src/gallium/drivers/radeon/r600_pipe_common.h

   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 "radeon/radeon_winsys.h"
  38
  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
  46 #define ATI_VENDOR_ID 0x1002
  47
  48 #define R600_RESOURCE_FLAG_TRANSFER             (PIPE_RESOURCE_FLAG_DRV_PRIV << 0)
  49 #define R600_RESOURCE_FLAG_FLUSHED_DEPTH        (PIPE_RESOURCE_FLAG_DRV_PRIV << 1)
  50 #define R600_RESOURCE_FLAG_FORCE_TILING         (PIPE_RESOURCE_FLAG_DRV_PRIV << 2)
  51 #define R600_RESOURCE_FLAG_DISABLE_DCC          (PIPE_RESOURCE_FLAG_DRV_PRIV << 3)
  52
  53 #define R600_CONTEXT_STREAMOUT_FLUSH            (1u << 0)
  54 /* Pipeline & streamout query controls. */
  55 #define R600_CONTEXT_START_PIPELINE_STATS       (1u << 1)
  56 #define R600_CONTEXT_STOP_PIPELINE_STATS        (1u << 2)
  57 #define R600_CONTEXT_PRIVATE_FLAG               (1u << 3)
  58
  59 /* special primitive types */
  60 #define R600_PRIM_RECTANGLE_LIST        PIPE_PRIM_MAX
  61
  62 /* Debug flags. */
  63 /* logging */
  64 #define DBG_TEX                 (1 << 0)
  65 /* gap - reuse */
  66 #define DBG_COMPUTE             (1 << 2)
  67 #define DBG_VM                  (1 << 3)
  68 /* gap - reuse */
  69 /* shader logging */
  70 #define DBG_FS                  (1 << 5)
  71 #define DBG_VS                  (1 << 6)
  72 #define DBG_GS                  (1 << 7)
  73 #define DBG_PS                  (1 << 8)
  74 #define DBG_CS                  (1 << 9)
  75 #define DBG_TCS                 (1 << 10)
  76 #define DBG_TES                 (1 << 11)
  77 #define DBG_NO_IR               (1 << 12)
  78 #define DBG_NO_TGSI             (1 << 13)
  79 #define DBG_NO_ASM              (1 << 14)
  80 #define DBG_PREOPT_IR           (1 << 15)
  81 #define DBG_CHECK_IR            (1 << 16)
  82 /* gaps */
  83 #define DBG_TEST_DMA            (1 << 20)
  84 /* Bits 21-31 are reserved for the r600g driver. */
  85 /* features */
  86 #define DBG_NO_ASYNC_DMA        (1llu << 32)
  87 #define DBG_NO_HYPERZ           (1llu << 33)
  88 #define DBG_NO_DISCARD_RANGE    (1llu << 34)
  89 #define DBG_NO_2D_TILING        (1llu << 35)
  90 #define DBG_NO_TILING           (1llu << 36)
  91 #define DBG_SWITCH_ON_EOP       (1llu << 37)
  92 #define DBG_FORCE_DMA           (1llu << 38)
  93 #define DBG_PRECOMPILE          (1llu << 39)
  94 #define DBG_INFO                (1llu << 40)
  95 #define DBG_NO_WC               (1llu << 41)
  96 #define DBG_CHECK_VM            (1llu << 42)
  97 #define DBG_NO_DCC              (1llu << 43)
  98 #define DBG_NO_DCC_CLEAR        (1llu << 44)
  99 #define DBG_NO_RB_PLUS          (1llu << 45)
 100 #define DBG_SI_SCHED            (1llu << 46)
 101 #define DBG_MONOLITHIC_SHADERS  (1llu << 47)
 102 #define DBG_NO_CE               (1llu << 48)
 103 #define DBG_UNSAFE_MATH         (1llu << 49)
 104 #define DBG_NO_DCC_FB           (1llu << 50)
 105
 106 #define R600_MAP_BUFFER_ALIGNMENT 64
 107 #define R600_MAX_VIEWPORTS        16
 108
 109 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 1024
 110
 111 enum r600_coherency {
 112         R600_COHERENCY_NONE, /* no cache flushes needed */
 113         R600_COHERENCY_SHADER,
 114         R600_COHERENCY_CB_META,
 115 };
 116
 117 #ifdef PIPE_ARCH_BIG_ENDIAN
 118 #define R600_BIG_ENDIAN 1
 119 #else
 120 #define R600_BIG_ENDIAN 0
 121 #endif
 122
 123 struct r600_common_context;
 124 struct r600_perfcounters;
 125 struct tgsi_shader_info;
 126 struct r600_qbo_state;
 127
 128 struct radeon_shader_reloc {
 129         char name[32];
 130         uint64_t offset;
 131 };
 132
 133 struct radeon_shader_binary {
 134         /** Shader code */
 135         unsigned char *code;
 136         unsigned code_size;
 137
 138         /** Config/Context register state that accompanies this shader.
