extern "C" __declspec(dllexport) void CPUsolveCSRDouble(int *row_offset, int *col, double *val,
const int nnz, const int N, double *_rhs, double *_x) {
std::string name = "s";
LocalVector<double> x;
LocalVector<double> rhs;
LocalMatrix<double> mat;
x.Allocate(name, N);
x.Zeros();
rhs.Allocate(name, N);
mat.AllocateCSR(name, nnz, N, N);
mat.CopyFromCSR(row_offset, col, val);
rhs.CopyFromData(_rhs);
CG<LocalMatrix<double>, LocalVector<double>, double> ls;
Jacobi<LocalMatrix<double>, LocalVector<double>, double> p;
ls.SetOperator(mat);
ls.SetPreconditioner(p);
ls.Build();
ls.Solve(rhs, &x);
x.CopyToData(_x);
mat.Clear();
x.Clear();
rhs.Clear();
ls.Clear();
}
TEST_F(beamFEATest, TransformsLocalToGlobalCoords) {
// test that identity matrix is recovered if local axes == global axes
Eigen::Vector3d nx(1.0, 0.0, 0.0);
Eigen::Vector3d nz(0.0, 0.0, 1.0);
assembleK3D.calcAelem(nx, nz);
LocalMatrix Aelem = assembleK3D.getAelem();
LocalMatrix expected;
expected.setIdentity();
for (size_t i = 0; i < Aelem.rows(); ++i) {
EXPECT_DOUBLE_EQ(expected(i), Aelem(i));
}
}
LocalMatrix chemReductionGIA::orthcomp( LocalMatrix & inMat )
{
// initialize it so that they have the same dimension
LocalMatrix outMat;
Eigen::FullPivLU<LocalMatrix> lu_decomp(inMat.transpose());
outMat = lu_decomp.kernel();
// notice that if the kernel returns a matrix with dimension zero,
// then the returned matrix will be a column-vector filled with zeros
// therefore we do a safety check here, and set the number of columns to zero
if ( outMat.cols() == 1 && outMat.col(0).norm() == 0.0 )
outMat = LocalMatrix::Zero(inMat.rows(), 0);
return outMat;
}
extern "C" __declspec(dllexport) void solveCSRSingle(int *row_offset, int *col, float *val,
const int nnz, const int N, float *_rhs, float *_x) {
std::string name = "s";
LocalVector<float> x;
LocalVector<float> rhs;
LocalMatrix<float> mat;
x.Allocate(name, N);
x.Zeros();
rhs.Allocate(name, N);
mat.AllocateCSR(name, nnz, N, N);
mat.CopyFromCSR(row_offset, col, val);
rhs.CopyFromData(_rhs);
// mat.Check();
/* rhs.SetDataPtr(&_rhs, name, N);
x.SetDataPtr(&_x, name, N);
mat.SetDataPtrCSR(&row_offset, &col, &val, name, nnz, N, N);
*/
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
CG<LocalMatrix<float>, LocalVector<float>, float> ls;
MultiColoredILU<LocalMatrix<float>, LocalVector<float>, float> p;
ls.SetOperator(mat);
ls.SetPreconditioner(p);
ls.Build();
ls.Solve(rhs, &x);
mat.MoveToHost();
x.MoveToHost();
rhs.MoveToHost();
/*
mat.LeaveDataPtrCSR(&row_offset, &col, &val);
rhs.LeaveDataPtr(&_rhs);
x.LeaveDataPtr(&_x);
*/
x.CopyToData(_x);
mat.Clear();
x.Clear();
rhs.Clear();
ls.Clear();
}
int main(int argc, char* argv[]) {
if (argc == 1) {
std::cerr << argv[0] << " <matrix> [Num threads]" << std::endl;
exit(1);
}
init_paralution();
if (argc > 2) {
set_omp_threads_paralution(atoi(argv[2]));
}
info_paralution();
LocalVector<double> x;
LocalVector<double> rhs;
LocalMatrix<double> mat;
mat.ReadFileMTX(std::string(argv[1]));
// Compute and apply (R)CMK ordering
LocalVector<int> cmk;
// mat.CMK(&cmk);
mat.RCMK(&cmk);
mat.Permute(cmk);
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
