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

   1 /****************************************************************************
   2 * Copyright (C) 2014-2016 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 #include "archrast/archrast.h"
  46
  47 // x.8 fixed point precision values
  48 #define FIXED_POINT_SHIFT 8
  49 #define FIXED_POINT_SCALE 256
  50
  51 // x.16 fixed point precision values
  52 #define FIXED_POINT16_SHIFT 16
  53 #define FIXED_POINT16_SCALE 65536
  54
  55 struct SWR_CONTEXT;
  56 struct DRAW_CONTEXT;
  57
  58 struct TRI_FLAGS
  59 {
  60     uint32_t frontFacing : 1;
  61     uint32_t yMajor : 1;
  62     uint32_t coverageMask : (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
  63     uint32_t reserved : 32 - 1 - 1 - (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM);
  64     float pointSize;
  65     uint32_t primID;
  66     uint32_t renderTargetArrayIndex;
  67     uint32_t viewportIndex;
  68 };
  69
  70 //////////////////////////////////////////////////////////////////////////
  71 /// SWR_TRIANGLE_DESC
  72 /////////////////////////////////////////////////////////////////////////
  73 struct SWR_TRIANGLE_DESC
  74 {
  75     float I[3];
  76     float J[3];
  77     float Z[3];
  78     float OneOverW[3];
  79     float recipDet;
  80
  81     float *pRecipW;
  82     float *pAttribs;
  83     float *pPerspAttribs;
  84     float *pSamplePos;
  85     float *pUserClipBuffer;
  86
  87     uint64_t coverageMask[SWR_MAX_NUM_MULTISAMPLES];
  88     uint64_t innerCoverageMask; // Conservative rasterization inner coverage: marked covered if entire pixel is covered
  89     uint64_t anyCoveredSamples;
  90
  91     TRI_FLAGS triFlags;
  92 };
  93
  94 struct TRIANGLE_WORK_DESC
  95 {
  96     float *pTriBuffer;
  97     float *pAttribs;
  98     float *pUserClipBuffer;
  99     uint32_t numAttribs;
 100     TRI_FLAGS triFlags;
 101 };
 102
 103 struct CLEAR_DESC
 104 {
 105     SWR_RECT rect;
 106     uint32_t attachmentMask;
 107     uint32_t renderTargetArrayIndex;
 108     float clearRTColor[4];  // RGBA_32F
 109     float clearDepth;   // [0..1]
 110     uint8_t clearStencil;
 111 };
 112
 113 struct DISCARD_INVALIDATE_TILES_DESC
 114 {
 115     uint32_t attachmentMask;
 116     SWR_RECT rect;
 117     SWR_TILE_STATE newTileState;
 118     bool createNewTiles;
 119     bool fullTilesOnly;
 120 };
 121
 122 struct SYNC_DESC
 123 {
 124     PFN_CALLBACK_FUNC pfnCallbackFunc;
 125     uint64_t userData;
 126     uint64_t userData2;
 127     uint64_t userData3;
 128 };
 129
 130 struct STORE_TILES_DESC
 131 {
 132     uint32_t attachmentMask;
 133     SWR_TILE_STATE postStoreTileState;
 134     SWR_RECT rect;
 135 };
 136
 137 struct COMPUTE_DESC
 138 {
 139     uint32_t threadGroupCountX;
 140     uint32_t threadGroupCountY;
 141     uint32_t threadGroupCountZ;
 142 };
 143
 144 typedef void(*PFN_WORK_FUNC)(DRAW_CONTEXT* pDC, uint32_t workerId, uint32_t macroTile, void* pDesc);
 145
 146 enum WORK_TYPE
 147 {
 148     SYNC,
 149     DRAW,
 150     CLEAR,
 151     DISCARDINVALIDATETILES,
 152     STORETILES,
 153     SHUTDOWN,
 154 };
 155
 156 OSALIGNSIMD(struct) BE_WORK
 157 {
 158     WORK_TYPE type;
 159     PFN_WORK_FUNC pfnWork;
 160     union
 161     {
 162         SYNC_DESC sync;
 163         TRIANGLE_WORK_DESC tri;
 164         CLEAR_DESC clear;
 165         DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles;
 166         STORE_TILES_DESC storeTiles;
 167     } desc;
 168 };
 169
 170 struct DRAW_WORK
 171 {
 172     DRAW_CONTEXT*   pDC;
 173     union
 174     {
 175         uint32_t   numIndices;      // DrawIndexed: Number of indices for draw.
