, cb
->buf
, 4 * cb
->num_dw
);
644 cs
->cdw
+= cb
->num_dw
;
647 void r600_trace_emit(struct r600_context
*rctx
);
649 static INLINE
void r600_emit_atom(struct r600_context
*rctx
, struct r600_atom
*atom
)
651 atom
->emit(rctx
, atom
);
653 if (rctx
->screen
->trace_bo
) {
654 r600_trace_emit(rctx
);
658 static INLINE
void r600_set_cso_state(struct r600_cso_state
*state
, void *cso
)
661 state
->atom
.dirty
= cso
!= NULL
;
664 static INLINE
void r600_set_cso_state_with_cb(struct r600_cso_state
*state
, void *cso
,
665 struct r600_command_buffer
*cb
)
668 state
->atom
.num_dw
= cb
->num_dw
;
669 r600_set_cso_state(state
, cso
);
672 /* compute_memory_pool.c */
673 struct compute_memory_pool
;
674 void compute_memory_pool_delete(struct compute_memory_pool
* pool
);
675 struct compute_memory_pool
* compute_memory_pool_new(
676 struct r600_screen
*rscreen
);
678 /* evergreen_state.c */
679 struct pipe_sampler_view
*
680 evergreen_create_sampler_view_custom(struct pipe_context
*ctx
,
681 struct pipe_resource
*texture
,
682 const struct pipe_sampler_view
*state
,
683 unsigned width0
, unsigned height0
);
684 void evergreen_init_common_regs(struct r600_command_buffer
*cb
,
685 enum chip_class ctx_chip_class
,
686 enum radeon_family ctx_family
,
688 void cayman_init_common_regs(struct r600_command_buffer
*cb
,
689 enum chip_class ctx_chip_class
,
690 enum radeon_family ctx_family
,
693 void evergreen_init_state_functions(struct r600_context
*rctx
);
694 void evergreen_init_atom_start_cs(struct r600_context
*rctx
);
695 void evergreen_update_ps_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
696 void evergreen_update_vs_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
697 void *evergreen_create_db_flush_dsa(struct r600_context
*rctx
);
698 void *evergreen_create_resolve_blend(struct r600_context
*rctx
);
699 void *evergreen_create_decompress_blend(struct r600_context
*rctx
);
700 boolean
evergreen_is_format_supported(struct pipe_screen
*screen
,
701 enum pipe_format format
,
702 enum pipe_texture_target target
,
703 unsigned sample_count
,
705 void evergreen_init_color_surface(struct r600_context
*rctx
,
706 struct r600_surface
*surf
);
707 void evergreen_init_color_surface_rat(struct r600_context
*rctx
,
708 struct r600_surface
*surf
);
709 void evergreen_update_db_shader_control(struct r600_context
* rctx
);
712 void r600_copy_buffer(struct pipe_context
*ctx
, struct pipe_resource
*dst
, unsigned dstx
,
713 struct pipe_resource
*src
, const struct pipe_box
*src_box
);
714 void r600_screen_clear_buffer(struct r600_screen
*rscreen
, struct pipe_resource
*dst
,
715 unsigned offset
, unsigned size
, unsigned char value
);
716 void r600_init_blit_functions(struct r600_context
*rctx
);
717 void r600_blit_decompress_depth(struct pipe_context
*ctx
,
718 struct r600_texture
*texture
,
719 struct r600_texture
*staging
,
720 unsigned first_level
, unsigned last_level
,
721 unsigned first_layer
, unsigned last_layer
,
722 unsigned first_sample
, unsigned last_sample
);
723 void r600_decompress_depth_textures(struct r600_context
*rctx
,
724 struct r600_samplerview_state
*textures
);
725 void r600_decompress_color_textures(struct r600_context
*rctx
,
726 struct r600_samplerview_state
*textures
);
729 bool r600_init_resource(struct r600_screen
*rscreen
,
730 struct r600_resource
*res
,
731 unsigned size
, unsigned alignment
,
732 bool use_reusable_pool
, unsigned usage
);
733 struct pipe_resource
*r600_buffer_create(struct pipe_screen
*screen
,
734 const struct pipe_resource
*templ
,
738 boolean
r600_rings_is_buffer_referenced(struct r600_context
*ctx
,
739 struct radeon_winsys_cs_handle
*buf
,
740 enum radeon_bo_usage usage
);
741 void *r600_buffer_mmap_sync_with_rings(struct r600_context
*ctx
,
