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)
83 #define DBG_NO_ASYNC_DMA (1 << 10)
84 #define DBG_NO_HYPERZ (1 << 11)
85 #define DBG_NO_DISCARD_RANGE (1 << 12)
86 #define DBG_NO_2D_TILING (1 << 13)
87 #define DBG_NO_TILING (1 << 14)
88 #define DBG_SWITCH_ON_EOP (1 << 15)
89 #define DBG_FORCE_DMA (1 << 16)
90 #define DBG_PRECOMPILE (1 << 17)
91 #define DBG_INFO (1 << 18)
92 /* The maximum allowed bit is 20. */
94 #define R600_MAP_BUFFER_ALIGNMENT 64
96 struct r600_common_context
;
98 struct radeon_shader_reloc
{
103 struct radeon_shader_binary
{
108 /** Config/Context register state that accompanies this shader.
109 * This is a stream of dword pairs. First dword contains the
110 * register address, the second dword contains the value.*/
111 unsigned char *config
;
112 unsigned config_size
;
114 /** The number of bytes of config information for each global symbol.
116 unsigned config_size_per_symbol
;
118 /** Constant data accessed by the shader. This will be uploaded
119 * into a constant buffer. */
120 unsigned char *rodata
;
121 unsigned rodata_size
;
123 /** List of symbol offsets for the shader */
124 uint64_t *global_symbol_offsets
;
125 unsigned global_symbol_count
;
127 struct radeon_shader_reloc
*relocs
;
128 unsigned reloc_count
;
130 /** Set to 1 if the disassembly for this binary has been dumped to
135 struct r600_resource
{
138 /* Winsys objects. */
139 struct pb_buffer
*buf
;
140 struct radeon_winsys_cs_handle
*cs_buf
;
141 uint64_t gpu_address
;
143 /* Resource state. */
144 enum radeon_bo_domain domains
;
146 /* The buffer range which is initialized (with a write transfer,
147 * streamout, DMA, or as a random access target). The rest of
148 * the buffer is considered invalid and can be mapped unsynchronized.
150 * This allows unsychronized mapping of a buffer range which hasn't
151 * been used yet. It's for applications which forget to use
152 * the unsynchronized map flag and expect the driver to figure it out.
154 struct util_range valid_buffer_range
;
156 /* For buffers only. This indicates that a write operation has been
157 * performed by TC L2, but the cache hasn't been flushed.
158 * Any hw block which doesn't use or bypasses TC L2 should check this
159 * flag and flush the cache before using the buffer.
161 * For example, TC L2 must be flushed if a buffer which has been
162 * modified by a shader store instruction is about to be used as
163 * an index buffer. The reason is that VGT DMA index fetching doesn't
169 struct r600_transfer
{
170 struct pipe_transfer transfer
;
171 struct r600_resource
*staging
;
175 struct r600_fmask_info
{
180 unsigned bank_height
;
181 unsigned slice_tile_max
;
182 unsigned tile_mode_index
;
185 struct r600_cmask_info
{
189 unsigned slice_tile_max
;
190 unsigned base_address_reg
;
193 struct r600_texture
{
194 struct r600_resource resource
;
197 unsigned pitch_override
;
199 unsigned dirty_level_mask
; /* each bit says if that mipmap is compressed */
200 struct r600_texture
*flushed_depth_texture
;
201 boolean is_flushing_texture
;
202 struct radeon_surf surface
;
204 /* Colorbuffer compression and fast clear. */
205 struct r600_fmask_info fmask
;
206 struct r600_cmask_info cmask
;
207 struct r600_resource
*cmask_buffer
;
208 unsigned cb_color_info
; /* fast clear enable bit */
209 unsigned color_clear_value
[2];
211 /* Depth buffer compression and fast clear. */
212 struct r600_resource
*htile_buffer
;
213 bool depth_cleared
; /* if it was cleared at least once */
214 float depth_clear_value
;
216 bool non_disp_tiling
; /* R600-Cayman only */
219 struct r600_surface
{
220 struct pipe_surface base
;
222 bool color_initialized
;
223 bool depth_initialized
;
225 /* Misc. color flags. */
226 bool alphatest_bypass
;
229 /* Color registers. */
230 unsigned cb_color_info
;
231 unsigned cb_color_base
;
232 unsigned cb_color_view
;
233 unsigned cb_color_size
; /* R600 only */
234 unsigned cb_color_dim
; /* EG only */
235 unsigned cb_color_pitch
