2 * Copyright 2013 Advanced Micro Devices, Inc.
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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * Authors: Marek Olšák <maraeo@gmail.com>
28 * This file contains common screen and context structures and functions
29 * for r600g and radeonsi.
32 #ifndef R600_PIPE_COMMON_H
33 #define R600_PIPE_COMMON_H
37 #include "radeon/radeon_winsys.h"
39 #include "util/disk_cache.h"
40 #include "util/u_blitter.h"
41 #include "util/list.h"
42 #include "util/u_range.h"
43 #include "util/slab.h"
44 #include "util/u_suballoc.h"
45 #include "util/u_transfer.h"
47 #define ATI_VENDOR_ID 0x1002
49 #define R600_RESOURCE_FLAG_TRANSFER (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
50 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
51 #define R600_RESOURCE_FLAG_FORCE_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
52 #define R600_RESOURCE_FLAG_DISABLE_DCC (PIPE_RESOURCE_FLAG_DRV_PRIV << 3)
53 #define R600_RESOURCE_FLAG_UNMAPPABLE (PIPE_RESOURCE_FLAG_DRV_PRIV << 4)
55 #define R600_CONTEXT_STREAMOUT_FLUSH (1u << 0)
56 /* Pipeline & streamout query controls. */
57 #define R600_CONTEXT_START_PIPELINE_STATS (1u << 1)
58 #define R600_CONTEXT_STOP_PIPELINE_STATS (1u << 2)
59 #define R600_CONTEXT_PRIVATE_FLAG (1u << 3)
61 /* special primitive types */
62 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
66 #define DBG_TEX (1 << 0)
68 #define DBG_COMPUTE (1 << 2)
69 #define DBG_VM (1 << 3)
72 #define DBG_FS (1 << 5)
73 #define DBG_VS (1 << 6)
74 #define DBG_GS (1 << 7)
75 #define DBG_PS (1 << 8)
76 #define DBG_CS (1 << 9)
77 #define DBG_TCS (1 << 10)
78 #define DBG_TES (1 << 11)
79 #define DBG_NO_IR (1 << 12)
80 #define DBG_NO_TGSI (1 << 13)
81 #define DBG_NO_ASM (1 << 14)
82 #define DBG_PREOPT_IR (1 << 15)
83 #define DBG_CHECK_IR (1 << 16)
84 #define DBG_NO_OPT_VARIANT (1 << 17)
86 #define DBG_TEST_DMA (1 << 20)
87 /* Bits 21-31 are reserved for the r600g driver. */
89 #define DBG_NO_ASYNC_DMA (1llu << 32)
90 #define DBG_NO_HYPERZ (1llu << 33)
91 #define DBG_NO_DISCARD_RANGE (1llu << 34)
92 #define DBG_NO_2D_TILING (1llu << 35)
93 #define DBG_NO_TILING (1llu << 36)
94 #define DBG_SWITCH_ON_EOP (1llu << 37)
95 #define DBG_FORCE_DMA (1llu << 38)
96 #define DBG_PRECOMPILE (1llu << 39)
97 #define DBG_INFO (1llu << 40)
98 #define DBG_NO_WC (1llu << 41)
99 #define DBG_CHECK_VM (1llu << 42)
100 #define DBG_NO_DCC (1llu << 43)
101 #define DBG_NO_DCC_CLEAR (1llu << 44)
102 #define DBG_NO_RB_PLUS (1llu << 45)
103 #define DBG_SI_SCHED (1llu << 46)
104 #define DBG_MONOLITHIC_SHADERS (1llu << 47)
105 #define DBG_NO_CE (1llu << 48)
106 #define DBG_UNSAFE_MATH (1llu << 49)
107 #define DBG_NO_DCC_FB (1llu << 50)
109 #define R600_MAP_BUFFER_ALIGNMENT 64
110 #define R600_MAX_VIEWPORTS 16
112 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 1024
114 enum r600_coherency
{
115 R600_COHERENCY_NONE
, /* no cache flushes needed */
116 R600_COHERENCY_SHADER
,
117 R600_COHERENCY_CB_META
,
120 #ifdef PIPE_ARCH_BIG_ENDIAN
121 #define R600_BIG_ENDIAN 1
123 #define R600_BIG_ENDIAN 0
126 struct r600_common_context
;
127 struct r600_perfcounters
;
128 struct tgsi_shader_info
;
129 struct r600_qbo_state
;
131 struct radeon_shader_reloc
{
136 struct radeon_shader_binary
{
141 /** Config/Context register state that accompanies this shader.
