#include "core/clip.h"
#include "core/utils.h"
-#include "common/simdintrin.h"
#include "common/os.h"
static const SWR_RECT g_MaxScissorRect = { 0, 0, KNOB_MAX_SCISSOR_X, KNOB_MAX_SCISSOR_Y };
memset(pContextMem, 0, sizeof(SWR_CONTEXT));
SWR_CONTEXT *pContext = new (pContextMem) SWR_CONTEXT();
- pContext->driverType = pCreateInfo->driver;
pContext->privateStateSize = pCreateInfo->privateStateSize;
- pContext->dcRing.Init(KNOB_MAX_DRAWS_IN_FLIGHT);
- pContext->dsRing.Init(KNOB_MAX_DRAWS_IN_FLIGHT);
+ pContext->MAX_DRAWS_IN_FLIGHT = KNOB_MAX_DRAWS_IN_FLIGHT;
+ if (pCreateInfo->MAX_DRAWS_IN_FLIGHT != 0)
+ {
+ pContext->MAX_DRAWS_IN_FLIGHT = pCreateInfo->MAX_DRAWS_IN_FLIGHT;
+ }
- pContext->pMacroTileManagerArray = (MacroTileMgr*)AlignedMalloc(sizeof(MacroTileMgr) * KNOB_MAX_DRAWS_IN_FLIGHT, 64);
- pContext->pDispatchQueueArray = (DispatchQueue*)AlignedMalloc(sizeof(DispatchQueue) * KNOB_MAX_DRAWS_IN_FLIGHT, 64);
+ pContext->dcRing.Init(pContext->MAX_DRAWS_IN_FLIGHT);
+ pContext->dsRing.Init(pContext->MAX_DRAWS_IN_FLIGHT);
- for (uint32_t dc = 0; dc < KNOB_MAX_DRAWS_IN_FLIGHT; ++dc)
+ pContext->pMacroTileManagerArray = (MacroTileMgr*)AlignedMalloc(sizeof(MacroTileMgr) * pContext->MAX_DRAWS_IN_FLIGHT, 64);
+ pContext->pDispatchQueueArray = (DispatchQueue*)AlignedMalloc(sizeof(DispatchQueue) * pContext->MAX_DRAWS_IN_FLIGHT, 64);
+
+ for (uint32_t dc = 0; dc < pContext->MAX_DRAWS_IN_FLIGHT; ++dc)
{
pContext->dcRing[dc].pArena = new CachingArena(pContext->cachingArenaAllocator);
new (&pContext->pMacroTileManagerArray[dc]) MacroTileMgr(*pContext->dcRing[dc].pArena);
pContext->dsRing[dc].pArena = new CachingArena(pContext->cachingArenaAllocator);
}
- pContext->threadInfo.MAX_WORKER_THREADS = KNOB_MAX_WORKER_THREADS;
- pContext->threadInfo.MAX_NUMA_NODES = KNOB_MAX_NUMA_NODES;
- pContext->threadInfo.MAX_CORES_PER_NUMA_NODE = KNOB_MAX_CORES_PER_NUMA_NODE;
- pContext->threadInfo.MAX_THREADS_PER_CORE = KNOB_MAX_THREADS_PER_CORE;
- pContext->threadInfo.SINGLE_THREADED = KNOB_SINGLE_THREADED;
-
if (pCreateInfo->pThreadInfo)
{
pContext->threadInfo = *pCreateInfo->pThreadInfo;
}
+ else
+ {
+ pContext->threadInfo.MAX_WORKER_THREADS = KNOB_MAX_WORKER_THREADS;
+ pContext->threadInfo.BASE_NUMA_NODE = KNOB_BASE_NUMA_NODE;
+ pContext->threadInfo.BASE_CORE = KNOB_BASE_CORE;
+ pContext->threadInfo.BASE_THREAD = KNOB_BASE_THREAD;
+ pContext->threadInfo.MAX_NUMA_NODES = KNOB_MAX_NUMA_NODES;
+ pContext->threadInfo.MAX_CORES_PER_NUMA_NODE = KNOB_MAX_CORES_PER_NUMA_NODE;
+ pContext->threadInfo.MAX_THREADS_PER_CORE = KNOB_MAX_THREADS_PER_CORE;
+ pContext->threadInfo.SINGLE_THREADED = KNOB_SINGLE_THREADED;
+ }
+
+ if (pCreateInfo->pApiThreadInfo)
+ {
+ pContext->apiThreadInfo = *pCreateInfo->pApiThreadInfo;
+ }
+ else
+ {
+ pContext->apiThreadInfo.bindAPIThread0 = true;
+ pContext->apiThreadInfo.numAPIReservedThreads = 1;
+ pContext->apiThreadInfo.numAPIThreadsPerCore = 1;
+ }
memset(&pContext->WaitLock, 0, sizeof(pContext->WaitLock));
memset(&pContext->FifosNotEmpty, 0, sizeof(pContext->FifosNotEmpty));
CreateThreadPool(pContext, &pContext->threadPool);
+ if (pContext->apiThreadInfo.bindAPIThread0)
+ {
+ BindApiThread(pContext, 0);
+ }
+
pContext->ppScratch = new uint8_t*[pContext->NumWorkerThreads];
- pContext->pStats = new SWR_STATS[pContext->NumWorkerThreads];
+ pContext->pStats = (SWR_STATS*)AlignedMalloc(sizeof(SWR_STATS) * pContext->NumWorkerThreads, 64);
#if defined(KNOB_ENABLE_AR)
// Setup ArchRast thread contexts which includes +1 for API thread.
