src/gallium/drivers/freedreno/freedreno_context.h

   1 /*
   2  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
   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:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #ifndef FREEDRENO_CONTEXT_H_
  28 #define FREEDRENO_CONTEXT_H_
  29
  30 #include "pipe/p_context.h"
  31 #include "indices/u_primconvert.h"
  32 #include "util/u_blitter.h"
  33 #include "util/list.h"
  34 #include "util/slab.h"
  35 #include "util/u_string.h"
  36
  37 #include "freedreno_batch.h"
  38 #include "freedreno_screen.h"
  39 #include "freedreno_gmem.h"
  40 #include "freedreno_util.h"
  41
  42 #define BORDER_COLOR_UPLOAD_SIZE (2 * PIPE_MAX_SAMPLERS * BORDERCOLOR_SIZE)
  43
  44 struct fd_vertex_stateobj;
  45
  46 struct fd_texture_stateobj {
  47         struct pipe_sampler_view *textures[PIPE_MAX_SAMPLERS];
  48         unsigned num_textures;
  49         unsigned valid_textures;
  50         struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
  51         unsigned num_samplers;
  52         unsigned valid_samplers;
  53         /* number of samples per sampler, 2 bits per sampler: */
  54         uint32_t samples;
  55 };
  56
  57 struct fd_program_stateobj {
  58         void *vp, *fp;
  59
  60         /* rest only used by fd2.. split out: */
  61         uint8_t num_exports;
  62         /* Indexed by semantic name or TGSI_SEMANTIC_COUNT + semantic index
  63          * for TGSI_SEMANTIC_GENERIC.  Special vs exports (position and point-
  64          * size) are not included in this
  65          */
  66         uint8_t export_linkage[63];
  67 };
  68
  69 struct fd_constbuf_stateobj {
  70         struct pipe_constant_buffer cb[PIPE_MAX_CONSTANT_BUFFERS];
  71         uint32_t enabled_mask;
  72 };
  73
  74 struct fd_shaderbuf_stateobj {
  75         struct pipe_shader_buffer sb[PIPE_MAX_SHADER_BUFFERS];
  76         uint32_t enabled_mask;
  77 };
  78
  79 struct fd_shaderimg_stateobj {
  80         struct pipe_image_view si[PIPE_MAX_SHADER_IMAGES];
  81         uint32_t enabled_mask;
  82 };
  83
  84 struct fd_vertexbuf_stateobj {
  85         struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
  86         unsigned count;
  87         uint32_t enabled_mask;
  88 };
  89
  90 struct fd_vertex_stateobj {
  91         struct pipe_vertex_element pipe[PIPE_MAX_ATTRIBS];
  92         unsigned num_elements;
  93 };
  94
  95 struct fd_streamout_stateobj {
  96         struct pipe_stream_output_target *targets[PIPE_MAX_SO_BUFFERS];
  97         unsigned num_targets;
  98         /* Track offset from vtxcnt for streamout data.  This counter
  99          * is just incremented by # of vertices on each draw until
 100          * reset or new streamout buffer bound.
 101          *
 102          * When we eventually have GS, the CPU won't actually know the
 103          * number of vertices per draw, so I think we'll have to do
 104          * something more clever.
