Home | History | Annotate | Download | only in radeon
      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 
     24 /**
     25  * This file contains common screen and context structures and functions
     26  * for r600g and radeonsi.
     27  */
     28 
     29 #ifndef R600_PIPE_COMMON_H
     30 #define R600_PIPE_COMMON_H
     31 
     32 #include <stdio.h>
     33 
     34 #include "amd/common/ac_binary.h"
     35 
     36 #include "radeon/radeon_winsys.h"
     37 
     38 #include "util/disk_cache.h"
     39 #include "util/u_blitter.h"
     40 #include "util/list.h"
     41 #include "util/u_range.h"
     42 #include "util/slab.h"
     43 #include "util/u_suballoc.h"
     44 #include "util/u_transfer.h"
     45 #include "util/u_threaded_context.h"
     46 
     47 struct u_log_context;
     48 struct si_screen;
     49 struct si_context;
     50 
     51 #define R600_RESOURCE_FLAG_TRANSFER		(PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
     52 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH	(PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
     53 #define R600_RESOURCE_FLAG_FORCE_TILING		(PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
     54 #define R600_RESOURCE_FLAG_DISABLE_DCC		(PIPE_RESOURCE_FLAG_DRV_PRIV << 3)
     55 #define R600_RESOURCE_FLAG_UNMAPPABLE		(PIPE_RESOURCE_FLAG_DRV_PRIV << 4)
     56 #define R600_RESOURCE_FLAG_READ_ONLY		(PIPE_RESOURCE_FLAG_DRV_PRIV << 5)
     57 
     58 /* Debug flags. */
     59 enum {
     60 	/* Shader logging options: */
     61 	DBG_VS = PIPE_SHADER_VERTEX,
     62 	DBG_PS = PIPE_SHADER_FRAGMENT,
     63 	DBG_GS = PIPE_SHADER_GEOMETRY,
     64 	DBG_TCS = PIPE_SHADER_TESS_CTRL,
     65 	DBG_TES = PIPE_SHADER_TESS_EVAL,
     66 	DBG_CS = PIPE_SHADER_COMPUTE,
     67 	DBG_NO_IR,
     68 	DBG_NO_TGSI,
     69 	DBG_NO_ASM,
     70 	DBG_PREOPT_IR,
     71 
     72 	/* Shader compiler options the shader cache should be aware of: */
     73 	DBG_FS_CORRECT_DERIVS_AFTER_KILL,
     74 	DBG_UNSAFE_MATH,
     75 	DBG_SI_SCHED,
     76 
     77 	/* Shader compiler options (with no effect on the shader cache): */
     78 	DBG_CHECK_IR,
     79 	DBG_PRECOMPILE,
     80 	DBG_NIR,
     81 	DBG_MONOLITHIC_SHADERS,
     82 	DBG_NO_OPT_VARIANT,
     83 
     84 	/* Information logging options: */
     85 	DBG_INFO,
     86 	DBG_TEX,
     87 	DBG_COMPUTE,
     88 	DBG_VM,
     89 
     90 	/* Driver options: */
     91 	DBG_FORCE_DMA,
     92 	DBG_NO_ASYNC_DMA,
     93 	DBG_NO_WC,
     94 	DBG_CHECK_VM,
     95 	DBG_RESERVE_VMID,
     96 
     97 	/* 3D engine options: */
     98 	DBG_SWITCH_ON_EOP,
     99 	DBG_NO_OUT_OF_ORDER,
    100 	DBG_NO_DPBB,
    101 	DBG_NO_DFSM,
    102 	DBG_DPBB,
    103 	DBG_DFSM,
    104 	DBG_NO_HYPERZ,
    105 	DBG_NO_RB_PLUS,
    106 	DBG_NO_2D_TILING,
    107 	DBG_NO_TILING,
    108 	DBG_NO_DCC,
    109 	DBG_NO_DCC_CLEAR,
    110 	DBG_NO_DCC_FB,
    111 	DBG_NO_DCC_MSAA,
    112 	DBG_DCC_MSAA,
    113 
    114 	/* Tests: */
    115 	DBG_TEST_DMA,
    116 	DBG_TEST_VMFAULT_CP,
    117 	DBG_TEST_VMFAULT_SDMA,
    118 	DBG_TEST_VMFAULT_SHADER,
    119 };
    120 
    121 #define DBG_ALL_SHADERS		(((1 << (DBG_CS + 1)) - 1))
    122 #define DBG(name)		(1ull << DBG_##name)
    123 
    124 #define R600_MAP_BUFFER_ALIGNMENT 64
    125 
    126 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 1024
    127 
    128 struct r600_common_context;
    129 struct r600_perfcounters;
    130 struct tgsi_shader_info;
    131 struct r600_qbo_state;
    132 
    133 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
    134  * at the moment.