142 * This is a stream of dword pairs. First dword contains the
143 * register address, the second dword contains the value.*/
144 unsigned char *config
;
145 unsigned config_size
;
147 /** The number of bytes of config information for each global symbol.
149 unsigned config_size_per_symbol
;
151 /** Constant data accessed by the shader. This will be uploaded
152 * into a constant buffer. */
153 unsigned char *rodata
;
154 unsigned rodata_size
;
156 /** List of symbol offsets for the shader */
157 uint64_t *global_symbol_offsets
;
158 unsigned global_symbol_count
;
160 struct radeon_shader_reloc
*relocs
;
161 unsigned reloc_count
;
163 /** Disassembled shader in a string. */
165 char *llvm_ir_string
;
168 void radeon_shader_binary_init(struct radeon_shader_binary
*b
);
169 void radeon_shader_binary_clean(struct radeon_shader_binary
*b
);
171 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
174 struct r600_resource
{
177 /* Winsys objects. */
178 struct pb_buffer
*buf
;
179 uint64_t gpu_address
;
180 /* Memory usage if the buffer placement is optimal. */
184 /* Resource properties. */
186 unsigned bo_alignment
;
187 enum radeon_bo_domain domains
;
188 enum radeon_bo_flag flags
;
189 unsigned bind_history
;
191 /* The buffer range which is initialized (with a write transfer,
192 * streamout, DMA, or as a random access target). The rest of
193 * the buffer is considered invalid and can be mapped unsynchronized.
195 * This allows unsychronized mapping of a buffer range which hasn't
196 * been used yet. It's for applications which forget to use
197 * the unsynchronized map flag and expect the driver to figure it out.
199 struct util_range valid_buffer_range
;
201 /* For buffers only. This indicates that a write operation has been
202 * performed by TC L2, but the cache hasn't been flushed.
203 * Any hw block which doesn't use or bypasses TC L2 should check this
204 * flag and flush the cache before using the buffer.
206 * For example, TC L2 must be flushed if a buffer which has been
207 * modified by a shader store instruction is about to be used as
208 * an index buffer. The reason is that VGT DMA index fetching doesn't
213 /* Whether the resource has been exported via resource_get_handle. */
215 unsigned external_usage
; /* PIPE_HANDLE_USAGE_* */
218 struct r600_transfer
{
219 struct pipe_transfer transfer
;
220 struct r600_resource
*staging
;
224 struct r600_fmask_info
{
228 unsigned pitch_in_pixels
;
229 unsigned bank_height
;
230 unsigned slice_tile_max
;
231 unsigned tile_mode_index
;
234 struct r600_cmask_info
{
238 unsigned slice_tile_max
;
239 unsigned base_address_reg
;
242 struct r600_texture
{
243 struct r600_resource resource
;
246 unsigned num_level0_transfers
;
247 enum pipe_format db_render_format
;
252 unsigned dirty_level_mask
; /* each bit says if that mipmap is compressed */
253 unsigned stencil_dirty_level_mask
; /* each bit says if that mipmap is compressed */
254 struct r600_texture
*flushed_depth_texture
;
255 struct radeon_surf surface
;
257 /* Colorbuffer compression and fast clear. */
258 struct r600_fmask_info fmask
;
259 struct r600_cmask_info cmask
;
260 struct r600_resource
*cmask_buffer
;
261 uint64_t dcc_offset
; /* 0 = disabled */
262 unsigned cb_color_info
; /* fast clear enable bit */
263 unsigned color_clear_value
[2];
264 unsigned last_msaa_resolve_target_micro_mode
;
266 /* Depth buffer compression and fast clear. */
267 struct r600_resource
*htile_buffer
;
268 bool tc_compatible_htile
;
269 bool depth_cleared
; /* if it was cleared at least once */
270 float depth_clear_value
;
271 bool stencil_cleared
; /* if it was cleared at least once */
272 uint8_t stencil_clear_value
;
274 bool non_disp_tiling
; /* R600-Cayman only */
276 /* Whether the texture is a displayable back buffer and needs DCC
277 * decompression, which is expensive. Therefore, it's enabled only
278 * if statistics suggest that it will pay off and it's allocated
279 * separately. It can't be bound as a sampler by apps. Limited to
280 * target == 2D and last_level == 0. If enabled, dcc_offset contains
281 * the absolute GPUVM address, not the relative one.
