void ML_Read_Matrix_Dimensions(const char *filename, int *numGlobalRows, Epetra_Comm &Comm)
{
char line[35], token1[35], token2[35], token3[35], token4[35], token5[35];
int lineLength = 1025;
FILE *fid = fopen(filename,"r");
int N, NZ;
if(fgets(line, lineLength, fid)==0) {
if (fid!=0) fclose(fid);
ML_Exit(Comm.MyPID(),"error opening matrix file", EXIT_FAILURE);
}
if(sscanf(line, "%s %s %s %s %s", token1, token2, token3, token4, token5 )==0) {
if (fid!=0) fclose(fid);
ML_Exit(Comm.MyPID(),"error reading matrix file header", EXIT_FAILURE);
}
if (strcmp(token1, "%%MatrixMarket") || strcmp(token2, "matrix") ||
strcmp(token3, "coordinate") || strcmp(token4, "real") ||
strcmp(token5, "general"))
{
if (fid!=0) fclose(fid);
ML_Exit(Comm.MyPID(),"error reading matrix file header", EXIT_FAILURE);
}
// Next, strip off header lines (which start with "%")
do {
if(fgets(line, lineLength, fid)==0) {
if (fid!=0) fclose(fid);
ML_Exit(Comm.MyPID(),"error reading matrix file comments", EXIT_FAILURE);
}
} while (line[0] == '%');
// Next get problem dimensions: M, N, NZ
if(sscanf(line, "%d %d %d", numGlobalRows, &N, &NZ)==0) {
if (fid!=0) fclose(fid);
ML_Exit(Comm.MyPID(),"error reading matrix file dimensions", EXIT_FAILURE);
}
} //ML_Read_Matrix_Dimensions()
int fevec6(Epetra_Comm& Comm, bool verbose)
{
int NumElements = 4;
Epetra_Map Map(NumElements, 0, Comm);
Epetra_FEVector x1(Map);
x1.PutScalar (0);
// let all processors set global entry 0 to 1
const int GID = 0;
const double value = 1;
x1.ReplaceGlobalValues(1, &GID, &value);
x1.GlobalAssemble (Insert);
if (Comm.MyPID()==0)
std::cout << "Entry " << GID << " after construct & set: "
<< x1[0][0] << std::endl;
x1.PutScalar(0);
// re-apply 1 to the vector, but only on the
// owning processor. should be enough to set
// the value (as non-local data in x1 should
// have been eliminated after calling
// GlobalAssemble).
if (Comm.MyPID()==0)
x1.ReplaceGlobalValues(1, &GID, &value);
x1.GlobalAssemble (Insert);
if (Comm.MyPID()==0) {
std::cout << "Entry " << GID << " after PutScalar & set: "
<< x1[0][0] << std::endl;
if (x1[0][0] != value) return -1;
}
return 0;
}
int generateHyprePrintOut(const char *filename, const Epetra_Comm &comm){
int MyPID = comm.MyPID();
int NumProc = comm.NumProc();
int N = 100;
int ilower = MyPID * N;
int iupper = (MyPID+1)*N-1;
double filePID = (double)MyPID/(double)100000;
std::ostringstream stream;
// Using setprecision() puts it in the std::string
stream << std::setiosflags(std::ios::fixed) << std::setprecision(5) << filePID;
// Then just ignore the first character
std::string fileName(filename);
fileName += stream.str().substr(1,7);
std::ofstream myfile(fileName.c_str());
if(myfile.is_open()){
myfile << ilower << " " << iupper << " " << ilower << " " << iupper << std::endl;
for(int i = ilower; i <= iupper; i++){
for(int j=i-5; j <= i+5; j++){
if(j >= 0 && j < N*NumProc)
myfile << i << " " << j << " " << (double)rand()/(double)RAND_MAX << std::endl;
}
}
myfile.close();
return 0;
} else {
std::cout << "\nERROR:\nCouldn't open file.\n";
return -1;
}
}
void random_distribution_1D(
itype nrows, // Number of global matrix rows
Epetra_Comm &comm, // Epetra communicator to be used in maps
Epetra_Map **rowMap, // OUTPUT: pointer to row map to be created
long long offsetEpetra64
)
{
// Randomly assign matrix rows to processor's row Map.
int me = comm.MyPID();
int np = comm.NumProc();
vector<itype> myGlobalElements(1.2 * (nrows / np) + 1);
int nMyRows = 0;
srandom(1);
double denom = (double) RAND_MAX + 1.;
for (itype i = 0; i < nrows; i++) {
int p = (int) ((double) np * (double) random() / denom);
if (p == me) {
if (nMyRows >= myGlobalElements.size())
myGlobalElements.resize(1.5*myGlobalElements.size());
myGlobalElements[nMyRows] = i + offsetEpetra64;
nMyRows++;
}
}
*rowMap = new Epetra_Map(nrows, nMyRows, &myGlobalElements[0], 0, comm);
}
bool global_check_for_flag_on_proc_0(const char* flag,
int numargs,
char** strargs,
const Epetra_Comm& comm)
{
int mypid = comm.MyPID();
int numprocs = comm.NumProc();
int flag_found = 0;
if (mypid==0) {
for(int i=0; i<numargs; ++i) {
if (strargs[i]==0) continue;
if (strcmp(flag, strargs[i]) == 0) {
flag_found = 1;
break;
}
}
}
if (numprocs > 1) {
comm.Broadcast(&flag_found, 1, 0);
}
bool return_value = flag_found==1 ? true : false;
return( return_value );
}
void show_matrix(const char *txt, const Epetra_RowMatrix &matrix, const Epetra_Comm &comm)
{
int me = comm.MyPID();
if (comm.NumProc() > 10){
if (me == 0){
std::cout << txt << std::endl;
std::cout << "Printed matrix format only works for 10 or fewer processes" << std::endl;
}
return;
}
int numRows = matrix.NumGlobalRows();
int numCols = matrix.NumGlobalCols();
if ((numRows > 200) || (numCols > 500)){
if (me == 0){
std::cerr << txt << std::endl;
std::cerr << "show_matrix: problem is too large to display" << std::endl;
}
return;
}
int *myA = new int [numRows * numCols];
make_my_A(matrix, myA, comm);
printMatrix(txt, myA, NULL, NULL, numRows, numCols, comm);
delete [] myA;
}
int HypreFileToCrsMatrix(const char *filename, const Epetra_Comm &comm, Epetra_CrsMatrix *&Matrix){
int MyPID = comm.MyPID();
// This double will be in the format we want for the extension besides the leading zero
double filePID = (double)MyPID/(double)100000;
std::ostringstream stream;
// Using setprecision() puts it in the string
stream << std::setiosflags(std::ios::fixed) << std::setprecision(5) << filePID;
// Then just ignore the first character
std::string fileName(filename);
fileName += stream.str().substr(1,7);
// Open the file
std::ifstream file(fileName.c_str());
string line;
if(file.is_open()){
std::getline(file, line);
int ilower, iupper;
std::istringstream istream(line);
// The first line of the file has the beginning and ending rows
istream >> ilower;
istream >> iupper;
// Using those we can create a row map
Epetra_Map RowMap(-1, iupper-ilower+1, 0, comm);
Matrix = new Epetra_CrsMatrix(Copy, RowMap, 0);
int currRow = -1;
int counter = 0;
std::vector<int> indices;
std::vector<double> values;
while(!file.eof()){
std::getline(file, line);
std::istringstream lineStr(line);
int row, col;
double val;
lineStr >> row;
lineStr >> col;
lineStr >> val;
if(currRow == -1) currRow = row; // First line
if(row == currRow){
// add to the vector
counter = counter + 1;
indices.push_back(col);
values.push_back(val);
} else {
Matrix->InsertGlobalValues(currRow, counter, &values[0], &indices[0]);
indices.clear();
values.clear();
counter = 0;
currRow = row;
// make a new vector
indices.push_back(col);
values.push_back(val);
counter = counter + 1;
}
}
Matrix->InsertGlobalValues(currRow, counter, &values[0], &indices[0]);
Matrix->Comm().Barrier();
Matrix->FillComplete();
file.close();
return 0;
} else {
int special_submap_import_test(Epetra_Comm& Comm)
{
int localProc = Comm.MyPID();
//set up ids_source and ids_target such that ids_source are only
//a subset of ids_target, and furthermore that ids_target are ordered
//such that the LIDs don't match up. In other words, even if gid 2 does
//exist in both ids_source and ids_target, it will correspond to different
//LIDs on at least 1 proc.
//
//This is to test a certain bug-fix in Epetra_Import where the 'RemoteLIDs'
//array wasn't being calculated correctly on all procs.
long long ids_source[1];
ids_source[0] = localProc*2+2;
long long ids_target[3];
ids_target[0] = localProc*2+2;
ids_target[1] = localProc*2+1;
ids_target[2] = localProc*2+0;
Epetra_Map map_source((long long) -1, 1, &ids_source[0], 0LL, Comm);
Epetra_Map map_target((long long) -1, 3, &ids_target[0], 0LL, Comm);
Epetra_Import importer(map_target, map_source);
Epetra_LongLongVector vec_source(map_source);
Epetra_LongLongVector vec_target(map_target);
vec_target.PutValue(0);
//set vec_source's contents so that entry[i] == GID[i].
long long* GIDs = map_source.MyGlobalElements64();
for(int i=0; i<map_source.NumMyElements(); ++i) {
vec_source[i] = GIDs[i];
}
//Import vec_source into vec_target. This should result in the contents
//of vec_target remaining 0 for the entries that don't exist in vec_source,
//and other entries should be equal to the corresponding GID in the map.
vec_target.Import(vec_source, importer, Insert);
GIDs = map_target.MyGlobalElements64();
int test_failed = 0;
//the test passes if the i-th entry in vec_target equals either 0 or
//GIDs[i].
for(int i=0; i<vec_target.MyLength(); ++i) {
if (vec_target[i] != GIDs[i] && vec_target[i] != 0) test_failed = 1;
}
int global_result;
Comm.MaxAll(&test_failed, &global_result, 1);
//If test didn't fail on any procs, global_result should be 0.
