src/gallium/drivers/swr/rasterizer/core/context.h

   1 /****************************************************************************
   2 * Copyright (C) 2014-2015 Intel Corporation.   All Rights Reserved.
   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
  20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21 * IN THE SOFTWARE.
  22 *
  23 * @file context.h
  24 *
  25 * @brief Definitions for SWR_CONTEXT and DRAW_CONTEXT
  26 *        The SWR_CONTEXT is our global context and contains the DC ring,
  27 *        thread state, etc.
  28 *
  29 *        The DRAW_CONTEXT contains all state associated with a draw operation.
  30 *
  31 ******************************************************************************/
  32 #pragma once
  33
  34 #include <condition_variable>
  35 #include <algorithm>
  36
  37 #include "core/api.h"
  38 #include "core/utils.h"
  39 #include "core/arena.h"
  40 #include "core/fifo.hpp"
  41 #include "core/knobs.h"
  42 #include "common/simdintrin.h"
  43 #include "core/threads.h"
  44 #include "ringbuffer.h"
  45
  46 // x.8 fixed point precision values
  47 #define FIXED_POINT_SHIFT 8
  48 #define FIXED_POINT_SCALE 256
  49
  50 // x.16 fixed point precision values
  51 #define FIXED_POINT16_SHIFT 16
  52 #define FIXED_POINT16_SCALE 65536
  53
  54 struct SWR_CONTEXT;
  55 struct DRAW_CONTEXT;
  56
  57 struct TRI_FLAGS
  58 {
  59     uint32_t frontFacing : 1;
  60     uint32_t yMajor : 1;
  61     uint32_t coverageMask : (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
  62     uint32_t reserved : 32 - 1 - 1 - (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
  63     float pointSize;
  64     uint32_t primID;
  65     uint32_t renderTargetArrayIndex;
  66 };
  67
  68 //////////////////////////////////////////////////////////////////////////
  69 /// SWR_TRIANGLE_DESC
  70 /////////////////////////////////////////////////////////////////////////
  71 struct SWR_TRIANGLE_DESC
  72 {
  73     float I[3];
  74     float J[3];
  75     float Z[3];
  76     float OneOverW[3];
  77     float recipDet;
  78
  79     float *pRecipW;
  80     float *pAttribs;
  81     float *pPerspAttribs;
  82     float *pSamplePos;
  83     float *pUserClipBuffer;
  84
  85     uint64_t coverageMask[SWR_MAX_NUM_MULTISAMPLES];
  86     uint64_t conservativeCoverageMask;
  87     uint64_t innerConservativeCoverageMask;
  88     uint64_t anyCoveredSamples;
  89
  90     TRI_FLAGS triFlags;
  91 };
  92
  93 struct TRIANGLE_WORK_DESC
  94 {
  95     float *pTriBuffer;
  96     float *pAttribs;
  97     float *pUserClipBuffer;
  98     uint32_t numAttribs;
  99     TRI_FLAGS triFlags;
 100 };
 101
 102 union CLEAR_FLAGS
 103 {
 104     struct
 105     {
 106         uint32_t mask : 3;
 107     };
 108     uint32_t bits;
 109 };
 110
 111 struct CLEAR_DESC
 112 {
 113     CLEAR_FLAGS flags;
 114     float clearRTColor[4];  // RGBA_32F
 115     float clearDepth;   // [0..1]
 116     uint8_t clearStencil;
 117 };
 118
 119 struct DISCARD_INVALIDATE_TILES_DESC
 120 {
 121     uint32_t attachmentMask;
 122     SWR_RECT rect;
 123     SWR_TILE_STATE newTileState;
 124     bool createNewTiles;
 125     bool fullTilesOnly;
 126 };
 127
 128 struct SYNC_DESC
 129 {
 130     PFN_CALLBACK_FUNC pfnCallbackFunc;
 131     uint64_t userData;
 132     uint64_t userData2;
 133     uint64_t userData3;
 134 };
 135
 136 struct QUERY_DESC
 137 {
 138     SWR_STATS* pStats;
 139 };
 140
 141 struct STORE_TILES_DESC
 142 {
 143     SWR_RENDERTARGET_ATTACHMENT attachment;
 144     SWR_TILE_STATE postStoreTileState;
 145 };
 146
 147 struct COMPUTE_DESC
 148 {
 149     uint32_t threadGroupCountX;
 150     uint32_t threadGroupCountY;
 151     uint32_t threadGroupCountZ;
 152 };
 153
 154 typedef void(*PFN_WORK_FUNC)(DRAW_CONTEXT* pDC, uint32_t workerId, uint32_t macroTile, void* pDesc);
 155
 156 enum WORK_TYPE
 157 {
 158     SYNC,
 159     DRAW,
 160     CLEAR,
 161     DISCARDINVALIDATETILES,
 162     STORETILES,
 163     QUERYSTATS,
 164 };
 165
 166 struct BE_WORK
 167 {
 168     WORK_TYPE type;
 169     PFN_WORK_FUNC pfnWork;
 170     union
 171     {
 172         SYNC_DESC sync;
 173         TRIANGLE_WORK_DESC tri;
 174         CLEAR_DESC clear;
 175         DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles;
 176         STORE_TILES_DESC storeTiles;
 177         QUERY_DESC queryStats;
 178     } desc;
 179 };
 180
 181 struct DRAW_WORK
 182 {
 183     DRAW_CONTEXT*   pDC;
 184     union
 185     {
 186         uint32_t   numIndices;      // DrawIndexed: Number of indices for draw.
