summaryrefslogtreecommitdiff
path: root/libs/ode-0.16.1/ode/src/fastldltfactor.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'libs/ode-0.16.1/ode/src/fastldltfactor.cpp')
-rw-r--r--libs/ode-0.16.1/ode/src/fastldltfactor.cpp462
1 files changed, 462 insertions, 0 deletions
diff --git a/libs/ode-0.16.1/ode/src/fastldltfactor.cpp b/libs/ode-0.16.1/ode/src/fastldltfactor.cpp
new file mode 100644
index 0000000..9c1b921
--- /dev/null
+++ b/libs/ode-0.16.1/ode/src/fastldltfactor.cpp
@@ -0,0 +1,462 @@
+/*************************************************************************
+ * *
+ * Open Dynamics Engine, Copyright (C) 2001,2002 Russell L. Smith. *
+ * All rights reserved. Email: russ@q12.org Web: www.q12.org *
+ * *
+ * This library is free software; you can redistribute it and/or *
+ * modify it under the terms of EITHER: *
+ * (1) The GNU Lesser General Public License as published by the Free *
+ * Software Foundation; either version 2.1 of the License, or (at *
+ * your option) any later version. The text of the GNU Lesser *
+ * General Public License is included with this library in the *
+ * file LICENSE.TXT. *
+ * (2) The BSD-style license that is included with this library in *
+ * the file LICENSE-BSD.TXT. *
+ * *
+ * This library is distributed in the hope that it will be useful, *
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of *
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the files *
+ * LICENSE.TXT and LICENSE-BSD.TXT for more details. *
+ * *
+ *************************************************************************/
+
+/*
+ * LDLT factorization related code of ThreadedEquationSolverLDLT
+ * Copyright (c) 2017-2019 Oleh Derevenko, odar@eleks.com (change all "a" to "e")
+ */
+
+
+#include <ode/common.h>
+#include <ode/matrix.h>
+#include <ode/matrix_coop.h>
+#include "config.h"
+#include "threaded_solver_ldlt.h"
+#include "threading_base.h"
+#include "resource_control.h"
+#include "error.h"
+
+#include "fastldltfactor_impl.h"
+
+
+/*static */
+void ThreadedEquationSolverLDLT::estimateCooperativeFactoringLDLTResourceRequirements(
+ dxResourceRequirementDescriptor *summaryRequirementsDescriptor,
+ unsigned allowedThreadCount, unsigned rowCount)
+{
+ dxThreadingBase *threading = summaryRequirementsDescriptor->getrelatedThreading();
+ unsigned limitedThreadCount = restrictFactoringLDLTAllowedThreadCount(threading, allowedThreadCount, rowCount);
+
+ if (limitedThreadCount > 1)
+ {
+ doEstimateCooperativeFactoringLDLTResourceRequirementsValidated(summaryRequirementsDescriptor, allowedThreadCount, rowCount);
+ }
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::cooperativelyFactorLDLT(
+ dxRequiredResourceContainer *resourceContainer, unsigned allowedThreadCount,
+ dReal *A, dReal *d, unsigned rowCount, unsigned rowSkip)
+{
+ dAASSERT(rowCount != 0);
+
+ dxThreadingBase *threading = resourceContainer->getThreadingInstance();
+ unsigned limitedThreadCount = restrictFactoringLDLTAllowedThreadCount(threading, allowedThreadCount, rowCount);
+
+ if (limitedThreadCount <= 1)
+ {
+ factorMatrixAsLDLT<FLDLT_D_STRIDE>(A, d, rowCount, rowSkip);
+ }
+ else
+ {
+ doCooperativelyFactorLDLTValidated(resourceContainer, limitedThreadCount, A, d, rowCount, rowSkip);