x.Allocate("x", mat.get_nrow());
rhs.Allocate("rhs", mat.get_nrow());
// Linear Solver
CG<LocalMatrix<double>, LocalVector<double>, double > ls;
// Preconditioner
ILU<LocalMatrix<double>, LocalVector<double>, double > p;
double tick, tack;
rhs.Ones();
x.Zeros();
ls.SetOperator(mat);
ls.SetPreconditioner(p);
ls.Build();
mat.info();
tick = paralution_time();
ls.Solve(rhs, &x);
tack = paralution_time();
std::cout << "Solver execution:" << (tack-tick)/1000000 << " sec" << std::endl;
// Revert CMK ordering on solution vector
x.PermuteBackward(cmk);
stop_paralution();
return 0;
}
void update_rotation(const std::vector<Node> &nodes, const BeamElement& elem, LocalMatrix &rotation, LocalMatrix &rotation_transposed) {
rotation.setZero();
rotation_transposed.setZero();
const int nn1 = elem.get_node_numbers()[0];
const int nn2 = elem.get_node_numbers()[1];
// calculate unit normal vector along local x-direction
Eigen::Vector3d nx = nodes[nn2] - nodes[nn1];
nx.normalize();
// calculate unit normal vector along y-direction
const Eigen::Vector3d ny = elem.get_props().normal_vec.normalized();
// calculate the unit normal vector in local z direction
Eigen::Vector3d nz = nx.cross(ny);
nz.normalize();
// update rotation matrix
rotation(0, 0) = nx(0);
rotation(0, 1) = nx(1);
rotation(0, 2) = nx(2);
rotation(1, 0) = ny(0);
rotation(1, 1) = ny(1);
rotation(1, 2) = ny(2);
rotation(2, 0) = nz(0);
rotation(2, 1) = nz(1);
rotation(2, 2) = nz(2);
rotation(3, 3) = nx(0);
rotation(3, 4) = nx(1);
rotation(3, 5) = nx(2);
rotation(4, 3) = ny(0);
rotation(4, 4) = ny(1);
rotation(4, 5) = ny(2);
rotation(5, 3) = nz(0);
rotation(5, 4) = nz(1);
rotation(5, 5) = nz(2);
rotation(6, 6) = nx(0);
rotation(6, 7) = nx(1);
rotation(6, 8) = nx(2);
rotation(7, 6) = ny(0);
rotation(7, 7) = ny(1);
rotation(7, 8) = ny(2);
rotation(8, 6) = nz(0);
rotation(8, 7) = nz(1);
rotation(8, 8) = nz(2);
rotation(9, 9) = nx(0);
rotation(9, 10) = nx(1);
rotation(9, 11) = nx(2);
rotation(10, 9) = ny(0);
rotation(10, 10) = ny(1);
rotation(10, 11) = ny(2);
rotation(11, 9) = nz(0);
rotation(11, 10) = nz(1);
rotation(11, 11) = nz(2);
// update transposed rotation matrix
rotation_transposed(0, 0) = nx(0);
rotation_transposed(0, 1) = ny(0);
rotation_transposed(0, 2) = nz(0);
rotation_transposed(1, 0) = nx(1);
rotation_transposed(1, 1) = ny(1);
rotation_transposed(1, 2) = nz(1);
rotation_transposed(2, 0) = nx(2);
rotation_transposed(2, 1) = ny(2);
rotation_transposed(2, 2) = nz(2);
rotation_transposed(3, 3) = nx(0);
rotation_transposed(3, 4) = ny(0);
rotation_transposed(3, 5) = nz(0);
rotation_transposed(4, 3) = nx(1);
rotation_transposed(4, 4) = ny(1);
rotation_transposed(4, 5) = nz(1);
rotation_transposed(5, 3) = nx(2);
rotation_transposed(5, 4) = ny(2);
rotation_transposed(5, 5) = nz(2);
rotation_transposed(6, 6) = nx(0);
rotation_transposed(6, 7) = ny(0);
rotation_transposed(6, 8) = nz(0);
rotation_transposed(7, 6) = nx(1);
rotation_transposed(7, 7) = ny(1);
rotation_transposed(7, 8) = nz(1);
rotation_transposed(8, 6) = nx(2);
rotation_transposed(8, 7) = ny(2);
rotation_transposed(8, 8) = nz(2);
rotation_transposed(9, 9) = nx(0);
rotation_transposed(9, 10) = ny(0);
