/*!
Get values of the specified elements from a global vector
@param v_type - Indicator for vector: 0: x; 1: rhs
@param ni - number of elements to get
@param ix - indices where to get them from (in global 1d numbering)
*/
void PETScLinearSolver::GetVecValues(const int v_type, PetscInt ni,const PetscInt ix[],
PetscScalar y[]) const
{
if(v_type == 0)
VecGetValues(x, ni, ix, y);
else
VecGetValues(b, ni, ix, y);
}
static PetscErrorCode FormJacobian1_block(SNES snes,Vec x,Mat *jac,Mat *B,MatStructure *flag,void *dummy)
{
Vec *xx, x1,x2;
PetscScalar xx_0, xx_1;
PetscInt index,nb;
PetscScalar A_00, A_01, A_10, A_11;
Mat j11, j12, j21, j22;
Mat **mats;
PetscErrorCode ierr;
PetscFunctionBegin;
/* get blocks for solution */
ierr = VecNestGetSubVecs( x, &nb, &xx );
CHKERRQ(ierr);
x1 = xx[0];
x2 = xx[1];
/* get solution values */
index = 0;
ierr = VecGetValues( x1,1, &index, &xx_0 );
CHKERRQ(ierr);
ierr = VecGetValues( x2,1, &index, &xx_1 );
CHKERRQ(ierr);
/* get block matrices */
ierr = MatNestGetSubMats(*jac,PETSC_NULL,PETSC_NULL,&mats);
CHKERRQ(ierr);
j11 = mats[0][0];
j12 = mats[0][1];
j21 = mats[1][0];
j22 = mats[1][1];
/* compute jacobian entries */
A_00 = 2.0*xx_0 + xx_1;
A_01 = xx_0;
A_10 = xx_1;
A_11 = xx_0 + 2.0*xx_1;
/* set jacobian values */
ierr = MatSetValue( j11, 0,0, A_00, INSERT_VALUES);
CHKERRQ(ierr);
ierr = MatSetValue( j12, 0,0, A_01, INSERT_VALUES);
CHKERRQ(ierr);
ierr = MatSetValue( j21, 0,0, A_10, INSERT_VALUES);
CHKERRQ(ierr);
ierr = MatSetValue( j22, 0,0, A_11, INSERT_VALUES);
CHKERRQ(ierr);
*flag = SAME_NONZERO_PATTERN;
/* Assemble sub matrix */
ierr = MatAssemblyBegin(*jac,MAT_FINAL_ASSEMBLY);
CHKERRQ(ierr);
ierr = MatAssemblyEnd(*jac,MAT_FINAL_ASSEMBLY);
CHKERRQ(ierr);
PetscFunctionReturn(0);
}
void linearSystemPETSc<fullMatrix<double> >::getFromSolution(int row, fullMatrix<double> &val) const
{
int blockSize;
_try(MatGetBlockSize(_a, &blockSize));
for (int i = 0; i < blockSize; i++) {
int ii = row*blockSize +i;
#ifdef PETSC_USE_COMPLEX
PetscScalar s;
VecGetValues ( _x, 1, &ii, &s);
val(i,0) = s.real();
#else
VecGetValues ( _x, 1, &ii, &val(i,0));
#endif
}
}
inline PetscErrorCode ScatIEMatMult(Mat mat,Vec xx,Vec yy)
{
PetscInt n,s,t;
PetscScalar tmp,v;
PetscInt xstart,xend;
PetscFunctionBegin;
VecGetOwnershipRange(xx,&xstart,&xend);
VecGetSize(yy,&n);
VecZeroEntries(yy);
for(s = 0; s <n ; s++)
{
tmp = 0;
for(t = xstart; t < xend; t++)
{
VecGetValues(xx,1,&t,&v);
tmp += ScatIE.CoefMatFast(s,t)*v;
}
VecSetValues(yy,1,&s,&tmp,ADD_VALUES);
}
VecAssemblyBegin(yy);
VecAssemblyEnd(yy);
PetscFunctionReturn(0);
}
double Integrate(Vec U, int *pt, unsigned int i, myCSField sf) {
if (i < sf->d){
double g = 0;
// Trapezoidal rule: \int_0^1 f(x) = (h/2) * \sum_{i=0}^{m} (f(x_i)+f(x_{i+1}))
// In the special case where f(0) = f(1) = f(x_0) = f(x_{m+1}) = 0
// \int_0^1 f(x) = 0.5*f(x_1) + { \sum_{i=1}^{m-1} 0.5*(f(x_i)+f(x_{i+1})) } + 0.5*f(x_m)
double h_2 = (dx(i, sf)/2.);
// u(x_0) = 0
pt[i] = 1;
g += Integrate(U, pt, i+1, sf) * h_2;
for (pt[i]=1;pt[i]<sf->mesh[i];pt[i]++) {
g += Integrate(U, pt, i+1, sf) * h_2;
pt[i] += 1;
g += Integrate(U, pt, i+1, sf) * h_2;
pt[i] -= 1;
}
pt[i] = sf->mesh[i];
g += Integrate(U, pt, i+1, sf) * h_2;
// u(x_{m+1}) = 0
return g;
}
int r = linIdx_Sys(sf->d, sf->mesh, pt,0,0);
double x[sf->d];
double Ux;
getPoint(pt, x, sf);
VecGetValues(U,1,&r,&Ux);
return QoIAtPoint(x, Ux, sf);
}
