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/u_blitter.h"
40 #include "util/list.h"
41 #include "util/u_range.h"
42 #include "util/u_slab.h"
43 #include "util/u_suballoc.h"
44 #include "util/u_transfer.h"
46 #define R600_RESOURCE_FLAG_TRANSFER (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
47 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
48 #define R600_RESOURCE_FLAG_FORCE_TILING (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
50 #define R600_QUERY_DRAW_CALLS (PIPE_QUERY_DRIVER_SPECIFIC + 0)
51 #define R600_QUERY_REQUESTED_VRAM (PIPE_QUERY_DRIVER_SPECIFIC + 1)
52 #define R600_QUERY_REQUESTED_GTT (PIPE_QUERY_DRIVER_SPECIFIC + 2)
53 #define R600_QUERY_BUFFER_WAIT_TIME (PIPE_QUERY_DRIVER_SPECIFIC + 3)
54 #define R600_QUERY_NUM_CS_FLUSHES (PIPE_QUERY_DRIVER_SPECIFIC + 4)
55 #define R600_QUERY_NUM_BYTES_MOVED (PIPE_QUERY_DRIVER_SPECIFIC + 5)
56 #define R600_QUERY_VRAM_USAGE (PIPE_QUERY_DRIVER_SPECIFIC + 6)
57 #define R600_QUERY_GTT_USAGE (PIPE_QUERY_DRIVER_SPECIFIC + 7)
58 #define R600_QUERY_GPU_TEMPERATURE (PIPE_QUERY_DRIVER_SPECIFIC + 8)
59 #define R600_QUERY_CURRENT_GPU_SCLK (PIPE_QUERY_DRIVER_SPECIFIC + 9)
60 #define R600_QUERY_CURRENT_GPU_MCLK (PIPE_QUERY_DRIVER_SPECIFIC + 10)
61 #define R600_QUERY_GPU_LOAD (PIPE_QUERY_DRIVER_SPECIFIC + 11)
63 #define R600_CONTEXT_STREAMOUT_FLUSH (1u << 0)
64 #define R600_CONTEXT_PRIVATE_FLAG (1u << 1)
66 /* special primitive types */
67 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
71 #define DBG_TEX (1 << 0)
72 #define DBG_TEXMIP (1 << 1)
73 #define DBG_COMPUTE (1 << 2)
74 #define DBG_VM (1 << 3)
75 #define DBG_TRACE_CS (1 << 4)
77 #define DBG_FS (1 << 5)
78 #define DBG_VS (1 << 6)
79 #define DBG_GS (1 << 7)
80 #define DBG_PS (1 << 8)
81 #define DBG_CS (1 << 9)
82 #define DBG_TCS (1 << 10)
83 #define DBG_TES (1 << 11)
84 #define DBG_NO_IR (1 << 12)
85 #define DBG_NO_TGSI (1 << 13)
86 #define DBG_NO_ASM (1 << 14)
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)
99 #define R600_MAP_BUFFER_ALIGNMENT 64
101 struct r600_common_context
;
103 struct radeon_shader_reloc
{
108 struct radeon_shader_binary
{
113 /** Config/Context register state that accompanies this shader.
114 * This is a stream of dword pairs. First dword contains the
115 * register address, the second dword contains the value.*/
116 unsigned char *config
;
117 unsigned config_size
;
119 /** The number of bytes of config information for each global symbol.
121 unsigned config_size_per_symbol
;
123 /** Constant data accessed by the shader. This will be uploaded
124 * into a constant buffer. */
125 unsigned char *rodata
;
126 unsigned rodata_size
;
128 /** List of symbol offsets for the shader */
129 uint64_t *global_symbol_offsets
;
130 unsigned global_symbol_count
;
132 struct radeon_shader_reloc
*relocs
;
133 unsigned reloc_count
;
135 /** Disassembled shader in a string. */
139 struct r600_resource
{
142 /* Winsys objects. */
143 struct pb_buffer
*buf
;
144 struct radeon_winsys_cs_handle
*cs_buf
;
145 uint64_t gpu_address
;
147 /* Resource state. */
148 enum radeon_bo_domain domains
;
150 /* The buffer range which is initialized (with a write transfer,
151 * streamout, DMA, or as a random access target). The rest of
152 * the buffer is considered invalid and can be mapped unsynchronized.
