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
  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
  49 #define ATI_VENDOR_ID 0x1002
  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
  57 #define R600_CONTEXT_STREAMOUT_FLUSH            (1u << 0)
  58 /* Pipeline & streamout query controls. */
  59 #define R600_CONTEXT_START_PIPELINE_STATS       (1u << 1)
  60 #define R600_CONTEXT_STOP_PIPELINE_STATS        (1u << 2)
  61 #define R600_CONTEXT_FLUSH_FOR_RENDER_COND      (1u << 3)
  62 #define R600_CONTEXT_PRIVATE_FLAG               (1u << 4)
  63
  64 /* special primitive types */
  65 #define R600_PRIM_RECTANGLE_LIST        PIPE_PRIM_MAX
  66
  67 #define R600_NOT_QUERY          0xffffffff
  68
  69 /* Debug flags. */
  70 enum {
  71         /* Shader logging options: */
  72         DBG_VS = PIPE_SHADER_VERTEX,
  73         DBG_PS = PIPE_SHADER_FRAGMENT,
  74         DBG_GS = PIPE_SHADER_GEOMETRY,
  75         DBG_TCS = PIPE_SHADER_TESS_CTRL,
  76         DBG_TES = PIPE_SHADER_TESS_EVAL,
  77         DBG_CS = PIPE_SHADER_COMPUTE,
  78         DBG_NO_IR,
  79         DBG_NO_TGSI,
  80         DBG_NO_ASM,
  81         DBG_PREOPT_IR,
  82
  83         /* Shader compiler options the shader cache should be aware of: */
  84         DBG_FS_CORRECT_DERIVS_AFTER_KILL,
  85         DBG_UNSAFE_MATH,
  86         DBG_SI_SCHED,
  87
  88         /* Shader compiler options (with no effect on the shader cache): */
  89         DBG_CHECK_IR,
  90         DBG_PRECOMPILE,
  91         DBG_NIR,
  92         DBG_MONOLITHIC_SHADERS,
  93         DBG_NO_OPT_VARIANT,
  94
  95         /* Information logging options: */
  96         DBG_INFO,
  97         DBG_TEX,
  98         DBG_COMPUTE,
  99         DBG_VM,
 100
 101         /* Driver options: */
 102         DBG_FORCE_DMA,
 103         DBG_NO_ASYNC_DMA,
 104         DBG_NO_DISCARD_RANGE,
 105         DBG_NO_WC,
 106         DBG_CHECK_VM,
 107         DBG_RESERVE_VMID,
 108
 109         /* 3D engine options: */
 110         DBG_SWITCH_ON_EOP,
 111         DBG_NO_OUT_OF_ORDER,
 112         DBG_NO_DPBB,
 113         DBG_NO_DFSM,
 114         DBG_DPBB,
 115         DBG_DFSM,
 116         DBG_NO_HYPERZ,
 117         DBG_NO_RB_PLUS,
 118         DBG_NO_2D_TILING,
 119         DBG_NO_TILING,
 120         DBG_NO_DCC,
 121         DBG_NO_DCC_CLEAR,
 122         DBG_NO_DCC_FB,
 123
 124         /* Tests: */
 125         DBG_TEST_DMA,
 126         DBG_TEST_VMFAULT_CP,
 127         DBG_TEST_VMFAULT_SDMA,
 128         DBG_TEST_VMFAULT_SHADER,
 129 };
 130
 131 #define DBG_ALL_SHADERS         (((1 << (DBG_CS + 1)) - 1))
 132 #define DBG(name)               (1ull << DBG_##name)
 133
 134 #define R600_MAP_BUFFER_ALIGNMENT 64
 135
 136 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 1024
 137
 138 enum r600_coherency {
 139         R600_COHERENCY_NONE, /* no cache flushes needed */
 140         R600_COHERENCY_SHADER,
 141         R600_COHERENCY_CB_META,
 142 };
 143
 144 #ifdef PIPE_ARCH_BIG_ENDIAN
 145 #define R600_BIG_ENDIAN 1
 146 #else
 147 #define R600_BIG_ENDIAN 0
 148 #endif
 149
 150 struct r600_common_context;
 151 struct r600_perfcounters;
 152 struct tgsi_shader_info;
 153 struct r600_qbo_state;
 154
 155 void si_radeon_shader_binary_init(struct ac_shader_binary *b);
 156 void si_radeon_shader_binary_clean(struct ac_shader_binary *b);
 157
 158 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
 159  * at the moment.