PetscErrorCode RetrieveVecPoints(Vec x, int Npt, int *Pos, double *ptValues)
{
PetscErrorCode ierr;
Vec T;
VecScatter scatter;
IS from, to;
ierr = VecCreateSeq(PETSC_COMM_SELF, Npt, &T);CHKERRQ(ierr);
ierr = ISCreateGeneral(PETSC_COMM_SELF,Npt, Pos,PETSC_COPY_VALUES, &from);CHKERRQ(ierr);
ierr = ISCreateStride(PETSC_COMM_SELF,Npt,0,1, &to);CHKERRQ(ierr);
ierr = VecScatterCreate(x,from,T,to,&scatter);CHKERRQ(ierr);
ierr = VecScatterBegin(scatter,x,T,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(scatter,x,T,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
int ix[Npt];
int i;
for(i=0; i<Npt; i++)
ix[i]=i;
ierr = VecGetValues(T,Npt,ix,ptValues);
ierr = ISDestroy(&from);CHKERRQ(ierr);
ierr = ISDestroy(&to);CHKERRQ(ierr);
ierr = VecScatterDestroy(&scatter);CHKERRQ(ierr);
ierr = VecDestroy(&T);CHKERRQ(ierr);
PetscFunctionReturn(0);
}
void operator| (PUP::er& p, Vec& v) {
PetscInt sz;
if (!p.isUnpacking()) {
VecGetSize(v, &sz);
}
p | sz;
if(p.isUnpacking()) {
VecCreateSeq(PETSC_COMM_WORLD, sz, &v);
VecSetFromOptions(v);
VecGetSize(v, &sz);
for (int i = 0; i < sz; i++) {
PetscScalar d;
p | d;
VecSetValue(v, i, d, INSERT_VALUES);
}
VecAssemblyBegin(v);
VecAssemblyEnd(v);
}
else {
for (int i = 0; i < sz; i++) {
PetscScalar d;
VecGetValues(v, 1, &i, &d);
p | d;
}
}
}
double SparseVector::operator ()(int i)
{
double value;
VecGetValues(vec_, 1, &i, &value);
return value;
}
static PetscErrorCode FormFunction1_block(SNES snes,Vec x,Vec f,void *dummy)
{
Vec *xx, *ff, x1,x2, f1,f2;
PetscScalar ff_0, ff_1;
PetscScalar xx_0, xx_1;
PetscInt index,nb;
PetscErrorCode ierr;
PetscFunctionBegin;
/* get blocks for function */
ierr = VecNestGetSubVecs( f, &nb, &ff );
CHKERRQ(ierr);
f1 = ff[0];
f2 = ff[1];
/* get blocks for solution */
ierr = VecNestGetSubVecs( x, &nb, &xx );
CHKERRQ(ierr);
x1 = xx[0];
x2 = xx[1];
/* get solution values */
index = 0;
ierr = VecGetValues( x1,1, &index, &xx_0 );
CHKERRQ(ierr);
ierr = VecGetValues( x2,1, &index, &xx_1 );
CHKERRQ(ierr);
/* Compute function */
ff_0 = xx_0*xx_0 + xx_0*xx_1 - 3.0;
ff_1 = xx_0*xx_1 + xx_1*xx_1 - 6.0;
/* set function values */
ierr = VecSetValue( f1, index, ff_0, INSERT_VALUES );
CHKERRQ(ierr);
ierr = VecSetValue( f2, index, ff_1, INSERT_VALUES );
CHKERRQ(ierr);
ierr = VecAssemblyBegin(f);
CHKERRQ(ierr);
ierr = VecAssemblyEnd(f);
CHKERRQ(ierr);
PetscFunctionReturn(0);
}
double PetscVector<double>::get(unsigned int idx) const
{
double y = 0;
PetscScalar py;
VecGetValues(vec, 1, (PetscInt*)&idx, &py);
y = py.real();
return y;
}
PetscErrorCode DiffIE(Vec originalVec, Vec reducedVec, Vec Distance, Vec nParticlePerCube)
{
// Find the distance between 2 solutions originalVec and reducedVec
PetscInt s,t;
PetscInt n,xstart,xend;
PetscScalar tmp,v,diff;
Vec vec;
VecScatter scat;
IS is;
PetscFunctionBegin;
VecGetOwnershipRange(originalVec,&xstart,&xend);
VecGetSize(reducedVec,&n);
VecCreate(PETSC_COMM_SELF,&vec);
VecSetSizes(vec,PETSC_DECIDE,n);
VecSetFromOptions(vec);
ISCreateStride(PETSC_COMM_WORLD,n,0,1,&is);
VecScatterCreate(reducedVec,PETSC_NULL,vec,is,&scat);
//VecScatterCreateToAll(reducedVec,&scat,&vec);
VecScatterBegin(scat,reducedVec,vec,INSERT_VALUES,SCATTER_FORWARD);
VecScatterEnd(scat,reducedVec,vec,INSERT_VALUES,SCATTER_FORWARD);
for(s=xstart;s<xend;s++)
{
t = ScatIE.FindCube(s);
VecGetValues(vec,1,&t,&tmp);
VecGetValues(originalVec,1,&s,&v);
diff = cabs(v - tmp);
