void dgSolver::CalculateJointForces(const dgBodyCluster& cluster, dgBodyInfo* const bodyArray, dgJointInfo* const jointArray, dgFloat32 timestep)
{
m_cluster = &cluster;
m_bodyArray = bodyArray;
m_jointArray = jointArray;
m_timestep = timestep;
m_invTimestep = (timestep > dgFloat32(0.0f)) ? dgFloat32(1.0f) / timestep : dgFloat32(0.0f);
m_invStepRK = dgFloat32 (0.25f);
m_timestepRK = m_timestep * m_invStepRK;
m_invTimestepRK = m_invTimestep * dgFloat32 (4.0f);
m_threadCounts = m_world->GetThreadCount();
m_solverPasses = m_world->GetSolverIterations();
dgInt32 mask = -dgInt32(DG_SOA_WORD_GROUP_SIZE - 1);
m_jointCount = ((m_cluster->m_jointCount + DG_SOA_WORD_GROUP_SIZE - 1) & mask) / DG_SOA_WORD_GROUP_SIZE;
m_bodyProxyArray = dgAlloca(dgBodyProxy, cluster.m_bodyCount);
m_bodyJacobiansPairs = dgAlloca(dgBodyJacobianPair, cluster.m_jointCount * 2);
m_soaRowStart = dgAlloca(dgInt32, cluster.m_jointCount / DG_SOA_WORD_GROUP_SIZE + 1);
InitWeights();
InitBodyArray();
InitJacobianMatrix();
CalculateForces();
}
void dgWorldDynamicUpdate::CalculateReactionForcesParallel (const dgIsland* const island, dgFloat32 timestep) const
{
dgParallelSolverSyncData syncData;
dgWorld* const world = (dgWorld*) this;
syncData.m_bodyLocks = dgAlloca (dgInt32, island->m_bodyCount + 1024);
syncData.m_jointConflicts = dgAlloca (dgParallelSolverSyncData::dgParallelJointMap, island->m_jointCount + 1024);
memset (syncData.m_bodyLocks, 0, island->m_bodyCount * sizeof (dgInt32));
dgInt32 bodyCount = island->m_bodyCount;
dgInt32 jointsCount = island->m_jointCount;
dgJointInfo* const constraintArrayPtr = (dgJointInfo*) &world->m_jointsMemory[0];
dgJointInfo* const constraintArray = &constraintArrayPtr[island->m_jointStart];
LinearizeJointParallelArray (&syncData, constraintArray, island);
dgJacobian* const internalForces = &world->m_solverMemory.m_internalForces[0];
internalForces[0].m_linear = dgVector::m_zero;
internalForces[0].m_angular = dgVector::m_zero;
const dgInt32 maxPasses = 4;
syncData.m_timestep = timestep;
syncData.m_invTimestep = (timestep > dgFloat32 (0.0f)) ? dgFloat32 (1.0f) / timestep : dgFloat32 (0.0f);
syncData.m_invStepRK = (dgFloat32 (1.0f) / dgFloat32 (maxPasses));
syncData.m_timestepRK = syncData.m_timestep * syncData.m_invStepRK;
syncData.m_invTimestepRK = syncData.m_invTimestep * dgFloat32 (maxPasses);
syncData.m_maxPasses = maxPasses;
syncData.m_passes = world->m_solverMode;
syncData.m_bodyCount = bodyCount;
syncData.m_jointCount = jointsCount;
syncData.m_atomicIndex = 0;
syncData.m_island = island;
InitilizeBodyArrayParallel (&syncData);
BuildJacobianMatrixParallel (&syncData);
SolverInitInternalForcesParallel (&syncData);
CalculateForcesGameModeParallel (&syncData);
IntegrateIslandParallel(&syncData);
}
void dgWorldDynamicUpdate::CalculateClusterReactionForces(const dgBodyCluster* const cluster, dgInt32 threadID, dgFloat32 timestep, dgFloat32 maxAccNorm) const
{
dTimeTrackerEvent(__FUNCTION__);
dgWorld* const world = (dgWorld*) this;
const dgInt32 bodyCount = cluster->m_bodyCount;
