• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright 2013 Advanced Micro Devices, Inc.
3  *
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:
10  *
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
13  * Software.
14  *
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
21  * SOFTWARE.
22  *
23  * Authors: Marek Olšák <maraeo@gmail.com>
24  *
25  */
26 
27 /**
28  * This file contains common screen and context structures and functions
29  * for r600g and radeonsi.
30  */
31 
32 #ifndef R600_PIPE_COMMON_H
33 #define R600_PIPE_COMMON_H
34 
35 #include <stdio.h>
36 
37 #include "amd/common/ac_binary.h"
38 
39 #include "radeon/radeon_winsys.h"
40 
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"
49 
50 struct u_log_context;
51 
52 #define ATI_VENDOR_ID 0x1002
53 
54 #define R600_RESOURCE_FLAG_TRANSFER		(PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
55 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH	(PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
56 #define R600_RESOURCE_FLAG_FORCE_TILING		(PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
57 #define R600_RESOURCE_FLAG_UNMAPPABLE		(PIPE_RESOURCE_FLAG_DRV_PRIV << 4)
58 
59 #define R600_CONTEXT_STREAMOUT_FLUSH		(1u << 0)
60 /* Pipeline & streamout query controls. */
61 #define R600_CONTEXT_START_PIPELINE_STATS	(1u << 1)
62 #define R600_CONTEXT_STOP_PIPELINE_STATS	(1u << 2)
63 #define R600_CONTEXT_FLUSH_FOR_RENDER_COND	(1u << 3)
64 #define R600_CONTEXT_PRIVATE_FLAG		(1u << 4)
65 
66 /* special primitive types */
67 #define R600_PRIM_RECTANGLE_LIST	PIPE_PRIM_MAX
68 
69 #define R600_NOT_QUERY		0xffffffff
70 
71 /* Debug flags. */
72 #define DBG_VS			(1 << PIPE_SHADER_VERTEX)
73 #define DBG_PS			(1 << PIPE_SHADER_FRAGMENT)
74 #define DBG_GS			(1 << PIPE_SHADER_GEOMETRY)
75 #define DBG_TCS			(1 << PIPE_SHADER_TESS_CTRL)
76 #define DBG_TES			(1 << PIPE_SHADER_TESS_EVAL)
77 #define DBG_CS			(1 << PIPE_SHADER_COMPUTE)
78 #define DBG_ALL_SHADERS		(DBG_FS - 1)
79 #define DBG_FS			(1 << 6) /* fetch shader */
80 #define DBG_TEX			(1 << 7)
81 #define DBG_NIR			(1 << 8)
82 #define DBG_COMPUTE		(1 << 9)
83 /* gap */
84 #define DBG_VM			(1 << 11)
85 #define DBG_NO_IR		(1 << 12)
86 #define DBG_NO_TGSI		(1 << 13)
87 #define DBG_NO_ASM		(1 << 14)
88 #define DBG_PREOPT_IR		(1 << 15)
89 #define DBG_CHECK_IR		(1 << 16)
90 #define DBG_NO_OPT_VARIANT	(1 << 17)
91 #define DBG_FS_CORRECT_DERIVS_AFTER_KILL (1 << 18)
92 /* gaps */
93 #define DBG_TEST_DMA		(1 << 20)
94 /* Bits 21-31 are reserved for the r600g driver. */
95 /* features */
96 #define DBG_NO_ASYNC_DMA	(1ull << 32)
97 #define DBG_NO_HYPERZ		(1ull << 33)
98 #define DBG_NO_DISCARD_RANGE	(1ull << 34)
99 #define DBG_NO_2D_TILING	(1ull << 35)
100 #define DBG_NO_TILING		(1ull << 36)
101 #define DBG_SWITCH_ON_EOP	(1ull << 37)
102 #define DBG_FORCE_DMA		(1ull << 38)
103 #define DBG_PRECOMPILE		(1ull << 39)
104 #define DBG_INFO		(1ull << 40)
105 #define DBG_NO_WC		(1ull << 41)
106 #define DBG_CHECK_VM		(1ull << 42)
107 /* gap */
108 #define DBG_UNSAFE_MATH		(1ull << 49)
109 #define DBG_TEST_VMFAULT_CP	(1ull << 51)
110 #define DBG_TEST_VMFAULT_SDMA	(1ull << 52)
111 #define DBG_TEST_VMFAULT_SHADER	(1ull << 53)
112 
113 #define R600_MAP_BUFFER_ALIGNMENT 64
114 #define R600_MAX_VIEWPORTS        16
115 
116 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 1024
117 
118 enum r600_coherency {
119 	R600_COHERENCY_NONE, /* no cache flushes needed */
120 	R600_COHERENCY_SHADER,
121 	R600_COHERENCY_CB_META,
122 };
123 
124 #ifdef PIPE_ARCH_BIG_ENDIAN
125 #define R600_BIG_ENDIAN 1
126 #else
127 #define R600_BIG_ENDIAN 0
128 #endif
129 
130 struct r600_common_context;
131 struct r600_perfcounters;
132 struct tgsi_shader_info;
133 struct r600_qbo_state;
134 
135 void radeon_shader_binary_init(struct ac_shader_binary *b);
136 void radeon_shader_binary_clean(struct ac_shader_binary *b);
137 
138 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
139  * at the moment.
