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 "amd/common/ac_binary.h"
39 #include "radeon/radeon_winsys.h"
41 #include "util/disk_cache.h"
42 #include "util/u_blitter.h"
43 #include "util/list.h"
44 #include "util/u_range.h"
45 #include "util/slab.h"
46 #include "util/u_suballoc.h"
47 #include "util/u_transfer.h"
48 #include "util/u_threaded_context.h"
50 #define ATI_VENDOR_ID 0x1002
52 #define R600_RESOURCE_FLAG_TRANSFER (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
53 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
54 #define R600_RESOURCE_FLAG_FORCE_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
55 #define R600_RESOURCE_FLAG_DISABLE_DCC (PIPE_RESOURCE_FLAG_DRV_PRIV << 3)
56 #define R600_RESOURCE_FLAG_UNMAPPABLE (PIPE_RESOURCE_FLAG_DRV_PRIV << 4)
58 #define R600_CONTEXT_STREAMOUT_FLUSH (1u << 0)
59 /* Pipeline & streamout query controls. */
60 #define R600_CONTEXT_START_PIPELINE_STATS (1u << 1)
61 #define R600_CONTEXT_STOP_PIPELINE_STATS (1u << 2)
62 #define R600_CONTEXT_PRIVATE_FLAG (1u << 3)
64 /* special primitive types */
65 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
69 #define DBG_TEX (1 << 0)
71 #define DBG_COMPUTE (1 << 2)
72 #define DBG_VM (1 << 3)
75 #define DBG_FS (1 << 5)
76 #define DBG_VS (1 << 6)
77 #define DBG_GS (1 << 7)
78 #define DBG_PS (1 << 8)
79 #define DBG_CS (1 << 9)
80 #define DBG_TCS (1 << 10)
81 #define DBG_TES (1 << 11)
82 #define DBG_NO_IR (1 << 12)
83 #define DBG_NO_TGSI (1 << 13)
84 #define DBG_NO_ASM (1 << 14)
85 #define DBG_PREOPT_IR (1 << 15)
86 #define DBG_CHECK_IR (1 << 16)
87 #define DBG_NO_OPT_VARIANT (1 << 17)
88 #define DBG_FS_CORRECT_DERIVS_AFTER_KILL (1 << 18)
90 #define DBG_TEST_DMA (1 << 20)
91 /* Bits 21-31 are reserved for the r600g driver. */
93 #define DBG_NO_ASYNC_DMA (1llu << 32)
94 #define DBG_NO_HYPERZ (1llu << 33)
95 #define DBG_NO_DISCARD_RANGE (1llu << 34)
96 #define DBG_NO_2D_TILING (1llu << 35)
97 #define DBG_NO_TILING (1llu << 36)
98 #define DBG_SWITCH_ON_EOP (1llu << 37)
99 #define DBG_FORCE_DMA (1llu << 38)
100 #define DBG_PRECOMPILE (1llu << 39)
101 #define DBG_INFO (1llu << 40)
102 #define DBG_NO_WC (1llu << 41)
103 #define DBG_CHECK_VM (1llu << 42)
104 #define DBG_NO_DCC (1llu << 43)
105 #define DBG_NO_DCC_CLEAR (1llu << 44)
106 #define DBG_NO_RB_PLUS (1llu << 45)
107 #define DBG_SI_SCHED (1llu << 46)
108 #define DBG_MONOLITHIC_SHADERS (1llu << 47)
109 #define DBG_NO_CE (1llu << 48)
110 #define DBG_UNSAFE_MATH (1llu << 49)
111 #define DBG_NO_DCC_FB (1llu << 50)
112 #define DBG_TEST_VMFAULT_CP (1llu << 51)
113 #define DBG_TEST_VMFAULT_SDMA (1llu << 52)
114 #define DBG_TEST_VMFAULT_SHADER (1llu << 53)
116 #define R600_MAP_BUFFER_ALIGNMENT 64
117 #define R600_MAX_VIEWPORTS 16
119 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 1024
121 enum r600_coherency
{
122 R600_COHERENCY_NONE
, /* no cache flushes needed */
123 R600_COHERENCY_SHADER
,
124 R600_COHERENCY_CB_META
,
127 #ifdef PIPE_ARCH_BIG_ENDIAN
128 #define R600_BIG_ENDIAN 1
130 #define R600_BIG_ENDIAN 0
133 struct r600_common_context
;
134 struct r600_perfcounters
;
135 struct tgsi_shader_info
;
136 struct r600_qbo_state
;
138 void radeon_shader_binary_init(struct ac_shader_binary
*b
);
139 void radeon_shader_binary_clean(struct ac_shader_binary
*b
);
141 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
144 struct r600_resource
{
145 struct threaded_resource b
;
147 /* Winsys objects. */
148 struct pb_buffer
*buf
;
149 uint64_t gpu_address
;
150 /* Memory usage if the buffer placement is optimal. */
154 /* Resource properties. */
156 unsigned bo_alignment
;
157 enum radeon_bo_domain domains
;
158 enum radeon_bo_flag flags
;
159 unsigned bind_history
;
161 /* The buffer range which is initialized (with a write transfer,
162 * streamout, DMA, or as a random access target). The rest of
163 * the buffer is considered invalid and can be mapped unsynchronized.