 139          * This is a stream of dword pairs.  First dword contains the
 140          * register address, the second dword contains the value.*/
 141         unsigned char *config;
 142         unsigned config_size;
 143
 144         /** The number of bytes of config information for each global symbol.
 145          */
 146         unsigned config_size_per_symbol;
 147
 148         /** Constant data accessed by the shader.  This will be uploaded
 149          * into a constant buffer. */
 150         unsigned char *rodata;
 151         unsigned rodata_size;
 152
 153         /** List of symbol offsets for the shader */
 154         uint64_t *global_symbol_offsets;
 155         unsigned global_symbol_count;
 156
 157         struct radeon_shader_reloc *relocs;
 158         unsigned reloc_count;
 159
 160         /** Disassembled shader in a string. */
 161         char *disasm_string;
 162         char *llvm_ir_string;
 163 };
 164
 165 void radeon_shader_binary_init(struct radeon_shader_binary *b);
 166 void radeon_shader_binary_clean(struct radeon_shader_binary *b);
 167
 168 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
 169  * at the moment.
 170  */
 171 struct r600_resource {
 172         struct u_resource               b;
 173
 174         /* Winsys objects. */
 175         struct pb_buffer                *buf;
 176         uint64_t                        gpu_address;
 177         /* Memory usage if the buffer placement is optimal. */
 178         uint64_t                        vram_usage;
 179         uint64_t                        gart_usage;
 180
 181         /* Resource properties. */
 182         uint64_t                        bo_size;
 183         unsigned                        bo_alignment;
 184         enum radeon_bo_domain           domains;
 185         enum radeon_bo_flag             flags;
 186         unsigned                        bind_history;
 187
 188         /* The buffer range which is initialized (with a write transfer,
 189          * streamout, DMA, or as a random access target). The rest of
 190          * the buffer is considered invalid and can be mapped unsynchronized.
 191          *
 192          * This allows unsychronized mapping of a buffer range which hasn't
 193          * been used yet. It's for applications which forget to use
 194          * the unsynchronized map flag and expect the driver to figure it out.
 195          */
 196         struct util_range               valid_buffer_range;
 197
 198         /* For buffers only. This indicates that a write operation has been
 199          * performed by TC L2, but the cache hasn't been flushed.
 200          * Any hw block which doesn't use or bypasses TC L2 should check this
 201          * flag and flush the cache before using the buffer.
 202          *
 203          * For example, TC L2 must be flushed if a buffer which has been
 204          * modified by a shader store instruction is about to be used as
 205          * an index buffer. The reason is that VGT DMA index fetching doesn't
 206          * use TC L2.
 207          */
 208         bool                            TC_L2_dirty;
 209
 210         /* Whether the resource has been exported via resource_get_handle. */
 211         bool                            is_shared;
 212         unsigned                        external_usage; /* PIPE_HANDLE_USAGE_* */
 213 };
 214
 215 struct r600_transfer {
 216         struct pipe_transfer            transfer;
 217         struct r600_resource            *staging;
 218         unsigned                        offset;
 219 };
 220
 221 struct r600_fmask_info {
 222         uint64_t offset;
 223         uint64_t size;
 224         unsigned alignment;
 225         unsigned pitch_in_pixels;
 226         unsigned bank_height;
 227         unsigned slice_tile_max;
 228         unsigned tile_mode_index;
 229 };
 230
 231 struct r600_cmask_info {
 232         uint64_t offset;
 233         uint64_t size;
 234         unsigned alignment;
 235         unsigned pitch;
 236         unsigned height;
 237         unsigned xalign;
 238         unsigned yalign;
 239         unsigned slice_tile_max;
 240         unsigned base_address_reg;
 241 };
 242
 243 struct r600_htile_info {
 244         unsigned pitch;
 245         unsigned height;
 246         unsigned xalign;
 247         unsigned yalign;
 248         unsigned alignment;
 249 };
 250
 251 struct r600_texture {
 252         struct r600_resource            resource;
 253
 254         uint64_t                        size;
 255         unsigned                        num_level0_transfers;
 256         enum pipe_format                db_render_format;
 257         bool                            is_depth;
 258         bool                            db_compatible;
 259         bool                            can_sample_z;
 260         bool                            can_sample_s;
 261         unsigned                        dirty_level_mask; /* each bit says if that mipmap is compressed */
 262         unsigned                        stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
 263         struct r600_texture             *flushed_depth_texture;
 264         struct radeon_surf              surface;
 265
 266         /* Colorbuffer compression and fast clear. */
 267         struct r600_fmask_info          fmask;
 268         struct r600_cmask_info          cmask;
 269         struct r600_resource            *cmask_buffer;
 270         uint64_t                        dcc_offset; /* 0 = disabled */
 271         unsigned                        cb_color_info; /* fast clear enable bit */
 272         unsigned                        color_clear_value[2];
 273         unsigned                        last_msaa_resolve_target_micro_mode;
 274
 275         /* Depth buffer compression and fast clear. */
 276         struct r600_htile_info          htile;
 277         struct r600_resource            *htile_buffer;
 278         bool                            tc_compatible_htile;
 279         bool                            depth_cleared; /* if it was cleared at least once */
 280         float                           depth_clear_value;
 281         bool                            stencil_cleared; /* if it was cleared at least once */
 282         uint8_t                         stencil_clear_value;
 283
 284         bool                            non_disp_tiling; /* R600-Cayman only */
 285
 286         /* Whether the texture is a displayable back buffer and needs DCC
 287          * decompression, which is expensive. Therefore, it's enabled only
 288          * if statistics suggest that it will pay off and it's allocated
 289          * separately. It can't be bound as a sampler by apps. Limited to
 290          * target == 2D and last_level == 0. If enabled, dcc_offset contains
 291          * the absolute GPUVM address, not the relative one.