 160  */
 161 struct r600_resource {
 162         struct threaded_resource        b;
 163
 164         /* Winsys objects. */
 165         struct pb_buffer                *buf;
 166         uint64_t                        gpu_address;
 167         /* Memory usage if the buffer placement is optimal. */
 168         uint64_t                        vram_usage;
 169         uint64_t                        gart_usage;
 170
 171         /* Resource properties. */
 172         uint64_t                        bo_size;
 173         unsigned                        bo_alignment;
 174         enum radeon_bo_domain           domains;
 175         enum radeon_bo_flag             flags;
 176         unsigned                        bind_history;
 177         int                             max_forced_staging_uploads;
 178
 179         /* The buffer range which is initialized (with a write transfer,
 180          * streamout, DMA, or as a random access target). The rest of
 181          * the buffer is considered invalid and can be mapped unsynchronized.
 182          *
 183          * This allows unsychronized mapping of a buffer range which hasn't
 184          * been used yet. It's for applications which forget to use
 185          * the unsynchronized map flag and expect the driver to figure it out.
 186          */
 187         struct util_range               valid_buffer_range;
 188
 189         /* For buffers only. This indicates that a write operation has been
 190          * performed by TC L2, but the cache hasn't been flushed.
 191          * Any hw block which doesn't use or bypasses TC L2 should check this
 192          * flag and flush the cache before using the buffer.
 193          *
 194          * For example, TC L2 must be flushed if a buffer which has been
 195          * modified by a shader store instruction is about to be used as
 196          * an index buffer. The reason is that VGT DMA index fetching doesn't
 197          * use TC L2.
 198          */
 199         bool                            TC_L2_dirty;
 200
 201         /* Whether the resource has been exported via resource_get_handle. */
 202         unsigned                        external_usage; /* PIPE_HANDLE_USAGE_* */
 203
 204         /* Whether this resource is referenced by bindless handles. */
 205         bool                            texture_handle_allocated;
 206         bool                            image_handle_allocated;
 207 };
 208
 209 struct r600_transfer {
 210         struct threaded_transfer        b;
 211         struct r600_resource            *staging;
 212         unsigned                        offset;
 213 };
 214
 215 struct r600_fmask_info {
 216         uint64_t offset;
 217         uint64_t size;
 218         unsigned alignment;
 219         unsigned pitch_in_pixels;
 220         unsigned bank_height;
 221         unsigned slice_tile_max;
 222         unsigned tile_mode_index;
 223         unsigned tile_swizzle;
 224 };
 225
 226 struct r600_cmask_info {
 227         uint64_t offset;
 228         uint64_t size;
 229         unsigned alignment;
 230         unsigned slice_tile_max;
 231         uint64_t base_address_reg;
 232 };
 233
 234 struct r600_texture {
 235         struct r600_resource            resource;
 236
 237         uint64_t                        size;
 238         unsigned                        num_level0_transfers;
 239         enum pipe_format                db_render_format;
 240         bool                            is_depth;
 241         bool                            db_compatible;
 242         bool                            can_sample_z;
 243         bool                            can_sample_s;
 244         unsigned                        dirty_level_mask; /* each bit says if that mipmap is compressed */
 245         unsigned                        stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
 246         struct r600_texture             *flushed_depth_texture;
 247         struct radeon_surf              surface;
 248
 249         /* Colorbuffer compression and fast clear. */
 250         struct r600_fmask_info          fmask;
 251         struct r600_cmask_info          cmask;
 252         struct r600_resource            *cmask_buffer;
 253         uint64_t                        dcc_offset; /* 0 = disabled */
 254         unsigned                        cb_color_info; /* fast clear enable bit */
 255         unsigned                        color_clear_value[2];
 256         unsigned                        last_msaa_resolve_target_micro_mode;
 257
 258         /* Depth buffer compression and fast clear. */
 259         uint64_t                        htile_offset;
 260         bool                            tc_compatible_htile;
 261         bool                            depth_cleared; /* if it was cleared at least once */
 262         float                           depth_clear_value;
 263         bool                            stencil_cleared; /* if it was cleared at least once */
 264         uint8_t                         stencil_clear_value;
 265         bool                            upgraded_depth; /* upgraded from unorm to Z32_FLOAT */
 266
 267         bool                            non_disp_tiling; /* R600-Cayman only */
 268
 269         /* Whether the texture is a displayable back buffer and needs DCC
 270          * decompression, which is expensive. Therefore, it's enabled only
 271          * if statistics suggest that it will pay off and it's allocated
 272          * separately. It can't be bound as a sampler by apps. Limited to
 273          * target == 2D and last_level == 0. If enabled, dcc_offset contains
 274          * the absolute GPUVM address, not the relative one.