    135  */
    136 struct r600_resource {
    137 	struct threaded_resource	b;
    138 
    139 	/* Winsys objects. */
    140 	struct pb_buffer		*buf;
    141 	uint64_t			gpu_address;
    142 	/* Memory usage if the buffer placement is optimal. */
    143 	uint64_t			vram_usage;
    144 	uint64_t			gart_usage;
    145 
    146 	/* Resource properties. */
    147 	uint64_t			bo_size;
    148 	unsigned			bo_alignment;
    149 	enum radeon_bo_domain		domains;
    150 	enum radeon_bo_flag		flags;
    151 	unsigned			bind_history;
    152 	int				max_forced_staging_uploads;
    153 
    154 	/* The buffer range which is initialized (with a write transfer,
    155 	 * streamout, DMA, or as a random access target). The rest of
    156 	 * the buffer is considered invalid and can be mapped unsynchronized.
    157 	 *
    158 	 * This allows unsychronized mapping of a buffer range which hasn't
    159 	 * been used yet. It's for applications which forget to use
    160 	 * the unsynchronized map flag and expect the driver to figure it out.
    161          */
    162 	struct util_range		valid_buffer_range;
    163 
    164 	/* For buffers only. This indicates that a write operation has been
    165 	 * performed by TC L2, but the cache hasn't been flushed.
    166 	 * Any hw block which doesn't use or bypasses TC L2 should check this
    167 	 * flag and flush the cache before using the buffer.
    168 	 *
    169 	 * For example, TC L2 must be flushed if a buffer which has been
    170 	 * modified by a shader store instruction is about to be used as
    171 	 * an index buffer. The reason is that VGT DMA index fetching doesn't
    172 	 * use TC L2.
    173 	 */
    174 	bool				TC_L2_dirty;
    175 
    176 	/* Whether the resource has been exported via resource_get_handle. */
    177 	unsigned			external_usage; /* PIPE_HANDLE_USAGE_* */
    178 
    179 	/* Whether this resource is referenced by bindless handles. */
    180 	bool				texture_handle_allocated;
    181 	bool				image_handle_allocated;
    182 };
    183 
    184 struct r600_transfer {
    185 	struct threaded_transfer	b;
    186 	struct r600_resource		*staging;
    187 	unsigned			offset;
    188 };
    189 
    190 struct r600_fmask_info {
    191 	uint64_t offset;
    192 	uint64_t size;
    193 	unsigned alignment;
    194 	unsigned pitch_in_pixels;
    195 	unsigned bank_height;
    196 	unsigned slice_tile_max;
    197 	unsigned tile_mode_index;
    198 	unsigned tile_swizzle;
    199 };
    200 
    201 struct r600_cmask_info {
    202 	uint64_t offset;
    203 	uint64_t size;
    204 	unsigned alignment;
    205 	unsigned slice_tile_max;
    206 	uint64_t base_address_reg;
    207 };
    208 
    209 struct r600_texture {
    210 	struct r600_resource		resource;
    211 
    212 	struct radeon_surf		surface;
    213 	uint64_t			size;
    214 	struct r600_texture		*flushed_depth_texture;
    215 
    216 	/* Colorbuffer compression and fast clear. */
    217 	struct r600_fmask_info		fmask;
    218 	struct r600_cmask_info		cmask;
    219 	struct r600_resource		*cmask_buffer;
    220 	uint64_t			dcc_offset; /* 0 = disabled */
    221 	unsigned			cb_color_info; /* fast clear enable bit */
    222 	unsigned			color_clear_value[2];
    223 	unsigned			last_msaa_resolve_target_micro_mode;
    224 	unsigned			num_level0_transfers;
    225 
    226 	/* Depth buffer compression and fast clear. */
    227 	uint64_t			htile_offset;
    228 	float				depth_clear_value;
    229 	uint16_t			dirty_level_mask; /* each bit says if that mipmap is compressed */
    230 	uint16_t			stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
    231 	enum pipe_format		db_render_format:16;
    232 	uint8_t				stencil_clear_value;