pContext->pArContext = new HANDLE[pContext->NumWorkerThreads+1];
- pContext->pArContext[pContext->NumWorkerThreads] = ArchRast::CreateThreadContext();
- _AR_EVENT(pContext->pArContext[pContext->NumWorkerThreads], ThreadStartApiEvent());
+ pContext->pArContext[pContext->NumWorkerThreads] = ArchRast::CreateThreadContext(ArchRast::AR_THREAD::API);
#endif
// Allocate scratch space for workers.
#if defined(KNOB_ENABLE_AR)
// Initialize worker thread context for ArchRast.
- pContext->pArContext[i] = ArchRast::CreateThreadContext();
- _AR_EVENT(pContext->pArContext[i], ThreadStartWorkerEvent());
+ pContext->pArContext[i] = ArchRast::CreateThreadContext(ArchRast::AR_THREAD::WORKER);
#endif
}
+#if defined(KNOB_ENABLE_AR)
+ // cache the API thread event manager, for use with sim layer
+ pCreateInfo->hArEventManager = pContext->pArContext[pContext->NumWorkerThreads];
+#endif
+
// State setup AFTER context is fully initialized
SetupDefaultState(pContext);
// initialize hot tile manager
pContext->pHotTileMgr = new HotTileMgr();
- // initialize function pointer tables
- InitClearTilesTable();
-
// initialize callback functions
pContext->pfnLoadTile = pCreateInfo->pfnLoadTile;
pContext->pfnStoreTile = pCreateInfo->pfnStoreTile;
pContext->pfnUpdateSoWriteOffset = pCreateInfo->pfnUpdateSoWriteOffset;
pContext->pfnUpdateStats = pCreateInfo->pfnUpdateStats;
pContext->pfnUpdateStatsFE = pCreateInfo->pfnUpdateStatsFE;
+
// pass pointer to bucket manager back to caller
#ifdef KNOB_ENABLE_RDTSC
void QueueWork(SWR_CONTEXT *pContext)
{
DRAW_CONTEXT* pDC = pContext->pCurDrawContext;
- uint32_t dcIndex = pDC->drawId % KNOB_MAX_DRAWS_IN_FLIGHT;
+ uint32_t dcIndex = pDC->drawId % pContext->MAX_DRAWS_IN_FLIGHT;
if (IsDraw)
{
if (IsDraw)
{
- InterlockedIncrement((volatile LONG*)&pContext->drawsOutstandingFE);
+ InterlockedIncrement(&pContext->drawsOutstandingFE);
}
_ReadWriteBarrier();
}
else
{
- AR_API_BEGIN(APIDrawWakeAllThreads, pDC->drawId);
+ RDTSC_BEGIN(APIDrawWakeAllThreads, pDC->drawId);
WakeAllThreads(pContext);
- AR_API_END(APIDrawWakeAllThreads, 1);
+ RDTSC_END(APIDrawWakeAllThreads, 1);
}
// Set current draw context to NULL so that next state call forces a new draw context to be created and populated.
DRAW_CONTEXT* GetDrawContext(SWR_CONTEXT *pContext, bool isSplitDraw = false)
{
- AR_API_BEGIN(APIGetDrawContext, 0);
+ RDTSC_BEGIN(APIGetDrawContext, 0);
// If current draw context is null then need to obtain a new draw context to use from ring.
if (pContext->pCurDrawContext == nullptr)
{
}
uint64_t curDraw = pContext->dcRing.GetHead();
- uint32_t dcIndex = curDraw % KNOB_MAX_DRAWS_IN_FLIGHT;
+ uint32_t dcIndex = curDraw % pContext->MAX_DRAWS_IN_FLIGHT;
if ((pContext->frameCount - pContext->lastFrameChecked) > 2 ||
(curDraw - pContext->lastDrawChecked) > 0x10000)
pContext->pCurDrawContext = pCurDrawContext;
// Assign next available entry in DS ring to this DC.
- uint32_t dsIndex = pContext->curStateId % KNOB_MAX_DRAWS_IN_FLIGHT;
+ uint32_t dsIndex = pContext->curStateId % pContext->MAX_DRAWS_IN_FLIGHT;
pCurDrawContext->pState = &pContext->dsRing[dsIndex];
// Copy previous state to current state.
SWR_ASSERT(pCurDrawContext->pArena->IsEmpty() == true);
+ // Reset dependency
pCurDrawContext->dependent = false;
+ pCurDrawContext->dependentFE = false;
+
pCurDrawContext->pContext = pContext;
pCurDrawContext->isCompute = false; // Dispatch has to set this to true.