 275          */
 276         struct r600_resource            *dcc_separate_buffer;
 277         /* When DCC is temporarily disabled, the separate buffer is here. */
 278         struct r600_resource            *last_dcc_separate_buffer;
 279         /* We need to track DCC dirtiness, because st/dri usually calls
 280          * flush_resource twice per frame (not a bug) and we don't wanna
 281          * decompress DCC twice. Also, the dirty tracking must be done even
 282          * if DCC isn't used, because it's required by the DCC usage analysis
 283          * for a possible future enablement.
 284          */
 285         bool                            separate_dcc_dirty;
 286         /* Statistics gathering for the DCC enablement heuristic. */
 287         bool                            dcc_gather_statistics;
 288         /* Estimate of how much this color buffer is written to in units of
 289          * full-screen draws: ps_invocations / (width * height)
 290          * Shader kills, late Z, and blending with trivial discards make it
 291          * inaccurate (we need to count CB updates, not PS invocations).
 292          */
 293         unsigned                        ps_draw_ratio;
 294         /* The number of clears since the last DCC usage analysis. */
 295         unsigned                        num_slow_clears;
 296
 297         /* Counter that should be non-zero if the texture is bound to a
 298          * framebuffer. Implemented in radeonsi only.
 299          */
 300         uint32_t                        framebuffers_bound;
 301 };
 302
 303 struct r600_surface {
 304         struct pipe_surface             base;
 305
 306         /* These can vary with block-compressed textures. */
 307         unsigned width0;
 308         unsigned height0;
 309
 310         bool color_initialized;
 311         bool depth_initialized;
 312
 313         /* Misc. color flags. */
 314         bool alphatest_bypass;
 315         bool export_16bpc;
 316         bool color_is_int8;
 317         bool color_is_int10;
 318         bool dcc_incompatible;
 319
 320         /* Color registers. */
 321         unsigned cb_color_info;
 322         unsigned cb_color_base;
 323         unsigned cb_color_view;
 324         unsigned cb_color_size;         /* R600 only */
 325         unsigned cb_color_dim;          /* EG only */
 326         unsigned cb_color_pitch;        /* EG and later */
 327         unsigned cb_color_slice;        /* EG and later */
 328         unsigned cb_color_attrib;       /* EG and later */
 329         unsigned cb_color_attrib2;      /* GFX9 and later */
 330         unsigned cb_dcc_control;        /* VI and later */
 331         unsigned cb_color_fmask;        /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
 332         unsigned cb_color_fmask_slice;  /* EG and later */
 333         unsigned cb_color_cmask;        /* CB_COLORn_TILE (r600 only) */
 334         unsigned cb_color_mask;         /* R600 only */
 335         unsigned spi_shader_col_format;         /* SI+, no blending, no alpha-to-coverage. */
 336         unsigned spi_shader_col_format_alpha;   /* SI+, alpha-to-coverage */
 337         unsigned spi_shader_col_format_blend;   /* SI+, blending without alpha. */
 338         unsigned spi_shader_col_format_blend_alpha; /* SI+, blending with alpha. */
 339         struct r600_resource *cb_buffer_fmask; /* Used for FMASK relocations. R600 only */
 340         struct r600_resource *cb_buffer_cmask; /* Used for CMASK relocations. R600 only */
 341
 342         /* DB registers. */
 343         uint64_t db_depth_base;         /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
 344         uint64_t db_stencil_base;       /* EG and later */
 345         uint64_t db_htile_data_base;
 346         unsigned db_depth_info;         /* R600 only, then SI and later */
 347         unsigned db_z_info;             /* EG and later */
 348         unsigned db_z_info2;            /* GFX9+ */
 349         unsigned db_depth_view;
 350         unsigned db_depth_size;
 351         unsigned db_depth_slice;        /* EG and later */
 352         unsigned db_stencil_info;       /* EG and later */
 353         unsigned db_stencil_info2;      /* GFX9+ */
 354         unsigned db_prefetch_limit;     /* R600 only */
 355         unsigned db_htile_surface;
 356         unsigned db_preload_control;    /* EG and later */
 357 };
 358
 359 struct r600_mmio_counter {
 360         unsigned busy;
 361         unsigned idle;
 362 };
 363
 364 union r600_mmio_counters {
 365         struct {
 366                 /* For global GPU load including SDMA. */
 367                 struct r600_mmio_counter gpu;
 368
 369                 /* GRBM_STATUS */
 370                 struct r600_mmio_counter spi;
 371                 struct r600_mmio_counter gui;
 372                 struct r600_mmio_counter ta;
