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
  85 #define VC5_MAX_FS_INPUTS 64
  86
  87 struct v3d_sampler_view {
  88         struct pipe_sampler_view base;
  89         uint32_t p0;
  90         uint32_t p1;
  91         /* Precomputed swizzles to pass in to the shader key. */
  92         uint8_t swizzle[4];
  93
  94         uint8_t texture_shader_state[32];
  95         /* V3D 4.x: Texture state struct. */
  96         struct v3d_bo *bo;
  97 };
  98
  99 struct v3d_sampler_state {
 100         struct pipe_sampler_state base;
 101         uint32_t p0;
 102         uint32_t p1;
 103
 104         /* V3D 3.x: Packed texture state. */
 105         uint8_t texture_shader_state[32];
 106         /* V3D 4.x: Sampler state struct. */
 107         struct v3d_bo *bo;
 108 };
 109
 110 struct v3d_texture_stateobj {
 111         struct pipe_sampler_view *textures[PIPE_MAX_SAMPLERS];
 112         unsigned num_textures;
 113         struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
 114         unsigned num_samplers;
 115         struct v3d_cl_reloc texture_state[PIPE_MAX_SAMPLERS];
 116 };
 117
 118 struct v3d_shader_uniform_info {
 119         enum quniform_contents *contents;
 120         uint32_t *data;
 121         uint32_t count;
 122 };
 123
 124 struct v3d_uncompiled_shader {
 125         /** A name for this program, so you can track it in shader-db output. */
 126         uint32_t program_id;
 127         /** How many variants of this program were compiled, for shader-db. */
 128         uint32_t compiled_variant_count;
 129         struct pipe_shader_state base;
 130         uint32_t num_tf_outputs;
 131         struct v3d_varying_slot *tf_outputs;
 132         uint16_t tf_specs[16];
 133         uint16_t tf_specs_psiz[16];
 134         uint32_t num_tf_specs;
 135
 136         /**
 137          * Flag for if the NIR in this shader originally came from TGSI.  If
 138          * so, we need to do some fixups at compile time, due to missing
 139          * information in TGSI that exists in NIR.
 140          */
 141         bool was_tgsi;
 142 };
 143
 144 struct v3d_compiled_shader {
 145         struct v3d_bo *bo;
 146
 147         union {
 148                 struct v3d_prog_data *base;
 149                 struct v3d_vs_prog_data *vs;
 150                 struct v3d_fs_prog_data *fs;
 151         } prog_data;
 152
 153         /**
 154          * VC5_DIRTY_* flags that, when set in v3d->dirty, mean that the
 155          * uniforms have to be rewritten (and therefore the shader state
 156          * reemitted).
 157          */
 158         uint32_t uniform_dirty_bits;
 159 };
 160
 161 struct v3d_program_stateobj {
 162         struct v3d_uncompiled_shader *bind_vs, *bind_fs;
 163         struct v3d_compiled_shader *cs, *vs, *fs;
 164
 165         struct v3d_bo *spill_bo;
 166         int spill_size_per_thread;
 167 };
 168
 169 struct v3d_constbuf_stateobj {
 170         struct pipe_constant_buffer cb[PIPE_MAX_CONSTANT_BUFFERS];
 171         uint32_t enabled_mask;
 172         uint32_t dirty_mask;
 173 };
 174
 175 struct v3d_vertexbuf_stateobj {
 176         struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
 177         unsigned count;
 178         uint32_t enabled_mask;
 179         uint32_t dirty_mask;
 180 };
 181
 182 struct v3d_vertex_stateobj {
 183         struct pipe_vertex_element pipe[VC5_MAX_ATTRIBUTES];
 184         unsigned num_elements;
 185
 186         uint8_t attrs[12 * VC5_MAX_ATTRIBUTES];
 187         struct v3d_bo *default_attribute_values;
 188 };
 189
 190 struct v3d_streamout_stateobj {
 191         struct pipe_stream_output_target *targets[PIPE_MAX_SO_BUFFERS];
 192         unsigned num_targets;
 193 };
 194
 195 /* Hash table key for v3d->jobs */
 196 struct v3d_job_key {
 197         struct pipe_surface *cbufs[4];
 198         struct pipe_surface *zsbuf;
 199 };
 200
 201 enum v3d_ez_state {
 202         VC5_EZ_UNDECIDED = 0,
 203         VC5_EZ_GT_GE,
 204         VC5_EZ_LT_LE,
 205         VC5_EZ_DISABLED,
 206 };
 207
 208 /**
 209  * A complete bin/render job.