140  */
141 struct r600_resource {
142 	struct threaded_resource	b;
143 
144 	/* Winsys objects. */
145 	struct pb_buffer		*buf;
146 	uint64_t			gpu_address;
147 	/* Memory usage if the buffer placement is optimal. */
148 	uint64_t			vram_usage;
149 	uint64_t			gart_usage;
150 
151 	/* Resource properties. */
152 	uint64_t			bo_size;
153 	unsigned			bo_alignment;
154 	enum radeon_bo_domain		domains;
155 	enum radeon_bo_flag		flags;
156 	unsigned			bind_history;
157 
158 	/* The buffer range which is initialized (with a write transfer,
159 	 * streamout, DMA, or as a random access target). The rest of
160 	 * the buffer is considered invalid and can be mapped unsynchronized.
161 	 *
162 	 * This allows unsychronized mapping of a buffer range which hasn't
163 	 * been used yet. It's for applications which forget to use
164 	 * the unsynchronized map flag and expect the driver to figure it out.
165          */
166 	struct util_range		valid_buffer_range;
167 
168 	/* Whether the resource has been exported via resource_get_handle. */
169 	unsigned			external_usage; /* PIPE_HANDLE_USAGE_* */
170 
171 	/* Whether this resource is referenced by bindless handles. */
172 	bool				texture_handle_allocated;
173 	bool				image_handle_allocated;
174 
175 	/*
176 	 * EG/Cayman only - for RAT operations hw need an immediate buffer
177 	 * to store results in.
178 	 */
179 	struct r600_resource            *immed_buffer;
180 };
181 
182 struct r600_transfer {
183 	struct threaded_transfer	b;
184 	struct r600_resource		*staging;
185 	unsigned			offset;
186 };
187 
188 struct r600_fmask_info {
189 	uint64_t offset;
190 	uint64_t size;
191 	unsigned alignment;
192 	unsigned pitch_in_pixels;
193 	unsigned bank_height;
194 	unsigned slice_tile_max;
195 	unsigned tile_mode_index;
196 	unsigned tile_swizzle;
197 };
198 
199 struct r600_cmask_info {
200 	uint64_t offset;
201 	uint64_t size;
202 	unsigned alignment;
203 	unsigned slice_tile_max;
204 	uint64_t base_address_reg;
205 };
206 
207 struct r600_texture {
208 	struct r600_resource		resource;
209 
210 	uint64_t			size;
211 	unsigned			num_level0_transfers;
212 	enum pipe_format		db_render_format;
213 	bool				is_depth;
214 	bool				db_compatible;
215 	bool				can_sample_z;
216 	bool				can_sample_s;
217 	unsigned			dirty_level_mask; /* each bit says if that mipmap is compressed */
218 	unsigned			stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
219 	struct r600_texture		*flushed_depth_texture;
220 	struct radeon_surf		surface;
221 
222 	/* Colorbuffer compression and fast clear. */
223 	struct r600_fmask_info		fmask;
224 	struct r600_cmask_info		cmask;
225 	struct r600_resource		*cmask_buffer;
226 	unsigned			cb_color_info; /* fast clear enable bit */
227 	unsigned			color_clear_value[2];
228 	unsigned			last_msaa_resolve_target_micro_mode;
229 
230 	/* Depth buffer compression and fast clear. */
231 	uint64_t			htile_offset;
232 	bool				depth_cleared; /* if it was cleared at least once */
233 	float				depth_clear_value;
234 	bool				stencil_cleared; /* if it was cleared at least once */
235 	uint8_t				stencil_clear_value;
236 
237 	bool				non_disp_tiling; /* R600-Cayman only */
238 
239 	/* Counter that should be non-zero if the texture is bound to a
240 	 * framebuffer. Implemented in radeonsi only.