 187         uint32_t   numVerts;        // Draw: Number of verts (triangles, lines, etc)
 188     };
 189     union
 190     {
 191         const int32_t* pIB;        // DrawIndexed: App supplied indices
 192         uint32_t   startVertex;    // Draw: Starting vertex in VB to render from.
 193     };
 194     int32_t    baseVertex;
 195     uint32_t   numInstances;        // Number of instances
 196     uint32_t   startInstance;       // Instance offset
 197     uint32_t   startPrimID;         // starting primitiveID for this draw batch
 198     uint32_t   startVertexID;       // starting VertexID for this draw batch (only needed for non-indexed draws)
 199     SWR_FORMAT type;                // index buffer type
 200 };
 201
 202 typedef void(*PFN_FE_WORK_FUNC)(SWR_CONTEXT* pContext, DRAW_CONTEXT* pDC, uint32_t workerId, void* pDesc);
 203 struct FE_WORK
 204 {
 205     WORK_TYPE type;
 206     PFN_FE_WORK_FUNC pfnWork;
 207     union
 208     {
 209         SYNC_DESC sync;
 210         DRAW_WORK draw;
 211         CLEAR_DESC clear;
 212         DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles;
 213         STORE_TILES_DESC storeTiles;
 214         QUERY_DESC queryStats;
 215     } desc;
 216 };
 217
 218 struct GUARDBAND
 219 {
 220     float left, right, top, bottom;
 221 };
 222
 223 struct PA_STATE;
 224
 225 // function signature for pipeline stages that execute after primitive assembly
 226 typedef void(*PFN_PROCESS_PRIMS)(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[],
 227     uint32_t primMask, simdscalari primID);
 228
 229 OSALIGNLINE(struct) API_STATE
 230 {
 231     // Vertex Buffers
 232     SWR_VERTEX_BUFFER_STATE vertexBuffers[KNOB_NUM_STREAMS];
 233
 234     // Index Buffer
 235     SWR_INDEX_BUFFER_STATE  indexBuffer;
 236
 237     // FS - Fetch Shader State
 238     PFN_FETCH_FUNC          pfnFetchFunc;
 239
 240     // VS - Vertex Shader State
 241     PFN_VERTEX_FUNC         pfnVertexFunc;
 242
 243     // GS - Geometry Shader State
 244     PFN_GS_FUNC             pfnGsFunc;
 245     SWR_GS_STATE            gsState;
 246
 247     // CS - Compute Shader
 248     PFN_CS_FUNC             pfnCsFunc;
 249     uint32_t                totalThreadsInGroup;
 250     uint32_t                totalSpillFillSize;
 251
 252     // FE - Frontend State
 253     SWR_FRONTEND_STATE      frontendState;
 254
 255     // SOS - Streamout Shader State
 256     PFN_SO_FUNC             pfnSoFunc[MAX_SO_STREAMS];
 257
 258     // Streamout state
 259     SWR_STREAMOUT_STATE     soState;
 260     mutable SWR_STREAMOUT_BUFFER soBuffer[MAX_SO_STREAMS];
 261
 262     // Tessellation State
 263     PFN_HS_FUNC             pfnHsFunc;
 264     PFN_DS_FUNC             pfnDsFunc;
 265     SWR_TS_STATE            tsState;
 266
 267     // Number of attributes used by the frontend (vs, so, gs)
 268     uint32_t                feNumAttributes;
 269
 270     PRIMITIVE_TOPOLOGY      topology;
 271     bool                    forceFront;
 272
 273     // RS - Rasterizer State
 274     SWR_RASTSTATE           rastState;
 275     // floating point multisample offsets
 276     float samplePos[SWR_MAX_NUM_MULTISAMPLES * 2];
 277
 278     GUARDBAND               gbState;
 279
 280     SWR_VIEWPORT            vp[KNOB_NUM_VIEWPORTS_SCISSORS];
 281     SWR_VIEWPORT_MATRIX     vpMatrix[KNOB_NUM_VIEWPORTS_SCISSORS];
 282
 283     BBOX                    scissorRects[KNOB_NUM_VIEWPORTS_SCISSORS];
 284     BBOX                    scissorInFixedPoint;
 285
 286     // Backend state
 287     SWR_BACKEND_STATE       backendState;
 288
 289     // PS - Pixel shader state
 290     SWR_PS_STATE            psState;
 291
 292     SWR_DEPTH_STENCIL_STATE depthStencilState;
 293
 294     // OM - Output Merger State
 295     SWR_BLEND_STATE         blendState;
 296     PFN_BLEND_JIT_FUNC      pfnBlendFunc[SWR_NUM_RENDERTARGETS];
 297
 298     // Stats are incremented when this is true.