 373                 struct r600_mmio_counter gds;
 374                 struct r600_mmio_counter vgt;
 375                 struct r600_mmio_counter ia;
 376                 struct r600_mmio_counter sx;
 377                 struct r600_mmio_counter wd;
 378                 struct r600_mmio_counter bci;
 379                 struct r600_mmio_counter sc;
 380                 struct r600_mmio_counter pa;
 381                 struct r600_mmio_counter db;
 382                 struct r600_mmio_counter cp;
 383                 struct r600_mmio_counter cb;
 384
 385                 /* SRBM_STATUS2 */
 386                 struct r600_mmio_counter sdma;
 387
 388                 /* CP_STAT */
 389                 struct r600_mmio_counter pfp;
 390                 struct r600_mmio_counter meq;
 391                 struct r600_mmio_counter me;
 392                 struct r600_mmio_counter surf_sync;
 393                 struct r600_mmio_counter cp_dma;
 394                 struct r600_mmio_counter scratch_ram;
 395         } named;
 396         unsigned array[0];
 397 };
 398
 399 struct r600_memory_object {
 400         struct pipe_memory_object       b;
 401         struct pb_buffer                *buf;
 402         uint32_t                        stride;
 403         uint32_t                        offset;
 404 };
 405
 406 struct r600_common_screen {
 407         struct pipe_screen              b;
 408         struct radeon_winsys            *ws;
 409         enum radeon_family              family;
 410         enum chip_class                 chip_class;
 411         struct radeon_info              info;
 412         uint64_t                        debug_flags;
 413         bool                            has_cp_dma;
 414         bool                            has_streamout;
 415         bool                            has_rbplus;     /* if RB+ registers exist */
 416         bool                            rbplus_allowed; /* if RB+ is allowed */
 417
 418         struct disk_cache               *disk_shader_cache;
 419
 420         struct slab_parent_pool         pool_transfers;
 421
 422         /* Texture filter settings. */
 423         int                             force_aniso; /* -1 = disabled */
 424
 425         /* Auxiliary context. Mainly used to initialize resources.
 426          * It must be locked prior to using and flushed before unlocking. */
 427         struct pipe_context             *aux_context;
 428         mtx_t                           aux_context_lock;
 429
 430         /* This must be in the screen, because UE4 uses one context for
 431          * compilation and another one for rendering.
 432          */
 433         unsigned                        num_compilations;
 434         /* Along with ST_DEBUG=precompile, this should show if applications
 435          * are loading shaders on demand. This is a monotonic counter.
 436          */
 437         unsigned                        num_shaders_created;
 438         unsigned                        num_shader_cache_hits;
 439
 440         /* GPU load thread. */
 441         mtx_t                           gpu_load_mutex;
 442         thrd_t                          gpu_load_thread;
 443         union r600_mmio_counters        mmio_counters;
 444         volatile unsigned               gpu_load_stop_thread; /* bool */
 445
 446         char                            renderer_string[100];
 447
 448         /* Performance counters. */
 449         struct r600_perfcounters        *perfcounters;
 450
 451         /* If pipe_screen wants to recompute and re-emit the framebuffer,
 452          * sampler, and image states of all contexts, it should atomically
 453          * increment this.
 454          *
 455          * Each context will compare this with its own last known value of
 456          * the counter before drawing and re-emit the states accordingly.
 457          */
 458         unsigned                        dirty_tex_counter;
 459
 460         /* Atomically increment this counter when an existing texture's
 461          * metadata is enabled or disabled in a way that requires changing
 462          * contexts' compressed texture binding masks.
 463          */
 464         unsigned                        compressed_colortex_counter;
 465
 466         struct {
 467                 /* Context flags to set so that all writes from earlier jobs
 468                  * in the CP are seen by L2 clients.
 469                  */
 470                 unsigned cp_to_L2;
 471
 472                 /* Context flags to set so that all writes from earlier jobs
 473                  * that end in L2 are seen by CP.
 474                  */
 475                 unsigned L2_to_cp;
 476
 477                 /* Context flags to set so that all writes from earlier
 478                  * compute jobs are seen by L2 clients.