+ }
+}
+
+
+/*static */
+unsigned ThreadedEquationSolverLDLT::restrictFactoringLDLTAllowedThreadCount(
+ dxThreadingBase *threading, unsigned allowedThreadCount, unsigned rowCount)
+{
+ unsigned limitedThreadCount = 1;
+
+#if dCOOPERATIVE_ENABLED
+ const unsigned int solvingBlockStep = FSL1S_BLOCK_SIZE; // Required by the implementation
+ unsigned solvingMaximalBlockCount = deriveSolvingL1StripeBlockCount(rowCount, solvingBlockStep);
+ dIASSERT(deriveSolvingL1StripeThreadCount(FLDLT_COOPERATIVE_BLOCK_COUNT_MINIMUM - 1, 2) > 1);
+
+ if (solvingMaximalBlockCount >= FLDLT_COOPERATIVE_BLOCK_COUNT_MINIMUM)
+ {
+ limitedThreadCount = threading->calculateThreadingLimitedThreadCount(allowedThreadCount, false);
+ }
+#endif // #if dCOOPERATIVE_ENABLED
+
+ return limitedThreadCount;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::doEstimateCooperativeFactoringLDLTResourceRequirementsValidated(
+ dxResourceRequirementDescriptor *summaryRequirementsDescriptor,
+ unsigned allowedThreadCount, unsigned rowCount)
+{
+ const unsigned int solvingBlockStep = FSL1S_BLOCK_SIZE; // Required by the implementation
+ unsigned solvingTotalBlockCount = deriveSolvingL1StripeBlockCount(rowCount, solvingBlockStep);
+ dIASSERT(solvingTotalBlockCount >= 1);
+
+ unsigned solvingLastBlockIndex = solvingTotalBlockCount - 1;
+
+ const unsigned factorizingBlockARows = FFL1S_REGULAR_A_ROWS;
+ unsigned factorizingMaximalBlockCount = deriveScalingAndFactorizingL1StripeBlockCountFromSolvingBlockIndex(solvingLastBlockIndex, solvingBlockStep, factorizingBlockARows);
+
+ unsigned blockSolvingMaximumThreads = deriveSolvingL1StripeThreadCount(solvingLastBlockIndex, allowedThreadCount);
+ unsigned blockFactorizingMaximumThreads = deriveScalingAndFactorizingL1StripeThreadCount(factorizingMaximalBlockCount, allowedThreadCount);
+ unsigned simultaneousCallCount = 1 // Final synchronization point
+ + 2 // intermediate synchronization points
+ + dMACRO_MAX(blockSolvingMaximumThreads, blockFactorizingMaximumThreads);
+
+ FactorizationSolvingL1StripeMemoryEstimates solvingMemoryEstimates;
+ FactorizationScalingAndFactorizingL1StripeMemoryEstimates scalingAndFactorizingEstimates;
+ sizeint solvingMemoryRequired = estimateCooperativelySolvingL1Stripe_XMemoryRequirement(solvingTotalBlockCount, solvingMemoryEstimates);
+ sizeint factorizingMemoryRequired = estimateCooperativelyScalingAndFactorizingL1Stripe_XMemoryRequirement(blockFactorizingMaximumThreads, scalingAndFactorizingEstimates);
+ sizeint totalSizeRequired = solvingMemoryRequired + factorizingMemoryRequired;
+ const unsigned memoryAlignmentRequired = ALLOCATION_DEFAULT_ALIGNMENT;
+
+ unsigned featureRequirement = dxResourceRequirementDescriptor::STOCK_CALLWAIT_REQUIRED;
+ summaryRequirementsDescriptor->mergeAnotherDescriptorIn(totalSizeRequired, memoryAlignmentRequired, simultaneousCallCount, featureRequirement);
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::doCooperativelyFactorLDLTValidated(
+ dxRequiredResourceContainer *resourceContainer, unsigned allowedThreadCount,
+ dReal *A, dReal *d, unsigned rowCount, unsigned rowSkip)
+{
+ dIASSERT(allowedThreadCount > 1);
+
+ const unsigned int solvingBlockStep = FSL1S_BLOCK_SIZE; // Required by the implementation