 210  *
 211  * This is all of the state necessary to submit a bin/render to the kernel.
 212  * We want to be able to have multiple in progress at a time, so that we don't
 213  * need to flush an existing CL just to switch to rendering to a new render
 214  * target (which would mean reading back from the old render target when
 215  * starting to render to it again).
 216  */
 217 struct v3d_job {
 218         struct v3d_context *v3d;
 219         struct v3d_cl bcl;
 220         struct v3d_cl rcl;
 221         struct v3d_cl indirect;
 222         struct v3d_bo *tile_alloc;
 223         struct v3d_bo *tile_state;
 224         uint32_t shader_rec_count;
 225
 226         struct drm_v3d_submit_cl submit;
 227
 228         /**
 229          * Set of all BOs referenced by the job.  This will be used for making
 230          * the list of BOs that the kernel will need to have paged in to
 231          * execute our job.
 232          */
 233         struct set *bos;
 234
 235         /** Sum of the sizes of the BOs referenced by the job. */
 236         uint32_t referenced_size;
 237
 238         struct set *write_prscs;
 239
 240         /* Size of the submit.bo_handles array. */
 241         uint32_t bo_handles_size;
 242
 243         /** @{ Surfaces to submit rendering for. */
 244         struct pipe_surface *cbufs[4];
 245         struct pipe_surface *zsbuf;
 246         /** @} */
 247         /** @{
 248          * Bounding box of the scissor across all queued drawing.
 249          *
 250          * Note that the max values are exclusive.
 251          */
 252         uint32_t draw_min_x;
 253         uint32_t draw_min_y;
 254         uint32_t draw_max_x;
 255         uint32_t draw_max_y;
 256         /** @} */
 257         /** @{
 258          * Width/height of the color framebuffer being rendered to,
 259          * for VC5_TILE_RENDERING_MODE_CONFIG.
 260         */
 261         uint32_t draw_width;
 262         uint32_t draw_height;
 263         /** @} */
 264         /** @{ Tile information, depending on MSAA and float color buffer. */
 265         uint32_t draw_tiles_x; /** @< Number of tiles wide for framebuffer. */
 266         uint32_t draw_tiles_y; /** @< Number of tiles high for framebuffer. */
 267
 268         uint32_t tile_width; /** @< Width of a tile. */
 269         uint32_t tile_height; /** @< Height of a tile. */
 270         /** maximum internal_bpp of all color render targets. */
 271         uint32_t internal_bpp;
 272
 273         /** Whether the current rendering is in a 4X MSAA tile buffer. */
 274         bool msaa;
 275         /** @} */
 276
 277         /* Bitmask of PIPE_CLEAR_* of buffers that were cleared before the
 278          * first rendering.
 279          */
 280         uint32_t cleared;
 281         /* Bitmask of PIPE_CLEAR_* of buffers that have been rendered to
 282          * (either clears or draws).
 283          */
 284         uint32_t resolve;
 285         uint32_t clear_color[4][4];
 286         float clear_z;
 287         uint8_t clear_s;
 288
 289         /**
 290          * Set if some drawing (triangles, blits, or just a glClear()) has
 291          * been done to the FBO, meaning that we need to
 292          * DRM_IOCTL_VC5_SUBMIT_CL.
 293          */
 294         bool needs_flush;
 295
 296         /**
 297          * Set if there is a nonzero address for OCCLUSION_QUERY_COUNTER.  If
 298          * so, we need to disable it and flush before ending the CL, to keep
 299          * the next tile from starting with it enabled.
 300          */
 301         bool oq_enabled;
 302
 303         /**
 304          * Set when a packet enabling TF on all further primitives has been
 305          * emitted.
 306          */
 307         bool tf_enabled;
 308
 309         /**
 310          * Current EZ state for drawing. Updated at the start of draw after
 311          * we've decided on the shader being rendered.
 312          */
 313         enum v3d_ez_state ez_state;
 314         /**
 315          * The first EZ state that was used for drawing with a decided EZ
 316          * direction (so either UNDECIDED, GT, or LT).
 317          */
 318         enum v3d_ez_state first_ez_state;
 319
 320         /**
 321          * Number of draw calls (not counting full buffer clears) queued in
 322          * the current job.
 323          */
 324         uint32_t draw_calls_queued;
 325
 326         struct v3d_job_key key;
 327 };
 328
 329 struct v3d_context {
 330         struct pipe_context base;
 331
 332         int fd;
 333         struct v3d_screen *screen;
 334
 335         /** The 3D rendering job for the currently bound FBO. */
 336         struct v3d_job *job;
 337
 338         /* Map from struct v3d_job_key to the job for that FBO.