// const dgInt32 jointCount = island->m_jointCount;
const dgInt32 jointCount = cluster->m_activeJointCount;
dgJacobian* const internalForces = &m_solverMemory.m_internalForcesBuffer[cluster->m_bodyStart];
dgBodyInfo* const bodyArrayPtr = (dgBodyInfo*)&world->m_bodiesMemory[0];
dgJointInfo* const constraintArrayPtr = (dgJointInfo*)&world->m_jointsMemory[0];
dgBodyInfo* const bodyArray = &bodyArrayPtr[cluster->m_bodyStart];
dgJointInfo* const constraintArray = &constraintArrayPtr[cluster->m_jointStart];
dgJacobianMatrixElement* const matrixRow = &m_solverMemory.m_jacobianBuffer[cluster->m_rowsStart];
const dgInt32 derivativesEvaluationsRK4 = 4;
dgFloat32 invTimestep = (timestep > dgFloat32(0.0f)) ? dgFloat32(1.0f) / timestep : dgFloat32(0.0f);
dgFloat32 invStepRK = (dgFloat32(1.0f) / dgFloat32(derivativesEvaluationsRK4));
dgFloat32 timestepRK = timestep * invStepRK;
dgFloat32 invTimestepRK = invTimestep * dgFloat32(derivativesEvaluationsRK4);
dgAssert(bodyArray[0].m_body == world->m_sentinelBody);
dgVector speedFreeze2(world->m_freezeSpeed2 * dgFloat32(0.1f));
dgVector freezeOmega2(world->m_freezeOmega2 * dgFloat32(0.1f));
dgJointAccelerationDecriptor joindDesc;
joindDesc.m_timeStep = timestepRK;
joindDesc.m_invTimeStep = invTimestepRK;
joindDesc.m_firstPassCoefFlag = dgFloat32(0.0f);
dgInt32 skeletonCount = 0;
dgInt32 skeletonMemorySizeInBytes = 0;
dgInt32 lru = dgAtomicExchangeAndAdd(&dgSkeletonContainer::m_lruMarker, 1);
dgSkeletonContainer* skeletonArray[DG_MAX_SKELETON_JOINT_COUNT];
dgInt32 memorySizes[DG_MAX_SKELETON_JOINT_COUNT];
for (dgInt32 i = 1; i < bodyCount; i++) {
dgDynamicBody* const body = (dgDynamicBody*)bodyArray[i].m_body;
dgSkeletonContainer* const container = body->GetSkeleton();
if (container && (container->m_lru != lru)) {
container->m_lru = lru;
memorySizes[skeletonCount] = container->GetMemoryBufferSizeInBytes(constraintArray, matrixRow);
skeletonMemorySizeInBytes += memorySizes[skeletonCount];
skeletonArray[skeletonCount] = container;
skeletonCount++;
dgAssert(skeletonCount < dgInt32(sizeof(skeletonArray) / sizeof(skeletonArray[0])));
}
}
dgInt8* const skeletonMemory = (dgInt8*)dgAlloca(dgVector, skeletonMemorySizeInBytes / sizeof(dgVector));
dgAssert((dgInt64(skeletonMemory) & 0x0f) == 0);
skeletonMemorySizeInBytes = 0;
for (dgInt32 i = 0; i < skeletonCount; i++) {
skeletonArray[i]->InitMassMatrix(constraintArray, matrixRow, &skeletonMemory[skeletonMemorySizeInBytes]);
skeletonMemorySizeInBytes += memorySizes[i];
}
const dgInt32 passes = world->m_solverMode;
for (dgInt32 step = 0; step < derivativesEvaluationsRK4; step++) {
for (dgInt32 i = 0; i < jointCount; i++) {
dgJointInfo* const jointInfo = &constraintArray[i];
dgConstraint* const constraint = jointInfo->m_joint;
joindDesc.m_rowsCount = jointInfo->m_pairCount;
joindDesc.m_rowMatrix = &matrixRow[jointInfo->m_pairStart];
constraint->JointAccelerations(&joindDesc);
}
joindDesc.m_firstPassCoefFlag = dgFloat32(1.0f);
dgFloat32 accNorm(maxAccNorm * dgFloat32(2.0f));