rotation_transposed(9, 11) = nz(0);
rotation_transposed(10, 9) = nx(1);
rotation_transposed(10, 10) = ny(1);
rotation_transposed(10, 11) = nz(1);
rotation_transposed(11, 9) = nx(2);
rotation_transposed(11, 10) = ny(2);
rotation_transposed(11, 11) = nz(2);
}
int main(int argc, char* argv[]) {
if (argc == 1) {
std::cerr << argv[0] << " <matrix> [Num threads]" << std::endl;
exit(1);
}
init_paralution();
if (argc > 2) {
set_omp_threads_paralution(atoi(argv[2]));
}
info_paralution();
// int ii;
LocalVector<double> x;
LocalVector<double> rhs;
LocalMatrix<double> mat;
struct timeval ti1,ti2;//timer
mat.ReadFileMTX(std::string(argv[1]));
mat.info();
x.Allocate("x", mat.get_nrow());
rhs.Allocate("rhs", mat.get_nrow());
x.info();
rhs.info();
rhs.Ones();
gettimeofday(&ti1,NULL); /* read starttime in t1 */
mat.Apply(rhs, &x);
gettimeofday(&ti2,NULL); /* read endtime in t2 */
fflush(stderr);
fprintf(stderr, "\nTime cost host spmv code microseconds: %ld microseconds\n",
((ti2.tv_sec - ti1.tv_sec)*1000000L
+ti2.tv_usec) - ti1.tv_usec
);
std::cout << "\ndot=" << x.Dot(rhs) << std::endl;
mat.ConvertToBCSR();
mat.info();
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
mat.info();
rhs.Ones();
// exit(1);
gettimeofday(&ti1,NULL); /* read starttime in t1 */
mat.Apply(rhs, &x);
gettimeofday(&ti2,NULL); /* read endtime in t2 */
fflush(stderr);
fprintf(stderr, "\nTime cost for accelerator spmv microseconds: %ld microseconds\n",
((ti2.tv_sec - ti1.tv_sec)*1000000L
+ti2.tv_usec) - ti1.tv_usec
);
std::cout << "\ndot=" << x.Dot(rhs) << std::endl;
stop_paralution();
return 0;
}
int main(int argc, char* argv[]) {
if (argc == 1) {
std::cerr << argv[0] << " <matrix> [Num threads]" << std::endl;
exit(1);
}
init_paralution();
if (argc > 2) {
set_omp_threads_paralution(atoi(argv[2]));
}
info_paralution();
LocalVector<double> b, b_old, *b_k, *b_k1, *b_tmp;
LocalMatrix<double> mat;
mat.ReadFileMTX(std::string(argv[1]));
// Gershgorin spectrum approximation
double glambda_min, glambda_max;
// Power method spectrum approximation
double plambda_min, plambda_max;
// Maximum number of iteration for the power method
int iter_max = 10000;
double tick, tack;
// Gershgorin approximation of the eigenvalues
mat.Gershgorin(glambda_min, glambda_max);
std::cout << "Gershgorin : Lambda min = " << glambda_min
<< "; Lambda max = " << glambda_max << std::endl;
mat.MoveToAccelerator();
b.MoveToAccelerator();
b_old.MoveToAccelerator();
b.Allocate("b_k+1", mat.get_nrow());
b_k1 = &b;
b_old.Allocate("b_k", mat.get_nrow());
b_k = &b_old;
b_k->Ones();
mat.info();
tick = paralution_time();
// compute lambda max
for (int i=0; i<=iter_max; ++i) {
mat.Apply(*b_k, b_k1);
// std::cout << b_k1->Dot(*b_k) << std::endl;
b_k1->Scale(double(1.0)/b_k1->Norm());
b_tmp = b_k1;
b_k1 = b_k;
b_k = b_tmp;
}
// get lambda max (Rayleigh quotient)
mat.Apply(*b_k, b_k1);
plambda_max = b_k1->Dot(*b_k) ;
tack = paralution_time();
std::cout << "Power method (lambda max) execution:" << (tack-tick)/1000000 << " sec" << std::endl;
mat.AddScalarDiagonal(double(-1.0)*plambda_max);
b_k->Ones();
tick = paralution_time();
// compute lambda min