 339          */
 340         struct hash_table *jobs;
 341
 342         /**
 343          * Map from v3d_resource to a job writing to that resource.
 344          *
 345          * Primarily for flushing jobs rendering to textures that are now
 346          * being read from.
 347          */
 348         struct hash_table *write_jobs;
 349
 350         struct slab_child_pool transfer_pool;
 351         struct blitter_context *blitter;
 352
 353         /** bitfield of VC5_DIRTY_* */
 354         uint32_t dirty;
 355
 356         struct primconvert_context *primconvert;
 357
 358         struct hash_table *fs_cache, *vs_cache;
 359         uint32_t next_uncompiled_program_id;
 360         uint64_t next_compiled_program_id;
 361
 362         struct v3d_compiler_state *compiler_state;
 363
 364         uint8_t prim_mode;
 365
 366         /** Maximum index buffer valid for the current shader_rec. */
 367         uint32_t max_index;
 368
 369         /** Sync object that our RCL will update as its out_sync. */
 370         uint32_t out_sync;
 371
 372         struct u_upload_mgr *uploader;
 373
 374         /** @{ Current pipeline state objects */
 375         struct pipe_scissor_state scissor;
 376         struct pipe_blend_state *blend;
 377         struct v3d_rasterizer_state *rasterizer;
 378         struct v3d_depth_stencil_alpha_state *zsa;
 379
 380         struct v3d_texture_stateobj verttex, fragtex;
 381
 382         struct v3d_program_stateobj prog;
 383
 384         struct v3d_vertex_stateobj *vtx;
 385
 386         struct {
 387                 struct pipe_blend_color f;
 388                 uint16_t hf[4];
 389         } blend_color;
 390         struct pipe_stencil_ref stencil_ref;
 391         unsigned sample_mask;
 392         struct pipe_framebuffer_state framebuffer;
 393
 394         /* Per render target, whether we should swap the R and B fields in the
 395          * shader's color output and in blending.  If render targets disagree
 396          * on the R/B swap and use the constant color, then we would need to
 397          * fall back to in-shader blending.
 398          */
 399         uint8_t swap_color_rb;
 400
 401         /* Per render target, whether we should treat the dst alpha values as
 402          * one in blending.
 403          *
 404          * For RGBX formats, the tile buffer's alpha channel will be
 405          * undefined.
 406          */
 407         uint8_t blend_dst_alpha_one;
 408
 409         bool active_queries;
 410
 411         uint32_t tf_prims_generated;
 412         uint32_t prims_generated;
 413
 414         struct pipe_poly_stipple stipple;
 415         struct pipe_clip_state clip;
 416         struct pipe_viewport_state viewport;
 417         struct v3d_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
 418         struct v3d_vertexbuf_stateobj vertexbuf;
 419         struct v3d_streamout_stateobj streamout;
 420         struct v3d_bo *current_oq;
 421         /** @} */
 422 };
 423
 424 struct v3d_rasterizer_state {
 425         struct pipe_rasterizer_state base;
 426
 427         float point_size;
 428
 429         /**
 430          * Half-float (1/8/7 bits) value of polygon offset units for
 431          * VC5_PACKET_DEPTH_OFFSET
 432          */
 433         uint16_t offset_units;
 434         /**
 435          * Half-float (1/8/7 bits) value of polygon offset scale for
 436          * VC5_PACKET_DEPTH_OFFSET
 437          */
 438         uint16_t offset_factor;
 439 };
 440
 441 struct v3d_depth_stencil_alpha_state {
 442         struct pipe_depth_stencil_alpha_state base;
 443
 444         enum v3d_ez_state ez_state;
 445
 446         uint8_t stencil_front[6];
 447         uint8_t stencil_back[6];
 448 };
 449
 450 #define perf_debug(...) do {                            \
 451         if (unlikely(V3D_DEBUG & V3D_DEBUG_PERF))       \
 452                 fprintf(stderr, __VA_ARGS__);           \
 453 } while (0)
 454
 455 static inline struct v3d_context *
 456 v3d_context(struct pipe_context *pcontext)
 457 {
 458         return (struct v3d_context *)pcontext;
 459 }
 460
 461 static inline struct v3d_sampler_view *
 462 v3d_sampler_view(struct pipe_sampler_view *psview)
 463 {
 464         return (struct v3d_sampler_view *)psview;
 465 }
 466
 467 static inline struct v3d_sampler_state *