//If test failed on any proc, global_result should be 1.
return global_result;
}
void show_matrix(const char *txt, const Epetra_LinearProblem &problem, const Epetra_Comm &comm)
{
int me = comm.MyPID();
if (comm.NumProc() > 10){
if (me == 0){
std::cout << txt << std::endl;
std::cout << "Printed matrix format only works for 10 or fewer processes" << std::endl;
}
return;
}
Epetra_RowMatrix *matrix = problem.GetMatrix();
Epetra_MultiVector *lhs = problem.GetLHS();
Epetra_MultiVector *rhs = problem.GetRHS();
int numRows = matrix->NumGlobalRows();
int numCols = matrix->NumGlobalCols();
if ((numRows > 200) || (numCols > 500)){
if (me == 0){
std::cerr << txt << std::endl;
std::cerr << "show_matrix: problem is too large to display" << std::endl;
}
return;
}
int *myA = new int [numRows * numCols];
make_my_A(*matrix, myA, comm);
int *myX = new int [numCols];
int *myB = new int [numRows];
memset(myX, 0, sizeof(int) * numCols);
memset(myB, 0, sizeof(int) * numRows);
const Epetra_BlockMap &lhsMap = lhs->Map();
const Epetra_BlockMap &rhsMap = rhs->Map();
int base = lhsMap.IndexBase();
for (int j=0; j < lhsMap.NumMyElements(); j++){
int colGID = lhsMap.GID(j);
myX[colGID - base] = me + 1;
}
for (int i=0; i < rhsMap.NumMyElements(); i++){
int rowGID = rhsMap.GID(i);
myB[rowGID - base] = me + 1;
}
printMatrix(txt, myA, myX, myB, numRows, numCols, comm);
delete [] myA;
delete [] myX;
delete [] myB;
}
Teuchos::RCP< Epetra_LinearProblem >
build_problem(Teuchos::ParameterList& test_params,
const Epetra_Comm& comm)
{
Teuchos::Time timer("build_problem");
timer.start();
Epetra_CrsMatrix* A;
Epetra_Vector* b = NULL;
std::string mm_file("not specified");
std::string rhs_mm_file("not specified");
helper::GetParameter(test_params, "mm_file", mm_file);
helper::GetParameter(test_params, "rhs_mm_file", rhs_mm_file);
std::string hb_file("not specified");
helper::GetParameter(test_params, "hb_file", hb_file);
if (mm_file != "not specified") {
if (comm.MyPID() == 0) {
std::cout << "Matrix-Market file: " << mm_file << std::endl;
}
A = read_matrix_mm(mm_file, comm);
if (rhs_mm_file != "not specified") {
if (comm.MyPID() == 0) {
std::cout << "Matrix-Market file: " << rhs_mm_file << std::endl;
}
b = read_vector_mm(rhs_mm_file, comm);
}
}
else if (hb_file != "not specified") {
read_matrix_hb(hb_file, comm, A, b);
}
else {
throw std::runtime_error("No matrix file specified.");
}
Teuchos::RCP<Epetra_LinearProblem> problem = build_problem_mm(test_params, A, b);
timer.stop();
if (comm.MyPID() == 0) {
std::cout << "proc 0 time to read matrix & create problem: " << timer.totalElapsedTime()
<< std::endl;
}
return problem;
}
// Do something with the given communicator. In this case, we just
// print Epetra's version to the given output stream, on Process 0.
void
exampleRoutine (const Epetra_Comm& comm,
std::ostream& out)
{
if (comm.MyPID () == 0) {
// On (MPI) Process 0, print out the Epetra software version.
out << Epetra_Version () << std::endl << std::endl;
}
}
static void print_out(const Epetra_Comm& Comm, const int level, const char* what)
{
if (Comm.MyPID() == 0 && ML_Get_PrintLevel() > 2)
#ifdef TFLOP
printf("Amesos (level %d) : Building %s\n", level, what);
#else
std::cout << "Amesos (level " << level << ") : Building " << what << "\n";
#endif
}
//==============================================================================
Poisson2dOperator::Poisson2dOperator(int nx, int ny, const Epetra_Comm & comm)
: nx_(nx),
ny_(ny),
useTranspose_(false),
comm_(comm),
map_(0),
numImports_(0),
importIDs_(0),
importMap_(0),
importer_(0),
importX_(0),
Label_(0) {
Label_ = "2D Poisson Operator";
int numProc = comm.NumProc(); // Get number of processors
int myPID = comm.MyPID(); // My rank
if (2*numProc > ny) { // ny must be >= 2*numProc (to avoid degenerate cases)
ny = 2*numProc;
ny_ = ny;
std::cout << " Increasing ny to " << ny << " to avoid degenerate distribution on " << numProc << " processors." << std::endl;
}
int chunkSize = ny/numProc;
int remainder = ny%numProc;
if (myPID+1 <= remainder) chunkSize++; // add on remainder
myny_ = chunkSize;
map_ = new Epetra_Map(-1LL, ((long long)nx)*chunkSize, 0, comm_);
if (numProc>1) {
// Build import GID list to build import map and importer
if (myPID>0) numImports_ += nx;
if (myPID+1<numProc) numImports_ += nx;
if (numImports_>0) importIDs_ = new long long[numImports_];
long long * ptr = importIDs_;
long long minGID = map_->MinMyGID64();
long long maxGID = map_->MaxMyGID64();
if (myPID>0) for (int i=0; i< nx; i++) *ptr++ = minGID - nx + i;
if (myPID+1<numProc) for (int i=0; i< nx; i++) *ptr++ = maxGID + i +1;
// At the end of the above step importIDs_ will have a list of global IDs that are needed
// to compute the matrix multiplication operation on this processor. Now build import map
// and importer
importMap_ = new Epetra_Map(-1LL, numImports_, importIDs_, 0LL, comm_);
importer_ = new Epetra_Import(*importMap_, *map_);
}
}
int alternate_import_constructor_test(Epetra_Comm& Comm) {
int rv=0;
int nodes_per_proc=10;
int numprocs = Comm.NumProc();
int mypid = Comm.MyPID();
// Only run if we have multiple procs & MPI
if(numprocs==0) return 0;
#ifndef HAVE_MPI
return 0;
#endif
// Build Map 1 - linear
Epetra_Map Map1((long long)-1,nodes_per_proc,(long long)0,Comm);
// Build Map 2 - mod striped
std::vector<long long> MyGIDs(nodes_per_proc);
for(int i=0; i<nodes_per_proc; i++)
MyGIDs[i] = (mypid*nodes_per_proc + i) % numprocs;
Epetra_Map Map2((long long)-1,nodes_per_proc,&MyGIDs[0],(long long)0,Comm);
// For testing
Epetra_LongLongVector Source(Map1), Target(Map2);
// Build Import 1 - normal
Epetra_Import Import1(Map2,Map1);
rv = rv|| test_import_gid("Alt test: 2 map constructor",Source,Target, Import1);
// Build Import 2 - no-comm constructor
int Nremote=Import1.NumRemoteIDs();
const int * RemoteLIDs = Import1.RemoteLIDs();
std::vector<int> RemotePIDs(Nremote+1); // I hate you, stl vector....
std::vector<int> AllPIDs;
Epetra_Util::GetPids(Import1,AllPIDs,true);
for(int i=0; i<Nremote; i++) {
RemotePIDs[i]=AllPIDs[RemoteLIDs[i]];
}
Epetra_Import Import2(Import1.TargetMap(),Import1.SourceMap(),Nremote,&RemotePIDs[0],Import1.NumExportIDs(),Import1.ExportLIDs(),Import1.ExportPIDs());
rv = rv || test_import_gid("Alt test: no comm constructor",Source,Target,Import2);
// Build Import 3 - Remotes only
Epetra_Import Import3(Import1.TargetMap(),Import1.SourceMap(),Nremote,&RemotePIDs[0]);
rv = rv || test_import_gid("Alt test: remote only constructor",Source,Target, Import3);
return rv;
}
int combine_mode_test(Epetra_Comm& Comm)
{
int localProc = Comm.MyPID();
long long ids_source[1];
ids_source[0] = localProc*2+2;
long long ids_target[3];
ids_target[0] = localProc*2+2;
ids_target[1] = localProc*2+1;
ids_target[2] = localProc*2+0;
Epetra_Map map_source((long long) -1, 1, &ids_source[0], 0LL, Comm);
Epetra_Map map_target((long long) -1, 3, &ids_target[0], 0LL, Comm);
Epetra_Import importer(map_target, map_source);
Epetra_LongLongVector vec_source(map_source);
Epetra_LongLongVector vec_target(map_target);
vec_target.PutValue(0);
//set vec_source's contents so that entry[i] == GID[i].
long long* GIDs = map_source.MyGlobalElements64();
for(int i=0; i<map_source.NumMyElements(); ++i) {
vec_source[i] = GIDs[i];
}
//Import vec_source into vec_target. This should result in the contents
//of vec_target remaining 0 for the entries that don't exist in vec_source,
//and other entries should be equal to the corresponding GID in the map.
vec_target.Import(vec_source, importer, Insert);
GIDs = map_target.MyGlobalElements64();
int test_failed = 0;
//the test passes if the i-th entry in vec_target equals either 0 or
//GIDs[i].
for(int i=0; i<vec_target.MyLength(); ++i) {
if (vec_target[i] != GIDs[i] && vec_target[i] != 0) test_failed = 1;
}
int global_result;
Comm.MaxAll(&test_failed, &global_result, 1);
//If test didn't fail on any procs, global_result should be 0.
