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 #define DBG_NO_OPT_VARIANT      (1 << 17)
  83 /* gaps */
  84 #define DBG_TEST_DMA            (1 << 20)
  85 /* Bits 21-31 are reserved for the r600g driver. */
  86 /* features */
  87 #define DBG_NO_ASYNC_DMA        (1llu << 32)
  88 #define DBG_NO_HYPERZ           (1llu << 33)
  89 #define DBG_NO_DISCARD_RANGE    (1llu << 34)
  90 #define DBG_NO_2D_TILING        (1llu << 35)
  91 #define DBG_NO_TILING           (1llu << 36)
  92 #define DBG_SWITCH_ON_EOP       (1llu << 37)
  93 #define DBG_FORCE_DMA           (1llu << 38)
  94 #define DBG_PRECOMPILE          (1llu << 39)
  95 #define DBG_INFO                (1llu << 40)
  96 #define DBG_NO_WC               (1llu << 41)
  97 #define DBG_CHECK_VM            (1llu << 42)
  98 #define DBG_NO_DCC              (1llu << 43)
  99 #define DBG_NO_DCC_CLEAR        (1llu << 44)
 100 #define DBG_NO_RB_PLUS          (1llu << 45)
 101 #define DBG_SI_SCHED            (1llu << 46)
 102 #define DBG_MONOLITHIC_SHADERS  (1llu << 47)
 103 #define DBG_NO_CE               (1llu << 48)
 104 #define DBG_UNSAFE_MATH         (1llu << 49)
 105 #define DBG_NO_DCC_FB           (1llu << 50)
 106
 107 #define R600_MAP_BUFFER_ALIGNMENT 64
 108 #define R600_MAX_VIEWPORTS        16
 109
 110 #define SI_MAX_VARIABLE_THREADS_PER_BLOCK 1024
 111
 112 enum r600_coherency {
 113         R600_COHERENCY_NONE, /* no cache flushes needed */
 114         R600_COHERENCY_SHADER,
 115         R600_COHERENCY_CB_META,
 116 };
 117
 118 #ifdef PIPE_ARCH_BIG_ENDIAN
 119 #define R600_BIG_ENDIAN 1
 120 #else
 121 #define R600_BIG_ENDIAN 0
 122 #endif
 123
 124 struct r600_common_context;
 125 struct r600_perfcounters;
 126 struct tgsi_shader_info;
 127 struct r600_qbo_state;
 128
 129 struct radeon_shader_reloc {
 130         char name[32];
 131         uint64_t offset;
 132 };
 133
 134 struct radeon_shader_binary {
 135         /** Shader code */
 136         unsigned char *code;
 137         unsigned code_size;
 138
 139         /** Config/Context register state that accompanies this shader.
 140          * This is a stream of dword pairs.  First dword contains the
 141          * register address, the second dword contains the value.*/
 142         unsigned char *config;
 143         unsigned config_size;
 144
 145         /** The number of bytes of config information for each global symbol.
 146          */
 147         unsigned config_size_per_symbol;
 148
 149         /** Constant data accessed by the shader.  This will be uploaded
 150          * into a constant buffer. */
 151         unsigned char *rodata;
 152         unsigned rodata_size;
 153
 154         /** List of symbol offsets for the shader */
 155         uint64_t *global_symbol_offsets;
 156         unsigned global_symbol_count;
 157
 158         struct radeon_shader_reloc *relocs;
 159         unsigned reloc_count;
 160
 161         /** Disassembled shader in a string. */
 162         char *disasm_string;
 163         char *llvm_ir_string;
 164 };
 165
 166 void radeon_shader_binary_init(struct radeon_shader_binary *b);
 167 void radeon_shader_binary_clean(struct radeon_shader_binary *b);
 168
 169 /* Only 32-bit buffer allocations are supported, gallium doesn't support more
 170  * at the moment.
 171  */
 172 struct r600_resource {
 173         struct u_resource               b;
 174
 175         /* Winsys objects. */
 176         struct pb_buffer                *buf;
 177         uint64_t                        gpu_address;
 178         /* Memory usage if the buffer placement is optimal. */
 179         uint64_t                        vram_usage;
 180         uint64_t                        gart_usage;
 181
 182         /* Resource properties. */
 183         uint64_t                        bo_size;
 184         unsigned                        bo_alignment;
 185         enum radeon_bo_domain           domains;
 186         enum radeon_bo_flag             flags;
 187         unsigned                        bind_history;
 188
 189         /* The buffer range which is initialized (with a write transfer,
 190          * streamout, DMA, or as a random access target). The rest of
 191          * the buffer is considered invalid and can be mapped unsynchronized.
 192          *
 193          * This allows unsychronized mapping of a buffer range which hasn't
 194          * been used yet. It's for applications which forget to use
 195          * the unsynchronized map flag and expect the driver to figure it out.
 196          */
 197         struct util_range               valid_buffer_range;
 198
 199         /* For buffers only. This indicates that a write operation has been
 200          * performed by TC L2, but the cache hasn't been flushed.