 176         uint32_t   numVerts;        // Draw: Number of verts (triangles, lines, etc)
 177     };
 178     union
 179     {
 180         const int32_t* pIB;        // DrawIndexed: App supplied indices
 181         uint32_t   startVertex;    // Draw: Starting vertex in VB to render from.
 182     };
 183     int32_t    baseVertex;
 184     uint32_t   numInstances;        // Number of instances
 185     uint32_t   startInstance;       // Instance offset
 186     uint32_t   startPrimID;         // starting primitiveID for this draw batch
 187     uint32_t   startVertexID;       // starting VertexID for this draw batch (only needed for non-indexed draws)
 188     SWR_FORMAT type;                // index buffer type
 189 };
 190
 191 typedef void(*PFN_FE_WORK_FUNC)(SWR_CONTEXT* pContext, DRAW_CONTEXT* pDC, uint32_t workerId, void* pDesc);
 192 struct FE_WORK
 193 {
 194     WORK_TYPE type;
 195     PFN_FE_WORK_FUNC pfnWork;
 196     union
 197     {
 198         SYNC_DESC sync;
 199         DRAW_WORK draw;
 200         CLEAR_DESC clear;
 201         DISCARD_INVALIDATE_TILES_DESC discardInvalidateTiles;
 202         STORE_TILES_DESC storeTiles;
 203     } desc;
 204 };
 205
 206 struct GUARDBANDS
 207 {
 208     float left[KNOB_NUM_VIEWPORTS_SCISSORS];
 209     float right[KNOB_NUM_VIEWPORTS_SCISSORS];
 210     float top[KNOB_NUM_VIEWPORTS_SCISSORS];
 211     float bottom[KNOB_NUM_VIEWPORTS_SCISSORS];
 212 };
 213
 214 struct PA_STATE;
 215
 216 // function signature for pipeline stages that execute after primitive assembly
 217 typedef void(*PFN_PROCESS_PRIMS)(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[],
 218     uint32_t primMask, simdscalari primID, simdscalari viewportIdx);
 219
 220 #if ENABLE_AVX512_SIMD16
 221 // function signature for pipeline stages that execute after primitive assembly
 222 typedef void(SIMDAPI *PFN_PROCESS_PRIMS_SIMD16)(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simd16vector prims[],
 223     uint32_t primMask, simd16scalari primID, simd16scalari viewportIdx);
 224
 225 #endif
 226 OSALIGNLINE(struct) API_STATE
 227 {
 228     // Vertex Buffers
 229     SWR_VERTEX_BUFFER_STATE vertexBuffers[KNOB_NUM_STREAMS];
 230
 231     // Index Buffer
 232     SWR_INDEX_BUFFER_STATE  indexBuffer;
 233
 234     // FS - Fetch Shader State
 235     PFN_FETCH_FUNC          pfnFetchFunc;
 236
 237     // VS - Vertex Shader State
 238     PFN_VERTEX_FUNC         pfnVertexFunc;
 239
 240     // GS - Geometry Shader State
 241     PFN_GS_FUNC             pfnGsFunc;
 242     SWR_GS_STATE            gsState;
 243
 244     // CS - Compute Shader
 245     PFN_CS_FUNC             pfnCsFunc;
 246     uint32_t                totalThreadsInGroup;
 247     uint32_t                totalSpillFillSize;
 248     uint32_t                scratchSpaceSize;
 249     uint32_t                scratchSpaceNumInstances;
 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     GUARDBANDS               gbState;
 278
 279     SWR_VIEWPORT            vp[KNOB_NUM_VIEWPORTS_SCISSORS];
 280     SWR_VIEWPORT_MATRICES   vpMatrices;
 281
 282     SWR_RECT                scissorRects[KNOB_NUM_VIEWPORTS_SCISSORS];
 283     SWR_RECT                scissorsInFixedPoint[KNOB_NUM_VIEWPORTS_SCISSORS];
 284     bool                    scissorsTileAligned;
 285
 286     // Backend state
 287     SWR_BACKEND_STATE       backendState;
 288