//If test failed on any proc, global_result should be 1.
return global_result;
}
//============================================================================
void Ifpack_BreakForDebugger(Epetra_Comm& Comm)
{
char hostname[80];
char buf[80];
if (Comm.MyPID() == 0) cout << "Host and Process Ids for tasks" << endl;
for (int i = 0; i <Comm.NumProc() ; i++) {
if (i == Comm.MyPID() ) {
#if defined(TFLOP) || defined(JANUS_STLPORT)
sprintf(buf, "Host: %s PID: %d", "janus", getpid());
#elif defined(_WIN32)
sprintf(buf,"Windows compiler, unknown hostname and PID!");
#else
gethostname(hostname, sizeof(hostname));
sprintf(buf, "Host: %s\tComm.MyPID(): %d\tPID: %d",
hostname, Comm.MyPID(), getpid());
#endif
printf("%s\n",buf);
fflush(stdout);
#if !( defined(_WIN32) )
sleep(1);
#endif
}
}
if(Comm.MyPID() == 0) {
printf("\n");
printf("** Pausing to attach debugger...\n");
printf("** You may now attach debugger to the processes listed above.\n");
printf( "**\n");
printf( "** Enter a character to continue > "); fflush(stdout);
char go;
scanf("%c",&go);
}
Comm.Barrier();
}
//==============================================================================
// Epetra_BlockMap constructor function for a Epetra-defined uniform linear distribution of constant size elements.
void Epetra_BlockMap::ConstructAutoUniform(long long NumGlobal_Elements, int Element_Size, int Index_Base, const Epetra_Comm& comm, bool IsLongLong)
{
// Each processor gets roughly numGlobalPoints/p points
// This routine automatically defines a linear partitioning of a
// map with numGlobalPoints across the processors
// specified in the given Epetra_Comm
if (NumGlobal_Elements < 0)
throw ReportError("NumGlobal_Elements = " + toString(NumGlobal_Elements) + ". Should be >= 0.", -1);
if (Element_Size <= 0)
throw ReportError("ElementSize = " + toString(Element_Size) + ". Should be > 0.", -2);
BlockMapData_ = new Epetra_BlockMapData(NumGlobal_Elements, Element_Size, Index_Base, comm, IsLongLong);
int NumProc = comm.NumProc();
BlockMapData_->ConstantElementSize_ = true;
BlockMapData_->LinearMap_ = true;
int MyPID = comm.MyPID();
if(BlockMapData_->NumGlobalElements_ / NumProc > (long long) std::numeric_limits<int>::max())
throw ReportError("Epetra_BlockMap::ConstructAutoUniform: Error. Not enough space for elements on each processor", -99);
BlockMapData_->NumMyElements_ = (int) (BlockMapData_->NumGlobalElements_ / NumProc);
int remainder = (int) (BlockMapData_->NumGlobalElements_ % NumProc); // remainder will fit int
int start_index = MyPID * (BlockMapData_->NumMyElements_ + 1);
if (MyPID < remainder)
BlockMapData_->NumMyElements_++;
else
start_index -= (MyPID - remainder);
BlockMapData_->NumGlobalPoints_ = BlockMapData_->NumGlobalElements_ * BlockMapData_->ElementSize_;
BlockMapData_->NumMyPoints_ = BlockMapData_->NumMyElements_ * BlockMapData_->ElementSize_;
BlockMapData_->MinMyElementSize_ = BlockMapData_->ElementSize_;
BlockMapData_->MaxMyElementSize_ = BlockMapData_->ElementSize_;
BlockMapData_->MinElementSize_ = BlockMapData_->ElementSize_;
BlockMapData_->MaxElementSize_ = BlockMapData_->ElementSize_;
BlockMapData_->MinAllGID_ = BlockMapData_->IndexBase_;
BlockMapData_->MaxAllGID_ = BlockMapData_->MinAllGID_ + BlockMapData_->NumGlobalElements_ - 1;
BlockMapData_->MinMyGID_ = start_index + BlockMapData_->IndexBase_;
BlockMapData_->MaxMyGID_ = BlockMapData_->MinMyGID_ + BlockMapData_->NumMyElements_ - 1;
BlockMapData_->DistributedGlobal_ = IsDistributedGlobal(BlockMapData_->NumGlobalElements_, BlockMapData_->NumMyElements_);
EndOfConstructorOps();
}
static int make_my_A(const Epetra_RowMatrix &matrix, int *myA, const Epetra_Comm &comm)
{
int me = comm.MyPID();
const Epetra_Map &rowmap = matrix.RowMatrixRowMap();
const Epetra_Map &colmap = matrix.RowMatrixColMap();
int myRows = matrix.NumMyRows();
int numRows = matrix.NumGlobalRows();
int numCols = matrix.NumGlobalCols();
int base = rowmap.IndexBase();
int maxRow = matrix.MaxNumEntries();
memset(myA, 0, sizeof(int) * numRows * numCols);
int *myIndices = new int [maxRow];
double *tmp = new double [maxRow];
int rowLen = 0;
for (int i=0; i< myRows; i++){
int rc = matrix.ExtractMyRowCopy(i, maxRow, rowLen, tmp, myIndices);
if (rc){
if (me == 0){
std::cout << "Error in make_my_A" << std::endl;
}
return 1;
}
int *row = myA + (numCols * (rowmap.GID(i) - base));
for (int j=0; j < rowLen; j++){
int colGID = colmap.GID(myIndices[j]);
row[colGID - base] = me + 1;
}
}
if (maxRow){
delete [] myIndices;
delete [] tmp;
}
return 0;
}
int fevec7(Epetra_Comm& Comm, bool verbose)
{
const int NumVectors = 4;
const int NumElements = 4;
Epetra_Map Map(NumElements, 0, Comm);
std::vector<double> mydata(NumElements*NumVectors, 1.0);
Epetra_FEVector x1(View, Map, &mydata[0], NumElements, NumVectors);
x1.PutScalar (0);
// let all processors set global entry 0 to 1
const int GID = 0;
const double value = 1;
x1.ReplaceGlobalValues(1, &GID, &value);
x1.GlobalAssemble (Insert);
if (Comm.MyPID()==0 && x1[0][0] != value) return -1;
return 0;
}
void build_maps(
itype nrows, // Number of global matrix rows
bool testEpetra64,// Flag indicating whether to adjust global row/column
// indices to exercise Epetra64 capability.
Epetra_Comm &comm, // Epetra communicator to be used in maps
Epetra_Map **vectorMap, // OUTPUT: Map to be used for the vector
Epetra_Map **rowMap, // OUTPUT: Map to be used for the matrix rows
Epetra_Map **colMap, // OUTPUT: Map to be used for the matrix cols
long long &offsetEpetra64, // OUTPUT for testing Epetra64: add offsetEpetra64
// to all row/column indices.
bool verbose // print out generated maps
)
{
// Function to build the maps for 1D or 2D matrix distribution.
// Output for 1D includes rowMap and NULL colMap and vectorMap.
// Output for 2D includes rowMap, colMap and vectorMap.
int me = comm.MyPID();
int np = comm.NumProc();
*rowMap = NULL;
*colMap = NULL;
*vectorMap = NULL;
// offsetEpetra64 = (testEpetra64 ? (long long) INT_MAX - (long long) 5 : 0);
offsetEpetra64 = (testEpetra64 ? (long long) 2 * INT_MAX : 0);
// Generate 1D row-based decomposition.
if ((me == 0) && verbose)
cout << endl
<< "1D Distribution: " << endl
<< " np = " << np << endl;
// Linear map similar to Trilinos default.
itype nMyRows = nrows / np + (nrows % np > me);
itype myFirstRow = me * (nrows / np) + MIN(nrows % np, me);
itype *myGlobalRows = new itype[nMyRows];
for (itype i = 0; i < nMyRows; i++)
myGlobalRows[i] = i + myFirstRow + offsetEpetra64;
*rowMap = new Epetra_Map(nrows, nMyRows, &myGlobalRows[0], 0, comm);
delete [] myGlobalRows;
}
int check_graph_sharing(Epetra_Comm& Comm)
{
int numLocalElems = 5;
int localProc = Comm.MyPID();
int firstElem = localProc*numLocalElems;
int err;
Epetra_Map map(-1, numLocalElems, 0, Comm);
Epetra_CrsMatrix* A = new Epetra_CrsMatrix(Copy, map, 1);
for (int i=0; i<numLocalElems; ++i) {
int row = firstElem+i;
int col = row;
double val = 1.0;
err = A->InsertGlobalValues(row, 1, &val, &col);
if (err != 0) {
cerr << "A->InsertGlobalValues("<<row<<") returned err="<<err<<endl;
return(err);
}
}
A->FillComplete(false);
Epetra_CrsMatrix B(Copy, A->Graph());
delete A;
for (int i=0; i<numLocalElems; ++i) {
int row = firstElem+i;
int col = row;
double val = 1.0;
err = B.ReplaceGlobalValues(row, 1, &val, &col);
if (err != 0) {
cerr << "B.InsertGlobalValues("<<row<<") returned err="<<err<<endl;
return(err);
}
}
return(0);
}
int rectangular(const Epetra_Comm& Comm, bool verbose)
{
int mypid = Comm.MyPID();
int numlocalrows = 3;
Epetra_Map rowmap((long long) -1, numlocalrows, 0, Comm);
long long numglobalrows = numlocalrows*Comm.NumProc();
long long numcols = 2*numglobalrows;
Epetra_FECrsGraph fegraph(Copy, rowmap, numcols);
long long* cols = new long long[numcols];
for(int j=0; j<numcols; ++j) cols[j] = j;
Epetra_Map domainmap((long long) -1, numcols, 0, Comm);
long long firstlocalrow = numlocalrows*mypid;
long long lastlocalrow = numlocalrows*(mypid+1)-1;
for(long long i=0; i<numglobalrows; ++i) {
//if i is a local row, then skip it. We want each processor to only
//load rows that belong on other processors.
if (i >= firstlocalrow && i <= lastlocalrow) continue;
EPETRA_CHK_ERR( fegraph.InsertGlobalIndices(1, &i, numcols, &(cols[0])) );
}
EPETRA_CHK_ERR( fegraph.GlobalAssemble(domainmap, rowmap) );
if (verbose) {
std::cout << "********************** fegraph **********************" << std::endl;
std::cout << fegraph << std::endl;
}
delete [] cols;
return(0);
}
int test_bug2890(Epetra_Comm& Comm, bool verbose)
{
//This function tests the AZ_random1() function in AztecOO.
//The implementation of the Park and Miller random number
//generator was incorrect and resulted in an overflow condition.
//This is *not* a complete test of AztecOO's RNG.