VecSetValues(Distance,1,&t,&diff,ADD_VALUES);
tmp = 1;
VecSetValues(nParticlePerCube,1,&t,&tmp,ADD_VALUES);
}
VecDestroy(&vec);
VecScatterDestroy(&scat);
ISDestroy(&is);
PetscFunctionReturn(0);
}
/*! Solve for x in Ax=b
*/
void HeatSolverBTCS::solve(std::vector<double>& x)
{
PetscInt i;
PetscScalar s;
KSPSolve(ksp, rhs, temp);
KSPGetConvergedReason(ksp, &reason);
for (i = 0; i < nx*ny; i++) {
VecGetValues(temp, 1, &i, &s);
x[i] = s;
}
}
PetscErrorCode SolveInit(FEMInf fem, int L, PetscScalar *e0, Vec *x) {
PetscErrorCode ierr;
Mat H, S;
ierr = CalcMat(fem, L, &H, &S); CHKERRQ(ierr);
EPS eps;
ierr = PrintTimeStamp(fem->comm, "EPS", NULL); CHKERRQ(ierr);
ierr = EPSCreate(fem->comm, &eps); CHKERRQ(ierr);
ierr = EPSSetTarget(eps, -0.6); CHKERRQ(ierr);
ierr = EPSSetWhichEigenpairs(eps, EPS_TARGET_MAGNITUDE); CHKERRQ(ierr);
ierr = EPSSetOperators(eps, H, S); CHKERRQ(ierr);
if(S == NULL) {
ierr = EPSSetProblemType(eps, EPS_NHEP); CHKERRQ(ierr);
} else {
ierr = EPSSetProblemType(eps, EPS_GNHEP); CHKERRQ(ierr);
}
Vec x0[1]; MatCreateVecs(H, &x0[0], NULL);
int num; FEMInfGetSize(fem, &num);
for(int i = 0; i < num; i++) {
VecSetValue(x0[0], i, 0.5, INSERT_VALUES);
}
VecAssemblyBegin(x0[0]); VecAssemblyEnd(x0[0]);
EPSSetInitialSpace(eps, 1, x0);
ierr = EPSSetFromOptions(eps); CHKERRQ(ierr);
ierr = EPSSolve(eps); CHKERRQ(ierr);
PetscInt nconv;
EPSGetConverged(eps, &nconv);
if(nconv == 0)
SETERRQ(fem->comm, 1, "Failed to digonalize in init state\n");
Vec x_ans;
MatCreateVecs(H, &x_ans, NULL);
EPSGetEigenpair(eps, 0, e0, NULL, x_ans, NULL);
EPSDestroy(&eps);
PetscScalar v[1]; PetscInt idx[1] = {1};
VecGetValues(x_ans, 1, idx, v);
PetscScalar scale_factor = v[0] / cabs(v[0]);
VecScale( x_ans, 1.0/scale_factor);
PetscScalar norm0;
Vec Sx; MatCreateVecs(S, &Sx, NULL);
MatMult(S, x_ans, Sx); VecDot(x_ans, Sx, &norm0);
VecScale(x_ans, 1.0/sqrt(norm0));
*x = x_ans;
return 0;
}
Vector<Scalar>* PetscVector<Scalar>::change_sign()
{
PetscScalar* y = malloc_with_check(this->size, this);
int *idx = malloc_with_check(this->size, this);
for (unsigned int i = 0; i < this->size; i++) idx[i] = i;
VecGetValues(vec, this->size, idx, y);
for (unsigned int i = 0; i < this->size; i++) y[i] *= -1.;
VecSetValues(vec, this->size, idx, y, INSERT_VALUES);
free_with_check(y);
free_with_check(idx);
return this;
}
void linearSystemPETSc<scalar>::getFromSolution(int row, scalar &val) const
{
#if defined(PETSC_USE_COMPLEX)
PetscScalar *tmp;
_try(VecGetArray(_x, &tmp));
PetscScalar s = tmp[row];
_try(VecRestoreArray(_x, &tmp));
val = s.real();
#else
_try(VecGetValues(_x, 1, &row, &val));
#endif
}
void linearSystemPETSc<scalar>::getFromRightHandSide(int row, scalar &val) const
{
#if defined(PETSC_USE_COMPLEX)
PetscScalar *tmp;
_try(VecGetArray(_b, &tmp));
PetscScalar s = tmp[row];
_try(VecRestoreArray(_b, &tmp));
// FIXME specialize this routine
val = s.real();
#else
_try(VecGetValues(_b, 1, &row, &val));
#endif
}
/**
* Copies a PETSc vector into an array of doubles. They must be the same size
*
* @param vec the PETSc vector
* @param array a pointer to the double array
*/
PetscErrorCode _pvec_to_array(Vec *vec, double *array) {
PetscErrorCode ierr;
PetscInt size;
ierr = VecGetSize(*vec, &size); CHKERRQ(ierr);
// Make array of indices
int indices[size];
for (int i = 0; i < size; i++) {
indices[i] = i;
}
// Get values from PETSc
ierr = VecGetValues(*vec, size, indices, array); CHKERRQ(ierr);
