src/gallium/drivers/swr/rasterizer/core/api.cpp

   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 api.cpp
  24 *
  25 * @brief API implementation
  26 *
  27 ******************************************************************************/
  28
  29 #include <cfloat>
  30 #include <cmath>
  31 #include <cstdio>
  32 #include <new>
  33
  34 #include "core/api.h"
  35 #include "core/backend.h"
  36 #include "core/context.h"
  37 #include "core/depthstencil.h"
  38 #include "core/frontend.h"
  39 #include "core/rasterizer.h"
  40 #include "core/rdtsc_core.h"
  41 #include "core/threads.h"
  42 #include "core/tilemgr.h"
  43 #include "core/clip.h"
  44 #include "core/utils.h"
  45
  46 #include "common/os.h"
  47
  48 static const SWR_RECT g_MaxScissorRect = { 0, 0, KNOB_MAX_SCISSOR_X, KNOB_MAX_SCISSOR_Y };
  49
  50 void SetupDefaultState(SWR_CONTEXT *pContext);
  51
  52 static INLINE SWR_CONTEXT* GetContext(HANDLE hContext)
  53 {
  54     return (SWR_CONTEXT*)hContext;
  55 }
  56
  57 void WakeAllThreads(SWR_CONTEXT *pContext)
  58 {
  59     pContext->FifosNotEmpty.notify_all();
  60 }
  61
  62 //////////////////////////////////////////////////////////////////////////
  63 /// @brief Create SWR Context.
  64 /// @param pCreateInfo - pointer to creation info.
  65 HANDLE SwrCreateContext(
  66     SWR_CREATECONTEXT_INFO* pCreateInfo)
  67 {
  68     RDTSC_RESET();
  69     RDTSC_INIT(0);
  70
  71     void* pContextMem = AlignedMalloc(sizeof(SWR_CONTEXT), KNOB_SIMD_WIDTH * 4);
  72     memset(pContextMem, 0, sizeof(SWR_CONTEXT));
  73     SWR_CONTEXT *pContext = new (pContextMem) SWR_CONTEXT();
  74
  75     pContext->privateStateSize = pCreateInfo->privateStateSize;
  76
  77     pContext->dcRing.Init(KNOB_MAX_DRAWS_IN_FLIGHT);
  78     pContext->dsRing.Init(KNOB_MAX_DRAWS_IN_FLIGHT);
  79
  80     pContext->pMacroTileManagerArray = (MacroTileMgr*)AlignedMalloc(sizeof(MacroTileMgr) * KNOB_MAX_DRAWS_IN_FLIGHT, 64);
  81     pContext->pDispatchQueueArray = (DispatchQueue*)AlignedMalloc(sizeof(DispatchQueue) * KNOB_MAX_DRAWS_IN_FLIGHT, 64);
  82
  83     for (uint32_t dc = 0; dc < KNOB_MAX_DRAWS_IN_FLIGHT; ++dc)
  84     {
  85         pContext->dcRing[dc].pArena = new CachingArena(pContext->cachingArenaAllocator);
  86         new (&pContext->pMacroTileManagerArray[dc]) MacroTileMgr(*pContext->dcRing[dc].pArena);
  87         new (&pContext->pDispatchQueueArray[dc]) DispatchQueue();
  88
  89         pContext->dsRing[dc].pArena = new CachingArena(pContext->cachingArenaAllocator);
  90     }
  91
  92     pContext->threadInfo.MAX_WORKER_THREADS        = KNOB_MAX_WORKER_THREADS;
  93     pContext->threadInfo.MAX_NUMA_NODES            = KNOB_MAX_NUMA_NODES;
  94     pContext->threadInfo.MAX_CORES_PER_NUMA_NODE   = KNOB_MAX_CORES_PER_NUMA_NODE;
  95     pContext->threadInfo.MAX_THREADS_PER_CORE      = KNOB_MAX_THREADS_PER_CORE;
  96     pContext->threadInfo.SINGLE_THREADED           = KNOB_SINGLE_THREADED;
  97
  98     if (pCreateInfo->pThreadInfo)
  99     {
 100         pContext->threadInfo = *pCreateInfo->pThreadInfo;
 101     }
 102
 103     memset(&pContext->WaitLock, 0, sizeof(pContext->WaitLock));
 104     memset(&pContext->FifosNotEmpty, 0, sizeof(pContext->FifosNotEmpty));
 105     new (&pContext->WaitLock) std::mutex();
 106     new (&pContext->FifosNotEmpty) std::condition_variable();
 107
 108     CreateThreadPool(pContext, &pContext->threadPool);
 109
 110     pContext->ppScratch = new uint8_t*[pContext->NumWorkerThreads];
 111     pContext->pStats = new SWR_STATS[pContext->NumWorkerThreads];
 112
 113 #if defined(KNOB_ENABLE_AR)
 114     // Setup ArchRast thread contexts which includes +1 for API thread.
 115     pContext->pArContext = new HANDLE[pContext->NumWorkerThreads+1];
 116     pContext->pArContext[pContext->NumWorkerThreads] = ArchRast::CreateThreadContext(ArchRast::AR_THREAD::API);
 117 #endif
 118
 119     // Allocate scratch space for workers.
 120     ///@note We could lazily allocate this but its rather small amount of memory.
 121     for (uint32_t i = 0; i < pContext->NumWorkerThreads; ++i)
 122     {
 123 #if defined(_WIN32)
 124         uint32_t numaNode = pContext->threadPool.pThreadData ?
 125             pContext->threadPool.pThreadData[i].numaId : 0;
 126         pContext->ppScratch[i] = (uint8_t*)VirtualAllocExNuma(
 127             GetCurrentProcess(), nullptr, 32 * sizeof(KILOBYTE),
 128             MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE,
 129             numaNode);
 130 #else
 131         pContext->ppScratch[i] = (uint8_t*)AlignedMalloc(32 * sizeof(KILOBYTE), KNOB_SIMD_WIDTH * 4);
 132 #endif
 133
 134 #if defined(KNOB_ENABLE_AR)
 135         // Initialize worker thread context for ArchRast.
 136         pContext->pArContext[i] = ArchRast::CreateThreadContext(ArchRast::AR_THREAD::WORKER);
 137 #endif
 138     }
 139
 140     // State setup AFTER context is fully initialized
 141     SetupDefaultState(pContext);
 142
 143     // initialize hot tile manager
 144     pContext->pHotTileMgr = new HotTileMgr();
 145
 146     // initialize callback functions
 147     pContext->pfnLoadTile = pCreateInfo->pfnLoadTile;
 148     pContext->pfnStoreTile = pCreateInfo->pfnStoreTile;
 149     pContext->pfnClearTile = pCreateInfo->pfnClearTile;
 150     pContext->pfnUpdateSoWriteOffset = pCreateInfo->pfnUpdateSoWriteOffset;
 151     pContext->pfnUpdateStats = pCreateInfo->pfnUpdateStats;
 152     pContext->pfnUpdateStatsFE = pCreateInfo->pfnUpdateStatsFE;
 153
 154
 155     // pass pointer to bucket manager back to caller
 156 #ifdef KNOB_ENABLE_RDTSC
 157     pCreateInfo->pBucketMgr = &gBucketMgr;
 158 #endif
 159
 160     pCreateInfo->contextSaveSize = sizeof(API_STATE);
 161
 162     StartThreadPool(pContext, &pContext->threadPool);
 163
 164     return (HANDLE)pContext;
 165 }
 166
 167 void CopyState(DRAW_STATE& dst, const DRAW_STATE& src)
 168 {
 169     memcpy(&dst.state, &src.state, sizeof(API_STATE));
 170 }
 171
 172 template<bool IsDraw>
 173 void QueueWork(SWR_CONTEXT *pContext)
 174 {
 175     DRAW_CONTEXT* pDC = pContext->pCurDrawContext;
 176     uint32_t dcIndex = pDC->drawId % KNOB_MAX_DRAWS_IN_FLIGHT;
 177
 178     if (IsDraw)
 179     {
 180         pDC->pTileMgr = &pContext->pMacroTileManagerArray[dcIndex];
 181         pDC->pTileMgr->initialize();
 182     }
 183
 184     // Each worker thread looks at a DC for both FE and BE work at different times and so we
 185     // multiply threadDone by 2.  When the threadDone counter has reached 0 then all workers
 186     // have moved past this DC. (i.e. Each worker has checked this DC for both FE and BE work and
 187     // then moved on if all work is done.)
 188     pContext->pCurDrawContext->threadsDone = pContext->NumFEThreads + pContext->NumBEThreads;
 189
 190     if (IsDraw)
 191     {
 192         InterlockedIncrement((volatile LONG*)&pContext->drawsOutstandingFE);
 193     }
 194
 195     _ReadWriteBarrier();
 196     {
 197         std::unique_lock<std::mutex> lock(pContext->WaitLock);
 198         pContext->dcRing.Enqueue();
 199     }
 200
 201     if (pContext->threadInfo.SINGLE_THREADED)
 202     {
 203         // flush denormals to 0
 204         uint32_t mxcsr = _mm_getcsr();
 205         _mm_setcsr(mxcsr | _MM_FLUSH_ZERO_ON | _MM_DENORMALS_ZERO_ON);
 206
 207         if (IsDraw)
 208         {
 209             uint32_t curDraw[2] = { pContext->pCurDrawContext->drawId, pContext->pCurDrawContext->drawId };
 210             WorkOnFifoFE(pContext, 0, curDraw[0]);
 211             WorkOnFifoBE(pContext, 0, curDraw[1], pContext->singleThreadLockedTiles, 0, 0);
 212         }
 213         else
 214         {
 215             uint32_t curDispatch = pContext->pCurDrawContext->drawId;
 216             WorkOnCompute(pContext, 0, curDispatch);
 217         }
 218
 219         // Dequeue the work here, if not already done, since we're single threaded (i.e. no workers).