283 struct r600_resource
*dcc_separate_buffer
;
284 /* When DCC is temporarily disabled, the separate buffer is here. */
285 struct r600_resource
*last_dcc_separate_buffer
;
286 /* We need to track DCC dirtiness, because st/dri usually calls
287 * flush_resource twice per frame (not a bug) and we don't wanna
288 * decompress DCC twice. Also, the dirty tracking must be done even
289 * if DCC isn't used, because it's required by the DCC usage analysis
290 * for a possible future enablement.
292 bool separate_dcc_dirty
;
293 /* Statistics gathering for the DCC enablement heuristic. */
294 bool dcc_gather_statistics
;
295 /* Estimate of how much this color buffer is written to in units of
296 * full-screen draws: ps_invocations / (width * height)
297 * Shader kills, late Z, and blending with trivial discards make it
298 * inaccurate (we need to count CB updates, not PS invocations).
300 unsigned ps_draw_ratio
;
301 /* The number of clears since the last DCC usage analysis. */
302 unsigned num_slow_clears
;
304 /* Counter that should be non-zero if the texture is bound to a
305 * framebuffer. Implemented in radeonsi only.
307 uint32_t framebuffers_bound
;
310 struct r600_surface
{
311 struct pipe_surface base
;
313 bool color_initialized
;
314 bool depth_initialized
;
316 /* Misc. color flags. */
317 bool alphatest_bypass
;
322 /* Color registers. */
323 unsigned cb_color_info
;
324 unsigned cb_color_base
;
325 unsigned cb_color_view
;
326 unsigned cb_color_size
; /* R600 only */
327 unsigned cb_color_dim
; /* EG only */
328 unsigned cb_color_pitch
; /* EG and later */
329 unsigned cb_color_slice
; /* EG and later */
330 unsigned cb_color_attrib
; /* EG and later */
331 unsigned cb_dcc_control
; /* VI and later */
332 unsigned cb_color_fmask
; /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
333 unsigned cb_color_fmask_slice
; /* EG and later */
334 unsigned cb_color_cmask
; /* CB_COLORn_TILE (r600 only) */
335 unsigned cb_color_mask
; /* R600 only */
336 unsigned spi_shader_col_format
; /* SI+, no blending, no alpha-to-coverage. */
337 unsigned spi_shader_col_format_alpha
; /* SI+, alpha-to-coverage */
338 unsigned spi_shader_col_format_blend
; /* SI+, blending without alpha. */
339 unsigned spi_shader_col_format_blend_alpha
; /* SI+, blending with alpha. */
340 struct r600_resource
*cb_buffer_fmask
; /* Used for FMASK relocations. R600 only */
341 struct r600_resource
*cb_buffer_cmask
; /* Used for CMASK relocations. R600 only */
344 unsigned db_depth_info
; /* R600 only, then SI and later */
345 unsigned db_z_info
; /* EG and later */
346 unsigned db_depth_base
; /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
347 unsigned db_depth_view
;
348 unsigned db_depth_size
;
349 unsigned db_depth_slice
; /* EG and later */
350 unsigned db_stencil_base
; /* EG and later */
351 unsigned db_stencil_info
; /* EG and later */
352 unsigned db_prefetch_limit
; /* R600 only */
353 unsigned db_htile_surface
;
354 unsigned db_htile_data_base
;
355 unsigned db_preload_control
; /* EG and later */
358 struct r600_mmio_counter
{
363 union r600_mmio_counters
{
365 /* For global GPU load including SDMA. */
366 struct r600_mmio_counter gpu
;
369 struct r600_mmio_counter spi
;
370 struct r600_mmio_counter gui
;
371 struct r600_mmio_counter ta
;
372 struct r600_mmio_counter gds
;
373 struct r600_mmio_counter vgt
;
374 struct r600_mmio_counter ia
;
375 struct r600_mmio_counter sx
;
376 struct r600_mmio_counter wd
;
377 struct r600_mmio_counter bci
;
378 struct r600_mmio_counter sc
;
379 struct r600_mmio_counter pa
;
380 struct r600_mmio_counter db
;
381 struct r600_mmio_counter cp
;
382 struct r600_mmio_counter cb
;
385 struct r600_mmio_counter sdma
;
388 struct r600_mmio_counter pfp
;
389 struct r600_mmio_counter meq
;
390 struct r600_mmio_counter me
;
391 struct r600_mmio_counter surf_sync
;
392 struct r600_mmio_counter dma
;
393 struct r600_mmio_counter scratch_ram
;
394 struct r600_mmio_counter ce
;
399 struct r600_common_screen
{
400 struct pipe_screen b
;
401 struct radeon_winsys
*ws
;
402 enum radeon_family family
;
403 enum chip_class chip_class
;
404 struct radeon_info info
;
405 uint64_t debug_flags
;
409 struct disk_cache
*disk_shader_cache
;
411 struct slab_parent_pool pool_transfers
;
413 /* Texture filter settings. */
414 int force_aniso
; /* -1 = disabled */
416 /* Auxiliary context. Mainly used to initialize resources.