 479                  */
 480                 unsigned compute_to_L2;
 481         } barrier_flags;
 482
 483         void (*query_opaque_metadata)(struct r600_common_screen *rscreen,
 484                                       struct r600_texture *rtex,
 485                                       struct radeon_bo_metadata *md);
 486
 487         void (*apply_opaque_metadata)(struct r600_common_screen *rscreen,
 488                                     struct r600_texture *rtex,
 489                                     struct radeon_bo_metadata *md);
 490 };
 491
 492 /* This encapsulates a state or an operation which can emitted into the GPU
 493  * command stream. */
 494 struct r600_atom {
 495         void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
 496         unsigned short          id;
 497 };
 498
 499 struct r600_ring {
 500         struct radeon_winsys_cs         *cs;
 501         void (*flush)(void *ctx, unsigned flags,
 502                       struct pipe_fence_handle **fence);
 503 };
 504
 505 /* Saved CS data for debugging features. */
 506 struct radeon_saved_cs {
 507         uint32_t                        *ib;
 508         unsigned                        num_dw;
 509
 510         struct radeon_bo_list_item      *bo_list;
 511         unsigned                        bo_count;
 512 };
 513
 514 struct r600_common_context {
 515         struct pipe_context b; /* base class */
 516
 517         struct r600_common_screen       *screen;
 518         struct radeon_winsys            *ws;
 519         struct radeon_winsys_ctx        *ctx;
 520         enum radeon_family              family;
 521         enum chip_class                 chip_class;
 522         struct r600_ring                gfx;
 523         struct r600_ring                dma;
 524         struct pipe_fence_handle        *last_gfx_fence;
 525         struct pipe_fence_handle        *last_sdma_fence;
 526         struct r600_resource            *eop_bug_scratch;
 527         unsigned                        num_gfx_cs_flushes;
 528         unsigned                        initial_gfx_cs_size;
 529         unsigned                        gpu_reset_counter;
 530         unsigned                        last_dirty_tex_counter;
 531         unsigned                        last_compressed_colortex_counter;
 532         unsigned                        last_num_draw_calls;
 533
 534         struct threaded_context         *tc;
 535         struct u_suballocator           *allocator_zeroed_memory;
 536         struct slab_child_pool          pool_transfers;
 537         struct slab_child_pool          pool_transfers_unsync; /* for threaded_context */
 538
 539         /* Current unaccounted memory usage. */
 540         uint64_t                        vram;
 541         uint64_t                        gtt;
 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_L2_invalidates;
 568         unsigned                        num_L2_writebacks;
 569         unsigned                        num_resident_handles;
 570         uint64_t                        num_alloc_tex_transfer_bytes;
 571         unsigned                        last_tex_ps_draw_ratio; /* for query */
 572
 573         /* Render condition. */
 574         struct r600_atom                render_cond_atom;
 575         struct pipe_query               *render_cond;
 576         unsigned                        render_cond_mode;
 577         bool                            render_cond_invert;
 578         bool                            render_cond_force_off; /* for u_blitter */
 579
 580         /* Statistics gathering for the DCC enablement heuristic. It can't be
 581          * in r600_texture because r600_texture can be shared by multiple
 582          * contexts. This is for back buffers only. We shouldn't get too many
 583          * of those.
 584          *
 585          * X11 DRI3 rotates among a finite set of back buffers. They should
 586          * all fit in this array. If they don't, separate DCC might never be
 587          * enabled by DCC stat gathering.