154 * This allows unsychronized mapping of a buffer range which hasn't
155 * been used yet. It's for applications which forget to use
156 * the unsynchronized map flag and expect the driver to figure it out.
158 struct util_range valid_buffer_range
;
160 /* For buffers only. This indicates that a write operation has been
161 * performed by TC L2, but the cache hasn't been flushed.
162 * Any hw block which doesn't use or bypasses TC L2 should check this
163 * flag and flush the cache before using the buffer.
165 * For example, TC L2 must be flushed if a buffer which has been
166 * modified by a shader store instruction is about to be used as
167 * an index buffer. The reason is that VGT DMA index fetching doesn't
173 struct r600_transfer
{
174 struct pipe_transfer transfer
;
175 struct r600_resource
*staging
;
179 struct r600_fmask_info
{
184 unsigned bank_height
;
185 unsigned slice_tile_max
;
186 unsigned tile_mode_index
;
189 struct r600_cmask_info
{
193 unsigned slice_tile_max
;
194 unsigned base_address_reg
;
197 struct r600_texture
{
198 struct r600_resource resource
;
201 unsigned pitch_override
;
203 unsigned dirty_level_mask
; /* each bit says if that mipmap is compressed */
204 struct r600_texture
*flushed_depth_texture
;
205 boolean is_flushing_texture
;
206 struct radeon_surf surface
;
208 /* Colorbuffer compression and fast clear. */
209 struct r600_fmask_info fmask
;
210 struct r600_cmask_info cmask
;
211 struct r600_resource
*cmask_buffer
;
212 unsigned cb_color_info
; /* fast clear enable bit */
213 unsigned color_clear_value
[2];
215 /* Depth buffer compression and fast clear. */
216 struct r600_resource
*htile_buffer
;
217 bool depth_cleared
; /* if it was cleared at least once */
218 float depth_clear_value
;
220 bool non_disp_tiling
; /* R600-Cayman only */
223 struct r600_surface
{
224 struct pipe_surface base
;
226 bool color_initialized
;
227 bool depth_initialized
;
229 /* Misc. color flags. */
230 bool alphatest_bypass
;
233 /* Color registers. */
234 unsigned cb_color_info
;
235 unsigned cb_color_base
;
236 unsigned cb_color_view
;
237 unsigned cb_color_size
; /* R600 only */
238 unsigned cb_color_dim
; /* EG only */
239 unsigned cb_color_pitch
; /* EG and later */
240 unsigned cb_color_slice
; /* EG and later */
241 unsigned cb_color_attrib
; /* EG and later */
242 unsigned cb_color_fmask
; /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
243 unsigned cb_color_fmask_slice
; /* EG and later */
244 unsigned cb_color_cmask
; /* CB_COLORn_TILE (r600 only) */
245 unsigned cb_color_mask
; /* R600 only */
246 struct r600_resource
*cb_buffer_fmask
; /* Used for FMASK relocations. R600 only */
247 struct r600_resource
*cb_buffer_cmask
; /* Used for CMASK relocations. R600 only */
250 unsigned db_depth_info
; /* R600 only, then SI and later */
251 unsigned db_z_info
; /* EG and later */
252 unsigned db_depth_base
; /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
253 unsigned db_depth_view
;
254 unsigned db_depth_size
;
255 unsigned db_depth_slice
; /* EG and later */
256 unsigned db_stencil_base
; /* EG and later */
257 unsigned db_stencil_info
; /* EG and later */
258 unsigned db_prefetch_limit
; /* R600 only */
259 unsigned db_htile_surface
;
260 unsigned db_htile_data_base
;
261 unsigned db_preload_control
; /* EG and later */
262 unsigned pa_su_poly_offset_db_fmt_cntl
;
265 struct r600_tiling_info
{
266 unsigned num_channels
;
268 unsigned group_bytes
;
271 struct r600_common_screen
{
272 struct pipe_screen b
;
273 struct radeon_winsys
*ws
;
274 enum radeon_family family
;
275 enum chip_class chip_class
;
276 struct radeon_info info
;
277 struct r600_tiling_info tiling_info
;
278 uint64_t debug_flags
;
282 /* Auxiliary context. Mainly used to initialize resources.