 299     bool enableStats;
 300
 301     struct
 302     {
 303         uint32_t colorHottileEnable : 8;
 304         uint32_t depthHottileEnable: 1;
 305         uint32_t stencilHottileEnable : 1;
 306     };
 307
 308     PFN_QUANTIZE_DEPTH      pfnQuantizeDepth;
 309 };
 310
 311 class MacroTileMgr;
 312 class DispatchQueue;
 313
 314 struct RenderOutputBuffers
 315 {
 316     uint8_t* pColor[SWR_NUM_RENDERTARGETS];
 317     uint8_t* pDepth;
 318     uint8_t* pStencil;
 319 };
 320
 321 // Plane equation A/B/C coeffs used to evaluate I/J barycentric coords
 322 struct BarycentricCoeffs
 323 {
 324     simdscalar vIa;
 325     simdscalar vIb;
 326     simdscalar vIc;
 327
 328     simdscalar vJa;
 329     simdscalar vJb;
 330     simdscalar vJc;
 331
 332     simdscalar vZa;
 333     simdscalar vZb;
 334     simdscalar vZc;
 335
 336     simdscalar vRecipDet;
 337
 338     simdscalar vAOneOverW;
 339     simdscalar vBOneOverW;
 340     simdscalar vCOneOverW;
 341 };
 342
 343 // pipeline function pointer types
 344 typedef void(*PFN_BACKEND_FUNC)(DRAW_CONTEXT*, uint32_t, uint32_t, uint32_t, SWR_TRIANGLE_DESC&, RenderOutputBuffers&);
 345 typedef void(*PFN_OUTPUT_MERGER)(SWR_PS_CONTEXT &, uint8_t* (&)[SWR_NUM_RENDERTARGETS], uint32_t, const SWR_BLEND_STATE*,
 346                                  const PFN_BLEND_JIT_FUNC (&)[SWR_NUM_RENDERTARGETS], simdscalar&, simdscalar);
 347 typedef void(*PFN_CALC_PIXEL_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &);
 348 typedef void(*PFN_CALC_SAMPLE_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT&);
 349 typedef void(*PFN_CALC_CENTROID_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &, const uint64_t *const, const uint32_t,
 350                                               const simdscalar, const simdscalar);
 351
 352 struct BACKEND_FUNCS
 353 {
 354     PFN_BACKEND_FUNC pfnBackend;
 355 };
 356
 357 // Draw State
 358 struct DRAW_STATE
 359 {
 360     API_STATE state;
 361
 362     void* pPrivateState;  // Its required the driver sets this up for each draw.
 363
 364     // pipeline function pointers, filled in by API thread when setting up the draw
 365     BACKEND_FUNCS backendFuncs;
 366     PFN_PROCESS_PRIMS pfnProcessPrims;
 367
 368     CachingArena* pArena;     // This should only be used by API thread.
 369 };
 370
 371 // Draw Context
 372 //    The api thread sets up a draw context that exists for the life of the draw.
 373 //    This draw context maintains all of the state needed for the draw operation.
 374 struct DRAW_CONTEXT
 375 {
 376     SWR_CONTEXT*    pContext;
 377     union
 378     {
 379         MacroTileMgr*   pTileMgr;
 380         DispatchQueue*  pDispatch;      // Queue for thread groups. (isCompute)
 381     };
 382     DRAW_STATE*     pState;
 383     CachingArena*   pArena;
 384
 385     uint32_t        drawId;
 386     bool            dependent;
 387     bool            isCompute;      // Is this DC a compute context?
 388     bool            cleanupState;   // True if this is the last draw using an entry in the state ring.
