src/gallium/drivers/v3d/v3d_context.h

   1 /*
   2  * Copyright © 2014-2017 Broadcom
   3  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  22  * IN THE SOFTWARE.
  23  */
  24
  25 #ifndef VC5_CONTEXT_H
  26 #define VC5_CONTEXT_H
  27
  28 #ifdef V3D_VERSION
  29 #include "broadcom/common/v3d_macros.h"
  30 #endif
  31
  32 #include <stdio.h>
  33
  34 #include "pipe/p_context.h"
  35 #include "pipe/p_state.h"
  36 #include "util/bitset.h"
  37 #include "util/slab.h"
  38 #include "xf86drm.h"
  39 #include "v3d_drm.h"
  40 #include "v3d_screen.h"
  41
  42 struct v3d_job;
  43 struct v3d_bo;
  44 void v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo);
  45
  46 #include "v3d_bufmgr.h"
  47 #include "v3d_resource.h"
  48 #include "v3d_cl.h"
  49
  50 #ifdef USE_V3D_SIMULATOR
  51 #define using_v3d_simulator true
  52 #else
  53 #define using_v3d_simulator false
  54 #endif
  55
  56 #define VC5_DIRTY_BLEND         (1 <<  0)
  57 #define VC5_DIRTY_RASTERIZER    (1 <<  1)
  58 #define VC5_DIRTY_ZSA           (1 <<  2)
  59 #define VC5_DIRTY_FRAGTEX       (1 <<  3)
  60 #define VC5_DIRTY_VERTTEX       (1 <<  4)
  61
  62 #define VC5_DIRTY_BLEND_COLOR   (1 <<  7)
  63 #define VC5_DIRTY_STENCIL_REF   (1 <<  8)
  64 #define VC5_DIRTY_SAMPLE_STATE  (1 <<  9)
  65 #define VC5_DIRTY_FRAMEBUFFER   (1 << 10)
  66 #define VC5_DIRTY_STIPPLE       (1 << 11)
  67 #define VC5_DIRTY_VIEWPORT      (1 << 12)
  68 #define VC5_DIRTY_CONSTBUF      (1 << 13)
  69 #define VC5_DIRTY_VTXSTATE      (1 << 14)
  70 #define VC5_DIRTY_VTXBUF        (1 << 15)
  71 #define VC5_DIRTY_SCISSOR       (1 << 17)
  72 #define VC5_DIRTY_FLAT_SHADE_FLAGS (1 << 18)
  73 #define VC5_DIRTY_PRIM_MODE     (1 << 19)
  74 #define VC5_DIRTY_CLIP          (1 << 20)
  75 #define VC5_DIRTY_UNCOMPILED_VS (1 << 21)
  76 #define VC5_DIRTY_UNCOMPILED_FS (1 << 22)
  77 #define VC5_DIRTY_COMPILED_CS   (1 << 23)
  78 #define VC5_DIRTY_COMPILED_VS   (1 << 24)
  79 #define VC5_DIRTY_COMPILED_FS   (1 << 25)
  80 #define VC5_DIRTY_FS_INPUTS     (1 << 26)
  81 #define VC5_DIRTY_STREAMOUT     (1 << 27)
  82 #define VC5_DIRTY_OQ            (1 << 28)
  83 #define VC5_DIRTY_CENTROID_FLAGS (1 << 29)
  84 #define VC5_DIRTY_NOPERSPECTIVE_FLAGS (1 << 30)
  85 #define VC5_DIRTY_SSBO          (1 << 31)
  86
  87 #define VC5_MAX_FS_INPUTS 64
  88
  89 struct v3d_sampler_view {
  90         struct pipe_sampler_view base;
  91         uint32_t p0;
  92         uint32_t p1;
  93         /* Precomputed swizzles to pass in to the shader key. */
  94         uint8_t swizzle[4];
  95
  96         uint8_t texture_shader_state[32];
  97         /* V3D 4.x: Texture state struct. */
  98         struct v3d_bo *bo;
  99
 100         /* Actual texture to be read by this sampler view.  May be different
 101          * from base.texture in the case of having a shadow tiled copy of a
 102          * raster texture.