742 struct r600_resource
*resource
,
744 const char * r600_llvm_gpu_string(enum radeon_family family
);
748 void r600_init_query_functions(struct r600_context
*rctx
);
749 void r600_suspend_nontimer_queries(struct r600_context
*ctx
);
750 void r600_resume_nontimer_queries(struct r600_context
*ctx
);
752 /* r600_resource.c */
753 void r600_init_context_resource_functions(struct r600_context
*r600
);
756 int r600_pipe_shader_create(struct pipe_context
*ctx
,
757 struct r600_pipe_shader
*shader
,
758 struct r600_shader_key key
);
760 int r600_compute_shader_create(struct pipe_context
* ctx
,
761 LLVMModuleRef mod
, struct r600_bytecode
* bytecode
);
763 void r600_pipe_shader_destroy(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
766 struct pipe_sampler_view
*
767 r600_create_sampler_view_custom(struct pipe_context
*ctx
,
768 struct pipe_resource
*texture
,
769 const struct pipe_sampler_view
*state
,
770 unsigned width_first_level
, unsigned height_first_level
);
771 void r600_init_state_functions(struct r600_context
*rctx
);
772 void r600_init_atom_start_cs(struct r600_context
*rctx
);
773 void r600_update_ps_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
774 void r600_update_vs_state(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
775 void *r600_create_db_flush_dsa(struct r600_context
*rctx
);
776 void *r600_create_resolve_blend(struct r600_context
*rctx
);
777 void *r700_create_resolve_blend(struct r600_context
*rctx
);
778 void *r600_create_decompress_blend(struct r600_context
*rctx
);
779 bool r600_adjust_gprs(struct r600_context
*rctx
);
780 boolean
r600_is_format_supported(struct pipe_screen
*screen
,
781 enum pipe_format format
,
782 enum pipe_texture_target target
,
783 unsigned sample_count
,
785 void r600_update_db_shader_control(struct r600_context
* rctx
);
788 void r600_init_screen_texture_functions(struct pipe_screen
*screen
);
789 void r600_init_surface_functions(struct r600_context
*r600
);
790 uint32_t r600_translate_texformat(struct pipe_screen
*screen
, enum pipe_format format
,
791 const unsigned char *swizzle_view
,
792 uint32_t *word4_p
, uint32_t *yuv_format_p
);
793 unsigned r600_texture_get_offset(struct r600_texture
*rtex
,
794 unsigned level
, unsigned layer
);
795 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
796 struct pipe_resource
*texture
,
797 const struct pipe_surface
*templ
,
798 unsigned width
, unsigned height
);
800 unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format
,
801 const unsigned char *swizzle_view
,
804 /* r600_hw_context.c */
805 void r600_get_backend_mask(struct r600_context
*ctx
);
806 void r600_context_flush(struct r600_context
*ctx
, unsigned flags
);
807 void r600_begin_new_cs(struct r600_context
*ctx
);
808 void r600_context_emit_fence(struct r600_context
*ctx
, struct r600_resource
*fence
,
809 unsigned offset
, unsigned value
);
810 void r600_flush_emit(struct r600_context
*ctx
);
811 void r600_need_cs_space(struct r600_context
*ctx
, unsigned num_dw
, boolean count_draw_in
);
812 void r600_need_dma_space(struct r600_context
*ctx
, unsigned num_dw
);
813 void r600_cp_dma_copy_buffer(struct r600_context
*rctx
,
814 struct pipe_resource
*dst
, uint64_t dst_offset
,
815 struct pipe_resource
*src
, uint64_t src_offset
,
817 void evergreen_cp_dma_clear_buffer(struct r600_context
*rctx
,
818 struct pipe_resource
*dst
, uint64_t offset
,
819 unsigned size
, uint32_t clear_value
);
820 void r600_dma_copy(struct r600_context
*rctx
,
821 struct pipe_resource
*dst
,
822 struct pipe_resource
*src
,
826 boolean
r600_dma_blit(struct pipe_context
*ctx
,
827 struct pipe_resource
*dst
,
829 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
830 struct pipe_resource
*src
,
832 const struct pipe_box
*src_box
);