165 * This allows unsychronized mapping of a buffer range which hasn't
166 * been used yet. It's for applications which forget to use
167 * the unsynchronized map flag and expect the driver to figure it out.
169 struct util_range valid_buffer_range
;
171 /* For buffers only. This indicates that a write operation has been
172 * performed by TC L2, but the cache hasn't been flushed.
173 * Any hw block which doesn't use or bypasses TC L2 should check this
174 * flag and flush the cache before using the buffer.
176 * For example, TC L2 must be flushed if a buffer which has been
177 * modified by a shader store instruction is about to be used as
178 * an index buffer. The reason is that VGT DMA index fetching doesn't
183 /* Whether the resource has been exported via resource_get_handle. */
184 unsigned external_usage
; /* PIPE_HANDLE_USAGE_* */
186 /* Whether this resource is referenced by bindless handles. */
187 bool texture_handle_allocated
;
188 bool image_handle_allocated
;
191 struct r600_transfer
{
192 struct threaded_transfer b
;
193 struct r600_resource
*staging
;
197 struct r600_fmask_info
{
201 unsigned pitch_in_pixels
;
202 unsigned bank_height
;
203 unsigned slice_tile_max
;
204 unsigned tile_mode_index
;
207 struct r600_cmask_info
{
211 unsigned slice_tile_max
;
212 uint64_t base_address_reg
;
215 struct r600_texture
{
216 struct r600_resource resource
;
219 unsigned num_level0_transfers
;
220 enum pipe_format db_render_format
;
225 unsigned dirty_level_mask
; /* each bit says if that mipmap is compressed */
226 unsigned stencil_dirty_level_mask
; /* each bit says if that mipmap is compressed */
227 struct r600_texture
*flushed_depth_texture
;
228 struct radeon_surf surface
;
230 /* Colorbuffer compression and fast clear. */
231 struct r600_fmask_info fmask
;
232 struct r600_cmask_info cmask
;
233 struct r600_resource
*cmask_buffer
;
234 uint64_t dcc_offset
; /* 0 = disabled */
235 unsigned cb_color_info
; /* fast clear enable bit */
236 unsigned color_clear_value
[2];
237 unsigned last_msaa_resolve_target_micro_mode
;
239 /* Depth buffer compression and fast clear. */
240 uint64_t htile_offset
;
241 bool tc_compatible_htile
;
242 bool depth_cleared
; /* if it was cleared at least once */
243 float depth_clear_value
;
244 bool stencil_cleared
; /* if it was cleared at least once */
245 uint8_t stencil_clear_value
;
247 bool non_disp_tiling
; /* R600-Cayman only */
249 /* Whether the texture is a displayable back buffer and needs DCC
250 * decompression, which is expensive. Therefore, it's enabled only
251 * if statistics suggest that it will pay off and it's allocated
252 * separately. It can't be bound as a sampler by apps. Limited to
253 * target == 2D and last_level == 0. If enabled, dcc_offset contains
254 * the absolute GPUVM address, not the relative one.
256 struct r600_resource
*dcc_separate_buffer
;
257 /* When DCC is temporarily disabled, the separate buffer is here. */
258 struct r600_resource
*last_dcc_separate_buffer
;
259 /* We need to track DCC dirtiness, because st/dri usually calls
260 * flush_resource twice per frame (not a bug) and we don't wanna
261 * decompress DCC twice. Also, the dirty tracking must be done even
262 * if DCC isn't used, because it's required by the DCC usage analysis
263 * for a possible future enablement.
265 bool separate_dcc_dirty
;
266 /* Statistics gathering for the DCC enablement heuristic. */
267 bool dcc_gather_statistics
;
268 /* Estimate of how much this color buffer is written to in units of
269 * full-screen draws: ps_invocations / (width * height)
270 * Shader kills, late Z, and blending with trivial discards make it
271 * inaccurate (we need to count CB updates, not PS invocations).
