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 = (SWR_STATS*)AlignedMalloc(sizeof(SWR_STATS) * pContext->NumWorkerThreads, 64);
#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);
}
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)
{
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;
}
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;
//enqueue
QueueDraw(pContext);
- AR_API_END(APISync, 1);
+ RDTSC_END(APISync, 1);
}
void SwrStallBE(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(
};
- // disable clipper if viewport transform is disabled
+ // Disable clipper if viewport transform is disabled
if (pState->state.frontendState.vpTransformDisable)
{
pState->pfnProcessPrims = pfnBinner;
#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) &&
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));
+ 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);
//enqueue DC
QueueDraw(pContext);
- AR_API_EVENT(DrawIndexedInstancedSplitEvent(pDC->drawId));
+ AR_API_EVENT(DrawIndexedInstancedSplitEvent(pDC->drawId, ArchRast::IndexedInstancedSplit));
pIB += maxIndicesPerDraw * indexSize;
remainingIndices -= numIndicesForDraw;
pDC = GetDrawContext(pContext);
pDC->pState->state.rastState.cullMode = oldCullMode;
- AR_API_END(APIDrawIndexed, numIndices * numInstances);
+ RDTSC_END(APIDrawIndexed, numIndices * numInstances);
}
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.
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;
AR_API_EVENT(SwrStoreTilesEvent(pDC->drawId));
- AR_API_END(APIStoreTiles, 1);
+ RDTSC_END(APIStoreTiles, 1);
}
//////////////////////////////////////////////////////////////////////////
SWR_CONTEXT *pContext = GetContext(hContext);
DRAW_CONTEXT* pDC = GetDrawContext(pContext);
- AR_API_BEGIN(APIClearRenderTarget, pDC->drawId);
+ RDTSC_BEGIN(APIClearRenderTarget, pDC->drawId);
pDC->FeWork.type = CLEAR;
pDC->FeWork.pfnWork = ProcessClear;
// enqueue draw
QueueDraw(pContext);
- AR_API_END(APIClearRenderTarget, 1);
+ RDTSC_END(APIClearRenderTarget, 1);
}
//////////////////////////////////////////////////////////////////////////
{
out_funcs.pfnSwrCreateContext = SwrCreateContext;
out_funcs.pfnSwrDestroyContext = SwrDestroyContext;
+ out_funcs.pfnSwrBindApiThread = SwrBindApiThread;
out_funcs.pfnSwrSaveState = SwrSaveState;
out_funcs.pfnSwrRestoreState = SwrRestoreState;
out_funcs.pfnSwrSync = SwrSync;