417 * It must be locked prior to using and flushed before unlocking. */
418 struct pipe_context
*aux_context
;
419 pipe_mutex aux_context_lock
;
421 /* This must be in the screen, because UE4 uses one context for
422 * compilation and another one for rendering.
424 unsigned num_compilations
;
425 /* Along with ST_DEBUG=precompile, this should show if applications
426 * are loading shaders on demand. This is a monotonic counter.
428 unsigned num_shaders_created
;
429 unsigned num_shader_cache_hits
;
431 /* GPU load thread. */
432 pipe_mutex gpu_load_mutex
;
433 pipe_thread gpu_load_thread
;
434 union r600_mmio_counters mmio_counters
;
435 volatile unsigned gpu_load_stop_thread
; /* bool */
437 char renderer_string
[100];
439 /* Performance counters. */
440 struct r600_perfcounters
*perfcounters
;
442 /* If pipe_screen wants to recompute and re-emit the framebuffer,
443 * sampler, and image states of all contexts, it should atomically
446 * Each context will compare this with its own last known value of
447 * the counter before drawing and re-emit the states accordingly.
449 unsigned dirty_tex_counter
;
451 /* Atomically increment this counter when an existing texture's
452 * metadata is enabled or disabled in a way that requires changing
453 * contexts' compressed texture binding masks.
455 unsigned compressed_colortex_counter
;
458 /* Context flags to set so that all writes from earlier jobs
459 * in the CP are seen by L2 clients.
463 /* Context flags to set so that all writes from earlier
464 * compute jobs are seen by L2 clients.
466 unsigned compute_to_L2
;
469 void (*query_opaque_metadata
)(struct r600_common_screen
*rscreen
,
470 struct r600_texture
*rtex
,
471 struct radeon_bo_metadata
*md
);
473 void (*apply_opaque_metadata
)(struct r600_common_screen
*rscreen
,
474 struct r600_texture
*rtex
,
475 struct radeon_bo_metadata
*md
);
478 /* This encapsulates a state or an operation which can emitted into the GPU
481 void (*emit
)(struct r600_common_context
*ctx
, struct r600_atom
*state
);
486 struct r600_so_target
{
487 struct pipe_stream_output_target b
;
489 /* The buffer where BUFFER_FILLED_SIZE is stored. */
490 struct r600_resource
*buf_filled_size
;
491 unsigned buf_filled_size_offset
;
492 bool buf_filled_size_valid
;
494 unsigned stride_in_dw
;
497 struct r600_streamout
{
498 struct r600_atom begin_atom
;
500 unsigned num_dw_for_end
;
502 unsigned enabled_mask
;
503 unsigned num_targets
;
504 struct r600_so_target
*targets
[PIPE_MAX_SO_BUFFERS
];
506 unsigned append_bitmask
;
509 /* External state which comes from the vertex shader,
510 * it must be set explicitly when binding a shader. */
511 unsigned *stride_in_dw
;
512 unsigned enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
514 /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
515 unsigned hw_enabled_mask
;
517 /* The state of VGT_STRMOUT_(CONFIG|EN). */
518 struct r600_atom enable_atom
;
519 bool streamout_enabled
;
520 bool prims_gen_query_enabled
;
521 int num_prims_gen_queries
;
524 struct r600_signed_scissor
{
531 struct r600_scissors
{
532 struct r600_atom atom
;
534 struct pipe_scissor_state states