 201          * Any hw block which doesn't use or bypasses TC L2 should check this
 202          * flag and flush the cache before using the buffer.
 203          *
 204          * For example, TC L2 must be flushed if a buffer which has been
 205          * modified by a shader store instruction is about to be used as
 206          * an index buffer. The reason is that VGT DMA index fetching doesn't
 207          * use TC L2.
 208          */
 209         bool                            TC_L2_dirty;
 210
 211         /* Whether the resource has been exported via resource_get_handle. */
 212         bool                            is_shared;
 213         unsigned                        external_usage; /* PIPE_HANDLE_USAGE_* */
 214 };
 215
 216 struct r600_transfer {
 217         struct pipe_transfer            transfer;
 218         struct r600_resource            *staging;
 219         unsigned                        offset;
 220 };
 221
 222 struct r600_fmask_info {
 223         uint64_t offset;
 224         uint64_t size;
 225         unsigned alignment;
 226         unsigned pitch_in_pixels;
 227         unsigned bank_height;
 228         unsigned slice_tile_max;
 229         unsigned tile_mode_index;
 230 };
 231
 232 struct r600_cmask_info {
 233         uint64_t offset;
 234         uint64_t size;
 235         unsigned alignment;
 236         unsigned slice_tile_max;
 237         unsigned base_address_reg;
 238 };
 239
 240 struct r600_texture {
 241         struct r600_resource            resource;
 242
 243         uint64_t                        size;
 244         unsigned                        num_level0_transfers;
 245         enum pipe_format                db_render_format;
 246         bool                            is_depth;
 247         bool                            db_compatible;
 248         bool                            can_sample_z;
 249         bool                            can_sample_s;
 250         unsigned                        dirty_level_mask; /* each bit says if that mipmap is compressed */
 251         unsigned                        stencil_dirty_level_mask; /* each bit says if that mipmap is compressed */
 252         struct r600_texture             *flushed_depth_texture;
 253         struct radeon_surf              surface;
 254
 255         /* Colorbuffer compression and fast clear. */
 256         struct r600_fmask_info          fmask;
 257         struct r600_cmask_info          cmask;
 258         struct r600_resource            *cmask_buffer;
 259         uint64_t                        dcc_offset; /* 0 = disabled */
 260         unsigned                        cb_color_info; /* fast clear enable bit */
 261         unsigned                        color_clear_value[2];
 262         unsigned                        last_msaa_resolve_target_micro_mode;
 263
 264         /* Depth buffer compression and fast clear. */
 265         struct r600_resource            *htile_buffer;
 266         bool                            tc_compatible_htile;
 267         bool                            depth_cleared; /* if it was cleared at least once */
 268         float                           depth_clear_value;
 269         bool                            stencil_cleared; /* if it was cleared at least once */
 270         uint8_t                         stencil_clear_value;
 271
 272         bool                            non_disp_tiling; /* R600-Cayman only */
 273
 274         /* Whether the texture is a displayable back buffer and needs DCC
 275          * decompression, which is expensive. Therefore, it's enabled only
 276          * if statistics suggest that it will pay off and it's allocated
 277          * separately. It can't be bound as a sampler by apps. Limited to
 278          * target == 2D and last_level == 0. If enabled, dcc_offset contains
 279          * the absolute GPUVM address, not the relative one.
 280          */
 281         struct r600_resource            *dcc_separate_buffer;
 282         /* When DCC is temporarily disabled, the separate buffer is here. */
 283         struct r600_resource            *last_dcc_separate_buffer;
 284         /* We need to track DCC dirtiness, because st/dri usually calls
 285          * flush_resource twice per frame (not a bug) and we don't wanna
 286          * decompress DCC twice. Also, the dirty tracking must be done even
 287          * if DCC isn't used, because it's required by the DCC usage analysis
 288          * for a possible future enablement.
 289          */
 290         bool                            separate_dcc_dirty;
 291         /* Statistics gathering for the DCC enablement heuristic. */
 292         bool                            dcc_gather_statistics;
 293         /* Estimate of how much this color buffer is written to in units of
 294          * full-screen draws: ps_invocations / (width * height)
 295          * Shader kills, late Z, and blending with trivial discards make it
 296          * inaccurate (we need to count CB updates, not PS invocations).
 297          */
 298         unsigned                        ps_draw_ratio;
 299         /* The number of clears since the last DCC usage analysis. */
 300         unsigned                        num_slow_clears;
 301
 302         /* Counter that should be non-zero if the texture is bound to a
 303          * framebuffer. Implemented in radeonsi only.