273 unsigned ps_draw_ratio
;
274 /* The number of clears since the last DCC usage analysis. */
275 unsigned num_slow_clears
;
277 /* Counter that should be non-zero if the texture is bound to a
278 * framebuffer. Implemented in radeonsi only.
280 uint32_t framebuffers_bound
;
283 struct r600_surface
{
284 struct pipe_surface base
;
286 /* These can vary with block-compressed textures. */
290 bool color_initialized
;
291 bool depth_initialized
;
293 /* Misc. color flags. */
294 bool alphatest_bypass
;
298 bool dcc_incompatible
;
300 /* Color registers. */
301 unsigned cb_color_info
;
302 unsigned cb_color_base
;
303 unsigned cb_color_view
;
304 unsigned cb_color_size
; /* R600 only */
305 unsigned cb_color_dim
; /* EG only */
306 unsigned cb_color_pitch
; /* EG and later */
307 unsigned cb_color_slice
; /* EG and later */
308 unsigned cb_color_attrib
; /* EG and later */
309 unsigned cb_color_attrib2
; /* GFX9 and later */
310 unsigned cb_dcc_control
; /* VI and later */
311 unsigned cb_color_fmask
; /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
312 unsigned cb_color_fmask_slice
; /* EG and later */
313 unsigned cb_color_cmask
; /* CB_COLORn_TILE (r600 only) */
314 unsigned cb_color_mask
; /* R600 only */
315 unsigned spi_shader_col_format
; /* SI+, no blending, no alpha-to-coverage. */
316 unsigned spi_shader_col_format_alpha
; /* SI+, alpha-to-coverage */
317 unsigned spi_shader_col_format_blend
; /* SI+, blending without alpha. */
318 unsigned spi_shader_col_format_blend_alpha
; /* SI+, blending with alpha. */
319 struct r600_resource
*cb_buffer_fmask
; /* Used for FMASK relocations. R600 only */
320 struct r600_resource
*cb_buffer_cmask
; /* Used for CMASK relocations. R600 only */
323 uint64_t db_depth_base
; /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
324 uint64_t db_stencil_base
; /* EG and later */
325 uint64_t db_htile_data_base
;
326 unsigned db_depth_info
; /* R600 only, then SI and later */
327 unsigned db_z_info
; /* EG and later */
328 unsigned db_z_info2
; /* GFX9+ */
329 unsigned db_depth_view
;
330 unsigned db_depth_size
;
331 unsigned db_depth_slice
; /* EG and later */
332 unsigned db_stencil_info
; /* EG and later */
333 unsigned db_stencil_info2
; /* GFX9+ */
334 unsigned db_prefetch_limit
; /* R600 only */
335 unsigned db_htile_surface
;
336 unsigned db_preload_control
; /* EG and later */
339 struct r600_mmio_counter
{
344 union r600_mmio_counters
{
346 /* For global GPU load including SDMA. */
347 struct r600_mmio_counter gpu
;
350 struct r600_mmio_counter spi
;
351 struct r600_mmio_counter gui
;
352 struct r600_mmio_counter ta
;
353 struct r600_mmio_counter gds
;
354 struct r600_mmio_counter vgt
;
355 struct r600_mmio_counter ia
;
356 struct r600_mmio_counter sx
;
357 struct r600_mmio_counter wd
;
358 struct r600_mmio_counter bci
;
359 struct r600_mmio_counter sc
;
360 struct r600_mmio_counter pa
;
361 struct r600_mmio_counter db
;
362 struct r600_mmio_counter cp
;
363 struct r600_mmio_counter cb
;
366 struct r600_mmio_counter sdma
;
369 struct r600_mmio_counter pfp
;
370 struct r600_mmio_counter meq
;
371 struct r600_mmio_counter me
;
372 struct r600_mmio_counter surf_sync
;
373 struct r600_mmio_counter dma
;
374 struct r600_mmio_counter scratch_ram
;
375 struct r600_mmio_counter ce
;
380 struct r600_common_screen
{
381 struct pipe_screen b
;
382 struct radeon_winsys
*ws
;
383 enum radeon_family family
;
384 enum chip_class chip_class
;
385 struct radeon_info info
;
386 uint64_t debug_flags
;
389 bool has_rbplus
; /* if RB+ registers exist */
390 bool rbplus_allowed
; /* if RB+ is allowed */
392 struct disk_cache
*disk_shader_cache
;
394 struct slab_parent_pool pool_transfers
;
396 /* Texture filter settings. */
397 int force_aniso
; /* -1 = disabled */
399 /* Auxiliary context. Mainly used to initialize resources.