241 	 */
242 	uint32_t			framebuffers_bound;
243 };
244 
245 struct r600_surface {
246 	struct pipe_surface		base;
247 
248 	/* These can vary with block-compressed textures. */
249 	unsigned width0;
250 	unsigned height0;
251 
252 	bool color_initialized;
253 	bool depth_initialized;
254 
255 	/* Misc. color flags. */
256 	bool alphatest_bypass;
257 	bool export_16bpc;
258 	bool color_is_int8;
259 	bool color_is_int10;
260 
261 	/* Color registers. */
262 	unsigned cb_color_info;
263 	unsigned cb_color_base;
264 	unsigned cb_color_view;
265 	unsigned cb_color_size;		/* R600 only */
266 	unsigned cb_color_dim;		/* EG only */
267 	unsigned cb_color_pitch;	/* EG and later */
268 	unsigned cb_color_slice;	/* EG and later */
269 	unsigned cb_color_attrib;	/* EG and later */
270 	unsigned cb_color_fmask;	/* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
271 	unsigned cb_color_fmask_slice;	/* EG and later */
272 	unsigned cb_color_cmask;	/* CB_COLORn_TILE (r600 only) */
273 	unsigned cb_color_mask;		/* R600 only */
274 	struct r600_resource *cb_buffer_fmask; /* Used for FMASK relocations. R600 only */
275 	struct r600_resource *cb_buffer_cmask; /* Used for CMASK relocations. R600 only */
276 
277 	/* DB registers. */
278 	uint64_t db_depth_base;		/* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
279 	uint64_t db_stencil_base;	/* EG and later */
280 	uint64_t db_htile_data_base;
281 	unsigned db_depth_info;		/* R600 only, then SI and later */
282 	unsigned db_z_info;		/* EG and later */
283 	unsigned db_depth_view;
284 	unsigned db_depth_size;
285 	unsigned db_depth_slice;	/* EG and later */
286 	unsigned db_stencil_info;	/* EG and later */
287 	unsigned db_prefetch_limit;	/* R600 only */
288 	unsigned db_htile_surface;
289 	unsigned db_preload_control;	/* EG and later */
290 };
291 
292 struct r600_mmio_counter {
293 	unsigned busy;
294 	unsigned idle;
295 };
296 
297 union r600_mmio_counters {
298 	struct {
299 		/* For global GPU load including SDMA. */
300 		struct r600_mmio_counter gpu;
301 
302 		/* GRBM_STATUS */
303 		struct r600_mmio_counter spi;
304 		struct r600_mmio_counter gui;
305 		struct r600_mmio_counter ta;
306 		struct r600_mmio_counter gds;
307 		struct r600_mmio_counter vgt;
308 		struct r600_mmio_counter ia;
309 		struct r600_mmio_counter sx;
310 		struct r600_mmio_counter wd;
311 		struct r600_mmio_counter bci;
312 		struct r600_mmio_counter sc;
313 		struct r600_mmio_counter pa;
314 		struct r600_mmio_counter db;
315 		struct r600_mmio_counter cp;
316 		struct r600_mmio_counter cb;
317 
318 		/* SRBM_STATUS2 */
319 		struct r600_mmio_counter sdma;
320 
321 		/* CP_STAT */
322 		struct r600_mmio_counter pfp;
323 		struct r600_mmio_counter meq;
324 		struct r600_mmio_counter me;
325 		struct r600_mmio_counter surf_sync;
326 		struct r600_mmio_counter cp_dma;
327 		struct r600_mmio_counter scratch_ram;
328 	} named;
329 	unsigned array[0];
330 };
331 
332 struct r600_memory_object {
333 	struct pipe_memory_object	b;
334 	struct pb_buffer		*buf;
335 	uint32_t			stride;
336 	uint32_t			offset;
337 };
338 
339 struct r600_common_screen {
340 	struct pipe_screen		b;
341 	struct radeon_winsys		*ws;
342 	enum radeon_family		family;
343 	enum chip_class			chip_class;
344 	struct radeon_info		info;
345 	uint64_t			debug_flags;
346 	bool				has_cp_dma;
347 	bool				has_streamout;
348 
349 	struct disk_cache		*disk_shader_cache;
350 
351 	struct slab_parent_pool		pool_transfers;
352 
353 	/* Texture filter settings. */
354 	int				force_aniso; /* -1 = disabled */
355 
356 	/* Auxiliary context. Mainly used to initialize resources.
357 	 * It must be locked prior to using and flushed before unlocking. */
358 	struct pipe_context		*aux_context;
359 	mtx_t				aux_context_lock;
360 
361 	/* This must be in the screen, because UE4 uses one context for
362 	 * compilation and another one for rendering.
363 	 */
364 	unsigned			num_compilations;
365 	/* Along with ST_DEBUG=precompile, this should show if applications
366 	 * are loading shaders on demand. This is a monotonic counter.