[R600_MAX_VIEWPORTS
];
537 struct r600_viewports
{
538 struct r600_atom atom
;
540 unsigned depth_range_dirty_mask
;
541 struct pipe_viewport_state states
[R600_MAX_VIEWPORTS
];
542 struct r600_signed_scissor as_scissor
[R600_MAX_VIEWPORTS
];
546 struct radeon_winsys_cs
*cs
;
547 void (*flush
)(void *ctx
, unsigned flags
,
548 struct pipe_fence_handle
**fence
);
551 /* Saved CS data for debugging features. */
552 struct radeon_saved_cs
{
556 struct radeon_bo_list_item
*bo_list
;
560 struct r600_common_context
{
561 struct pipe_context b
; /* base class */
563 struct r600_common_screen
*screen
;
564 struct radeon_winsys
*ws
;
565 struct radeon_winsys_ctx
*ctx
;
566 enum radeon_family family
;
567 enum chip_class chip_class
;
568 struct r600_ring gfx
;
569 struct r600_ring dma
;
570 struct pipe_fence_handle
*last_gfx_fence
;
571 struct pipe_fence_handle
*last_sdma_fence
;
572 unsigned num_gfx_cs_flushes
;
573 unsigned initial_gfx_cs_size
;
574 unsigned gpu_reset_counter
;
575 unsigned last_dirty_tex_counter
;
576 unsigned last_compressed_colortex_counter
;
578 struct u_suballocator
*allocator_zeroed_memory
;
579 struct slab_child_pool pool_transfers
;
581 /* Current unaccounted memory usage. */
586 struct r600_streamout streamout
;
587 struct r600_scissors scissors
;
588 struct r600_viewports viewports
;
589 bool scissor_enabled
;
591 bool vs_writes_viewport_index
;
592 bool vs_disables_clipping_viewport
;
594 /* Additional context states. */
595 unsigned flags
; /* flush flags */
598 /* Maintain the list of active queries for pausing between IBs. */
599 int num_occlusion_queries
;
600 int num_perfect_occlusion_queries
;
601 struct list_head active_queries
;
602 unsigned num_cs_dw_queries_suspend
;
604 unsigned num_draw_calls
;
605 unsigned num_spill_draw_calls
;
606 unsigned num_compute_calls
;
607 unsigned num_spill_compute_calls
;
608 unsigned num_dma_calls
;
609 unsigned num_cp_dma_calls
;
610 unsigned num_vs_flushes
;
611 unsigned num_ps_flushes
;
612 unsigned num_cs_flushes
;
613 unsigned num_fb_cache_flushes
;
614 unsigned num_L2_invalidates
;
615 unsigned num_L2_writebacks
;
616 uint64_t num_alloc_tex_transfer_bytes
;
617 unsigned last_tex_ps_draw_ratio
; /* for query */
619 /* Render condition. */
620 struct r600_atom render_cond_atom
;
621 struct pipe_query
*render_cond
;
622 unsigned render_cond_mode
;
623 bool render_cond_invert
;
624 bool render_cond_force_off
; /* for u_blitter */
626 /* MSAA sample locations.
627 * The first index is the sample index.
628 * The second index is the coordinate: X, Y. */
629 float sample_locations_1x
[1][2];
630 float sample_locations_2x
[2][2];
631 float sample_locations_4x
[4][2];
632 float sample_locations_8x
[8][2];
633 float sample_locations_16x
[16][2];
635 /* Statistics gathering for the DCC enablement heuristic. It can't be
636 * in r600_texture because r600_texture can be shared by multiple
637 * contexts. This is for back buffers only. We shouldn't get too many
640 * X11 DRI3 rotates among a finite set of back buffers. They should
641 * all fit in this array. If they don't, separate DCC might never be
642 * enabled by DCC stat gathering.
