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
  86 #define VC5_MAX_FS_INPUTS 64
  87
  88 struct v3d_sampler_view {
  89         struct pipe_sampler_view base;
  90         uint32_t p0;
  91         uint32_t p1;
  92         /* Precomputed swizzles to pass in to the shader key. */
  93         uint8_t swizzle[4];
  94
  95         uint8_t texture_shader_state[32];
  96         /* V3D 4.x: Texture state struct. */
  97         struct v3d_bo *bo;
  98 };
  99
 100 struct v3d_sampler_state {
 101         struct pipe_sampler_state base;
 102         uint32_t p0;
 103         uint32_t p1;
 104
 105         /* V3D 3.x: Packed texture state. */
 106         uint8_t texture_shader_state[32];
 107         /* V3D 4.x: Sampler state struct. */
 108         struct v3d_bo *bo;
 109 };
 110
 111 struct v3d_texture_stateobj {
 112         struct pipe_sampler_view *textures[PIPE_MAX_SAMPLERS];
 113         unsigned num_textures;
 114         struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
 115         unsigned num_samplers;
 116         struct v3d_cl_reloc texture_state[PIPE_MAX_SAMPLERS];
 117 };
 118
 119 struct v3d_shader_uniform_info {
 120         enum quniform_contents *contents;
 121         uint32_t *data;
 122         uint32_t count;
 123 };
 124
 125 struct v3d_uncompiled_shader {
 126         /** A name for this program, so you can track it in shader-db output. */
 127         uint32_t program_id;
 128         /** How many variants of this program were compiled, for shader-db. */
 129         uint32_t compiled_variant_count;
 130         struct pipe_shader_state base;
 131         uint32_t num_tf_outputs;
 132         struct v3d_varying_slot *tf_outputs;
 133         uint16_t tf_specs[16];
 134         uint16_t tf_specs_psiz[16];
 135         uint32_t num_tf_specs;
 136
 137         /**
 138          * Flag for if the NIR in this shader originally came from TGSI.  If
 139          * so, we need to do some fixups at compile time, due to missing
 140          * information in TGSI that exists in NIR.
 141          */
 142         bool was_tgsi;
 143 };
 144
 145 struct v3d_compiled_shader {
 146         struct v3d_bo *bo;
 147
 148         union {
 149                 struct v3d_prog_data *base;
 150                 struct v3d_vs_prog_data *vs;
 151                 struct v3d_fs_prog_data *fs;
 152         } prog_data;
 153
 154         /**
 155          * VC5_DIRTY_* flags that, when set in v3d->dirty, mean that the
 156          * uniforms have to be rewritten (and therefore the shader state
 157          * reemitted).
 158          */
 159         uint32_t uniform_dirty_bits;
 160 };
 161
 162 struct v3d_program_stateobj {
 163         struct v3d_uncompiled_shader *bind_vs, *bind_fs;
 164         struct v3d_compiled_shader *cs, *vs, *fs;
 165
 166         struct v3d_bo *spill_bo;
 167         int spill_size_per_thread;
 168 };
 169
 170 struct v3d_constbuf_stateobj {
 171         struct pipe_constant_buffer cb[PIPE_MAX_CONSTANT_BUFFERS];
 172         uint32_t enabled_mask;
 173         uint32_t dirty_mask;
 174 };
 175
 176 struct v3d_vertexbuf_stateobj {
 177         struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
 178         unsigned count;
 179         uint32_t enabled_mask;
 180         uint32_t dirty_mask;
 181 };
 182
 183 struct v3d_vertex_stateobj {
 184         struct pipe_vertex_element pipe[VC5_MAX_ATTRIBUTES];
 185         unsigned num_elements;
 186
 187         uint8_t attrs[16 * VC5_MAX_ATTRIBUTES];
 188         struct v3d_bo *default_attribute_values;
 189 };
 190
 191 struct v3d_streamout_stateobj {
 192         struct pipe_stream_output_target *targets[PIPE_MAX_SO_BUFFERS];
 193         /* Number of vertices we've written into the buffer so far. */
 194         uint32_t offsets[PIPE_MAX_SO_BUFFERS];
 195         unsigned num_targets;
 196 };
 197
 198 /* Hash table key for v3d->jobs */
 199 struct v3d_job_key {
 200         struct pipe_surface *cbufs[4];
 201         struct pipe_surface *zsbuf;
 202 };
 203
 204 enum v3d_ez_state {
 205         VC5_EZ_UNDECIDED = 0,
 206         VC5_EZ_GT_GE,
 207         VC5_EZ_LT_LE,
 208         VC5_EZ_DISABLED,
 209 };
 210
 211 /**
 212  * A complete bin/render job.