    233 	bool				tc_compatible_htile:1;
    234 	bool				depth_cleared:1; /* if it was cleared at least once */
    235 	bool				stencil_cleared:1; /* if it was cleared at least once */
    236 	bool				upgraded_depth:1; /* upgraded from unorm to Z32_FLOAT */
    237 	bool				is_depth:1;
    238 	bool				db_compatible:1;
    239 	bool				can_sample_z:1;
    240 	bool				can_sample_s:1;
    241 
    242 	/* We need to track DCC dirtiness, because st/dri usually calls
    243 	 * flush_resource twice per frame (not a bug) and we don't wanna
    244 	 * decompress DCC twice. Also, the dirty tracking must be done even
    245 	 * if DCC isn't used, because it's required by the DCC usage analysis
    246 	 * for a possible future enablement.
    247 	 */
    248 	bool				separate_dcc_dirty:1;
    249 	/* Statistics gathering for the DCC enablement heuristic. */
    250 	bool				dcc_gather_statistics:1;
    251 	/* Counter that should be non-zero if the texture is bound to a
    252 	 * framebuffer.
    253 	 */
    254 	unsigned                        framebuffers_bound;
    255 	/* Whether the texture is a displayable back buffer and needs DCC
    256 	 * decompression, which is expensive. Therefore, it's enabled only
    257 	 * if statistics suggest that it will pay off and it's allocated
    258 	 * separately. It can't be bound as a sampler by apps. Limited to
    259 	 * target == 2D and last_level == 0. If enabled, dcc_offset contains
    260 	 * the absolute GPUVM address, not the relative one.
    261 	 */
    262 	struct r600_resource		*dcc_separate_buffer;
    263 	/* When DCC is temporarily disabled, the separate buffer is here. */
    264 	struct r600_resource		*last_dcc_separate_buffer;
    265 	/* Estimate of how much this color buffer is written to in units of
    266 	 * full-screen draws: ps_invocations / (width * height)
    267 	 * Shader kills, late Z, and blending with trivial discards make it
    268 	 * inaccurate (we need to count CB updates, not PS invocations).
    269 	 */
    270 	unsigned			ps_draw_ratio;
    271 	/* The number of clears since the last DCC usage analysis. */
    272 	unsigned			num_slow_clears;
    273 };
    274 
    275 struct r600_surface {
    276 	struct pipe_surface		base;
    277 
    278 	/* These can vary with block-compressed textures. */
    279 	uint16_t width0;
    280 	uint16_t height0;
    281 
    282 	bool color_initialized:1;
    283 	bool depth_initialized:1;
    284 
    285 	/* Misc. color flags. */
    286 	bool color_is_int8:1;
    287 	bool color_is_int10:1;
    288 	bool dcc_incompatible:1;
    289 
    290 	/* Color registers. */
    291 	unsigned cb_color_info;
    292 	unsigned cb_color_view;
    293 	unsigned cb_color_attrib;
    294 	unsigned cb_color_attrib2;	/* GFX9 and later */
    295 	unsigned cb_dcc_control;	/* VI and later */
    296 	unsigned spi_shader_col_format:8;	/* no blending, no alpha-to-coverage. */
    297 	unsigned spi_shader_col_format_alpha:8;	/* alpha-to-coverage */
    298 	unsigned spi_shader_col_format_blend:8;	/* blending without alpha. */
    299 	unsigned spi_shader_col_format_blend_alpha:8; /* blending with alpha. */
    300 
    301 	/* DB registers. */
    302 	uint64_t db_depth_base;		/* DB_Z_READ/WRITE_BASE */
    303 	uint64_t db_stencil_base;
    304 	uint64_t db_htile_data_base;
    305 	unsigned db_depth_info;
    306 	unsigned db_z_info;
    307 	unsigned db_z_info2;		/* GFX9+ */
    308 	unsigned db_depth_view;
    309 	unsigned db_depth_size;
    310 	unsigned db_depth_slice;
    311 	unsigned db_stencil_info;
    312 	unsigned db_stencil_info2;	/* GFX9+ */
    313 	unsigned db_htile_surface;
    314 };
    315 
    316 struct r600_mmio_counter {