367 	 */
368 	unsigned			num_shaders_created;
369 	unsigned			num_shader_cache_hits;
370 
371 	/* GPU load thread. */
372 	mtx_t				gpu_load_mutex;
373 	thrd_t				gpu_load_thread;
374 	union r600_mmio_counters	mmio_counters;
375 	volatile unsigned		gpu_load_stop_thread; /* bool */
376 
377 	char				renderer_string[100];
378 
379 	/* Performance counters. */
380 	struct r600_perfcounters	*perfcounters;
381 
382 	/* If pipe_screen wants to recompute and re-emit the framebuffer,
383 	 * sampler, and image states of all contexts, it should atomically
384 	 * increment this.
385 	 *
386 	 * Each context will compare this with its own last known value of
387 	 * the counter before drawing and re-emit the states accordingly.
388 	 */
389 	unsigned			dirty_tex_counter;
390 
391 	/* Atomically increment this counter when an existing texture's
392 	 * metadata is enabled or disabled in a way that requires changing
393 	 * contexts' compressed texture binding masks.
394 	 */
395 	unsigned			compressed_colortex_counter;
396 
397 	struct {
398 		/* Context flags to set so that all writes from earlier jobs
399 		 * in the CP are seen by L2 clients.
400 		 */
401 		unsigned cp_to_L2;
402 
403 		/* Context flags to set so that all writes from earlier jobs
404 		 * that end in L2 are seen by CP.
405 		 */
406 		unsigned L2_to_cp;
407 
408 		/* Context flags to set so that all writes from earlier
409 		 * compute jobs are seen by L2 clients.
410 		 */
411 		unsigned compute_to_L2;
412 	} barrier_flags;
413 
414 	void (*query_opaque_metadata)(struct r600_common_screen *rscreen,
415 				      struct r600_texture *rtex,
416 				      struct radeon_bo_metadata *md);
417 
418 	void (*apply_opaque_metadata)(struct r600_common_screen *rscreen,
419 				    struct r600_texture *rtex,
420 				    struct radeon_bo_metadata *md);
421 };
422 
423 /* This encapsulates a state or an operation which can emitted into the GPU
424  * command stream. */
425 struct r600_atom {
426 	void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
427 	unsigned		num_dw;
428 	unsigned short		id;
429 };
430 
431 struct r600_so_target {
432 	struct pipe_stream_output_target b;
433 
434 	/* The buffer where BUFFER_FILLED_SIZE is stored. */
435 	struct r600_resource	*buf_filled_size;
436 	unsigned		buf_filled_size_offset;
437 	bool			buf_filled_size_valid;
438 
439 	unsigned		stride_in_dw;
440 };
441 
442 struct r600_streamout {
443 	struct r600_atom		begin_atom;
444 	bool				begin_emitted;
445 	unsigned			num_dw_for_end;
446 
447 	unsigned			enabled_mask;
448 	unsigned			num_targets;
449 	struct r600_so_target		*targets[PIPE_MAX_SO_BUFFERS];
450 
451 	unsigned			append_bitmask;
452 	bool				suspended;
453 
454 	/* External state which comes from the vertex shader,
455 	 * it must be set explicitly when binding a shader. */
456 	uint16_t			*stride_in_dw;
457 	unsigned			enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
458 
459 	/* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
460 	unsigned			hw_enabled_mask;
461 
462 	/* The state of VGT_STRMOUT_(CONFIG|EN). */
463 	struct r600_atom		enable_atom;
464 	bool				streamout_enabled;
465 	bool				prims_gen_query_enabled;
466 	int				num_prims_gen_queries;
467 };
468 
469 struct r600_signed_scissor {
470 	int minx;
471 	int miny;
472 	int maxx;
473 	int maxy;
474 };
475 
476 struct r600_scissors {
477 	struct r600_atom		atom;
478 	unsigned			dirty_mask;
479 	struct pipe_scissor_state	states[R600_MAX_VIEWPORTS];
480 };
481 
482 struct r600_viewports {
483 	struct r600_atom		atom;
484 	unsigned			dirty_mask;
485 	unsigned			depth_range_dirty_mask;
486 	struct pipe_viewport_state	states[R600_MAX_VIEWPORTS];
487 	struct r600_signed_scissor	as_scissor[R600_MAX_VIEWPORTS];
488 };
489 
490 struct r600_ring {
491 	struct radeon_winsys_cs		*cs;
492 	void (*flush)(void *ctx, unsigned flags,
493 		      struct pipe_fence_handle **fence);
494 };
495 
496 /* Saved CS data for debugging features. */
497 struct radeon_saved_cs {
498 	uint32_t			*ib;
499 	unsigned			num_dw;
500 
501 	struct radeon_bo_list_item	*bo_list;
502 	unsigned			bo_count;
503 };
504 
505 struct r600_common_context {
506 	struct pipe_context b; /* base class */
507 
508 	struct r600_common_screen	*screen;
509 	struct radeon_winsys		*ws;
510 	struct radeon_winsys_ctx	*ctx;
511 	enum radeon_family		family;
512 	enum chip_class			chip_class;