283 * It must be locked prior to using and flushed before unlocking. */
284 struct pipe_context
*aux_context
;
285 pipe_mutex aux_context_lock
;
287 struct r600_resource
*trace_bo
;
291 /* GPU load thread. */
292 pipe_mutex gpu_load_mutex
;
293 pipe_thread gpu_load_thread
;
294 unsigned gpu_load_counter_busy
;
295 unsigned gpu_load_counter_idle
;
296 volatile unsigned gpu_load_stop_thread
; /* bool */
299 /* This encapsulates a state or an operation which can emitted into the GPU
302 void (*emit
)(struct r600_common_context
*ctx
, struct r600_atom
*state
);
307 struct r600_so_target
{
308 struct pipe_stream_output_target b
;
310 /* The buffer where BUFFER_FILLED_SIZE is stored. */
311 struct r600_resource
*buf_filled_size
;
312 unsigned buf_filled_size_offset
;
313 bool buf_filled_size_valid
;
315 unsigned stride_in_dw
;
318 struct r600_streamout
{
319 struct r600_atom begin_atom
;
321 unsigned num_dw_for_end
;
323 unsigned enabled_mask
;
324 unsigned num_targets
;
325 struct r600_so_target
*targets
[PIPE_MAX_SO_BUFFERS
];
327 unsigned append_bitmask
;
330 /* External state which comes from the vertex shader,
331 * it must be set explicitly when binding a shader. */
332 unsigned *stride_in_dw
;
333 unsigned enabled_stream_buffers_mask
; /* stream0 buffers0-3 in 4 LSB */
335 /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
336 unsigned hw_enabled_mask
;
338 /* The state of VGT_STRMOUT_(CONFIG|EN). */
339 struct r600_atom enable_atom
;
340 bool streamout_enabled
;
341 bool prims_gen_query_enabled
;
342 int num_prims_gen_queries
;
346 struct radeon_winsys_cs
*cs
;
348 void (*flush
)(void *ctx
, unsigned flags
,
349 struct pipe_fence_handle
**fence
);
353 struct r600_ring gfx
;
354 struct r600_ring dma
;
357 struct r600_common_context
{
358 struct pipe_context b
; /* base class */
360 struct r600_common_screen
*screen
;
361 struct radeon_winsys
*ws
;
362 enum radeon_family family
;
363 enum chip_class chip_class
;
364 struct r600_rings rings
;
365 unsigned initial_gfx_cs_size
;
366 unsigned gpu_reset_counter
;
368 struct u_upload_mgr
*uploader
;
369 struct u_suballocator
*allocator_so_filled_size
;
370 struct util_slab_mempool pool_transfers
;
372 /* Current unaccounted memory usage. */
377 struct r600_streamout streamout
;
379 /* Additional context states. */
380 unsigned flags
; /* flush flags */
383 /* The list of active queries. Only one query of each type can be active. */
384 int num_occlusion_queries
;
385 /* Keep track of non-timer queries, because they should be suspended
386 * during context flushing.
387 * The timer queries (TIME_ELAPSED) shouldn't be suspended for blits,
388 * but they should be suspended between IBs. */
389 struct list_head active_nontimer_queries
;
390 struct list_head active_timer_queries
;
391 unsigned num_cs_dw_nontimer_queries_suspend
;
392 unsigned num_cs_dw_timer_queries_suspend
;
393 /* If queries have been suspended. */
394 bool queries_suspended_for_flush
;
395 /* Additional hardware info. */
396 unsigned backend_mask
;
397 unsigned max_db
; /* for OQ */
399 unsigned num_draw_calls
;
401 /* Render condition. */
402 struct pipe_query
*current_render_cond
;
403 unsigned current_render_cond_mode
;
404 boolean current_render_cond_cond
;
405 boolean predicate_drawing
;
406 /* For context flushing. */
407 struct pipe_query
*saved_render_cond
;
408 boolean saved_render_cond_cond
;
409 unsigned saved_render_cond_mode
;
411 /* MSAA sample locations.
412 * The first index is the sample index.