645 struct r600_texture
*tex
;
646 /* Query queue: 0 = usually active, 1 = waiting, 2 = readback. */
647 struct pipe_query
*ps_stats
[3];
648 /* If all slots are used and another slot is needed,
649 * the least recently used slot is evicted based on this. */
650 int64_t last_use_timestamp
;
654 struct pipe_debug_callback debug
;
655 struct pipe_device_reset_callback device_reset_callback
;
657 void *query_result_shader
;
659 /* Copy one resource to another using async DMA. */
660 void (*dma_copy
)(struct pipe_context
*ctx
,
661 struct pipe_resource
*dst
,
663 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
664 struct pipe_resource
*src
,
666 const struct pipe_box
*src_box
);
668 void (*dma_clear_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*dst
,
669 uint64_t offset
, uint64_t size
, unsigned value
);
671 void (*clear_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*dst
,
672 uint64_t offset
, uint64_t size
, unsigned value
,
673 enum r600_coherency coher
);
675 void (*blit_decompress_depth
)(struct pipe_context
*ctx
,
676 struct r600_texture
*texture
,
677 struct r600_texture
*staging
,
678 unsigned first_level
, unsigned last_level
,
679 unsigned first_layer
, unsigned last_layer
,
680 unsigned first_sample
, unsigned last_sample
);
682 void (*decompress_dcc
)(struct pipe_context
*ctx
,
683 struct r600_texture
*rtex
);
685 /* Reallocate the buffer and update all resource bindings where
686 * the buffer is bound, including all resource descriptors. */
687 void (*invalidate_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*buf
);
689 /* Enable or disable occlusion queries. */
690 void (*set_occlusion_query_state
)(struct pipe_context
*ctx
, bool enable
);
692 void (*save_qbo_state
)(struct pipe_context
*ctx
, struct r600_qbo_state
*st
);
694 /* This ensures there is enough space in the command stream. */
695 void (*need_gfx_cs_space
)(struct pipe_context
*ctx
, unsigned num_dw
,
696 bool include_draw_vbo
);
698 void (*set_atom_dirty
)(struct r600_common_context
*ctx
,
699 struct r600_atom
*atom
, bool dirty
);
701 void (*check_vm_faults
)(struct r600_common_context
*ctx
,
702 struct radeon_saved_cs
*saved
,
703 enum ring_type ring
);
707 bool r600_rings_is_buffer_referenced(struct r600_common_context
*ctx
,
708 struct pb_buffer
*buf
,
709 enum radeon_bo_usage usage
);
710 void *r600_buffer_map_sync_with_rings(struct r600_common_context
*ctx
,
711 struct r600_resource
*resource
,
713 void r600_buffer_subdata(struct pipe_context
*ctx
,
714 struct pipe_resource
*buffer
,
715 unsigned usage
, unsigned offset
,
716 unsigned size
, const void *data
);
717 void r600_init_resource_fields(struct r600_common_screen
*rscreen
,
718 struct r600_resource
*res
,
719 uint64_t size
, unsigned alignment
);
720 bool r600_alloc_resource(struct r600_common_screen
*rscreen
,
721 struct r600_resource
*res
);
722 struct pipe_resource
*r600_buffer_create(struct pipe_screen
*screen
,
723 const struct pipe_resource
*templ
,
725 struct pipe_resource
* r600_aligned_buffer_create(struct pipe_screen
*screen
,
730 struct pipe_resource
*
731 r600_buffer_from_user_memory(struct pipe_screen
*screen
,
732 const struct pipe_resource
*templ
,
735 r600_invalidate_resource(struct pipe_context
*ctx
,
736 struct pipe_resource
*resource
);
738 /* r600_common_pipe.c */
739 void r600_gfx_write_event_eop(struct r600_common_context
*ctx
,
740 unsigned event
, unsigned event_flags
,
742 struct r600_resource
*buf
, uint64_t va
,
743 uint32_t old_fence
, uint32_t new_fence
);
744 unsigned r600_gfx_write_fence_dwords(struct r600_common_screen
*screen
);
745 void r600_gfx_wait_fence(struct r600_common_context
*ctx
,
746 uint64_t va
, uint32_t ref
, uint32_t mask
);
747 void r600_draw_rectangle(struct blitter_context
*blitter
,
748 int x1
, int y1
, int x2
, int y2
, float depth
,
749 enum blitter_attrib_type type
,
750 const union pipe_color_union
*attrib
);
751 bool r600_common_screen_init(struct r600_common_screen
*rscreen
,
752 struct radeon_winsys
*ws
);
753 void r600_destroy_common_screen(struct r600_common_screen
*rscreen
);
754 void r600_preflush_suspend_features(struct r600_common_context
*ctx
);
755 void r600_postflush_resume_features(struct r600_common_context