 105          */
 106         unsigned offsets[PIPE_MAX_SO_BUFFERS];
 107 };
 108
 109 #define MAX_GLOBAL_BUFFERS 16
 110 struct fd_global_bindings_stateobj {
 111         struct pipe_resource *buf[MAX_GLOBAL_BUFFERS];
 112         uint32_t enabled_mask;
 113 };
 114
 115 /* group together the vertex and vertexbuf state.. for ease of passing
 116  * around, and because various internal operations (gmem<->mem, etc)
 117  * need their own vertex state:
 118  */
 119 struct fd_vertex_state {
 120         struct fd_vertex_stateobj *vtx;
 121         struct fd_vertexbuf_stateobj vertexbuf;
 122 };
 123
 124 /* global 3d pipeline dirty state: */
 125 enum fd_dirty_3d_state {
 126         FD_DIRTY_BLEND       = BIT(0),
 127         FD_DIRTY_RASTERIZER  = BIT(1),
 128         FD_DIRTY_ZSA         = BIT(2),
 129         FD_DIRTY_BLEND_COLOR = BIT(3),
 130         FD_DIRTY_STENCIL_REF = BIT(4),
 131         FD_DIRTY_SAMPLE_MASK = BIT(5),
 132         FD_DIRTY_FRAMEBUFFER = BIT(6),
 133         FD_DIRTY_STIPPLE     = BIT(7),
 134         FD_DIRTY_VIEWPORT    = BIT(8),
 135         FD_DIRTY_VTXSTATE    = BIT(9),
 136         FD_DIRTY_VTXBUF      = BIT(10),
 137
 138         FD_DIRTY_SCISSOR     = BIT(12),
 139         FD_DIRTY_STREAMOUT   = BIT(13),
 140         FD_DIRTY_UCP         = BIT(14),
 141         FD_DIRTY_BLEND_DUAL  = BIT(15),
 142
 143         /* These are a bit redundent with fd_dirty_shader_state, and possibly
 144          * should be removed.  (But OTOH kinda convenient in some places)
 145          */
 146         FD_DIRTY_PROG        = BIT(16),
 147         FD_DIRTY_CONST       = BIT(17),
 148         FD_DIRTY_TEX         = BIT(18),
 149
 150         /* only used by a2xx.. possibly can be removed.. */
 151         FD_DIRTY_TEXSTATE    = BIT(19),
 152 };
 153
 154 /* per shader-stage dirty state: */
 155 enum fd_dirty_shader_state {
 156         FD_DIRTY_SHADER_PROG  = BIT(0),
 157         FD_DIRTY_SHADER_CONST = BIT(1),
 158         FD_DIRTY_SHADER_TEX   = BIT(2),
 159         FD_DIRTY_SHADER_SSBO  = BIT(3),
 160         FD_DIRTY_SHADER_IMAGE = BIT(4),
 161 };
 162
 163 struct fd_context {
 164         struct pipe_context base;
 165
 166         struct fd_device *dev;
 167         struct fd_screen *screen;
 168         struct fd_pipe *pipe;
 169
 170         struct util_queue flush_queue;
 171
 172         struct blitter_context *blitter;
 173         void *clear_rs_state;
 174         struct primconvert_context *primconvert;
 175
 176         /* slab for pipe_transfer allocations: */
 177         struct slab_child_pool transfer_pool;
 178
 179         /**
 180          * query related state:
 181          */
 182         /*@{*/
 183         /* slabs for fd_hw_sample and fd_hw_sample_period allocations: */
 184         struct slab_mempool sample_pool;
 185         struct slab_mempool sample_period_pool;
 186
 187         /* sample-providers for hw queries: */
 188         const struct fd_hw_sample_provider *hw_sample_providers[MAX_HW_SAMPLE_PROVIDERS];
 189
 190         /* list of active queries: */
 191         struct list_head hw_active_queries;
 192
 193         /* sample-providers for accumulating hw queries: */
 194         const struct fd_acc_sample_provider *acc_sample_providers[MAX_HW_SAMPLE_PROVIDERS];
 195
 196         /* list of active accumulating queries: */
 197         struct list_head acc_active_queries;
 198         /*@}*/
 199
 200         /* table with PIPE_PRIM_MAX entries mapping PIPE_PRIM_x to
 201          * DI_PT_x value to use for draw initiator.  There are some
 202          * slight differences between generation:
 203          */
 204         const uint8_t *primtypes;
 205         uint32_t primtype_mask;
 206
 207         /* shaders used by clear, and gmem->mem blits: */
 208         struct fd_program_stateobj solid_prog; // TODO move to screen?
 209
 210         /* shaders used by mem->gmem blits: */
 211         struct fd_program_stateobj blit_prog[MAX_RENDER_TARGETS]; // TODO move to screen?