for (int i=0; i<=iter_max; ++i) {
mat.Apply(*b_k, b_k1);
// std::cout << b_k1->Dot(*b_k) + plambda_max << std::endl;
b_k1->Scale(double(1.0)/b_k1->Norm());
b_tmp = b_k1;
b_k1 = b_k;
b_k = b_tmp;
}
// get lambda min (Rayleigh quotient)
mat.Apply(*b_k, b_k1);
plambda_min = (b_k1->Dot(*b_k) + plambda_max);
// back to the original matrix
mat.AddScalarDiagonal(plambda_max);
tack = paralution_time();
std::cout << "Power method (lambda min) execution:" << (tack-tick)/1000000 << " sec" << std::endl;
std::cout << "Power method Lambda min = " << plambda_min
<< "; Lambda max = " << plambda_max
<< "; iter=2x" << iter_max << std::endl;
LocalVector<double> x;
LocalVector<double> rhs;
x.CloneBackend(mat);
rhs.CloneBackend(mat);
x.Allocate("x", mat.get_nrow());
rhs.Allocate("rhs", mat.get_nrow());
// Chebyshev iteration
Chebyshev<LocalMatrix<double>, LocalVector<double>, double > ls;
rhs.Ones();
x.Zeros();
ls.SetOperator(mat);
ls.Set(plambda_min, plambda_max);
ls.Build();
tick = paralution_time();
ls.Solve(rhs, &x);
tack = paralution_time();
std::cout << "Solver execution:" << (tack-tick)/1000000 << " sec" << std::endl;
// PCG + Chebyshev polynomial
CG<LocalMatrix<double>, LocalVector<double>, double > cg;
AIChebyshev<LocalMatrix<double>, LocalVector<double>, double > p;
// damping factor
plambda_min = plambda_max / 7;
p.Set(3, plambda_min, plambda_max);
rhs.Ones();
x.Zeros();
cg.SetOperator(mat);
cg.SetPreconditioner(p);
cg.Build();
tick = paralution_time();
cg.Solve(rhs, &x);
tack = paralution_time();
std::cout << "Solver execution:" << (tack-tick)/1000000 << " sec" << std::endl;
stop_paralution();
return 0;
}
int main(int argc, char* argv[]) {
if (argc == 1) {
std::cerr << argv[0] << " <matrix> <initial_guess> <rhs> [Num threads]" << std::endl;
exit(1);
}
init_paralution();
// if (argc > 4) {
// set_omp_threads_paralution(atoi(argv[]));
// }
set_omp_threads_paralution(8);
info_paralution();
struct timeval now;
double tick, tack, b,s, sol_norm, diff_norm, ones_norm;
double *phi_ptr=NULL;
int *bubmap_ptr=NULL, phisize, maxbmap, setlssd, lvst_offst;
int xdim, ydim, zdim, defvex_perdirec;
#ifdef BUBFLO
xdim=atoi(argv[5]);
setlssd=atoi(argv[6]);
defvex_perdirec=atoi(argv[7]);
lvst_offst=atoi(argv[8]);
phisize=(xdim+2*lvst_offst)*(ydim+2*lvst_offst)*(zdim+2*lvst_offst);
#endif
LocalVector<double> x;
LocalVector<double> rhs;
LocalMatrix<double> mat;
LocalVector<double> Dinvhalf_min;
LocalVector<double> Dinvhalf_plus;
#ifdef GUUS
LocalMatrix<double> Zin;
LocalVector<double> refsol;
LocalVector<double> refones;
#endif
mat.ReadFileMTX(std::string(argv[1]));
mat.info();
#ifdef GUUS
Zin.ReadFileMTX(std::string(argv[2]));
Zin.info();
refsol.Allocate("refsol", mat.get_nrow());
refones.Allocate("refones", mat.get_nrow());
//refsol.Ones();
refsol.ReadFileASCII(std::string(argv[4]));
refones.Ones();
#endif
x.Allocate("x", mat.get_nrow());
rhs.Allocate("rhs", mat.get_nrow());
// Linear Solver
DPCG<LocalMatrix<double>, LocalVector<double>, double > ls;
MultiElimination<LocalMatrix<double>, LocalVector<double>, double > p;
Jacobi<LocalMatrix<double>, LocalVector<double>, double > j_p;