+ unsigned solvingTotalBlockCount = deriveSolvingL1StripeBlockCount(rowCount, solvingBlockStep);
+ dIASSERT(solvingTotalBlockCount >= 1);
+
+ unsigned solvingLastBlockIndex = solvingTotalBlockCount - 1;
+
+ const unsigned factorizingBlockARows = FFL1S_REGULAR_A_ROWS;
+ unsigned factorizingMaximalBlockCount = deriveScalingAndFactorizingL1StripeBlockCountFromSolvingBlockIndex(solvingLastBlockIndex, solvingBlockStep, factorizingBlockARows);
+
+ unsigned blockFactorizingMaximumThreads = deriveScalingAndFactorizingL1StripeThreadCount(factorizingMaximalBlockCount, allowedThreadCount);
+
+ dCallWaitID completionWait = resourceContainer->getStockCallWait();
+ dAASSERT(completionWait != NULL);
+
+ FactorizationSolvingL1StripeMemoryEstimates solvingMemoryEstimates;
+ FactorizationScalingAndFactorizingL1StripeMemoryEstimates scalingAndFactorizingEstimates;
+ sizeint solvingMemoryRequired = estimateCooperativelySolvingL1Stripe_XMemoryRequirement(solvingTotalBlockCount, solvingMemoryEstimates);
+ sizeint factorizingMemoryRequired = estimateCooperativelyScalingAndFactorizingL1Stripe_XMemoryRequirement(blockFactorizingMaximumThreads, scalingAndFactorizingEstimates);
+ sizeint totalSizeRequired = solvingMemoryRequired + factorizingMemoryRequired;
+ dIASSERT(totalSizeRequired <= resourceContainer->getMemoryBufferSize());
+
+ void *bufferAllocated = resourceContainer->getMemoryBufferPointer();
+ dIASSERT(bufferAllocated != NULL);
+ dIASSERT(dALIGN_PTR(bufferAllocated, ALLOCATION_DEFAULT_ALIGNMENT) == bufferAllocated);
+
+ atomicord32 solvingBlockCompletionProgress;
+ cellindexint *solvingBlockProgressDescriptors;
+ FactorizationSolveL1StripeCellContext *solvingCellContexts;
+
+ FactorizationFactorizeL1StripeContext *factorizingFactorizationContext;
+
+ void *bufferCurrentLocation = bufferAllocated;
+ bufferCurrentLocation = markCooperativelySolvingL1Stripe_XMemoryStructuresOut(bufferCurrentLocation, solvingMemoryEstimates, solvingBlockProgressDescriptors, solvingCellContexts);
+ bufferCurrentLocation = markCooperativelyScalingAndFactorizingL1Stripe_XMemoryStructuresOut(bufferCurrentLocation, scalingAndFactorizingEstimates, factorizingFactorizationContext);
+ dIVERIFY(bufferCurrentLocation <= (uint8 *)bufferAllocated + totalSizeRequired);
+
+ dCallReleaseeID calculationFinishReleasee;
+ dxThreadingBase *threading = resourceContainer->getThreadingInstance();
+ threading->PostThreadedCall(NULL, &calculationFinishReleasee, 1, NULL, completionWait, &factotLDLT_completion_callback, NULL, 0, "FactorLDLT Completion");
+
+ FactorLDLTWorkerContext workerContext(threading, allowedThreadCount, A, d, solvingTotalBlockCount, rowCount, rowSkip,
+ solvingBlockCompletionProgress, solvingBlockProgressDescriptors, solvingCellContexts,
+ factorizingFactorizationContext,
+ calculationFinishReleasee); // The variable must exist in the outer scope
+
+ dIASSERT(solvingTotalBlockCount >= FLDLT_COOPERATIVE_BLOCK_COUNT_MINIMUM);
+ dSASSERT(FLDLT_COOPERATIVE_BLOCK_COUNT_MINIMUM > 2);
+
+ scaleAndFactorizeL1FirstRowStripe_2<FLDLT_D_STRIDE>(workerContext.m_ARow, workerContext.m_d, workerContext.m_rowSkip);
+ workerContext.incrementForNextBlock();
+
+ const unsigned blockIndex = 1;
+ dIASSERT(blockIndex == workerContext.m_solvingBlockIndex);