SWR_ASSERT(isSplitDraw == false, "Split draw should only be used when obtaining a new DC");
}
- AR_API_END(APIGetDrawContext, 0);
+ RDTSC_END(APIGetDrawContext, 0);
return pContext->pCurDrawContext;
}
DestroyThreadPool(pContext, &pContext->threadPool);
// free the fifos
- for (uint32_t i = 0; i < KNOB_MAX_DRAWS_IN_FLIGHT; ++i)
+ for (uint32_t i = 0; i < pContext->MAX_DRAWS_IN_FLIGHT; ++i)
{
- delete[] pContext->dcRing[i].dynState.pStats;
+ AlignedFree(pContext->dcRing[i].dynState.pStats);
delete pContext->dcRing[i].pArena;
delete pContext->dsRing[i].pArena;
pContext->pMacroTileManagerArray[i].~MacroTileMgr();
}
delete[] pContext->ppScratch;
- delete[] pContext->pArContext;
- delete[] pContext->pStats;
+ AlignedFree(pContext->pStats);
delete(pContext->pHotTileMgr);
AlignedFree(GetContext(hContext));
}
+void SwrBindApiThread(HANDLE hContext, uint32_t apiThreadId)
+{
+ SWR_CONTEXT *pContext = GetContext(hContext);
+ BindApiThread(pContext, apiThreadId);
+}
+
void SWR_API SwrSaveState(
HANDLE hContext,
void* pOutputStateBlock,
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
- AR_API_BEGIN(APISync, 0);
+ RDTSC_BEGIN(APISync, 0);
pDC->FeWork.type = SYNC;
pDC->FeWork.pfnWork = ProcessSync;
pDC->retireCallback.userData2 = userData2;
pDC->retireCallback.userData3 = userData3;
+ AR_API_EVENT(SwrSyncEvent(pDC->drawId));
+
//enqueue
QueueDraw(pContext);
- AR_API_END(APISync, 1);
+ RDTSC_END(APISync, 1);
+}
+
+void SwrStallBE(HANDLE hContext)
+{
+ SWR_CONTEXT* pContext = GetContext(hContext);
+ DRAW_CONTEXT* pDC = GetDrawContext(pContext);
+
+ pDC->dependent = true;
}
void SwrWaitForIdle(HANDLE hContext)
{
SWR_CONTEXT *pContext = GetContext(hContext);
- AR_API_BEGIN(APIWaitForIdle, 0);
+ RDTSC_BEGIN(APIWaitForIdle, 0);
while (!pContext->dcRing.IsEmpty())
{
_mm_pause();
}
- AR_API_END(APIWaitForIdle, 1);
+ RDTSC_END(APIWaitForIdle, 1);
}
void SwrWaitForIdleFE(HANDLE hContext)
{
SWR_CONTEXT *pContext = GetContext(hContext);
- AR_API_BEGIN(APIWaitForIdle, 0);
+ RDTSC_BEGIN(APIWaitForIdle, 0);
while (pContext->drawsOutstandingFE > 0)
{
_mm_pause();
}
- AR_API_END(APIWaitForIdle, 1);
+ RDTSC_END(APIWaitForIdle, 1);
}
void SwrSetVertexBuffers(
HANDLE hContext,
PFN_CS_FUNC pfnCsFunc,
uint32_t totalThreadsInGroup,
- uint32_t totalSpillFillSize)
+ uint32_t totalSpillFillSize,
+ uint32_t scratchSpaceSizePerInstance,
+ uint32_t numInstances)
{
API_STATE* pState = GetDrawState(GetContext(hContext));
pState->pfnCsFunc = pfnCsFunc;
pState->totalThreadsInGroup = totalThreadsInGroup;
pState->totalSpillFillSize = totalSpillFillSize;
+ pState->scratchSpaceSize = scratchSpaceSizePerInstance;
+ pState->scratchSpaceNumInstances = numInstances;
}
void SwrSetTsState(
// update guardband multipliers for the viewport
void updateGuardbands(API_STATE *pState)
{
- uint32_t numGbs = pState->gsState.emitsRenderTargetArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
+ uint32_t numGbs = pState->backendState.readViewportArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
for(uint32_t i = 0; i < numGbs; ++i)
{
API_STATE* pState = GetDrawState(pContext);
memcpy(&pState->vp[0], pViewports, sizeof(SWR_VIEWPORT) * numViewports);
-
- if (pMatrices != nullptr)
- {
- // @todo Faster to copy portions of the SOA or just copy all of it?
- memcpy(&pState->vpMatrices, pMatrices, sizeof(SWR_VIEWPORT_MATRICES));
- }
- else
- {
- // Compute default viewport transform.
- for (uint32_t i = 0; i < numViewports; ++i)
- {
- if (pContext->driverType == DX)
- {
- pState->vpMatrices.m00[i] = pState->vp[i].width / 2.0f;
- pState->vpMatrices.m11[i] = -pState->vp[i].height / 2.0f;
- pState->vpMatrices.m22[i] = pState->vp[i].maxZ - pState->vp[i].minZ;
- pState->vpMatrices.m30[i] = pState->vp[i].x + pState->vpMatrices.m00[i];
- pState->vpMatrices.m31[i] = pState->vp[i].y - pState->vpMatrices.m11[i];
- pState->vpMatrices.m32[i] = pState->vp[i].minZ;
- }
- else
- {
- // Standard, with the exception that Y is inverted.