MultiColoredILU<LocalMatrix<double>, LocalVector<double>, double > mcilu_p;
ILU<LocalMatrix<double>, LocalVector<double>, double > ilu_p;
MultiColoredSGS<LocalMatrix<double>, LocalVector<double>, double > mcsgs_p;
FSAI<LocalMatrix <double>, LocalVector<double>, double > fsai_p ;
SPAI<LocalMatrix <double>, LocalVector <double>, double > spai_p ;
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef SCALIN
mat.ExtractInverseDiagonal_sqrt(&Dinvhalf_min, -1);
mat.ExtractInverseDiagonal_sqrt(&Dinvhalf_plus, 1);
mat.DiagonalMatrixMult(Dinvhalf_min);
mat.DiagonalMatrixMult_fromL(Dinvhalf_min);
//x.PointWiseMult(Dinvhalf_plus);
rhs.PointWiseMult(Dinvhalf_min);
#endif
/////////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver MCSGS" << std::endl;
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.Setlvst_offst(lvst_offst);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
ls.SetOperator(mat);
ls.SetPreconditioner (mcsgs_p) ;
mcsgs_p.SetPrecondMatrixFormat(HYB);
ls.Init(0.0, 1e-6, 1e8, 200000);
#ifdef BUBFLO
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef GUUS
x.MoveToHost();
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
#endif
//x.WriteFileASCII("x_solution1e3shell_ilu01.rec");
ls.Clear();
/////////////////////////////////////////////////////////////////
/////////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver FSAI" << std::endl;
refones.Ones();
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
//x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
fsai_p.Set (2) ;
ls.SetOperator(mat);
ls.SetPreconditioner (fsai_p) ;
fsai_p.SetPrecondMatrixFormat(HYB);
ls.Init(0.0, 1e-6, 1e8, 200000);
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef BUBFLO
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
// mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef GUUS
x.MoveToHost();
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
#endif
//x.WriteFileASCII("x_solution1e3shell_ilu01.rec");
ls.Clear();
// //
///////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver ILU-p" << std::endl;
refones.Ones();
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
//x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
ilu_p.Set(0);
ls.SetOperator(mat);
ls.SetPreconditioner(ilu_p);
ls.Init(0.0, 1e-6, 1e8, 20000);
ls.RecordResidualHistory();
// mat.ConvertToCSR();
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef BUBFLO
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
//ls.Verbose(2);
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef SCALIN
x.PointWiseMult(Dinvhalf_min);
#endif
x.MoveToHost();
// x.WriteFileASCII("x_solution_shell_scal.rec");
#ifdef GUUS
// ls.RecordHistory("res_ongpu_ilu-p.rec");
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
x.MoveToHost();
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
#endif
ls.Clear();
/////////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver ME-ILU-J" << std::endl;
refones.Ones();
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
//x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
p.Set(j_p, 1);
ls.SetOperator(mat);
ls.SetPreconditioner(p);
ls.Init(0.0, 1e-6, 1e8, 200000);