 304          */
 305         uint32_t                        framebuffers_bound;
 306 };
 307
 308 struct r600_surface {
 309         struct pipe_surface             base;
 310
 311         bool color_initialized;
 312         bool depth_initialized;
 313
 314         /* Misc. color flags. */
 315         bool alphatest_bypass;
 316         bool export_16bpc;
 317         bool color_is_int8;
 318
 319         /* Color registers. */
 320         unsigned cb_color_info;
 321         unsigned cb_color_base;
 322         unsigned cb_color_view;
 323         unsigned cb_color_size;         /* R600 only */
 324         unsigned cb_color_dim;          /* EG only */
 325         unsigned cb_color_pitch;        /* EG and later */
 326         unsigned cb_color_slice;        /* EG and later */
 327         unsigned cb_color_attrib;       /* EG and later */
 328         unsigned cb_dcc_control;        /* VI and later */
 329         unsigned cb_color_fmask;        /* CB_COLORn_FMASK (EG and later) or CB_COLORn_FRAG (r600) */
 330         unsigned cb_color_fmask_slice;  /* EG and later */
 331         unsigned cb_color_cmask;        /* CB_COLORn_TILE (r600 only) */
 332         unsigned cb_color_mask;         /* R600 only */
 333         unsigned spi_shader_col_format;         /* SI+, no blending, no alpha-to-coverage. */
 334         unsigned spi_shader_col_format_alpha;   /* SI+, alpha-to-coverage */
 335         unsigned spi_shader_col_format_blend;   /* SI+, blending without alpha. */
 336         unsigned spi_shader_col_format_blend_alpha; /* SI+, blending with alpha. */
 337         struct r600_resource *cb_buffer_fmask; /* Used for FMASK relocations. R600 only */
 338         struct r600_resource *cb_buffer_cmask; /* Used for CMASK relocations. R600 only */
 339
 340         /* DB registers. */
 341         unsigned db_depth_info;         /* R600 only, then SI and later */
 342         unsigned db_z_info;             /* EG and later */
 343         unsigned db_depth_base;         /* DB_Z_READ/WRITE_BASE (EG and later) or DB_DEPTH_BASE (r600) */
 344         unsigned db_depth_view;
 345         unsigned db_depth_size;
 346         unsigned db_depth_slice;        /* EG and later */
 347         unsigned db_stencil_base;       /* EG and later */
 348         unsigned db_stencil_info;       /* EG and later */
 349         unsigned db_prefetch_limit;     /* R600 only */
 350         unsigned db_htile_surface;
 351         unsigned db_htile_data_base;
 352         unsigned db_preload_control;    /* EG and later */
 353 };
 354
 355 struct r600_common_screen {
 356         struct pipe_screen              b;
 357         struct radeon_winsys            *ws;
 358         enum radeon_family              family;
 359         enum chip_class                 chip_class;
 360         struct radeon_info              info;
 361         uint64_t                        debug_flags;
 362         bool                            has_cp_dma;
 363         bool                            has_streamout;
 364
 365         struct slab_parent_pool         pool_transfers;
 366
 367         /* Texture filter settings. */
 368         int                             force_aniso; /* -1 = disabled */
 369
 370         /* Auxiliary context. Mainly used to initialize resources.
 371          * It must be locked prior to using and flushed before unlocking. */
 372         struct pipe_context             *aux_context;
 373         pipe_mutex                      aux_context_lock;
 374
 375         /* This must be in the screen, because UE4 uses one context for
 376          * compilation and another one for rendering.
 377          */
 378         unsigned                        num_compilations;
 379         /* Along with ST_DEBUG=precompile, this should show if applications
 380          * are loading shaders on demand. This is a monotonic counter.
 381          */
 382         unsigned                        num_shaders_created;
 383         unsigned                        num_shader_cache_hits;
 384
 385         /* GPU load thread. */
 386         pipe_mutex                      gpu_load_mutex;
 387         pipe_thread                     gpu_load_thread;
 388         unsigned                        gpu_load_counter_busy;
 389         unsigned                        gpu_load_counter_idle;
 390         volatile unsigned               gpu_load_stop_thread; /* bool */
 391
 392         char                            renderer_string[100];
 393
 394         /* Performance counters. */
 395         struct r600_perfcounters        *perfcounters;
 396
 397         /* If pipe_screen wants to re-emit the framebuffer state of all
 398          * contexts, it should atomically increment this. Each context will
 399          * compare this with its own last known value of the counter before
 400          * drawing and re-emit the framebuffer state accordingly.
 401          */
 402         unsigned                        dirty_fb_counter;
 403
 404         /* Atomically increment this counter when an existing texture's
 405          * metadata is enabled or disabled in a way that requires changing
 406          * contexts' compressed texture binding masks.
 407          */
 408         unsigned                        compressed_colortex_counter;
 409
 410         /* Atomically increment this counter when an existing texture's
 411          * backing buffer or tile mode parameters have changed that requires
 412          * recomputation of shader descriptors.
 413          */
 414         unsigned                        dirty_tex_descriptor_counter;
 415
 416         struct {
 417                 /* Context flags to set so that all writes from earlier jobs
 418                  * in the CP are seen by L2 clients.
 419                  */
 420                 unsigned cp_to_L2;
 421
 422                 /* Context flags to set so that all writes from earlier
 423                  * compute jobs are seen by L2 clients.