- pState->vpMatrices.m00[i] = (pState->vp[i].width - pState->vp[i].x) / 2.0f;
- pState->vpMatrices.m11[i] = (pState->vp[i].y - pState->vp[i].height) / 2.0f;
- pState->vpMatrices.m22[i] = (pState->vp[i].maxZ - pState->vp[i].minZ) / 2.0f;
- pState->vpMatrices.m30[i] = pState->vp[i].x + pState->vpMatrices.m00[i];
- pState->vpMatrices.m31[i] = pState->vp[i].height + pState->vpMatrices.m11[i];
- pState->vpMatrices.m32[i] = pState->vp[i].minZ + pState->vpMatrices.m22[i];
-
- // Now that the matrix is calculated, clip the view coords to screen size.
- // OpenGL allows for -ve x,y in the viewport.
- pState->vp[i].x = std::max(pState->vp[i].x, 0.0f);
- pState->vp[i].y = std::max(pState->vp[i].y, 0.0f);
- }
- }
- }
+ // @todo Faster to copy portions of the SOA or just copy all of it?
+ memcpy(&pState->vpMatrices, pMatrices, sizeof(SWR_VIEWPORT_MATRICES));
updateGuardbands(pState);
}
void SetupMacroTileScissors(DRAW_CONTEXT *pDC)
{
API_STATE *pState = &pDC->pState->state;
- uint32_t numScissors = pState->gsState.emitsViewportArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
+ uint32_t numScissors = pState->backendState.readViewportArrayIndex ? KNOB_NUM_VIEWPORTS_SCISSORS : 1;
pState->scissorsTileAligned = true;
for (uint32_t index = 0; index < numScissors; ++index)
tileAligned = (scissorInFixedPoint.xmin % KNOB_TILE_X_DIM) == 0;
tileAligned &= (scissorInFixedPoint.ymin % KNOB_TILE_Y_DIM) == 0;
tileAligned &= (scissorInFixedPoint.xmax % KNOB_TILE_X_DIM) == 0;
- tileAligned &= (scissorInFixedPoint.xmax % KNOB_TILE_Y_DIM) == 0;
+ tileAligned &= (scissorInFixedPoint.ymax % KNOB_TILE_Y_DIM) == 0;
pState->scissorsTileAligned &= tileAligned;
}
}
+
// templated backend function tables
-extern PFN_BACKEND_FUNC gBackendNullPs[SWR_MULTISAMPLE_TYPE_COUNT];
-extern PFN_BACKEND_FUNC gBackendSingleSample[SWR_INPUT_COVERAGE_COUNT][2][2];
-extern PFN_BACKEND_FUNC gBackendPixelRateTable[SWR_MULTISAMPLE_TYPE_COUNT][SWR_MSAA_SAMPLE_PATTERN_COUNT][SWR_INPUT_COVERAGE_COUNT][2][2][2];
-extern PFN_BACKEND_FUNC gBackendSampleRateTable[SWR_MULTISAMPLE_TYPE_COUNT][SWR_INPUT_COVERAGE_COUNT][2][2];
+
void SetupPipeline(DRAW_CONTEXT *pDC)
{
DRAW_STATE* pState = pDC->pState;
const SWR_RASTSTATE &rastState = pState->state.rastState;
const SWR_PS_STATE &psState = pState->state.psState;
BACKEND_FUNCS& backendFuncs = pState->backendFuncs;
- const uint32_t forcedSampleCount = (rastState.forcedSampleCount) ? 1 : 0;
// setup backend
if (psState.pfnPixelShader == nullptr)
}
else
{
- const bool bMultisampleEnable = ((rastState.sampleCount > SWR_MULTISAMPLE_1X) || rastState.forcedSampleCount) ? 1 : 0;
+ const uint32_t forcedSampleCount = (rastState.forcedSampleCount) ? 1 : 0;
+ const bool bMultisampleEnable = ((rastState.sampleCount > SWR_MULTISAMPLE_1X) || forcedSampleCount) ? 1 : 0;
const uint32_t centroid = ((psState.barycentricsMask & SWR_BARYCENTRIC_CENTROID_MASK) > 0) ? 1 : 0;
- const uint32_t canEarlyZ = (psState.forceEarlyZ || (!psState.writesODepth && !psState.usesSourceDepth && !psState.usesUAV)) ? 1 : 0;
-
+ const uint32_t canEarlyZ = (psState.forceEarlyZ || (!psState.writesODepth && !psState.usesUAV)) ? 1 : 0;