 289     SWR_DEPTH_BOUNDS_STATE  depthBoundsState;
 290
 291     // PS - Pixel shader state
 292     SWR_PS_STATE            psState;
 293
 294     SWR_DEPTH_STENCIL_STATE depthStencilState;
 295
 296     // OM - Output Merger State
 297     SWR_BLEND_STATE         blendState;
 298     PFN_BLEND_JIT_FUNC      pfnBlendFunc[SWR_NUM_RENDERTARGETS];
 299
 300     struct
 301     {
 302         uint32_t enableStatsFE : 1;             // Enable frontend pipeline stats
 303         uint32_t enableStatsBE : 1;             // Enable backend pipeline stats
 304         uint32_t colorHottileEnable : 8;        // Bitmask of enabled color hottiles
 305         uint32_t depthHottileEnable: 1;         // Enable depth buffer hottile
 306         uint32_t stencilHottileEnable : 1;      // Enable stencil buffer hottile
 307     };
 308
 309     PFN_QUANTIZE_DEPTH      pfnQuantizeDepth;
 310 };
 311
 312 class MacroTileMgr;
 313 class DispatchQueue;
 314
 315 struct RenderOutputBuffers
 316 {
 317     uint8_t* pColor[SWR_NUM_RENDERTARGETS];
 318     uint8_t* pDepth;
 319     uint8_t* pStencil;
 320 };
 321
 322 // Plane equation A/B/C coeffs used to evaluate I/J barycentric coords
 323 struct BarycentricCoeffs
 324 {
 325     simdscalar vIa;
 326     simdscalar vIb;
 327     simdscalar vIc;
 328
 329     simdscalar vJa;
 330     simdscalar vJb;
 331     simdscalar vJc;
 332
 333     simdscalar vZa;
 334     simdscalar vZb;
 335     simdscalar vZc;
 336
 337     simdscalar vRecipDet;
 338
 339     simdscalar vAOneOverW;
 340     simdscalar vBOneOverW;
 341     simdscalar vCOneOverW;
 342 };
 343
 344 // pipeline function pointer types
 345 typedef void(*PFN_BACKEND_FUNC)(DRAW_CONTEXT*, uint32_t, uint32_t, uint32_t, SWR_TRIANGLE_DESC&, RenderOutputBuffers&);
 346 typedef void(*PFN_OUTPUT_MERGER)(SWR_PS_CONTEXT &, uint8_t* (&)[SWR_NUM_RENDERTARGETS], uint32_t, const SWR_BLEND_STATE*,
 347                                  const PFN_BLEND_JIT_FUNC (&)[SWR_NUM_RENDERTARGETS], simdscalar&, simdscalar);
 348 typedef void(*PFN_CALC_PIXEL_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &);
 349 typedef void(*PFN_CALC_SAMPLE_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT&);
 350 typedef void(*PFN_CALC_CENTROID_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &, const uint64_t *const, const uint32_t,
 351                                               const simdscalar, const simdscalar);
 352
 353 struct BACKEND_FUNCS
 354 {
 355     PFN_BACKEND_FUNC pfnBackend;
 356 };
 357
 358 // Draw State
 359 struct DRAW_STATE
 360 {
 361     API_STATE state;
 362
 363     void* pPrivateState;  // Its required the driver sets this up for each draw.
 364
 365     // pipeline function pointers, filled in by API thread when setting up the draw
 366     BACKEND_FUNCS backendFuncs;
 367     PFN_PROCESS_PRIMS pfnProcessPrims;
 368 #if USE_SIMD16_FRONTEND
 369     PFN_PROCESS_PRIMS_SIMD16 pfnProcessPrims_simd16;
 370 #endif
 371
 372     CachingArena* pArena;     // This should only be used by API thread.
 373 };
 374
 375 struct DRAW_DYNAMIC_STATE
 376 {
 377     void Reset(uint32_t numThreads)
 378     {
 379         SWR_STATS* pSavePtr = pStats;
 380         memset(this, 0, sizeof(*this));
 381         pStats = pSavePtr;
 382         memset(pStats, 0, sizeof(SWR_STATS) * numThreads);
 383     }
 384     ///@todo Currently assumes only a single FE can do stream output for a draw.