513 	struct r600_ring		gfx;
514 	struct r600_ring		dma;
515 	struct pipe_fence_handle	*last_gfx_fence;
516 	struct pipe_fence_handle	*last_sdma_fence;
517 	struct r600_resource		*eop_bug_scratch;
518 	unsigned			num_gfx_cs_flushes;
519 	unsigned			initial_gfx_cs_size;
520 	unsigned			gpu_reset_counter;
521 	unsigned			last_dirty_tex_counter;
522 	unsigned			last_compressed_colortex_counter;
523 	unsigned			last_num_draw_calls;
524 
525 	struct threaded_context		*tc;
526 	struct u_suballocator		*allocator_zeroed_memory;
527 	struct slab_child_pool		pool_transfers;
528 	struct slab_child_pool		pool_transfers_unsync; /* for threaded_context */
529 
530 	/* Current unaccounted memory usage. */
531 	uint64_t			vram;
532 	uint64_t			gtt;
533 
534 	/* States. */
535 	struct r600_streamout		streamout;
536 	struct r600_scissors		scissors;
537 	struct r600_viewports		viewports;
538 	bool				scissor_enabled;
539 	bool				clip_halfz;
540 	bool				vs_writes_viewport_index;
541 	bool				vs_disables_clipping_viewport;
542 
543 	/* Additional context states. */
544 	unsigned flags; /* flush flags */
545 
546 	/* Queries. */
547 	/* Maintain the list of active queries for pausing between IBs. */
548 	int				num_occlusion_queries;
549 	int				num_perfect_occlusion_queries;
550 	struct list_head		active_queries;
551 	unsigned			num_cs_dw_queries_suspend;
552 	/* Misc stats. */
553 	unsigned			num_draw_calls;
554 	unsigned			num_decompress_calls;
555 	unsigned			num_mrt_draw_calls;
556 	unsigned			num_prim_restart_calls;
557 	unsigned			num_spill_draw_calls;
558 	unsigned			num_compute_calls;
559 	unsigned			num_spill_compute_calls;
560 	unsigned			num_dma_calls;
561 	unsigned			num_cp_dma_calls;
562 	unsigned			num_vs_flushes;
563 	unsigned			num_ps_flushes;
564 	unsigned			num_cs_flushes;
565 	unsigned			num_cb_cache_flushes;
566 	unsigned			num_db_cache_flushes;
567 	unsigned			num_resident_handles;
568 	uint64_t			num_alloc_tex_transfer_bytes;
569 
570 	/* Render condition. */
571 	struct r600_atom		render_cond_atom;
572 	struct pipe_query		*render_cond;
573 	unsigned			render_cond_mode;
574 	bool				render_cond_invert;
575 	bool				render_cond_force_off; /* for u_blitter */
576 
577 	/* MSAA sample locations.
578 	 * The first index is the sample index.
579 	 * The second index is the coordinate: X, Y. */
580 	float				sample_locations_1x[1][2];
581 	float				sample_locations_2x[2][2];
582 	float				sample_locations_4x[4][2];
583 	float				sample_locations_8x[8][2];
584 	float				sample_locations_16x[16][2];
585 
586 	struct pipe_debug_callback	debug;
587 	struct pipe_device_reset_callback device_reset_callback;
588 	struct u_log_context		*log;
589 
590 	void				*query_result_shader;
591 
592 	/* Copy one resource to another using async DMA. */
593 	void (*dma_copy)(struct pipe_context *ctx,
594 			 struct pipe_resource *dst,
595 			 unsigned dst_level,
596 			 unsigned dst_x, unsigned dst_y, unsigned dst_z,
597 			 struct pipe_resource *src,
598 			 unsigned src_level,
599 			 const struct pipe_box *src_box);
600 
601 	void (*dma_clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
602 				 uint64_t offset, uint64_t size, unsigned value);
603 
604 	void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
605 			     uint64_t offset, uint64_t size, unsigned value,
606 			     enum r600_coherency coher);
607 
608 	void (*blit_decompress_depth)(struct pipe_context *ctx,
609 				      struct r600_texture *texture,
610 				      struct r600_texture *staging,
611 				      unsigned first_level, unsigned last_level,
612 				      unsigned first_layer, unsigned last_layer,
613 				      unsigned first_sample, unsigned last_sample);
614 
615 	/* Reallocate the buffer and update all resource bindings where
616 	 * the buffer is bound, including all resource descriptors. */
617 	void (*invalidate_buffer)(struct pipe_context *ctx, struct pipe_resource *buf);
618 
619 	/* Update all resource bindings where the buffer is bound, including
620 	 * all resource descriptors. This is invalidate_buffer without
621 	 * the invalidation. */
622 	void (*rebind_buffer)(struct pipe_context *ctx, struct pipe_resource *buf,
623 			      uint64_t old_gpu_address);
624 
625 	void (*save_qbo_state)(struct pipe_context *ctx, struct r600_qbo_state *st);
626 
627 	/* This ensures there is enough space in the command stream. */