SWR_BARYCENTRICS_MASK barycentricsMask = (SWR_BARYCENTRICS_MASK)psState.barycentricsMask;
// select backend function
{
// always need to generate I & J per sample for Z interpolation
barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_SAMPLE_MASK);
- backendFuncs.pfnBackend = gBackendPixelRateTable[rastState.sampleCount][rastState.samplePattern][psState.inputCoverage][centroid][forcedSampleCount][canEarlyZ];
+ backendFuncs.pfnBackend = gBackendPixelRateTable[rastState.sampleCount][rastState.bIsCenterPattern][psState.inputCoverage]
+ [centroid][forcedSampleCount][canEarlyZ]
+ ;
}
else
{
}
break;
case SWR_SHADING_RATE_SAMPLE:
- SWR_ASSERT(rastState.samplePattern == SWR_MSAA_STANDARD_PATTERN);
+ SWR_ASSERT(rastState.bIsCenterPattern != true);
// always need to generate I & J per sample for Z interpolation
barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_SAMPLE_MASK);
backendFuncs.pfnBackend = gBackendSampleRateTable[rastState.sampleCount][psState.inputCoverage][centroid][canEarlyZ];
break;
}
}
-
+
+ SWR_ASSERT(backendFuncs.pfnBackend);
+
PFN_PROCESS_PRIMS pfnBinner;
+#if USE_SIMD16_FRONTEND
+ PFN_PROCESS_PRIMS_SIMD16 pfnBinner_simd16;
+#endif
switch (pState->state.topology)
{
case TOP_POINT_LIST:
pState->pfnProcessPrims = ClipPoints;
pfnBinner = BinPoints;
+#if USE_SIMD16_FRONTEND
+ pState->pfnProcessPrims_simd16 = ClipPoints_simd16;
+ pfnBinner_simd16 = BinPoints_simd16;
+#endif
break;
case TOP_LINE_LIST:
case TOP_LINE_STRIP:
case TOP_LISTSTRIP_ADJ:
pState->pfnProcessPrims = ClipLines;
pfnBinner = BinLines;
+#if USE_SIMD16_FRONTEND
+ pState->pfnProcessPrims_simd16 = ClipLines_simd16;
+ pfnBinner_simd16 = BinLines_simd16;
+#endif
break;
default:
pState->pfnProcessPrims = ClipTriangles;
pfnBinner = GetBinTrianglesFunc((rastState.conservativeRast > 0));
+#if USE_SIMD16_FRONTEND
+ pState->pfnProcessPrims_simd16 = ClipTriangles_simd16;
+ pfnBinner_simd16 = GetBinTrianglesFunc_simd16((rastState.conservativeRast > 0));
+#endif
break;
};
- // disable clipper if viewport transform is disabled
+
+ // Disable clipper if viewport transform is disabled
if (pState->state.frontendState.vpTransformDisable)
{
pState->pfnProcessPrims = pfnBinner;
+#if USE_SIMD16_FRONTEND
+ pState->pfnProcessPrims_simd16 = pfnBinner_simd16;
+#endif
}
+ // Disable rasterizer and backend if no pixel, no depth/stencil, and no attributes
if ((pState->state.psState.pfnPixelShader == nullptr) &&
(pState->state.depthStencilState.depthTestEnable == FALSE) &&
(pState->state.depthStencilState.depthWriteEnable == FALSE) &&
(pState->state.backendState.numAttributes == 0))
{
pState->pfnProcessPrims = nullptr;
+#if USE_SIMD16_FRONTEND
+ pState->pfnProcessPrims_simd16 = nullptr;
+#endif
}
if (pState->state.soState.rasterizerDisable == true)
{
pState->pfnProcessPrims = nullptr;
+#if USE_SIMD16_FRONTEND
+ pState->pfnProcessPrims_simd16 = nullptr;
+#endif
}
+
// set up the frontend attribute count
pState->state.feNumAttributes = 0;
const SWR_BACKEND_STATE& backendState = pState->state.backendState;
// have to check for the special case where depth/stencil test is enabled but depthwrite is disabled.