*ctx
);
756 bool r600_common_context_init(struct r600_common_context
*rctx
,
757 struct r600_common_screen
*rscreen
,
758 unsigned context_flags
);
759 void r600_common_context_cleanup(struct r600_common_context
*rctx
);
760 bool r600_can_dump_shader(struct r600_common_screen
*rscreen
,
762 bool r600_extra_shader_checks(struct r600_common_screen
*rscreen
,
764 void r600_screen_clear_buffer(struct r600_common_screen
*rscreen
, struct pipe_resource
*dst
,
765 uint64_t offset
, uint64_t size
, unsigned value
);
766 struct pipe_resource
*r600_resource_create_common(struct pipe_screen
*screen
,
767 const struct pipe_resource
*templ
);
768 const char *r600_get_llvm_processor_name(enum radeon_family family
);
769 void r600_need_dma_space(struct r600_common_context
*ctx
, unsigned num_dw
,
770 struct r600_resource
*dst
, struct r600_resource
*src
);
771 void radeon_save_cs(struct radeon_winsys
*ws
, struct radeon_winsys_cs
*cs
,
772 struct radeon_saved_cs
*saved
);
773 void radeon_clear_saved_cs(struct radeon_saved_cs
*saved
);
774 bool r600_check_device_reset(struct r600_common_context
*rctx
);
776 /* r600_gpu_load.c */
777 void r600_gpu_load_kill_thread(struct r600_common_screen
*rscreen
);
778 uint64_t r600_begin_counter(struct r600_common_screen
*rscreen
, unsigned type
);
779 unsigned r600_end_counter(struct r600_common_screen
*rscreen
, unsigned type
,
782 /* r600_perfcounters.c */
783 void r600_perfcounters_destroy(struct r600_common_screen
*rscreen
);
786 void r600_init_screen_query_functions(struct r600_common_screen
*rscreen
);
787 void r600_query_init(struct r600_common_context
*rctx
);
788 void r600_suspend_queries(struct r600_common_context
*ctx
);
789 void r600_resume_queries(struct r600_common_context
*ctx
);
790 void r600_query_fix_enabled_rb_mask(struct r600_common_screen
*rscreen
);
792 /* r600_streamout.c */
793 void r600_streamout_buffers_dirty(struct r600_common_context
*rctx
);
794 void r600_set_streamout_targets(struct pipe_context
*ctx
,
795 unsigned num_targets
,
796 struct pipe_stream_output_target
**targets
,
797 const unsigned *offset
);
798 void r600_emit_streamout_end(struct r600_common_context
*rctx
);
799 void r600_update_prims_generated_query_state(struct r600_common_context
*rctx
,
800 unsigned type
, int diff
);
801 void r600_streamout_init(struct r600_common_context
*rctx
);
803 /* r600_test_dma.c */
804 void r600_test_dma(struct r600_common_screen
*rscreen
);
807 bool r600_prepare_for_dma_blit(struct r600_common_context
*rctx
,
808 struct r600_texture
*rdst
,
809 unsigned dst_level
, unsigned dstx
,
810 unsigned dsty
, unsigned dstz
,
811 struct r600_texture
*rsrc
,
813 const struct pipe_box
*src_box
);
814 void r600_texture_get_fmask_info(struct r600_common_screen
*rscreen
,
815 struct r600_texture
*rtex
,
817 struct r600_fmask_info
*out
);
818 void r600_texture_get_cmask_info(struct r600_common_screen
*rscreen
,
819 struct r600_texture
*rtex
,
820 struct r600_cmask_info
*out
);
821 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
822 struct pipe_resource
*texture
,
823 struct r600_texture
**staging
);
824 void r600_print_texture_info(struct r600_texture
*rtex
, FILE *f
);
825 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
826 const struct pipe_resource
*templ
);
827 bool vi_dcc_formats_compatible(enum pipe_format format1
,
828 enum pipe_format format2
);
829 void vi_dcc_disable_if_incompatible_format(struct r600_common_context
*rctx
,
830 struct pipe_resource
*tex
,
832 enum pipe_format view_format
);
833 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
834 struct pipe_resource
*texture
,
835 const struct pipe_surface
*templ
,
836 unsigned width
, unsigned height
);
837 unsigned r600_translate_colorswap(enum pipe_format format
, bool do_endian_swap
);
838 void vi_separate_dcc_start_query(struct pipe_context
*ctx
,
839 struct r600_texture
*tex
);
840 void vi_separate_dcc_stop_query(struct pipe_context
*ctx
,
841 struct r600_texture
*tex
);
842 void vi_separate_dcc_process_and_reset_stats(struct pipe_context
*ctx
,
843 struct r600_texture
*tex
);
844 void vi_dcc_clear_level(struct r600_common_context
*rctx
,
845 struct r600_texture
*rtex
,
846 unsigned level