833 void r600_emit_streamout_begin(struct r600_context
*ctx
, struct r600_atom
*atom
);
834 void r600_emit_streamout_end(struct r600_context
*ctx
);
837 * evergreen_hw_context.c
839 void evergreen_flush_vgt_streamout(struct r600_context
*ctx
);
840 void evergreen_set_streamout_enable(struct r600_context
*ctx
, unsigned buffer_enable_bit
);
841 void evergreen_dma_copy(struct r600_context
*rctx
,
842 struct pipe_resource
*dst
,
843 struct pipe_resource
*src
,
847 boolean
evergreen_dma_blit(struct pipe_context
*ctx
,
848 struct pipe_resource
*dst
,
850 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
851 struct pipe_resource
*src
,
853 const struct pipe_box
*src_box
);
855 /* r600_state_common.c */
856 void r600_init_common_state_functions(struct r600_context
*rctx
);
857 void r600_emit_cso_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
858 void r600_emit_alphatest_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
859 void r600_emit_blend_color(struct r600_context
*rctx
, struct r600_atom
*atom
);
860 void r600_emit_vgt_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
861 void r600_emit_clip_misc_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
862 void r600_emit_stencil_ref(struct r600_context
*rctx
, struct r600_atom
*atom
);
863 void r600_emit_viewport_state(struct r600_context
*rctx
, struct r600_atom
*atom
);
864 void r600_emit_shader(struct r600_context
*rctx
, struct r600_atom
*a
);
865 void r600_init_atom(struct r600_context
*rctx
, struct r600_atom
*atom
, unsigned id
,
866 void (*emit
)(struct r600_context
*ctx
, struct r600_atom
*state
),
868 void r600_vertex_buffers_dirty(struct r600_context
*rctx
);
869 void r600_sampler_views_dirty(struct r600_context
*rctx
,
870 struct r600_samplerview_state
*state
);
871 void r600_sampler_states_dirty(struct r600_context
*rctx
,
872 struct r600_sampler_states
*state
);
873 void r600_constant_buffers_dirty(struct r600_context
*rctx
, struct r600_constbuf_state
*state
);
874 void r600_streamout_buffers_dirty(struct r600_context
*rctx
);
875 void r600_draw_rectangle(struct blitter_context
*blitter
,
876 int x1
, int y1
, int x2
, int y2
, float depth
,
877 enum blitter_attrib_type type
, const union pipe_color_union
*attrib
);
878 uint32_t r600_translate_stencil_op(int s_op
);
879 uint32_t r600_translate_fill(uint32_t func
);
880 unsigned r600_tex_wrap(unsigned wrap
);
881 unsigned r600_tex_filter(unsigned filter
);
882 unsigned r600_tex_mipfilter(unsigned filter
);
883 unsigned r600_tex_compare(unsigned compare
);
884 bool sampler_state_needs_border_color(const struct pipe_sampler_state
*state
);
887 struct pipe_video_decoder
*r600_uvd_create_decoder(struct pipe_context
*context
,
888 enum pipe_video_profile profile
,
889 enum pipe_video_entrypoint entrypoint
,
890 enum pipe_video_chroma_format chroma_format
,
891 unsigned width
, unsigned height
,
892 unsigned max_references
, bool expect_chunked_decode
);
894 struct pipe_video_buffer
*r600_video_buffer_create(struct pipe_context
*pipe
,
895 const struct pipe_video_buffer
*tmpl
);
897 int r600_uvd_get_video_param(struct pipe_screen
*screen
,
898 enum pipe_video_profile profile
,
899 enum pipe_video_cap param
);
902 * Helpers for building command buffers
905 #define PKT3_SET_CONFIG_REG 0x68
906 #define PKT3_SET_CONTEXT_REG 0x69
907 #define PKT3_SET_CTL_CONST 0x6F
908 #define PKT3_SET_LOOP_CONST 0x6C
910 #define R600_CONFIG_REG_OFFSET 0x08000
911 #define R600_CONTEXT_REG_OFFSET 0x28000
912 #define R600_CTL_CONST_OFFSET 0x3CFF0
913 #define R600_LOOP_CONST_OFFSET 0X0003E200
914 #define EG_LOOP_CONST_OFFSET 0x0003A200
916 #define PKT_TYPE_S(x) (((x) & 0x3) << 30)
917 #define PKT_COUNT_S(x) (((x) & 0x3FFF) << 16)
918 #define PKT3_IT_OPCODE_S(x) (((x) & 0xFF) << 8)