413 * The second index is the coordinate: X, Y. */
414 float sample_locations_1x
[1][2];
415 float sample_locations_2x
[2][2];
416 float sample_locations_4x
[4][2];
417 float sample_locations_8x
[8][2];
418 float sample_locations_16x
[16][2];
420 /* The list of all texture buffer objects in this context.
421 * This list is walked when a buffer is invalidated/reallocated and
422 * the GPU addresses are updated. */
423 struct list_head texture_buffers
;
425 /* Copy one resource to another using async DMA. */
426 void (*dma_copy
)(struct pipe_context
*ctx
,
427 struct pipe_resource
*dst
,
429 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
430 struct pipe_resource
*src
,
432 const struct pipe_box
*src_box
);
434 void (*clear_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*dst
,
435 unsigned offset
, unsigned size
, unsigned value
,
436 bool is_framebuffer
);
438 void (*blit_decompress_depth
)(struct pipe_context
*ctx
,
439 struct r600_texture
*texture
,
440 struct r600_texture
*staging
,
441 unsigned first_level
, unsigned last_level
,
442 unsigned first_layer
, unsigned last_layer
,
443 unsigned first_sample
, unsigned last_sample
);
445 /* Reallocate the buffer and update all resource bindings where
446 * the buffer is bound, including all resource descriptors. */
447 void (*invalidate_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*buf
);
449 /* Enable or disable occlusion queries. */
450 void (*set_occlusion_query_state
)(struct pipe_context
*ctx
, bool enable
);
452 /* This ensures there is enough space in the command stream. */
453 void (*need_gfx_cs_space
)(struct pipe_context
*ctx
, unsigned num_dw
,
454 bool include_draw_vbo
);
458 boolean
r600_rings_is_buffer_referenced(struct r600_common_context
*ctx
,
459 struct radeon_winsys_cs_handle
*buf
,
460 enum radeon_bo_usage usage
);
461 void *r600_buffer_map_sync_with_rings(struct r600_common_context
*ctx
,
462 struct r600_resource
*resource
,
464 bool r600_init_resource(struct r600_common_screen
*rscreen
,
465 struct r600_resource
*res
,
466 unsigned size
, unsigned alignment
,
467 bool use_reusable_pool
);
468 struct pipe_resource
*r600_buffer_create(struct pipe_screen
*screen
,
469 const struct pipe_resource
*templ
,
471 struct pipe_resource
*
472 r600_buffer_from_user_memory(struct pipe_screen
*screen
,
473 const struct pipe_resource
*templ
,
476 /* r600_common_pipe.c */
477 void r600_draw_rectangle(struct blitter_context
*blitter
,
478 int x1
, int y1
, int x2
, int y2
, float depth
,
479 enum blitter_attrib_type type
,
480 const union pipe_color_union
*attrib
);
481 bool r600_common_screen_init(struct r600_common_screen
*rscreen
,
482 struct radeon_winsys
*ws
);
483 void r600_destroy_common_screen(struct r600_common_screen
*rscreen
);
484 void r600_preflush_suspend_features(struct r600_common_context
*ctx
);
485 void r600_postflush_resume_features(struct r600_common_context
*ctx
);
486 bool r600_common_context_init(struct r600_common_context
*rctx
,
487 struct r600_common_screen
*rscreen
);
488 void r600_common_context_cleanup(struct r600_common_context
*rctx
);
489 void r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
);
490 bool r600_can_dump_shader(struct r600_common_screen
*rscreen
,
491 const struct tgsi_token
*tokens
);
492 void r600_screen_clear_buffer(struct r600_common_screen
*rscreen
, struct pipe_resource
*dst
,
493 unsigned offset
, unsigned size
, unsigned value
,
494 bool is_framebuffer
);
495 struct pipe_resource
*r600_resource_create_common(struct pipe_screen
*screen
,
496 const struct pipe_resource
*templ
);
497 const char *r600_get_llvm_processor_name(enum radeon_family family
);
498 void r600_need_dma_space(struct r600_common_context
*ctx
, unsigned num_dw
);
500 /* r600_gpu_load.c */
501 void r600_gpu_load_kill_thread(struct r600_common_screen
*rscreen
);