 212         struct fd_program_stateobj blit_z, blit_zs;
 213
 214         /* Stats/counters:
 215          */
 216         struct {
 217                 uint64_t prims_emitted;
 218                 uint64_t prims_generated;
 219                 uint64_t draw_calls;
 220                 uint64_t batch_total, batch_sysmem, batch_gmem, batch_nondraw, batch_restore;
 221                 uint64_t staging_uploads, shadow_uploads;
 222                 uint64_t vs_regs, fs_regs;
 223         } stats;
 224
 225         /* Current batch.. the rule here is that you can deref ctx->batch
 226          * in codepaths from pipe_context entrypoints.  But not in code-
 227          * paths from fd_batch_flush() (basically, the stuff that gets
 228          * called from GMEM code), since in those code-paths the batch
 229          * you care about is not necessarily the same as ctx->batch.
 230          */
 231         struct fd_batch *batch;
 232
 233         /* NULL if there has been rendering since last flush.  Otherwise
 234          * keeps a reference to the last fence so we can re-use it rather
 235          * than having to flush no-op batch.
 236          */
 237         struct pipe_fence_handle *last_fence;
 238
 239         /* Are we in process of shadowing a resource? Used to detect recursion
 240          * in transfer_map, and skip unneeded synchronization.
 241          */
 242         bool in_shadow : 1;
 243
 244         /* Ie. in blit situation where we no longer care about previous framebuffer
 245          * contents.  Main point is to eliminate blits from fd_try_shadow_resource().
 246          * For example, in case of texture upload + gen-mipmaps.
 247          */
 248         bool in_blit : 1;
 249
 250         struct pipe_scissor_state scissor;
 251
 252         /* we don't have a disable/enable bit for scissor, so instead we keep
 253          * a disabled-scissor state which matches the entire bound framebuffer
 254          * and use that when scissor is not enabled.
 255          */
 256         struct pipe_scissor_state disabled_scissor;
 257
 258         /* Current gmem/tiling configuration.. gets updated on render_tiles()
 259          * if out of date with current maximal-scissor/cpp:
 260          *
 261          * (NOTE: this is kind of related to the batch, but moving it there
 262          * means we'd always have to recalc tiles ever batch)
 263          */
 264         struct fd_gmem_stateobj gmem;
 265         struct fd_vsc_pipe      vsc_pipe[32];
 266         struct fd_tile          tile[512];
 267
 268         /* which state objects need to be re-emit'd: */
 269         enum fd_dirty_3d_state dirty;
 270
 271         /* per shader-stage dirty status: */
 272         enum fd_dirty_shader_state dirty_shader[PIPE_SHADER_TYPES];
 273
 274         void *compute;
 275         struct pipe_blend_state *blend;
 276         struct pipe_rasterizer_state *rasterizer;
 277         struct pipe_depth_stencil_alpha_state *zsa;
 278
 279         struct fd_texture_stateobj tex[PIPE_SHADER_TYPES];
 280
 281         struct fd_program_stateobj prog;
 282
 283         struct fd_vertex_state vtx;
 284
 285         struct pipe_blend_color blend_color;
 286         struct pipe_stencil_ref stencil_ref;
 287         unsigned sample_mask;
 288         /* local context fb state, for when ctx->batch is null: */
 289         struct pipe_framebuffer_state framebuffer;
 290         struct pipe_poly_stipple stipple;
 291         struct pipe_viewport_state viewport;
 292         struct fd_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
 293         struct fd_shaderbuf_stateobj shaderbuf[PIPE_SHADER_TYPES];
 294         struct fd_shaderimg_stateobj shaderimg[PIPE_SHADER_TYPES];