 220         while (CompleteDrawContext(pContext, pContext->pCurDrawContext) > 0) {}
 221
 222         // restore csr
 223         _mm_setcsr(mxcsr);
 224     }
 225     else
 226     {
 227         AR_API_BEGIN(APIDrawWakeAllThreads, pDC->drawId);
 228         WakeAllThreads(pContext);
 229         AR_API_END(APIDrawWakeAllThreads, 1);
 230     }
 231
 232     // Set current draw context to NULL so that next state call forces a new draw context to be created and populated.
 233     pContext->pPrevDrawContext = pContext->pCurDrawContext;
 234     pContext->pCurDrawContext = nullptr;
 235 }
 236
 237 INLINE void QueueDraw(SWR_CONTEXT* pContext)
 238 {
 239     QueueWork<true>(pContext);
 240 }
 241
 242 INLINE void QueueDispatch(SWR_CONTEXT* pContext)
 243 {
 244     QueueWork<false>(pContext);
 245 }
 246
 247 DRAW_CONTEXT* GetDrawContext(SWR_CONTEXT *pContext, bool isSplitDraw = false)
 248 {
 249     AR_API_BEGIN(APIGetDrawContext, 0);
 250     // If current draw context is null then need to obtain a new draw context to use from ring.
 251     if (pContext->pCurDrawContext == nullptr)
 252     {
 253         // Need to wait for a free entry.
 254         while (pContext->dcRing.IsFull())
 255         {
 256             _mm_pause();
 257         }
 258
 259         uint64_t curDraw = pContext->dcRing.GetHead();
 260         uint32_t dcIndex = curDraw % KNOB_MAX_DRAWS_IN_FLIGHT;
 261
 262         if ((pContext->frameCount - pContext->lastFrameChecked) > 2 ||
 263             (curDraw - pContext->lastDrawChecked) > 0x10000)
 264         {
 265             // Take this opportunity to clean-up old arena allocations
 266             pContext->cachingArenaAllocator.FreeOldBlocks();
 267
 268             pContext->lastFrameChecked = pContext->frameCount;
 269             pContext->lastDrawChecked = curDraw;
 270         }
 271
 272         DRAW_CONTEXT* pCurDrawContext = &pContext->dcRing[dcIndex];
 273         pContext->pCurDrawContext = pCurDrawContext;
 274
 275         // Assign next available entry in DS ring to this DC.
 276         uint32_t dsIndex = pContext->curStateId % KNOB_MAX_DRAWS_IN_FLIGHT;
 277         pCurDrawContext->pState = &pContext->dsRing[dsIndex];
 278
 279         // Copy previous state to current state.
 280         if (pContext->pPrevDrawContext)
 281         {
 282             DRAW_CONTEXT* pPrevDrawContext = pContext->pPrevDrawContext;
 283
 284             // If we're splitting our draw then we can just use the same state from the previous
 285             // draw. In this case, we won't increment the DS ring index so the next non-split
 286             // draw can receive the state.
 287             if (isSplitDraw == false)
 288             {
 289                 CopyState(*pCurDrawContext->pState, *pPrevDrawContext->pState);
 290
 291                 // Should have been cleaned up previously
 292                 SWR_ASSERT(pCurDrawContext->pState->pArena->IsEmpty() == true);
 293
 294                 pCurDrawContext->pState->pPrivateState = nullptr;
 295
 296                 pContext->curStateId++;  // Progress state ring index forward.
 297             }
 298             else
 299             {
 300                 // If its a split draw then just copy the state pointer over
 301                 // since its the same draw.
 302                 pCurDrawContext->pState = pPrevDrawContext->pState;
 303                 SWR_ASSERT(pPrevDrawContext->cleanupState == false);
 304             }
 305         }
 306         else
 307         {
 308             SWR_ASSERT(pCurDrawContext->pState->pArena->IsEmpty() == true);
 309             pContext->curStateId++;  // Progress state ring index forward.
 310         }
 311
 312         SWR_ASSERT(pCurDrawContext->pArena->IsEmpty() == true);
 313
 314         // Reset dependency
 315         pCurDrawContext->dependent = false;
 316         pCurDrawContext->dependentFE = false;
 317
 318         pCurDrawContext->pContext = pContext;
 319         pCurDrawContext->isCompute = false; // Dispatch has to set this to true.
 320
 321         pCurDrawContext->doneFE = false;
 322         pCurDrawContext->FeLock = 0;
 323         pCurDrawContext->threadsDone = 0;
 324         pCurDrawContext->retireCallback.pfnCallbackFunc = nullptr;
 325
 326         pCurDrawContext->dynState.Reset(pContext->NumWorkerThreads);
 327
 328         // Assign unique drawId for this DC
 329         pCurDrawContext->drawId = pContext->dcRing.GetHead();
 330
 331         pCurDrawContext->cleanupState = true;
 332     }
 333     else
 334     {
 335         SWR_ASSERT(isSplitDraw == false, "Split draw should only be used when obtaining a new DC");
 336     }
 337
 338     AR_API_END(APIGetDrawContext, 0);
 339     return pContext->pCurDrawContext;
 340 }
 341
 342 API_STATE* GetDrawState(SWR_CONTEXT *pContext)
 343 {
 344     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
 345     SWR_ASSERT(pDC->pState != nullptr);
 346
 347     return &pDC->pState->state;
 348 }
 349
 350 void SwrDestroyContext(HANDLE hContext)
 351 {
 352     SWR_CONTEXT *pContext = GetContext(hContext);
 353     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
 354
 355     pDC->FeWork.type = SHUTDOWN;
 356     pDC->FeWork.pfnWork = ProcessShutdown;
 357
 358     //enqueue
 359     QueueDraw(pContext);
 360
 361     DestroyThreadPool(pContext, &pContext->threadPool);
 362
 363     // free the fifos
 364     for (uint32_t i = 0; i < KNOB_MAX_DRAWS_IN_FLIGHT; ++i)
 365     {
 366         delete[] pContext->dcRing[i].dynState.pStats;
 367         delete pContext->dcRing[i].pArena;
 368         delete pContext->dsRing[i].pArena;
 369         pContext->pMacroTileManagerArray[i].~MacroTileMgr();
 370         pContext->pDispatchQueueArray[i].~DispatchQueue();
 371     }
 372
 373     AlignedFree(pContext->pDispatchQueueArray);
 374     AlignedFree(pContext->pMacroTileManagerArray);
 375
 376     // Free scratch space.
 377     for (uint32_t i = 0; i < pContext->NumWorkerThreads; ++i)
 378     {
 379 #if defined(_WIN32)
 380         VirtualFree(pContext->ppScratch[i], 0, MEM_RELEASE);
 381 #else
 382         AlignedFree(pContext->ppScratch[i]);
 383 #endif
 384
 385 #if defined(KNOB_ENABLE_AR)
 386         ArchRast::DestroyThreadContext(pContext->pArContext[i]);
 387 #endif
 388     }
 389
 390     delete[] pContext->ppScratch;
 391     delete[] pContext->pStats;
 392
 393     delete(pContext->pHotTileMgr);
 394
 395     pContext->~SWR_CONTEXT();
 396     AlignedFree(GetContext(hContext));
 397 }
 398
 399 void SWR_API SwrSaveState(
 400     HANDLE hContext,
 401     void* pOutputStateBlock,
 402     size_t memSize)
 403 {
 404     SWR_CONTEXT *pContext = GetContext(hContext);
 405     auto pSrc = GetDrawState(pContext);
 406     SWR_ASSERT(pOutputStateBlock && memSize >= sizeof(*pSrc));
 407
 408     memcpy(pOutputStateBlock, pSrc, sizeof(*pSrc));
 409 }
 410
 411 void SWR_API SwrRestoreState(
 412     HANDLE hContext,
 413     const void* pStateBlock,
 414     size_t memSize)
 415 {
 416     SWR_CONTEXT *pContext = GetContext(hContext);
 417     auto pDst = GetDrawState(pContext);
 418     SWR_ASSERT(pStateBlock && memSize >= sizeof(*pDst));
 419
 420     memcpy(pDst, pStateBlock, sizeof(*pDst));
 421 }
 422
 423 void SetupDefaultState(SWR_CONTEXT *pContext)
 424 {
 425     API_STATE* pState = GetDrawState(pContext);
 426
 427     pState->rastState.cullMode = SWR_CULLMODE_NONE;
 428     pState->rastState.frontWinding = SWR_FRONTWINDING_CCW;
 429
 430     pState->depthBoundsState.depthBoundsTestEnable = false;
 431     pState->depthBoundsState.depthBoundsTestMinValue = 0.0f;
 432     pState->depthBoundsState.depthBoundsTestMaxValue = 1.0f;
 433 }
 434
 435 void SwrSync(HANDLE hContext, PFN_CALLBACK_FUNC pfnFunc, uint64_t userData, uint64_t userData2, uint64_t userData3)
 436 {
 437     SWR_ASSERT(pfnFunc != nullptr);
 438
 439     SWR_CONTEXT *pContext = GetContext(hContext);
 440     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
 441
 442     AR_API_BEGIN(APISync, 0);
 443