+
+ initializeCooperativelySolvingL1Stripe_XMemoryStructures(blockIndex, solvingBlockCompletionProgress, solvingBlockProgressDescriptors, solvingCellContexts);
+ unsigned secondBlockSolvingThreadCount = deriveSolvingL1StripeThreadCount(blockIndex, allowedThreadCount);
+
+ dCallReleaseeID secondBlockSolvingSyncReleasee;
+ threading->PostThreadedCall(NULL, &secondBlockSolvingSyncReleasee, secondBlockSolvingThreadCount, NULL, NULL, &factotLDLT_solvingCompleteSync_callback, &workerContext, 0, "FactorLDLT Solving Complete Sync");
+
+ if (secondBlockSolvingThreadCount > 1)
+ {
+ threading->PostThreadedCallsGroup(NULL, secondBlockSolvingThreadCount - 1, secondBlockSolvingSyncReleasee, &factotLDLT_solvingComplete_callback, &workerContext, "FactorLDLT Solving Complete");
+ }
+
+ factotLDLT_solvingComplete(workerContext, secondBlockSolvingThreadCount - 1);
+ threading->AlterThreadedCallDependenciesCount(secondBlockSolvingSyncReleasee, -1);
+
+ threading->WaitThreadedCallExclusively(NULL, completionWait, NULL, "FactorLDLT End Wait");
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_solvingComplete_callback(void *callContext, dcallindex_t callInstanceIndex, dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ FactorLDLTWorkerContext *ptrContext = (FactorLDLTWorkerContext *)callContext;
+
+ factotLDLT_solvingComplete(*ptrContext, dCAST_TO_SMALLER(unsigned, callInstanceIndex));
+
+ return 1;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::factotLDLT_solvingComplete(FactorLDLTWorkerContext &ref_context, unsigned ownThreadIndex)
+{
+ participateSolvingL1Stripe_X<FSL1S_BLOCK_SIZE, FSL1S_REGULAR_B_ROWS>(ref_context.m_A, ref_context.m_ARow, ref_context.m_solvingBlockIndex, ref_context.m_rowSkip,
+ ref_context.m_refSolvingBlockCompletionProgress, ref_context.m_solvingBlockProgressDescriptors, ref_context.m_solvingCellContexts, ownThreadIndex);
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_solvingCompleteSync_callback(void *callContext, dcallindex_t dUNUSED(callInstanceIndex), dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ FactorLDLTWorkerContext *ptrContext = (FactorLDLTWorkerContext *)callContext;
+
+ factotLDLT_solvingCompleteSync(*ptrContext);
+
+ return 1;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::factotLDLT_solvingCompleteSync(FactorLDLTWorkerContext &ref_workerContext)
+{
+ unsigned solvingBlockIndex = ref_workerContext.m_solvingBlockIndex;
+ FactorizationFactorizeL1StripeContext *factorizingFactorizationContext = ref_workerContext.m_factorizingFactorizationContext;
+
+ const unsigned int solvingBlockStep = FSL1S_BLOCK_SIZE;
+ const unsigned factorizingBlockARows = FFL1S_REGULAR_A_ROWS;
+ unsigned factorizingBlockCount = deriveScalingAndFactorizingL1StripeBlockCountFromSolvingBlockIndex(solvingBlockIndex, solvingBlockStep, factorizingBlockARows);
+ unsigned blockFactorizingThreadCount = deriveScalingAndFactorizingL1StripeThreadCount(factorizingBlockCount, ref_workerContext.m_allowedThreadCount);
+ initializeCooperativelyScalingAndFactorizingL1Stripe_XMemoryStructures(factorizingFactorizationContext, blockFactorizingThreadCount);
+
+ dCallReleaseeID blockFactorizingSyncReleasee;
+
+ dxThreadingBase *threading = ref_workerContext.m_threading;