return 0;
}
bool multiply(const std::vector<double>& input, std::vector<double>& output ) const
{
for (int i = 0; i < n; ++i)
{
VecSet(temp_input,input[i]);
}
MatMult(A, temp_input, temp_output) ;
for (int i = 0; i < n; ++i)
{
PetscInt tempi=i;
PetscScalar tempv;
VecGetValues(temp_output, 1,&tempi,&tempv);
output[i] = tempv;
}
return true;
}
std::complex<double> PetscVector<std::complex<double> >::get(unsigned int idx) const
{
std::complex<double> y = 0;
VecGetValues(vec, 1, (PetscInt*)&idx, &y);
return y;
}
PetscInt main(PetscInt argc,char **args)
{
ptrdiff_t N0=256,N1=256,N2=256,N3=2,dim[4];
fftw_plan bplan,fplan;
fftw_complex *out;
double *in1,*in2;
ptrdiff_t alloc_local,local_n0,local_0_start;
ptrdiff_t local_n1,local_1_start;
PetscInt i,j,indx,n1;
PetscInt size,rank,n,N,*in,N_factor,NM;
PetscScalar *data_fin,value1,one=1.57,zero=0.0;
PetscScalar a,*x_arr,*y_arr,*z_arr,enorm;
Vec fin,fout,fout1,ini,final;
PetscRandom rnd;
PetscErrorCode ierr;
VecScatter vecscat,vecscat1;
IS indx1,indx2;
PetscInt *indx3,k,l,*indx4;
PetscInt low,tempindx,tempindx1;
ierr = PetscInitialize(&argc,&args,(char*)0,help);if (ierr) return ierr;
#if defined(PETSC_USE_COMPLEX)
SETERRQ(PETSC_COMM_WORLD,PETSC_ERR_SUP, "This example requires real numbers. Your current scalar type is complex");
#endif
ierr = MPI_Comm_size(PETSC_COMM_WORLD, &size);CHKERRQ(ierr);
ierr = MPI_Comm_rank(PETSC_COMM_WORLD, &rank);CHKERRQ(ierr);
PetscRandomCreate(PETSC_COMM_WORLD,&rnd);
alloc_local = fftw_mpi_local_size_3d_transposed(N0,N1,N2/2+1,PETSC_COMM_WORLD,&local_n0,&local_0_start,&local_n1,&local_1_start);
/* printf("The value alloc_local is %ld from process %d\n",alloc_local,rank); */
printf("The value local_n0 is %ld from process %d\n",local_n0,rank);
/* printf("The value local_0_start is %ld from process %d\n",local_0_start,rank);*/
/* printf("The value local_n1 is %ld from process %d\n",local_n1,rank); */
/* printf("The value local_1_start is %ld from process %d\n",local_1_start,rank);*/
/* Allocate space for input and output arrays */
in1=(double*)fftw_malloc(sizeof(double)*alloc_local*2);
in2=(double*)fftw_malloc(sizeof(double)*alloc_local*2);
out=(fftw_complex*)fftw_malloc(sizeof(fftw_complex)*alloc_local);
N=2*N0*N1*(N2/2+1);N_factor=N0*N1*N2;
n=2*local_n0*N1*(N2/2+1);n1=local_n1*N0*2*N1;
/* printf("The value N is %d from process %d\n",N,rank); */
/* printf("The value n is %d from process %d\n",n,rank); */
/* printf("The value n1 is %d from process %d\n",n1,rank); */
/* Creating data vector and accompanying array with VeccreateMPIWithArray */
ierr = VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)in1,&fin);CHKERRQ(ierr);
ierr = VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)out,&fout);CHKERRQ(ierr);
ierr = VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)in2,&fout1);CHKERRQ(ierr);
/* VecGetSize(fin,&size); */
/* printf("The size is %d\n",size); */
VecSet(fin,one);
VecSet(fout,zero);
VecSet(fout1,zero);
VecAssemblyBegin(fin);
VecAssemblyEnd(fin);
/* VecView(fin,PETSC_VIEWER_STDOUT_WORLD); */
VecGetArray(fin,&x_arr);
VecGetArray(fout1,&z_arr);
VecGetArray(fout,&y_arr);
fplan=fftw_mpi_plan_dft_r2c_3d(N0,N1,N2,(double*)x_arr,(fftw_complex*)y_arr,PETSC_COMM_WORLD,FFTW_ESTIMATE);
bplan=fftw_mpi_plan_dft_c2r_3d(N0,N1,N2,(fftw_complex*)y_arr,(double*)z_arr,PETSC_COMM_WORLD,FFTW_ESTIMATE);
fftw_execute(fplan);