//
//A more robust check is to compile AztecOO with gcc -ftrapv and run
//a Krylov method that invokes AZ_random_vector().
int seed = -127773;
double rand_num;
rand_num = AZ_srandom1(&seed);
if (verbose && Comm.MyPID() == 0)
printf("test_bug2890: rand_num = %e (should be in [0,1])\n",rand_num);
if ( (rand_num > 1) || (rand_num < 0) )
return 1; // rand_num should be in [0,1]
else
return 0;
}
void MPIWrapper::allGatherCompact(const Epetra_Comm &Comm, FieldContainer<Scalar> &gatheredValues,
FieldContainer<Scalar> &myValues, FieldContainer<int> &offsets)
{
int mySize = myValues.size();
int totalSize;
Comm.SumAll(&mySize, &totalSize, 1);
int myOffset = 0;
Comm.ScanSum(&mySize,&myOffset,1);
myOffset -= mySize;
gatheredValues.resize(totalSize);
for (int i=0; i<mySize; i++)
{
gatheredValues[myOffset+i] = myValues[i];
}
MPIWrapper::entryWiseSum(Comm, gatheredValues);
offsets.resize(Comm.NumProc());
offsets[Comm.MyPID()] = myOffset;
MPIWrapper::entryWiseSum(Comm, offsets);
}
void Trilinos_Util_distrib_vbr_matrix(const Epetra_Comm & Comm,
int *N_global, int *N_blk_global,
int *n_nonzeros, int *n_blk_nonzeros,
int *N_update, int **update,
double **val, int **indx, int **rpntr, int **cpntr,
int **bpntr, int **bindx,
double **x, double **b, double **xexact)
#undef DEBUG
{
int i, n_global_nonzeros, n_global_blk_nonzeros;
int N_local;
int j, row, have_xexact = 0 ;
int *rpntr1, *bindx1, *bpntr1, *indx1;
double *val1, *b1, *x1, *xexact1=0;
int MyPID = Comm.MyPID();
int NumProc = Comm.NumProc();
printf("Processor %d of %d entering distrib_matrix.\n",
MyPID,NumProc) ;
/*************** Distribute global matrix to all processors ************/
if(MyPID == 0)
{
if ((*xexact) != NULL) have_xexact = 1;
printf("%s", "Broadcasting exact solution\n");
}
if(NumProc > 1)
{
Comm.Broadcast( N_global, 1, 0);
Comm.Broadcast( N_blk_global, 1, 0);
Comm.Broadcast( n_nonzeros, 1, 0);
Comm.Broadcast( n_blk_nonzeros, 1, 0);
Comm.Broadcast( &have_xexact, 1, 0);
printf("Processor %d of %d done with global parameter broadcast.\n",
MyPID,NumProc) ;
if(MyPID != 0)
{
*bpntr = (int *) calloc(*N_blk_global+1,sizeof(int)) ;
*rpntr = (int *) calloc(*N_blk_global+1,sizeof(int)) ;
*bindx = (int *) calloc(*n_blk_nonzeros+1,sizeof(int)) ;
*indx = (int *) calloc(*n_blk_nonzeros+1,sizeof(int)) ;
*val = (double *) calloc(*n_nonzeros+1,sizeof(double)) ;
printf("Processor %d of %d done with global calloc.\n",
MyPID,NumProc) ;
}
Comm.Broadcast( (*bpntr), (*N_blk_global+1), 0);
Comm.Broadcast( (*rpntr), (*N_blk_global+1), 0);
Comm.Broadcast( (*bindx), (*n_blk_nonzeros+1), 0);
Comm.Broadcast( (*indx), (*n_blk_nonzeros+1), 0);
Comm.Broadcast( (*val), (*n_nonzeros+1), 0);
printf("Processor %d of %d done with matrix broadcast.\n",
MyPID,NumProc) ;
/* Set rhs and initialize guess */
if(MyPID != 0)
{
(*b) = (double *) calloc(*N_global,sizeof(double)) ;
(*x) = (double *) calloc(*N_global,sizeof(double)) ;
if (have_xexact)
(*xexact) = (double *) calloc(*N_global,sizeof(double)) ;
}
Comm.Broadcast( (*x), (*N_global), 0);
Comm.Broadcast( (*b), (*N_global), 0);
if (have_xexact)
Comm.Broadcast((*xexact), (*N_global), 0);
printf("Processor %d of %d done with rhs/guess broadcast.\n",
MyPID,NumProc) ;
}
/********************** Generate update map *************************/
//read_update(N_update, update, proc_config, *N_blk_global, 1, linear) ;
Epetra_Map map(*N_blk_global, 0, Comm);
*N_update = map.NumMyElements();
(*update) = (int *) calloc(*N_update,sizeof(int)) ;
map.MyGlobalElements(*update);
printf("Processor %d of %d has %d rows of %d total block rows.\n",
MyPID,NumProc,*N_update,*N_blk_global) ;
/*************** Construct local matrix from global matrix ************/
/* The local matrix is a copy of the rows assigned to this processor.
It is stored in MSR format and still has global indices
*/
if(NumProc > 1)
{
n_global_nonzeros = *n_nonzeros;
n_global_blk_nonzeros = *n_blk_nonzeros;
*n_nonzeros = 0;
*n_blk_nonzeros = 0;
N_local = 0;
for (i=0; i<*N_update; i++)
{
row = (*update)[i];
*n_nonzeros += (*indx)[(*bpntr)[row+1]] - (*indx)[(*bpntr)[row]];
*n_blk_nonzeros += (*bpntr)[row+1] - (*bpntr)[row];
N_local += (*rpntr)[row+1] - (*rpntr)[row];
}
printf("Processor %d of %d has %d nonzeros of %d total nonzeros.\n",
MyPID,NumProc,
*n_nonzeros,n_global_nonzeros) ;
printf("Processor %d of %d has %d block nonzeros of %d total block nonzeros.\n",
MyPID,NumProc,
*n_blk_nonzeros,n_global_blk_nonzeros) ;
printf("Processor %d of %d has %d equations of %d total equations.\n",
MyPID,NumProc,
N_local,*N_global) ;
#ifdef DEBUG
{ double sum1 = 0.0;
for (i=0;i<*N_global; i++) sum1 += (*b)[i];
printf("Processor %d of %d has sum of b = %12.4g.\n",
MyPID,NumProc,sum1) ;
}
#endif /* DEBUG */
/* Allocate memory for local matrix */
bpntr1 = (int *) calloc(*N_update+1,sizeof(int)) ;
rpntr1 = (int *) calloc(*N_update+1,sizeof(int)) ;
bindx1 = (int *) calloc(*n_blk_nonzeros+1,sizeof(int)) ;
indx1 = (int *) calloc(*n_blk_nonzeros+1,sizeof(int)) ;
val1 = (double *) calloc(*n_nonzeros+1,sizeof(double)) ;
b1 = (double *) calloc(N_local,sizeof(double)) ;
x1 = (double *) calloc(N_local,sizeof(double)) ;
if (have_xexact)
xexact1 = (double *) calloc(N_local,sizeof(double)) ;
{
int cur_blk_size, indx_offset, len_val, row_offset, row_offset1;
double *val_ptr, *val1_ptr;
bpntr1[0] = 0;
indx1[0] = 0;
rpntr1[0] = 0;
for (i=0; i<*N_update; i++)
{
row = (*update)[i];
cur_blk_size = (*rpntr)[row+1] - (*rpntr)[row];
rpntr1[i+1] = rpntr1[i] + cur_blk_size;
row_offset = (*rpntr)[row];
row_offset1 = rpntr1[i];
for (j = 0; j<cur_blk_size; j++)
{
b1[row_offset1+j] = (*b)[row_offset+j];
x1[row_offset1+j] = (*x)[row_offset+j];
if (have_xexact) xexact1[row_offset1+j] = (*xexact)[row_offset+j];
}
bpntr1[i+1] = bpntr1[i];
#ifdef DEBUG
printf("Proc %d of %d: Global row = %d: Local row = %d: b = %12.4g: x = %12.4g: bindx = %d: val = %12.4g \n",
MyPID,NumProc,
row, i, b1[i], x1[i], bindx1[i], val1[i]) ;
#endif
indx_offset = (*indx)[(*bpntr)[row]] - indx1[bpntr1[i]];
for (j = (*bpntr)[row]; j < (*bpntr)[row+1]; j++)
{
indx1[bpntr1 [i+1] + 1] = (*indx)[j+1] - indx_offset;
bindx1[bpntr1 [i+1] ] = (*bindx)[j];
bpntr1[i+1] ++;
}
len_val = indx1[bpntr1[i+1]] - indx1[bpntr1[i]];
val_ptr = (*val)+(*indx)[(*bpntr)[row]];
val1_ptr = val1+indx1[bpntr1[i]];
for (j = 0; j<len_val; j++)
{
*val1_ptr = *val_ptr;
val_ptr++; val1_ptr++;
}
}
}
printf("Processor %d of %d done with extracting local operators.\n",
MyPID,NumProc) ;
if (have_xexact)
{
printf(
"The residual using VBR format and exact solution on processor %d is %12.4g\n",
MyPID,
Trilinos_Util_svbrres (N_local, *N_global, *N_update, val1, indx1, bindx1,
rpntr1, (*rpntr), bpntr1, bpntr1+1,
(*xexact), b1));
}
/* Release memory for global matrix, rhs and solution */
free ((void *) (*val));
free ((void *) (*indx));
free ((void *) (*bindx));
free ((void *) (*bpntr));
free ((void *) (*rpntr));
free ((void *) (*b));
free ((void *) (*x));
if (have_xexact) free((void *) *xexact);
/* Return local matrix through same pointers. */
*val = val1;
*indx = indx1;
*bindx = bindx1;
*bpntr = bpntr1;
*rpntr = rpntr1;
*b = b1;
*x = x1;
if (have_xexact) *xexact = xexact1;
}
if (have_xexact && NumProc == 1)
{
printf(
"The residual using VBR format and exact solution on processor %d is %12.4g\n",
MyPID,
Trilinos_Util_svbrres (*N_global, *N_global, *N_update, (*val), (*indx), (*bindx),
(*rpntr), (*rpntr), (*bpntr), (*bpntr)+1,
(*xexact), (*b)));
}
printf("Processor %d of %d leaving distrib_matrix.\n",
MyPID,NumProc) ;
/* end distrib_matrix */
}
//
// Amesos_TestMultiSolver.cpp reads in a matrix in Harwell-Boeing format,
// calls one of the sparse direct solvers, using blocked right hand sides
// and computes the error and residual.
//
// TestSolver ignores the Harwell-Boeing right hand sides, creating
// random right hand sides instead.
//
// Amesos_TestMultiSolver can test either A x = b or A^T x = b.
// This can be a bit confusing because sparse direct solvers
// use compressed column storage - the transpose of Trilinos'
// sparse row storage.
//
// Matrices:
// readA - Serial. As read from the file.
// transposeA - Serial. The transpose of readA.