; /* EG and later */
236 unsigned cb_color_slice
; /* EG and later */
237 unsigned cb_color_attrib
; /* EG and later */
238 unsigned cb_color_fmask
; /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
239 unsigned cb_color_fmask_slice
; /* EG and later */
240 unsigned cb_color_cmask
; /* CB_COLORn_TILE (r600 only) */
241 unsigned cb_color_mask
; /* R600 only */
242 struct r600_resource
*cb_buffer_fmask
; /* Used for FMASK relocations. R600 only */
243 struct r600_resource
*cb_buffer_cmask
; /* Used for CMASK relocations. R600 only */
246 unsigned db_depth_info
; /* R600 only, then SI and later */
247 unsigned db_z_info
; /* EG and later */
248 unsigned db_depth_base
; /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
249 unsigned db_depth_view
;
250 unsigned db_depth_size
;
251 unsigned db_depth_slice
; /* EG and later */
252 unsigned db_stencil_base
; /* EG and later */
253 unsigned db_stencil_info
; /* EG and later */
254 unsigned db_prefetch_limit
; /* R600 only */
255 unsigned db_htile_surface
;
256 unsigned db_htile_data_base
;
257 unsigned db_preload_control
; /* EG and later */
258 unsigned pa_su_poly_offset_db_fmt_cntl
;
261 struct r600_tiling_info
{
262 unsigned num_channels
;
264 unsigned group_bytes
;
267 struct r600_common_screen
{
268 struct pipe_screen b
;
269 struct radeon_winsys
*ws
;
270 enum radeon_family family
;
271 enum chip_class chip_class
;
272 struct radeon_info info
;
273 struct r600_tiling_info tiling_info
;
274 unsigned debug_flags
;
278 /* Auxiliary context. Mainly used to initialize resources.
279 * It must be locked prior to using and flushed before unlocking. */
280 struct pipe_context
*aux_context
;
281 pipe_mutex aux_context_lock
;
283 struct r600_resource
*trace_bo
;
287 /* GPU load thread. */
288 pipe_mutex gpu_load_mutex
;
289 pipe_thread gpu_load_thread
;
290 unsigned gpu_load_counter_busy
;
291 unsigned gpu_load_counter_idle
;
292 volatile unsigned gpu_load_stop_thread
; /* bool */
295 /* This encapsulates a state or an operation which can emitted into the GPU
298 void (*emit
)(struct r600_common_context
*ctx
, struct r600_atom
*state
);
303 struct r600_so_target
{
304 struct pipe_stream_output_target b
;
306 /* The buffer where BUFFER_FILLED_SIZE is stored. */
307 struct r600_resource
*buf_filled_size
;
308 unsigned buf_filled_size_offset
;
309 bool buf_filled_size_valid
;
311 unsigned stride_in_dw
;
314 struct r600_streamout
{
315 struct r600_atom begin_atom
;
317 unsigned num_dw_for_end
;
319 unsigned enabled_mask
;
320 unsigned num_targets
;
321 struct r600_so_target
*targets
[PIPE_MAX_SO_BUFFERS
];
323 unsigned append_bitmask
;
326 /* External state which comes from the vertex shader,
327 * it must be set explicitly when binding a shader. */
328 unsigned *stride_in_dw
;
330 /* The state of VGT_STRMOUT_(CONFIG|EN). */
331 struct r600_atom enable_atom
;
332 bool streamout_enabled
;
333 bool prims_gen_query_enabled
;
334 int num_prims_gen_queries
;
338 struct radeon_winsys_cs
*cs
;
340 void (*flush
)(void *ctx
, unsigned flags
,
341 struct pipe_fence_handle
**fence
);
345 struct r600_ring gfx
;
346 struct r600_ring dma
;
349 struct r600_common_context
{
350 struct pipe_context b
; /* base class */
352 struct r600_common_screen
*screen
;
353 struct radeon_winsys
*ws
;
354 enum radeon_family family
;
355 enum chip_class chip_class
;
356 struct r600_rings rings
;
357 unsigned initial_gfx_cs_size
;
358 unsigned gpu_reset_counter
;
360 struct u_upload_mgr
*uploader
;
361 struct u_suballocator
*allocator_so_filled_size
;
362 struct util_slab_mempool pool_transfers
;
364 /* Current unaccounted memory usage. */
369 struct r600_streamout streamout
;
371 /* Additional context states. */
372 unsigned flags
; /* flush flags */
375 /* The list of active queries. Only one query of each type can be active. */
376 int num_occlusion_queries
;
377 /* Keep track of non-timer queries, because they should be suspended
378 * during context flushing.