fftw_execute(bplan);
VecRestoreArray(fin,&x_arr);
VecRestoreArray(fout1,&z_arr);
VecRestoreArray(fout,&y_arr);
/* a = 1.0/(PetscReal)N_factor; */
/* ierr = VecScale(fout1,a);CHKERRQ(ierr); */
VecCreate(PETSC_COMM_WORLD,&ini);
VecCreate(PETSC_COMM_WORLD,&final);
VecSetSizes(ini,local_n0*N1*N2,N_factor);
VecSetSizes(final,local_n0*N1*N2,N_factor);
/* VecSetSizes(ini,PETSC_DECIDE,N_factor); */
/* VecSetSizes(final,PETSC_DECIDE,N_factor); */
VecSetFromOptions(ini);
VecSetFromOptions(final);
if (N2%2==0) NM=N2+2;
else NM=N2+1;
ierr = VecGetOwnershipRange(fin,&low,NULL);
printf("The local index is %d from %d\n",low,rank);
ierr = PetscMalloc1(local_n0*N1*N2,&indx3);
ierr = PetscMalloc1(local_n0*N1*N2,&indx4);
for (i=0; i<local_n0; i++) {
for (j=0;j<N1;j++) {
for (k=0;k<N2;k++) {
tempindx = i*N1*N2 + j*N2 + k;
tempindx1 = i*N1*NM + j*NM + k;
indx3[tempindx]=local_0_start*N1*N2+tempindx;
indx4[tempindx]=low+tempindx1;
}
/* printf("index3 %d from proc %d is \n",indx3[tempindx],rank); */
/* printf("index4 %d from proc %d is \n",indx4[tempindx],rank); */
}
}
VecGetValues(fin,local_n0*N1*N2,indx4,x_arr);
VecSetValues(ini,local_n0*N1*N2,indx3,x_arr,INSERT_VALUES);
VecAssemblyBegin(ini);
VecAssemblyEnd(ini);
VecGetValues(fout1,local_n0*N1*N2,indx4,y_arr);
VecSetValues(final,local_n0*N1*N2,indx3,y_arr,INSERT_VALUES);
VecAssemblyBegin(final);
VecAssemblyEnd(final);
printf("The local index value is %ld from %d",local_n0*N1*N2,rank);
/*
for (i=0;i<N0;i++) {
for (j=0;j<N1;j++) {
indx=i*N1*NM+j*NM;
ISCreateStride(PETSC_COMM_WORLD,N2,indx,1,&indx1);
indx=i*N1*N2+j*N2;
ISCreateStride(PETSC_COMM_WORLD,N2,indx,1,&indx2);
VecScatterCreate(fin,indx1,ini,indx2,&vecscat);
VecScatterBegin(vecscat,fin,ini,INSERT_VALUES,SCATTER_FORWARD);
VecScatterEnd(vecscat,fin,ini,INSERT_VALUES,SCATTER_FORWARD);
VecScatterCreate(fout1,indx1,final,indx2,&vecscat1);
VecScatterBegin(vecscat1,fout1,final,INSERT_VALUES,SCATTER_FORWARD);
VecScatterEnd(vecscat1,fout1,final,INSERT_VALUES,SCATTER_FORWARD);
}
}
*/
a = 1.0/(PetscReal)N_factor;
ierr = VecScale(fout1,a);CHKERRQ(ierr);
ierr = VecScale(final,a);CHKERRQ(ierr);
VecAssemblyBegin(ini);
VecAssemblyEnd(ini);
VecAssemblyBegin(final);
VecAssemblyEnd(final);
/* VecView(final,PETSC_VIEWER_STDOUT_WORLD); */
ierr = VecAXPY(final,-1.0,ini);CHKERRQ(ierr);
ierr = VecNorm(final,NORM_1,&enorm);CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD," Error norm of |x - z| = %e\n",enorm);CHKERRQ(ierr);
fftw_destroy_plan(fplan);
fftw_destroy_plan(bplan);
fftw_free(in1); ierr = VecDestroy(&fin);CHKERRQ(ierr);
fftw_free(out); ierr = VecDestroy(&fout);CHKERRQ(ierr);
fftw_free(in2); ierr = VecDestroy(&fout1);CHKERRQ(ierr);
ierr = PetscFinalize();
return ierr;
}
int main(int argc, char** args){
PetscErrorCode err;
PetscViewer fd = NULL;
Mat invL1 = NULL, invU1 = NULL, invL2 = NULL, invU2 = NULL, H12 = NULL, H21 = NULL;
Vec order = NULL, r = NULL; //, or = NULL; //dimension: n: n1 + n2
Vec seeds = NULL;
// Vec r1 = NULL, q1 = NULL, t1_1 = NULL, t1_2 = NULL, t1_3 = NULL, t1_4 = NULL, t1_5 = NULL; // dimension: n1
// Vec r2 = NULL, q2 = NULL, q_tilda = NULL, t2_1 = NULL, t2_2 = NULL, t2_3 = NULL; // dimension: n2
PetscRandom rand;
PetscLogDouble tic, toc, total_time, time;
PetscInt n, i;
PetscMPIInt rank, size;
PetscInt seed;
PetscScalar c, val;
PetscInt QN = 100;
// Initialize PETSC and MPI
err = PetscInitialize(&argc, &args, (char*) 0, help); CHKERRQ(err);
