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/u_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
 111 struct fd_context {
 112         struct pipe_context base;
 113
 114         struct fd_device *dev;
 115         struct fd_screen *screen;
 116
 117         struct blitter_context *blitter;
 118         struct primconvert_context *primconvert;
 119
 120         /* slab for pipe_transfer allocations: */
 121         struct util_slab_mempool transfer_pool;
 122
 123         /* slabs for fd_hw_sample and fd_hw_sample_period allocations: */
 124         struct util_slab_mempool sample_pool;
 125         struct util_slab_mempool sample_period_pool;
 126
 127         /* sample-providers for hw queries: */
 128         const struct fd_hw_sample_provider *sample_providers[MAX_HW_SAMPLE_PROVIDERS];
 129
 130         /* list of active queries: */
 131         struct list_head active_queries;
 132
 133         /* table with PIPE_PRIM_MAX entries mapping PIPE_PRIM_x to
 134          * DI_PT_x value to use for draw initiator.  There are some
 135          * slight differences between generation:
 136          */
 137         const uint8_t *primtypes;
 138         uint32_t primtype_mask;
 139
 140         /* shaders used by clear, and gmem->mem blits: */
 141         struct fd_program_stateobj solid_prog; // TODO move to screen?
 142
 143         /* shaders used by mem->gmem blits: */
 144         struct fd_program_stateobj blit_prog[MAX_RENDER_TARGETS]; // TODO move to screen?
 145         struct fd_program_stateobj blit_z, blit_zs;
 146
 147         /* Stats/counters:
 148          */
 149         struct {
 150                 uint64_t prims_emitted;
 151                 uint64_t prims_generated;
 152                 uint64_t draw_calls;
 153                 uint64_t batch_total, batch_sysmem, batch_gmem, batch_restore;
 154         } stats;
 155
 156         /* Current batch.. the rule here is that you can deref ctx->batch
 157          * in codepaths from pipe_context entrypoints.  But not in code-
 158          * paths from fd_batch_flush() (basically, the stuff that gets
 159          * called from GMEM code), since in those code-paths the batch
 160          * you care about is not necessarily the same as ctx->batch.
 161          */
 162         struct fd_batch *batch;
 163
 164         /* Keep track if WAIT_FOR_IDLE is needed for registers we need
 165          * to update via RMW:
 166          */
 167         bool needs_wfi;
 168
 169         /* Do we need to re-emit RB_FRAME_BUFFER_DIMENSION?  At least on a3xx
 170          * it is not a banked context register, so it needs a WFI to update.
 171          * Keep track if it has actually changed, to avoid unneeded WFI.
 172          * */
 173         bool needs_rb_fbd;
 174
 175         struct pipe_scissor_state scissor;
 176
 177         /* we don't have a disable/enable bit for scissor, so instead we keep
 178          * a disabled-scissor state which matches the entire bound framebuffer
 179          * and use that when scissor is not enabled.
 180          */
 181         struct pipe_scissor_state disabled_scissor;
 182
 183         /* Current gmem/tiling configuration.. gets updated on render_tiles()
 184          * if out of date with current maximal-scissor/cpp:
 185          *
 186          * (NOTE: this is kind of related to the batch, but moving it there
 187          * means we'd always have to recalc tiles ever batch)
 188          */
 189         struct fd_gmem_stateobj gmem;
 190         struct fd_vsc_pipe      pipe[8];
 191         struct fd_tile          tile[512];
 192
 193         /* which state objects need to be re-emit'd: */
 194         enum {
 195                 FD_DIRTY_BLEND       = (1 <<  0),
 196                 FD_DIRTY_RASTERIZER  = (1 <<  1),
 197                 FD_DIRTY_ZSA         = (1 <<  2),
 198                 FD_DIRTY_FRAGTEX     = (1 <<  3),
 199                 FD_DIRTY_VERTTEX     = (1 <<  4),
 200                 FD_DIRTY_TEXSTATE    = (1 <<  5),
 201
 202                 FD_SHADER_DIRTY_VP   = (1 <<  6),
 203                 FD_SHADER_DIRTY_FP   = (1 <<  7),
 204                 /* skip geom/tcs/tes/compute */
 205                 FD_DIRTY_PROG        = FD_SHADER_DIRTY_FP | FD_SHADER_DIRTY_VP,
 206
 207                 FD_DIRTY_BLEND_COLOR = (1 << 12),
 208                 FD_DIRTY_STENCIL_REF = (1 << 13),
 209                 FD_DIRTY_SAMPLE_MASK = (1 << 14),
 210                 FD_DIRTY_FRAMEBUFFER = (1 << 15),
 211                 FD_DIRTY_STIPPLE     = (1 << 16),
 212                 FD_DIRTY_VIEWPORT    = (1 << 17),
 213                 FD_DIRTY_CONSTBUF    = (1 << 18),
 214                 FD_DIRTY_VTXSTATE    = (1 << 19),
 215                 FD_DIRTY_VTXBUF      = (1 << 20),
 216                 FD_DIRTY_INDEXBUF    = (1 << 21),
 217                 FD_DIRTY_SCISSOR     = (1 << 22),
 218                 FD_DIRTY_STREAMOUT   = (1 << 23),
 219                 FD_DIRTY_UCP         = (1 << 24),
 220                 FD_DIRTY_BLEND_DUAL  = (1 << 25),
 221         } dirty;
 222
 223         struct pipe_blend_state *blend;
 224         struct pipe_rasterizer_state *rasterizer;
 225         struct pipe_depth_stencil_alpha_state *zsa;
 226
 227         struct fd_texture_stateobj verttex, fragtex;
 228
 229         struct fd_program_stateobj prog;
 230
 231         struct fd_vertex_state vtx;
 232
 233         struct pipe_blend_color blend_color;
 234         struct pipe_stencil_ref stencil_ref;
 235         unsigned sample_mask;
 236         struct pipe_poly_stipple stipple;
 237         struct pipe_viewport_state viewport;
 238         struct fd_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
 239         struct pipe_index_buffer indexbuf;
 240         struct fd_streamout_stateobj streamout;
 241         struct pipe_clip_state ucp;
 242
 243         struct pipe_query *cond_query;
 244         bool cond_cond; /* inverted rendering condition */
 245         uint cond_mode;
 246
 247         /* Are we in process of shadowing a resource? Used to detect recursion
 248          * in transfer_map, and skip unneeded synchronization.