pState->state.depthHottileEnable = ((!(pState->state.depthStencilState.depthTestEnable &&
!pState->state.depthStencilState.depthWriteEnable &&
+ !pState->state.depthBoundsState.depthBoundsTestEnable &&
pState->state.depthStencilState.depthTestFunc == ZFUNC_ALWAYS)) &&
(pState->state.depthStencilState.depthTestEnable ||
- pState->state.depthStencilState.depthWriteEnable)) ? true : false;
+ pState->state.depthStencilState.depthWriteEnable ||
+ pState->state.depthBoundsState.depthBoundsTestEnable)) ? true : false;
pState->state.stencilHottileEnable = (((!(pState->state.depthStencilState.stencilTestEnable &&
!pState->state.depthStencilState.stencilWriteEnable &&
(pState->state.depthStencilState.stencilTestEnable ||
pState->state.depthStencilState.stencilWriteEnable)) ? true : false;
- uint32_t numRTs = pState->state.psState.numRenderTargets;
- pState->state.colorHottileEnable = 0;
+
+ uint32_t hotTileEnable = pState->state.psState.renderTargetMask;
+
+ // Disable hottile for surfaces with no writes
if (psState.pfnPixelShader != nullptr)
{
- for (uint32_t rt = 0; rt < numRTs; ++rt)
+ DWORD rt;
+ uint32_t rtMask = pState->state.psState.renderTargetMask;
+ while (_BitScanForward(&rt, rtMask))
{
- pState->state.colorHottileEnable |=
- (!pState->state.blendState.renderTarget[rt].writeDisableAlpha ||
- !pState->state.blendState.renderTarget[rt].writeDisableRed ||
- !pState->state.blendState.renderTarget[rt].writeDisableGreen ||
- !pState->state.blendState.renderTarget[rt].writeDisableBlue) ? (1 << rt) : 0;
+ rtMask &= ~(1 << rt);
+
+ if (pState->state.blendState.renderTarget[rt].writeDisableAlpha &&
+ pState->state.blendState.renderTarget[rt].writeDisableRed &&
+ pState->state.blendState.renderTarget[rt].writeDisableGreen &&
+ pState->state.blendState.renderTarget[rt].writeDisableBlue)
+ {
+ hotTileEnable &= ~(1 << rt);
+ }
}
}
+ pState->state.colorHottileEnable = hotTileEnable;
+
+
// Setup depth quantization function
if (pState->state.depthHottileEnable)
{
case R32_FLOAT: pState->state.pfnQuantizeDepth = QuantizeDepth < R32_FLOAT > ; break;
case R24_UNORM_X8_TYPELESS: pState->state.pfnQuantizeDepth = QuantizeDepth < R24_UNORM_X8_TYPELESS > ; break;
case R16_UNORM: pState->state.pfnQuantizeDepth = QuantizeDepth < R16_UNORM > ; break;
- default: SWR_ASSERT(false, "Unsupported depth format for depth quantiztion.");
+ default: SWR_INVALID("Unsupported depth format for depth quantiztion.");
pState->state.pfnQuantizeDepth = QuantizeDepth < R32_FLOAT > ;
}
}
SetupMacroTileScissors(pDC);
SetupPipeline(pDC);
}
+
+
}
//////////////////////////////////////////////////////////////////////////
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
- AR_API_BEGIN(APIDraw, pDC->drawId);
- AR_API_EVENT(DrawInstancedEvent(pDC->drawId, topology, numVertices, startVertex, numInstances, startInstance));
+ RDTSC_BEGIN(APIDraw, pDC->drawId);
+ AR_API_EVENT(DrawInstancedEvent(pDC->drawId, ArchRast::Instanced, topology, numVertices, startVertex, numInstances, startInstance));
uint32_t maxVertsPerDraw = MaxVertsPerDraw(pDC, numVertices, topology);
uint32_t primsPerDraw = GetNumPrims(topology, maxVertsPerDraw);
//enqueue DC
QueueDraw(pContext);
+ AR_API_EVENT(DrawInstancedSplitEvent(pDC->drawId, ArchRast::InstancedSplit));
+
remainingVerts -= numVertsForDraw;
draw++;
}
pDC = GetDrawContext(pContext);
pDC->pState->state.rastState.cullMode = oldCullMode;
- AR_API_END(APIDraw, numVertices * numInstances);
+ RDTSC_END(APIDraw, numVertices * numInstances);
}
//////////////////////////////////////////////////////////////////////////
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
API_STATE* pState = &pDC->pState->state;
- AR_API_BEGIN(APIDrawIndexed, pDC->drawId);
- AR_API_EVENT(DrawIndexedInstancedEvent(pDC->drawId, topology, numIndices, indexOffset, baseVertex, numInstances, startInstance));
+ RDTSC_BEGIN(APIDrawIndexed, pDC->drawId);
+ AR_API_EVENT(DrawIndexedInstancedEvent(pDC->drawId, ArchRast::IndexedInstancedSplit, topology, numIndices, indexOffset, baseVertex, numInstances, startInstance));
uint32_t maxIndicesPerDraw = MaxVertsPerDraw(pDC, numIndices, topology);
uint32_t primsPerDraw = GetNumPrims(topology, maxIndicesPerDraw);
case R16_UINT: indexSize = sizeof(uint16_t); break;
case R8_UINT: indexSize = sizeof(uint8_t); break;
default:
- SWR_ASSERT(0);
+ SWR_INVALID("Invalid index buffer format: %d", pState->indexBuffer.format);
}
int draw = 0;
// When breaking up draw, we need to obtain new draw context for each iteration.
bool isSplitDraw = (draw > 0) ? true : false;
+
pDC = GetDrawContext(pContext, isSplitDraw);
InitDraw(pDC, isSplitDraw);
//enqueue DC
QueueDraw(pContext);
+ AR_API_EVENT(DrawIndexedInstancedSplitEvent(pDC->drawId, ArchRast::IndexedInstancedSplit));
+
pIB += maxIndicesPerDraw * indexSize;
remainingIndices -= numIndicesForDraw;
draw++;
}
- // restore culling state
+ // Restore culling state
pDC = GetDrawContext(pContext);
pDC->pState->state.rastState.cullMode = oldCullMode;
-
- AR_API_END(APIDrawIndexed, numIndices * numInstances);
+
+ RDTSC_END(APIDrawIndexed, numIndices * numInstances);
}
//enqueue
QueueDraw(pContext);
+
+ AR_API_EVENT(SwrInvalidateTilesEvent(pDC->drawId));
}
//////////////////////////////////////////////////////////////////////////
//enqueue
QueueDraw(pContext);
+
+ AR_API_EVENT(SwrDiscardRectEvent(pDC->drawId));
}
//////////////////////////////////////////////////////////////////////////
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
- AR_API_BEGIN(APIDispatch, pDC->drawId);
+ RDTSC_BEGIN(APIDispatch, pDC->drawId);
AR_API_EVENT(DispatchEvent(pDC->drawId, threadGroupCountX, threadGroupCountY, threadGroupCountZ));
pDC->isCompute = true; // This is a compute context.