 424                  */
 425                 unsigned compute_to_L2;
 426         } barrier_flags;
 427
 428         void (*query_opaque_metadata)(struct r600_common_screen *rscreen,
 429                                       struct r600_texture *rtex,
 430                                       struct radeon_bo_metadata *md);
 431
 432         void (*apply_opaque_metadata)(struct r600_common_screen *rscreen,
 433                                     struct r600_texture *rtex,
 434                                     struct radeon_bo_metadata *md);
 435 };
 436
 437 /* This encapsulates a state or an operation which can emitted into the GPU
 438  * command stream. */
 439 struct r600_atom {
 440         void (*emit)(struct r600_common_context *ctx, struct r600_atom *state);
 441         unsigned                num_dw;
 442         unsigned short          id;
 443 };
 444
 445 struct r600_so_target {
 446         struct pipe_stream_output_target b;
 447
 448         /* The buffer where BUFFER_FILLED_SIZE is stored. */
 449         struct r600_resource    *buf_filled_size;
 450         unsigned                buf_filled_size_offset;
 451         bool                    buf_filled_size_valid;
 452
 453         unsigned                stride_in_dw;
 454 };
 455
 456 struct r600_streamout {
 457         struct r600_atom                begin_atom;
 458         bool                            begin_emitted;
 459         unsigned                        num_dw_for_end;
 460
 461         unsigned                        enabled_mask;
 462         unsigned                        num_targets;
 463         struct r600_so_target           *targets[PIPE_MAX_SO_BUFFERS];
 464
 465         unsigned                        append_bitmask;
 466         bool                            suspended;
 467
 468         /* External state which comes from the vertex shader,
 469          * it must be set explicitly when binding a shader. */
 470         unsigned                        *stride_in_dw;
 471         unsigned                        enabled_stream_buffers_mask; /* stream0 buffers0-3 in 4 LSB */
 472
 473         /* The state of VGT_STRMOUT_BUFFER_(CONFIG|EN). */
 474         unsigned                        hw_enabled_mask;
 475
 476         /* The state of VGT_STRMOUT_(CONFIG|EN). */
 477         struct r600_atom                enable_atom;
 478         bool                            streamout_enabled;
 479         bool                            prims_gen_query_enabled;
 480         int                             num_prims_gen_queries;
 481 };
 482
 483 struct r600_signed_scissor {
 484         int minx;
 485         int miny;
 486         int maxx;
 487         int maxy;
 488 };
 489
 490 struct r600_scissors {
 491         struct r600_atom                atom;
 492         unsigned                        dirty_mask;
 493         struct pipe_scissor_state       states[R600_MAX_VIEWPORTS];
 494 };
 495
 496 struct r600_viewports {
 497         struct r600_atom                atom;
 498         unsigned                        dirty_mask;
 499         unsigned                        depth_range_dirty_mask;
 500         struct pipe_viewport_state      states[R600_MAX_VIEWPORTS];
 501         struct r600_signed_scissor      as_scissor[R600_MAX_VIEWPORTS];
 502 };
 503
 504 struct r600_ring {
 505         struct radeon_winsys_cs         *cs;
 506         void (*flush)(void *ctx, unsigned flags,
 507                       struct pipe_fence_handle **fence);
 508 };
 509
 510 /* Saved CS data for debugging features. */
 511 struct radeon_saved_cs {
 512         uint32_t                        *ib;
 513         unsigned                        num_dw;
 514
 515         struct radeon_bo_list_item      *bo_list;
 516         unsigned                        bo_count;
 517 };
 518
 519 struct r600_common_context {
 520         struct pipe_context b; /* base class */
 521
 522         struct r600_common_screen       *screen;
 523         struct radeon_winsys            *ws;
 524         struct radeon_winsys_ctx        *ctx;
 525         enum radeon_family              family;
 526         enum chip_class                 chip_class;
 527         struct r600_ring                gfx;
 528         struct r600_ring                dma;
 529         struct pipe_fence_handle        *last_gfx_fence;
 530         struct pipe_fence_handle        *last_sdma_fence;
 531         unsigned                        num_gfx_cs_flushes;
 532         unsigned                        initial_gfx_cs_size;
 533         unsigned                        gpu_reset_counter;
 534         unsigned                        last_dirty_fb_counter;
 535         unsigned                        last_compressed_colortex_counter;
 536         unsigned                        last_dirty_tex_descriptor_counter;
 537
 538         struct u_upload_mgr             *uploader;
 539         struct u_suballocator           *allocator_zeroed_memory;
 540         struct slab_child_pool          pool_transfers;
 541
 542         /* Current unaccounted memory usage. */
 543         uint64_t                        vram;
 544         uint64_t                        gtt;
 545
 546         /* States. */
 547         struct r600_streamout           streamout;
 548         struct r600_scissors            scissors;
 549         struct r600_viewports           viewports;
 550         bool                            scissor_enabled;
 551         bool                            clip_halfz;
 552         bool                            vs_writes_viewport_index;
 553         bool                            vs_disables_clipping_viewport;
 554
 555         /* Additional context states. */
 556         unsigned flags; /* flush flags */
 557
 558         /* Queries. */
 559         /* Maintain the list of active queries for pausing between IBs. */
 560         int                             num_occlusion_queries;
 561         int                             num_perfect_occlusion_queries;
 562         struct list_head                active_queries;
 563         unsigned                        num_cs_dw_queries_suspend;
 564         /* Additional hardware info. */
 565         unsigned                        backend_mask;
 566         unsigned                        max_db; /* for OQ */
 567         /* Misc stats. */
 568         unsigned                        num_draw_calls;
 569         unsigned                        num_spill_draw_calls;
 570         unsigned                        num_compute_calls;
 571         unsigned                        num_spill_compute_calls;
 572         unsigned                        num_dma_calls;
 573         unsigned                        num_cp_dma_calls;
 574         unsigned                        num_vs_flushes;
 575         unsigned                        num_ps_flushes;
 576         unsigned                        num_cs_flushes;
 577         uint64_t                        num_alloc_tex_transfer_bytes;
 578         unsigned                        last_tex_ps_draw_ratio; /* for query */
 579
 580         /* Render condition. */
 581         struct r600_atom                render_cond_atom;
 582         struct pipe_query               *render_cond;
 583         unsigned                        render_cond_mode;
 584         bool                            render_cond_invert;
 585         bool                            render_cond_force_off; /* for u_blitter */
 586
 587         /* MSAA sample locations.