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef BUBFLO
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef GUUS
x.MoveToHost();
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
#endif
//x.WriteFileASCII("x_solution1e3shell_ilu01.rec");
ls.Clear();
///////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver ME-ILU-SGS" << std::endl;
refones.Ones();
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
//x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
//p.Init(mcsgs_p, 1);
ls.SetOperator(mat);
ls.SetPreconditioner(mcsgs_p);
ls.Init(0.0, 1e-6, 1e8, 200000);
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef BUBFLO
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
// mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef GUUS
x.MoveToHost();
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
#endif
//x.WriteFileASCII("x_solution1e3shell_ilu01.rec");
ls.Clear();
//
// /////////////////////////////////////////////////////////////////
//
// /////////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver ME-ILU-ILU(0,1)" << std::endl;
refones.Ones();
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
//x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
// mcilu_p.Init(0);
//
// p.Init(mcilu_p, 1, 0.0);
mcilu_p.Set(0);
p.Set(mcilu_p, 1, 0.0);
ls.SetOperator(mat);
ls.SetPreconditioner(p);
// p.SetPrecondMatrixFormat(HYB);
ls.Init(0.0, 1e-6, 1e8, 200000);
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef BUBFLO
// ls.SetNVectors(4);
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
//
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
//
// ls.Verbose(2);
mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef GUUS
x.MoveToHost();
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
//x.WriteFileASCII("x_solution1e3shell_ilu01.rec");
#endif
ls.Clear();
// /////////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver ILU(0,1)" << std::endl;
refones.Ones();
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
//x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
mcilu_p.Set(0,1);
ls.SetOperator(mat);
ls.SetPreconditioner(mcilu_p);
ls.Init(0.0, 1e-6, 1e8, 200000);
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef BUBFLO
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
// mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef GUUS
x.MoveToHost();
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
// x.WriteFileASCII("x_solution1e3shell_ilu01.rec");
#endif
ls.Clear();
/////////////////////////////////////////////////////////////////
// /////////////////////////////////////////////////////////////////
std::cout << "-----------------------------------------------" << std::endl;
std::cout << "DPCG solver jacobi" << std::endl;
refones.Ones();
#ifdef GUUS
rhs.ReadFileASCII(std::string(argv[3]));
x.SetRandom(0.0,1.0,1000);
ls.SetZ(Zin);
#endif
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
rhs.ReadFileASCII(std::string(argv[3]));
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
//x.SetRandom(0.0,1.0,1000);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
mat.ConvertToCSR();
#endif
ls.SetOperator(mat);
ls.SetPreconditioner(j_p);