    317 	unsigned busy;
    318 	unsigned idle;
    319 };
    320 
    321 union r600_mmio_counters {
    322 	struct {
    323 		/* For global GPU load including SDMA. */
    324 		struct r600_mmio_counter gpu;
    325 
    326 		/* GRBM_STATUS */
    327 		struct r600_mmio_counter spi;
    328 		struct r600_mmio_counter gui;
    329 		struct r600_mmio_counter ta;
    330 		struct r600_mmio_counter gds;
    331 		struct r600_mmio_counter vgt;
    332 		struct r600_mmio_counter ia;
    333 		struct r600_mmio_counter sx;
    334 		struct r600_mmio_counter wd;
    335 		struct r600_mmio_counter bci;
    336 		struct r600_mmio_counter sc;
    337 		struct r600_mmio_counter pa;
    338 		struct r600_mmio_counter db;
    339 		struct r600_mmio_counter cp;
    340 		struct r600_mmio_counter cb;
    341 
    342 		/* SRBM_STATUS2 */
    343 		struct r600_mmio_counter sdma;
    344 
    345 		/* CP_STAT */
    346 		struct r600_mmio_counter pfp;
    347 		struct r600_mmio_counter meq;
    348 		struct r600_mmio_counter me;
    349 		struct r600_mmio_counter surf_sync;
    350 		struct r600_mmio_counter cp_dma;
    351 		struct r600_mmio_counter scratch_ram;
    352 	} named;
    353 	unsigned array[0];
    354 };
    355 
    356 struct r600_memory_object {
    357 	struct pipe_memory_object	b;
    358 	struct pb_buffer		*buf;
    359 	uint32_t			stride;
    360 	uint32_t			offset;
    361 };
    362 
    363 /* This encapsulates a state or an operation which can emitted into the GPU
    364  * command stream. */
    365 struct r600_atom {
    366 	void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
    367 	unsigned short		id;
    368 };
    369 
    370 struct r600_ring {
    371 	struct radeon_winsys_cs		*cs;
    372 	void (*flush)(void *ctx, unsigned flags,
    373 		      struct pipe_fence_handle **fence);
    374 };
    375 
    376 /* Saved CS data for debugging features. */
    377 struct radeon_saved_cs {
    378 	uint32_t			*ib;
    379 	unsigned			num_dw;
    380 
    381 	struct radeon_bo_list_item	*bo_list;
    382 	unsigned			bo_count;
    383 };
    384 
    385 struct r600_common_context {
    386 	struct pipe_context b; /* base class */
    387 
    388 	struct si_screen		*screen;
    389 	struct radeon_winsys		*ws;
    390 	struct radeon_winsys_ctx	*ctx;
    391 	enum radeon_family		family;
    392 	enum chip_class			chip_class;
    393 	struct r600_ring		gfx;
    394 	struct r600_ring		dma;
    395 	struct pipe_fence_handle	*last_gfx_fence;
    396 	struct pipe_fence_handle	*last_sdma_fence;
    397 	struct r600_resource		*eop_bug_scratch;
    398 	struct u_upload_mgr		*cached_gtt_allocator;
    399 	unsigned			num_gfx_cs_flushes;
    400 	unsigned			initial_gfx_cs_size;
    401 	unsigned			gpu_reset_counter;
    402 	unsigned			last_dirty_tex_counter;
    403 	unsigned			last_compressed_colortex_counter;
    404 	unsigned			last_num_draw_calls;
    405 
    406 	struct threaded_context		*tc;
    407 	struct u_suballocator		*allocator_zeroed_memory;
    408 	struct slab_child_pool		pool_transfers;
    409 	struct slab_child_pool		pool_transfers_unsync; /* for threaded_context */
    410 
    411 	/* Current unaccounted memory usage. */
    412 	uint64_t			vram;
    413 	uint64_t			gtt;
    414 
    415 	/* Additional context states. */
    416 	unsigned flags; /* flush flags */
    417 
    418 	/* Queries. */
    419 	/* Maintain the list of active queries for pausing between IBs. */
    420 	int				num_occlusion_queries;
    421 	int				num_perfect_occlusion_queries;
    422 	struct list_head		active_queries;
    423 	unsigned			num_cs_dw_queries_suspend;
    424 	/* Misc stats. */
    425 	unsigned			num_draw_calls;
    426 	unsigned			num_decompress_calls;