 292          */
 293         struct r600_resource            *dcc_separate_buffer;
 294         /* When DCC is temporarily disabled, the separate buffer is here. */
 295         struct r600_resource            *last_dcc_separate_buffer;
 296         /* We need to track DCC dirtiness, because st/dri usually calls
 297          * flush_resource twice per frame (not a bug) and we don't wanna
 298          * decompress DCC twice. Also, the dirty tracking must be done even
 299          * if DCC isn't used, because it's required by the DCC usage analysis
 300          * for a possible future enablement.
 301          */
 302         bool                            separate_dcc_dirty;
 303         /* Statistics gathering for the DCC enablement heuristic. */
 304         bool                            dcc_gather_statistics;
 305         /* Estimate of how much this color buffer is written to in units of
 306          * full-screen draws: ps_invocations / (width * height)
 307          * Shader kills, late Z, and blending with trivial discards make it
 308          * inaccurate (we need to count CB updates, not PS invocations).
 309          */
 310         unsigned                        ps_draw_ratio;
 311         /* The number of clears since the last DCC usage analysis. */
 312         unsigned                        num_slow_clears;
 313
 314         /* Counter that should be non-zero if the texture is bound to a
 315          * framebuffer. Implemented in radeonsi only.
 316          */
 317         uint32_t                        framebuffers_bound;
 318 };
 319
 320 struct r600_surface {
 321         struct pipe_surface             base;
 322         const struct radeon_surf_level  *level_info;
 323
 324         bool color_initialized;
 325         bool depth_initialized;
 326
 327         /* Misc. color flags. */
 328         bool alphatest_bypass;
 329         bool export_16bpc;
 330         bool color_is_int8;
 331
 332         /* Color registers. */
 333         unsigned cb_color_info;
 334         unsigned cb_color_base;
 335         unsigned cb_color_view;
 336         unsigned cb_color_size;         /* R600 only */
 337         unsigned cb_color_dim;          /* EG only */
 338         unsigned cb_color_pitch;        /* EG and later */
 339         unsigned cb_color_slice;        /* EG and later */
 340         unsigned cb_color_attrib;       /* EG and later */
 341         unsigned cb_dcc_control;        /* VI and later */
 342         unsigned cb_color_fmask;        /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
 343         unsigned cb_color_fmask_slice;  /* EG and later */
 344         unsigned cb_color_cmask;        /* CB_COLORn_TILE (r600 only) */
 345         unsigned cb_color_mask;         /* R600 only */
 346         unsigned spi_shader_col_format;         /* SI+, no blending, no alpha-to-coverage. */
 347         unsigned spi_shader_col_format_alpha;   /* SI+, alpha-to-coverage */
 348         unsigned spi_shader_col_format_blend;   /* SI+, blending without alpha. */
 349         unsigned spi_shader_col_format_blend_alpha; /* SI+, blending with alpha. */
 350         struct r600_resource *cb_buffer_fmask; /* Used for FMASK relocations. R600 only */
 351         struct r600_resource *cb_buffer_cmask; /* Used for CMASK relocations. R600 only */
 352
 353         /* DB registers. */
 354         unsigned db_depth_info;         /* R600 only, then SI and later */
 355         unsigned db_z_info;             /* EG and later */
 356         unsigned db_depth_base;         /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
 357         unsigned db_depth_view;
 358         unsigned db_depth_size;
 359         unsigned db_depth_slice;        /* EG and later */
 360         unsigned db_stencil_base;       /* EG and later */
 361         unsigned db_stencil_info;       /* EG and later */
 362         unsigned db_prefetch_limit;     /* R600 only */
 363         unsigned db_htile_surface;
 364         unsigned db_htile_data_base;
 365         unsigned db_preload_control;    /* EG and later */
 366 };
 367
 368 struct r600_common_screen {
 369         struct pipe_screen              b;
 370         struct radeon_winsys            *ws;
 371         enum radeon_family              family;
 372         enum chip_class                 chip_class;
 373         struct radeon_info              info;
 374         uint64_t                        debug_flags;
 375         bool                            has_cp_dma;
 376         bool                            has_streamout;
 377
 378         struct slab_parent_pool         pool_transfers;
 379
 380         /* Texture filter settings. */
 381         int                             force_aniso; /* -1 = disabled */
 382
 383         /* Auxiliary context. Mainly used to initialize resources.