 444     pDC->FeWork.type = SYNC;
 445     pDC->FeWork.pfnWork = ProcessSync;
 446
 447     // Setup callback function
 448     pDC->retireCallback.pfnCallbackFunc = pfnFunc;
 449     pDC->retireCallback.userData = userData;
 450     pDC->retireCallback.userData2 = userData2;
 451     pDC->retireCallback.userData3 = userData3;
 452
 453     AR_API_EVENT(SwrSyncEvent(pDC->drawId));
 454
 455     //enqueue
 456     QueueDraw(pContext);
 457
 458     AR_API_END(APISync, 1);
 459 }
 460
 461 void SwrWaitForIdle(HANDLE hContext)
 462 {
 463     SWR_CONTEXT *pContext = GetContext(hContext);
 464
 465     AR_API_BEGIN(APIWaitForIdle, 0);
 466
 467     while (!pContext->dcRing.IsEmpty())
 468     {
 469         _mm_pause();
 470     }
 471
 472     AR_API_END(APIWaitForIdle, 1);
 473 }
 474
 475 void SwrWaitForIdleFE(HANDLE hContext)
 476 {
 477     SWR_CONTEXT *pContext = GetContext(hContext);
 478
 479     AR_API_BEGIN(APIWaitForIdle, 0);
 480
 481     while (pContext->drawsOutstandingFE > 0)
 482     {
 483         _mm_pause();
 484     }
 485
 486     AR_API_END(APIWaitForIdle, 1);
 487 }
 488
 489 void SwrSetVertexBuffers(
 490     HANDLE hContext,
 491     uint32_t numBuffers,
 492     const SWR_VERTEX_BUFFER_STATE* pVertexBuffers)
 493 {
 494     API_STATE* pState = GetDrawState(GetContext(hContext));
 495
 496     for (uint32_t i = 0; i < numBuffers; ++i)
 497     {
 498         const SWR_VERTEX_BUFFER_STATE *pVB = &pVertexBuffers[i];
 499         pState->vertexBuffers[pVB->index] = *pVB;
 500     }
 501 }
 502
 503 void SwrSetIndexBuffer(
 504     HANDLE hContext,
 505     const SWR_INDEX_BUFFER_STATE* pIndexBuffer)
 506 {
 507     API_STATE* pState = GetDrawState(GetContext(hContext));
 508
 509     pState->indexBuffer = *pIndexBuffer;
 510 }
 511
 512 void SwrSetFetchFunc(
 513     HANDLE hContext,
 514     PFN_FETCH_FUNC    pfnFetchFunc)
 515 {
 516     API_STATE* pState = GetDrawState(GetContext(hContext));
 517
 518     pState->pfnFetchFunc = pfnFetchFunc;
 519 }
 520
 521 void SwrSetSoFunc(
 522     HANDLE hContext,
 523     PFN_SO_FUNC    pfnSoFunc,
 524     uint32_t streamIndex)
 525 {
 526     API_STATE* pState = GetDrawState(GetContext(hContext));
 527
 528     SWR_ASSERT(streamIndex < MAX_SO_STREAMS);
 529
 530     pState->pfnSoFunc[streamIndex] = pfnSoFunc;
 531 }
 532
 533 void SwrSetSoState(
 534     HANDLE hContext,
 535     SWR_STREAMOUT_STATE* pSoState)
 536 {
 537     API_STATE* pState = GetDrawState(GetContext(hContext));
 538
 539     pState->soState = *pSoState;
 540 }
 541
 542 void SwrSetSoBuffers(
 543     HANDLE hContext,
 544     SWR_STREAMOUT_BUFFER* pSoBuffer,
 545     uint32_t slot)
 546 {
 547     API_STATE* pState = GetDrawState(GetContext(hContext));
 548
 549     SWR_ASSERT((slot < 4), "There are only 4 SO buffer slots [0, 3]\nSlot requested: %d", slot);
 550
 551     pState->soBuffer[slot] = *pSoBuffer;
 552 }
 553
 554 void SwrSetVertexFunc(
 555     HANDLE hContext,
 556     PFN_VERTEX_FUNC pfnVertexFunc)
 557 {
 558     API_STATE* pState = GetDrawState(GetContext(hContext));
 559
 560     pState->pfnVertexFunc = pfnVertexFunc;
 561 }
 562
 563 void SwrSetFrontendState(
 564     HANDLE hContext,
 565     SWR_FRONTEND_STATE *pFEState)
 566 {
 567     API_STATE* pState = GetDrawState(GetContext(hContext));
 568     pState->frontendState = *pFEState;
 569 }
 570
 571 void SwrSetGsState(
 572     HANDLE hContext,
 573     SWR_GS_STATE *pGSState)
 574 {
 575     API_STATE* pState = GetDrawState(GetContext(hContext));
 576     pState->gsState = *pGSState;
 577 }
 578
 579 void SwrSetGsFunc(
 580     HANDLE hContext,
 581     PFN_GS_FUNC pfnGsFunc)
 582 {
 583     API_STATE* pState = GetDrawState(GetContext(hContext));
 584     pState->pfnGsFunc = pfnGsFunc;
 585 }
 586
 587 void SwrSetCsFunc(
 588     HANDLE hContext,
 589     PFN_CS_FUNC pfnCsFunc,
 590     uint32_t totalThreadsInGroup,
 591     uint32_t totalSpillFillSize,
 592     uint32_t scratchSpaceSizePerInstance,
 593     uint32_t numInstances)
 594 {
 595     API_STATE* pState = GetDrawState(GetContext(hContext));
 596     pState->pfnCsFunc = pfnCsFunc;
 597     pState->totalThreadsInGroup = totalThreadsInGroup;
 598     pState->totalSpillFillSize = totalSpillFillSize;
 599     pState->scratchSpaceSize = scratchSpaceSizePerInstance;
 600     pState->scratchSpaceNumInstances = numInstances;
 601 }
 602
 603 void SwrSetTsState(
 604     HANDLE hContext,
 605     SWR_TS_STATE *pState)
 606 {
 607     API_STATE* pApiState = GetDrawState(GetContext(hContext));
 608     pApiState->tsState = *pState;
 609 }
 610
 611 void SwrSetHsFunc(
 612     HANDLE hContext,
 613     PFN_HS_FUNC pfnFunc)
 614 {
 615     API_STATE* pApiState = GetDrawState(GetContext(hContext));
 616     pApiState->pfnHsFunc = pfnFunc;
 617 }
 618
 619 void SwrSetDsFunc(
 620     HANDLE hContext,
 621     PFN_DS_FUNC pfnFunc)
 622 {
 623     API_STATE* pApiState = GetDrawState(GetContext(hContext));
 624     pApiState->pfnDsFunc = pfnFunc;
 625 }
 626
 627 void SwrSetDepthStencilState(
 628     HANDLE hContext,
 629     SWR_DEPTH_STENCIL_STATE *pDSState)
 630 {
 631     API_STATE* pState = GetDrawState(GetContext(hContext));
 632
 633     pState->depthStencilState = *pDSState;
 634 }
 635
 636 void SwrSetBackendState(
 637     HANDLE hContext,
 638     SWR_BACKEND_STATE *pBEState)
 639 {
 640     API_STATE* pState = GetDrawState(GetContext(hContext));
 641
 642     pState->backendState = *pBEState;
 643 }
 644
 645 void SwrSetDepthBoundsState(
 646     HANDLE hContext,
 647     SWR_DEPTH_BOUNDS_STATE *pDBState)
 648 {
 649     API_STATE* pState = GetDrawState(GetContext(hContext));
 650
 651     pState->depthBoundsState = *pDBState;
 652 }
 653
 654 void SwrSetPixelShaderState(
 655     HANDLE hContext,
 656     SWR_PS_STATE *pPSState)
 657 {
 658     API_STATE *pState = GetDrawState(GetContext(hContext));
 659     pState->psState = *pPSState;
 660 }
 661
 662 void SwrSetBlendState(
 663     HANDLE hContext,
 664     SWR_BLEND_STATE *pBlendState)
 665 {
 666     API_STATE *pState = GetDrawState(GetContext(hContext));
 667     memcpy(&pState->blendState, pBlendState, sizeof(SWR_BLEND_STATE));
 668 }
 669
 670 void SwrSetBlendFunc(
 671     HANDLE hContext,
 672     uint32_t renderTarget,
 673     PFN_BLEND_JIT_FUNC pfnBlendFunc)
 674 {
 675     SWR_ASSERT(renderTarget < SWR_NUM_RENDERTARGETS);
 676     API_STATE *pState = GetDrawState(GetContext(hContext));
 677     pState->pfnBlendFunc[renderTarget] = pfnBlendFunc;
 678 }
 679
 680 // update guardband multipliers for the viewport
 681 void updateGuardbands(API_STATE *pState)
 682 {
 683     uint32_t numGbs = pState->gsState.emitsRenderTargetArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
 684
 685     for(uint32_t i = 0; i < numGbs; ++i)
 686     {
 687         // guardband center is viewport center
 688         pState->gbState.left[i] = KNOB_GUARDBAND_WIDTH / pState->vp[i].width;
 689         pState->gbState.right[i] = KNOB_GUARDBAND_WIDTH / pState->vp[i].width;
 690         pState->gbState.top[i] = KNOB_GUARDBAND_HEIGHT / pState->vp[i].height;
 691         pState->gbState.bottom[i] = KNOB_GUARDBAND_HEIGHT / pState->vp[i].height;
 692     }
 693 }
 694
 695 void SwrSetRastState(
 696     HANDLE hContext,
 697     const SWR_RASTSTATE *pRastState)
 698 {
 699     SWR_CONTEXT *pContext = GetContext(hContext);
 700     API_STATE* pState = GetDrawState(pContext);
 701
 702     memcpy(&pState->rastState, pRastState, sizeof(SWR_RASTSTATE));
 703 }
 704
 705 void SwrSetViewports(
 706     HANDLE hContext,
 707     uint32_t numViewports,
 708     const SWR_VIEWPORT* pViewports,
 709     const SWR_VIEWPORT_MATRICES* pMatrices)
 710 {
 711     SWR_ASSERT(numViewports <= KNOB_NUM_VIEWPORTS_SCISSORS,
 712         "Invalid number of viewports.");
 713
 714     SWR_CONTEXT *pContext = GetContext(hContext);
 715     API_STATE* pState = GetDrawState(pContext);
 716
 717     memcpy(&pState->vp[0], pViewports, sizeof(SWR_VIEWPORT) * numViewports);
 718     // @todo Faster to copy portions of the SOA or just copy all of it?