502 uint64_t r600_gpu_load_begin(struct r600_common_screen
*rscreen
);
503 unsigned r600_gpu_load_end(struct r600_common_screen
*rscreen
, uint64_t begin
);
506 void r600_query_init(struct r600_common_context
*rctx
);
507 void r600_suspend_nontimer_queries(struct r600_common_context
*ctx
);
508 void r600_resume_nontimer_queries(struct r600_common_context
*ctx
);
509 void r600_suspend_timer_queries(struct r600_common_context
*ctx
);
510 void r600_resume_timer_queries(struct r600_common_context
*ctx
);
511 void r600_query_init_backend_mask(struct r600_common_context
*ctx
);
513 /* r600_streamout.c */
514 void r600_streamout_buffers_dirty(struct r600_common_context
*rctx
);
515 void r600_set_streamout_targets(struct pipe_context
*ctx
,
516 unsigned num_targets
,
517 struct pipe_stream_output_target
**targets
,
518 const unsigned *offset
);
519 void r600_emit_streamout_end(struct r600_common_context
*rctx
);
520 void r600_update_prims_generated_query_state(struct r600_common_context
*rctx
,
521 unsigned type
, int diff
);
522 void r600_streamout_init(struct r600_common_context
*rctx
);
525 void r600_texture_get_fmask_info(struct r600_common_screen
*rscreen
,
526 struct r600_texture
*rtex
,
528 struct r600_fmask_info
*out
);
529 void r600_texture_get_cmask_info(struct r600_common_screen
*rscreen
,
530 struct r600_texture
*rtex
,
531 struct r600_cmask_info
*out
);
532 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
533 struct pipe_resource
*texture
,
534 struct r600_texture
**staging
);
535 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
536 const struct pipe_resource
*templ
);
537 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
538 struct pipe_resource
*texture
,
539 const struct pipe_surface
*templ
,
540 unsigned width
, unsigned height
);
541 unsigned r600_translate_colorswap(enum pipe_format format
);
542 void evergreen_do_fast_color_clear(struct r600_common_context
*rctx
,
543 struct pipe_framebuffer_state
*fb
,
544 struct r600_atom
*fb_state
,
546 const union pipe_color_union
*color
);
547 void r600_init_screen_texture_functions(struct r600_common_screen
*rscreen
);
548 void r600_init_context_texture_functions(struct r600_common_context
*rctx
);
551 extern const uint32_t eg_sample_locs_2x
[4];
552 extern const unsigned eg_max_dist_2x
;
553 extern const uint32_t eg_sample_locs_4x
[4];
554 extern const unsigned eg_max_dist_4x
;
555 void cayman_get_sample_position(struct pipe_context
*ctx
, unsigned sample_count
,
556 unsigned sample_index
, float *out_value
);
557 void cayman_init_msaa(struct pipe_context
*ctx
);
558 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
559 void cayman_emit_msaa_config(struct radeon_winsys_cs
*cs
, int nr_samples
,
560 int ps_iter_samples
, int overrast_samples
);
563 /* Inline helpers. */
565 static inline struct r600_resource
*r600_resource(struct pipe_resource
*r
)
567 return (struct r600_resource
*)r
;
571 r600_resource_reference(struct r600_resource
**ptr
, struct r600_resource
*res
)
573 pipe_resource_reference((struct pipe_resource
**)ptr
,
574 (struct pipe_resource
*)res
);
577 static inline unsigned r600_tex_aniso_filter(unsigned filter
)
579 if (filter
<= 1) return 0;
580 if (filter
<= 2) return 1;
581 if (filter
<= 4) return 2;
582 if (filter
<= 8) return 3;
586 static inline unsigned r600_wavefront_size(enum radeon_family family
)
606 #define COMPUTE_DBG(rscreen, fmt, args...) \
608 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
611 #define R600_ERR(fmt, args...) \
612 fprintf(stderr, "EE %s:%d %s - "fmt, __FILE__, __LINE__, __func__, ##args)
614 /* For MSAA sample positions. */
615 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
616 (((s0x) & 0xf) | (((s0y) & 0xf) << 4) | \
617 (((s1x) & 0xf) << 8) | (((s1y) & 0xf) << 12) | \
618 (((s2x) & 0xf) << 16) | (((s2y) & 0xf) << 20) | \
619 (((s3x) & 0xf) << 24) | (((s3y) & 0xf) << 28))