 588          * The first index is the sample index.
 589          * The second index is the coordinate: X, Y. */
 590         float                           sample_locations_1x[1][2];
 591         float                           sample_locations_2x[2][2];
 592         float                           sample_locations_4x[4][2];
 593         float                           sample_locations_8x[8][2];
 594         float                           sample_locations_16x[16][2];
 595
 596         /* Statistics gathering for the DCC enablement heuristic. It can't be
 597          * in r600_texture because r600_texture can be shared by multiple
 598          * contexts. This is for back buffers only. We shouldn't get too many
 599          * of those.
 600          *
 601          * X11 DRI3 rotates among a finite set of back buffers. They should
 602          * all fit in this array. If they don't, separate DCC might never be
 603          * enabled by DCC stat gathering.
 604          */
 605         struct {
 606                 struct r600_texture             *tex;
 607                 /* Query queue: 0 = usually active, 1 = waiting, 2 = readback. */
 608                 struct pipe_query               *ps_stats[3];
 609                 /* If all slots are used and another slot is needed,
 610                  * the least recently used slot is evicted based on this. */
 611                 int64_t                         last_use_timestamp;
 612                 bool                            query_active;
 613         } dcc_stats[5];
 614
 615         struct pipe_debug_callback      debug;
 616         struct pipe_device_reset_callback device_reset_callback;
 617
 618         void                            *query_result_shader;
 619
 620         /* Copy one resource to another using async DMA. */
 621         void (*dma_copy)(struct pipe_context *ctx,
 622                          struct pipe_resource *dst,
 623                          unsigned dst_level,
 624                          unsigned dst_x, unsigned dst_y, unsigned dst_z,
 625                          struct pipe_resource *src,
 626                          unsigned src_level,
 627                          const struct pipe_box *src_box);
 628
 629         void (*clear_buffer)(struct pipe_context *ctx, struct pipe_resource *dst,
 630                              uint64_t offset, uint64_t size, unsigned value,
 631                              enum r600_coherency coher);
 632
 633         void (*blit_decompress_depth)(struct pipe_context *ctx,
 634                                       struct r600_texture *texture,
 635                                       struct r600_texture *staging,
 636                                       unsigned first_level, unsigned last_level,
 637                                       unsigned first_layer, unsigned last_layer,
 638                                       unsigned first_sample, unsigned last_sample);
 639
 640         void (*decompress_dcc)(struct pipe_context *ctx,
 641                                struct r600_texture *rtex);
 642
 643         /* Reallocate the buffer and update all resource bindings where
 644          * the buffer is bound, including all resource descriptors. */
 645         void (*invalidate_buffer)(struct pipe_context *ctx, struct pipe_resource *buf);
 646
 647         /* Enable or disable occlusion queries. */
 648         void (*set_occlusion_query_state)(struct pipe_context *ctx, bool enable);
 649
 650         void (*save_qbo_state)(struct pipe_context *ctx, struct r600_qbo_state *st);
 651
 652         /* This ensures there is enough space in the command stream. */
 653         void (*need_gfx_cs_space)(struct pipe_context *ctx, unsigned num_dw,
 654                                   bool include_draw_vbo);
 655
 656         void (*set_atom_dirty)(struct r600_common_context *ctx,
 657                                struct r600_atom *atom, bool dirty);
 658
 659         void (*check_vm_faults)(struct r600_common_context *ctx,
 660                                 struct radeon_saved_cs *saved,
 661                                 enum ring_type ring);
 662 };
 663
 664 /* r600_buffer.c */
 665 bool r600_rings_is_buffer_referenced(struct r600_common_context *ctx,
 666                                      struct pb_buffer *buf,
 667                                      enum radeon_bo_usage usage);
 668 void *r600_buffer_map_sync_with_rings(struct r600_common_context *ctx,
 669                                       struct r600_resource *resource,
 670                                       unsigned usage);
 671 void r600_buffer_subdata(struct pipe_context *ctx,
 672                          struct pipe_resource *buffer,
 673                          unsigned usage, unsigned offset,
 674                          unsigned size, const void *data);
 675 void r600_init_resource_fields(struct r600_common_screen *rscreen,