    427 	unsigned			num_mrt_draw_calls;
    428 	unsigned			num_prim_restart_calls;
    429 	unsigned			num_spill_draw_calls;
    430 	unsigned			num_compute_calls;
    431 	unsigned			num_spill_compute_calls;
    432 	unsigned			num_dma_calls;
    433 	unsigned			num_cp_dma_calls;
    434 	unsigned			num_vs_flushes;
    435 	unsigned			num_ps_flushes;
    436 	unsigned			num_cs_flushes;
    437 	unsigned			num_cb_cache_flushes;
    438 	unsigned			num_db_cache_flushes;
    439 	unsigned			num_L2_invalidates;
    440 	unsigned			num_L2_writebacks;
    441 	unsigned			num_resident_handles;
    442 	uint64_t			num_alloc_tex_transfer_bytes;
    443 	unsigned			last_tex_ps_draw_ratio; /* for query */
    444 
    445 	/* Render condition. */
    446 	struct r600_atom		render_cond_atom;
    447 	struct pipe_query		*render_cond;
    448 	unsigned			render_cond_mode;
    449 	bool				render_cond_invert;
    450 	bool				render_cond_force_off; /* for u_blitter */
    451 
    452 	/* Statistics gathering for the DCC enablement heuristic. It can't be
    453 	 * in r600_texture because r600_texture can be shared by multiple
    454 	 * contexts. This is for back buffers only. We shouldn't get too many
    455 	 * of those.
    456 	 *
    457 	 * X11 DRI3 rotates among a finite set of back buffers. They should
    458 	 * all fit in this array. If they don't, separate DCC might never be
    459 	 * enabled by DCC stat gathering.
    460 	 */
    461 	struct {
    462 		struct r600_texture		*tex;
    463 		/* Query queue: 0 = usually active, 1 = waiting, 2 = readback. */
    464 		struct pipe_query		*ps_stats[3];
    465 		/* If all slots are used and another slot is needed,
    466 		 * the least recently used slot is evicted based on this. */
    467 		int64_t				last_use_timestamp;
    468 		bool				query_active;
    469 	} dcc_stats[5];
    470 
    471 	struct pipe_device_reset_callback device_reset_callback;
    472 	struct u_log_context		*log;
    473 
    474 	void				*query_result_shader;
    475 
    476 	/* Copy one resource to another using async DMA. */
    477 	void (*dma_copy)(struct pipe_context *ctx,
    478 			 struct pipe_resource *dst,
    479 			 unsigned dst_level,
    480 			 unsigned dst_x, unsigned dst_y, unsigned dst_z,
    481 			 struct pipe_resource *src,
    482 			 unsigned src_level,
    483 			 const struct pipe_box *src_box);
    484 
    485 	void (*dma_clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
    486 				 uint64_t offset, uint64_t size, unsigned value);
    487 
    488 	void (*blit_decompress_depth)(struct pipe_context *ctx,
    489 				      struct r600_texture *texture,
    490 				      struct r600_texture *staging,
    491 				      unsigned first_level, unsigned last_level,
    492 				      unsigned first_layer, unsigned last_layer,
    493 				      unsigned first_sample, unsigned last_sample);
    494 
    495 	void (*decompress_dcc)(struct pipe_context *ctx,
    496 			       struct r600_texture *rtex);
    497 
    498 	/* Reallocate the buffer and update all resource bindings where
    499 	 * the buffer is bound, including all resource descriptors. */
    500 	void (*invalidate_buffer)(struct pipe_context *ctx, struct pipe_resource *buf);
    501 
    502 	/* Update all resource bindings where the buffer is bound, including
    503 	 * all resource descriptors. This is invalidate_buffer without
    504 	 * the invalidation. */
    505 	void (*rebind_buffer)(struct pipe_context *ctx, struct pipe_resource *buf,
    506 			      uint64_t old_gpu_address);
    507 
    508 	/* Enable or disable occlusion queries. */
    509 	void (*set_occlusion_query_state)(struct pipe_context *ctx,
    510 					  bool old_enable,