// serialA - if (transpose) then transposeA else readA
// distributedA - readA distributed to all processes
// passA - if ( distributed ) then distributedA else serialA
//
//
int Amesos_TestMultiSolver( Epetra_Comm &Comm, char *matrix_file, int numsolves,
SparseSolverType SparseSolver, bool transpose,
int special, AMESOS_MatrixType matrix_type ) {
int iam = Comm.MyPID() ;
// int hatever;
// if ( iam == 0 ) std::cin >> hatever ;
Comm.Barrier();
Epetra_Map * readMap;
Epetra_CrsMatrix * readA;
Epetra_Vector * readx;
Epetra_Vector * readb;
Epetra_Vector * readxexact;
std::string FileName = matrix_file ;
int FN_Size = FileName.size() ;
std::string LastFiveBytes = FileName.substr( EPETRA_MAX(0,FN_Size-5), FN_Size );
std::string LastFourBytes = FileName.substr( EPETRA_MAX(0,FN_Size-4), FN_Size );
bool NonContiguousMap = false;
if ( LastFiveBytes == ".triU" ) {
NonContiguousMap = true;
// Call routine to read in unsymmetric Triplet matrix
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra( matrix_file, false, Comm, readMap, readA, readx,
readb, readxexact, NonContiguousMap ) );
} else {
if ( LastFiveBytes == ".triS" ) {
NonContiguousMap = true;
// Call routine to read in symmetric Triplet matrix
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra( matrix_file, true, Comm,
readMap, readA, readx,
readb, readxexact, NonContiguousMap ) );
} else {
if ( LastFourBytes == ".mtx" ) {
EPETRA_CHK_ERR( Trilinos_Util_ReadMatrixMarket2Epetra( matrix_file, Comm, readMap,
readA, readx, readb, readxexact) );
} else {
// Call routine to read in HB problem
Trilinos_Util_ReadHb2Epetra( matrix_file, Comm, readMap, readA, readx,
readb, readxexact) ;
}
}
}
Epetra_CrsMatrix transposeA(Copy, *readMap, 0);
Epetra_CrsMatrix *serialA ;
if ( transpose ) {
assert( CrsMatrixTranspose( readA, &transposeA ) == 0 );
serialA = &transposeA ;
} else {
serialA = readA ;
}
// Create uniform distributed map
Epetra_Map map(readMap->NumGlobalElements(), 0, Comm);
Epetra_Map* map_;
if( NonContiguousMap ) {
//
// map gives us NumMyElements and MyFirstElement;
//
int NumGlobalElements = readMap->NumGlobalElements();
int NumMyElements = map.NumMyElements();
int MyFirstElement = map.MinMyGID();
std::vector<int> MapMap_( NumGlobalElements );
readMap->MyGlobalElements( &MapMap_[0] ) ;
Comm.Broadcast( &MapMap_[0], NumGlobalElements, 0 ) ;
map_ = new Epetra_Map( NumGlobalElements, NumMyElements, &MapMap_[MyFirstElement], 0, Comm);
} else {
map_ = new Epetra_Map( map ) ;
}
// Create Exporter to distribute read-in matrix and vectors
Epetra_Export exporter(*readMap, *map_);
Epetra_CrsMatrix A(Copy, *map_, 0);
Epetra_RowMatrix * passA = 0;
Epetra_MultiVector * passx = 0;
Epetra_MultiVector * passb = 0;
Epetra_MultiVector * passxexact = 0;
Epetra_MultiVector * passresid = 0;
Epetra_MultiVector * passtmp = 0;
Epetra_MultiVector x(*map_,numsolves);
Epetra_MultiVector b(*map_,numsolves);
Epetra_MultiVector xexact(*map_,numsolves);
Epetra_MultiVector resid(*map_,numsolves);
Epetra_MultiVector tmp(*map_,numsolves);
Epetra_MultiVector serialx(*readMap,numsolves);
Epetra_MultiVector serialb(*readMap,numsolves);
Epetra_MultiVector serialxexact(*readMap,numsolves);
Epetra_MultiVector serialresid(*readMap,numsolves);
Epetra_MultiVector serialtmp(*readMap,numsolves);
bool distribute_matrix = ( matrix_type == AMESOS_Distributed ) ;
if ( distribute_matrix ) {
//
// Initialize x, b and xexact to the values read in from the file
//
A.Export(*serialA, exporter, Add);
Comm.Barrier();
assert(A.FillComplete()==0);
Comm.Barrier();
passA = &A;
passx = &x;
passb = &b;
passxexact = &xexact;
passresid = &resid;
passtmp = &tmp;
} else {
passA = serialA;
passx = &serialx;
passb = &serialb;
passxexact = &serialxexact;
passresid = &serialresid;
passtmp = &serialtmp;
}
passxexact->SetSeed(131) ;
passxexact->Random();
passx->SetSeed(11231) ;
passx->Random();
passb->PutScalar( 0.0 );
passA->Multiply( transpose, *passxexact, *passb ) ;
Epetra_MultiVector CopyB( *passb ) ;
double Anorm = passA->NormInf() ;
SparseDirectTimingVars::SS_Result.Set_Anorm(Anorm) ;
Epetra_LinearProblem Problem( (Epetra_RowMatrix *) passA,
(Epetra_MultiVector *) passx,
(Epetra_MultiVector *) passb );
double max_resid = 0.0;
for ( int j = 0 ; j < special+1 ; j++ ) {
Epetra_Time TotalTime( Comm ) ;
if ( false ) {
#ifdef TEST_UMFPACK
unused code
} else if ( SparseSolver == UMFPACK ) {
UmfpackOO umfpack( (Epetra_RowMatrix *) passA,
(Epetra_MultiVector *) passx,
(Epetra_MultiVector *) passb ) ;
umfpack.SetTrans( transpose ) ;
umfpack.Solve() ;
#endif
#ifdef TEST_SUPERLU
} else if ( SparseSolver == SuperLU ) {
SuperluserialOO superluserial( (Epetra_RowMatrix *) passA,
(Epetra_MultiVector *) passx,
(Epetra_MultiVector *) passb ) ;
superluserial.SetPermc( SuperLU_permc ) ;
superluserial.SetTrans( transpose ) ;
superluserial.SetUseDGSSV( special == 0 ) ;
superluserial.Solve() ;
#endif
#ifdef HAVE_AMESOS_SLUD
} else if ( SparseSolver == SuperLUdist ) {
SuperludistOO superludist( Problem ) ;
superludist.SetTrans( transpose ) ;
EPETRA_CHK_ERR( superludist.Solve( true ) ) ;
#endif
#ifdef HAVE_AMESOS_SLUD2
} else if ( SparseSolver == SuperLUdist2 ) {
Superludist2_OO superludist2( Problem ) ;
superludist2.SetTrans( transpose ) ;
EPETRA_CHK_ERR( superludist2.Solve( true ) ) ;
#endif
#ifdef TEST_SPOOLES
} else if ( SparseSolver == SPOOLES ) {
SpoolesOO spooles( (Epetra_RowMatrix *) passA,
(Epetra_MultiVector *) passx,
(Epetra_MultiVector *) passb ) ;
spooles.SetTrans( transpose ) ;
spooles.Solve() ;
#endif
#ifdef HAVE_AMESOS_DSCPACK
} else if ( SparseSolver == DSCPACK ) {
Teuchos::ParameterList ParamList ;
Amesos_Dscpack dscpack( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( dscpack.SetParameters( ParamList ) );
EPETRA_CHK_ERR( dscpack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_UMFPACK
} else if ( SparseSolver == UMFPACK ) {
Teuchos::ParameterList ParamList ;
Amesos_Umfpack umfpack( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( umfpack.SetParameters( ParamList ) );
EPETRA_CHK_ERR( umfpack.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( umfpack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_KLU
} else if ( SparseSolver == KLU ) {
Teuchos::ParameterList ParamList ;
Amesos_Klu klu( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( klu.SetParameters( ParamList ) );
EPETRA_CHK_ERR( klu.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( klu.SymbolicFactorization( ) );
EPETRA_CHK_ERR( klu.NumericFactorization( ) );
EPETRA_CHK_ERR( klu.Solve( ) );
#endif
#ifdef HAVE_AMESOS_PARAKLETE
} else if ( SparseSolver == PARAKLETE ) {
Teuchos::ParameterList ParamList ;
Amesos_Paraklete paraklete( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( paraklete.SetParameters( ParamList ) );
EPETRA_CHK_ERR( paraklete.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( paraklete.SymbolicFactorization( ) );
EPETRA_CHK_ERR( paraklete.NumericFactorization( ) );
EPETRA_CHK_ERR( paraklete.Solve( ) );
#endif
#ifdef HAVE_AMESOS_SLUS
} else if ( SparseSolver == SuperLU ) {
Epetra_SLU superluserial( &Problem ) ;
EPETRA_CHK_ERR( superluserial.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( superluserial.SymbolicFactorization( ) );
EPETRA_CHK_ERR( superluserial.NumericFactorization( ) );
EPETRA_CHK_ERR( superluserial.Solve( ) );
#endif
#ifdef HAVE_AMESOS_LAPACK
} else if ( SparseSolver == LAPACK ) {
Teuchos::ParameterList ParamList ;
ParamList.set( "MaxProcs", -3 );
Amesos_Lapack lapack( Problem ) ;
EPETRA_CHK_ERR( lapack.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( lapack.SymbolicFactorization( ) );
EPETRA_CHK_ERR( lapack.NumericFactorization( ) );
EPETRA_CHK_ERR( lapack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_TAUCS
} else if ( SparseSolver == TAUCS ) {
Teuchos::ParameterList ParamList ;
Amesos_Taucs taucs( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( taucs.SetParameters( ParamList ) );
EPETRA_CHK_ERR( taucs.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( taucs.SymbolicFactorization( ) );
EPETRA_CHK_ERR( taucs.NumericFactorization( ) );
EPETRA_CHK_ERR( taucs.Solve( ) );
#endif
#ifdef HAVE_AMESOS_PARDISO
} else if ( SparseSolver == PARDISO ) {
Teuchos::ParameterList ParamList ;
Amesos_Pardiso pardiso( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( pardiso.SetParameters( ParamList ) );
EPETRA_CHK_ERR( pardiso.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( pardiso.SymbolicFactorization( ) );
EPETRA_CHK_ERR( pardiso.NumericFactorization( ) );
EPETRA_CHK_ERR( pardiso.Solve( ) );
#endif
#ifdef HAVE_AMESOS_PARKLETE
} else if ( SparseSolver == PARKLETE ) {
Teuchos::ParameterList ParamList ;
Amesos_Parklete parklete( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( parklete.SetParameters( ParamList ) );
EPETRA_CHK_ERR( parklete.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( parklete.SymbolicFactorization( ) );
EPETRA_CHK_ERR( parklete.NumericFactorization( ) );
EPETRA_CHK_ERR( parklete.Solve( ) );
#endif
#ifdef HAVE_AMESOS_MUMPS
} else if ( SparseSolver == MUMPS ) {
Teuchos::ParameterList ParamList ;
Amesos_Mumps mumps( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( mumps.SetParameters( ParamList ) );