+ if (solvingBlockIndex != ref_workerContext.m_totalBlockCount - 1)
+ {
+ threading->PostThreadedCall(NULL, &blockFactorizingSyncReleasee, blockFactorizingThreadCount, NULL, NULL, &factotLDLT_scalingAndFactorizingCompleteSync_callback, &ref_workerContext, 0, "FactorLDLT S'n'F Sync");
+ }
+ else
+ {
+ blockFactorizingSyncReleasee = ref_workerContext.m_calculationFinishReleasee;
+
+ if (blockFactorizingThreadCount > 1)
+ {
+ threading->AlterThreadedCallDependenciesCount(blockFactorizingSyncReleasee, blockFactorizingThreadCount - 1);
+ }
+ }
+
+ if (blockFactorizingThreadCount > 1)
+ {
+ threading->PostThreadedCallsGroup(NULL, blockFactorizingThreadCount - 1, blockFactorizingSyncReleasee, &factotLDLT_scalingAndFactorizingComplete_callback, &ref_workerContext, "FactorLDLT S'n'F Complete");
+ }
+
+ factotLDLT_scalingAndFactorizingComplete(ref_workerContext, blockFactorizingThreadCount - 1);
+ threading->AlterThreadedCallDependenciesCount(blockFactorizingSyncReleasee, -1);
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_scalingAndFactorizingComplete_callback(void *callContext, dcallindex_t callInstanceIndex, dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ FactorLDLTWorkerContext *ptrContext = (FactorLDLTWorkerContext *)callContext;
+
+ factotLDLT_scalingAndFactorizingComplete(*ptrContext, dCAST_TO_SMALLER(unsigned, callInstanceIndex));
+
+ return 1;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::factotLDLT_scalingAndFactorizingComplete(FactorLDLTWorkerContext &ref_workerContext, unsigned ownThreadIndex)
+{
+ unsigned factorizationRow = ref_workerContext.m_solvingBlockIndex * FSL1S_BLOCK_SIZE;
+ participateScalingAndFactorizingL1Stripe_X<FFL1S_REGULAR_A_ROWS, FLDLT_D_STRIDE>(ref_workerContext.m_ARow, ref_workerContext.m_d, factorizationRow,
+ ref_workerContext.m_rowSkip, ref_workerContext.m_factorizingFactorizationContext, ownThreadIndex);
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_scalingAndFactorizingCompleteSync_callback(void *callContext, dcallindex_t dUNUSED(callInstanceIndex), dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ FactorLDLTWorkerContext *ptrContext = (FactorLDLTWorkerContext *)callContext;
+
+ factotLDLT_scalingAndFactorizingCompleteSync(*ptrContext);
+
+ return 1;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::factotLDLT_scalingAndFactorizingCompleteSync(FactorLDLTWorkerContext &ref_workerContext)
+{
+ ref_workerContext.incrementForNextBlock();
+
+ unsigned blockIndex = ref_workerContext.m_solvingBlockIndex;
+ dIASSERT(blockIndex < ref_workerContext.m_totalBlockCount);
+
+ atomicord32 &refSolvingBlockCompletionProgress = ref_workerContext.m_refSolvingBlockCompletionProgress;
+ cellindexint *solvingBlockProgressDescriptors = ref_workerContext.m_solvingBlockProgressDescriptors;
+ FactorizationSolveL1StripeCellContext *solvingCellContexts = ref_workerContext.m_solvingCellContexts;
+
+ initializeCooperativelySolvingL1Stripe_XMemoryStructures(blockIndex, refSolvingBlockCompletionProgress, solvingBlockProgressDescriptors, solvingCellContexts);
+ unsigned blockSolvingThreadCount = deriveSolvingL1StripeThreadCount(blockIndex, ref_workerContext.m_allowedThreadCount);
+
+ dCallReleaseeID blockSolvingSyncReleasee;
+