    511 					  bool old_perfect_enable);
    512 
    513 	void (*save_qbo_state)(struct pipe_context *ctx, struct r600_qbo_state *st);
    514 
    515 	/* This ensures there is enough space in the command stream. */
    516 	void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
    517 				  bool include_draw_vbo);
    518 
    519 	void (*set_atom_dirty)(struct r600_common_context *ctx,
    520 			       struct r600_atom *atom, bool dirty);
    521 
    522 	void (*check_vm_faults)(struct r600_common_context *ctx,
    523 				struct radeon_saved_cs *saved,
    524 				enum ring_type ring);
    525 };
    526 
    527 /* r600_buffer_common.c */
    528 bool si_rings_is_buffer_referenced(struct r600_common_context *ctx,
    529 				   struct pb_buffer *buf,
    530 				   enum radeon_bo_usage usage);
    531 void *si_buffer_map_sync_with_rings(struct r600_common_context *ctx,
    532 				    struct r600_resource *resource,
    533 				    unsigned usage);
    534 void si_init_resource_fields(struct si_screen *sscreen,
    535 			     struct r600_resource *res,
    536 			     uint64_t size, unsigned alignment);
    537 bool si_alloc_resource(struct si_screen *sscreen,
    538 		       struct r600_resource *res);
    539 struct pipe_resource *si_aligned_buffer_create(struct pipe_screen *screen,
    540 					       unsigned flags,
    541 					       unsigned usage,
    542 					       unsigned size,
    543 					       unsigned alignment);
    544 void si_replace_buffer_storage(struct pipe_context *ctx,
    545 			       struct pipe_resource *dst,
    546 			       struct pipe_resource *src);
    547 void si_init_screen_buffer_functions(struct si_screen *sscreen);
    548 void si_init_buffer_functions(struct si_context *sctx);
    549 
    550 /* r600_common_pipe.c */
    551 void si_gfx_write_event_eop(struct r600_common_context *ctx,
    552 			    unsigned event, unsigned event_flags,
    553 			    unsigned data_sel,
    554 			    struct r600_resource *buf, uint64_t va,
    555 			    uint32_t new_fence, unsigned query_type);
    556 unsigned si_gfx_write_fence_dwords(struct si_screen *screen);
    557 void si_gfx_wait_fence(struct r600_common_context *ctx,
    558 		       uint64_t va, uint32_t ref, uint32_t mask);
    559 bool si_common_context_init(struct r600_common_context *rctx,
    560 			    struct si_screen *sscreen,
    561 			    unsigned context_flags);
    562 void si_common_context_cleanup(struct r600_common_context *rctx);
    563 void si_screen_clear_buffer(struct si_screen *sscreen, struct pipe_resource *dst,
    564 			    uint64_t offset, uint64_t size, unsigned value);
    565 void si_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
    566 		       struct r600_resource *dst, struct r600_resource *src);
    567 void si_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
    568 		struct radeon_saved_cs *saved, bool get_buffer_list);
    569 void si_clear_saved_cs(struct radeon_saved_cs *saved);
    570 bool si_check_device_reset(struct r600_common_context *rctx);
    571 
    572 /* r600_gpu_load.c */
    573 void si_gpu_load_kill_thread(struct si_screen *sscreen);
    574 uint64_t si_begin_counter(struct si_screen *sscreen, unsigned type);
    575 unsigned si_end_counter(struct si_screen *sscreen, unsigned type,
    576 			uint64_t begin);
    577 
    578 /* r600_perfcounters.c */
    579 void si_perfcounters_destroy(struct si_screen *sscreen);
    580 
    581 /* r600_query.c */
    582 void si_init_screen_query_functions(struct si_screen *sscreen);
    583 void si_init_query_functions(struct r600_common_context *rctx);
    584 void si_suspend_queries(struct r600_common_context *ctx);