919 #define PKT3_PREDICATE(x) (((x) >> 0) & 0x1)
920 #define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT_COUNT_S(count) | PKT3_IT_OPCODE_S(op) | PKT3_PREDICATE(predicate))
922 #define RADEON_CP_PACKET3_COMPUTE_MODE 0x00000002
924 /*Evergreen Compute packet3*/
925 #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)
927 static INLINE
void r600_store_value(struct r600_command_buffer
*cb
, unsigned value
)
929 cb
->buf
[cb
->num_dw
++] = value
;
932 static INLINE
void r600_store_array(struct r600_command_buffer
*cb
, unsigned num
, unsigned *ptr
)
934 assert(cb
->num_dw
+num
<= cb
->max_num_dw
);
935 memcpy(&cb
->buf
[cb
->num_dw
], ptr
, num
* sizeof(ptr
[0]));
939 static INLINE
void r600_store_config_reg_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
941 assert(reg
< R600_CONTEXT_REG_OFFSET
);
942 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
943 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_CONFIG_REG
, num
, 0);
944 cb
->buf
[cb
->num_dw
++] = (reg
- R600_CONFIG_REG_OFFSET
) >> 2;
948 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
951 static INLINE
void r600_store_context_reg_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
953 assert(reg
>= R600_CONTEXT_REG_OFFSET
&& reg
< R600_CTL_CONST_OFFSET
);
954 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
955 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_CONTEXT_REG
, num
, 0) | cb
->pkt_flags
;
956 cb
->buf
[cb
->num_dw
++] = (reg
- R600_CONTEXT_REG_OFFSET
) >> 2;
960 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
963 static INLINE
void r600_store_ctl_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
965 assert(reg
>= R600_CTL_CONST_OFFSET
);
966 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
967 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_CTL_CONST
, num
, 0) | cb
->pkt_flags
;
968 cb
->buf
[cb
->num_dw
++] = (reg
- R600_CTL_CONST_OFFSET
) >> 2;
971 static INLINE
void r600_store_loop_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
973 assert(reg
>= R600_LOOP_CONST_OFFSET
);
974 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
975 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_LOOP_CONST
, num
, 0);
976 cb
->buf
[cb
->num_dw
++] = (reg
- R600_LOOP_CONST_OFFSET
) >> 2;
980 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
983 static INLINE
void eg_store_loop_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
985 assert(reg
>= EG_LOOP_CONST_OFFSET
);
986 assert(cb
->num_dw
+2+num
<= cb
->max_num_dw
);
987 cb
->buf
[cb
->num_dw
++] = PKT3(PKT3_SET_LOOP_CONST
, num
, 0) | cb
->pkt_flags
;
988 cb
->buf
[cb
->num_dw
++] = (reg
- EG_LOOP_CONST_OFFSET
) >> 2;
991 static INLINE
void r600_store_config_reg(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
993 r600_store_config_reg_seq(cb
, reg
, 1);
994 r600_store_value(cb
, value
);
997 static INLINE
void r600_store_context_reg(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
999 r600_store_context_reg_seq(cb
, reg
, 1);
1000 r600_store_value(cb
, value
);
1003 static INLINE
void r600_store_ctl_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
1005 r600_store_ctl_const_seq(cb
, reg
, 1);
1006 r600_store_value(cb
, value
);
1009 static INLINE
void r600_store_loop_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
1011 r600_store_loop_const_seq(cb
, reg
, 1);
1012 r600_store_value(cb
, value
);
1015 static INLINE
void eg_store_loop_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
1017 eg_store_loop_const_seq(cb
, reg
, 1);
1018 r600_store_value(cb
, value
);
1021 void r600_init_command_buffer(struct r600_command_buffer
*cb
, unsigned num_dw
);
1022 void r600_release_command_buffer(struct r600_command_buffer
*cb
);
1025 * Helpers for emitting state into a command stream directly.