400 * It must be locked prior to using and flushed before unlocking. */
401 struct pipe_context
*aux_context
;
402 mtx_t aux_context_lock
;
404 /* This must be in the screen, because UE4 uses one context for
405 * compilation and another one for rendering.
407 unsigned num_compilations
;
408 /* Along with ST_DEBUG=precompile, this should show if applications
409 * are loading shaders on demand. This is a monotonic counter.
411 unsigned num_shaders_created
;
412 unsigned num_shader_cache_hits
;
414 /* GPU load thread. */
415 mtx_t gpu_load_mutex
;
416 thrd_t gpu_load_thread
;
417 union r600_mmio_counters mmio_counters
;
418 volatile unsigned gpu_load_stop_thread
; /* bool */
420 char renderer_string
[100];
422 /* Performance counters. */
423 struct r600_perfcounters
*perfcounters
;
425 /* If pipe_screen wants to recompute and re-emit the framebuffer,
426 * sampler, and image states of all contexts, it should atomically
429 * Each context will compare this with its own last known value of
430 * the counter before drawing and re-emit the states accordingly.
432 unsigned dirty_tex_counter
;
434 /* Atomically increment this counter when an existing texture's
435 * metadata is enabled or disabled in a way that requires changing
436 * contexts' compressed texture binding masks.
438 unsigned compressed_colortex_counter
;
441 /* Context flags to set so that all writes from earlier jobs
442 * in the CP are seen by L2 clients.
446 /* Context flags to set so that all writes from earlier
447 * compute jobs are seen by L2 clients.
449 unsigned compute_to_L2
;
452 void (*query_opaque_metadata
)(struct r600_common_screen
*rscreen
,
453 struct r600_texture
*rtex
,
454 struct radeon_bo_metadata
*md
);
456 void (*apply_opaque_metadata
)(struct r600_common_screen
*rscreen
,
457 struct r600_texture
*rtex
,
458 struct radeon_bo_metadata
*md
);
461 /* This encapsulates a state or an operation which can emitted into the GPU
464 void (*emit
)(struct r600_common_context
*ctx
, struct r600_atom
*state
);
469 struct r600_so_target
{
470 struct pipe_stream_output_target b
;
472 /* The buffer where BUFFER_FILLED_SIZE is stored. */
473 struct r600_resource
*buf_filled_size
;
474 unsigned buf_filled_size_offset
;
475 bool buf_filled_size_valid
;
477 unsigned stride_in_dw
;
480 struct r600_streamout
{
481 struct r600_atom begin_atom
;
483 unsigned num_dw_for_end
;
485 unsigned enabled_mask
;
486 unsigned num_targets
;
487 struct r600_so_target
*targets
[PIPE_MAX_SO_BUFFERS
];
489 unsigned append_bitmask
;
492 /* External state which comes from the vertex shader,
493 * it must be set explicitly when binding a shader. */
494 uint16_t *stride_in_dw
;
495 unsigned enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
497 /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
498 unsigned hw_enabled_mask
;
500 /* The state of VGT_STRMOUT_(CONFIG|EN). */
501 struct r600_atom enable_atom
;
502 bool streamout_enabled
;
503 bool prims_gen_query_enabled
;
504 int num_prims_gen_queries
;
507 struct r600_signed_scissor
{
514 struct r600_scissors
{
515 struct r600_atom atom
;
517 struct pipe_scissor_state states
[R600_MAX_VIEWPORTS
];
520 struct r600_viewports
{
521 struct r600_atom atom
;
523 unsigned depth_range_dirty_mask
;
524 struct pipe_viewport_state states
[R600_MAX_VIEWPORTS
];
525 struct r600_signed_scissor as_scissor
[R600_MAX_VIEWPORTS
];
529 struct radeon_winsys_cs
*cs
;
530 void (*flush
)(void *ctx
, unsigned flags
,
531 struct pipe_fence_handle
**fence
);
534 /* Saved CS data for debugging features. */
535 struct radeon_saved_cs
{
539 struct radeon_bo_list_item
*bo_list
;
543 struct r600_common_context
{
544 struct pipe_context b
; /* base class */
546 struct r600_common_screen
*screen
;
547 struct radeon_winsys
*ws
;
548 struct radeon_winsys_ctx
*ctx
;
549 enum radeon_family family
;
550 enum chip_class chip_class
;
551 struct r600_ring gfx
;
552 struct r600_ring dma
;
553 struct pipe_fence_handle
*last_gfx_fence
;
554 struct pipe_fence_handle
