PEER NEES NEESit

• HOME
• USER
• DEVELOPER
• PROJECTS
• SUPPORT
• SITE MAP

• Dev Doc
• API
• Source
• Download
• Builds
• Bug Reports
• Message Board

DistributedSuperLU.cpp Go to the documentation of this file. /* ****************************************************************** ** ** OpenSees - Open System for Earthquake Engineering Simulation ** ** Pacific Earthquake Engineering Research Center ** ** ** ** ** ** (C) Copyright 1999, The Regents of the University of California ** ** All Rights Reserved. ** ** ** ** Commercial use of this program without express permission of the ** ** University of California, Berkeley, is strictly prohibited. See ** ** file 'COPYRIGHT' in main directory for information on usage and ** ** redistribution, and for a DISCLAIMER OF ALL WARRANTIES. ** ** ** ** Developed by: ** ** Frank McKenna (fmckenna@ce.berkeley.edu) ** ** Gregory L. Fenves (fenves@ce.berkeley.edu) ** ** Filip C. Filippou (filippou@ce.berkeley.edu) ** ** ** ** ****************************************************************** */ // $Revision: 1.1 $ // $Date: 2005/12/06 22:21:03 $ // $Source: /usr/local/cvs/OpenSees/SRC/system_of_eqn/linearSOE/sparseGEN/DistributedSuperLU.cpp,v $ // Written: fmk // // Description: This file contains the implementation of DistributedSuperLU // // What: "@(#) DistributedSuperLU.h, revA" #include <DistributedSuperLU.h> #include <SparseGenColLinSOE.h> #include <f2c.h> #include <math.h> #include <Channel.h> #include <FEM_ObjectBroker.h> #include <ID.h> DistributedSuperLU::DistributedSuperLU(int npR, int npC) :SparseGenColLinSolver(SOLVER_TAGS_DistributedSuperLU), gridInit(false), npRow(npR), npCol(npC), processID(0), numChannels(0), theChannels(0) { } DistributedSuperLU::DistributedSuperLU() :SparseGenColLinSolver(SOLVER_TAGS_DistributedSuperLU), gridInit(false), npRow(0), npCol(0), processID(0), numChannels(0), theChannels(0) { } DistributedSuperLU::~DistributedSuperLU() { //Destroy_LU(theSOE->size, &grid, &LUstruct); ScalePermstructFree(&ScalePermstruct); LUstructFree(&LUstruct); //superlu_gridexit(&grid); if (theChannels != 0) delete [] theChannels; // MPI_Comm_free(&comm_SuperLU); } int DistributedSuperLU::solve(void) { if (theSOE == 0) { opserr << "WARNING DistributedSuperLU::solve(void)- "; opserr << " No LinearSOE object has been set\n"; return -1; } // check for quick return if (theSOE->size == 0) return 0; // if subprocess recv A & B from p0 if (processID != 0) { Channel *theChannel = theChannels[0]; theChannel->recvVector(0, 0, (*theSOE->vectB)); Vector vectA(theSOE->A, theSOE->nnz); theChannel->recvVector(0, 0, vectA); } // // if main process, send B & A to all, solve and send back X, B & result else { Vector vectA(theSOE->A, theSOE->nnz); // send B & A to p1 through n-1 for (int j=0; j<numChannels; j++) { Channel *theChannel = theChannels[j]; theChannel->sendVector(0, 0, *(theSOE->vectB)); theChannel->sendVector(0, 0, vectA); } } int info; int iam = grid.iam; if (iam < (npRow * npCol)) { int n = theSOE->size; int nnz = theSOE->nnz; int ldb = n; int nrhs = 1; static double berr[1]; // first copy B into X double *Xptr = theSOE->X; double *Bptr = theSOE->B; for (int i=0; i<n; i++) *(Xptr++) = *(Bptr++); Xptr = theSOE->X; // // set the Fact options: // leave DOFACT if never been factored // otherwise use SamePattern if factored at least once // if ((options.Fact == FACTORED) && (theSOE->factored == false)) { options.Fact = SamePattern; for (int i=0; i<nnz; i++) rowA[i] = theSOE->rowA[i]; } // // solve & mark as factored // pdgssvx_ABglobal(&options, &A, &ScalePermstruct, Xptr, ldb, nrhs, &grid, &LUstruct, berr, &stat, &info); if (theSOE->factored == false) { options.Fact = FACTORED; theSOE->factored = true; } } /* // synch processes again if (processID != 0) { static ID idData(1); idData(0) = info; Channel *theChannel = theChannels[0]; theChannel->sendID(0, 