 103          */
 104         struct pipe_resource *texture;
 105 };
 106
 107 struct v3d_sampler_state {
 108         struct pipe_sampler_state base;
 109         uint32_t p0;
 110         uint32_t p1;
 111
 112         /* V3D 3.x: Packed texture state. */
 113         uint8_t texture_shader_state[32];
 114         /* V3D 4.x: Sampler state struct. */
 115         struct v3d_bo *bo;
 116 };
 117
 118 struct v3d_texture_stateobj {
 119         struct pipe_sampler_view *textures[PIPE_MAX_SAMPLERS];
 120         unsigned num_textures;
 121         struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
 122         unsigned num_samplers;
 123         struct v3d_cl_reloc texture_state[PIPE_MAX_SAMPLERS];
 124 };
 125
 126 struct v3d_shader_uniform_info {
 127         enum quniform_contents *contents;
 128         uint32_t *data;
 129         uint32_t count;
 130 };
 131
 132 struct v3d_uncompiled_shader {
 133         /** A name for this program, so you can track it in shader-db output. */
 134         uint32_t program_id;
 135         /** How many variants of this program were compiled, for shader-db. */
 136         uint32_t compiled_variant_count;
 137         struct pipe_shader_state base;
 138         uint32_t num_tf_outputs;
 139         struct v3d_varying_slot *tf_outputs;
 140         uint16_t tf_specs[16];
 141         uint16_t tf_specs_psiz[16];
 142         uint32_t num_tf_specs;
 143
 144         /**
 145          * Flag for if the NIR in this shader originally came from TGSI.  If
 146          * so, we need to do some fixups at compile time, due to missing
 147          * information in TGSI that exists in NIR.
 148          */
 149         bool was_tgsi;
 150 };
 151
 152 struct v3d_compiled_shader {
 153         struct pipe_resource *resource;
 154         uint32_t offset;
 155
 156         union {
 157                 struct v3d_prog_data *base;
 158                 struct v3d_vs_prog_data *vs;
 159                 struct v3d_fs_prog_data *fs;
 160         } prog_data;
 161
 162         /**
 163          * VC5_DIRTY_* flags that, when set in v3d->dirty, mean that the
 164          * uniforms have to be rewritten (and therefore the shader state
 165          * reemitted).
 166          */
 167         uint32_t uniform_dirty_bits;
 168 };
 169
 170 struct v3d_program_stateobj {
 171         struct v3d_uncompiled_shader *bind_vs, *bind_fs;
 172         struct v3d_compiled_shader *cs, *vs, *fs;
 173
 174         struct v3d_bo *spill_bo;
 175         int spill_size_per_thread;
 176 };
 177
 178 struct v3d_constbuf_stateobj {
 179         struct pipe_constant_buffer cb[PIPE_MAX_CONSTANT_BUFFERS];
 180         uint32_t enabled_mask;
 181         uint32_t dirty_mask;
 182 };
 183
 184 struct v3d_vertexbuf_stateobj {
 185         struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
 186         unsigned count;
 187         uint32_t enabled_mask;
 188         uint32_t dirty_mask;
 189 };
 190
 191 struct v3d_vertex_stateobj {
 192         struct pipe_vertex_element pipe[VC5_MAX_ATTRIBUTES];
 193         unsigned num_elements;
 194
 195         uint8_t attrs[16 * VC5_MAX_ATTRIBUTES];
 196         struct pipe_resource *defaults;
 197         uint32_t defaults_offset;
 198 };
 199
 200 struct v3d_streamout_stateobj {
 201         struct pipe_stream_output_target *targets[PIPE_MAX_SO_BUFFERS];
 202         /* Number of vertices we've written into the buffer so far. */
 203         uint32_t offsets[PIPE_MAX_SO_BUFFERS];
 204         unsigned num_targets;
 205 };
 206
 207 struct v3d_ssbo_stateobj {
 208         struct pipe_shader_buffer sb[PIPE_MAX_SHADER_BUFFERS];
 209         uint32_t enabled_mask;
 210 };
 211
 212 /* Hash table key for v3d->jobs */
 213 struct v3d_job_key {
 214         struct pipe_surface *cbufs[4];
 215         struct pipe_surface *zsbuf;
 216 };
 217
 218 enum v3d_ez_state {
 219         VC5_EZ_UNDECIDED = 0,
 220         VC5_EZ_GT_GE,
 221         VC5_EZ_LT_LE,
 222         VC5_EZ_DISABLED,
 223 };
 224
 225 /**
 226  * A complete bin/render job.