 588          */
 589         struct {
 590                 struct r600_texture             *tex;
 591                 /* Query queue: 0 = usually active, 1 = waiting, 2 = readback. */
 592                 struct pipe_query               *ps_stats[3];
 593                 /* If all slots are used and another slot is needed,
 594                  * the least recently used slot is evicted based on this. */
 595                 int64_t                         last_use_timestamp;
 596                 bool                            query_active;
 597         } dcc_stats[5];
 598
 599         struct pipe_device_reset_callback device_reset_callback;
 600         struct u_log_context            *log;
 601
 602         void                            *query_result_shader;
 603
 604         /* Copy one resource to another using async DMA. */
 605         void (*dma_copy)(struct pipe_context *ctx,
 606                          struct pipe_resource *dst,
 607                          unsigned dst_level,
 608                          unsigned dst_x, unsigned dst_y, unsigned dst_z,
 609                          struct pipe_resource *src,
 610                          unsigned src_level,
 611                          const struct pipe_box *src_box);
 612
 613         void (*dma_clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
 614                                  uint64_t offset, uint64_t size, unsigned value);
 615
 616         void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
 617                              uint64_t offset, uint64_t size, unsigned value,
 618                              enum r600_coherency coher);
 619
 620         void (*blit_decompress_depth)(struct pipe_context *ctx,
 621                                       struct r600_texture *texture,
 622                                       struct r600_texture *staging,
 623                                       unsigned first_level, unsigned last_level,
 624                                       unsigned first_layer, unsigned last_layer,
 625                                       unsigned first_sample, unsigned last_sample);
 626
 627         void (*decompress_dcc)(struct pipe_context *ctx,
 628                                struct r600_texture *rtex);
 629
 630         /* Reallocate the buffer and update all resource bindings where
 631          * the buffer is bound, including all resource descriptors. */
 632         void (*invalidate_buffer)(struct pipe_context *ctx, struct pipe_resource *buf);
 633
 634         /* Update all resource bindings where the buffer is bound, including
 635          * all resource descriptors. This is invalidate_buffer without
 636          * the invalidation. */
 637         void (*rebind_buffer)(struct pipe_context *ctx, struct pipe_resource *buf,
 638                               uint64_t old_gpu_address);
 639
 640         /* Enable or disable occlusion queries. */
 641         void (*set_occlusion_query_state)(struct pipe_context *ctx,
 642                                           bool old_enable,
 643                                           bool old_perfect_enable);
 644
 645         void (*save_qbo_state)(struct pipe_context *ctx, struct r600_qbo_state *st);
 646
 647         /* This ensures there is enough space in the command stream. */
 648         void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
 649                                   bool include_draw_vbo);
 650
 651         void (*set_atom_dirty)(struct r600_common_context *ctx,
 652                                struct r600_atom *atom, bool dirty);
 653
 654         void (*check_vm_faults)(struct r600_common_context *ctx,
 655                                 struct radeon_saved_cs *saved,
 656                                 enum ring_type ring);
 657 };
 658
 659 /* r600_buffer_common.c */
 660 bool si_rings_is_buffer_referenced(struct r600_common_context *ctx,
 661                                    struct pb_buffer *buf,
 662                                    enum radeon_bo_usage usage);
 663 void *si_buffer_map_sync_with_rings(struct r600_common_context *ctx,
 664                                     struct r600_resource *resource,
 665                                     unsigned usage);
 666 void si_buffer_subdata(struct pipe_context *ctx,
 667                        struct pipe_resource *buffer,
 668                        unsigned usage, unsigned offset,
 669                        unsigned size, const void *data);
 670 void si_init_resource_fields(struct r600_common_screen *rscreen,
 671                              struct r600_resource *res,
 672                              uint64_t size, unsigned alignment);
 673 bool si_alloc_resource(struct r600_common_screen *rscreen,
 674                        struct r600_resource *res);
 675 struct pipe_resource *si_buffer_create(struct pipe_screen *screen,
 676                                        const struct pipe_resource *templ,
 677                                        unsigned alignment);
 678 struct pipe_resource *si_aligned_buffer_create(struct pipe_screen *screen,
 679                                                unsigned flags,
 680                                                unsigned usage,
 681                                                unsigned size,
 682                                                unsigned alignment);
 683 struct pipe_resource *
 684 si_buffer_from_user_memory(struct pipe_screen *screen,