*last_sdma_fence
;
555 unsigned num_gfx_cs_flushes
;
556 unsigned initial_gfx_cs_size
;
557 unsigned gpu_reset_counter
;
558 unsigned last_dirty_tex_counter
;
559 unsigned last_compressed_colortex_counter
;
561 struct threaded_context
*tc
;
562 struct u_suballocator
*allocator_zeroed_memory
;
563 struct slab_child_pool pool_transfers
;
564 struct slab_child_pool pool_transfers_unsync
; /* for threaded_context */
566 /* Current unaccounted memory usage. */
571 struct r600_streamout streamout
;
572 struct r600_scissors scissors
;
573 struct r600_viewports viewports
;
574 bool scissor_enabled
;
576 bool vs_writes_viewport_index
;
577 bool vs_disables_clipping_viewport
;
579 /* Additional context states. */
580 unsigned flags
; /* flush flags */
583 /* Maintain the list of active queries for pausing between IBs. */
584 int num_occlusion_queries
;
585 int num_perfect_occlusion_queries
;
586 struct list_head active_queries
;
587 unsigned num_cs_dw_queries_suspend
;
589 unsigned num_draw_calls
;
590 unsigned num_prim_restart_calls
;
591 unsigned num_spill_draw_calls
;
592 unsigned num_compute_calls
;
593 unsigned num_spill_compute_calls
;
594 unsigned num_dma_calls
;
595 unsigned num_cp_dma_calls
;
596 unsigned num_vs_flushes
;
597 unsigned num_ps_flushes
;
598 unsigned num_cs_flushes
;
599 unsigned num_cb_cache_flushes
;
600 unsigned num_db_cache_flushes
;
601 unsigned num_L2_invalidates
;
602 unsigned num_L2_writebacks
;
603 unsigned num_resident_handles
;
604 uint64_t num_alloc_tex_transfer_bytes
;
605 unsigned last_tex_ps_draw_ratio
; /* for query */
607 /* Render condition. */
608 struct r600_atom render_cond_atom
;
609 struct pipe_query
*render_cond
;
610 unsigned render_cond_mode
;
611 bool render_cond_invert
;
612 bool render_cond_force_off
; /* for u_blitter */
614 /* MSAA sample locations.
615 * The first index is the sample index.
616 * The second index is the coordinate: X, Y. */
617 float sample_locations_1x
[1][2];
618 float sample_locations_2x
[2][2];
619 float sample_locations_4x
[4][2];
620 float sample_locations_8x
[8][2];
621 float sample_locations_16x
[16][2];
623 /* Statistics gathering for the DCC enablement heuristic. It can't be
624 * in r600_texture because r600_texture can be shared by multiple
625 * contexts. This is for back buffers only. We shouldn't get too many
628 * X11 DRI3 rotates among a finite set of back buffers. They should
629 * all fit in this array. If they don't, separate DCC might never be
630 * enabled by DCC stat gathering.
633 struct r600_texture
*tex
;
634 /* Query queue: 0 = usually active, 1 = waiting, 2 = readback. */
635 struct pipe_query
*ps_stats
[3];
636 /* If all slots are used and another slot is needed,
637 * the least recently used slot is evicted based on this. */
638 int64_t last_use_timestamp
;
642 struct pipe_debug_callback debug
;
643 struct pipe_device_reset_callback device_reset_callback
;
645 void *query_result_shader
;
647 /* Copy one resource to another using async DMA. */
648 void (*dma_copy
)(struct pipe_context
*ctx
,
649 struct pipe_resource
*dst
,
651 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
652 struct pipe_resource
*src
,
654 const struct pipe_box
*src_box
);
656 void (*dma_clear_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*dst
,
657 uint64_t offset
, uint64_t size
, unsigned value
);
659 void (*clear_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*dst
,
660 uint64_t offset
, uint64_t size
, unsigned value
,
661 enum r600_coherency coher
);
663 void (*blit_decompress_depth
)(struct pipe_context
*ctx
,
664 struct r600_texture
*texture
,
665 struct r600_texture
*staging
,
666 unsigned first_level
, unsigned last_level
,
667 unsigned first_layer
, unsigned last_layer
,
668 unsigned first_sample
, unsigned last_sample
);
670 void (*decompress_dcc
)(struct pipe_context
*ctx
,
671 struct r600_texture
*rtex
);
673 /* Reallocate the buffer and update all resource bindings where
674 * the buffer is bound, including all resource descriptors. */