628 	void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
629 				  bool include_draw_vbo);
630 
631 	void (*set_atom_dirty)(struct r600_common_context *ctx,
632 			       struct r600_atom *atom, bool dirty);
633 
634 	void (*check_vm_faults)(struct r600_common_context *ctx,
635 				struct radeon_saved_cs *saved,
636 				enum ring_type ring);
637 };
638 
639 /* r600_buffer_common.c */
640 bool r600_rings_is_buffer_referenced(struct r600_common_context *ctx,
641 				     struct pb_buffer *buf,
642 				     enum radeon_bo_usage usage);
643 void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx,
644                                       struct r600_resource *resource,
645                                       unsigned usage);
646 void r600_buffer_subdata(struct pipe_context *ctx,
647 			 struct pipe_resource *buffer,
648 			 unsigned usage, unsigned offset,
649 			 unsigned size, const void *data);
650 void r600_init_resource_fields(struct r600_common_screen *rscreen,
651 			       struct r600_resource *res,
652 			       uint64_t size, unsigned alignment);
653 bool r600_alloc_resource(struct r600_common_screen *rscreen,
654 			 struct r600_resource *res);
655 struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
656 					 const struct pipe_resource *templ,
657 					 unsigned alignment);
658 struct pipe_resource * r600_aligned_buffer_create(struct pipe_screen *screen,
659 						  unsigned flags,
660 						  unsigned usage,
661 						  unsigned size,
662 						  unsigned alignment);
663 struct pipe_resource *
664 r600_buffer_from_user_memory(struct pipe_screen *screen,
665 			     const struct pipe_resource *templ,
666 			     void *user_memory);
667 void
668 r600_invalidate_resource(struct pipe_context *ctx,
669 			 struct pipe_resource *resource);
670 void r600_replace_buffer_storage(struct pipe_context *ctx,
671 				 struct pipe_resource *dst,
672 				 struct pipe_resource *src);
673 
674 /* r600_common_pipe.c */
675 void r600_gfx_write_event_eop(struct r600_common_context *ctx,
676 			      unsigned event, unsigned event_flags,
677 			      unsigned data_sel,
678 			      struct r600_resource *buf, uint64_t va,
679 			      uint32_t new_fence, unsigned query_type);
680 unsigned r600_gfx_write_fence_dwords(struct r600_common_screen *screen);
681 void r600_gfx_wait_fence(struct r600_common_context *ctx,
682 			 uint64_t va, uint32_t ref, uint32_t mask);
683 void r600_draw_rectangle(struct blitter_context *blitter,
684 			 void *vertex_elements_cso,
685 			 blitter_get_vs_func get_vs,
686 			 int x1, int y1, int x2, int y2,
687 			 float depth, unsigned num_instances,
688 			 enum blitter_attrib_type type,
689 			 const union blitter_attrib *attrib);
690 bool r600_common_screen_init(struct r600_common_screen *rscreen,
691 			     struct radeon_winsys *ws);
692 void r600_destroy_common_screen(struct r600_common_screen *rscreen);
693 void r600_preflush_suspend_features(struct r600_common_context *ctx);
694 void r600_postflush_resume_features(struct r600_common_context *ctx);
695 bool r600_common_context_init(struct r600_common_context *rctx,
696 			      struct r600_common_screen *rscreen,
697 			      unsigned context_flags);
698 void r600_common_context_cleanup(struct r600_common_context *rctx);
699 bool r600_can_dump_shader(struct r600_common_screen *rscreen,
700 			  unsigned processor);
701 bool r600_extra_shader_checks(struct r600_common_screen *rscreen,
702 			      unsigned processor);
703 void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
704 			      uint64_t offset, uint64_t size, unsigned value);
705 struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
706 						  const struct pipe_resource *templ);
707 const char *r600_get_llvm_processor_name(enum radeon_family family);
708 void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
709 			 struct r600_resource *dst, struct r600_resource *src);
710 void radeon_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
711 		    struct radeon_saved_cs *saved, bool get_buffer_list);
712 void radeon_clear_saved_cs(struct radeon_saved_cs *saved);
713 bool r600_check_device_reset(struct r600_common_context *rctx);
714 
715 /* r600_gpu_load.c */
716 void r600_gpu_load_kill_thread(struct r600_common_screen *rscreen);
717 uint64_t r600_begin_counter(struct r600_common_screen *rscreen, unsigned type);
718 unsigned r600_end_counter(struct r600_common_screen *rscreen, unsigned type,