 384          * It must be locked prior to using and flushed before unlocking. */
 385         struct pipe_context             *aux_context;
 386         pipe_mutex                      aux_context_lock;
 387
 388         /* This must be in the screen, because UE4 uses one context for
 389          * compilation and another one for rendering.
 390          */
 391         unsigned                        num_compilations;
 392         /* Along with ST_DEBUG=precompile, this should show if applications
 393          * are loading shaders on demand. This is a monotonic counter.
 394          */
 395         unsigned                        num_shaders_created;
 396
 397         /* GPU load thread. */
 398         pipe_mutex                      gpu_load_mutex;
 399         pipe_thread                     gpu_load_thread;
 400         unsigned                        gpu_load_counter_busy;
 401         unsigned                        gpu_load_counter_idle;
 402         volatile unsigned               gpu_load_stop_thread; /* bool */
 403
 404         char                            renderer_string[100];
 405
 406         /* Performance counters. */
 407         struct r600_perfcounters        *perfcounters;
 408
 409         /* If pipe_screen wants to re-emit the framebuffer state of all
 410          * contexts, it should atomically increment this. Each context will
 411          * compare this with its own last known value of the counter before
 412          * drawing and re-emit the framebuffer state accordingly.
 413          */
 414         unsigned                        dirty_fb_counter;
 415
 416         /* Atomically increment this counter when an existing texture's
 417          * metadata is enabled or disabled in a way that requires changing
 418          * contexts' compressed texture binding masks.
 419          */
 420         unsigned                        compressed_colortex_counter;
 421
 422         /* Atomically increment this counter when an existing texture's
 423          * backing buffer or tile mode parameters have changed that requires
 424          * recomputation of shader descriptors.
 425          */
 426         unsigned                        dirty_tex_descriptor_counter;
 427
 428         struct {
 429                 /* Context flags to set so that all writes from earlier jobs
 430                  * in the CP are seen by L2 clients.
 431                  */
 432                 unsigned cp_to_L2;
 433
 434                 /* Context flags to set so that all writes from earlier
 435                  * compute jobs are seen by L2 clients.
 436                  */
 437                 unsigned compute_to_L2;
 438         } barrier_flags;
 439
 440         void (*query_opaque_metadata)(struct r600_common_screen *rscreen,
 441                                       struct r600_texture *rtex,
 442                                       struct radeon_bo_metadata *md);
 443
 444         void (*apply_opaque_metadata)(struct r600_common_screen *rscreen,
 445                                     struct r600_texture *rtex,
 446                                     struct radeon_bo_metadata *md);
 447 };
 448
 449 /* This encapsulates a state or an operation which can emitted into the GPU
 450  * command stream. */
 451 struct r600_atom {
 452         void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
 453         unsigned                num_dw;
 454         unsigned short          id;
 455 };
 456
 457 struct r600_so_target {
 458         struct pipe_stream_output_target b;
 459
 460         /* The buffer where BUFFER_FILLED_SIZE is stored. */
 461         struct r600_resource    *buf_filled_size;
 462         unsigned                buf_filled_size_offset;
 463         bool                    buf_filled_size_valid;
 464
 465         unsigned                stride_in_dw;
 466 };
 467
 468 struct r600_streamout {
 469         struct r600_atom                begin_atom;
 470         bool                            begin_emitted;
 471         unsigned                        num_dw_for_end;
 472
 473         unsigned                        enabled_mask;
 474         unsigned                        num_targets;
 475         struct r600_so_target           *targets[PIPE_MAX_SO_BUFFERS];
 476
 477         unsigned                        append_bitmask;
 478         bool                            suspended;
 479
 480         /* External state which comes from the vertex shader,
 481          * it must be set explicitly when binding a shader. */
 482         unsigned                        *stride_in_dw;
 483         unsigned                        enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
 484
 485         /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
 486         unsigned                        hw_enabled_mask;
 487
 488         /* The state of VGT_STRMOUT_(CONFIG|EN). */
 489         struct r600_atom                enable_atom;
 490         bool                            streamout_enabled;
 491         bool                            prims_gen_query_enabled;
 492         int                             num_prims_gen_queries;
 493 };
 494
 495 struct r600_signed_scissor {
 496         int minx;
 497         int miny;
 498         int maxx;
 499         int maxy;
 500 };
 501
 502 struct r600_scissors {
 503         struct r600_atom                atom;
 504         unsigned                        dirty_mask;
 505         struct pipe_scissor_state       states[R600_MAX_VIEWPORTS];