EPETRA_CHK_ERR( mumps.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( mumps.SymbolicFactorization( ) );
EPETRA_CHK_ERR( mumps.NumericFactorization( ) );
EPETRA_CHK_ERR( mumps.Solve( ) );
#endif
#ifdef HAVE_AMESOS_SCALAPACK
} else if ( SparseSolver == SCALAPACK ) {
Teuchos::ParameterList ParamList ;
Amesos_Scalapack scalapack( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( scalapack.SetParameters( ParamList ) );
EPETRA_CHK_ERR( scalapack.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( scalapack.SymbolicFactorization( ) );
EPETRA_CHK_ERR( scalapack.NumericFactorization( ) );
EPETRA_CHK_ERR( scalapack.Solve( ) );
#endif
#ifdef HAVE_AMESOS_SUPERLUDIST
} else if ( SparseSolver == SUPERLUDIST ) {
Teuchos::ParameterList ParamList ;
Amesos_Superludist superludist( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( superludist.SetParameters( ParamList ) );
EPETRA_CHK_ERR( superludist.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( superludist.SymbolicFactorization( ) );
EPETRA_CHK_ERR( superludist.NumericFactorization( ) );
EPETRA_CHK_ERR( superludist.Solve( ) );
#endif
#ifdef HAVE_AMESOS_SUPERLU
} else if ( SparseSolver == SUPERLU ) {
Teuchos::ParameterList ParamList ;
Amesos_Superlu superlu( Problem ) ;
ParamList.set( "MaxProcs", -3 );
EPETRA_CHK_ERR( superlu.SetParameters( ParamList ) );
EPETRA_CHK_ERR( superlu.SetUseTranspose( transpose ) );
EPETRA_CHK_ERR( superlu.SymbolicFactorization( ) );
EPETRA_CHK_ERR( superlu.NumericFactorization( ) );
EPETRA_CHK_ERR( superlu.Solve( ) );
#endif
#ifdef TEST_SPOOLESSERIAL
} else if ( SparseSolver == SPOOLESSERIAL ) {
SpoolesserialOO spoolesserial( (Epetra_RowMatrix *) passA,
(Epetra_MultiVector *) passx,
(Epetra_MultiVector *) passb ) ;
spoolesserial.Solve() ;
#endif
} else {
SparseDirectTimingVars::log_file << "Solver not implemented yet" << std::endl ;
std::cerr << "\n\n#################### Requested solver not available (Or not tested with blocked RHS) on this platform #####################\n" << std::endl ;
}
SparseDirectTimingVars::SS_Result.Set_Total_Time( TotalTime.ElapsedTime() );
// SparseDirectTimingVars::SS_Result.Set_First_Time( 0.0 );
// SparseDirectTimingVars::SS_Result.Set_Middle_Time( 0.0 );
// SparseDirectTimingVars::SS_Result.Set_Last_Time( 0.0 );
//
// Compute the error = norm(xcomp - xexact )
//
std::vector <double> error(numsolves) ;
double max_error = 0.0;
passresid->Update(1.0, *passx, -1.0, *passxexact, 0.0);
passresid->Norm2(&error[0]);
for ( int i = 0 ; i< numsolves; i++ )
if ( error[i] > max_error ) max_error = error[i] ;
SparseDirectTimingVars::SS_Result.Set_Error(max_error) ;
// passxexact->Norm2(&error[0] ) ;
// passx->Norm2(&error ) ;
//
// Compute the residual = norm(Ax - b)
//
std::vector <double> residual(numsolves) ;
passtmp->PutScalar(0.0);
passA->Multiply( transpose, *passx, *passtmp);
passresid->Update(1.0, *passtmp, -1.0, *passb, 0.0);
// passresid->Update(1.0, *passtmp, -1.0, CopyB, 0.0);
passresid->Norm2(&residual[0]);
for ( int i = 0 ; i< numsolves; i++ )
if ( residual[i] > max_resid ) max_resid = residual[i] ;
SparseDirectTimingVars::SS_Result.Set_Residual(max_resid) ;
std::vector <double> bnorm(numsolves);
passb->Norm2( &bnorm[0] ) ;
SparseDirectTimingVars::SS_Result.Set_Bnorm(bnorm[0]) ;
std::vector <double> xnorm(numsolves);
passx->Norm2( &xnorm[0] ) ;
SparseDirectTimingVars::SS_Result.Set_Xnorm(xnorm[0]) ;
if ( false && iam == 0 ) {
std::cout << " Amesos_TestMutliSolver.cpp " << std::endl ;
for ( int i = 0 ; i< numsolves && i < 10 ; i++ ) {
std::cout << "i=" << i
<< " error = " << error[i]
<< " xnorm = " << xnorm[i]
<< " residual = " << residual[i]
<< " bnorm = " << bnorm[i]
<< std::endl ;
}
std::cout << std::endl << " max_resid = " << max_resid ;
std::cout << " max_error = " << max_error << std::endl ;
std::cout << " Get_residual() again = " << SparseDirectTimingVars::SS_Result.Get_Residual() << std::endl ;
}
}
delete readA;
delete readx;
delete readb;
delete readxexact;
delete readMap;
delete map_;
Comm.Barrier();
return 0 ;
}
int tcubed_test(int NumGlobalElements, bool verbose, Epetra_Comm& Comm,
bool includeUV, bool useP, const std::string& prec,
const std::string& prec_method)
{
int ierr = 0;
int MyPID = Comm.MyPID();
if (MyPID == 0) {
std::cout << "********** "
<< "Testing includeUV = " << includeUV << " useP = " << useP
<< " Preconditioner = " << prec
<< " Preconditioner method = " << prec_method
<< " **********" << std::endl;
}
try {
double nonlinear_factor = 1.0;
double left_bc = 0.0;
double right_bc = 2.07;
// Create the FiniteElementProblem class. This creates all required
// Epetra objects for the problem and allows calls to the
// function (RHS) and Jacobian evaluation routines.
Tcubed_FiniteElementProblem Problem(NumGlobalElements, Comm);
// Get the vector from the Problem
Epetra_Vector& soln = Problem.getSolution();
// Initialize Solution
soln.PutScalar(1.0);
// Create initial guess for the null vector of jacobian
// Only needed for Moore-Spence
Teuchos::RCP<NOX::Abstract::Vector> nullVec =
Teuchos::rcp(new NOX::Epetra::Vector(soln));
nullVec->init(1.0); // initial value 1.0
// Begin LOCA Solver ************************************
// Create parameter list
Teuchos::RCP<Teuchos::ParameterList> paramList =
Teuchos::rcp(new Teuchos::ParameterList);
// Create LOCA sublist
Teuchos::ParameterList& locaParamsList = paramList->sublist("LOCA");
// Create the stepper sublist and set the stepper parameters
Teuchos::ParameterList& locaStepperList =
locaParamsList.sublist("Stepper");
locaStepperList.set("Continuation Method", "Arc Length");
locaStepperList.set("Continuation Parameter", "Nonlinear Factor");
locaStepperList.set("Initial Value", nonlinear_factor);
locaStepperList.set("Max Value", 2.0);
locaStepperList.set("Min Value", 0.05);
locaStepperList.set("Max Steps", 20);
locaStepperList.set("Max Nonlinear Iterations", 15);
locaStepperList.set("Bordered Solver Method", "Nested");
Teuchos::ParameterList& nestedList =
locaStepperList.sublist("Nested Bordered Solver");
nestedList.set("Bordered Solver Method", "Householder");
nestedList.set("Include UV In Preconditioner", includeUV);
nestedList.set("Use P For Preconditioner", useP);
// Create bifurcation sublist
Teuchos::ParameterList& bifurcationList =
locaParamsList.sublist("Bifurcation");
bifurcationList.set("Type", "Turning Point");
bifurcationList.set("Bifurcation Parameter", "Right BC");
bifurcationList.set("Formulation", "Minimally Augmented");
bifurcationList.set("Symmetric Jacobian", false);
bifurcationList.set("Update Null Vectors Every Continuation Step", true);
bifurcationList.set("Update Null Vectors Every Nonlinear Iteration",
false);
bifurcationList.set("Transpose Solver Method",prec);
bifurcationList.set("Multiply Null Vectors by Mass Matrix", true);
// The others don't seem to work with "Solve df/dp"
if (prec == "Explicit Transpose")
bifurcationList.set("Initial Null Vector Computation", "Solve df/dp");
else {
bifurcationList.set("Initial A Vector", nullVec);
bifurcationList.set("Initial B Vector", nullVec);
}
bifurcationList.set("Bordered Solver Method", "Householder");
bifurcationList.set("Include UV In Preconditioner", includeUV);
bifurcationList.set("Use P For Preconditioner", useP);
bifurcationList.set("Preconditioner Method", prec_method);
//bifurcationList.set("Formulation", "Moore-Spence");
//bifurcationList.set("Solver Method", "Phipps Bordering");
//bifurcationList.set("Solver Method", "Salinger Bordering");
//bifurcationList.set("Initial Null Vector", nullVec);
//bifurcationList.set("Length Normalization Vector", nullVec);
// Create predictor sublist
Teuchos::ParameterList& predictorList =
locaParamsList.sublist("Predictor");
predictorList.set("Method", "Secant");
// Create step size sublist
Teuchos::ParameterList& stepSizeList = locaParamsList.sublist("Step Size");
stepSizeList.set("Method", "Adaptive");
stepSizeList.set("Initial Step Size", 0.1);
stepSizeList.set("Min Step Size", 1.0e-3);
stepSizeList.set("Max Step Size", 2000.0);
stepSizeList.set("Aggressiveness", 0.1);
// Create the "Solver" parameters sublist to be used with NOX Solvers
Teuchos::ParameterList& nlParams = paramList->sublist("NOX");
// Create the NOX printing parameter list
Teuchos::ParameterList& nlPrintParams = nlParams.sublist("Printing");
nlPrintParams.set("MyPID", MyPID);
nlPrintParams.set("Output Precision", 4);
if (verbose)
nlPrintParams.set("Output Information",
NOX::Utils::OuterIteration +
NOX::Utils::OuterIterationStatusTest +
NOX::Utils::InnerIteration +
NOX::Utils::Details +
NOX::Utils::LinearSolverDetails +
NOX::Utils::Warning +
NOX::Utils::TestDetails +
NOX::Utils::Error +
NOX::Utils::StepperIteration +
NOX::Utils::StepperDetails +
NOX::Utils::StepperParameters);
else
nlPrintParams.set("Output Information", NOX::Utils::Error);
// Create the "Linear Solver" sublist for Newton's method
Teuchos::ParameterList& dirParams = nlParams.sublist("Direction");
Teuchos::ParameterList& newParams = dirParams.sublist("Newton");
Teuchos::ParameterList& lsParams = newParams.sublist("Linear Solver");
lsParams.set("Aztec Solver", "GMRES");
lsParams.set("Max Iterations", 200);
lsParams.set("Tolerance", 1e-6);
lsParams.set("Output Frequency", 50);