+ dxThreadingBase *threading = ref_workerContext.m_threading;
+ if (blockIndex != ref_workerContext.m_totalBlockCount - 1 || ref_workerContext.m_rowCount % FSL1S_REGULAR_B_ROWS == 0)
+ {
+ threading->PostThreadedCall(NULL, &blockSolvingSyncReleasee, blockSolvingThreadCount, NULL, NULL, &factotLDLT_solvingCompleteSync_callback, &ref_workerContext, 0, "FactorLDLT Solving Complete Sync");
+
+ if (blockSolvingThreadCount > 1)
+ {
+ threading->PostThreadedCallsGroup(NULL, blockSolvingThreadCount - 1, blockSolvingSyncReleasee, &factotLDLT_solvingComplete_callback, &ref_workerContext, "FactorLDLT Solving Complete");
+ }
+
+ factotLDLT_solvingComplete(ref_workerContext, blockSolvingThreadCount - 1);
+ }
+ else
+ {
+ dSASSERT(FSL1S_REGULAR_B_ROWS == 2);
+ dSASSERT(FSL1S_FINAL_B_ROWS == 1);
+
+ threading->PostThreadedCall(NULL, &blockSolvingSyncReleasee, blockSolvingThreadCount, NULL, NULL, &factotLDLT_solvingFinalSync_callback, &ref_workerContext, 0, "FactorLDLT Solving Final Sync");
+
+ if (blockSolvingThreadCount > 1)
+ {
+ threading->PostThreadedCallsGroup(NULL, blockSolvingThreadCount - 1, blockSolvingSyncReleasee, &factotLDLT_solvingFinal_callback, &ref_workerContext, "FactorLDLT Solving Final");
+ }
+
+ factotLDLT_solvingFinal(ref_workerContext, blockSolvingThreadCount - 1);
+ }
+
+ threading->AlterThreadedCallDependenciesCount(blockSolvingSyncReleasee, -1);
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_solvingFinal_callback(void *callContext, dcallindex_t callInstanceIndex, dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ FactorLDLTWorkerContext *ptrContext = (FactorLDLTWorkerContext *)callContext;
+
+ factotLDLT_solvingFinal(*ptrContext, dCAST_TO_SMALLER(unsigned, callInstanceIndex));
+
+ return 1;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::factotLDLT_solvingFinal(FactorLDLTWorkerContext &ref_context, unsigned ownThreadIndex)
+{
+ participateSolvingL1Stripe_X<FSL1S_BLOCK_SIZE, FSL1S_FINAL_B_ROWS>(ref_context.m_A, ref_context.m_ARow, ref_context.m_solvingBlockIndex, ref_context.m_rowSkip,
+ ref_context.m_refSolvingBlockCompletionProgress, ref_context.m_solvingBlockProgressDescriptors, ref_context.m_solvingCellContexts, ownThreadIndex);
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_solvingFinalSync_callback(void *callContext, dcallindex_t dUNUSED(callInstanceIndex), dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ FactorLDLTWorkerContext *ptrContext = (FactorLDLTWorkerContext *)callContext;
+
+ factotLDLT_solvingFinalSync(*ptrContext);
+
+ return 1;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::factotLDLT_solvingFinalSync(FactorLDLTWorkerContext &ref_workerContext)
+{
+ unsigned solvingBlockIndex = ref_workerContext.m_solvingBlockIndex;
+ FactorizationFactorizeL1StripeContext *factorizingFactorizationContext = ref_workerContext.m_factorizingFactorizationContext;
+
+ const unsigned int solvingBlockStep = FSL1S_BLOCK_SIZE;
+ const unsigned factorizingBlockARows = FFL1S_FINAL_A_ROWS;
+ unsigned factorizingBlockCount = deriveScalingAndFactorizingL1StripeBlockCountFromSolvingBlockIndex(solvingBlockIndex, solvingBlockStep, factorizingBlockARows);
+ unsigned blockFactorizingThreadCount = deriveScalingAndFactorizingL1StripeThreadCount(factorizingBlockCount, ref_workerContext.m_allowedThreadCount);