 719     memcpy(&pState->vpMatrices, pMatrices, sizeof(SWR_VIEWPORT_MATRICES));
 720
 721     updateGuardbands(pState);
 722 }
 723
 724 void SwrSetScissorRects(
 725     HANDLE hContext,
 726     uint32_t numScissors,
 727     const SWR_RECT* pScissors)
 728 {
 729     SWR_ASSERT(numScissors <= KNOB_NUM_VIEWPORTS_SCISSORS,
 730         "Invalid number of scissor rects.");
 731
 732     API_STATE* pState = GetDrawState(GetContext(hContext));
 733     memcpy(&pState->scissorRects[0], pScissors, numScissors * sizeof(pScissors[0]));
 734 };
 735
 736 void SetupMacroTileScissors(DRAW_CONTEXT *pDC)
 737 {
 738     API_STATE *pState = &pDC->pState->state;
 739     uint32_t numScissors = pState->gsState.emitsViewportArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
 740     pState->scissorsTileAligned = true;
 741
 742     for (uint32_t index = 0; index < numScissors; ++index)
 743     {
 744         SWR_RECT &scissorInFixedPoint = pState->scissorsInFixedPoint[index];
 745
 746         // Set up scissor dimensions based on scissor or viewport
 747         if (pState->rastState.scissorEnable)
 748         {
 749             scissorInFixedPoint = pState->scissorRects[index];
 750         }
 751         else
 752         {
 753             // the vp width and height must be added to origin un-rounded then the result round to -inf.
 754             // The cast to int works for rounding assuming all [left, right, top, bottom] are positive.
 755             scissorInFixedPoint.xmin = (int32_t)pState->vp[index].x;
 756             scissorInFixedPoint.xmax = (int32_t)(pState->vp[index].x + pState->vp[index].width);
 757             scissorInFixedPoint.ymin = (int32_t)pState->vp[index].y;
 758             scissorInFixedPoint.ymax = (int32_t)(pState->vp[index].y + pState->vp[index].height);
 759         }
 760
 761         // Clamp to max rect
 762         scissorInFixedPoint &= g_MaxScissorRect;
 763
 764         // Test for tile alignment
 765         bool tileAligned;
 766         tileAligned  = (scissorInFixedPoint.xmin % KNOB_TILE_X_DIM) == 0;
 767         tileAligned &= (scissorInFixedPoint.ymin % KNOB_TILE_Y_DIM) == 0;
 768         tileAligned &= (scissorInFixedPoint.xmax % KNOB_TILE_X_DIM) == 0;
 769         tileAligned &= (scissorInFixedPoint.ymax % KNOB_TILE_Y_DIM) == 0;
 770
 771         pState->scissorsTileAligned &= tileAligned;
 772
 773         // Scale to fixed point
 774         scissorInFixedPoint.xmin *= FIXED_POINT_SCALE;
 775         scissorInFixedPoint.xmax *= FIXED_POINT_SCALE;
 776         scissorInFixedPoint.ymin *= FIXED_POINT_SCALE;
 777         scissorInFixedPoint.ymax *= FIXED_POINT_SCALE;
 778
 779         // Make scissor inclusive
 780         scissorInFixedPoint.xmax -= 1;
 781         scissorInFixedPoint.ymax -= 1;
 782     }
 783 }
 784
 785
 786 // templated backend function tables
 787 extern PFN_BACKEND_FUNC gBackendNullPs[SWR_MULTISAMPLE_TYPE_COUNT];
 788 extern PFN_BACKEND_FUNC gBackendSingleSample[SWR_INPUT_COVERAGE_COUNT][2][2];
 789 extern PFN_BACKEND_FUNC gBackendSampleRateTable[SWR_MULTISAMPLE_TYPE_COUNT][SWR_INPUT_COVERAGE_COUNT][2][2];
 790
 791 void SetupPipeline(DRAW_CONTEXT *pDC)
 792 {
 793     DRAW_STATE* pState = pDC->pState;
 794     const SWR_RASTSTATE &rastState = pState->state.rastState;
 795     const SWR_PS_STATE &psState = pState->state.psState;
 796     BACKEND_FUNCS& backendFuncs = pState->backendFuncs;
 797
 798     // setup backend
 799     if (psState.pfnPixelShader == nullptr)
 800     {
 801         backendFuncs.pfnBackend = gBackendNullPs[pState->state.rastState.sampleCount];
 802     }
 803     else
 804     {
 805         const uint32_t forcedSampleCount = (rastState.forcedSampleCount) ? 1 : 0;
 806         const bool bMultisampleEnable = ((rastState.sampleCount > SWR_MULTISAMPLE_1X) || forcedSampleCount) ? 1 : 0;
 807         const uint32_t centroid = ((psState.barycentricsMask & SWR_BARYCENTRIC_CENTROID_MASK) > 0) ? 1 : 0;
 808         const uint32_t canEarlyZ = (psState.forceEarlyZ || (!psState.writesODepth && !psState.usesUAV)) ? 1 : 0;
 809         SWR_BARYCENTRICS_MASK barycentricsMask = (SWR_BARYCENTRICS_MASK)psState.barycentricsMask;
 810
 811         // select backend function
 812         switch(psState.shadingRate)
 813         {
 814         case SWR_SHADING_RATE_PIXEL:
 815             if(bMultisampleEnable)
 816             {
 817                 // always need to generate I & J per sample for Z interpolation
 818                 barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_SAMPLE_MASK);
 819                 backendFuncs.pfnBackend = gBackendPixelRateTable[rastState.sampleCount][rastState.bIsCenterPattern][psState.inputCoverage]
 820                                                                 [centroid][forcedSampleCount][canEarlyZ]
 821                     ;
 822             }
 823             else
 824             {
 825                 // always need to generate I & J per pixel for Z interpolation
 826                 barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_PIXEL_MASK);
 827                 backendFuncs.pfnBackend = gBackendSingleSample[psState.inputCoverage][centroid][canEarlyZ];
 828             }
 829             break;
 830         case SWR_SHADING_RATE_SAMPLE:
 831             SWR_ASSERT(rastState.bIsCenterPattern != true);
 832             // always need to generate I & J per sample for Z interpolation
 833             barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_SAMPLE_MASK);
 834             backendFuncs.pfnBackend = gBackendSampleRateTable[rastState.sampleCount][psState.inputCoverage][centroid][canEarlyZ];
 835             break;
 836         default:
 837             SWR_ASSERT(0 && "Invalid shading rate");
 838             break;
 839         }
 840     }
 841
 842     PFN_PROCESS_PRIMS pfnBinner;
 843 #if USE_SIMD16_FRONTEND
 844     PFN_PROCESS_PRIMS_SIMD16 pfnBinner_simd16;
 845 #endif
 846     switch (pState->state.topology)
 847     {
 848     case TOP_POINT_LIST:
 849         pState->pfnProcessPrims = ClipPoints;
 850         pfnBinner = BinPoints;
 851 #if USE_SIMD16_FRONTEND
 852         pState->pfnProcessPrims_simd16 = ClipPoints_simd16;
 853         pfnBinner_simd16 = BinPoints_simd16;
 854 #endif
 855         break;
 856     case TOP_LINE_LIST:
 857     case TOP_LINE_STRIP:
 858     case TOP_LINE_LOOP:
 859     case TOP_LINE_LIST_ADJ:
 860     case TOP_LISTSTRIP_ADJ:
 861         pState->pfnProcessPrims = ClipLines;
 862         pfnBinner = BinLines;
 863 #if USE_SIMD16_FRONTEND
 864         pState->pfnProcessPrims_simd16 = ClipLines_simd16;
 865         pfnBinner_simd16 = BinLines_simd16;
 866 #endif
 867         break;
 868     default:
 869         pState->pfnProcessPrims = ClipTriangles;
 870         pfnBinner = GetBinTrianglesFunc((rastState.conservativeRast > 0));
 871 #if USE_SIMD16_FRONTEND
 872         pState->pfnProcessPrims_simd16 = ClipTriangles_simd16;
 873         pfnBinner_simd16 = GetBinTrianglesFunc_simd16((rastState.conservativeRast > 0));
 874 #endif
 875         break;
 876     };
 877
 878
 879     // disable clipper if viewport transform is disabled
 880     if (pState->state.frontendState.vpTransformDisable)
 881     {
 882         pState->pfnProcessPrims = pfnBinner;
 883 #if USE_SIMD16_FRONTEND
 884         pState->pfnProcessPrims_simd16 = pfnBinner_simd16;
 885 #endif
 886     }
 887
 888     if ((pState->state.psState.pfnPixelShader == nullptr) &&
 889         (pState->state.depthStencilState.depthTestEnable == FALSE) &&
 890         (pState->state.depthStencilState.depthWriteEnable == FALSE) &&
 891         (pState->state.depthStencilState.stencilTestEnable == FALSE) &&
 892         (pState->state.depthStencilState.stencilWriteEnable == FALSE) &&
 893         (pState->state.backendState.numAttributes == 0))
 894     {
 895         pState->pfnProcessPrims = nullptr;
 896 #if USE_SIMD16_FRONTEND
 897         pState->pfnProcessPrims_simd16 = nullptr;
 898 #endif
 899     }
 900
 901     if (pState->state.soState.rasterizerDisable == true)
 902     {
 903         pState->pfnProcessPrims = nullptr;
 904 #if USE_SIMD16_FRONTEND
 905         pState->pfnProcessPrims_simd16 = nullptr;
 906 #endif
 907     }
 908
 909
 910     // set up the frontend attribute count
 911     pState->state.feNumAttributes = 0;
 912     const SWR_BACKEND_STATE& backendState = pState->state.backendState;
 913     if (backendState.swizzleEnable)
 914     {
 915         // attribute swizzling is enabled, iterate over the map and record the max attribute used
 916         for (uint32_t i = 0; i < backendState.numAttributes; ++i)
 917         {
 918             pState->state.feNumAttributes = std::max(pState->state.feNumAttributes, (uint32_t)backendState.swizzleMap[i].sourceAttrib + 1);
 919         }
 920     }
 921     else
 922     {
 923         pState->state.feNumAttributes = pState->state.backendState.numAttributes;
 924     }
 925
 926     if (pState->state.soState.soEnable)
 927     {
 928         uint32_t streamMasks = 0;
 929         for (uint32_t i = 0; i < 4; ++i)
 930         {
 931             streamMasks |= pState->state.soState.streamMasks[i];
 932         }
 933
 934         DWORD maxAttrib;
 935         if (_BitScanReverse(&maxAttrib, streamMasks))
 936         {
 937             pState->state.feNumAttributes = std::max(pState->state.feNumAttributes, (uint32_t)(maxAttrib + 1));
 938         }
 939     }
 940
 941     // complicated logic to test for cases where we don't need backing hottile memory for a draw
 942     // have to check for the special case where depth/stencil test is enabled but depthwrite is disabled.