 227  *
 228  * This is all of the state necessary to submit a bin/render to the kernel.
 229  * We want to be able to have multiple in progress at a time, so that we don't
 230  * need to flush an existing CL just to switch to rendering to a new render
 231  * target (which would mean reading back from the old render target when
 232  * starting to render to it again).
 233  */
 234 struct v3d_job {
 235         struct v3d_context *v3d;
 236         struct v3d_cl bcl;
 237         struct v3d_cl rcl;
 238         struct v3d_cl indirect;
 239         struct v3d_bo *tile_alloc;
 240         struct v3d_bo *tile_state;
 241         uint32_t shader_rec_count;
 242
 243         struct drm_v3d_submit_cl submit;
 244
 245         /**
 246          * Set of all BOs referenced by the job.  This will be used for making
 247          * the list of BOs that the kernel will need to have paged in to
 248          * execute our job.
 249          */
 250         struct set *bos;
 251
 252         /** Sum of the sizes of the BOs referenced by the job. */
 253         uint32_t referenced_size;
 254
 255         struct set *write_prscs;
 256
 257         /* Size of the submit.bo_handles array. */
 258         uint32_t bo_handles_size;
 259
 260         /** @{ Surfaces to submit rendering for. */
 261         struct pipe_surface *cbufs[4];
 262         struct pipe_surface *zsbuf;
 263         /** @} */
 264         /** @{
 265          * Bounding box of the scissor across all queued drawing.
 266          *
 267          * Note that the max values are exclusive.
 268          */
 269         uint32_t draw_min_x;
 270         uint32_t draw_min_y;
 271         uint32_t draw_max_x;
 272         uint32_t draw_max_y;
 273         /** @} */
 274         /** @{
 275          * Width/height of the color framebuffer being rendered to,
 276          * for VC5_TILE_RENDERING_MODE_CONFIG.
 277         */
 278         uint32_t draw_width;
 279         uint32_t draw_height;
 280         /** @} */
 281         /** @{ Tile information, depending on MSAA and float color buffer. */
 282         uint32_t draw_tiles_x; /** @< Number of tiles wide for framebuffer. */
 283         uint32_t draw_tiles_y; /** @< Number of tiles high for framebuffer. */
 284
 285         uint32_t tile_width; /** @< Width of a tile. */
 286         uint32_t tile_height; /** @< Height of a tile. */
 287         /** maximum internal_bpp of all color render targets. */
 288         uint32_t internal_bpp;
 289
 290         /** Whether the current rendering is in a 4X MSAA tile buffer. */
 291         bool msaa;
 292         /** @} */
 293
 294         /* Bitmask of PIPE_CLEAR_* of buffers that were cleared before the
 295          * first rendering.
 296          */
 297         uint32_t clear;
 298         /* Bitmask of PIPE_CLEAR_* of buffers that have been read by a draw
 299          * call without having been cleared first.
 300          */
 301         uint32_t load;
 302         /* Bitmask of PIPE_CLEAR_* of buffers that have been rendered to
 303          * (either clears or draws) and should be stored.