 676                                struct r600_resource *res,
 677                                uint64_t size, unsigned alignment);
 678 bool r600_alloc_resource(struct r600_common_screen *rscreen,
 679                          struct r600_resource *res);
 680 struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
 681                                          const struct pipe_resource *templ,
 682                                          unsigned alignment);
 683 struct pipe_resource * r600_aligned_buffer_create(struct pipe_screen *screen,
 684                                                   unsigned bind,
 685                                                   unsigned usage,
 686                                                   unsigned size,
 687                                                   unsigned alignment);
 688 struct pipe_resource *
 689 r600_buffer_from_user_memory(struct pipe_screen *screen,
 690                              const struct pipe_resource *templ,
 691                              void *user_memory);
 692 void
 693 r600_invalidate_resource(struct pipe_context *ctx,
 694                          struct pipe_resource *resource);
 695
 696 /* r600_common_pipe.c */
 697 void r600_gfx_write_event_eop(struct r600_common_context *ctx,
 698                               unsigned event, unsigned event_flags,
 699                               unsigned data_sel,
 700                               struct r600_resource *buf, uint64_t va,
 701                               uint32_t old_fence, uint32_t new_fence);
 702 unsigned r600_gfx_write_fence_dwords(struct r600_common_screen *screen);
 703 void r600_gfx_wait_fence(struct r600_common_context *ctx,
 704                          uint64_t va, uint32_t ref, uint32_t mask);
 705 void r600_draw_rectangle(struct blitter_context *blitter,
 706                          int x1, int y1, int x2, int y2, float depth,
 707                          enum blitter_attrib_type type,
 708                          const union pipe_color_union *attrib);
 709 bool r600_common_screen_init(struct r600_common_screen *rscreen,
 710                              struct radeon_winsys *ws);
 711 void r600_destroy_common_screen(struct r600_common_screen *rscreen);
 712 void r600_preflush_suspend_features(struct r600_common_context *ctx);
 713 void r600_postflush_resume_features(struct r600_common_context *ctx);
 714 bool r600_common_context_init(struct r600_common_context *rctx,
 715                               struct r600_common_screen *rscreen,
 716                               unsigned context_flags);
 717 void r600_common_context_cleanup(struct r600_common_context *rctx);
 718 bool r600_can_dump_shader(struct r600_common_screen *rscreen,
 719                           unsigned processor);
 720 bool r600_extra_shader_checks(struct r600_common_screen *rscreen,
 721                               unsigned processor);
 722 void r600_screen_clear_buffer(struct r600_common_screen *rscreen, struct pipe_resource *dst,
 723                               uint64_t offset, uint64_t size, unsigned value,
 724                               enum r600_coherency coher);
 725 struct pipe_resource *r600_resource_create_common(struct pipe_screen *screen,
 726                                                   const struct pipe_resource *templ);
 727 const char *r600_get_llvm_processor_name(enum radeon_family family);
 728 void r600_need_dma_space(struct r600_common_context *ctx, unsigned num_dw,
 729                          struct r600_resource *dst, struct r600_resource *src);
 730 void r600_dma_emit_wait_idle(struct r600_common_context *rctx);
 731 void radeon_save_cs(struct radeon_winsys *ws, struct radeon_winsys_cs *cs,
 732                     struct radeon_saved_cs *saved);
 733 void radeon_clear_saved_cs(struct radeon_saved_cs *saved);
 734 bool r600_check_device_reset(struct r600_common_context *rctx);
 735
 736 /* r600_gpu_load.c */
 737 void r600_gpu_load_kill_thread(struct r600_common_screen *rscreen);
 738 uint64_t r600_gpu_load_begin(struct r600_common_screen *rscreen);
 739 unsigned r600_gpu_load_end(struct r600_common_screen *rscreen, uint64_t begin);
 740
 741 /* r600_perfcounters.c */
 742 void r600_perfcounters_destroy(struct r600_common_screen *rscreen);
 743
 744 /* r600_query.c */
 745 void r600_init_screen_query_functions(struct r600_common_screen *rscreen);
 746 void r600_query_init(struct r600_common_context *rctx);
 747 void r600_suspend_queries(struct r600_common_context *ctx);
 748 void r600_resume_queries(struct r600_common_context *ctx);
 749 void r600_query_init_backend_mask(struct r600_common_context *ctx);
 750
 751 /* r600_streamout.c */
 752 void r600_streamout_buffers_dirty(struct r600_common_context *rctx);
 753 void r600_set_streamout_targets(struct pipe_context *ctx,
 754                                 unsigned num_targets,