// p.SetPrecondMatrixFormat(HYB);
ls.Init(0.0, 1e-6, 1e8, 200000);
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
#endif
#ifdef BUBFLO
// ls.SetNVectors(4);
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
//
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b << " sec" << std::endl;
//
// ls.Verbose(2);
mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s= (tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef GUUS
x.MoveToHost();
sol_norm=x.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
//x.WriteFileASCII("x_solution1e3shell_ilu01.rec");
#endif
ls.Clear();
//
cout<<"########################################################################"<<endl;
cout<<"Everything complete stopping paralution now."<<endl;
stop_paralution();
return 0;
}
int main(int argc, char* argv[]) {
if (argc == 1) {
std::cerr << argv[0] << " <matrix> [Num threads]" << std::endl;
exit(1);
}
init_paralution();
if (argc > 2) {
set_omp_threads_paralution(atoi(argv[2]));
}
info_paralution();
LocalVector<double> x;
LocalVector<double> rhs;
LocalMatrix<double> mat;
mat.ReadFileMTX(std::string(argv[1]));
mat.info();
x.Allocate("x", mat.get_nrow());
rhs.Allocate("rhs", mat.get_nrow());
x.info();
rhs.info();
rhs.Ones();
mat.Apply(rhs, &x);
std::cout << "dot=" << x.Dot(rhs) << std::endl;
mat.ConvertToELL();
mat.info();
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
mat.info();
rhs.Ones();
mat.Apply(rhs, &x);
std::cout << "dot=" << x.Dot(rhs) << std::endl;
stop_paralution();
return 0;
}
int main(int argc, char* argv[]) {
if (argc == 1) {
std::cerr << argv[0] << " <matrix> <initial_guess> <rhs> [Num threads]" << std::endl;
exit(1);
}
init_paralution();
// if (argc > 4) {
// set_omp_threads_paralution(atoi(argv[5]));
// }
set_omp_threads_paralution(8);
info_paralution();
struct timeval now;
double tick, tack, b=0.0f,s=0.0f, lprep=0.0f, sol_norm, diff_norm, ones_norm;
double *phi_ptr=NULL;
int *bubmap_ptr=NULL, phisize, maxbmap, setlssd, lvst_offst;
int xdim, ydim, zdim, defvex_perdirec, defvex_perdirec_y, defvex_perdirec_z;
DPCG<LocalMatrix<double>, LocalVector<double>, double > ls;
#ifdef BUBFLO
xdim=atoi(argv[5]);
setlssd=atoi(argv[6]);
defvex_perdirec=atoi(argv[7]);
lvst_offst=atoi(argv[8]);
phisize=(xdim+2*lvst_offst)*(ydim+2*lvst_offst)*(zdim+2*lvst_offst);
#endif
LocalVector<double> x;
LocalVector<double>refsol;
LocalVector<double>refones;
LocalVector<double>chk_r;
LocalVector<double> rhs;
LocalMatrix<double> mat;
LocalVector<double> Dinvhalf_min;
LocalVector<double> Dinvhalf_plus;
#ifdef GUUS
LocalMatrix<double> Zin;
#endif
mat.ReadFileMTX(std::string(argv[1]));
mat.info();
#ifdef GUUS
Zin.ReadFileMTX(std::string(argv[2]));
Zin.info();
#endif
x.Allocate("x", mat.get_nrow());
refsol.Allocate("refsol", mat.get_nrow());
refones.Allocate("refones", mat.get_nrow());
rhs.Allocate("rhs", mat.get_nrow());
chk_r.Allocate("chk_r", mat.get_nrow());
#ifdef BUBFLO
x.ReadFileASCII(std::string(argv[2]));
#endif
rhs.ReadFileASCII(std::string(argv[3]));
#ifdef GUUS
x.SetRandom(0.0,1.0,1000);
refsol.ReadFileASCII(std::string(argv[4]));
refones.Ones();
#endif
//refsol.Ones();
//
// // Uncomment for GPU