 943     pState->state.depthHottileEnable = ((!(pState->state.depthStencilState.depthTestEnable &&
 944                                            !pState->state.depthStencilState.depthWriteEnable &&
 945                                            !pState->state.depthBoundsState.depthBoundsTestEnable &&
 946                                            pState->state.depthStencilState.depthTestFunc == ZFUNC_ALWAYS)) &&
 947                                         (pState->state.depthStencilState.depthTestEnable ||
 948                                          pState->state.depthStencilState.depthWriteEnable ||
 949                                          pState->state.depthBoundsState.depthBoundsTestEnable)) ? true : false;
 950
 951     pState->state.stencilHottileEnable = (((!(pState->state.depthStencilState.stencilTestEnable &&
 952                                              !pState->state.depthStencilState.stencilWriteEnable &&
 953                                               pState->state.depthStencilState.stencilTestFunc == ZFUNC_ALWAYS)) ||
 954                                           // for stencil we have to check the double sided state as well
 955                                           (!(pState->state.depthStencilState.doubleSidedStencilTestEnable &&
 956                                              !pState->state.depthStencilState.stencilWriteEnable &&
 957                                               pState->state.depthStencilState.backfaceStencilTestFunc == ZFUNC_ALWAYS))) &&
 958                                           (pState->state.depthStencilState.stencilTestEnable  ||
 959                                            pState->state.depthStencilState.stencilWriteEnable)) ? true : false;
 960
 961     uint32_t numRTs = pState->state.psState.numRenderTargets;
 962     pState->state.colorHottileEnable = 0;
 963     if (psState.pfnPixelShader != nullptr)
 964     {
 965         for (uint32_t rt = 0; rt < numRTs; ++rt)
 966         {
 967             pState->state.colorHottileEnable |=
 968                 (!pState->state.blendState.renderTarget[rt].writeDisableAlpha ||
 969                  !pState->state.blendState.renderTarget[rt].writeDisableRed ||
 970                  !pState->state.blendState.renderTarget[rt].writeDisableGreen ||
 971                  !pState->state.blendState.renderTarget[rt].writeDisableBlue) ? (1 << rt) : 0;
 972         }
 973     }
 974
 975     // Setup depth quantization function
 976     if (pState->state.depthHottileEnable)
 977     {
 978         switch (pState->state.rastState.depthFormat)
 979         {
 980         case R32_FLOAT_X8X24_TYPELESS: pState->state.pfnQuantizeDepth = QuantizeDepth < R32_FLOAT_X8X24_TYPELESS > ; break;
 981         case R32_FLOAT: pState->state.pfnQuantizeDepth = QuantizeDepth < R32_FLOAT > ; break;
 982         case R24_UNORM_X8_TYPELESS: pState->state.pfnQuantizeDepth = QuantizeDepth < R24_UNORM_X8_TYPELESS > ; break;
 983         case R16_UNORM: pState->state.pfnQuantizeDepth = QuantizeDepth < R16_UNORM > ; break;
 984         default: SWR_INVALID("Unsupported depth format for depth quantiztion.");
 985             pState->state.pfnQuantizeDepth = QuantizeDepth < R32_FLOAT > ;
 986         }
 987     }
 988     else
 989     {
 990         // set up pass-through quantize if depth isn't enabled
 991         pState->state.pfnQuantizeDepth = QuantizeDepth < R32_FLOAT > ;
 992     }
 993 }
 994
 995 //////////////////////////////////////////////////////////////////////////
 996 /// @brief InitDraw
 997 /// @param pDC - Draw context to initialize for this draw.
 998 void InitDraw(
 999     DRAW_CONTEXT *pDC,
1000     bool isSplitDraw)
1001 {
1002     // We don't need to re-setup the scissors/pipeline state again for split draw.
1003     if (isSplitDraw == false)
1004     {
1005         SetupMacroTileScissors(pDC);
1006         SetupPipeline(pDC);
1007     }
1008
1009
1010 }
1011
1012 //////////////////////////////////////////////////////////////////////////
1013 /// @brief We can split the draw for certain topologies for better performance.
1014 /// @param totalVerts - Total vertices for draw
1015 /// @param topology - Topology used for draw
1016 uint32_t MaxVertsPerDraw(
1017     DRAW_CONTEXT* pDC,
1018     uint32_t totalVerts,
1019     PRIMITIVE_TOPOLOGY topology)
1020 {
1021     API_STATE& state = pDC->pState->state;
1022
1023     uint32_t vertsPerDraw = totalVerts;
1024
1025     if (state.soState.soEnable)
1026     {
1027         return totalVerts;
1028     }
1029
1030     switch (topology)
1031     {
1032     case TOP_POINT_LIST:
1033     case TOP_TRIANGLE_LIST:
1034         vertsPerDraw = KNOB_MAX_PRIMS_PER_DRAW;
1035         break;
1036
1037     case TOP_PATCHLIST_1:
1038     case TOP_PATCHLIST_2:
1039     case TOP_PATCHLIST_3:
1040     case TOP_PATCHLIST_4:
1041     case TOP_PATCHLIST_5:
1042     case TOP_PATCHLIST_6:
1043     case TOP_PATCHLIST_7:
1044     case TOP_PATCHLIST_8:
1045     case TOP_PATCHLIST_9:
1046     case TOP_PATCHLIST_10:
1047     case TOP_PATCHLIST_11:
1048     case TOP_PATCHLIST_12:
1049     case TOP_PATCHLIST_13:
1050     case TOP_PATCHLIST_14:
1051     case TOP_PATCHLIST_15:
1052     case TOP_PATCHLIST_16:
1053     case TOP_PATCHLIST_17:
1054     case TOP_PATCHLIST_18:
1055     case TOP_PATCHLIST_19:
1056     case TOP_PATCHLIST_20:
1057     case TOP_PATCHLIST_21:
1058     case TOP_PATCHLIST_22:
1059     case TOP_PATCHLIST_23:
1060     case TOP_PATCHLIST_24:
1061     case TOP_PATCHLIST_25:
1062     case TOP_PATCHLIST_26:
1063     case TOP_PATCHLIST_27:
1064     case TOP_PATCHLIST_28:
1065     case TOP_PATCHLIST_29:
1066     case TOP_PATCHLIST_30:
1067     case TOP_PATCHLIST_31:
1068     case TOP_PATCHLIST_32:
1069         if (pDC->pState->state.tsState.tsEnable)
1070         {
1071             uint32_t vertsPerPrim = topology - TOP_PATCHLIST_BASE;
1072             vertsPerDraw = vertsPerPrim * KNOB_MAX_TESS_PRIMS_PER_DRAW;
1073         }
1074         break;
1075
1076     // The Primitive Assembly code can only handle 1 RECT at a time.
1077     case TOP_RECT_LIST:
1078         vertsPerDraw = 3;
1079         break;
1080
1081     default:
1082         // We are not splitting up draws for other topologies.
1083         break;
1084     }
1085
1086     return vertsPerDraw;
1087 }
1088
1089
1090 //////////////////////////////////////////////////////////////////////////
1091 /// @brief DrawInstanced
1092 /// @param hContext - Handle passed back from SwrCreateContext
1093 /// @param topology - Specifies topology for draw.
1094 /// @param numVerts - How many vertices to read sequentially from vertex data (per instance).
1095 /// @param startVertex - Specifies start vertex for draw. (vertex data)
1096 /// @param numInstances - How many instances to render.