 755                                 struct pipe_stream_output_target **targets,
 756                                 const unsigned *offset);
 757 void r600_emit_streamout_end(struct r600_common_context *rctx);
 758 void r600_update_prims_generated_query_state(struct r600_common_context *rctx,
 759                                              unsigned type, int diff);
 760 void r600_streamout_init(struct r600_common_context *rctx);
 761
 762 /* r600_test_dma.c */
 763 void r600_test_dma(struct r600_common_screen *rscreen);
 764
 765 /* r600_texture.c */
 766 bool r600_prepare_for_dma_blit(struct r600_common_context *rctx,
 767                                 struct r600_texture *rdst,
 768                                 unsigned dst_level, unsigned dstx,
 769                                 unsigned dsty, unsigned dstz,
 770                                 struct r600_texture *rsrc,
 771                                 unsigned src_level,
 772                                 const struct pipe_box *src_box);
 773 void r600_texture_get_fmask_info(struct r600_common_screen *rscreen,
 774                                  struct r600_texture *rtex,
 775                                  unsigned nr_samples,
 776                                  struct r600_fmask_info *out);
 777 void r600_texture_get_cmask_info(struct r600_common_screen *rscreen,
 778                                  struct r600_texture *rtex,
 779                                  struct r600_cmask_info *out);
 780 bool r600_init_flushed_depth_texture(struct pipe_context *ctx,
 781                                      struct pipe_resource *texture,
 782                                      struct r600_texture **staging);
 783 void r600_print_texture_info(struct r600_texture *rtex, FILE *f);
 784 struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 785                                         const struct pipe_resource *templ);
 786 bool vi_dcc_formats_compatible(enum pipe_format format1,
 787                                enum pipe_format format2);
 788 void vi_dcc_disable_if_incompatible_format(struct r600_common_context *rctx,
 789                                            struct pipe_resource *tex,
 790                                            unsigned level,
 791                                            enum pipe_format view_format);
 792 struct pipe_surface *r600_create_surface_custom(struct pipe_context *pipe,
 793                                                 struct pipe_resource *texture,
 794                                                 const struct pipe_surface *templ,
 795                                                 unsigned width, unsigned height);
 796 unsigned r600_translate_colorswap(enum pipe_format format, bool do_endian_swap);
 797 void vi_separate_dcc_start_query(struct pipe_context *ctx,
 798                                  struct r600_texture *tex);
 799 void vi_separate_dcc_stop_query(struct pipe_context *ctx,
 800                                 struct r600_texture *tex);
 801 void vi_separate_dcc_process_and_reset_stats(struct pipe_context *ctx,
 802                                              struct r600_texture *tex);
 803 void vi_dcc_clear_level(struct r600_common_context *rctx,
 804                         struct r600_texture *rtex,
 805                         unsigned level, unsigned clear_value);
 806 void evergreen_do_fast_color_clear(struct r600_common_context *rctx,
 807                                    struct pipe_framebuffer_state *fb,
 808                                    struct r600_atom *fb_state,
 809                                    unsigned *buffers, unsigned *dirty_cbufs,
 810                                    const union pipe_color_union *color);
 811 bool r600_texture_disable_dcc(struct r600_common_context *rctx,
 812                               struct r600_texture *rtex);
 813 void r600_init_screen_texture_functions(struct r600_common_screen *rscreen);
 814 void r600_init_context_texture_functions(struct r600_common_context *rctx);
 815
 816 /* r600_viewport.c */
 817 void evergreen_apply_scissor_bug_workaround(struct r600_common_context *rctx,
 818                                             struct pipe_scissor_state *scissor);
 819 void r600_viewport_set_rast_deps(struct r600_common_context *rctx,
 820                                  bool scissor_enable, bool clip_halfz);
 821 void r600_update_vs_writes_viewport_index(struct r600_common_context *rctx,
 822                                           struct tgsi_shader_info *info);
 823 void r600_init_viewport_functions(struct r600_common_context *rctx);
 824
 825 /* cayman_msaa.c */
 826 extern const uint32_t eg_sample_locs_2x[4];
 827 extern const unsigned eg_max_dist_2x;
 828 extern const uint32_t eg_sample_locs_4x[4];
 829 extern const unsigned eg_max_dist_4x;
 830 void cayman_get_sample_position(struct pipe_context *ctx, unsigned sample_count,