675 void (*invalidate_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*buf
);
677 /* Update all resource bindings where the buffer is bound, including
678 * all resource descriptors. This is invalidate_buffer without
679 * the invalidation. */
680 void (*rebind_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*buf
,
681 uint64_t old_gpu_address
);
683 /* Enable or disable occlusion queries. */
684 void (*set_occlusion_query_state
)(struct pipe_context
*ctx
, bool enable
);
686 void (*save_qbo_state
)(struct pipe_context
*ctx
, struct r600_qbo_state
*st
);
688 /* This ensures there is enough space in the command stream. */
689 void (*need_gfx_cs_space
)(struct pipe_context
*ctx
, unsigned num_dw
,
690 bool include_draw_vbo
);
692 void (*set_atom_dirty
)(struct r600_common_context
*ctx
,
693 struct r600_atom
*atom
, bool dirty
);
695 void (*check_vm_faults
)(struct r600_common_context
*ctx
,
696 struct radeon_saved_cs
*saved
,
697 enum ring_type ring
);
700 /* r600_buffer_common.c */
701 bool r600_rings_is_buffer_referenced(struct r600_common_context
*ctx
,
702 struct pb_buffer
*buf
,
703 enum radeon_bo_usage usage
);
704 void *r600_buffer_map_sync_with_rings(struct r600_common_context
*ctx
,
705 struct r600_resource
*resource
,
707 void r600_buffer_subdata(struct pipe_context
*ctx
,
708 struct pipe_resource
*buffer
,
709 unsigned usage
, unsigned offset
,
710 unsigned size
, const void *data
);
711 void r600_init_resource_fields(struct r600_common_screen
*rscreen
,
712 struct r600_resource
*res
,
713 uint64_t size
, unsigned alignment
);
714 bool r600_alloc_resource(struct r600_common_screen
*rscreen
,
715 struct r600_resource
*res
);
716 struct pipe_resource
*r600_buffer_create(struct pipe_screen
*screen
,
717 const struct pipe_resource
*templ
,
719 struct pipe_resource
* r600_aligned_buffer_create(struct pipe_screen
*screen
,
724 struct pipe_resource
*
725 r600_buffer_from_user_memory(struct pipe_screen
*screen
,
726 const struct pipe_resource
*templ
,
729 r600_invalidate_resource(struct pipe_context
*ctx
,
730 struct pipe_resource
*resource
);
731 void r600_replace_buffer_storage(struct pipe_context
*ctx
,
732 struct pipe_resource
*dst
,
733 struct pipe_resource
*src
);
735 /* r600_common_pipe.c */
736 void r600_gfx_write_event_eop(struct r600_common_context
*ctx
,
737 unsigned event
, unsigned event_flags
,
739 struct r600_resource
*buf
, uint64_t va
,
740 uint32_t old_fence
, uint32_t new_fence
);
741 unsigned r600_gfx_write_fence_dwords(struct r600_common_screen
*screen
);
742 void r600_gfx_wait_fence(struct r600_common_context
*ctx
,
743 uint64_t va
, uint32_t ref
, uint32_t mask
);
744 void r600_draw_rectangle(struct blitter_context
*blitter
,
745 int x1
, int y1
, int x2
, int y2
, float depth
,
746 enum blitter_attrib_type type
,
747 const union pipe_color_union
*attrib
);
748 bool r600_common_screen_init(struct r600_common_screen
*rscreen
,
749 struct radeon_winsys
*ws
, unsigned flags
);
750 void r600_destroy_common_screen(struct r600_common_screen
*rscreen
);
751 void r600_preflush_suspend_features(struct r600_common_context
*ctx
);
752 void r600_postflush_resume_features(struct r600_common_context
*ctx
);
753 bool r600_common_context_init(struct r600_common_context
*rctx
,
754 struct r600_common_screen
*rscreen
,
755 unsigned context_flags
);
756 void r600_common_context_cleanup(struct r600_common_context
*rctx
);
757 bool r600_can_dump_shader(struct r600_common_screen
*rscreen
,
759 bool r600_extra_shader_checks(struct r600_common_screen
*rscreen
,
761 void r600_screen_clear_buffer(struct r600_common_screen
*rscreen
, struct pipe_resource
*dst
,
762 uint64_t offset
, uint64_t size
, unsigned value
);
763 struct pipe_resource
*r600_resource_create_common(struct pipe_screen
*screen
,
764 const struct pipe_resource
*templ
);
765 const char *r600_get_llvm_processor_name(enum radeon_family family
);
766 void r600_need_dma_space(struct r600_common_context
*ctx