 506 };
 507
 508 struct r600_viewports {
 509         struct r600_atom                atom;
 510         unsigned                        dirty_mask;
 511         unsigned                        depth_range_dirty_mask;
 512         struct pipe_viewport_state      states[R600_MAX_VIEWPORTS];
 513         struct r600_signed_scissor      as_scissor[R600_MAX_VIEWPORTS];
 514 };
 515
 516 struct r600_ring {
 517         struct radeon_winsys_cs         *cs;
 518         void (*flush)(void *ctx, unsigned flags,
 519                       struct pipe_fence_handle **fence);
 520 };
 521
 522 /* Saved CS data for debugging features. */
 523 struct radeon_saved_cs {
 524         uint32_t                        *ib;
 525         unsigned                        num_dw;
 526
 527         struct radeon_bo_list_item      *bo_list;
 528         unsigned                        bo_count;
 529 };
 530
 531 struct r600_common_context {
 532         struct pipe_context b; /* base class */
 533
 534         struct r600_common_screen       *screen;
 535         struct radeon_winsys            *ws;
 536         struct radeon_winsys_ctx        *ctx;
 537         enum radeon_family              family;
 538         enum chip_class                 chip_class;
 539         struct r600_ring                gfx;
 540         struct r600_ring                dma;
 541         struct pipe_fence_handle        *last_gfx_fence;
 542         struct pipe_fence_handle        *last_sdma_fence;
 543         unsigned                        num_gfx_cs_flushes;
 544         unsigned                        initial_gfx_cs_size;
 545         unsigned                        gpu_reset_counter;
 546         unsigned                        last_dirty_fb_counter;
 547         unsigned                        last_compressed_colortex_counter;
 548         unsigned                        last_dirty_tex_descriptor_counter;
 549
 550         struct u_upload_mgr             *uploader;
 551         struct u_suballocator           *allocator_zeroed_memory;
 552         struct slab_child_pool          pool_transfers;
 553
 554         /* Current unaccounted memory usage. */
 555         uint64_t                        vram;
 556         uint64_t                        gtt;
 557
 558         /* States. */
 559         struct r600_streamout           streamout;
 560         struct r600_scissors            scissors;
 561         struct r600_viewports           viewports;
 562         bool                            scissor_enabled;
 563         bool                            clip_halfz;
 564         bool                            vs_writes_viewport_index;
 565         bool                            vs_disables_clipping_viewport;
 566
 567         /* Additional context states. */
 568         unsigned flags; /* flush flags */
 569
 570         /* Queries. */
 571         /* Maintain the list of active queries for pausing between IBs. */
 572         int                             num_occlusion_queries;
 573         int                             num_perfect_occlusion_queries;
 574         struct list_head                active_queries;
 575         unsigned                        num_cs_dw_queries_suspend;
 576         /* Additional hardware info. */
 577         unsigned                        backend_mask;
 578         unsigned                        max_db; /* for OQ */
 579         /* Misc stats. */
 580         unsigned                        num_draw_calls;
 581         unsigned                        num_spill_draw_calls;
 582         unsigned                        num_compute_calls;
 583         unsigned                        num_spill_compute_calls;
 584         unsigned                        num_dma_calls;
 585         unsigned                        num_vs_flushes;
 586         unsigned                        num_ps_flushes;
 587         unsigned                        num_cs_flushes;
 588         uint64_t                        num_alloc_tex_transfer_bytes;
 589         unsigned                        last_tex_ps_draw_ratio; /* for query */
 590
 591         /* Render condition. */
 592         struct r600_atom                render_cond_atom;
 593         struct pipe_query               *render_cond;
 594         unsigned                        render_cond_mode;
 595         bool                            render_cond_invert;
 596         bool                            render_cond_force_off; /* for u_blitter */
 597
 598         /* MSAA sample locations.
 599          * The first index is the sample index.
 600          * The second index is the coordinate: X, Y. */
 601         float                           sample_locations_1x[1][2];
 602         float                           sample_locations_2x[2][2];
 603         float                           sample_locations_4x[4][2];
 604         float                           sample_locations_8x[8][2];
 605         float                           sample_locations_16x[16][2];
 606
 607         /* Statistics gathering for the DCC enablement heuristic. It can't be
 608          * in r600_texture because r600_texture can be shared by multiple
 609          * contexts. This is for back buffers only. We shouldn't get too many
 610          * of those.
 611          *
 612          * X11 DRI3 rotates among a finite set of back buffers. They should
 613          * all fit in this array. If they don't, separate DCC might never be
 614          * enabled by DCC stat gathering.