 295         struct fd_streamout_stateobj streamout;
 296         struct fd_global_bindings_stateobj global_bindings;
 297         struct pipe_clip_state ucp;
 298
 299         struct pipe_query *cond_query;
 300         bool cond_cond; /* inverted rendering condition */
 301         uint cond_mode;
 302
 303         struct pipe_debug_callback debug;
 304
 305         /* GMEM/tile handling fxns: */
 306         void (*emit_tile_init)(struct fd_batch *batch);
 307         void (*emit_tile_prep)(struct fd_batch *batch, struct fd_tile *tile);
 308         void (*emit_tile_mem2gmem)(struct fd_batch *batch, struct fd_tile *tile);
 309         void (*emit_tile_renderprep)(struct fd_batch *batch, struct fd_tile *tile);
 310         void (*emit_tile_gmem2mem)(struct fd_batch *batch, struct fd_tile *tile);
 311         void (*emit_tile_fini)(struct fd_batch *batch);   /* optional */
 312
 313         /* optional, for GMEM bypass: */
 314         void (*emit_sysmem_prep)(struct fd_batch *batch);
 315         void (*emit_sysmem_fini)(struct fd_batch *batch);
 316
 317         /* draw: */
 318         bool (*draw_vbo)(struct fd_context *ctx, const struct pipe_draw_info *info,
 319                         unsigned index_offset);
 320         bool (*clear)(struct fd_context *ctx, unsigned buffers,
 321                         const union pipe_color_union *color, double depth, unsigned stencil);
 322
 323         /* compute: */
 324         void (*launch_grid)(struct fd_context *ctx, const struct pipe_grid_info *info);
 325
 326         /* constant emit:  (note currently not used/needed for a2xx) */
 327         void (*emit_const)(struct fd_ringbuffer *ring, gl_shader_stage type,
 328                         uint32_t regid, uint32_t offset, uint32_t sizedwords,
 329                         const uint32_t *dwords, struct pipe_resource *prsc);
 330         /* emit bo addresses as constant: */
 331         void (*emit_const_bo)(struct fd_ringbuffer *ring, gl_shader_stage type, boolean write,
 332                         uint32_t regid, uint32_t num, struct pipe_resource **prscs, uint32_t *offsets);
 333
 334         /* indirect-branch emit: */
 335         void (*emit_ib)(struct fd_ringbuffer *ring, struct fd_ringbuffer *target);
 336
 337         /* query: */
 338         struct fd_query * (*create_query)(struct fd_context *ctx, unsigned query_type);
 339         void (*query_prepare)(struct fd_batch *batch, uint32_t num_tiles);
 340         void (*query_prepare_tile)(struct fd_batch *batch, uint32_t n,
 341                         struct fd_ringbuffer *ring);
 342         void (*query_set_stage)(struct fd_batch *batch, enum fd_render_stage stage);
 343
 344         /* blitter: */
 345         void (*blit)(struct fd_context *ctx, const struct pipe_blit_info *info);
 346
 347         /* simple gpu "memcpy": */
 348         void (*mem_to_mem)(struct fd_ringbuffer *ring, struct pipe_resource *dst,
 349                         unsigned dst_off, struct pipe_resource *src, unsigned src_off,
 350                         unsigned sizedwords);
 351
 352         /*
 353          * Common pre-cooked VBO state (used for a3xx and later):
 354          */
 355
 356         /* for clear/gmem->mem vertices, and mem->gmem */
 357         struct pipe_resource *solid_vbuf;
 358
 359         /* for mem->gmem tex coords: */
 360         struct pipe_resource *blit_texcoord_vbuf;
 361
 362         /* vertex state for solid_vbuf:
 363          *    - solid_vbuf / 12 / R32G32B32_FLOAT
 364          */
 365         struct fd_vertex_state solid_vbuf_state;
 366
 367         /* vertex state for blit_prog:
 368          *    - blit_texcoord_vbuf / 8 / R32G32_FLOAT
 369          *    - solid_vbuf / 12 / R32G32B32_FLOAT