, unsigned num_dw
,
767 struct r600_resource
*dst
, struct r600_resource
*src
);
768 void radeon_save_cs(struct radeon_winsys
*ws
, struct radeon_winsys_cs
*cs
,
769 struct radeon_saved_cs
*saved
);
770 void radeon_clear_saved_cs(struct radeon_saved_cs
*saved
);
771 bool r600_check_device_reset(struct r600_common_context
*rctx
);
773 /* r600_gpu_load.c */
774 void r600_gpu_load_kill_thread(struct r600_common_screen
*rscreen
);
775 uint64_t r600_begin_counter(struct r600_common_screen
*rscreen
, unsigned type
);
776 unsigned r600_end_counter(struct r600_common_screen
*rscreen
, unsigned type
,
779 /* r600_perfcounters.c */
780 void r600_perfcounters_destroy(struct r600_common_screen
*rscreen
);
783 void r600_init_screen_query_functions(struct r600_common_screen
*rscreen
);
784 void r600_query_init(struct r600_common_context
*rctx
);
785 void r600_suspend_queries(struct r600_common_context
*ctx
);
786 void r600_resume_queries(struct r600_common_context
*ctx
);
787 void r600_query_fix_enabled_rb_mask(struct r600_common_screen
*rscreen
);
789 /* r600_streamout.c */
790 void r600_streamout_buffers_dirty(struct r600_common_context
*rctx
);
791 void r600_set_streamout_targets(struct pipe_context
*ctx
,
792 unsigned num_targets
,
793 struct pipe_stream_output_target
**targets
,
794 const unsigned *offset
);
795 void r600_emit_streamout_end(struct r600_common_context
*rctx
);
796 void r600_update_prims_generated_query_state(struct r600_common_context
*rctx
,
797 unsigned type
, int diff
);
798 void r600_streamout_init(struct r600_common_context
*rctx
);
800 /* r600_test_dma.c */
801 void r600_test_dma(struct r600_common_screen
*rscreen
);
804 bool r600_prepare_for_dma_blit(struct r600_common_context
*rctx
,
805 struct r600_texture
*rdst
,
806 unsigned dst_level
, unsigned dstx
,
807 unsigned dsty
, unsigned dstz
,
808 struct r600_texture
*rsrc
,
810 const struct pipe_box
*src_box
);
811 void r600_texture_get_fmask_info(struct r600_common_screen
*rscreen
,
812 struct r600_texture
*rtex
,
814 struct r600_fmask_info
*out
);
815 void r600_texture_get_cmask_info(struct r600_common_screen
*rscreen
,
816 struct r600_texture
*rtex
,
817 struct r600_cmask_info
*out
);
818 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
819 struct pipe_resource
*texture
,
820 struct r600_texture
**staging
);
821 void r600_print_texture_info(struct r600_common_screen
*rscreen
,
822 struct r600_texture
*rtex
, FILE *f
);
823 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
824 const struct pipe_resource
*templ
);
825 bool vi_dcc_formats_compatible(enum pipe_format format1
,
826 enum pipe_format format2
);
827 bool vi_dcc_formats_are_incompatible(struct pipe_resource
*tex
,
829 enum pipe_format view_format
);
830 void vi_disable_dcc_if_incompatible_format(struct r600_common_context
*rctx
,
831 struct pipe_resource
*tex
,
833 enum pipe_format view_format
);
834 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
835 struct pipe_resource
*texture
,
836 const struct pipe_surface
*templ
,
837 unsigned width0
, unsigned height0
,
838 unsigned width
, unsigned height
);
839 unsigned r600_translate_colorswap(enum pipe_format format
, bool do_endian_swap
);
840 void vi_separate_dcc_start_query(struct pipe_context
*ctx
,
841 struct r600_texture
*tex
);
842 void vi_separate_dcc_stop_query(struct pipe_context
*ctx
,
843 struct r600_texture
*tex
);
844 void vi_separate_dcc_process_and_reset_stats(struct pipe_context
*ctx
,
845 struct r600_texture
*tex
);
846 void vi_dcc_clear_level(struct r600_common_context
*rctx
,
847 struct r600_texture
*rtex
,
848 unsigned level
, unsigned clear_value
);
849 void evergreen_do_fast_color_clear(struct r600_common_context
*rctx
,
850 struct pipe_framebuffer_state
*fb
,
851 struct r600_atom
*fb_state
,
852 unsigned *buffers
, ubyte
*dirty_cbufs
,
853 const union pipe_color_union
*color
);
854 bool r600_texture_disable_dcc(struct r600_common_context
*rctx
,
855 struct r600_texture
*rtex
);