 615          */
 616         struct {
 617                 struct r600_texture             *tex;
 618                 /* Query queue: 0 = usually active, 1 = waiting, 2 = readback. */
 619                 struct pipe_query               *ps_stats[3];
 620                 /* If all slots are used and another slot is needed,
 621                  * the least recently used slot is evicted based on this. */
 622                 int64_t                         last_use_timestamp;
 623                 bool                            query_active;
 624         } dcc_stats[5];
 625
 626         struct pipe_debug_callback      debug;
 627         struct pipe_device_reset_callback device_reset_callback;
 628
 629         void                            *query_result_shader;
 630
 631         /* Copy one resource to another using async DMA. */
 632         void (*dma_copy)(struct pipe_context *ctx,
 633                          struct pipe_resource *dst,
 634                          unsigned dst_level,
 635                          unsigned dst_x, unsigned dst_y, unsigned dst_z,
 636                          struct pipe_resource *src,
 637                          unsigned src_level,
 638                          const struct pipe_box *src_box);
 639
 640         void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
 641                              uint64_t offset, uint64_t size, unsigned value,
 642                              enum r600_coherency coher);
 643
 644         void (*blit_decompress_depth)(struct pipe_context *ctx,
 645                                       struct r600_texture *texture,
 646                                       struct r600_texture *staging,
 647                                       unsigned first_level, unsigned last_level,
 648                                       unsigned first_layer, unsigned last_layer,
 649                                       unsigned first_sample, unsigned last_sample);
 650
 651         void (*decompress_dcc)(struct pipe_context *ctx,
 652                                struct r600_texture *rtex);
 653
 654         /* Reallocate the buffer and update all resource bindings where
 655          * the buffer is bound, including all resource descriptors. */
 656         void (*invalidate_buffer)(struct pipe_context *ctx, struct pipe_resource *buf);
 657
 658         /* Enable or disable occlusion queries. */
 659         void (*set_occlusion_query_state)(struct pipe_context *ctx, bool enable);
 660
 661         void (*save_qbo_state)(struct pipe_context *ctx, struct r600_qbo_state *st);
 662
 663         /* This ensures there is enough space in the command stream. */
 664         void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
 665                                   bool include_draw_vbo);
 666
 667         void (*set_atom_dirty)(struct r600_common_context *ctx,
 668                                struct r600_atom *atom, bool dirty);
 669
 670         void (*check_vm_faults)(struct r600_common_context *ctx,
 671                                 struct radeon_saved_cs *saved,
 672                                 enum ring_type ring);
 673 };
 674
 675 /* r600_buffer.c */
 676 bool r600_rings_is_buffer_referenced(struct r600_common_context *ctx,
 677                                      struct pb_buffer *buf,
 678                                      enum radeon_bo_usage usage);
 679 void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx,
 680                                       struct r600_resource *resource,
 681                                       unsigned usage);
 682 void r600_buffer_subdata(struct pipe_context *ctx,
 683                          struct pipe_resource *buffer,
 684                          unsigned usage, unsigned offset,
 685                          unsigned size, const void *data);
 686 void r600_init_resource_fields(struct r600_common_screen *rscreen,
 687                                struct r600_resource *res,
 688                                uint64_t size, unsigned alignment);
 689 bool r600_alloc_resource(struct r600_common_screen *rscreen,
 690                          struct r600_resource *res);
 691 struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 692                                          const struct pipe_resource *templ,
 693                                          unsigned alignment);
 694 struct pipe_resource * r600_aligned_buffer_create(struct pipe_screen *screen,
 695                                                   unsigned bind,
 696                                                   unsigned usage,
 697                                                   unsigned size,
 698                                                   unsigned alignment);
 699 struct pipe_resource *
 700 r600_buffer_from_user_memory(struct pipe_screen *screen,
 701                              const struct pipe_resource *templ,
 702                              void *user_memory);
 703 void
 704 r600_invalidate_resource(struct pipe_context *ctx,
 705                          struct pipe_resource *resource);
 706
 707 /* r600_common_pipe.c */
 708 void r600_gfx_write_fence(struct r600_common_context *ctx, struct r600_resource *buf,
 709                           uint64_t va, uint32_t old_value, uint32_t new_value);
 710 unsigned r600_gfx_write_fence_dwords(struct r600_common_screen *screen);
 711 void r600_gfx_wait_fence(struct r600_common_context *ctx,
 712                          uint64_t va, uint32_t ref, uint32_t mask);
 713 void r600_draw_rectangle(struct blitter_context *blitter,
 714                          int x1, int y1, int x2, int y2, float depth,
 715                          enum blitter_attrib_type type,
 716                          const union pipe_color_union *attrib);
 717 bool r600_common_screen_init(struct r600_common_screen *rscreen,
 718                              struct radeon_winsys *ws);
 719 void r600_destroy_common_screen(struct r600_common_screen *rscreen);
 720 void r600_preflush_suspend_features(struct r600_common_context *ctx);