719 			  uint64_t begin);
720 
721 /* r600_perfcounters.c */
722 void r600_perfcounters_destroy(struct r600_common_screen *rscreen);
723 
724 /* r600_query.c */
725 void r600_init_screen_query_functions(struct r600_common_screen *rscreen);
726 void r600_query_init(struct r600_common_context *rctx);
727 void r600_suspend_queries(struct r600_common_context *ctx);
728 void r600_resume_queries(struct r600_common_context *ctx);
729 void r600_query_fix_enabled_rb_mask(struct r600_common_screen *rscreen);
730 
731 /* r600_streamout.c */
732 void r600_streamout_buffers_dirty(struct r600_common_context *rctx);
733 void r600_set_streamout_targets(struct pipe_context *ctx,
734 				unsigned num_targets,
735 				struct pipe_stream_output_target **targets,
736 				const unsigned *offset);
737 void r600_emit_streamout_end(struct r600_common_context *rctx);
738 void r600_update_prims_generated_query_state(struct r600_common_context *rctx,
739 					     unsigned type, int diff);
740 void r600_streamout_init(struct r600_common_context *rctx);
741 
742 /* r600_test_dma.c */
743 void r600_test_dma(struct r600_common_screen *rscreen);
744 
745 /* r600_texture.c */
746 bool r600_prepare_for_dma_blit(struct r600_common_context *rctx,
747 				struct r600_texture *rdst,
748 				unsigned dst_level, unsigned dstx,
749 				unsigned dsty, unsigned dstz,
750 				struct r600_texture *rsrc,
751 				unsigned src_level,
752 				const struct pipe_box *src_box);
753 void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
754 				 struct r600_texture *rtex,
755 				 unsigned nr_samples,
756 				 struct r600_fmask_info *out);
757 void r600_texture_get_cmask_info(struct r600_common_screen *rscreen,
758 				 struct r600_texture *rtex,
759 				 struct r600_cmask_info *out);
760 bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
761 				     struct pipe_resource *texture,
762 				     struct r600_texture **staging);
763 void r600_print_texture_info(struct r600_common_screen *rscreen,
764 			     struct r600_texture *rtex, struct u_log_context *log);
765 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
766 					const struct pipe_resource *templ);
767 struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
768 						struct pipe_resource *texture,
769 						const struct pipe_surface *templ,
770 						unsigned width0, unsigned height0,
771 						unsigned width, unsigned height);
772 unsigned r600_translate_colorswap(enum pipe_format format, bool do_endian_swap);
773 void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
774 				   struct pipe_framebuffer_state *fb,
775 				   struct r600_atom *fb_state,
776 				   unsigned *buffers, ubyte *dirty_cbufs,
777 				   const union pipe_color_union *color);
778 void r600_init_screen_texture_functions(struct r600_common_screen *rscreen);
779 void r600_init_context_texture_functions(struct r600_common_context *rctx);
780 void eg_resource_alloc_immed(struct r600_common_screen *rscreen,
781 			     struct r600_resource *res,
782 			     unsigned immed_size);
783 
784 /* r600_viewport.c */
785 void evergreen_apply_scissor_bug_workaround(struct r600_common_context *rctx,
786 					    struct pipe_scissor_state *scissor);
787 void r600_viewport_set_rast_deps(struct r600_common_context *rctx,
788 				 bool scissor_enable, bool clip_halfz);
789 void r600_update_vs_writes_viewport_index(struct r600_common_context *rctx,
790 					  struct tgsi_shader_info *info);
791 void r600_init_viewport_functions(struct r600_common_context *rctx);
792 
793 /* cayman_msaa.c */
794 extern const uint32_t eg_sample_locs_2x[4];
795 extern const unsigned eg_max_dist_2x;
796 extern const uint32_t eg_sample_locs_4x[4];
797 extern const unsigned eg_max_dist_4x;
798 void cayman_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
799 				unsigned sample_index, float *out_value);
800 void cayman_init_msaa(struct pipe_context *ctx);
801 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
802 void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
803 			     int ps_iter_samples, int overrast_samples,
804 			     unsigned sc_mode_cntl_1);
805 
806 
807 /* Inline helpers. */
808 
r600_resource(struct pipe_resource * r)809 static inline struct r600_resource *r600_resource(struct pipe_resource *r)
810 {
811 	return (struct r600_resource*)r;
812 }
813 
814 static inline void