//lsParams.set("Scaling", "None");
//lsParams.set("Scaling", "Row Sum");
lsParams.set("Compute Scaling Manually", false);
//lsParams.set("Preconditioner", "Ifpack");
lsParams.set("Preconditioner", "New Ifpack");
lsParams.set("Ifpack Preconditioner", "ILU");
// Create and initialize the parameter vector
LOCA::ParameterVector pVector;
pVector.addParameter("Nonlinear Factor",nonlinear_factor);
pVector.addParameter("Left BC", left_bc);
pVector.addParameter("Right BC", right_bc);
// Create the interface between the test problem and the nonlinear solver
// This is created by the user using inheritance of the abstract base class
Teuchos::RCP<Problem_Interface> interface =
Teuchos::rcp(new Problem_Interface(Problem));
Teuchos::RCP<LOCA::Epetra::Interface::TimeDependent> iReq = interface;
Teuchos::RCP<NOX::Epetra::Interface::Jacobian> iJac = interface;
// Create the Epetra_RowMatrixfor the Jacobian/Preconditioner
Teuchos::RCP<Epetra_RowMatrix> Amat =
Teuchos::rcp(&Problem.getJacobian(),false);
// Create scaling object
Teuchos::RCP<NOX::Epetra::Scaling> scaling = Teuchos::null;
// scaling = Teuchos::rcp(new NOX::Epetra::Scaling);
// Teuchos::RCP<Epetra_Vector> scalingVector =
// Teuchos::rcp(new Epetra_Vector(soln.Map()));
// //scaling->addRowSumScaling(NOX::Epetra::Scaling::Left, scalingVector);
// scaling->addColSumScaling(NOX::Epetra::Scaling::Right, scalingVector);
// Create transpose scaling object
// Teuchos::RCP<NOX::Epetra::Scaling> trans_scaling = Teuchos::null;
// trans_scaling = Teuchos::rcp(new NOX::Epetra::Scaling);
// Teuchos::RCP<Epetra_Vector> transScalingVector =
// Teuchos::rcp(new Epetra_Vector(soln.Map()));
// trans_scaling->addRowSumScaling(NOX::Epetra::Scaling::Right,
// transScalingVector);
// trans_scaling->addColSumScaling(NOX::Epetra::Scaling::Left,
// transScalingVector);
// //bifurcationList.set("Transpose Scaling", trans_scaling);
// Create the linear systems
Teuchos::RCP<NOX::Epetra::LinearSystemAztecOO> linsys =
Teuchos::rcp(new NOX::Epetra::LinearSystemAztecOO(nlPrintParams,
lsParams,
iReq, iJac, Amat, soln,
scaling));
// Create the loca vector
NOX::Epetra::Vector locaSoln(soln);
// Create Epetra factory
Teuchos::RCP<LOCA::Abstract::Factory> epetraFactory =
Teuchos::rcp(new LOCA::Epetra::Factory);
// Create global data object
Teuchos::RCP<LOCA::GlobalData> globalData =
LOCA::createGlobalData(paramList, epetraFactory);
// Create the Group
Teuchos::RCP<LOCA::Epetra::Group> grp =
Teuchos::rcp(new LOCA::Epetra::Group(globalData, nlPrintParams, iReq,
locaSoln, linsys, linsys,
pVector));
grp->computeF();
// Create the Solver convergence test
//NOX::StatusTest::NormWRMS wrms(1.0e-2, 1.0e-8);
Teuchos::RCP<NOX::StatusTest::NormF> wrms =
Teuchos::rcp(new NOX::StatusTest::NormF(1.0e-12));
Teuchos::RCP<NOX::StatusTest::MaxIters> maxiters =
Teuchos::rcp(new NOX::StatusTest::MaxIters(locaStepperList.get("Max Nonlinear Iterations", 10)));
Teuchos::RCP<NOX::StatusTest::Combo> combo =
Teuchos::rcp(new NOX::StatusTest::Combo(NOX::StatusTest::Combo::OR));
combo->addStatusTest(wrms);
combo->addStatusTest(maxiters);
// Create the stepper
LOCA::Stepper stepper(globalData, grp, combo, paramList);
LOCA::Abstract::Iterator::IteratorStatus status = stepper.run();
if (status != LOCA::Abstract::Iterator::Finished) {
ierr = 1;
if (globalData->locaUtils->isPrintType(NOX::Utils::Error))
globalData->locaUtils->out()
<< "Stepper failed to converge!" << std::endl;
}
// Get the final solution from the stepper
Teuchos::RCP<const LOCA::Epetra::Group> finalGroup =
Teuchos::rcp_dynamic_cast<const LOCA::Epetra::Group>(stepper.getSolutionGroup());
const NOX::Epetra::Vector& finalSolution =
dynamic_cast<const NOX::Epetra::Vector&>(finalGroup->getX());
// Output the parameter list
if (globalData->locaUtils->isPrintType(NOX::Utils::StepperParameters)) {
globalData->locaUtils->out()
<< std::endl << "Final Parameters" << std::endl
<< "****************" << std::endl;
stepper.getList()->print(globalData->locaUtils->out());
globalData->locaUtils->out() << std::endl;
}
// Check some statistics on the solution
NOX::TestCompare testCompare(globalData->locaUtils->out(),
*(globalData->locaUtils));
if (globalData->locaUtils->isPrintType(NOX::Utils::TestDetails))
globalData->locaUtils->out()
<< std::endl
<< "***** Checking solution statistics *****"
<< std::endl;
// Check number of steps
int numSteps = stepper.getStepNumber();
int numSteps_expected = 7;
ierr += testCompare.testValue(numSteps, numSteps_expected, 0.0,
"number of continuation steps",
NOX::TestCompare::Absolute);
// Check number of failed steps
int numFailedSteps = stepper.getNumFailedSteps();
int numFailedSteps_expected = 0;
ierr += testCompare.testValue(numFailedSteps, numFailedSteps_expected, 0.0,
"number of failed continuation steps",
NOX::TestCompare::Absolute);
// Check final value of continuation parameter
double factor_final = finalGroup->getParam("Nonlinear Factor");
double factor_expected = 2.0;
ierr += testCompare.testValue(factor_final, factor_expected, 1.0e-14,
"final value of continuation parameter",
NOX::TestCompare::Relative);
// Check final value of bifurcation parameter
double right_bc_final = finalGroup->getParam("Right BC");
double right_bc_expected = 1.47241293;
ierr += testCompare.testValue(right_bc_final, right_bc_expected, 1.0e-7,
"final value of bifurcation parameter",
NOX::TestCompare::Relative);
// Check norm of solution
double norm_x = finalSolution.norm();
double norm_x_expected = 12.038464;
ierr += testCompare.testValue(norm_x, norm_x_expected, 1.0e-7,
"norm of final solution",
NOX::TestCompare::Relative);
LOCA::destroyGlobalData(globalData);
}
catch (std::exception& e) {
std::cout << e.what() << std::endl;
ierr = 1;
}
catch (const char *s) {
std::cout << s << std::endl;
ierr = 1;
}
catch (...) {
std::cout << "Caught unknown exception!" << std::endl;
ierr = 1;
}
return ierr ;
}
void Trilinos_Util_ReadHpc2Epetra(char *data_file,
const Epetra_Comm &comm,
Epetra_Map *& map,
Epetra_CrsMatrix *& A,
Epetra_Vector *& x,
Epetra_Vector *& b,
Epetra_Vector *&xexact) {
FILE *in_file ;
int l;
int * lp = &l;
double v;
double * vp = &v;
#ifdef DEBUG
bool debug = true;
#else
bool debug = false;
#endif
int size = comm.NumProc();
int rank = comm.MyPID();
printf("Reading matrix info from %s...\n",data_file);
in_file = fopen( data_file, "r");
if (in_file == NULL)
{
printf("Error: Cannot open file: %s\n",data_file);
exit(1);
}
int numGlobalEquations, total_nnz;
fscanf(in_file,"%d",&numGlobalEquations);
fscanf(in_file,"%d",&total_nnz);
map = new Epetra_Map(numGlobalEquations, 0, comm); // Create map with uniform distribution
A = new Epetra_CrsMatrix(Copy, *map, 0); // Construct matrix
x = new Epetra_Vector(*map);
b = new Epetra_Vector(*map);
xexact = new Epetra_Vector(*map);
int numMyEquations = map->NumMyPoints();
// Allocate arrays that are of length numMyEquations
// Find max nnz per row for this processor
int max_nnz = 0;
for (int i=0; i<numGlobalEquations; i++) {
fscanf(in_file, "%d",lp); /* row #, nnz in row */
if (map->MyGID(i)) max_nnz = EPETRA_MAX(max_nnz,l);
}
// Allocate arrays that are of length local_nnz
double * list_of_vals = new double[max_nnz];
int *list_of_inds = new int [max_nnz];
{for (int i=0; i<numGlobalEquations; i++)
{
int cur_nnz;
fscanf(in_file, "%d",&cur_nnz);
if (map->MyGID(i)) // See if nnz for row should be added
{
if (debug) cout << "Process "<<rank
<<" of "<<size<<" getting row "<<i<<endl;
int nnz_kept = 0;
for (int j=0; j<cur_nnz; j++)
{
fscanf(in_file, "%lf %d",vp,lp);
if (v!=0.0) {
list_of_vals[nnz_kept] = v;
list_of_inds[nnz_kept] = l;
nnz_kept++;
}
}
A->InsertGlobalValues(i, nnz_kept, list_of_vals, list_of_inds);
}
else
for (int j=0; j<cur_nnz; j++) fscanf(in_file, "%lf %d",vp,lp); // otherwise read and discard
}}
double xt, bt, xxt;
{for (int i=0; i<numGlobalEquations; i++)
{
if (map->MyGID(i)) // See if entry should be added
{
if (debug) cout << "Process "<<rank<<" of "
<<size<<" getting RHS "<<i<<endl;
fscanf(in_file, "%lf %lf %lf",&xt, &bt, &xxt);
int cur_local_row = map->LID(i);
(*x)[cur_local_row] = xt;
(*b)[cur_local_row] = bt;
(*xexact)[cur_local_row] = xxt;
}
else
fscanf(in_file, "%lf %lf %lf",vp, vp, vp); // or thrown away
}}
fclose(in_file);
if (debug)
cout << "Process "<<rank<<" of "<<size<<" has "<<numMyEquations
<< " rows. Min global row "<< map->MinMyGID()
<<" Max global row "<< map->MaxMyGID() <<endl
<<" and "<<A->NumMyNonzeros()<<" nonzeros."<<endl;
A->FillComplete();
Epetra_Vector bcomp(*map);
A->Multiply(false, *xexact, bcomp);
double residual;
bcomp.Norm2(&residual);
if (comm.MyPID()==0) cout << "Norm of computed b = " << residual << endl;
b->Norm2(&residual);
if (comm.MyPID()==0) cout << "Norm of given b = " << residual << endl;
bcomp.Update(-1.0, *b, 1.0);
bcomp.Norm2(&residual);
if (comm.MyPID()==0) cout << "Norm of difference between computed b and given b for xexact = " << residual << endl;
delete [] list_of_vals;
delete []list_of_inds;
return;
}
void Trilinos_Util_ReadHb2EpetraVbr(char *data_file, char * partitioning,
const Epetra_Comm &comm,
Epetra_BlockMap *& map,
Epetra_VbrMatrix *& A,
Epetra_Vector *& x,
Epetra_Vector *& b,
Epetra_Vector *&xexact) {
/* Read matrix file and distribute among processors.