 685                            const struct pipe_resource *templ,
 686                            void *user_memory);
 687 void si_invalidate_resource(struct pipe_context *ctx,
 688                             struct pipe_resource *resource);
 689 void si_replace_buffer_storage(struct pipe_context *ctx,
 690                                struct pipe_resource *dst,
 691                                struct pipe_resource *src);
 692
 693 /* r600_common_pipe.c */
 694 void si_gfx_write_event_eop(struct r600_common_context *ctx,
 695                             unsigned event, unsigned event_flags,
 696                             unsigned data_sel,
 697                             struct r600_resource *buf, uint64_t va,
 698                             uint32_t new_fence, unsigned query_type);
 699 unsigned si_gfx_write_fence_dwords(struct r600_common_screen *screen);
 700 void si_gfx_wait_fence(struct r600_common_context *ctx,
 701                        uint64_t va, uint32_t ref, uint32_t mask);
 702 bool si_common_screen_init(struct r600_common_screen *rscreen,
 703                            struct radeon_winsys *ws);
 704 void si_destroy_common_screen(struct r600_common_screen *rscreen);
 705 void si_preflush_suspend_features(struct r600_common_context *ctx);
 706 void si_postflush_resume_features(struct r600_common_context *ctx);
 707 bool si_common_context_init(struct r600_common_context *rctx,
 708                             struct r600_common_screen *rscreen,
 709                             unsigned context_flags);
 710 void si_common_context_cleanup(struct r600_common_context *rctx);
 711 bool si_can_dump_shader(struct r600_common_screen *rscreen,
 712                         unsigned processor);
 713 bool si_extra_shader_checks(struct r600_common_screen *rscreen,
 714                             unsigned processor);
 715 void si_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
 716                             uint64_t offset, uint64_t size, unsigned value);
 717 struct pipe_resource *si_resource_create_common(struct pipe_screen *screen,
 718                                                 const struct pipe_resource *templ);
 719 void si_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
 720                        struct r600_resource *dst, struct r600_resource *src);
 721 void si_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
 722                 struct radeon_saved_cs *saved, bool get_buffer_list);
 723 void si_clear_saved_cs(struct radeon_saved_cs *saved);
 724 bool si_check_device_reset(struct r600_common_context *rctx);
 725
 726 /* r600_gpu_load.c */
 727 void si_gpu_load_kill_thread(struct r600_common_screen *rscreen);
 728 uint64_t si_begin_counter(struct r600_common_screen *rscreen, unsigned type);
 729 unsigned si_end_counter(struct r600_common_screen *rscreen, unsigned type,
 730                         uint64_t begin);
 731
 732 /* r600_perfcounters.c */
 733 void si_perfcounters_destroy(struct r600_common_screen *rscreen);
 734
 735 /* r600_query.c */
 736 void si_init_screen_query_functions(struct r600_common_screen *rscreen);
 737 void si_init_query_functions(struct r600_common_context *rctx);
 738 void si_suspend_queries(struct r600_common_context *ctx);
 739 void si_resume_queries(struct r600_common_context *ctx);
 740
 741 /* r600_test_dma.c */
 742 void si_test_dma(struct r600_common_screen *rscreen);
 743
 744 /* r600_texture.c */
 745 bool si_prepare_for_dma_blit(struct r600_common_context *rctx,
 746                              struct r600_texture *rdst,
 747                              unsigned dst_level, unsigned dstx,
 748                              unsigned dsty, unsigned dstz,
 749                              struct r600_texture *rsrc,
 750                              unsigned src_level,
 751                              const struct pipe_box *src_box);
 752 void si_texture_get_fmask_info(struct r600_common_screen *rscreen,
 753                                struct r600_texture *rtex,
 754                                unsigned nr_samples,
 755                                struct r600_fmask_info *out);
 756 bool si_init_flushed_depth_texture(struct pipe_context *ctx,
 757                                    struct pipe_resource *texture,
 758                                    struct r600_texture **staging);
 759 void si_print_texture_info(struct r600_common_screen *rscreen,
 760                            struct r600_texture *rtex, struct u_log_context *log);
 761 struct pipe_resource *si_texture_create(struct pipe_screen *screen,
 762                                         const struct pipe_resource *templ);
 763 bool vi_dcc_formats_compatible(enum pipe_format format1,
 764                                enum pipe_format format2);
 765 bool vi_dcc_formats_are_incompatible(struct pipe_resource *tex,
 766                                      unsigned level,
 767                                      enum pipe_format view_format);
 768 void vi_disable_dcc_if_incompatible_format(struct r600_common_context *rctx,
 769                                            struct pipe_resource *tex,
 770                                            unsigned level,
 771                                            enum pipe_format view_format);
 772 struct pipe_surface *si_create_surface_custom(struct pipe_context *pipe,
 773                                               struct pipe_resource *texture,