1097 /// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data)
1098 void DrawInstanced(
1099     HANDLE hContext,
1100     PRIMITIVE_TOPOLOGY topology,
1101     uint32_t numVertices,
1102     uint32_t startVertex,
1103     uint32_t numInstances = 1,
1104     uint32_t startInstance = 0)
1105 {
1106     if (KNOB_TOSS_DRAW)
1107     {
1108         return;
1109     }
1110
1111     SWR_CONTEXT *pContext = GetContext(hContext);
1112     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1113
1114     AR_API_BEGIN(APIDraw, pDC->drawId);
1115     AR_API_EVENT(DrawInstancedEvent(pDC->drawId, topology, numVertices, startVertex, numInstances, startInstance));
1116
1117     uint32_t maxVertsPerDraw = MaxVertsPerDraw(pDC, numVertices, topology);
1118     uint32_t primsPerDraw = GetNumPrims(topology, maxVertsPerDraw);
1119     uint32_t remainingVerts = numVertices;
1120
1121     API_STATE    *pState = &pDC->pState->state;
1122     pState->topology = topology;
1123     pState->forceFront = false;
1124
1125     // disable culling for points/lines
1126     uint32_t oldCullMode = pState->rastState.cullMode;
1127     if (topology == TOP_POINT_LIST)
1128     {
1129         pState->rastState.cullMode = SWR_CULLMODE_NONE;
1130         pState->forceFront = true;
1131     }
1132     else if (topology == TOP_RECT_LIST)
1133     {
1134         pState->rastState.cullMode = SWR_CULLMODE_NONE;
1135     }
1136
1137     int draw = 0;
1138     while (remainingVerts)
1139     {
1140         uint32_t numVertsForDraw = (remainingVerts < maxVertsPerDraw) ?
1141         remainingVerts : maxVertsPerDraw;
1142
1143         bool isSplitDraw = (draw > 0) ? true : false;
1144         DRAW_CONTEXT* pDC = GetDrawContext(pContext, isSplitDraw);
1145         InitDraw(pDC, isSplitDraw);
1146
1147         pDC->FeWork.type = DRAW;
1148         pDC->FeWork.pfnWork = GetProcessDrawFunc(
1149             false,  // IsIndexed
1150             false, // bEnableCutIndex
1151             pState->tsState.tsEnable,
1152             pState->gsState.gsEnable,
1153             pState->soState.soEnable,
1154             pDC->pState->pfnProcessPrims != nullptr);
1155         pDC->FeWork.desc.draw.numVerts = numVertsForDraw;
1156         pDC->FeWork.desc.draw.startVertex = startVertex;
1157         pDC->FeWork.desc.draw.numInstances = numInstances;
1158         pDC->FeWork.desc.draw.startInstance = startInstance;
1159         pDC->FeWork.desc.draw.startPrimID = draw * primsPerDraw;
1160         pDC->FeWork.desc.draw.startVertexID = draw * maxVertsPerDraw;
1161
1162         pDC->cleanupState = (remainingVerts == numVertsForDraw);
1163
1164         //enqueue DC
1165         QueueDraw(pContext);
1166
1167         AR_API_EVENT(DrawInstancedSplitEvent(pDC->drawId));
1168
1169         remainingVerts -= numVertsForDraw;
1170         draw++;
1171     }
1172
1173     // restore culling state
1174     pDC = GetDrawContext(pContext);
1175     pDC->pState->state.rastState.cullMode = oldCullMode;
1176
1177     AR_API_END(APIDraw, numVertices * numInstances);
1178 }
1179
1180 //////////////////////////////////////////////////////////////////////////
1181 /// @brief SwrDraw
1182 /// @param hContext - Handle passed back from SwrCreateContext
1183 /// @param topology - Specifies topology for draw.
1184 /// @param startVertex - Specifies start vertex in vertex buffer for draw.
1185 /// @param primCount - Number of vertices.
1186 void SwrDraw(
1187     HANDLE hContext,
1188     PRIMITIVE_TOPOLOGY topology,
1189     uint32_t startVertex,
1190     uint32_t numVertices)
1191 {
1192     DrawInstanced(hContext, topology, numVertices, startVertex);
1193 }
1194
1195 //////////////////////////////////////////////////////////////////////////
1196 /// @brief SwrDrawInstanced
1197 /// @param hContext - Handle passed back from SwrCreateContext
1198 /// @param topology - Specifies topology for draw.
1199 /// @param numVertsPerInstance - How many vertices to read sequentially from vertex data.
1200 /// @param numInstances - How many instances to render.
1201 /// @param startVertex - Specifies start vertex for draw. (vertex data)
1202 /// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data)
1203 void SwrDrawInstanced(
1204     HANDLE hContext,
1205     PRIMITIVE_TOPOLOGY topology,
1206     uint32_t numVertsPerInstance,
1207     uint32_t numInstances,
1208     uint32_t startVertex,
1209     uint32_t startInstance
1210     )
1211 {
1212     DrawInstanced(hContext, topology, numVertsPerInstance, startVertex, numInstances, startInstance);
1213 }
1214
1215 //////////////////////////////////////////////////////////////////////////
1216 /// @brief DrawIndexedInstanced
1217 /// @param hContext - Handle passed back from SwrCreateContext
1218 /// @param topology - Specifies topology for draw.
1219 /// @param numIndices - Number of indices to read sequentially from index buffer.
1220 /// @param indexOffset - Starting index into index buffer.
1221 /// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed.
1222 /// @param numInstances - Number of instances to render.
1223 /// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data)
1224 void DrawIndexedInstance(
1225     HANDLE hContext,
1226     PRIMITIVE_TOPOLOGY topology,
1227     uint32_t numIndices,
1228     uint32_t indexOffset,
1229     int32_t baseVertex,
1230     uint32_t numInstances = 1,
1231     uint32_t startInstance = 0)
1232 {
1233     if (KNOB_TOSS_DRAW)
1234     {
1235         return;
1236     }
1237
1238     SWR_CONTEXT *pContext = GetContext(hContext);
1239     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1240     API_STATE* pState = &pDC->pState->state;
1241
1242     AR_API_BEGIN(APIDrawIndexed, pDC->drawId);
1243     AR_API_EVENT(DrawIndexedInstancedEvent(pDC->drawId, topology, numIndices, indexOffset, baseVertex, numInstances, startInstance));
1244
1245     uint32_t maxIndicesPerDraw = MaxVertsPerDraw(pDC, numIndices, topology);
1246     uint32_t primsPerDraw = GetNumPrims(topology, maxIndicesPerDraw);
1247     uint32_t remainingIndices = numIndices;
1248
1249     uint32_t indexSize = 0;
1250     switch (pState->indexBuffer.format)
1251     {
1252     case R32_UINT: indexSize = sizeof(uint32_t); break;
1253     case R16_UINT: indexSize = sizeof(uint16_t); break;
1254     case R8_UINT: indexSize = sizeof(uint8_t); break;
1255     default:
1256         SWR_INVALID("Invalid index buffer format: %d", pState->indexBuffer.format);
1257     }
1258
1259     int draw = 0;
1260     uint8_t *pIB = (uint8_t*)pState->indexBuffer.pIndices;
1261     pIB += (uint64_t)indexOffset * (uint64_t)indexSize;
1262
1263     pState->topology = topology;
1264     pState->forceFront = false;
1265
1266     // disable culling for points/lines
1267     uint32_t oldCullMode = pState->rastState.cullMode;
1268     if (topology == TOP_POINT_LIST)
1269     {
1270         pState->rastState.cullMode = SWR_CULLMODE_NONE;
1271         pState->forceFront = true;
1272     }
1273     else if (topology == TOP_RECT_LIST)
1274     {
1275         pState->rastState.cullMode = SWR_CULLMODE_NONE;
1276     }
1277
1278     while (remainingIndices)
1279     {
1280         uint32_t numIndicesForDraw = (remainingIndices < maxIndicesPerDraw) ?
1281         remainingIndices : maxIndicesPerDraw;
1282
1283         // When breaking up draw, we need to obtain new draw context for each iteration.
1284         bool isSplitDraw = (draw > 0) ? true : false;
1285
1286         pDC = GetDrawContext(pContext, isSplitDraw);
1287         InitDraw(pDC, isSplitDraw);
1288
1289         pDC->FeWork.type = DRAW;
1290         pDC->FeWork.pfnWork = GetProcessDrawFunc(
1291             true,   // IsIndexed
1292             pState->frontendState.bEnableCutIndex,
1293             pState->tsState.tsEnable,
1294             pState->gsState.gsEnable,
1295             pState->soState.soEnable,
1296             pDC->pState->pfnProcessPrims != nullptr);
1297         pDC->FeWork.desc.draw.pDC = pDC;
1298         pDC->FeWork.desc.draw.numIndices = numIndicesForDraw;
1299         pDC->FeWork.desc.draw.pIB = (int*)pIB;
1300         pDC->FeWork.desc.draw.type = pDC->pState->state.indexBuffer.format;
1301
1302         pDC->FeWork.desc.draw.numInstances = numInstances;
1303         pDC->FeWork.desc.draw.startInstance = startInstance;
1304         pDC->FeWork.desc.draw.baseVertex = baseVertex;
1305         pDC->FeWork.desc.draw.startPrimID = draw * primsPerDraw;
1306
1307         pDC->cleanupState = (remainingIndices == numIndicesForDraw);
1308
1309         //enqueue DC
1310         QueueDraw(pContext);
1311
1312         AR_API_EVENT(DrawIndexedInstancedSplitEvent(pDC->drawId));
1313
1314         pIB += maxIndicesPerDraw * indexSize;
1315         remainingIndices -= numIndicesForDraw;
1316         draw++;
1317     }
1318
1319     // Restore culling state
1320     pDC = GetDrawContext(pContext);
1321     pDC->pState->state.rastState.cullMode = oldCullMode;
1322
1323     AR_API_END(APIDrawIndexed, numIndices * numInstances);
1324 }
1325
1326
1327 //////////////////////////////////////////////////////////////////////////
1328 /// @brief DrawIndexed
1329 /// @param hContext - Handle passed back from SwrCreateContext
1330 /// @param topology - Specifies topology for draw.