379 * The timer queries (TIME_ELAPSED) shouldn't be suspended. */
380 struct list_head active_nontimer_queries
;
381 unsigned num_cs_dw_nontimer_queries_suspend
;
382 /* If queries have been suspended. */
383 bool nontimer_queries_suspended
;
384 /* Additional hardware info. */
385 unsigned backend_mask
;
386 unsigned max_db
; /* for OQ */
388 unsigned num_draw_calls
;
390 /* Render condition. */
391 struct pipe_query
*current_render_cond
;
392 unsigned current_render_cond_mode
;
393 boolean current_render_cond_cond
;
394 boolean predicate_drawing
;
395 /* For context flushing. */
396 struct pipe_query
*saved_render_cond
;
397 boolean saved_render_cond_cond
;
398 unsigned saved_render_cond_mode
;
400 /* MSAA sample locations.
401 * The first index is the sample index.
402 * The second index is the coordinate: X, Y. */
403 float sample_locations_1x
[1][2];
404 float sample_locations_2x
[2][2];
405 float sample_locations_4x
[4][2];
406 float sample_locations_8x
[8][2];
407 float sample_locations_16x
[16][2];
409 /* The list of all texture buffer objects in this context.
410 * This list is walked when a buffer is invalidated/reallocated and
411 * the GPU addresses are updated. */
412 struct list_head texture_buffers
;
414 /* Copy one resource to another using async DMA. */
415 void (*dma_copy
)(struct pipe_context
*ctx
,
416 struct pipe_resource
*dst
,
418 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
419 struct pipe_resource
*src
,
421 const struct pipe_box
*src_box
);
423 void (*clear_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*dst
,
424 unsigned offset
, unsigned size
, unsigned value
,
425 bool is_framebuffer
);
427 void (*blit_decompress_depth
)(struct pipe_context
*ctx
,
428 struct r600_texture
*texture
,
429 struct r600_texture
*staging
,
430 unsigned first_level
, unsigned last_level
,
431 unsigned first_layer
, unsigned last_layer
,
432 unsigned first_sample
, unsigned last_sample
);
434 /* Reallocate the buffer and update all resource bindings where
435 * the buffer is bound, including all resource descriptors. */
436 void (*invalidate_buffer
)(struct pipe_context
*ctx
, struct pipe_resource
*buf
);
438 /* Enable or disable occlusion queries. */
439 void (*set_occlusion_query_state
)(struct pipe_context
*ctx
, bool enable
);
441 /* This ensures there is enough space in the command stream. */
442 void (*need_gfx_cs_space
)(struct pipe_context
*ctx
, unsigned num_dw
,
443 bool include_draw_vbo
);
447 boolean
r600_rings_is_buffer_referenced(struct r600_common_context
*ctx
,
448 struct radeon_winsys_cs_handle
*buf
,
449 enum radeon_bo_usage usage
);
450 void *r600_buffer_map_sync_with_rings(struct r600_common_context
*ctx
,
451 struct r600_resource
*resource
,
453 bool r600_init_resource(struct r600_common_screen
*rscreen
,
454 struct r600_resource
*res
,
455 unsigned size
, unsigned alignment
,
456 bool use_reusable_pool
);
457 struct pipe_resource
*r600_buffer_create(struct pipe_screen
*screen
,
458 const struct pipe_resource
*templ
,
460 struct pipe_resource
*
461 r600_buffer_from_user_memory(struct pipe_screen
*screen
,
462 const struct pipe_resource
*templ
,
465 /* r600_common_pipe.c */
466 void r600_draw_rectangle(struct blitter_context
*blitter
,
467 int x1
, int y1
, int x2
, int y2
, float depth
,
468 enum blitter_attrib_type type
,
469 const union pipe_color_union
*attrib
);
470 bool r600_common_screen_init(struct r600_common_screen
*rscreen
,
471 struct radeon_winsys
*ws
);
472 void r600_destroy_common_screen(struct r600_common_screen
*rscreen
);
473 void r600_preflush_suspend_features(struct r600_common_context
*ctx
);
474 void r600_postflush_resume_features(struct r600_common_context
*ctx
);
475 bool r600_common_context_init(struct r600_common_context
*rctx
,
476 struct r600_common_screen
*rscreen
);
477 void r600_common_context_cleanup(struct r600_common_context
*rctx
);
478 void r600_context_add_resource_size(struct pipe_context
*ctx
, struct pipe_resource
*r
);
479 bool r600_can_dump_shader(struct r600_common_screen
*rscreen
,
480 const struct tgsi_token
*tokens
);
481 void r600_screen_clear_buffer(struct r600_common_screen
*rscreen
, struct pipe_resource
*dst
,
482 unsigned offset
, unsigned size
, unsigned value