, unsigned clear_value
);
847 void evergreen_do_fast_color_clear(struct r600_common_context
*rctx
,
848 struct pipe_framebuffer_state
*fb
,
849 struct r600_atom
*fb_state
,
850 unsigned *buffers
, unsigned *dirty_cbufs
,
851 const union pipe_color_union
*color
);
852 bool r600_texture_disable_dcc(struct r600_common_context
*rctx
,
853 struct r600_texture
*rtex
);
854 void r600_init_screen_texture_functions(struct r600_common_screen
*rscreen
);
855 void r600_init_context_texture_functions(struct r600_common_context
*rctx
);
857 /* r600_viewport.c */
858 void evergreen_apply_scissor_bug_workaround(struct r600_common_context
*rctx
,
859 struct pipe_scissor_state
*scissor
);
860 void r600_viewport_set_rast_deps(struct r600_common_context
*rctx
,
861 bool scissor_enable
, bool clip_halfz
);
862 void r600_update_vs_writes_viewport_index(struct r600_common_context
*rctx
,
863 struct tgsi_shader_info
*info
);
864 void r600_init_viewport_functions(struct r600_common_context
*rctx
);
867 extern const uint32_t eg_sample_locs_2x
[4];
868 extern const unsigned eg_max_dist_2x
;
869 extern const uint32_t eg_sample_locs_4x
[4];
870 extern const unsigned eg_max_dist_4x
;
871 void cayman_get_sample_position(struct pipe_context
*ctx
, unsigned sample_count
,
872 unsigned sample_index
, float *out_value
);
873 void cayman_init_msaa(struct pipe_context
*ctx
);
874 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
875 void cayman_emit_msaa_config(struct radeon_winsys_cs
*cs
, int nr_samples
,
876 int ps_iter_samples
, int overrast_samples
,
877 unsigned sc_mode_cntl_1
);
880 /* Inline helpers. */
882 static inline struct r600_resource
*r600_resource(struct pipe_resource
*r
)
884 return (struct r600_resource
*)r
;
888 r600_resource_reference(struct r600_resource
**ptr
, struct r600_resource
*res
)
890 pipe_resource_reference((struct pipe_resource
**)ptr
,
891 (struct pipe_resource
*)res
);
895 r600_texture_reference(struct r600_texture
**ptr
, struct r600_texture
*res
)
897 pipe_resource_reference((struct pipe_resource
**)ptr
, &res
->resource
.b
.b
);
901 r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
)
903 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
904 struct r600_resource
*res
= (struct r600_resource
*)r
;
907 /* Add memory usage for need_gfx_cs_space */
908 rctx
->vram
+= res
->vram_usage
;
909 rctx
->gtt
+= res
->gart_usage
;
913 static inline bool r600_get_strmout_en(struct r600_common_context
*rctx
)
915 return rctx
->streamout
.streamout_enabled
||
916 rctx
->streamout
.prims_gen_query_enabled
;
919 #define SQ_TEX_XY_FILTER_POINT 0x00
920 #define SQ_TEX_XY_FILTER_BILINEAR 0x01
921 #define SQ_TEX_XY_FILTER_ANISO_POINT 0x02
922 #define SQ_TEX_XY_FILTER_ANISO_BILINEAR 0x03
924 static inline unsigned eg_tex_filter(unsigned filter
, unsigned max_aniso
)
926 if (filter
== PIPE_TEX_FILTER_LINEAR
)
927 return max_aniso
> 1 ? SQ_TEX_XY_FILTER_ANISO_BILINEAR
928 : SQ_TEX_XY_FILTER_BILINEAR
;
930 return max_aniso
> 1 ? SQ_TEX_XY_FILTER_ANISO_POINT
931 : SQ_TEX_XY_FILTER_POINT
;
934 static inline unsigned r600_tex_aniso_filter(unsigned filter
)
947 static inline unsigned r600_wavefront_size(enum radeon_family family
)
967 static inline enum radeon_bo_priority
968 r600_get_sampler_view_priority(struct r600_resource
*res
)
970 if (res
->b
.b
.target
== PIPE_BUFFER
)
971 return RADEON_PRIO_SAMPLER_BUFFER
;
973 if (res
->b
.b
.nr_samples
> 1)
974 return RADEON_PRIO_SAMPLER_TEXTURE_MSAA
;
976 return RADEON_PRIO_SAMPLER_TEXTURE
;
980 r600_can_sample_zs(struct r600_texture
*tex
, bool stencil_sampler
)
982 return (stencil_sampler
&& tex
->can_sample_s
) ||
983 (!stencil_sampler
&& tex
->can_sample_z
);
986 #define COMPUTE_DBG(rscreen, fmt, args...) \
988 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
991 #define R600_ERR(fmt, args...) \
992 fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
994 /* For MSAA sample positions. */
995 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
996 (((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) | \
997 (((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) | \
998 (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) | \
999 (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))