pTaskData->threadGroupCountZ = threadGroupCountZ;
uint32_t totalThreadGroups = threadGroupCountX * threadGroupCountY * threadGroupCountZ;
- uint32_t dcIndex = pDC->drawId % KNOB_MAX_DRAWS_IN_FLIGHT;
+ uint32_t dcIndex = pDC->drawId % pContext->MAX_DRAWS_IN_FLIGHT;
pDC->pDispatch = &pContext->pDispatchQueueArray[dcIndex];
pDC->pDispatch->initialize(totalThreadGroups, pTaskData, &ProcessComputeBE);
QueueDispatch(pContext);
- AR_API_END(APIDispatch, threadGroupCountX * threadGroupCountY * threadGroupCountZ);
+ RDTSC_END(APIDispatch, threadGroupCountX * threadGroupCountY * threadGroupCountZ);
}
// Deswizzles, converts and stores current contents of the hot tiles to surface
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
- AR_API_BEGIN(APIStoreTiles, pDC->drawId);
+ RDTSC_BEGIN(APIStoreTiles, pDC->drawId);
pDC->FeWork.type = STORETILES;
pDC->FeWork.pfnWork = ProcessStoreTiles;
//enqueue
QueueDraw(pContext);
- AR_API_END(APIStoreTiles, 1);
+ AR_API_EVENT(SwrStoreTilesEvent(pDC->drawId));
+
+ RDTSC_END(APIStoreTiles, 1);
}
//////////////////////////////////////////////////////////////////////////
/// @brief SwrClearRenderTarget - Clear attached render targets / depth / stencil
/// @param hContext - Handle passed back from SwrCreateContext
-/// @param clearMask - combination of SWR_CLEAR_COLOR / SWR_CLEAR_DEPTH / SWR_CLEAR_STENCIL flags (or SWR_CLEAR_NONE)
+/// @param attachmentMask - combination of SWR_ATTACHMENT_*_BIT attachments to clear
+/// @param renderTargetArrayIndex - the RT array index to clear
/// @param clearColor - color use for clearing render targets
/// @param z - depth value use for clearing depth buffer
/// @param stencil - stencil value used for clearing stencil buffer
/// @param clearRect - The pixel-coordinate rectangle to clear in all cleared buffers
void SWR_API SwrClearRenderTarget(
HANDLE hContext,
- uint32_t clearMask,
+ uint32_t attachmentMask,
+ uint32_t renderTargetArrayIndex,
const float clearColor[4],
float z,
uint8_t stencil,
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
- AR_API_BEGIN(APIClearRenderTarget, pDC->drawId);
-
- CLEAR_FLAGS flags;
- flags.bits = 0;
- flags.mask = clearMask;
+ RDTSC_BEGIN(APIClearRenderTarget, pDC->drawId);
pDC->FeWork.type = CLEAR;
pDC->FeWork.pfnWork = ProcessClear;
pDC->FeWork.desc.clear.rect = clearRect;
pDC->FeWork.desc.clear.rect &= g_MaxScissorRect;
- pDC->FeWork.desc.clear.flags = flags;
+ pDC->FeWork.desc.clear.attachmentMask = attachmentMask;
+ pDC->FeWork.desc.clear.renderTargetArrayIndex = renderTargetArrayIndex;
pDC->FeWork.desc.clear.clearDepth = z;
pDC->FeWork.desc.clear.clearRTColor[0] = clearColor[0];
pDC->FeWork.desc.clear.clearRTColor[1] = clearColor[1];
// enqueue draw
QueueDraw(pContext);
- AR_API_END(APIClearRenderTarget, 1);
+ RDTSC_END(APIClearRenderTarget, 1);
}
//////////////////////////////////////////////////////////////////////////
/// @brief Enables stats counting
/// @param hContext - Handle passed back from SwrCreateContext
/// @param enable - If true then counts are incremented.
-void SwrEnableStats(
+void SwrEnableStatsFE(
+ HANDLE hContext,
+ bool enable)
+{
+ SWR_CONTEXT *pContext = GetContext(hContext);
+ DRAW_CONTEXT* pDC = GetDrawContext(pContext);
+
+ pDC->pState->state.enableStatsFE = enable;
+}
+
+//////////////////////////////////////////////////////////////////////////
+/// @brief Enables stats counting
+/// @param hContext - Handle passed back from SwrCreateContext
+/// @param enable - If true then counts are incremented.