 389     volatile bool   doneFE;         // Is FE work done for this draw?
 390
 391     FE_WORK         FeWork;
 392
 393     volatile OSALIGNLINE(uint32_t)   FeLock;
 394     volatile int64_t    threadsDone;
 395
 396     SYNC_DESC       retireCallback; // Call this func when this DC is retired.
 397 };
 398
 399 static_assert((sizeof(DRAW_CONTEXT) & 63) == 0, "Invalid size for DRAW_CONTEXT");
 400
 401 INLINE const API_STATE& GetApiState(const DRAW_CONTEXT* pDC)
 402 {
 403     SWR_ASSERT(pDC != nullptr);
 404     SWR_ASSERT(pDC->pState != nullptr);
 405
 406     return pDC->pState->state;
 407 }
 408
 409 INLINE void* GetPrivateState(const DRAW_CONTEXT* pDC)
 410 {
 411     SWR_ASSERT(pDC != nullptr);
 412     SWR_ASSERT(pDC->pState != nullptr);
 413
 414     return pDC->pState->pPrivateState;
 415 }
 416
 417 class HotTileMgr;
 418
 419 struct SWR_CONTEXT
 420 {
 421     // Draw Context Ring
 422     //  Each draw needs its own state in order to support mulitple draws in flight across multiple threads.
 423     //  We maintain N draw contexts configured as a ring. The size of the ring limits the maximum number
 424     //  of draws that can be in flight at any given time.
 425     //
 426     //  Description:
 427     //  1. State - When an application first sets state we'll request a new draw context to use.
 428     //     a. If there are no available draw contexts then we'll have to wait until one becomes free.
 429     //     b. If one is available then set pCurDrawContext to point to it and mark it in use.
 430     //     c. All state calls set state on pCurDrawContext.
 431     //  2. Draw - Creates submits a work item that is associated with current draw context.
 432     //     a. Set pPrevDrawContext = pCurDrawContext
 433     //     b. Set pCurDrawContext to NULL.
 434     //  3. State - When an applications sets state after draw
 435     //     a. Same as step 1.
 436     //     b. State is copied from prev draw context to current.
 437     RingBuffer<DRAW_CONTEXT> dcRing;
 438
 439     DRAW_CONTEXT *pCurDrawContext;    // This points to DC entry in ring for an unsubmitted draw.
 440     DRAW_CONTEXT *pPrevDrawContext;   // This points to DC entry for the previous context submitted that we can copy state from.
 441
 442     MacroTileMgr* pMacroTileManagerArray;
 443     DispatchQueue* pDispatchQueueArray;
 444
 445     // Draw State Ring
 446     //  When draw are very large (lots of primitives) then the API thread will break these up.
 447     //  These split draws all have identical state. So instead of storing the state directly
 448     //  in the Draw Context (DC) we instead store it in a Draw State (DS). This allows multiple DCs
 449     //  to reference a single entry in the DS ring.
 450     RingBuffer<DRAW_STATE> dsRing;
 451
 452     uint32_t curStateId;               // Current index to the next available entry in the DS ring.
 453
 454     uint32_t NumWorkerThreads;
 455     uint32_t NumFEThreads;
 456     uint32_t NumBEThreads;
 457
 458     THREAD_POOL threadPool; // Thread pool associated with this context
 459
 460     std::condition_variable FifosNotEmpty;
 461     std::mutex WaitLock;
 462
 463     DRIVER_TYPE driverType;
 464
 465     uint32_t privateStateSize;
 466
 467     HotTileMgr *pHotTileMgr;
 468
 469     // tile load/store functions, passed in at create context time
 470     PFN_LOAD_TILE pfnLoadTile;
 471     PFN_STORE_TILE pfnStoreTile;
 472     PFN_CLEAR_TILE pfnClearTile;
 473
 474     // Global Stats
 475     SWR_STATS stats[KNOB_MAX_NUM_THREADS];
 476
 477     // Scratch space for workers.
 478     uint8_t* pScratch[KNOB_MAX_NUM_THREADS];
 479
 480     CachingAllocator cachingArenaAllocator;
 481     uint32_t frameCount;
 482 };
 483
 484 void WaitForDependencies(SWR_CONTEXT *pContext, uint64_t drawId);
 485 void WakeAllThreads(SWR_CONTEXT *pContext);
 486
 487 #define UPDATE_STAT(name, count) if (GetApiState(pDC).enableStats) { pContext->stats[workerId].name += count; }
 488 #define SET_STAT(name, count) if (GetApiState(pDC).enableStats) { pContext->stats[workerId].name = count; }