Returns with this processor's set of rows */
int NumGlobalEquations = 0, NumMyNonzeros = 0;
double *val_msr = 0, *x_in = 0, *b_in = 0, *xexact_in = 0;
int *bindx_msr = 0;
/* Set exact solution to NULL */
xexact = NULL;
Trilinos_Util_read_hb(data_file, comm.MyPID(), &NumGlobalEquations, &NumMyNonzeros,
&val_msr, &bindx_msr, &x_in, &b_in, &xexact_in);
double *val = 0;
int NumGlobalElements = 0;
int *indx = 0, *rpntr = 0, *cpntr = 0, *bpntr = 0, *bindx = 0;
int NumMyBlockEntries = 0, NumMyElements = 0, * MyGlobalElements = 0;
Trilinos_Util_create_vbr(comm, partitioning,
&NumGlobalEquations, &NumGlobalElements,
&NumMyNonzeros, &NumMyBlockEntries,
&NumMyElements, &MyGlobalElements,
bindx_msr, val_msr,
&val, &indx, &rpntr, &cpntr,
&bpntr, &bindx);
if(comm.MyPID()==0)
{
free ((void *) val_msr);
free ((void *) bindx_msr);
free ((void *) cpntr);
}
int * ElementSizeList = 0;
if (NumMyElements>0) ElementSizeList = new int[NumMyElements];
for (int i=0; i<NumMyElements; i++) ElementSizeList[i] = rpntr[i+1] - rpntr[i];
map = new Epetra_BlockMap(-1, NumMyElements, MyGlobalElements,
ElementSizeList, 0, comm);
A = new Epetra_VbrMatrix(Copy, *map, 0);
/* Add block rows one-at-a-time */
{for (int i=0; i<NumMyElements; i++) {
int BlockRow = MyGlobalElements[i];
int NumBlockEntries = bpntr[i+1] - bpntr[i];
int *BlockIndices = bindx + bpntr[i];
int ierr = A->BeginInsertGlobalValues(BlockRow, NumBlockEntries, BlockIndices);
if (ierr!=0) {
cerr << "Error in BeginInsertGlobalValues(GlobalBlockRow = " << BlockRow
<< ") = " << ierr << endl;
abort();
}
int LDA = ElementSizeList[i];
int NumRows = LDA;
for (int j=bpntr[i]; j<bpntr[i+1]; j++) {
int NumCols = (indx[j+1] - indx[j])/LDA;
double * Values = val + indx[j];
ierr = A->SubmitBlockEntry(Values, LDA, NumRows, NumCols);
if (ierr!=0) {
cerr << "Error in SubmitBlockEntry, GlobalBlockRow = " << BlockRow
<< "GlobalBlockCol = " << BlockIndices[j] << "Error = " << ierr << endl;
abort();
}
}
ierr = A->EndSubmitEntries();
if (ierr!=0) {
cerr << "Error in EndSubmitEntries(GlobalBlockRow = " << BlockRow
<< ") = " << ierr << endl;
abort();
}
}}
int ierr=A->FillComplete();
if (ierr!=0) cerr << "Error in Epetra_VbrMatrix FillComplete ierr = " << ierr << endl;
xexact = new Epetra_Vector(Copy, *map, xexact_in);
x = new Epetra_Vector(Copy, *map, x_in);
b = new Epetra_Vector(Copy, *map, b_in);
if(comm.MyPID()==0)
{
free ((void *) val);
free ((void *) indx);
free ((void *) rpntr);
free ((void *) bpntr);
free ((void *) bindx);
free ((void *) b_in);
free ((void *) x_in);
free ((void *) xexact_in);
free ((void *) MyGlobalElements);
delete [] ElementSizeList;
}
return;
}
int TestOneMatrix( std::string HBname, std::string MMname, std::string TRIname, Epetra_Comm &Comm, bool verbose ) {
if ( Comm.MyPID() != 0 ) verbose = false ;
Epetra_Map * readMap = 0;
Epetra_CrsMatrix * HbA = 0;
Epetra_Vector * Hbx = 0;
Epetra_Vector * Hbb = 0;
Epetra_Vector * Hbxexact = 0;
Epetra_CrsMatrix * TriplesA = 0;
Epetra_Vector * Triplesx = 0;
Epetra_Vector * Triplesb = 0;
Epetra_Vector * Triplesxexact = 0;
Epetra_CrsMatrix * MatrixMarketA = 0;
Epetra_Vector * MatrixMarketx = 0;
Epetra_Vector * MatrixMarketb = 0;
Epetra_Vector * MatrixMarketxexact = 0;
int TRI_Size = TRIname.size() ;
std::string LastFiveBytes = TRIname.substr( EPETRA_MAX(0,TRI_Size-5), TRI_Size );
if ( LastFiveBytes == ".TimD" ) {
// Call routine to read in a file with a Tim Davis header and zero-based indexing
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra64( &TRIname[0], false, Comm,
readMap, TriplesA, Triplesx,
Triplesb, Triplesxexact, false, true, true ) );
delete readMap;
} else {
if ( LastFiveBytes == ".triU" ) {
// Call routine to read in unsymmetric Triplet matrix
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra64( &TRIname[0], false, Comm,
readMap, TriplesA, Triplesx,
Triplesb, Triplesxexact, false, false ) );
delete readMap;
} else {
if ( LastFiveBytes == ".triS" ) {
// Call routine to read in symmetric Triplet matrix
EPETRA_CHK_ERR( Trilinos_Util_ReadTriples2Epetra64( &TRIname[0], true, Comm,
readMap, TriplesA, Triplesx,
Triplesb, Triplesxexact, false, false ) );
delete readMap;
} else {
assert( false ) ;
}
}
}
EPETRA_CHK_ERR( Trilinos_Util_ReadMatrixMarket2Epetra64( &MMname[0], Comm, readMap,
MatrixMarketA, MatrixMarketx,
MatrixMarketb, MatrixMarketxexact) );
delete readMap;
// Call routine to read in HB problem
Trilinos_Util_ReadHb2Epetra64( &HBname[0], Comm, readMap, HbA, Hbx,
Hbb, Hbxexact) ;
#if 0
std::cout << " HbA " ;
HbA->Print( std::cout ) ;
std::cout << std::endl ;
std::cout << " MatrixMarketA " ;
MatrixMarketA->Print( std::cout ) ;
std::cout << std::endl ;
std::cout << " TriplesA " ;
TriplesA->Print( std::cout ) ;
std::cout << std::endl ;
#endif
int TripleErr = 0 ;
int MMerr = 0 ;
for ( int i = 0 ; i < 10 ; i++ )
{
double resid_Hb_Triples;
double resid_Hb_Matrix_Market;
double norm_A ;
Hbx->Random();
//
// Set the output vectors to different values:
//
Triplesb->PutScalar(1.1);
Hbb->PutScalar(1.2);
MatrixMarketb->PutScalar(1.3);
HbA->Multiply( false, *Hbx, *Hbb );
norm_A = HbA->NormOne( ) ;
TriplesA->Multiply( false, *Hbx, *Triplesb );
Triplesb->Update( 1.0, *Hbb, -1.0 ) ;
MatrixMarketA->Multiply( false, *Hbx, *MatrixMarketb );
MatrixMarketb->Update( 1.0, *Hbb, -1.0 ) ;
Triplesb->Norm1( &resid_Hb_Triples ) ;
MatrixMarketb->Norm1( &resid_Hb_Matrix_Market ) ;
TripleErr += ( resid_Hb_Triples > 1e-11 * norm_A ) ;
MMerr += ( resid_Hb_Matrix_Market > 1e-11 * norm_A ) ;
if ( verbose && resid_Hb_Triples > 1e-11 * norm_A )
std::cout << " resid_Hb_Triples = " << resid_Hb_Triples
<< " norm_A = " << norm_A << std::endl ;
if ( verbose && resid_Hb_Matrix_Market > 1e-11 * norm_A )
std::cout << " resid_Hb_Matrix_Market = " << resid_Hb_Matrix_Market
<< " norm_A = " << norm_A << std::endl ;
}
if ( verbose ) {
if ( TripleErr ) std::cout << " Error in reading " << HBname << " or " << TRIname << std::endl ;
if ( MMerr ) std::cout << " Error in reading " << HBname << " or " << MMname << std::endl ;
}
delete HbA;
delete Hbx;
delete Hbb;
delete Hbxexact;
delete TriplesA;
delete Triplesx;
delete Triplesb;
delete Triplesxexact;
delete MatrixMarketA;
delete MatrixMarketx;
delete MatrixMarketb;
delete MatrixMarketxexact;
delete readMap;
return TripleErr+MMerr ;
}