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