1331 /// @param numIndices - Number of indices to read sequentially from index buffer.
1332 /// @param indexOffset - Starting index into index buffer.
1333 /// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed.
1334 void SwrDrawIndexed(
1335     HANDLE hContext,
1336     PRIMITIVE_TOPOLOGY topology,
1337     uint32_t numIndices,
1338     uint32_t indexOffset,
1339     int32_t baseVertex
1340     )
1341 {
1342     DrawIndexedInstance(hContext, topology, numIndices, indexOffset, baseVertex);
1343 }
1344
1345 //////////////////////////////////////////////////////////////////////////
1346 /// @brief SwrDrawIndexedInstanced
1347 /// @param hContext - Handle passed back from SwrCreateContext
1348 /// @param topology - Specifies topology for draw.
1349 /// @param numIndices - Number of indices to read sequentially from index buffer.
1350 /// @param numInstances - Number of instances to render.
1351 /// @param indexOffset - Starting index into index buffer.
1352 /// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed.
1353 /// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data)
1354 void SwrDrawIndexedInstanced(
1355     HANDLE hContext,
1356     PRIMITIVE_TOPOLOGY topology,
1357     uint32_t numIndices,
1358     uint32_t numInstances,
1359     uint32_t indexOffset,
1360     int32_t baseVertex,
1361     uint32_t startInstance)
1362 {
1363     DrawIndexedInstance(hContext, topology, numIndices, indexOffset, baseVertex, numInstances, startInstance);
1364 }
1365
1366 //////////////////////////////////////////////////////////////////////////
1367 /// @brief SwrInvalidateTiles
1368 /// @param hContext - Handle passed back from SwrCreateContext
1369 /// @param attachmentMask - The mask specifies which surfaces attached to the hottiles to invalidate.
1370 /// @param invalidateRect - The pixel-coordinate rectangle to invalidate.  This will be expanded to
1371 ///                         be hottile size-aligned.
1372 void SWR_API SwrInvalidateTiles(
1373     HANDLE hContext,
1374     uint32_t attachmentMask,
1375     const SWR_RECT& invalidateRect)
1376 {
1377     if (KNOB_TOSS_DRAW)
1378     {
1379         return;
1380     }
1381
1382     SWR_CONTEXT *pContext = GetContext(hContext);
1383     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1384
1385     pDC->FeWork.type = DISCARDINVALIDATETILES;
1386     pDC->FeWork.pfnWork = ProcessDiscardInvalidateTiles;
1387     pDC->FeWork.desc.discardInvalidateTiles.attachmentMask = attachmentMask;
1388     pDC->FeWork.desc.discardInvalidateTiles.rect = invalidateRect;
1389     pDC->FeWork.desc.discardInvalidateTiles.rect &= g_MaxScissorRect;
1390     pDC->FeWork.desc.discardInvalidateTiles.newTileState = SWR_TILE_INVALID;
1391     pDC->FeWork.desc.discardInvalidateTiles.createNewTiles = false;
1392     pDC->FeWork.desc.discardInvalidateTiles.fullTilesOnly = false;
1393
1394     //enqueue
1395     QueueDraw(pContext);
1396
1397     AR_API_EVENT(SwrInvalidateTilesEvent(pDC->drawId));
1398 }
1399
1400 //////////////////////////////////////////////////////////////////////////
1401 /// @brief SwrDiscardRect
1402 /// @param hContext - Handle passed back from SwrCreateContext
1403 /// @param attachmentMask - The mask specifies which surfaces attached to the hottiles to discard.
1404 /// @param rect - The pixel-coordinate rectangle to discard.  Only fully-covered hottiles will be
1405 ///               discarded.
1406 void SWR_API SwrDiscardRect(
1407     HANDLE hContext,
1408     uint32_t attachmentMask,
1409     const SWR_RECT& rect)
1410 {
1411     if (KNOB_TOSS_DRAW)
1412     {
1413         return;
1414     }
1415
1416     SWR_CONTEXT *pContext = GetContext(hContext);
1417     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1418
1419     // Queue a load to the hottile
1420     pDC->FeWork.type = DISCARDINVALIDATETILES;
1421     pDC->FeWork.pfnWork = ProcessDiscardInvalidateTiles;
1422     pDC->FeWork.desc.discardInvalidateTiles.attachmentMask = attachmentMask;
1423     pDC->FeWork.desc.discardInvalidateTiles.rect = rect;
1424     pDC->FeWork.desc.discardInvalidateTiles.rect &= g_MaxScissorRect;
1425     pDC->FeWork.desc.discardInvalidateTiles.newTileState = SWR_TILE_RESOLVED;
1426     pDC->FeWork.desc.discardInvalidateTiles.createNewTiles = true;
1427     pDC->FeWork.desc.discardInvalidateTiles.fullTilesOnly = true;
1428
1429     //enqueue
1430     QueueDraw(pContext);
1431
1432     AR_API_EVENT(SwrDiscardRectEvent(pDC->drawId));
1433 }
1434
1435 //////////////////////////////////////////////////////////////////////////
1436 /// @brief SwrDispatch
1437 /// @param hContext - Handle passed back from SwrCreateContext
1438 /// @param threadGroupCountX - Number of thread groups dispatched in X direction
1439 /// @param threadGroupCountY - Number of thread groups dispatched in Y direction
1440 /// @param threadGroupCountZ - Number of thread groups dispatched in Z direction
1441 void SwrDispatch(
1442     HANDLE hContext,
1443     uint32_t threadGroupCountX,
1444     uint32_t threadGroupCountY,
1445     uint32_t threadGroupCountZ)
1446 {
1447     if (KNOB_TOSS_DRAW)
1448     {
1449         return;
1450     }
1451
1452     SWR_CONTEXT *pContext = GetContext(hContext);
1453     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1454
1455     AR_API_BEGIN(APIDispatch, pDC->drawId);
1456     AR_API_EVENT(DispatchEvent(pDC->drawId, threadGroupCountX, threadGroupCountY, threadGroupCountZ));
1457     pDC->isCompute = true;      // This is a compute context.
1458
1459     COMPUTE_DESC* pTaskData = (COMPUTE_DESC*)pDC->pArena->AllocAligned(sizeof(COMPUTE_DESC), 64);
1460
1461     pTaskData->threadGroupCountX = threadGroupCountX;
1462     pTaskData->threadGroupCountY = threadGroupCountY;
1463     pTaskData->threadGroupCountZ = threadGroupCountZ;
1464
1465     uint32_t totalThreadGroups = threadGroupCountX * threadGroupCountY * threadGroupCountZ;
1466     uint32_t dcIndex = pDC->drawId % KNOB_MAX_DRAWS_IN_FLIGHT;
1467     pDC->pDispatch = &pContext->pDispatchQueueArray[dcIndex];
1468     pDC->pDispatch->initialize(totalThreadGroups, pTaskData, &ProcessComputeBE);
1469
1470     QueueDispatch(pContext);
1471     AR_API_END(APIDispatch, threadGroupCountX * threadGroupCountY * threadGroupCountZ);
1472 }
1473
1474 // Deswizzles, converts and stores current contents of the hot tiles to surface
1475 // described by pState
1476 void SWR_API SwrStoreTiles(
1477     HANDLE hContext,
1478     uint32_t attachmentMask,
1479     SWR_TILE_STATE postStoreTileState,
1480     const SWR_RECT& storeRect)
1481 {
1482     if (KNOB_TOSS_DRAW)
1483     {
1484         return;
1485     }
1486
1487     SWR_CONTEXT *pContext = GetContext(hContext);
1488     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1489
1490     AR_API_BEGIN(APIStoreTiles, pDC->drawId);
1491
1492     pDC->FeWork.type = STORETILES;
1493     pDC->FeWork.pfnWork = ProcessStoreTiles;
1494     pDC->FeWork.desc.storeTiles.attachmentMask = attachmentMask;
1495     pDC->FeWork.desc.storeTiles.postStoreTileState = postStoreTileState;
1496     pDC->FeWork.desc.storeTiles.rect = storeRect;
1497     pDC->FeWork.desc.storeTiles.rect &= g_MaxScissorRect;
1498
1499     //enqueue
1500     QueueDraw(pContext);
1501
1502     AR_API_EVENT(SwrStoreTilesEvent(pDC->drawId));
1503
1504     AR_API_END(APIStoreTiles, 1);
1505 }
1506
1507 //////////////////////////////////////////////////////////////////////////
1508 /// @brief SwrClearRenderTarget - Clear attached render targets / depth / stencil
1509 /// @param hContext - Handle passed back from SwrCreateContext
1510 /// @param attachmentMask - combination of SWR_ATTACHMENT_*_BIT attachments to clear
1511 /// @param renderTargetArrayIndex - the RT array index to clear
1512 /// @param clearColor - color use for clearing render targets
1513 /// @param z - depth value use for clearing depth buffer
1514 /// @param stencil - stencil value used for clearing stencil buffer
1515 /// @param clearRect - The pixel-coordinate rectangle to clear in all cleared buffers
1516 void SWR_API SwrClearRenderTarget(
1517     HANDLE hContext,
1518     uint32_t attachmentMask,
1519     uint32_t renderTargetArrayIndex,
1520     const float clearColor[4],
1521     float z,
1522     uint8_t stencil,
1523     const SWR_RECT& clearRect)
1524 {
1525     if (KNOB_TOSS_DRAW)
1526     {
1527         return;
1528     }
1529
1530     SWR_CONTEXT *pContext = GetContext(hContext);
1531     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1532
1533     AR_API_BEGIN(APIClearRenderTarget, pDC->drawId);
1534
1535     pDC->FeWork.type = CLEAR;
1536     pDC->FeWork.pfnWork = ProcessClear;
1537     pDC->FeWork.desc.clear.rect = clearRect;
1538     pDC->FeWork.desc.clear.rect &= g_MaxScissorRect;
1539     pDC->FeWork.desc.clear.attachmentMask = attachmentMask;
1540     pDC->FeWork.desc.clear.renderTargetArrayIndex = renderTargetArrayIndex;
1541     pDC->FeWork.desc.clear.clearDepth = z;
1542     pDC->FeWork.desc.clear.clearRTColor[0] = clearColor[0];
1543     pDC->FeWork.desc.clear.clearRTColor[1] = clearColor[1];
1544     pDC->FeWork.desc.clear.clearRTColor[2] = clearColor[2];
1545     pDC->FeWork.desc.clear.clearRTColor[3] = clearColor[3];
1546     pDC->FeWork.desc.clear.clearStencil = stencil;
1547
1548     // enqueue draw
1549     QueueDraw(pContext);
1550
1551     AR_API_END(APIClearRenderTarget, 1);
1552 }
1553
1554 //////////////////////////////////////////////////////////////////////////
1555 /// @brief Returns a pointer to the private context state for the current
1556 ///        draw operation. This is used for external componets such as the
1557 ///        sampler.