+void SwrEnableStatsBE(
HANDLE hContext,
bool enable)
{
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
- pDC->pState->state.enableStats = enable;
+ pDC->pState->state.enableStatsBE = enable;
}
//////////////////////////////////////////////////////////////////////////
{
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
+ (void)pDC; // var used
RDTSC_ENDFRAME();
AR_API_EVENT(FrameEndEvent(pContext->frameCount, pDC->drawId));
pContext->frameCount++;
}
+
+void InitSimLoadTilesTable();
+void InitSimStoreTilesTable();
+void InitSimClearTilesTable();
+
+void InitClearTilesTable();
+void InitBackendFuncTables();
+
+//////////////////////////////////////////////////////////////////////////
+/// @brief Initialize swr backend and memory internal tables
+void SwrInit()
+{
+ InitSimLoadTilesTable();
+ InitSimStoreTilesTable();
+ InitSimClearTilesTable();
+
+ InitClearTilesTable();
+ InitBackendFuncTables();
+ InitRasterizerFunctions();
+}
+
+void SwrGetInterface(SWR_INTERFACE &out_funcs)
+{
+ out_funcs.pfnSwrCreateContext = SwrCreateContext;
+ out_funcs.pfnSwrDestroyContext = SwrDestroyContext;
+ out_funcs.pfnSwrBindApiThread = SwrBindApiThread;
+ out_funcs.pfnSwrSaveState = SwrSaveState;
+ out_funcs.pfnSwrRestoreState = SwrRestoreState;
+ out_funcs.pfnSwrSync = SwrSync;
+ out_funcs.pfnSwrStallBE = SwrStallBE;
+ out_funcs.pfnSwrWaitForIdle = SwrWaitForIdle;
+ out_funcs.pfnSwrWaitForIdleFE = SwrWaitForIdleFE;
+ out_funcs.pfnSwrSetVertexBuffers = SwrSetVertexBuffers;
+ out_funcs.pfnSwrSetIndexBuffer = SwrSetIndexBuffer;
+ out_funcs.pfnSwrSetFetchFunc = SwrSetFetchFunc;
+ out_funcs.pfnSwrSetSoFunc = SwrSetSoFunc;
+ out_funcs.pfnSwrSetSoState = SwrSetSoState;
+ out_funcs.pfnSwrSetSoBuffers = SwrSetSoBuffers;
+ out_funcs.pfnSwrSetVertexFunc = SwrSetVertexFunc;
+ out_funcs.pfnSwrSetFrontendState = SwrSetFrontendState;
+ out_funcs.pfnSwrSetGsState = SwrSetGsState;
+ out_funcs.pfnSwrSetGsFunc = SwrSetGsFunc;
+ out_funcs.pfnSwrSetCsFunc = SwrSetCsFunc;
+ out_funcs.pfnSwrSetTsState = SwrSetTsState;
+ out_funcs.pfnSwrSetHsFunc = SwrSetHsFunc;
+ out_funcs.pfnSwrSetDsFunc = SwrSetDsFunc;
+ out_funcs.pfnSwrSetDepthStencilState = SwrSetDepthStencilState;
+ out_funcs.pfnSwrSetBackendState = SwrSetBackendState;
+ out_funcs.pfnSwrSetDepthBoundsState = SwrSetDepthBoundsState;
+ out_funcs.pfnSwrSetPixelShaderState = SwrSetPixelShaderState;
+ out_funcs.pfnSwrSetBlendState = SwrSetBlendState;
+ out_funcs.pfnSwrSetBlendFunc = SwrSetBlendFunc;
+ out_funcs.pfnSwrDraw = SwrDraw;
+ out_funcs.pfnSwrDrawInstanced = SwrDrawInstanced;
+ out_funcs.pfnSwrDrawIndexed = SwrDrawIndexed;
+ out_funcs.pfnSwrDrawIndexedInstanced = SwrDrawIndexedInstanced;
+ out_funcs.pfnSwrInvalidateTiles = SwrInvalidateTiles;
+ out_funcs.pfnSwrDiscardRect = SwrDiscardRect;
+ out_funcs.pfnSwrDispatch = SwrDispatch;
+ out_funcs.pfnSwrStoreTiles = SwrStoreTiles;
+ out_funcs.pfnSwrClearRenderTarget = SwrClearRenderTarget;
+ out_funcs.pfnSwrSetRastState = SwrSetRastState;
+ out_funcs.pfnSwrSetViewports = SwrSetViewports;
+ out_funcs.pfnSwrSetScissorRects = SwrSetScissorRects;
+ out_funcs.pfnSwrGetPrivateContextState = SwrGetPrivateContextState;
+ out_funcs.pfnSwrAllocDrawContextMemory = SwrAllocDrawContextMemory;
+ out_funcs.pfnSwrEnableStatsFE = SwrEnableStatsFE;
+ out_funcs.pfnSwrEnableStatsBE = SwrEnableStatsBE;
+ out_funcs.pfnSwrEndFrame = SwrEndFrame;
+ out_funcs.pfnSwrInit = SwrInit;
+ out_funcs.pfnSwrLoadHotTile = SwrLoadHotTile;
+ out_funcs.pfnSwrStoreHotTileToSurface = SwrStoreHotTileToSurface;
+ out_funcs.pfnSwrStoreHotTileClear = SwrStoreHotTileClear;
+}