 721 void r600_postflush_resume_features(struct r600_common_context *ctx);
 722 bool r600_common_context_init(struct r600_common_context *rctx,
 723                               struct r600_common_screen *rscreen,
 724                               unsigned context_flags);
 725 void r600_common_context_cleanup(struct r600_common_context *rctx);
 726 bool r600_can_dump_shader(struct r600_common_screen *rscreen,
 727                           unsigned processor);
 728 bool r600_extra_shader_checks(struct r600_common_screen *rscreen,
 729                               unsigned processor);
 730 void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
 731                               uint64_t offset, uint64_t size, unsigned value,
 732                               enum r600_coherency coher);
 733 struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
 734                                                   const struct pipe_resource *templ);
 735 const char *r600_get_llvm_processor_name(enum radeon_family family);
 736 void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
 737                          struct r600_resource *dst, struct r600_resource *src);
 738 void r600_dma_emit_wait_idle(struct r600_common_context *rctx);
 739 void radeon_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
 740                     struct radeon_saved_cs *saved);
 741 void radeon_clear_saved_cs(struct radeon_saved_cs *saved);
 742 bool r600_check_device_reset(struct r600_common_context *rctx);
 743
 744 /* r600_gpu_load.c */
 745 void r600_gpu_load_kill_thread(struct r600_common_screen *rscreen);
 746 uint64_t r600_gpu_load_begin(struct r600_common_screen *rscreen);
 747 unsigned r600_gpu_load_end(struct r600_common_screen *rscreen, uint64_t begin);
 748
 749 /* r600_perfcounters.c */
 750 void r600_perfcounters_destroy(struct r600_common_screen *rscreen);
 751
 752 /* r600_query.c */
 753 void r600_init_screen_query_functions(struct r600_common_screen *rscreen);
 754 void r600_query_init(struct r600_common_context *rctx);
 755 void r600_suspend_queries(struct r600_common_context *ctx);
 756 void r600_resume_queries(struct r600_common_context *ctx);
 757 void r600_query_init_backend_mask(struct r600_common_context *ctx);
 758
 759 /* r600_streamout.c */
 760 void r600_streamout_buffers_dirty(struct r600_common_context *rctx);
 761 void r600_set_streamout_targets(struct pipe_context *ctx,
 762                                 unsigned num_targets,
 763                                 struct pipe_stream_output_target **targets,
 764                                 const unsigned *offset);
 765 void r600_emit_streamout_end(struct r600_common_context *rctx);
 766 void r600_update_prims_generated_query_state(struct r600_common_context *rctx,
 767                                              unsigned type, int diff);
 768 void r600_streamout_init(struct r600_common_context *rctx);
 769
 770 /* r600_test_dma.c */
 771 void r600_test_dma(struct r600_common_screen *rscreen);
 772
 773 /* r600_texture.c */
 774 bool r600_prepare_for_dma_blit(struct r600_common_context *rctx,
 775                                 struct r600_texture *rdst,
 776                                 unsigned dst_level, unsigned dstx,
 777                                 unsigned dsty, unsigned dstz,
 778                                 struct r600_texture *rsrc,
 779                                 unsigned src_level,
 780                                 const struct pipe_box *src_box);
 781 void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
 782                                  struct r600_texture *rtex,
 783                                  unsigned nr_samples,
 784                                  struct r600_fmask_info *out);
 785 void r600_texture_get_cmask_info(struct r600_common_screen *rscreen,
 786                                  struct r600_texture *rtex,
 787                                  struct r600_cmask_info *out);
 788 bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
 789                                      struct pipe_resource *texture,
 790                                      struct r600_texture **staging);
 791 void r600_print_texture_info(struct r600_texture *rtex, FILE *f);
 792 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 793                                         const struct pipe_resource *templ);
 794 bool vi_dcc_formats_compatible(enum pipe_format format1,
 795                                enum pipe_format format2);
 796 void vi_dcc_disable_if_incompatible_format(struct r600_common_context *rctx,
 797                                            struct pipe_resource *tex,
 798                                            unsigned level,
 799                                            enum pipe_format view_format);
 800 struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
 801                                                 struct pipe_resource *texture,
 802                                                 const struct pipe_surface *templ,
 803                                                 unsigned width, unsigned height);
 804 unsigned r600_translate_colorswap(enum pipe_format format, bool do_endian_swap);
 805 void vi_separate_dcc_start_query(struct pipe_context *ctx,
 806                                  struct r600_texture *tex);
 807 void vi_separate_dcc_stop_query(struct pipe_context *ctx,
 808                                 struct r600_texture *tex);
 809 void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx,
 810                                              struct r600_texture *tex);
 811 void vi_dcc_clear_level(struct r600_common_context *rctx,
 812                         struct r600_texture *rtex,
 813                         unsigned level, unsigned clear_value);
 814 void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
 815                                    struct pipe_framebuffer_state *fb,
 816                                    struct r600_atom *fb_state,