1027 static INLINE
unsigned r600_context_bo_reloc(struct r600_context
*ctx
,
1028 struct r600_ring
*ring
,
1029 struct r600_resource
*rbo
,
1030 enum radeon_bo_usage usage
)
1033 /* make sure that all previous ring use are flushed so everything
1034 * look serialized from driver pov
1036 if (!ring
->flushing
) {
1037 if (ring
== &ctx
->rings
.gfx
) {
1038 if (ctx
->rings
.dma
.cs
) {
1039 /* flush dma ring */
1040 ctx
->rings
.dma
.flush(ctx
, RADEON_FLUSH_ASYNC
);
1043 /* flush gfx ring */
1044 ctx
->rings
.gfx
.flush(ctx
, RADEON_FLUSH_ASYNC
);
1047 return ctx
->ws
->cs_add_reloc(ring
->cs
, rbo
->cs_buf
, usage
, rbo
->domains
) * 4;
1050 static INLINE
void r600_write_value(struct radeon_winsys_cs
*cs
, unsigned value
)
1052 cs
->buf
[cs
->cdw
++] = value
;
1055 static INLINE
void r600_write_array(struct radeon_winsys_cs
*cs
, unsigned num
, unsigned *ptr
)
1057 assert(cs
->cdw
+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1058 memcpy(&cs
->buf
[cs
->cdw
], ptr
, num
* sizeof(ptr
[0]));
1062 static INLINE
void r600_write_config_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1064 assert(reg
< R600_CONTEXT_REG_OFFSET
);
1065 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1066 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CONFIG_REG
, num
, 0);
1067 cs
->buf
[cs
->cdw
++] = (reg
- R600_CONFIG_REG_OFFSET
) >> 2;
1070 static INLINE
void r600_write_context_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1072 assert(reg
>= R600_CONTEXT_REG_OFFSET
&& reg
< R600_CTL_CONST_OFFSET
);
1073 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1074 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CONTEXT_REG
, num
, 0);
1075 cs
->buf
[cs
->cdw
++] = (reg
- R600_CONTEXT_REG_OFFSET
) >> 2;
1078 static INLINE
void r600_write_compute_context_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1080 r600_write_context_reg_seq(cs
, reg
, num
);
1081 /* Set the compute bit on the packet header */
1082 cs
->buf
[cs
->cdw
- 2] |= RADEON_CP_PACKET3_COMPUTE_MODE
;
1085 static INLINE
void r600_write_ctl_const_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
1087 assert(reg
>= R600_CTL_CONST_OFFSET
);
1088 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
1089 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CTL_CONST
, num
, 0);
1090 cs
->buf
[cs
->cdw
++] = (reg
- R600_CTL_CONST_OFFSET
) >> 2;
1093 static INLINE
void r600_write_config_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1095 r600_write_config_reg_seq(cs
, reg
, 1);
1096 r600_write_value(cs
, value
);
1099 static INLINE
void r600_write_context_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1101 r600_write_context_reg_seq(cs
, reg
, 1);
1102 r600_write_value(cs
, value
);
1105 static INLINE
void r600_write_compute_context_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1107 r600_write_compute_context_reg_seq(cs
, reg
, 1);
1108 r600_write_value(cs
, value
);
1111 static INLINE
void r600_write_context_reg_flag(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
, unsigned flag
)
1113 if (flag
& RADEON_CP_PACKET3_COMPUTE_MODE
) {
1114 r600_write_compute_context_reg(cs
, reg
, value
);
1116 r600_write_context_reg(cs
, reg
, value
);
1120 static INLINE
void r600_write_ctl_const(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
1122 r600_write_ctl_const_seq(cs
, reg
, 1);
1123 r600_write_value(cs
, value
);
1129 static INLINE
uint32_t S_FIXED(float value
, uint32_t frac_bits
)
1131 return value
* (1 << frac_bits
);
1133 #define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
1135 static inline unsigned r600_tex_aniso_filter(unsigned filter
)
1137 if (filter
<= 1) return 0;
1138 if (filter
<= 2) return 1;
1139 if (filter
<= 4) return 2;
1140 if (filter
<= 8) return 3;
1141 /* else */ return 4;
1144 /* 12.4 fixed-point */
1145 static INLINE
unsigned r600_pack_float_12p4(float x
)
1148 x
>= 4096 ? 0xffff : x
* 16;
1151 static INLINE
uint64_t r600_resource_va(struct pipe_screen
*screen
, struct pipe_resource
*resource
)
1153 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
1154 struct r600_resource
*rresource
= (struct r600_resource
*)resource
;
1156 return rscreen
->ws
->buffer_get_virtual_address(rresource
->cs_buf
);
1159 static INLINE
void r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
)
1161 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
1162 struct r600_resource
*rr
= (struct r600_resource
*)r
;
1169 * The idea is to compute a gross estimate of memory requirement of
1170 * each draw call. After each draw call, memory will be precisely
1171 * accounted. So the uncertainty is only on the current draw call.
1172 * In practice this gave very good estimate (+/- 10% of the target
1175 if (rr
->domains
& RADEON_DOMAIN_GTT
) {
1176 rctx
->gtt
+= rr
->buf
->size
;
1178 if (rr
->domains
& RADEON_DOMAIN_VRAM
) {
1179 rctx
->vram
+= rr
->buf
->size
;