err = MPI_Comm_size(PETSC_COMM_WORLD, &size); CHKERRQ(err);
err = MPI_Comm_rank(PETSC_COMM_WORLD, &rank); CHKERRQ(err);
err = PetscPrintf(PETSC_COMM_WORLD, "mpi size: %d\n", size); CHKERRQ(err);
// Read matrices and an ordering vector
err = PetscPrintf(PETSC_COMM_WORLD, "Read inputs (invL1, invU1, invL2, invU2, H12, H21, order)\n"); CHKERRQ(err);
err = loadMat("./data/invL1.dat", &invL1, PETSC_COMM_WORLD, MATMPIAIJ, &fd); CHKERRQ(err);
err = checkMat("invL1", invL1); CHKERRQ(err);
err = loadMat("./data/invU1.dat", &invU1, PETSC_COMM_WORLD, MATMPIAIJ, &fd); CHKERRQ(err);
err = checkMat("invU1", invU1); CHKERRQ(err);
err = loadMat("./data/invL2.dat", &invL2, PETSC_COMM_WORLD, MATMPIAIJ, &fd); CHKERRQ(err);
err = checkMat("invL2", invL2); CHKERRQ(err);
err = loadMat("./data/invU2.dat", &invU2, PETSC_COMM_WORLD, MATMPIAIJ, &fd); CHKERRQ(err);
err = checkMat("invU2", invU2); CHKERRQ(err);
err = loadMat("./data/H12.dat", &H12, PETSC_COMM_WORLD, MATMPIAIJ, &fd); CHKERRQ(err);
err = checkMat("H12", H12); CHKERRQ(err);
err = loadMat("./data/H21.dat", &H21, PETSC_COMM_WORLD, MATMPIAIJ, &fd); CHKERRQ(err);
err = checkMat("H21", H21); CHKERRQ(err);
err = loadVec("./data/order.dat", &order, PETSC_COMM_SELF, &fd); CHKERRQ(err); //all processes must have this vector for ordering the result vector.
err = checkVec("order", order); CHKERRQ(err);
// shift -1 for zero-based index
err = VecShift(order, -1); CHKERRQ(err);
err = VecGetSize(order, &n); CHKERRQ(err);
seed = 5;
c = 0.05;
err = PetscTime(&tic); CHKERRQ(err);
//err = BearQueryMat(seed, c, invL1, invU1, invL2, invU2, H12, H21, order); CHKERRQ(err);
//err = PetscTime(&toc); CHKERRQ(err);
//time = toc - tic;
//err = PetscPrintf(PETSC_COMM_WORLD, "running time: %f sec\n", time); CHKERRQ(err);
///* 100 times querying
err = VecCreateSeq(PETSC_COMM_SELF, QN, &seeds); CHKERRQ(err);
err = VecSetFromOptions(seeds); CHKERRQ(err);
err = PetscRandomCreate(PETSC_COMM_WORLD, &rand); CHKERRQ(err);
err = PetscRandomSetSeed(rand, 100); CHKERRQ(err);
err = PetscRandomSetInterval(rand, (PetscScalar) 0, (PetscScalar) n); CHKERRQ(err);
err = PetscRandomSetFromOptions(rand); CHKERRQ(err);
err = VecSetRandom(seeds, rand); CHKERRQ(err);
err = PetscRandomDestroy(&rand); CHKERRQ(err);
seed = 5; //seed is give by user on one-based index
c = 0.05;
i = 0;
err = VecDuplicate(order, &r); CHKERRQ(err);
for(i = 0; i < QN; i++){
err = VecGetValues(seeds, 1, &i, &val);
seed = (PetscInt) val;
//err = PetscPrintf(PETSC_COMM_SELF, "rank: %d, seed: %d\n", rank, seed);
err = PetscTime(&tic); CHKERRQ(err);
err = BearQuery(seed, c, invL1, invU1, invL2, invU2, H12, H21, order, r); CHKERRQ(err);
err = PetscTime(&toc); CHKERRQ(err);
time = toc - tic;
err = PetscPrintf(PETSC_COMM_WORLD, "running time: %f sec\n", time); CHKERRQ(err);
total_time += time;
}
err = PetscPrintf(PETSC_COMM_WORLD, "average running time: %f sec\n", total_time/QN); CHKERRQ(err);
err = VecDestroy(&r);
/* err = MatGetSize(H12, &n1, &n2); CHKERRQ(err);
n = n1 + n2;
err = PetscPrintf(PETSC_COMM_WORLD, "n1: %d, n2: %d\n", n1, n2); CHKERRQ(err);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n, &r); CHKERRQ(err);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n1, &q1); CHKERRQ(err);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n2, &q2); CHKERRQ(err);
err = VecSet(q1, 0); CHKERRQ(err);
err = VecSet(q2, 0); CHKERRQ(err);