 304          */
 305         uint32_t store;
 306         uint32_t clear_color[4][4];
 307         float clear_z;
 308         uint8_t clear_s;
 309
 310         /**
 311          * Set if some drawing (triangles, blits, or just a glClear()) has
 312          * been done to the FBO, meaning that we need to
 313          * DRM_IOCTL_VC5_SUBMIT_CL.
 314          */
 315         bool needs_flush;
 316
 317         /* Set if any shader has dirtied cachelines in the TMU that need to be
 318          * flushed before job end.
 319          */
 320         bool tmu_dirty_rcl;
 321
 322         /**
 323          * Set if a packet enabling TF has been emitted in the job (V3D 4.x).
 324          */
 325         bool tf_enabled;
 326
 327         /**
 328          * Current EZ state for drawing. Updated at the start of draw after
 329          * we've decided on the shader being rendered.
 330          */
 331         enum v3d_ez_state ez_state;
 332         /**
 333          * The first EZ state that was used for drawing with a decided EZ
 334          * direction (so either UNDECIDED, GT, or LT).
 335          */
 336         enum v3d_ez_state first_ez_state;
 337
 338         /**
 339          * Number of draw calls (not counting full buffer clears) queued in
 340          * the current job.
 341          */
 342         uint32_t draw_calls_queued;
 343
 344         struct v3d_job_key key;
 345 };
 346
 347 struct v3d_context {
 348         struct pipe_context base;
 349
 350         int fd;
 351         struct v3d_screen *screen;
 352
 353         /** The 3D rendering job for the currently bound FBO. */
 354         struct v3d_job *job;
 355
 356         /* Map from struct v3d_job_key to the job for that FBO.
 357          */
 358         struct hash_table *jobs;
 359
 360         /**
 361          * Map from v3d_resource to a job writing to that resource.
 362          *
 363          * Primarily for flushing jobs rendering to textures that are now
 364          * being read from.
 365          */
 366         struct hash_table *write_jobs;
 367
 368         struct slab_child_pool transfer_pool;
 369         struct blitter_context *blitter;
 370
 371         /** bitfield of VC5_DIRTY_* */
 372         uint32_t dirty;
 373
 374         struct primconvert_context *primconvert;
 375
 376         struct hash_table *fs_cache, *vs_cache;
 377         uint32_t next_uncompiled_program_id;
 378         uint64_t next_compiled_program_id;
 379
 380         struct v3d_compiler_state *compiler_state;
 381
 382         uint8_t prim_mode;
 383
 384         /** Maximum index buffer valid for the current shader_rec. */
 385         uint32_t max_index;
 386
 387         /** Sync object that our RCL or TFU job will update as its out_sync. */
 388         uint32_t out_sync;
 389
 390         /* Stream uploader used by gallium internals.  This could also be used
 391          * by driver internals, but we tend to use the v3d_cl.h interfaces
 392          * instead.
 393          */
 394         struct u_upload_mgr *uploader;
 395         /* State uploader used inside the driver.  This is for packing bits of
 396          * long-term state inside buffers, since the kernel interfaces
 397          * allocate a page at a time.
 398          */
 399         struct u_upload_mgr *state_uploader;
 400
 401         /** @{ Current pipeline state objects */
 402         struct pipe_scissor_state scissor;
 403         struct v3d_blend_state *blend;
 404         struct v3d_rasterizer_state *rasterizer;
 405         struct v3d_depth_stencil_alpha_state *zsa;
 406
 407         struct v3d_program_stateobj prog;
 408
 409         struct v3d_vertex_stateobj *vtx;
 410
 411         struct {
 412                 struct pipe_blend_color f;
 413                 uint16_t hf[4];
 414         } blend_color;
 415         struct pipe_stencil_ref stencil_ref;
 416         unsigned sample_mask;
 417         struct pipe_framebuffer_state framebuffer;
 418
 419         /* Per render target, whether we should swap the R and B fields in the
 420          * shader's color output and in blending.  If render targets disagree
 421          * on the R/B swap and use the constant color, then we would need to
 422          * fall back to in-shader blending.