 213  *
 214  * This is all of the state necessary to submit a bin/render to the kernel.
 215  * We want to be able to have multiple in progress at a time, so that we don't
 216  * need to flush an existing CL just to switch to rendering to a new render
 217  * target (which would mean reading back from the old render target when
 218  * starting to render to it again).
 219  */
 220 struct v3d_job {
 221         struct v3d_context *v3d;
 222         struct v3d_cl bcl;
 223         struct v3d_cl rcl;
 224         struct v3d_cl indirect;
 225         struct v3d_bo *tile_alloc;
 226         struct v3d_bo *tile_state;
 227         uint32_t shader_rec_count;
 228
 229         struct drm_v3d_submit_cl submit;
 230
 231         /**
 232          * Set of all BOs referenced by the job.  This will be used for making
 233          * the list of BOs that the kernel will need to have paged in to
 234          * execute our job.
 235          */
 236         struct set *bos;
 237
 238         /** Sum of the sizes of the BOs referenced by the job. */
 239         uint32_t referenced_size;
 240
 241         struct set *write_prscs;
 242
 243         /* Size of the submit.bo_handles array. */
 244         uint32_t bo_handles_size;
 245
 246         /** @{ Surfaces to submit rendering for. */
 247         struct pipe_surface *cbufs[4];
 248         struct pipe_surface *zsbuf;
 249         /** @} */
 250         /** @{
 251          * Bounding box of the scissor across all queued drawing.
 252          *
 253          * Note that the max values are exclusive.
 254          */
 255         uint32_t draw_min_x;
 256         uint32_t draw_min_y;
 257         uint32_t draw_max_x;
 258         uint32_t draw_max_y;
 259         /** @} */
 260         /** @{
 261          * Width/height of the color framebuffer being rendered to,
 262          * for VC5_TILE_RENDERING_MODE_CONFIG.
 263         */
 264         uint32_t draw_width;
 265         uint32_t draw_height;
 266         /** @} */
 267         /** @{ Tile information, depending on MSAA and float color buffer. */
 268         uint32_t draw_tiles_x; /** @< Number of tiles wide for framebuffer. */
 269         uint32_t draw_tiles_y; /** @< Number of tiles high for framebuffer. */
 270
 271         uint32_t tile_width; /** @< Width of a tile. */
 272         uint32_t tile_height; /** @< Height of a tile. */
 273         /** maximum internal_bpp of all color render targets. */
 274         uint32_t internal_bpp;
 275
 276         /** Whether the current rendering is in a 4X MSAA tile buffer. */
 277         bool msaa;
 278         /** @} */
 279
 280         /* Bitmask of PIPE_CLEAR_* of buffers that were cleared before the
 281          * first rendering.
 282          */
 283         uint32_t cleared;
 284         /* Bitmask of PIPE_CLEAR_* of buffers that have been rendered to
 285          * (either clears or draws).
 286          */
 287         uint32_t resolve;
 288         uint32_t clear_color[4][4];
 289         float clear_z;
 290         uint8_t clear_s;
 291
 292         /**
 293          * Set if some drawing (triangles, blits, or just a glClear()) has
 294          * been done to the FBO, meaning that we need to
 295          * DRM_IOCTL_VC5_SUBMIT_CL.
 296          */
 297         bool needs_flush;
 298
 299         /**
 300          * Set if there is a nonzero address for OCCLUSION_QUERY_COUNTER.  If
 301          * so, we need to disable it and flush before ending the CL, to keep
 302          * the next tile from starting with it enabled.
 303          */
 304         bool oq_enabled;
 305
 306         /**
 307          * Set when a packet enabling TF on all further primitives has been
 308          * emitted.
 309          */
 310         bool tf_enabled;
 311
 312         /**
 313          * Current EZ state for drawing. Updated at the start of draw after
 314          * we've decided on the shader being rendered.
 315          */
 316         enum v3d_ez_state ez_state;
 317         /**
 318          * The first EZ state that was used for drawing with a decided EZ
 319          * direction (so either UNDECIDED, GT, or LT).
 320          */
 321         enum v3d_ez_state first_ez_state;
 322
 323         /**
 324          * Number of draw calls (not counting full buffer clears) queued in
 325          * the current job.
 326          */
 327         uint32_t draw_calls_queued;
 328
 329         struct v3d_job_key key;
 330 };
 331
 332 struct v3d_context {
 333         struct pipe_context base;
 334
 335         int fd;
 336         struct v3d_screen *screen;
 337
 338         /** The 3D rendering job for the currently bound FBO. */
 339         struct v3d_job *job;
 340
 341         /* Map from struct v3d_job_key to the job for that FBO.