+ initializeCooperativelyScalingAndFactorizingL1Stripe_XMemoryStructures(factorizingFactorizationContext, blockFactorizingThreadCount);
+
+ dCallReleaseeID blockFactorizingSyncReleasee = ref_workerContext.m_calculationFinishReleasee;
+ dIASSERT(solvingBlockIndex == ref_workerContext.m_totalBlockCount - 1);
+
+ dxThreadingBase *threading = ref_workerContext.m_threading;
+
+ if (blockFactorizingThreadCount > 1)
+ {
+ threading->AlterThreadedCallDependenciesCount(blockFactorizingSyncReleasee, blockFactorizingThreadCount - 1);
+ threading->PostThreadedCallsGroup(NULL, blockFactorizingThreadCount - 1, blockFactorizingSyncReleasee, &factotLDLT_scalingAndFactorizingFinal_callback, &ref_workerContext, "FactorLDLT S'n'F Final");
+ }
+
+ factotLDLT_scalingAndFactorizingFinal(ref_workerContext, blockFactorizingThreadCount - 1);
+ threading->AlterThreadedCallDependenciesCount(blockFactorizingSyncReleasee, -1);
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_scalingAndFactorizingFinal_callback(void *callContext, dcallindex_t callInstanceIndex, dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ FactorLDLTWorkerContext *ptrContext = (FactorLDLTWorkerContext *)callContext;
+
+ factotLDLT_scalingAndFactorizingFinal(*ptrContext, dCAST_TO_SMALLER(unsigned, callInstanceIndex));
+
+ return 1;
+}
+
+/*static */
+void ThreadedEquationSolverLDLT::factotLDLT_scalingAndFactorizingFinal(FactorLDLTWorkerContext &ref_workerContext, unsigned ownThreadIndex)
+{
+ unsigned factorizationRow = ref_workerContext.m_solvingBlockIndex * FSL1S_BLOCK_SIZE;
+ participateScalingAndFactorizingL1Stripe_X<FFL1S_FINAL_A_ROWS, FLDLT_D_STRIDE>(ref_workerContext.m_ARow, ref_workerContext.m_d, factorizationRow,
+ ref_workerContext.m_rowSkip, ref_workerContext.m_factorizingFactorizationContext, ownThreadIndex);
+}
+
+
+/*static */
+int ThreadedEquationSolverLDLT::factotLDLT_completion_callback(void *dUNUSED(callContext), dcallindex_t dUNUSED(callInstanceIndex), dCallReleaseeID dUNUSED(callThisReleasee))
+{
+ // Do nothing
+ return 1;
+}
+
+
+//////////////////////////////////////////////////////////////////////////
+// Public interface functions
+
+
+/*extern ODE_API */
+void dFactorLDLT(dReal *A, dReal *d, int n, int nskip1)
+{
+ factorMatrixAsLDLT<1>(A, d, n, nskip1);
+}
+
+
+/*extern ODE_API */
+void dEstimateCooperativelyFactorLDLTResourceRequirements(dResourceRequirementsID requirements,
+ unsigned maximalAllowedThreadCount, unsigned maximalRowCount)
+{
+ dAASSERT(requirements != NULL);
+
+ dxResourceRequirementDescriptor *requirementsDescriptor = (dxResourceRequirementDescriptor *)requirements;
+ ThreadedEquationSolverLDLT::estimateCooperativeFactoringLDLTResourceRequirements(requirementsDescriptor, maximalAllowedThreadCount, maximalRowCount);
+}
+
+/*extern ODE_API */
+void dCooperativelyFactorLDLT(dResourceContainerID resources, unsigned allowedThreadCount,
+ dReal *A, dReal *d, unsigned rowCount, unsigned rowSkip)
+{
+ dAASSERT(resources != NULL);
+
+ dxRequiredResourceContainer *resourceContainer = (dxRequiredResourceContainer *)resources;
+ ThreadedEquationSolverLDLT::cooperativelyFactorLDLT(resourceContainer, allowedThreadCount, A, d, rowCount, rowSkip);
+}