 423          */
 424         uint8_t swap_color_rb;
 425
 426         /* Per render target, whether we should treat the dst alpha values as
 427          * one in blending.
 428          *
 429          * For RGBX formats, the tile buffer's alpha channel will be
 430          * undefined.
 431          */
 432         uint8_t blend_dst_alpha_one;
 433
 434         bool active_queries;
 435
 436         uint32_t tf_prims_generated;
 437         uint32_t prims_generated;
 438
 439         struct pipe_poly_stipple stipple;
 440         struct pipe_clip_state clip;
 441         struct pipe_viewport_state viewport;
 442         struct v3d_ssbo_stateobj ssbo[PIPE_SHADER_TYPES];
 443         struct v3d_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
 444         struct v3d_texture_stateobj tex[PIPE_SHADER_TYPES];
 445         struct v3d_vertexbuf_stateobj vertexbuf;
 446         struct v3d_streamout_stateobj streamout;
 447         struct v3d_bo *current_oq;
 448         struct pipe_debug_callback debug;
 449         /** @} */
 450 };
 451
 452 struct v3d_rasterizer_state {
 453         struct pipe_rasterizer_state base;
 454
 455         float point_size;
 456
 457         uint8_t depth_offset[9];
 458         uint8_t depth_offset_z16[9];
 459 };
 460
 461 struct v3d_depth_stencil_alpha_state {
 462         struct pipe_depth_stencil_alpha_state base;
 463
 464         enum v3d_ez_state ez_state;
 465
 466         uint8_t stencil_front[6];
 467         uint8_t stencil_back[6];
 468 };
 469
 470 struct v3d_blend_state {
 471         struct pipe_blend_state base;
 472
 473         /* Per-RT mask of whether blending is enabled. */
 474         uint8_t blend_enables;
 475 };
 476
 477 #define perf_debug(...) do {                            \
 478         if (unlikely(V3D_DEBUG & V3D_DEBUG_PERF))       \
 479                 fprintf(stderr, __VA_ARGS__);           \
 480         if (unlikely(v3d->debug.debug_message))         \
 481                 pipe_debug_message(&v3d->debug, PERF_INFO, __VA_ARGS__);    \
 482 } while (0)
 483
 484 #define foreach_bit(b, mask)                                            \
 485         for (uint32_t _m = (mask), b; _m && ({(b) = u_bit_scan(&_m); 1;});)
 486
 487 static inline struct v3d_context *
 488 v3d_context(struct pipe_context *pcontext)
 489 {
 490         return (struct v3d_context *)pcontext;
 491 }
 492
 493 static inline struct v3d_sampler_view *
 494 v3d_sampler_view(struct pipe_sampler_view *psview)
 495 {
 496         return (struct v3d_sampler_view *)psview;
 497 }
 498
 499 static inline struct v3d_sampler_state *
 500 v3d_sampler_state(struct pipe_sampler_state *psampler)
 501 {
 502         return (struct v3d_sampler_state *)psampler;
 503 }
 504
 505 struct pipe_context *v3d_context_create(struct pipe_screen *pscreen,
 506                                         void *priv, unsigned flags);
 507 void v3d_program_init(struct pipe_context *pctx);
 508 void v3d_program_fini(struct pipe_context *pctx);
 509 void v3d_query_init(struct pipe_context *pctx);
 510
 511 void v3d_simulator_init(struct v3d_screen *screen);
 512 void v3d_simulator_destroy(struct v3d_screen *screen);
 513 int v3d_simulator_ioctl(int fd, unsigned long request, void *arg);
 514 void v3d_simulator_open_from_handle(int fd, int handle, uint32_t size);
 515
 516 static inline int
 517 v3d_ioctl(int fd, unsigned long request, void *arg)
 518 {
 519         if (using_v3d_simulator)
 520                 return v3d_simulator_ioctl(fd, request, arg);
 521         else
 522                 return drmIoctl(fd, request, arg);