1558 ///        SWR is responsible for the allocation of the private context state.
1559 /// @param hContext - Handle passed back from SwrCreateContext
1560 VOID* SwrGetPrivateContextState(
1561     HANDLE hContext)
1562 {
1563     SWR_CONTEXT* pContext = GetContext(hContext);
1564     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1565     DRAW_STATE* pState = pDC->pState;
1566
1567     if (pState->pPrivateState == nullptr)
1568     {
1569         pState->pPrivateState = pState->pArena->AllocAligned(pContext->privateStateSize, KNOB_SIMD_WIDTH*sizeof(float));
1570     }
1571
1572     return pState->pPrivateState;
1573 }
1574
1575 //////////////////////////////////////////////////////////////////////////
1576 /// @brief Clients can use this to allocate memory for draw/dispatch
1577 ///        operations. The memory will automatically be freed once operation
1578 ///        has completed. Client can use this to allocate binding tables,
1579 ///        etc. needed for shader execution.
1580 /// @param hContext - Handle passed back from SwrCreateContext
1581 /// @param size - Size of allocation
1582 /// @param align - Alignment needed for allocation.
1583 VOID* SwrAllocDrawContextMemory(
1584     HANDLE hContext,
1585     uint32_t size,
1586     uint32_t align)
1587 {
1588     SWR_CONTEXT* pContext = GetContext(hContext);
1589     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1590
1591     return pDC->pState->pArena->AllocAligned(size, align);
1592 }
1593
1594 //////////////////////////////////////////////////////////////////////////
1595 /// @brief Enables stats counting
1596 /// @param hContext - Handle passed back from SwrCreateContext
1597 /// @param enable - If true then counts are incremented.
1598 void SwrEnableStatsFE(
1599     HANDLE hContext,
1600     bool enable)
1601 {
1602     SWR_CONTEXT *pContext = GetContext(hContext);
1603     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1604
1605     pDC->pState->state.enableStatsFE = enable;
1606 }
1607
1608 //////////////////////////////////////////////////////////////////////////
1609 /// @brief Enables stats counting
1610 /// @param hContext - Handle passed back from SwrCreateContext
1611 /// @param enable - If true then counts are incremented.
1612 void SwrEnableStatsBE(
1613     HANDLE hContext,
1614     bool enable)
1615 {
1616     SWR_CONTEXT *pContext = GetContext(hContext);
1617     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1618
1619     pDC->pState->state.enableStatsBE = enable;
1620 }
1621
1622 //////////////////////////////////////////////////////////////////////////
1623 /// @brief Mark end of frame - used for performance profiling
1624 /// @param hContext - Handle passed back from SwrCreateContext
1625 void SWR_API SwrEndFrame(
1626     HANDLE hContext)
1627 {
1628     SWR_CONTEXT *pContext = GetContext(hContext);
1629     DRAW_CONTEXT* pDC = GetDrawContext(pContext);
1630     (void)pDC; // var used
1631
1632     RDTSC_ENDFRAME();
1633     AR_API_EVENT(FrameEndEvent(pContext->frameCount, pDC->drawId));
1634
1635     pContext->frameCount++;
1636 }
1637
1638 void InitSimLoadTilesTable();
1639 void InitSimStoreTilesTable();
1640 void InitSimClearTilesTable();
1641
1642 void InitClearTilesTable();
1643 void InitBackendFuncTables();
1644
1645 //////////////////////////////////////////////////////////////////////////
1646 /// @brief Initialize swr backend and memory internal tables
1647 void SwrInit()
1648 {
1649     InitSimLoadTilesTable();
1650     InitSimStoreTilesTable();
1651     InitSimClearTilesTable();
1652
1653     InitClearTilesTable();
1654     InitBackendFuncTables();
1655 }
1656
1657 void SwrGetInterface(SWR_INTERFACE &out_funcs)
1658 {
1659     out_funcs.pfnSwrCreateContext = SwrCreateContext;
1660     out_funcs.pfnSwrDestroyContext = SwrDestroyContext;
1661     out_funcs.pfnSwrSaveState = SwrSaveState;
1662     out_funcs.pfnSwrRestoreState = SwrRestoreState;
1663     out_funcs.pfnSwrSync = SwrSync;
1664     out_funcs.pfnSwrWaitForIdle = SwrWaitForIdle;
1665     out_funcs.pfnSwrWaitForIdleFE = SwrWaitForIdleFE;
1666     out_funcs.pfnSwrSetVertexBuffers = SwrSetVertexBuffers;
1667     out_funcs.pfnSwrSetIndexBuffer = SwrSetIndexBuffer;
1668     out_funcs.pfnSwrSetFetchFunc = SwrSetFetchFunc;
1669     out_funcs.pfnSwrSetSoFunc = SwrSetSoFunc;
1670     out_funcs.pfnSwrSetSoState = SwrSetSoState;
1671     out_funcs.pfnSwrSetSoBuffers = SwrSetSoBuffers;
1672     out_funcs.pfnSwrSetVertexFunc = SwrSetVertexFunc;
1673     out_funcs.pfnSwrSetFrontendState = SwrSetFrontendState;
1674     out_funcs.pfnSwrSetGsState = SwrSetGsState;
1675     out_funcs.pfnSwrSetGsFunc = SwrSetGsFunc;
1676     out_funcs.pfnSwrSetCsFunc = SwrSetCsFunc;
1677     out_funcs.pfnSwrSetTsState = SwrSetTsState;
1678     out_funcs.pfnSwrSetHsFunc = SwrSetHsFunc;
1679     out_funcs.pfnSwrSetDsFunc = SwrSetDsFunc;
1680     out_funcs.pfnSwrSetDepthStencilState = SwrSetDepthStencilState;
1681     out_funcs.pfnSwrSetBackendState = SwrSetBackendState;
1682     out_funcs.pfnSwrSetDepthBoundsState = SwrSetDepthBoundsState;
1683     out_funcs.pfnSwrSetPixelShaderState = SwrSetPixelShaderState;
1684     out_funcs.pfnSwrSetBlendState = SwrSetBlendState;
1685     out_funcs.pfnSwrSetBlendFunc = SwrSetBlendFunc;
1686     out_funcs.pfnSwrDraw = SwrDraw;
1687     out_funcs.pfnSwrDrawInstanced = SwrDrawInstanced;
1688     out_funcs.pfnSwrDrawIndexed = SwrDrawIndexed;
1689     out_funcs.pfnSwrDrawIndexedInstanced = SwrDrawIndexedInstanced;
1690     out_funcs.pfnSwrInvalidateTiles = SwrInvalidateTiles;
1691     out_funcs.pfnSwrDiscardRect = SwrDiscardRect;
1692     out_funcs.pfnSwrDispatch = SwrDispatch;
1693     out_funcs.pfnSwrStoreTiles = SwrStoreTiles;
1694     out_funcs.pfnSwrClearRenderTarget = SwrClearRenderTarget;
1695     out_funcs.pfnSwrSetRastState = SwrSetRastState;
1696     out_funcs.pfnSwrSetViewports = SwrSetViewports;
1697     out_funcs.pfnSwrSetScissorRects = SwrSetScissorRects;
1698     out_funcs.pfnSwrGetPrivateContextState = SwrGetPrivateContextState;
1699     out_funcs.pfnSwrAllocDrawContextMemory = SwrAllocDrawContextMemory;
1700     out_funcs.pfnSwrEnableStatsFE = SwrEnableStatsFE;
1701     out_funcs.pfnSwrEnableStatsBE = SwrEnableStatsBE;
1702     out_funcs.pfnSwrEndFrame = SwrEndFrame;
1703     out_funcs.pfnSwrInit = SwrInit;
1704     out_funcs.pfnSwrLoadHotTile = SwrLoadHotTile;
1705     out_funcs.pfnSwrStoreHotTileToSurface = SwrStoreHotTileToSurface;
1706     out_funcs.pfnSwrStoreHotTileClear = SwrStoreHotTileClear;
1707 }