seed = seed - 1; // shift -1 for zero-based index
err = VecGetValues(order, 1, &seed, &val); CHKERRQ(err);
oseed = (PetscInt) val;
err = PetscPrintf(PETSC_COMM_WORLD, "Given seed: %d, Reorered seed: %d (0 ~ n-1)\n", seed, oseed); CHKERRQ(err);
if(oseed < n1){
err = VecSetValues(q1, 1, &oseed, &one, INSERT_VALUES); CHKERRQ(err);
}else{
oseed = oseed - n1;
err = VecSetValues(q2, 1, &oseed, &one, INSERT_VALUES); CHKERRQ(err);
//err = printVecSum(q2);
}
err = VecAssemblyBegin(q1); CHKERRQ(err);
err = VecAssemblyBegin(q2); CHKERRQ(err);
err = VecAssemblyEnd(q1); CHKERRQ(err);
err = VecAssemblyEnd(q2); CHKERRQ(err);
err = VecDuplicate(q1, &r1); CHKERRQ(err);
err = VecDuplicate(q1, &t1_1); CHKERRQ(err);
err = VecDuplicate(q1, &t1_2); CHKERRQ(err);
err = VecDuplicate(q1, &t1_3); CHKERRQ(err);
err = VecDuplicate(q1, &t1_4); CHKERRQ(err);
err = VecDuplicate(q1, &t1_5); CHKERRQ(err);
err = VecDuplicate(q2, &r2); CHKERRQ(err);
err = VecDuplicate(q2, &q_tilda); CHKERRQ(err);
err = VecDuplicate(q2, &t2_1); CHKERRQ(err);
err = VecDuplicate(q2, &t2_2); CHKERRQ(err);
err = VecDuplicate(q2, &t2_3); CHKERRQ(err);
// Start matrix-vec multiplications
err = MatMult(invL1, q1, t1_1); CHKERRQ(err);
err = MatMult(invU1, t1_1, t1_2); CHKERRQ(err);
err = MatMult(H21, t1_2, t2_1); CHKERRQ(err);
err = VecAXPBYPCZ(q_tilda, 1.0, -1.0, 0.0, q2, t2_1); CHKERRQ(err);
err = MatMult(invL2, q_tilda, t2_2); CHKERRQ(err);
err = MatMult(invU2, t2_2, r2); CHKERRQ(err);
err = MatMult(H12, r2, t1_3); CHKERRQ(err);
err = VecAXPBYPCZ(t1_4, 1.0, -1.0, 0.0, q1, t1_3); CHKERRQ(err);
err = MatMult(invL1, t1_4, t1_5); CHKERRQ(err);
err = MatMult(invU1, t1_5, r1); CHKERRQ(err);
//err = printVecSum(r1);
//err = VecView(r2, PETSC_VIEWER_STDOUT_WORLD);
// Concatenate r1 and r2
err = VecMerge(r1, r2, r); CHKERRQ(err);
err = VecScale(r, c); CHKERRQ(err);
//err = VecView(r, PETSC_VIEWER_STDOUT_WORLD);
err = VecDuplicate(r, &or); CHKERRQ(err);
err = VecReorder(r, order, or); CHKERRQ(err);
//err = VecView(or, PETSC_VIEWER_STDOUT_WORLD);*/
// Destory matrices and vectors
err = MatDestroy(&invL1); CHKERRQ(err);
err = MatDestroy(&invU1); CHKERRQ(err);
err = MatDestroy(&invL2); CHKERRQ(err);
err = MatDestroy(&invU2); CHKERRQ(err);
err = MatDestroy(&H12); CHKERRQ(err);
err = MatDestroy(&H21); CHKERRQ(err);
err = VecDestroy(&order); CHKERRQ(err);
err = VecDestroy(&r); CHKERRQ(err);
err = VecDestroy(&seeds); CHKERRQ(err);
//err = VecDestroy(&or); CHKERRQ(err);
/* err = VecDestroy(&r1); CHKERRQ(err);
err = VecDestroy(&q1); CHKERRQ(err);
err = VecDestroy(&t1_1); CHKERRQ(err);
err = VecDestroy(&t1_2); CHKERRQ(err);
err = VecDestroy(&t1_3); CHKERRQ(err);
err = VecDestroy(&t1_4); CHKERRQ(err);
err = VecDestroy(&t1_5); CHKERRQ(err);
err = VecDestroy(&r2); CHKERRQ(err);
err = VecDestroy(&q2); CHKERRQ(err);
err = VecDestroy(&q_tilda); CHKERRQ(err);
err = VecDestroy(&t2_1); CHKERRQ(err);
err = VecDestroy(&t2_2); CHKERRQ(err);
err = VecDestroy(&t2_3); CHKERRQ(err);*/
// Finalize
err = PetscFinalize(); CHKERRQ(err);
return 0;
}
PetscErrorCode cHamiltonianMatrix::measurement(){
double *ALLdepart = new double[Nt];
double *ALLentropy = new double[Nt];
gsl_matrix* density1 = gsl_matrix_alloc(L,Nt);//background fermion density
gsl_matrix* density2 = gsl_matrix_alloc(L,Nt);//background fermion density
gsl_vector* corr12 = gsl_vector_alloc(Nt);//correlation betwen fermion up and down.
