src/gallium/drivers/freedreno/freedreno_context.h

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29 #ifndef FREEDRENO_CONTEXT_H_
  30 #define FREEDRENO_CONTEXT_H_
  31
  32 #include "pipe/p_context.h"
  33 #include "indices/u_primconvert.h"
  34 #include "util/u_blitter.h"
  35 #include "util/list.h"
  36 #include "util/slab.h"
  37 #include "util/u_string.h"
  38
  39 #include "freedreno_batch.h"
  40 #include "freedreno_screen.h"
  41 #include "freedreno_gmem.h"
  42 #include "freedreno_util.h"
  43
  44 #define BORDER_COLOR_UPLOAD_SIZE (2 * PIPE_MAX_SAMPLERS * BORDERCOLOR_SIZE)
  45
  46 struct fd_vertex_stateobj;
  47
  48 struct fd_texture_stateobj {
  49         struct pipe_sampler_view *textures[PIPE_MAX_SAMPLERS];
  50         unsigned num_textures;
  51         unsigned valid_textures;
  52         struct pipe_sampler_state *samplers[PIPE_MAX_SAMPLERS];
  53         unsigned num_samplers;
  54         unsigned valid_samplers;
  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         uint32_t dirty_mask;
  73 };
  74
  75 struct fd_vertexbuf_stateobj {
  76         struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
  77         unsigned count;
  78         uint32_t enabled_mask;
  79         uint32_t dirty_mask;
  80 };
  81
  82 struct fd_vertex_stateobj {
  83         struct pipe_vertex_element pipe[PIPE_MAX_ATTRIBS];
  84         unsigned num_elements;
  85 };
  86
  87 struct fd_streamout_stateobj {
  88         struct pipe_stream_output_target *targets[PIPE_MAX_SO_BUFFERS];
  89         unsigned num_targets;
  90         /* Track offset from vtxcnt for streamout data.  This counter
  91          * is just incremented by # of vertices on each draw until
  92          * reset or new streamout buffer bound.
  93          *
  94          * When we eventually have GS, the CPU won't actually know the
  95          * number of vertices per draw, so I think we'll have to do
  96          * something more clever.
  97          */
  98         unsigned offsets[PIPE_MAX_SO_BUFFERS];
  99 };
 100
 101 /* group together the vertex and vertexbuf state.. for ease of passing
 102  * around, and because various internal operations (gmem<->mem, etc)
 103  * need their own vertex state:
 104  */
 105 struct fd_vertex_state {
 106         struct fd_vertex_stateobj *vtx;
 107         struct fd_vertexbuf_stateobj vertexbuf;
 108 };
 109
 110 /* global 3d pipeline dirty state: */
 111 enum fd_dirty_3d_state {
 112         FD_DIRTY_BLEND       = BIT(0),
 113         FD_DIRTY_RASTERIZER  = BIT(1),
 114         FD_DIRTY_ZSA         = BIT(2),
 115         FD_DIRTY_BLEND_COLOR = BIT(3),
 116         FD_DIRTY_STENCIL_REF = BIT(4),
 117         FD_DIRTY_SAMPLE_MASK = BIT(5),
 118         FD_DIRTY_FRAMEBUFFER = BIT(6),
 119         FD_DIRTY_STIPPLE     = BIT(7),
 120         FD_DIRTY_VIEWPORT    = BIT(8),
 121         FD_DIRTY_VTXSTATE    = BIT(9),
 122         FD_DIRTY_VTXBUF      = BIT(10),
 123         FD_DIRTY_INDEXBUF    = BIT(11),
 124         FD_DIRTY_SCISSOR     = BIT(12),
 125         FD_DIRTY_STREAMOUT   = BIT(13),
 126         FD_DIRTY_UCP         = BIT(14),
 127         FD_DIRTY_BLEND_DUAL  = BIT(15),
 128
 129         /* These are a bit redundent with fd_dirty_shader_state, and possibly
 130          * should be removed.  (But OTOH kinda convenient in some places)
 131          */
 132         FD_DIRTY_PROG        = BIT(16),
 133         FD_DIRTY_CONST       = BIT(17),
 134         FD_DIRTY_TEX         = BIT(18),
 135
 136         /* only used by a2xx.. possibly can be removed.. */
 137         FD_DIRTY_TEXSTATE    = BIT(19),
 138 };
 139
 140 /* per shader-stage dirty state: */
 141 enum fd_dirty_shader_state {
 142         FD_DIRTY_SHADER_PROG  = BIT(0),
 143         FD_DIRTY_SHADER_CONST = BIT(1),
 144         FD_DIRTY_SHADER_TEX   = BIT(2),
 145 };
 146
 147 struct fd_context {
 148         struct pipe_context base;
 149
 150         struct fd_device *dev;
 151         struct fd_screen *screen;
 152
 153         struct util_queue flush_queue;
 154
 155         struct blitter_context *blitter;
 156         void *clear_rs_state;
 157         struct primconvert_context *primconvert;
 158
 159         /* slab for pipe_transfer allocations: */
 160         struct slab_child_pool transfer_pool;
 161
 162         /* slabs for fd_hw_sample and fd_hw_sample_period allocations: */
 163         struct slab_mempool sample_pool;
 164         struct slab_mempool sample_period_pool;
 165
 166         /* sample-providers for hw queries: */
 167         const struct fd_hw_sample_provider *sample_providers[MAX_HW_SAMPLE_PROVIDERS];
 168
 169         /* list of active queries: */
 170         struct list_head active_queries;
 171
 172         /* table with PIPE_PRIM_MAX entries mapping PIPE_PRIM_x to
 173          * DI_PT_x value to use for draw initiator.  There are some
 174          * slight differences between generation:
 175          */
 176         const uint8_t *primtypes;
 177         uint32_t primtype_mask;
 178
 179         /* shaders used by clear, and gmem->mem blits: */
 180         struct fd_program_stateobj solid_prog; // TODO move to screen?