 370          */
 371         struct fd_vertex_state blit_vbuf_state;
 372 };
 373
 374 static inline struct fd_context *
 375 fd_context(struct pipe_context *pctx)
 376 {
 377         return (struct fd_context *)pctx;
 378 }
 379
 380 static inline void
 381 fd_context_assert_locked(struct fd_context *ctx)
 382 {
 383         pipe_mutex_assert_locked(ctx->screen->lock);
 384 }
 385
 386 static inline void
 387 fd_context_lock(struct fd_context *ctx)
 388 {
 389         mtx_lock(&ctx->screen->lock);
 390 }
 391
 392 static inline void
 393 fd_context_unlock(struct fd_context *ctx)
 394 {
 395         mtx_unlock(&ctx->screen->lock);
 396 }
 397
 398 /* mark all state dirty: */
 399 static inline void
 400 fd_context_all_dirty(struct fd_context *ctx)
 401 {
 402         ctx->dirty = ~0;
 403         for (unsigned i = 0; i < PIPE_SHADER_TYPES; i++)
 404                 ctx->dirty_shader[i] = ~0;
 405 }
 406
 407 static inline void
 408 fd_context_all_clean(struct fd_context *ctx)
 409 {
 410         ctx->dirty = 0;
 411         for (unsigned i = 0; i < PIPE_SHADER_TYPES; i++) {
 412                 /* don't mark compute state as clean, since it is not emitted
 413                  * during normal draw call.  The places that call _all_dirty(),
 414                  * it is safe to mark compute state dirty as well, but the
 415                  * inverse is not true.
 416                  */
 417                 if (i == PIPE_SHADER_COMPUTE)
 418                         continue;
 419                 ctx->dirty_shader[i] = 0;
 420         }
 421 }
 422
 423 static inline struct pipe_scissor_state *
 424 fd_context_get_scissor(struct fd_context *ctx)
 425 {
 426         if (ctx->rasterizer && ctx->rasterizer->scissor)
 427                 return &ctx->scissor;
 428         return &ctx->disabled_scissor;
 429 }
 430
 431 static inline bool
 432 fd_supported_prim(struct fd_context *ctx, unsigned prim)
 433 {
 434         return (1 << prim) & ctx->primtype_mask;
 435 }
 436
 437 static inline struct fd_batch *
 438 fd_context_batch(struct fd_context *ctx)
 439 {
 440         if (unlikely(!ctx->batch)) {
 441                 struct fd_batch *batch =
 442                         fd_batch_from_fb(&ctx->screen->batch_cache, ctx, &ctx->framebuffer);
 443                 util_copy_framebuffer_state(&batch->framebuffer, &ctx->framebuffer);
 444                 ctx->batch = batch;
 445                 fd_context_all_dirty(ctx);
 446         }
 447         return ctx->batch;
 448 }
 449
 450 static inline void
 451 fd_batch_set_stage(struct fd_batch *batch, enum fd_render_stage stage)
 452 {
 453         struct fd_context *ctx = batch->ctx;
 454
 455         /* special case: internal blits (like mipmap level generation)
 456          * go through normal draw path (via util_blitter_blit()).. but
 457          * we need to ignore the FD_STAGE_DRAW which will be set, so we
 458          * don't enable queries which should be paused during internal
 459          * blits:
 460          */
 461         if ((batch->stage == FD_STAGE_BLIT) &&
 462                         (stage != FD_STAGE_NULL))
 463                 return;
 464
 465         if (ctx->query_set_stage)
 466                 ctx->query_set_stage(batch, stage);
 467
 468         batch->stage = stage;
 469 }
 470
 471 void fd_context_setup_common_vbos(struct fd_context *ctx);
 472 void fd_context_cleanup_common_vbos(struct fd_context *ctx);
 473
 474 struct pipe_context * fd_context_init(struct fd_context *ctx,
 475                 struct pipe_screen *pscreen, const uint8_t *primtypes,
 476                 void *priv, unsigned flags);
 477
 478 void fd_context_destroy(struct pipe_context *pctx);
 479
 480 #endif /* FREEDRENO_CONTEXT_H_ */