 385     uint32_t SoWriteOffset[4];
 386     bool     SoWriteOffsetDirty[4];
 387
 388     SWR_STATS_FE statsFE;   // Only one FE thread per DC.
 389     SWR_STATS*   pStats;
 390 };
 391
 392 // Draw Context
 393 //    The api thread sets up a draw context that exists for the life of the draw.
 394 //    This draw context maintains all of the state needed for the draw operation.
 395 struct DRAW_CONTEXT
 396 {
 397     SWR_CONTEXT*    pContext;
 398     union
 399     {
 400         MacroTileMgr*   pTileMgr;
 401         DispatchQueue*  pDispatch;      // Queue for thread groups. (isCompute)
 402     };
 403     DRAW_STATE*     pState;             // Read-only state. Core should not update this outside of API thread.
 404     DRAW_DYNAMIC_STATE dynState;
 405
 406     CachingArena*   pArena;
 407
 408     uint32_t        drawId;
 409     bool            dependentFE;    // Frontend work is dependent on all previous FE
 410     bool            dependent;      // Backend work is dependent on all previous BE
 411     bool            isCompute;      // Is this DC a compute context?
 412     bool            cleanupState;   // True if this is the last draw using an entry in the state ring.
 413     volatile bool   doneFE;         // Is FE work done for this draw?
 414
 415     FE_WORK         FeWork;
 416
 417     volatile OSALIGNLINE(uint32_t)   FeLock;
 418     volatile int32_t    threadsDone;
 419
 420     SYNC_DESC       retireCallback; // Call this func when this DC is retired.
 421 };
 422
 423 static_assert((sizeof(DRAW_CONTEXT) & 63) == 0, "Invalid size for DRAW_CONTEXT");
 424
 425 INLINE const API_STATE& GetApiState(const DRAW_CONTEXT* pDC)
 426 {
 427     SWR_ASSERT(pDC != nullptr);
 428     SWR_ASSERT(pDC->pState != nullptr);
 429
 430     return pDC->pState->state;
 431 }
 432
 433 INLINE void* GetPrivateState(const DRAW_CONTEXT* pDC)
 434 {
 435     SWR_ASSERT(pDC != nullptr);
 436     SWR_ASSERT(pDC->pState != nullptr);
 437
 438     return pDC->pState->pPrivateState;
 439 }
 440
 441 class HotTileMgr;
 442
 443 struct SWR_CONTEXT
 444 {
 445     // Draw Context Ring
 446     //  Each draw needs its own state in order to support mulitple draws in flight across multiple threads.
 447     //  We maintain N draw contexts configured as a ring. The size of the ring limits the maximum number
 448     //  of draws that can be in flight at any given time.
 449     //
 450     //  Description:
 451     //  1. State - When an application first sets state we'll request a new draw context to use.
 452     //     a. If there are no available draw contexts then we'll have to wait until one becomes free.
 453     //     b. If one is available then set pCurDrawContext to point to it and mark it in use.
 454     //     c. All state calls set state on pCurDrawContext.
 455     //  2. Draw - Creates submits a work item that is associated with current draw context.
 456     //     a. Set pPrevDrawContext = pCurDrawContext
 457     //     b. Set pCurDrawContext to NULL.
 458     //  3. State - When an applications sets state after draw
 459     //     a. Same as step 1.
 460     //     b. State is copied from prev draw context to current.
 461     RingBuffer<DRAW_CONTEXT> dcRing;
 462
 463     DRAW_CONTEXT *pCurDrawContext;    // This points to DC entry in ring for an unsubmitted draw.
 464     DRAW_CONTEXT *pPrevDrawContext;   // This points to DC entry for the previous context submitted that we can copy state from.
 465
 466     MacroTileMgr* pMacroTileManagerArray;
 467     DispatchQueue* pDispatchQueueArray;
 468
 469     // Draw State Ring
 470     //  When draw are very large (lots of primitives) then the API thread will break these up.
 471     //  These split draws all have identical state. So instead of storing the state directly
 472     //  in the Draw Context (DC) we instead store it in a Draw State (DS). This allows multiple DCs
 473     //  to reference a single entry in the DS ring.