 181
 182         /* shaders used by mem->gmem blits: */
 183         struct fd_program_stateobj blit_prog[MAX_RENDER_TARGETS]; // TODO move to screen?
 184         struct fd_program_stateobj blit_z, blit_zs;
 185
 186         /* Stats/counters:
 187          */
 188         struct {
 189                 uint64_t prims_emitted;
 190                 uint64_t prims_generated;
 191                 uint64_t draw_calls;
 192                 uint64_t batch_total, batch_sysmem, batch_gmem, batch_restore;
 193         } stats;
 194
 195         /* Current batch.. the rule here is that you can deref ctx->batch
 196          * in codepaths from pipe_context entrypoints.  But not in code-
 197          * paths from fd_batch_flush() (basically, the stuff that gets
 198          * called from GMEM code), since in those code-paths the batch
 199          * you care about is not necessarily the same as ctx->batch.
 200          */
 201         struct fd_batch *batch;
 202
 203         struct pipe_fence_handle *last_fence;
 204
 205         /* Are we in process of shadowing a resource? Used to detect recursion
 206          * in transfer_map, and skip unneeded synchronization.
 207          */
 208         bool in_shadow : 1;
 209
 210         /* Ie. in blit situation where we no longer care about previous framebuffer
 211          * contents.  Main point is to eliminate blits from fd_try_shadow_resource().
 212          * For example, in case of texture upload + gen-mipmaps.
 213          */
 214         bool in_blit : 1;
 215
 216         struct pipe_scissor_state scissor;
 217
 218         /* we don't have a disable/enable bit for scissor, so instead we keep
 219          * a disabled-scissor state which matches the entire bound framebuffer
 220          * and use that when scissor is not enabled.
 221          */
 222         struct pipe_scissor_state disabled_scissor;
 223
 224         /* Current gmem/tiling configuration.. gets updated on render_tiles()
 225          * if out of date with current maximal-scissor/cpp:
 226          *
 227          * (NOTE: this is kind of related to the batch, but moving it there
 228          * means we'd always have to recalc tiles ever batch)
 229          */
 230         struct fd_gmem_stateobj gmem;
 231         struct fd_vsc_pipe      pipe[8];
 232         struct fd_tile          tile[512];
 233
 234         /* which state objects need to be re-emit'd: */
 235         enum fd_dirty_3d_state dirty;
 236
 237         /* per shader-stage dirty status: */
 238         enum fd_dirty_shader_state dirty_shader[PIPE_SHADER_TYPES];
 239
 240         struct pipe_blend_state *blend;
 241         struct pipe_rasterizer_state *rasterizer;
 242         struct pipe_depth_stencil_alpha_state *zsa;
 243
 244         struct fd_texture_stateobj tex[PIPE_SHADER_TYPES];
 245
 246         struct fd_program_stateobj prog;
 247
 248         struct fd_vertex_state vtx;
 249
 250         struct pipe_blend_color blend_color;
 251         struct pipe_stencil_ref stencil_ref;
 252         unsigned sample_mask;
 253         struct pipe_poly_stipple stipple;
 254         struct pipe_viewport_state viewport;
 255         struct fd_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
 256         struct pipe_index_buffer indexbuf;
 257         struct fd_streamout_stateobj streamout;
 258         struct pipe_clip_state ucp;
 259
 260         struct pipe_query *cond_query;
 261         bool cond_cond; /* inverted rendering condition */
 262         uint cond_mode;
 263
 264         struct pipe_debug_callback debug;
 265
 266         /* GMEM/tile handling fxns: */
 267         void (*emit_tile_init)(struct fd_batch *batch);
 268         void (*emit_tile_prep)(struct fd_batch *batch, struct fd_tile *tile);
 269         void (*emit_tile_mem2gmem)(struct fd_batch *batch, struct fd_tile *tile);
 270         void (*emit_tile_renderprep)(struct fd_batch *batch, struct fd_tile *tile);