 523 }
 524
 525 void v3d_set_shader_uniform_dirty_flags(struct v3d_compiled_shader *shader);
 526 struct v3d_cl_reloc v3d_write_uniforms(struct v3d_context *v3d,
 527                                        struct v3d_compiled_shader *shader,
 528                                        enum pipe_shader_type stage);
 529
 530 void v3d_flush(struct pipe_context *pctx);
 531 void v3d_job_init(struct v3d_context *v3d);
 532 struct v3d_job *v3d_get_job(struct v3d_context *v3d,
 533                             struct pipe_surface **cbufs,
 534                             struct pipe_surface *zsbuf);
 535 struct v3d_job *v3d_get_job_for_fbo(struct v3d_context *v3d);
 536 void v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo);
 537 void v3d_job_add_write_resource(struct v3d_job *job, struct pipe_resource *prsc);
 538 void v3d_job_submit(struct v3d_context *v3d, struct v3d_job *job);
 539 void v3d_flush_jobs_writing_resource(struct v3d_context *v3d,
 540                                      struct pipe_resource *prsc);
 541 void v3d_flush_jobs_reading_resource(struct v3d_context *v3d,
 542                                      struct pipe_resource *prsc);
 543 void v3d_update_compiled_shaders(struct v3d_context *v3d, uint8_t prim_mode);
 544
 545 bool v3d_rt_format_supported(const struct v3d_device_info *devinfo,
 546                              enum pipe_format f);
 547 bool v3d_tex_format_supported(const struct v3d_device_info *devinfo,
 548                               enum pipe_format f);
 549 uint8_t v3d_get_rt_format(const struct v3d_device_info *devinfo, enum pipe_format f);
 550 uint8_t v3d_get_tex_format(const struct v3d_device_info *devinfo, enum pipe_format f);
 551 uint8_t v3d_get_tex_return_size(const struct v3d_device_info *devinfo,
 552                                 enum pipe_format f,
 553                                 enum pipe_tex_compare compare);
 554 uint8_t v3d_get_tex_return_channels(const struct v3d_device_info *devinfo,
 555                                     enum pipe_format f);
 556 const uint8_t *v3d_get_format_swizzle(const struct v3d_device_info *devinfo,
 557                                       enum pipe_format f);
 558 void v3d_get_internal_type_bpp_for_output_format(const struct v3d_device_info *devinfo,
 559                                                  uint32_t format,
 560                                                  uint32_t *type,
 561                                                  uint32_t *bpp);
 562 bool v3d_tfu_supports_tex_format(const struct v3d_device_info *devinfo,
 563                                  uint32_t tex_format);
 564
 565 void v3d_init_query_functions(struct v3d_context *v3d);
 566 void v3d_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info);
 567 void v3d_blitter_save(struct v3d_context *v3d);
 568 boolean v3d_generate_mipmap(struct pipe_context *pctx,
 569                             struct pipe_resource *prsc,
 570                             enum pipe_format format,
 571                             unsigned int base_level,
 572                             unsigned int last_level,
 573                             unsigned int first_layer,
 574                             unsigned int last_layer);
 575
 576 struct v3d_fence *v3d_fence_create(struct v3d_context *v3d);
 577
 578 #ifdef v3dX
 579 #  include "v3dx_context.h"
 580 #else
 581 #  define v3dX(x) v3d33_##x
 582 #  include "v3dx_context.h"
 583 #  undef v3dX
 584
 585 #  define v3dX(x) v3d41_##x
 586 #  include "v3dx_context.h"
 587 #  undef v3dX
 588 #endif
 589
 590 #endif /* VC5_CONTEXT_H */