#ifdef GPURUN
mat.MoveToAccelerator();
x.MoveToAccelerator();
rhs.MoveToAccelerator();
chk_r.MoveToAccelerator();
Dinvhalf_min.MoveToAccelerator();
Dinvhalf_plus.MoveToAccelerator();
#endif
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef BUBFLO
if(setlssd){
LocalVector<double> phi;
LocalVector<int> bubmap;
phi.Allocate("PHI", phisize);
bubmap.Allocate("bubmap",mat.get_nrow());
phi.ReadFileASCII(std::string(argv[4]));
bubmap.LeaveDataPtr(&bubmap_ptr);
phi.LeaveDataPtr(&phi_ptr);
bubmap_create(phi_ptr, bubmap_ptr, xdim, xdim, xdim, mat.get_nrow(), &maxbmap, lvst_offst);
phi.Clear();
}
ls.Setxdim(xdim);
ls.SetNVectors_eachdirec(defvex_perdirec+1, defvex_perdirec+2, defvex_perdirec+3);
ls.Set_alldims(xdim, xdim, xdim);
ls.Setlvst_offst(lvst_offst);
ls.SetNVectors(defvex_perdirec);
ls.SetZlssd(setlssd);
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
lprep=(tack-tick)/1000000;
std::cout << "levelset_prep" << lprep << " sec" << std::endl;
// Linear Solver
// return 0;
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
#ifdef SCALIN
mat.ExtractInverseDiagonal_sqrt(&Dinvhalf_min, -1);
mat.ExtractInverseDiagonal_sqrt(&Dinvhalf_plus, 1);
mat.DiagonalMatrixMult(Dinvhalf_min);
mat.DiagonalMatrixMult_fromL(Dinvhalf_min);
//x.PointWiseMult(Dinvhalf_plus);
rhs.PointWiseMult(Dinvhalf_min);
// rhs.Scale(0.3);
#endif
#ifdef GUUS
ls.SetZ(Zin);
#endif
ls.SetOperator(mat);
ls.Init(0.0, 1e-6, 1e8, 200000);
// ls.RecordResidualHistory();
#ifdef BUBFLO
ls.MakeZ_CSR(); // requires xdim_ and novecni_ and zlssd_ to be set
if(setlssd)
ls.MakeZLSSD(bubmap_ptr, maxbmap); // bubmap must be ready and maxbmap available
#endif
//
// stop_paralution();
// return 0;
ls.Build();
#ifdef MATDIA
mat.ConvertToDIA();
#endif
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
b=(tack-tick)/1000000;
std::cout << "Building:" << b+lprep << " sec" << std::endl;
// ls.Verbose(2);
mat.info();
gettimeofday(&now, NULL);
tick = now.tv_sec*1000000.0+(now.tv_usec);
ls.Solve(rhs, &x);
gettimeofday(&now, NULL);
tack = now.tv_sec*1000000.0+(now.tv_usec);
s=(tack-tick)/1000000;
std::cout << "Solver execution:" << s << " sec" << std::endl;
std::cout << "Total execution:" << s+b << " sec" << std::endl;
#ifdef SCALIN
x.PointWiseMult(Dinvhalf_min);
#endif
//
#ifdef GUUS
// x.WriteFileASCII("x_solution_shell_inv_neumann.rec");
//ls.RecordHistory("res__ongpu_tns.rec");
x.MoveToHost();
x.WriteFileASCII("x_neumann.rec");
x.MoveToAccelerator();
sol_norm=x.Norm();
mat.Apply(x, &chk_r);
chk_r.ScaleAdd(double(-1.0), rhs);
cout<<"\n Real Residual Norm is "<<chk_r.Norm();
cout<<"\n Norm of Solution is "<<sol_norm<<endl;
cout<<"\n Norm of Reference Solution is "<<refsol.Norm()<<endl;
cout<<"\n Norm of Ones is "<<refones.Norm()<<endl;
x.MoveToHost();
refones.AddScale(x,(double)-1.0f);
x.AddScale(refsol,(double)-1.0f);
diff_norm=x.Norm();
ones_norm=refones.Norm();
cout<<"\n Relative Norm of Calculated Solution w.r.t. Reference is "<<((double)diff_norm/(double)sol_norm)<<endl;
cout<<"\n Relative Norm of Calculated Solution w.r.t. Ones is "<<((double)ones_norm/(double)sol_norm)<<endl;
#endif
ls.Clear();
stop_paralution();
return 0;
}