0, idData); } else { for (int j=0; j<numChannels; j++) { static ID idData(1); Channel *theChannel = theChannels[j]; theChannel->recvID(0, 0, idData); } } */ return 0; } int DistributedSuperLU::setSize() { int n = theSOE->size; // // init the super lu process grid // if (gridInit == false) { MPI_Comm comm_world; // MPI_Group group_world; // group_worker; comm_world = MPI_COMM_WORLD; MPI_Comm_group(comm_world, &group_world); // MPI_Group_excl(group_world, 1, 0, &group_worker); /* process 0 not member */ MPI_Comm_create(comm_world, group_world, &comm_SuperLU); superlu_gridinit(comm_SuperLU, npRow, npCol, &grid); // free old structures if resize already called } else { Destroy_LU(theSOE->size, &grid, &LUstruct); ScalePermstructFree(&ScalePermstruct); LUstructFree(&LUstruct); } // // Initialize the statistics variables. // PStatInit(&stat); // // Create compressed column matrix for A. // if (n > 0) { // create the SuperMatrix A int nnz = theSOE->nnz; rowA = new int[nnz]; for (int i=0; i<nnz; i++) rowA[i] = theSOE->rowA[i]; dCreate_CompCol_Matrix_dist(&A, n, n, nnz, theSOE->A, rowA, theSOE->colStartA, SLU_NC, SLU_D, SLU_GE); // // Initialize ScalePermstruct and LUstruct. // ScalePermstructInit(n, n, &ScalePermstruct); LUstructInit(n, n, &LUstruct); } // // set default options // set_default_options_Distributed(&options); // // Initialize the statistics variables. // PStatInit(&stat); return 0; } int DistributedSuperLU::setProcessID(int dTag) { processID = dTag; return 0; } int DistributedSuperLU::setChannels(int nChannels, Channel **theC) { numChannels = nChannels; if (theChannels != 0) delete [] theChannels; theChannels = new Channel *[numChannels]; if (theChannels == 0) { opserr << "DistributedSuperLU::sendSelf() - failed to allocate channel array of size: " << numChannels << endln; return -1; } for (int i=0; i<numChannels; i++) theChannels[i] = theC[i]; return 0; } int DistributedSuperLU::sendSelf(int cTag, Channel &theChannel) { int sendID =0; // if P0 check if already sent. If already sent use old processID; if not allocate a new process // id for remote part of object, enlarge channel * to hold a channel * for this remote object. // if not P0, send current processID if (processID == 0) { // check if already using this object bool found = false; for (int i=0; i<numChannels; i++) if (theChannels[i] == &theChannel) { sendID = i+1; found = true; } // if new object, enlarge Channel pointers to hold new channel * & allocate new ID if (found == false) { int nextNumChannels = numChannels + 1; Channel **nextChannels = new Channel *[nextNumChannels]; if (nextNumChannels == 0) { opserr << "DistributedSuperLU::sendSelf() - failed to allocate channel array of size: " << nextNumChannels << endln; return -1; } for (int i=0; i<numChannels; i++) nextChannels[i] = theChannels[i]; nextChannels[numChannels] = &theChannel; numChannels = nextNumChannels; if (theChannels != 0) delete [] theChannels; theChannels = nextChannels; // allocate new processID for remote object sendID = numChannels; } } else sendID = processID; static ID idData(3); idData(0) = sendID; idData(1) = npRow; idData(2) = npCol; int res = theChannel.sendID(0, cTag, idData); if (res < 0) { opserr <<"WARNING DistributedSuperLU::sendSelf() - failed to send data\n"; return -1; } return 0; } int DistributedSuperLU::recvSelf(int cTag, Channel &theChannel, FEM_ObjectBroker &theBroker) { static ID idData(3); int res = theChannel.recvID(0, cTag, idData); if (res < 0) { opserr <<"WARNING DistributedSuperLU::recvSelf() - failed to receive data\n"; return -1; } processID = idData(0); npRow = idData(1); npCol = idData(2); numChannels = 1; theChannels = new Channel *[1]; theChannels[0] = &theChannel; return 0; } Generated on Mon Oct 23 15:05:29 2006 for OpenSees by  1.5.0

---

opensees-support @ berkeley.edu

Supported by the National Science Foundation

©2006, UC Regents