for (dgInt32 i = 0; (i < passes) && (accNorm > maxAccNorm); i++) {
accNorm = dgFloat32(0.0f);
for (dgInt32 j = 0; j < jointCount; j++) {
dgJointInfo* const jointInfo = &constraintArray[j];
dgFloat32 accel = CalculateJointForceGaussSeidel(jointInfo, bodyArray, internalForces, matrixRow, maxAccNorm);
accNorm = (accel > accNorm) ? accel : accNorm;
}
}
for (dgInt32 j = 0; j < skeletonCount; j++) {
skeletonArray[j]->CalculateJointForce(constraintArray, bodyArray, internalForces, matrixRow);
}
if (timestepRK != dgFloat32(0.0f)) {
dgVector timestep4(timestepRK);
for (dgInt32 i = 1; i < bodyCount; i++) {
dgDynamicBody* const body = (dgDynamicBody*)bodyArray[i].m_body;
dgAssert(body->m_index == i);
if (body->IsRTTIType(dgBody::m_dynamicBodyRTTI)) {
const dgJacobian& forceAndTorque = internalForces[i];
dgVector force(body->m_externalForce + forceAndTorque.m_linear);
dgVector torque(body->m_externalTorque + forceAndTorque.m_angular);
dgVector velocStep((force.Scale4(body->m_invMass.m_w)) * timestep4);
dgVector omegaStep((body->m_invWorldInertiaMatrix.RotateVector(torque)) * timestep4);
body->m_veloc += velocStep;
body->m_omega += omegaStep;
dgAssert(body->m_veloc.m_w == dgFloat32(0.0f));
dgAssert(body->m_omega.m_w == dgFloat32(0.0f));
}
}
} else {
for (dgInt32 i = 1; i < bodyCount; i++) {
dgDynamicBody* const body = (dgDynamicBody*)bodyArray[i].m_body;
const dgVector& linearMomentum = internalForces[i].m_linear;
const dgVector& angularMomentum = internalForces[i].m_angular;
body->m_veloc += linearMomentum.Scale4(body->m_invMass.m_w);
body->m_omega += body->m_invWorldInertiaMatrix.RotateVector(angularMomentum);
}
}
}
dgInt32 hasJointFeeback = 0;
if (timestepRK != dgFloat32(0.0f)) {
for (dgInt32 i = 0; i < jointCount; i++) {
dgJointInfo* const jointInfo = &constraintArray[i];
dgConstraint* const constraint = jointInfo->m_joint;
const dgInt32 first = jointInfo->m_pairStart;
const dgInt32 count = jointInfo->m_pairCount;
for (dgInt32 j = 0; j < count; j++) {
dgJacobianMatrixElement* const row = &matrixRow[j + first];
dgFloat32 val = row->m_force;
dgAssert(dgCheckFloat(val));
row->m_jointFeebackForce->m_force = val;
row->m_jointFeebackForce->m_impact = row->m_maxImpact * timestepRK;
}
hasJointFeeback |= (constraint->m_updaFeedbackCallback ? 1 : 0);
}
const dgVector invTime(invTimestep);
const dgVector maxAccNorm2(maxAccNorm * maxAccNorm);
for (dgInt32 i = 1; i < bodyCount; i++) {
dgBody* const body = bodyArray[i].m_body;
CalculateNetAcceleration(body, invTime, maxAccNorm2);
}
if (hasJointFeeback) {
for (dgInt32 i = 0; i < jointCount; i++) {
if (constraintArray[i].m_joint->m_updaFeedbackCallback) {
constraintArray[i].m_joint->m_updaFeedbackCallback(*constraintArray[i].m_joint, timestep, threadID);
}
}
}
} else {
for (dgInt32 i = 1; i < bodyCount; i++) {
dgBody* const body = bodyArray[i].m_body;
dgAssert(body->IsRTTIType(dgBody::m_dynamicBodyRTTI) || body->IsRTTIType(dgBody::m_kinematicBodyRTTI));
body->m_accel = dgVector::m_zero;
body->m_alpha = dgVector::m_zero;
}
}
}