 817                                    unsigned *buffers, unsigned *dirty_cbufs,
 818                                    const union pipe_color_union *color);
 819 bool r600_texture_disable_dcc(struct r600_common_context *rctx,
 820                               struct r600_texture *rtex);
 821 void r600_init_screen_texture_functions(struct r600_common_screen *rscreen);
 822 void r600_init_context_texture_functions(struct r600_common_context *rctx);
 823
 824 /* r600_viewport.c */
 825 void evergreen_apply_scissor_bug_workaround(struct r600_common_context *rctx,
 826                                             struct pipe_scissor_state *scissor);
 827 void r600_viewport_set_rast_deps(struct r600_common_context *rctx,
 828                                  bool scissor_enable, bool clip_halfz);
 829 void r600_update_vs_writes_viewport_index(struct r600_common_context *rctx,
 830                                           struct tgsi_shader_info *info);
 831 void r600_init_viewport_functions(struct r600_common_context *rctx);
 832
 833 /* cayman_msaa.c */
 834 extern const uint32_t eg_sample_locs_2x[4];
 835 extern const unsigned eg_max_dist_2x;
 836 extern const uint32_t eg_sample_locs_4x[4];
 837 extern const unsigned eg_max_dist_4x;
 838 void cayman_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
 839                                 unsigned sample_index, float *out_value);
 840 void cayman_init_msaa(struct pipe_context *ctx);
 841 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
 842 void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
 843                              int ps_iter_samples, int overrast_samples,
 844                              unsigned sc_mode_cntl_1);
 845
 846
 847 /* Inline helpers. */
 848
 849 static inline struct r600_resource *r600_resource(struct pipe_resource *r)
 850 {
 851         return (struct r600_resource*)r;
 852 }
 853
 854 static inline void
 855 r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
 856 {
 857         pipe_resource_reference((struct pipe_resource **)ptr,
 858                                 (struct pipe_resource *)res);
 859 }
 860
 861 static inline void
 862 r600_texture_reference(struct r600_texture **ptr, struct r600_texture *res)
 863 {
 864         pipe_resource_reference((struct pipe_resource **)ptr, &res->resource.b.b);
 865 }
 866
 867 static inline void
 868 r600_context_add_resource_size(struct pipe_context *ctx, struct pipe_resource *r)
 869 {
 870         struct r600_common_context *rctx = (struct r600_common_context *)ctx;
 871         struct r600_resource *res = (struct r600_resource *)r;
 872
 873         if (res) {
 874                 /* Add memory usage for need_gfx_cs_space */
 875                 rctx->vram += res->vram_usage;
 876                 rctx->gtt += res->gart_usage;
 877         }
 878 }
 879
 880 static inline bool r600_get_strmout_en(struct r600_common_context *rctx)
 881 {
 882         return rctx->streamout.streamout_enabled ||
 883                rctx->streamout.prims_gen_query_enabled;
 884 }
 885
 886 #define     SQ_TEX_XY_FILTER_POINT                         0x00
 887 #define     SQ_TEX_XY_FILTER_BILINEAR                      0x01
 888 #define     SQ_TEX_XY_FILTER_ANISO_POINT                   0x02
 889 #define     SQ_TEX_XY_FILTER_ANISO_BILINEAR                0x03
 890
 891 static inline unsigned eg_tex_filter(unsigned filter, unsigned max_aniso)
 892 {
 893         if (filter == PIPE_TEX_FILTER_LINEAR)
 894                 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_BILINEAR
 895                                      : SQ_TEX_XY_FILTER_BILINEAR;
 896         else
 897                 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_POINT
 898                                      : SQ_TEX_XY_FILTER_POINT;
 899 }
 900
 901 static inline unsigned r600_tex_aniso_filter(unsigned filter)
 902 {
 903         if (filter < 2)
 904                 return 0;
 905         if (filter < 4)
 906                 return 1;
 907         if (filter < 8)
 908                 return 2;
 909         if (filter < 16)
 910                 return 3;
 911         return 4;
 912 }
 913
 914 static inline unsigned r600_wavefront_size(enum radeon_family family)
 915 {
 916         switch (family) {
 917         case CHIP_RV610:
 918         case CHIP_RS780:
 919         case CHIP_RV620:
 920         case CHIP_RS880:
 921                 return 16;
 922         case CHIP_RV630:
 923         case CHIP_RV635:
 924         case CHIP_RV730:
 925         case CHIP_RV710:
 926         case CHIP_PALM:
 927         case CHIP_CEDAR:
 928                 return 32;
 929         default:
 930                 return 64;
 931         }
 932 }
 933
 934 static inline enum radeon_bo_priority
 935 r600_get_sampler_view_priority(struct r600_resource *res)
 936 {
 937         if (res->b.b.target == PIPE_BUFFER)
 938                 return RADEON_PRIO_SAMPLER_BUFFER;
 939
 940         if (res->b.b.nr_samples > 1)
 941                 return RADEON_PRIO_SAMPLER_TEXTURE_MSAA;
 942
 943         return RADEON_PRIO_SAMPLER_TEXTURE;
 944 }
 945
 946 static inline bool
 947 r600_can_sample_zs(struct r600_texture *tex, bool stencil_sampler)
 948 {
 949         return (stencil_sampler && tex->can_sample_s) ||
 950                (!stencil_sampler && tex->can_sample_z);
 951 }
 952
 953 #define COMPUTE_DBG(rscreen, fmt, args...) \
 954         do { \
 955                 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
 956         } while (0);
 957
 958 #define R600_ERR(fmt, args...) \
 959         fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
 960
 961 /* For MSAA sample positions. */
 962 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y)  \
 963         (((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) |                  \
 964         (((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) |         \
 965         (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) |        \
 966          (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))
 967
 968 #endif