 474     RingBuffer<DRAW_STATE> dsRing;
 475
 476     uint32_t curStateId;               // Current index to the next available entry in the DS ring.
 477
 478     uint32_t NumWorkerThreads;
 479     uint32_t NumFEThreads;
 480     uint32_t NumBEThreads;
 481
 482     THREAD_POOL threadPool; // Thread pool associated with this context
 483     SWR_THREADING_INFO threadInfo;
 484
 485     std::condition_variable FifosNotEmpty;
 486     std::mutex WaitLock;
 487
 488     uint32_t privateStateSize;
 489
 490     HotTileMgr *pHotTileMgr;
 491
 492     // Callback functions, passed in at create context time
 493     PFN_LOAD_TILE               pfnLoadTile;
 494     PFN_STORE_TILE              pfnStoreTile;
 495     PFN_CLEAR_TILE              pfnClearTile;
 496     PFN_UPDATE_SO_WRITE_OFFSET  pfnUpdateSoWriteOffset;
 497     PFN_UPDATE_STATS            pfnUpdateStats;
 498     PFN_UPDATE_STATS_FE         pfnUpdateStatsFE;
 499
 500
 501     // Global Stats
 502     SWR_STATS* pStats;
 503
 504     // Scratch space for workers.
 505     uint8_t** ppScratch;
 506
 507     volatile int32_t  drawsOutstandingFE;
 508
 509     CachingAllocator cachingArenaAllocator;
 510     uint32_t frameCount;
 511
 512     uint32_t lastFrameChecked;
 513     uint64_t lastDrawChecked;
 514     TileSet singleThreadLockedTiles;
 515
 516     // ArchRast thread contexts.
 517     HANDLE* pArContext;
 518 };
 519
 520 #define UPDATE_STAT_BE(name, count) if (GetApiState(pDC).enableStatsBE) { pDC->dynState.pStats[workerId].name += count; }
 521 #define UPDATE_STAT_FE(name, count) if (GetApiState(pDC).enableStatsFE) { pDC->dynState.statsFE.name += count; }
 522
 523 // ArchRast instrumentation framework
 524 #define AR_WORKER_CTX  pContext->pArContext[workerId]
 525 #define AR_API_CTX     pContext->pArContext[pContext->NumWorkerThreads]
 526
 527 #ifdef KNOB_ENABLE_AR
 528     #define _AR_BEGIN(ctx, type, id)    ArchRast::Dispatch(ctx, ArchRast::Start(ArchRast::type, id))
 529     #define _AR_END(ctx, type, count)   ArchRast::Dispatch(ctx, ArchRast::End(ArchRast::type, count))
 530     #define _AR_EVENT(ctx, event)       ArchRast::Dispatch(ctx, ArchRast::event)
 531     #define _AR_FLUSH(ctx, id)          ArchRast::FlushDraw(ctx, id)
 532 #else
 533     #ifdef KNOB_ENABLE_RDTSC
 534         #define _AR_BEGIN(ctx, type, id) (void)ctx; RDTSC_START(type)
 535         #define _AR_END(ctx, type, id)   RDTSC_STOP(type, id, 0)
 536     #else
 537         #define _AR_BEGIN(ctx, type, id) (void)ctx
 538         #define _AR_END(ctx, type, id)
 539     #endif
 540     #define _AR_EVENT(ctx, event)
 541     #define _AR_FLUSH(ctx, id)
 542 #endif
 543
 544 // Use these macros for api thread.
 545 #define AR_API_BEGIN(type, id) _AR_BEGIN(AR_API_CTX, type, id)
 546 #define AR_API_END(type, count) _AR_END(AR_API_CTX, type, count)
 547 #define AR_API_EVENT(event) _AR_EVENT(AR_API_CTX, event)
 548
 549 // Use these macros for worker threads.
 550 #define AR_BEGIN(type, id) _AR_BEGIN(AR_WORKER_CTX, type, id)
 551 #define AR_END(type, count) _AR_END(AR_WORKER_CTX, type, count)
 552 #define AR_EVENT(event) _AR_EVENT(AR_WORKER_CTX, event)
 553 #define AR_FLUSH(id) _AR_FLUSH(AR_WORKER_CTX, id)