 342          */
 343         struct hash_table *jobs;
 344
 345         /**
 346          * Map from v3d_resource to a job writing to that resource.
 347          *
 348          * Primarily for flushing jobs rendering to textures that are now
 349          * being read from.
 350          */
 351         struct hash_table *write_jobs;
 352
 353         struct slab_child_pool transfer_pool;
 354         struct blitter_context *blitter;
 355
 356         /** bitfield of VC5_DIRTY_* */
 357         uint32_t dirty;
 358
 359         struct primconvert_context *primconvert;
 360
 361         struct hash_table *fs_cache, *vs_cache;
 362         uint32_t next_uncompiled_program_id;
 363         uint64_t next_compiled_program_id;
 364
 365         struct v3d_compiler_state *compiler_state;
 366
 367         uint8_t prim_mode;
 368
 369         /** Maximum index buffer valid for the current shader_rec. */
 370         uint32_t max_index;
 371
 372         /** Sync object that our RCL will update as its out_sync. */
 373         uint32_t out_sync;
 374
 375         struct u_upload_mgr *uploader;
 376
 377         /** @{ Current pipeline state objects */
 378         struct pipe_scissor_state scissor;
 379         struct v3d_blend_state *blend;
 380         struct v3d_rasterizer_state *rasterizer;
 381         struct v3d_depth_stencil_alpha_state *zsa;
 382
 383         struct v3d_texture_stateobj verttex, fragtex;
 384
 385         struct v3d_program_stateobj prog;
 386
 387         struct v3d_vertex_stateobj *vtx;
 388
 389         struct {
 390                 struct pipe_blend_color f;
 391                 uint16_t hf[4];
 392         } blend_color;
 393         struct pipe_stencil_ref stencil_ref;
 394         unsigned sample_mask;
 395         struct pipe_framebuffer_state framebuffer;
 396
 397         /* Per render target, whether we should swap the R and B fields in the
 398          * shader's color output and in blending.  If render targets disagree
 399          * on the R/B swap and use the constant color, then we would need to
 400          * fall back to in-shader blending.
 401          */
 402         uint8_t swap_color_rb;
 403
 404         /* Per render target, whether we should treat the dst alpha values as
 405          * one in blending.
 406          *
 407          * For RGBX formats, the tile buffer's alpha channel will be
 408          * undefined.
 409          */
 410         uint8_t blend_dst_alpha_one;
 411
 412         bool active_queries;
 413
 414         uint32_t tf_prims_generated;
 415         uint32_t prims_generated;
 416
 417         struct pipe_poly_stipple stipple;
 418         struct pipe_clip_state clip;
 419         struct pipe_viewport_state viewport;
 420         struct v3d_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
 421         struct v3d_vertexbuf_stateobj vertexbuf;
 422         struct v3d_streamout_stateobj streamout;
 423         struct v3d_bo *current_oq;
 424         /** @} */
 425 };
 426
 427 struct v3d_rasterizer_state {
 428         struct pipe_rasterizer_state base;
 429
 430         float point_size;
 431
 432         /**
 433          * Half-float (1/8/7 bits) value of polygon offset units for
 434          * VC5_PACKET_DEPTH_OFFSET
 435          */
 436         uint16_t offset_units;
 437         /**
 438          * The HW treats polygon offset units based on a Z24 buffer, so we
 439          * need to scale up offset_units if we're only Z16.