 831                                 unsigned sample_index, float *out_value);
 832 void cayman_init_msaa(struct pipe_context *ctx);
 833 void cayman_emit_msaa_sample_locs(struct radeon_winsys_cs *cs, int nr_samples);
 834 void cayman_emit_msaa_config(struct radeon_winsys_cs *cs, int nr_samples,
 835                              int ps_iter_samples, int overrast_samples,
 836                              unsigned sc_mode_cntl_1);
 837
 838
 839 /* Inline helpers. */
 840
 841 static inline struct r600_resource *r600_resource(struct pipe_resource *r)
 842 {
 843         return (struct r600_resource*)r;
 844 }
 845
 846 static inline void
 847 r600_resource_reference(struct r600_resource **ptr, struct r600_resource *res)
 848 {
 849         pipe_resource_reference((struct pipe_resource **)ptr,
 850                                 (struct pipe_resource *)res);
 851 }
 852
 853 static inline void
 854 r600_texture_reference(struct r600_texture **ptr, struct r600_texture *res)
 855 {
 856         pipe_resource_reference((struct pipe_resource **)ptr, &res->resource.b.b);
 857 }
 858
 859 static inline void
 860 r600_context_add_resource_size(struct pipe_context *ctx, struct pipe_resource *r)
 861 {
 862         struct r600_common_context *rctx = (struct r600_common_context *)ctx;
 863         struct r600_resource *res = (struct r600_resource *)r;
 864
 865         if (res) {
 866                 /* Add memory usage for need_gfx_cs_space */
 867                 rctx->vram += res->vram_usage;
 868                 rctx->gtt += res->gart_usage;
 869         }
 870 }
 871
 872 static inline bool r600_get_strmout_en(struct r600_common_context *rctx)
 873 {
 874         return rctx->streamout.streamout_enabled ||
 875                rctx->streamout.prims_gen_query_enabled;
 876 }
 877
 878 #define     SQ_TEX_XY_FILTER_POINT                         0x00
 879 #define     SQ_TEX_XY_FILTER_BILINEAR                      0x01
 880 #define     SQ_TEX_XY_FILTER_ANISO_POINT                   0x02
 881 #define     SQ_TEX_XY_FILTER_ANISO_BILINEAR                0x03
 882
 883 static inline unsigned eg_tex_filter(unsigned filter, unsigned max_aniso)
 884 {
 885         if (filter == PIPE_TEX_FILTER_LINEAR)
 886                 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_BILINEAR
 887                                      : SQ_TEX_XY_FILTER_BILINEAR;
 888         else
 889                 return max_aniso > 1 ? SQ_TEX_XY_FILTER_ANISO_POINT
 890                                      : SQ_TEX_XY_FILTER_POINT;
 891 }
 892
 893 static inline unsigned r600_tex_aniso_filter(unsigned filter)
 894 {
 895         if (filter < 2)
 896                 return 0;
 897         if (filter < 4)
 898                 return 1;
 899         if (filter < 8)
 900                 return 2;
 901         if (filter < 16)
 902                 return 3;
 903         return 4;
 904 }
 905
 906 static inline unsigned r600_wavefront_size(enum radeon_family family)
 907 {
 908         switch (family) {
 909         case CHIP_RV610:
 910         case CHIP_RS780:
 911         case CHIP_RV620:
 912         case CHIP_RS880:
 913                 return 16;
 914         case CHIP_RV630:
 915         case CHIP_RV635:
 916         case CHIP_RV730:
 917         case CHIP_RV710:
 918         case CHIP_PALM:
 919         case CHIP_CEDAR:
 920                 return 32;
 921         default:
 922                 return 64;
 923         }
 924 }
 925
 926 static inline enum radeon_bo_priority
 927 r600_get_sampler_view_priority(struct r600_resource *res)
 928 {
 929         if (res->b.b.target == PIPE_BUFFER)
 930                 return RADEON_PRIO_SAMPLER_BUFFER;
 931
 932         if (res->b.b.nr_samples > 1)
 933                 return RADEON_PRIO_SAMPLER_TEXTURE_MSAA;
 934
 935         return RADEON_PRIO_SAMPLER_TEXTURE;
 936 }
 937
 938 static inline bool
 939 r600_can_sample_zs(struct r600_texture *tex, bool stencil_sampler)
 940 {
 941         return (stencil_sampler && tex->can_sample_s) ||
 942                (!stencil_sampler && tex->can_sample_z);
 943 }
 944
 945 #define COMPUTE_DBG(rscreen, fmt, args...) \
 946         do { \
 947                 if ((rscreen->b.debug_flags & DBG_COMPUTE)) fprintf(stderr, fmt, ##args); \
 948         } while (0);
 949
 950 #define R600_ERR(fmt, args...) \
 951         fprintf(stderr, "EE %s:%d %s - " fmt, __FILE__, __LINE__, __func__, ##args)
 952
 953 /* For MSAA sample positions. */
 954 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y)  \
 955         (((s0x) & 0xf) | (((unsigned)(s0y) & 0xf) << 4) |                  \
 956         (((unsigned)(s1x) & 0xf) << 8) | (((unsigned)(s1y) & 0xf) << 12) |         \
 957         (((unsigned)(s2x) & 0xf) << 16) | (((unsigned)(s2y) & 0xf) << 20) |        \
 958          (((unsigned)(s3x) & 0xf) << 24) | (((unsigned)(s3y) & 0xf) << 28))
 959
 960 #endif