 774                                               const struct pipe_surface *templ,
 775                                               unsigned width0, unsigned height0,
 776                                               unsigned width, unsigned height);
 777 unsigned si_translate_colorswap(enum pipe_format format, bool do_endian_swap);
 778 void vi_separate_dcc_start_query(struct pipe_context *ctx,
 779                                  struct r600_texture *tex);
 780 void vi_separate_dcc_stop_query(struct pipe_context *ctx,
 781                                 struct r600_texture *tex);
 782 void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx,
 783                                              struct r600_texture *tex);
 784 void vi_dcc_clear_level(struct r600_common_context *rctx,
 785                         struct r600_texture *rtex,
 786                         unsigned level, unsigned clear_value);
 787 void si_do_fast_color_clear(struct r600_common_context *rctx,
 788                             struct pipe_framebuffer_state *fb,
 789                             struct r600_atom *fb_state,
 790                             unsigned *buffers, ubyte *dirty_cbufs,
 791                             const union pipe_color_union *color);
 792 bool si_texture_disable_dcc(struct r600_common_context *rctx,
 793                             struct r600_texture *rtex);
 794 void si_init_screen_texture_functions(struct r600_common_screen *rscreen);
 795 void si_init_context_texture_functions(struct r600_common_context *rctx);
 796
 797
 798 /* Inline helpers. */
 799
 800 static inline struct r600_resource *r600_resource(struct pipe_resource *r)
 801 {
 802         return (struct r600_resource*)r;
 803 }
 804
 805 static inline void
 806 r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
 807 {
 808         pipe_resource_reference((struct pipe_resource **)ptr,
 809                                 (struct pipe_resource *)res);
 810 }
 811
 812 static inline void
 813 r600_texture_reference(struct r600_texture **ptr, struct r600_texture *res)
 814 {
 815         pipe_resource_reference((struct pipe_resource **)ptr, &res->resource.b.b);
 816 }
 817
 818 static inline void
 819 r600_context_add_resource_size(struct pipe_context *ctx, struct pipe_resource *r)
 820 {
 821         struct r600_common_context *rctx = (struct r600_common_context *)ctx;
 822         struct r600_resource *res = (struct r600_resource *)r;
 823
 824         if (res) {
 825                 /* Add memory usage for need_gfx_cs_space */
 826                 rctx->vram += res->vram_usage;
 827                 rctx->gtt += res->gart_usage;
 828         }
 829 }
 830
 831 #define     SQ_TEX_XY_FILTER_POINT                         0x00
 832 #define     SQ_TEX_XY_FILTER_BILINEAR                      0x01
 833 #define     SQ_TEX_XY_FILTER_ANISO_POINT                   0x02
 834 #define     SQ_TEX_XY_FILTER_ANISO_BILINEAR                0x03
 835
 836 static inline unsigned eg_tex_filter(unsigned filter, unsigned max_aniso)
 837 {
 838         if (filter == PIPE_TEX_FILTER_LINEAR)
 839                 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_BILINEAR
 840                                      : SQ_TEX_XY_FILTER_BILINEAR;
 841         else
 842                 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_POINT
 843                                      : SQ_TEX_XY_FILTER_POINT;
 844 }
 845
 846 static inline unsigned r600_tex_aniso_filter(unsigned filter)
 847 {
 848         if (filter < 2)
 849                 return 0;
 850         if (filter < 4)
 851                 return 1;
 852         if (filter < 8)
 853                 return 2;
 854         if (filter < 16)
 855                 return 3;
 856         return 4;
 857 }
 858
 859 static inline enum radeon_bo_priority
 860 r600_get_sampler_view_priority(struct r600_resource *res)
 861 {
 862         if (res->b.b.target == PIPE_BUFFER)
 863                 return RADEON_PRIO_SAMPLER_BUFFER;
 864
 865         if (res->b.b.nr_samples > 1)
 866                 return RADEON_PRIO_SAMPLER_TEXTURE_MSAA;
 867
 868         return RADEON_PRIO_SAMPLER_TEXTURE;
 869 }
 870
 871 static inline bool
 872 r600_can_sample_zs(struct r600_texture *tex, bool stencil_sampler)
 873 {
 874         return (stencil_sampler && tex->can_sample_s) ||
 875                (!stencil_sampler && tex->can_sample_z);
 876 }
 877
 878 static inline bool
 879 vi_dcc_enabled(struct r600_texture *tex, unsigned level)
 880 {
 881         return tex->dcc_offset && level < tex->surface.num_dcc_levels;
 882 }
 883
 884 static inline bool
 885 r600_htile_enabled(struct r600_texture *tex, unsigned level)
 886 {
 887         return tex->htile_offset && level == 0;
 888 }
 889
 890 static inline bool
 891 vi_tc_compat_htile_enabled(struct r600_texture *tex, unsigned level)
 892 {
 893         assert(!tex->tc_compatible_htile || tex->htile_offset);
 894         return tex->tc_compatible_htile && level == 0;
 895 }
 896
 897 #define COMPUTE_DBG(rscreen, fmt, args...) \
 898         do { \
 899                 if ((rscreen->b.debug_flags & DBG(COMPUTE))) fprintf(stderr, fmt, ##args); \
 900         } while (0);
 901
 902 #define R600_ERR(fmt, args...) \
 903         fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
 904
 905 static inline int S_FIXED(float value, unsigned frac_bits)
 906 {
 907         return value * (1 << frac_bits);
 908 }
 909
 910 #endif