,
483 bool is_framebuffer
);
484 struct pipe_resource
*r600_resource_create_common(struct pipe_screen
*screen
,
485 const struct pipe_resource
*templ
);
486 const char *r600_get_llvm_processor_name(enum radeon_family family
);
487 void r600_need_dma_space(struct r600_common_context
*ctx
, unsigned num_dw
);
489 /* r600_gpu_load.c */
490 void r600_gpu_load_kill_thread(struct r600_common_screen
*rscreen
);
491 uint64_t r600_gpu_load_begin(struct r600_common_screen
*rscreen
);
492 unsigned r600_gpu_load_end(struct r600_common_screen
*rscreen
, uint64_t begin
);
495 void r600_query_init(struct r600_common_context
*rctx
);
496 void r600_suspend_nontimer_queries(struct r600_common_context
*ctx
);
497 void r600_resume_nontimer_queries(struct r600_common_context
*ctx
);
498 void r600_query_init_backend_mask(struct r600_common_context
*ctx
);
500 /* r600_streamout.c */
501 void r600_streamout_buffers_dirty(struct r600_common_context
*rctx
);
502 void r600_set_streamout_targets(struct pipe_context
*ctx
,
503 unsigned num_targets
,
504 struct pipe_stream_output_target
**targets
,
505 const unsigned *offset
);
506 void r600_emit_streamout_end(struct r600_common_context
*rctx
);
507 void r600_update_prims_generated_query_state(struct r600_common_context
*rctx
,
508 unsigned type
, int diff
);
509 void r600_streamout_init(struct r600_common_context
*rctx
);
512 void r600_texture_get_fmask_info(struct r600_common_screen
*rscreen
,
513 struct r600_texture
*rtex
,
515 struct r600_fmask_info
*out
);
516 void r600_texture_get_cmask_info(struct r600_common_screen
*rscreen
,
517 struct r600_texture
*rtex
,
518 struct r600_cmask_info
*out
);
519 bool r600_init_flushed_depth_texture(struct pipe_context
*ctx
,
520 struct pipe_resource
*texture
,
521 struct r600_texture
**staging
);
522 struct pipe_resource
*r600_texture_create(struct pipe_screen
*screen
,
523 const struct pipe_resource
*templ
);
524 struct pipe_surface
*r600_create_surface_custom(struct pipe_context
*pipe
,
525 struct pipe_resource
*texture
,
526 const struct pipe_surface
*templ
,
527 unsigned width
, unsigned height
);
528 unsigned r600_translate_colorswap(enum pipe_format format
);
529 void evergreen_do_fast_color_clear(struct r600_common_context
*rctx
,
530 struct pipe_framebuffer_state
*fb
,
531 struct r600_atom
*fb_state
,
533 const union pipe_color_union
*color
);
534 void r600_init_screen_texture_functions(struct r600_common_screen
*rscreen
);
535 void r600_init_context_texture_functions(struct r600_common_context
*rctx
);
538 extern const uint32_t eg_sample_locs_2x
[4];
539 extern const unsigned eg_max_dist_2x
;
540 extern const uint32_t eg_sample_locs_4x
[4];
541 extern const unsigned eg_max_dist_4x
;
542 void cayman_get_sample_position(struct pipe_context
*ctx
, unsigned sample_count
,
543 unsigned sample_index
, float *out_value
);
544 void cayman_init_msaa(struct pipe_context
*ctx
);
545 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs
*cs
, int nr_samples
);
546 void cayman_emit_msaa_config(struct radeon_winsys_cs
*cs
, int nr_samples
,
547 int ps_iter_samples
, int overrast_samples
);
550 /* Inline helpers. */
552 static inline struct r600_resource
*r600_resource(struct pipe_resource
*r
)
554 return (struct r600_resource
*)r
;
558 r600_resource_reference(struct r600_resource
**ptr
, struct r600_resource
*res
)
560 pipe_resource_reference((struct pipe_resource
**)ptr
,
561 (struct pipe_resource
*)res
);
564 static inline unsigned r600_tex_aniso_filter(unsigned filter
)
566 if (filter
<= 1) return 0;
567 if (filter
<= 2) return 1;
568 if (filter
<= 4) return 2;
569 if (filter
<= 8) return 3;
573 static inline unsigned r600_wavefront_size(enum radeon_family family
)
593 #define COMPUTE_DBG(rscreen, fmt, args...) \
595 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
598 #define R600_ERR(fmt, args...) \
599 fprintf(stderr, "EE %s:%d %s - "fmt, __FILE__, __LINE__, __func__, ##args)
601 /* For MSAA sample positions. */
602 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
603 (((s0x) & 0xf) | (((s0y) & 0xf) << 4) | \
604 (((s1x) & 0xf) << 8) | (((s1y) & 0xf) << 12) | \
605 (((s2x) & 0xf) << 16) | (((s2y) & 0xf) << 20) | \
606 (((s3x) & 0xf) << 24) | (((s3y) & 0xf) << 28))