    585 void si_resume_queries(struct r600_common_context *ctx);
    586 
    587 /* r600_texture.c */
    588 bool si_prepare_for_dma_blit(struct r600_common_context *rctx,
    589 			     struct r600_texture *rdst,
    590 			     unsigned dst_level, unsigned dstx,
    591 			     unsigned dsty, unsigned dstz,
    592 			     struct r600_texture *rsrc,
    593 			     unsigned src_level,
    594 			     const struct pipe_box *src_box);
    595 void si_texture_get_fmask_info(struct si_screen *sscreen,
    596 			       struct r600_texture *rtex,
    597 			       unsigned nr_samples,
    598 			       struct r600_fmask_info *out);
    599 void si_texture_get_cmask_info(struct si_screen *sscreen,
    600 			       struct r600_texture *rtex,
    601 			       struct r600_cmask_info *out);
    602 bool si_init_flushed_depth_texture(struct pipe_context *ctx,
    603 				   struct pipe_resource *texture,
    604 				   struct r600_texture **staging);
    605 void si_print_texture_info(struct si_screen *sscreen,
    606 			   struct r600_texture *rtex, struct u_log_context *log);
    607 struct pipe_resource *si_texture_create(struct pipe_screen *screen,
    608 					const struct pipe_resource *templ);
    609 bool vi_dcc_formats_compatible(enum pipe_format format1,
    610 			       enum pipe_format format2);
    611 bool vi_dcc_formats_are_incompatible(struct pipe_resource *tex,
    612 				     unsigned level,
    613 				     enum pipe_format view_format);
    614 void vi_disable_dcc_if_incompatible_format(struct r600_common_context *rctx,
    615 					   struct pipe_resource *tex,
    616 					   unsigned level,
    617 					   enum pipe_format view_format);
    618 struct pipe_surface *si_create_surface_custom(struct pipe_context *pipe,
    619 					      struct pipe_resource *texture,
    620 					      const struct pipe_surface *templ,
    621 					      unsigned width0, unsigned height0,
    622 					      unsigned width, unsigned height);
    623 unsigned si_translate_colorswap(enum pipe_format format, bool do_endian_swap);
    624 void vi_separate_dcc_try_enable(struct r600_common_context *rctx,
    625 				struct r600_texture *tex);
    626 void vi_separate_dcc_start_query(struct pipe_context *ctx,
    627 				 struct r600_texture *tex);
    628 void vi_separate_dcc_stop_query(struct pipe_context *ctx,
    629 				struct r600_texture *tex);
    630 void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx,
    631 					     struct r600_texture *tex);
    632 bool si_texture_disable_dcc(struct r600_common_context *rctx,
    633 			    struct r600_texture *rtex);
    634 void si_init_screen_texture_functions(struct si_screen *sscreen);
    635 void si_init_context_texture_functions(struct r600_common_context *rctx);
    636 
    637 
    638 /* Inline helpers. */
    639 
    640 static inline struct r600_resource *r600_resource(struct pipe_resource *r)
    641 {
    642 	return (struct r600_resource*)r;
    643 }
    644 
    645 static inline void
    646 r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
    647 {
    648 	pipe_resource_reference((struct pipe_resource **)ptr,
    649 				(struct pipe_resource *)res);
    650 }
    651 
    652 static inline void
    653 r600_texture_reference(struct r600_texture **ptr, struct r600_texture *res)
    654 {
    655 	pipe_resource_reference((struct pipe_resource **)ptr, &res->resource.b.b);
    656 }
    657 
    658 static inline bool
    659 vi_dcc_enabled(struct r600_texture *tex, unsigned level)
    660 {
    661 	return tex->dcc_offset && level < tex->surface.num_dcc_levels;
    662 }
    663 
    664 #define R600_ERR(fmt, args...) \
    665 	fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
    666 
    667 #endif
    668