 440          */
 441         uint16_t z16_offset_units;
 442         /**
 443          * Half-float (1/8/7 bits) value of polygon offset scale for
 444          * VC5_PACKET_DEPTH_OFFSET
 445          */
 446         uint16_t offset_factor;
 447 };
 448
 449 struct v3d_depth_stencil_alpha_state {
 450         struct pipe_depth_stencil_alpha_state base;
 451
 452         enum v3d_ez_state ez_state;
 453
 454         uint8_t stencil_front[6];
 455         uint8_t stencil_back[6];
 456 };
 457
 458 struct v3d_blend_state {
 459         struct pipe_blend_state base;
 460
 461         /* Per-RT mask of whether blending is enabled. */
 462         uint8_t blend_enables;
 463 };
 464
 465 #define perf_debug(...) do {                            \
 466         if (unlikely(V3D_DEBUG & V3D_DEBUG_PERF))       \
 467                 fprintf(stderr, __VA_ARGS__);           \
 468 } while (0)
 469
 470 static inline struct v3d_context *
 471 v3d_context(struct pipe_context *pcontext)
 472 {
 473         return (struct v3d_context *)pcontext;
 474 }
 475
 476 static inline struct v3d_sampler_view *
 477 v3d_sampler_view(struct pipe_sampler_view *psview)
 478 {
 479         return (struct v3d_sampler_view *)psview;
 480 }
 481
 482 static inline struct v3d_sampler_state *
 483 v3d_sampler_state(struct pipe_sampler_state *psampler)
 484 {
 485         return (struct v3d_sampler_state *)psampler;
 486 }
 487
 488 struct pipe_context *v3d_context_create(struct pipe_screen *pscreen,
 489                                         void *priv, unsigned flags);
 490 void v3d_program_init(struct pipe_context *pctx);
 491 void v3d_program_fini(struct pipe_context *pctx);
 492 void v3d_query_init(struct pipe_context *pctx);
 493
 494 void v3d_simulator_init(struct v3d_screen *screen);
 495 void v3d_simulator_destroy(struct v3d_screen *screen);
 496 int v3d_simulator_flush(struct v3d_context *v3d,
 497                         struct drm_v3d_submit_cl *args,
 498                         struct v3d_job *job);
 499 int v3d_simulator_ioctl(int fd, unsigned long request, void *arg);
 500 void v3d_simulator_open_from_handle(int fd, uint32_t winsys_stride,
 501                                     int handle, uint32_t size);
 502
 503 static inline int
 504 v3d_ioctl(int fd, unsigned long request, void *arg)
 505 {
 506         if (using_v3d_simulator)
 507                 return v3d_simulator_ioctl(fd, request, arg);
 508         else
 509                 return drmIoctl(fd, request, arg);
 510 }
 511
 512 void v3d_set_shader_uniform_dirty_flags(struct v3d_compiled_shader *shader);
 513 struct v3d_cl_reloc v3d_write_uniforms(struct v3d_context *v3d,
 514                                        struct v3d_compiled_shader *shader,
 515                                        struct v3d_constbuf_stateobj *cb,
 516                                        struct v3d_texture_stateobj *texstate);
 517
 518 void v3d_flush(struct pipe_context *pctx);
 519 void v3d_job_init(struct v3d_context *v3d);
 520 struct v3d_job *v3d_get_job(struct v3d_context *v3d,
 521                             struct pipe_surface **cbufs,
 522                             struct pipe_surface *zsbuf);
 523 struct v3d_job *v3d_get_job_for_fbo(struct v3d_context *v3d);
 524 void v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo);
 525 void v3d_job_add_write_resource(struct v3d_job *job, struct pipe_resource *prsc);
 526 void v3d_job_submit(struct v3d_context *v3d, struct v3d_job *job);
 527 void v3d_flush_jobs_writing_resource(struct v3d_context *v3d,
 528                                      struct pipe_resource *prsc);
 529 void v3d_flush_jobs_reading_resource(struct v3d_context *v3d,
 530                                      struct pipe_resource *prsc);
 531 void v3d_update_compiled_shaders(struct v3d_context *v3d, uint8_t prim_mode);
 532
 533 bool v3d_rt_format_supported(const struct v3d_device_info *devinfo,
 534                              enum pipe_format f);
 535 bool v3d_tex_format_supported(const struct v3d_device_info *devinfo,
 536                               enum pipe_format f);
 537 uint8_t v3d_get_rt_format(const struct v3d_device_info *devinfo, enum pipe_format f);
 538 uint8_t v3d_get_tex_format(const struct v3d_device_info *devinfo, enum pipe_format f);
 539 uint8_t v3d_get_tex_return_size(const struct v3d_device_info *devinfo,
 540                                 enum pipe_format f,
 541                                 enum pipe_tex_compare compare);
 542 uint8_t v3d_get_tex_return_channels(const struct v3d_device_info *devinfo,
 543                                     enum pipe_format f);
 544 const uint8_t *v3d_get_format_swizzle(const struct v3d_device_info *devinfo,
 545                                       enum pipe_format f);
 546 void v3d_get_internal_type_bpp_for_output_format(const struct v3d_device_info *devinfo,
 547                                                  uint32_t format,
 548                                                  uint32_t *type,
 549                                                  uint32_t *bpp);
 550
 551 void v3d_init_query_functions(struct v3d_context *v3d);
 552 void v3d_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info);
 553 void v3d_blitter_save(struct v3d_context *v3d);
 554
 555 struct v3d_fence *v3d_fence_create(struct v3d_context *v3d);
 556
 557 #ifdef v3dX
 558 #  include "v3dx_context.h"
 559 #else
 560 #  define v3dX(x) v3d33_##x
 561 #  include "v3dx_context.h"
 562 #  undef v3dX
 563
 564 #  define v3dX(x) v3d41_##x
 565 #  include "v3dx_context.h"
 566 #  undef v3dX
 567 #endif
 568
 569 #endif /* VC5_CONTEXT_H */