 249          */
 250         bool in_shadow;
 251
 252         /* Ie. in blit situation where we no longer care about previous framebuffer
 253          * contents.  Main point is to eliminate blits from fd_try_shadow_resource().
 254          * For example, in case of texture upload + gen-mipmaps.
 255          */
 256         bool discard;
 257
 258         struct pipe_debug_callback debug;
 259
 260         /* GMEM/tile handling fxns: */
 261         void (*emit_tile_init)(struct fd_batch *batch);
 262         void (*emit_tile_prep)(struct fd_batch *batch, struct fd_tile *tile);
 263         void (*emit_tile_mem2gmem)(struct fd_batch *batch, struct fd_tile *tile);
 264         void (*emit_tile_renderprep)(struct fd_batch *batch, struct fd_tile *tile);
 265         void (*emit_tile_gmem2mem)(struct fd_batch *batch, struct fd_tile *tile);
 266
 267         /* optional, for GMEM bypass: */
 268         void (*emit_sysmem_prep)(struct fd_batch *batch);
 269
 270         /* draw: */
 271         bool (*draw_vbo)(struct fd_context *ctx, const struct pipe_draw_info *info);
 272         void (*clear)(struct fd_context *ctx, unsigned buffers,
 273                         const union pipe_color_union *color, double depth, unsigned stencil);
 274
 275         /* constant emit:  (note currently not used/needed for a2xx) */
 276         void (*emit_const)(struct fd_ringbuffer *ring, enum shader_t type,
 277                         uint32_t regid, uint32_t offset, uint32_t sizedwords,
 278                         const uint32_t *dwords, struct pipe_resource *prsc);
 279         /* emit bo addresses as constant: */
 280         void (*emit_const_bo)(struct fd_ringbuffer *ring, enum shader_t type, boolean write,
 281                         uint32_t regid, uint32_t num, struct pipe_resource **prscs, uint32_t *offsets);
 282
 283         /* indirect-branch emit: */
 284         void (*emit_ib)(struct fd_ringbuffer *ring, struct fd_ringbuffer *target);
 285 };
 286
 287 static inline struct fd_context *
 288 fd_context(struct pipe_context *pctx)
 289 {
 290         return (struct fd_context *)pctx;
 291 }
 292
 293 static inline struct pipe_scissor_state *
 294 fd_context_get_scissor(struct fd_context *ctx)
 295 {
 296         if (ctx->rasterizer && ctx->rasterizer->scissor)
 297                 return &ctx->scissor;
 298         return &ctx->disabled_scissor;
 299 }
 300
 301 static inline bool
 302 fd_supported_prim(struct fd_context *ctx, unsigned prim)
 303 {
 304         return (1 << prim) & ctx->primtype_mask;
 305 }
 306
 307 static inline void
 308 fd_reset_wfi(struct fd_context *ctx)
 309 {
 310         ctx->needs_wfi = true;
 311 }
 312
 313 /* emit a WAIT_FOR_IDLE only if needed, ie. if there has not already
 314  * been one since last draw:
 315  */
 316 static inline void
 317 fd_wfi(struct fd_context *ctx, struct fd_ringbuffer *ring)
 318 {
 319         if (ctx->needs_wfi) {
 320                 OUT_WFI(ring);
 321                 ctx->needs_wfi = false;
 322         }
 323 }
 324
 325 /* emit a CP_EVENT_WRITE:
 326  */
 327 static inline void
 328 fd_event_write(struct fd_context *ctx, struct fd_ringbuffer *ring,
 329                 enum vgt_event_type evt)
 330 {
 331         OUT_PKT3(ring, CP_EVENT_WRITE, 1);
 332         OUT_RING(ring, evt);
 333         fd_reset_wfi(ctx);
 334 }
 335
 336 struct pipe_context * fd_context_init(struct fd_context *ctx,
 337                 struct pipe_screen *pscreen, const uint8_t *primtypes,
 338                 void *priv);
 339
 340 void fd_context_destroy(struct pipe_context *pctx);
 341
 342 #endif /* FREEDRENO_CONTEXT_H_ */