r600_resource_reference(struct r600_resource ** ptr,struct r600_resource * res)815 r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
816 {
817 	pipe_resource_reference((struct pipe_resource **)ptr,
818 				(struct pipe_resource *)res);
819 }
820 
821 static inline void
r600_texture_reference(struct r600_texture ** ptr,struct r600_texture * res)822 r600_texture_reference(struct r600_texture **ptr, struct r600_texture *res)
823 {
824 	pipe_resource_reference((struct pipe_resource **)ptr, &res->resource.b.b);
825 }
826 
827 static inline void
r600_context_add_resource_size(struct pipe_context * ctx,struct pipe_resource * r)828 r600_context_add_resource_size(struct pipe_context *ctx, struct pipe_resource *r)
829 {
830 	struct r600_common_context *rctx = (struct r600_common_context *)ctx;
831 	struct r600_resource *res = (struct r600_resource *)r;
832 
833 	if (res) {
834 		/* Add memory usage for need_gfx_cs_space */
835 		rctx->vram += res->vram_usage;
836 		rctx->gtt += res->gart_usage;
837 	}
838 }
839 
r600_get_strmout_en(struct r600_common_context * rctx)840 static inline bool r600_get_strmout_en(struct r600_common_context *rctx)
841 {
842 	return rctx->streamout.streamout_enabled ||
843 	       rctx->streamout.prims_gen_query_enabled;
844 }
845 
846 #define     SQ_TEX_XY_FILTER_POINT                         0x00
847 #define     SQ_TEX_XY_FILTER_BILINEAR                      0x01
848 #define     SQ_TEX_XY_FILTER_ANISO_POINT                   0x02
849 #define     SQ_TEX_XY_FILTER_ANISO_BILINEAR                0x03
850 
eg_tex_filter(unsigned filter,unsigned max_aniso)851 static inline unsigned eg_tex_filter(unsigned filter, unsigned max_aniso)
852 {
853 	if (filter == PIPE_TEX_FILTER_LINEAR)
854 		return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_BILINEAR
855 				     : SQ_TEX_XY_FILTER_BILINEAR;
856 	else
857 		return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_POINT
858 				     : SQ_TEX_XY_FILTER_POINT;
859 }
860 
r600_tex_aniso_filter(unsigned filter)861 static inline unsigned r600_tex_aniso_filter(unsigned filter)
862 {
863 	if (filter < 2)
864 		return 0;
865 	if (filter < 4)
866 		return 1;
867 	if (filter < 8)
868 		return 2;
869 	if (filter < 16)
870 		return 3;
871 	return 4;
872 }
873 
r600_wavefront_size(enum radeon_family family)874 static inline unsigned r600_wavefront_size(enum radeon_family family)
875 {
876 	switch (family) {
877 	case CHIP_RV610:
878 	case CHIP_RS780:
879 	case CHIP_RV620:
880 	case CHIP_RS880:
881 		return 16;
882 	case CHIP_RV630:
883 	case CHIP_RV635:
884 	case CHIP_RV730:
885 	case CHIP_RV710:
886 	case CHIP_PALM:
887 	case CHIP_CEDAR:
888 		return 32;
889 	default:
890 		return 64;
891 	}
892 }
893 
894 static inline enum radeon_bo_priority
r600_get_sampler_view_priority(struct r600_resource * res)895 r600_get_sampler_view_priority(struct r600_resource *res)
896 {
897 	if (res->b.b.target == PIPE_BUFFER)
898 		return RADEON_PRIO_SAMPLER_BUFFER;
899 
900 	if (res->b.b.nr_samples > 1)
901 		return RADEON_PRIO_SAMPLER_TEXTURE_MSAA;
902 
903 	return RADEON_PRIO_SAMPLER_TEXTURE;
904 }
905 
906 static inline bool
r600_can_sample_zs(struct r600_texture * tex,bool stencil_sampler)907 r600_can_sample_zs(struct r600_texture *tex, bool stencil_sampler)
908 {
909 	return (stencil_sampler && tex->can_sample_s) ||
910 	       (!stencil_sampler && tex->can_sample_z);
911 }
912 
913 static inline bool
r600_htile_enabled(struct r600_texture * tex,unsigned level)914 r600_htile_enabled(struct r600_texture *tex, unsigned level)
915 {
916 	return tex->htile_offset && level == 0;
917 }
918 
919 #define COMPUTE_DBG(rscreen, fmt, args...) \
920 	do { \
921 		if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
922 	} while (0);
923 
924 #define R600_ERR(fmt, args...) \
925 	fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
926 
927 /* For MSAA sample positions. */
928 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y)  \
929 	(((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) |		   \
930 	(((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) |	   \
931 	(((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) |	   \
932 	 (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))
933 
S_FIXED(float value,unsigned frac_bits)934 static inline int S_FIXED(float value, unsigned frac_bits)
935 {
936 	return value * (1 << frac_bits);
937 }
938 
939 #endif
940