856 void r600_init_screen_texture_functions(struct r600_common_screen
*rscreen
);
857 void r600_init_context_texture_functions(struct r600_common_context
*rctx
);
859 /* r600_viewport.c */
860 void evergreen_apply_scissor_bug_workaround(struct r600_common_context
*rctx
,
861 struct pipe_scissor_state
*scissor
);
862 void r600_viewport_set_rast_deps(struct r600_common_context
*rctx
,
863 bool scissor_enable
, bool clip_halfz
);
864 void r600_update_vs_writes_viewport_index(struct r600_common_context
*rctx
,
865 struct tgsi_shader_info
*info
);
866 void r600_init_viewport_functions(struct r600_common_context
*rctx
);
869 extern const uint32_t eg_sample_locs_2x
[4];
870 extern const unsigned eg_max_dist_2x
;
871 extern const uint32_t eg_sample_locs_4x
[4];
872 extern const unsigned eg_max_dist_4x
;
873 void cayman_get_sample_position(struct pipe_context
*ctx
, unsigned sample_count
,
874 unsigned sample_index
, float *out_value
);
875 void cayman_init_msaa(struct pipe_context
*ctx
);
876 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
877 void cayman_emit_msaa_config(struct radeon_winsys_cs
*cs
, int nr_samples
,
878 int ps_iter_samples
, int overrast_samples
,
879 unsigned sc_mode_cntl_1
);
882 /* Inline helpers. */
884 static inline struct r600_resource
*r600_resource(struct pipe_resource
*r
)
886 return (struct r600_resource
*)r
;
890 r600_resource_reference(struct r600_resource
**ptr
, struct r600_resource
*res
)
892 pipe_resource_reference((struct pipe_resource
**)ptr
,
893 (struct pipe_resource
*)res
);
897 r600_texture_reference(struct r600_texture
**ptr
, struct r600_texture
*res
)
899 pipe_resource_reference((struct pipe_resource
**)ptr
, &res
->resource
.b
.b
);
903 r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
)
905 struct r600_common_context
*rctx
= (struct r600_common_context
*)ctx
;
906 struct r600_resource
*res
= (struct r600_resource
*)r
;
909 /* Add memory usage for need_gfx_cs_space */
910 rctx
->vram
+= res
->vram_usage
;
911 rctx
->gtt
+= res
->gart_usage
;
915 static inline bool r600_get_strmout_en(struct r600_common_context
*rctx
)
917 return rctx
->streamout
.streamout_enabled
||
918 rctx
->streamout
.prims_gen_query_enabled
;
921 #define SQ_TEX_XY_FILTER_POINT 0x00
922 #define SQ_TEX_XY_FILTER_BILINEAR 0x01
923 #define SQ_TEX_XY_FILTER_ANISO_POINT 0x02
924 #define SQ_TEX_XY_FILTER_ANISO_BILINEAR 0x03
926 static inline unsigned eg_tex_filter(unsigned filter
, unsigned max_aniso
)
928 if (filter
== PIPE_TEX_FILTER_LINEAR
)
929 return max_aniso
> 1 ? SQ_TEX_XY_FILTER_ANISO_BILINEAR
930 : SQ_TEX_XY_FILTER_BILINEAR
;
932 return max_aniso
> 1 ? SQ_TEX_XY_FILTER_ANISO_POINT
933 : SQ_TEX_XY_FILTER_POINT
;
936 static inline unsigned r600_tex_aniso_filter(unsigned filter
)
949 static inline unsigned r600_wavefront_size(enum radeon_family family
)
969 static inline enum radeon_bo_priority
970 r600_get_sampler_view_priority(struct r600_resource
*res
)
972 if (res
->b
.b
.target
== PIPE_BUFFER
)
973 return RADEON_PRIO_SAMPLER_BUFFER
;
975 if (res
->b
.b
.nr_samples
> 1)
976 return RADEON_PRIO_SAMPLER_TEXTURE_MSAA
;
978 return RADEON_PRIO_SAMPLER_TEXTURE
;
982 r600_can_sample_zs(struct r600_texture
*tex
, bool stencil_sampler
)
984 return (stencil_sampler
&& tex
->can_sample_s
) ||
985 (!stencil_sampler
&& tex
->can_sample_z
);
989 vi_dcc_enabled(struct r600_texture
*tex
, unsigned level
)
991 return tex
->dcc_offset
&& level
< tex
->surface
.num_dcc_levels
;
994 #define COMPUTE_DBG(rscreen, fmt, args...) \
996 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
999 #define R600_ERR(fmt, args...) \
1000 fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
1002 /* For MSAA sample positions. */
1003 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
1004 (((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) | \
1005 (((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) | \
1006 (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) | \
1007 (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))