 271         void (*emit_tile_gmem2mem)(struct fd_batch *batch, struct fd_tile *tile);
 272         void (*emit_tile_fini)(struct fd_batch *batch);   /* optional */
 273
 274         /* optional, for GMEM bypass: */
 275         void (*emit_sysmem_prep)(struct fd_batch *batch);
 276         void (*emit_sysmem_fini)(struct fd_batch *batch);
 277
 278         /* draw: */
 279         bool (*draw_vbo)(struct fd_context *ctx, const struct pipe_draw_info *info);
 280         void (*clear)(struct fd_context *ctx, unsigned buffers,
 281                         const union pipe_color_union *color, double depth, unsigned stencil);
 282
 283         /* constant emit:  (note currently not used/needed for a2xx) */
 284         void (*emit_const)(struct fd_ringbuffer *ring, enum shader_t type,
 285                         uint32_t regid, uint32_t offset, uint32_t sizedwords,
 286                         const uint32_t *dwords, struct pipe_resource *prsc);
 287         /* emit bo addresses as constant: */
 288         void (*emit_const_bo)(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
 289                         uint32_t regid, uint32_t num, struct pipe_resource **prscs, uint32_t *offsets);
 290
 291         /* indirect-branch emit: */
 292         void (*emit_ib)(struct fd_ringbuffer *ring, struct fd_ringbuffer *target);
 293
 294         /*
 295          * Common pre-cooked VBO state (used for a3xx and later):
 296          */
 297
 298         /* for clear/gmem->mem vertices, and mem->gmem */
 299         struct pipe_resource *solid_vbuf;
 300
 301         /* for mem->gmem tex coords: */
 302         struct pipe_resource *blit_texcoord_vbuf;
 303
 304         /* vertex state for solid_vbuf:
 305          *    - solid_vbuf / 12 / R32G32B32_FLOAT
 306          */
 307         struct fd_vertex_state solid_vbuf_state;
 308
 309         /* vertex state for blit_prog:
 310          *    - blit_texcoord_vbuf / 8 / R32G32_FLOAT
 311          *    - solid_vbuf / 12 / R32G32B32_FLOAT
 312          */
 313         struct fd_vertex_state blit_vbuf_state;
 314 };
 315
 316 static inline struct fd_context *
 317 fd_context(struct pipe_context *pctx)
 318 {
 319         return (struct fd_context *)pctx;
 320 }
 321
 322 static inline void
 323 fd_context_assert_locked(struct fd_context *ctx)
 324 {
 325         pipe_mutex_assert_locked(ctx->screen->lock);
 326 }
 327
 328 static inline void
 329 fd_context_lock(struct fd_context *ctx)
 330 {
 331         mtx_lock(&ctx->screen->lock);
 332 }
 333
 334 static inline void
 335 fd_context_unlock(struct fd_context *ctx)
 336 {
 337         mtx_unlock(&ctx->screen->lock);
 338 }
 339
 340 /* mark all state dirty: */
 341 static inline void
 342 fd_context_all_dirty(struct fd_context *ctx)
 343 {
 344         ctx->dirty = ~0;
 345         for (unsigned i = 0; i < PIPE_SHADER_TYPES; i++)
 346                 ctx->dirty_shader[i] = ~0;
 347 }
 348
 349 static inline void
 350 fd_context_all_clean(struct fd_context *ctx)
 351 {
 352         ctx->dirty = 0;
 353         for (unsigned i = 0; i < PIPE_SHADER_TYPES; i++)
 354                 ctx->dirty_shader[i] = 0;
 355 }
 356
 357 static inline struct pipe_scissor_state *
 358 fd_context_get_scissor(struct fd_context *ctx)
 359 {
 360         if (ctx->rasterizer && ctx->rasterizer->scissor)
 361                 return &ctx->scissor;
 362         return &ctx->disabled_scissor;
 363 }
 364
 365 static inline bool
 366 fd_supported_prim(struct fd_context *ctx, unsigned prim)
 367 {
 368         return (1 << prim) & ctx->primtype_mask;
 369 }
 370
 371 void fd_context_setup_common_vbos(struct fd_context *ctx);
 372 void fd_context_cleanup_common_vbos(struct fd_context *ctx);
 373
 374 struct pipe_context * fd_context_init(struct fd_context *ctx,
 375                 struct pipe_screen *pscreen, const uint8_t *primtypes,
 376                 void *priv);
 377
 378 void fd_context_destroy(struct pipe_context *pctx);
 379
 380 #endif /* FREEDRENO_CONTEXT_H_ */