 468 v3d_sampler_state(struct pipe_sampler_state *psampler)
 469 {
 470         return (struct v3d_sampler_state *)psampler;
 471 }
 472
 473 struct pipe_context *v3d_context_create(struct pipe_screen *pscreen,
 474                                         void *priv, unsigned flags);
 475 void v3d_program_init(struct pipe_context *pctx);
 476 void v3d_program_fini(struct pipe_context *pctx);
 477 void v3d_query_init(struct pipe_context *pctx);
 478
 479 void v3d_simulator_init(struct v3d_screen *screen);
 480 void v3d_simulator_destroy(struct v3d_screen *screen);
 481 int v3d_simulator_flush(struct v3d_context *v3d,
 482                         struct drm_v3d_submit_cl *args,
 483                         struct v3d_job *job);
 484 int v3d_simulator_ioctl(int fd, unsigned long request, void *arg);
 485 void v3d_simulator_open_from_handle(int fd, uint32_t winsys_stride,
 486                                     int handle, uint32_t size);
 487
 488 static inline int
 489 v3d_ioctl(int fd, unsigned long request, void *arg)
 490 {
 491         if (using_v3d_simulator)
 492                 return v3d_simulator_ioctl(fd, request, arg);
 493         else
 494                 return drmIoctl(fd, request, arg);
 495 }
 496
 497 void v3d_set_shader_uniform_dirty_flags(struct v3d_compiled_shader *shader);
 498 struct v3d_cl_reloc v3d_write_uniforms(struct v3d_context *v3d,
 499                                        struct v3d_compiled_shader *shader,
 500                                        struct v3d_constbuf_stateobj *cb,
 501                                        struct v3d_texture_stateobj *texstate);
 502
 503 void v3d_flush(struct pipe_context *pctx);
 504 void v3d_job_init(struct v3d_context *v3d);
 505 struct v3d_job *v3d_get_job(struct v3d_context *v3d,
 506                             struct pipe_surface **cbufs,
 507                             struct pipe_surface *zsbuf);
 508 struct v3d_job *v3d_get_job_for_fbo(struct v3d_context *v3d);
 509 void v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo);
 510 void v3d_job_add_write_resource(struct v3d_job *job, struct pipe_resource *prsc);
 511 void v3d_job_submit(struct v3d_context *v3d, struct v3d_job *job);
 512 void v3d_flush_jobs_writing_resource(struct v3d_context *v3d,
 513                                      struct pipe_resource *prsc);
 514 void v3d_flush_jobs_reading_resource(struct v3d_context *v3d,
 515                                      struct pipe_resource *prsc);
 516 void v3d_update_compiled_shaders(struct v3d_context *v3d, uint8_t prim_mode);
 517
 518 bool v3d_rt_format_supported(const struct v3d_device_info *devinfo,
 519                              enum pipe_format f);
 520 bool v3d_tex_format_supported(const struct v3d_device_info *devinfo,
 521                               enum pipe_format f);
 522 uint8_t v3d_get_rt_format(const struct v3d_device_info *devinfo, enum pipe_format f);
 523 uint8_t v3d_get_tex_format(const struct v3d_device_info *devinfo, enum pipe_format f);
 524 uint8_t v3d_get_tex_return_size(const struct v3d_device_info *devinfo,
 525                                 enum pipe_format f,
 526                                 enum pipe_tex_compare compare);
 527 uint8_t v3d_get_tex_return_channels(const struct v3d_device_info *devinfo,
 528                                     enum pipe_format f);
 529 const uint8_t *v3d_get_format_swizzle(const struct v3d_device_info *devinfo,
 530                                       enum pipe_format f);
 531 void v3d_get_internal_type_bpp_for_output_format(const struct v3d_device_info *devinfo,
 532                                                  uint32_t format,
 533                                                  uint32_t *type,
 534                                                  uint32_t *bpp);
 535
 536 void v3d_init_query_functions(struct v3d_context *v3d);
 537 void v3d_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info);
 538 void v3d_blitter_save(struct v3d_context *v3d);
 539
 540 struct v3d_fence *v3d_fence_create(struct v3d_context *v3d);
 541
 542 #ifdef v3dX
 543 #  include "v3dx_context.h"
 544 #else
 545 #  define v3dX(x) v3d33_##x
 546 #  include "v3dx_context.h"
 547 #  undef v3dX
 548
 549 #  define v3dX(x) v3d41_##x
 550 #  include "v3dx_context.h"
 551 #  undef v3dX
 552 #endif
 553
 554 #endif /* VC5_CONTEXT_H */