// The density correlation is in fact proportional to the interacting energy.
double var_rank;
PetscScalar var_tmp, var_tmp2;
gsl_complex var_tmp_gsl;
Vec vectort;
VecScatter ctx;
ofstream output;
VecScatterCreateToZero(WFt[0],&ctx,&vectort);
if(rank==0) cout << size << endl;
gsl_matrix_complex* RDM = gsl_matrix_complex_alloc(dim2,dim2);
gsl_vector *eval_RDM = gsl_vector_alloc(dim2);
gsl_eigen_herm_workspace* w_RDM = gsl_eigen_herm_alloc(dim2);
for (int itime = 0; itime < Nt; ++itime) {
if (rank==0&&itime%10==0) cout << "this is time " << itime << endl;
// % ## departure ##
var_rank = 0.0;
for (int ivar = rstart; ivar < rend; ++ivar) {
ierr = VecGetValues(WFt[itime],1,&ivar,&var_tmp);CHKERRQ(ierr);
var_rank += pow(gsl_vector_get(rr,ivar)*PetscAbsComplex(var_tmp),2);
}
MPI_Reduce(&var_rank, &(ALLdepart[itime]), 1, MPI_DOUBLE, MPI_SUM, 0, PETSC_COMM_WORLD);
ALLdepart[itime] = sqrt(ALLdepart[itime]);
// % ## entropy ##
VecScatterBegin(ctx,WFt[itime],vectort,INSERT_VALUES,SCATTER_FORWARD);
VecScatterEnd(ctx,WFt[itime],vectort,INSERT_VALUES,SCATTER_FORWARD);
if(rank==0) {
int ivar;double eigen_RDM;
gsl_matrix_complex_set_zero(RDM);
for (int row2 = 0; row2 < dim2; ++row2) {
for (int col2 = row2; col2 < dim2; ++col2) {
var_tmp_gsl.dat[0] = 0.0; var_tmp_gsl.dat[1] = 0.0;
for (int jjj = 0; jjj < dim; ++jjj) {
ivar = row2*dim+jjj;
ierr = VecGetValues(vectort,1,&ivar,&var_tmp);CHKERRQ(ierr);
ivar = col2*dim+jjj;
ierr = VecGetValues(vectort,1,&ivar,&var_tmp2);CHKERRQ(ierr);
var_tmp_gsl.dat[0] += PetscRealPart(var_tmp*PetscConj(var_tmp2));
var_tmp_gsl.dat[1] += PetscImaginaryPart(var_tmp*PetscConj(var_tmp2));
}
gsl_matrix_complex_set(RDM,row2,col2,var_tmp_gsl);
if (col2 != row2) {
gsl_matrix_complex_set(RDM,col2,row2,gsl_matrix_complex_get(RDM,row2,col2));
}
}
}
gsl_eigen_herm(RDM,eval_RDM,w_RDM);
ALLentropy[itime] = 0;
for (ivar = 0; ivar < dim2; ++ivar) {
eigen_RDM = gsl_vector_get(eval_RDM, ivar);
// cout << eigen_RDM << endl;
ALLentropy[itime] += -eigen_RDM*log(eigen_RDM);
}
}
// % ## density distribution of impurity fermion
if(rank==0) {
int ivar;
for (int row2 = 0; row2 < dim2; ++row2) {
for (int jpar = 0; jpar < N2; ++jpar) {
double density_tmp=0;
for (int jjj = 0; jjj < dim; ++jjj) {
ivar = row2*dim+jjj;
ierr = VecGetValues(vectort,1,&ivar,&var_tmp);CHKERRQ(ierr);
density_tmp +=pow(PetscAbsComplex(var_tmp),2);
}
/*if (itime==0) {
if (rank==0) cout << "density_tmp=" << density_tmp << endl;
}*/
gsl_matrix_set(density2,gsl_matrix_get(basis2,jpar,row2)-1,itime,gsl_matrix_get(density2,gsl_matrix_get(basis2,jpar,row2)-1,itime)+density_tmp);
}
}
}
/*if (rank==0) {
cout << "density of impurity:" << endl;
for (int jpar =0; jpar < L; ++jpar) {
cout << gsl_matrix_get(density2,jpar,itime) << "\t";
}
cout << endl;
}*/
// % ## density distribution of majority fermions
if(rank==0) {
int ivar;
for (int jjj = 0; jjj < dim; ++jjj) {
for (int jpar = 0; jpar < N; ++jpar) {
double density_tmp=0;
for (int row2 = 0; row2 < dim2; ++row2) {
ivar = row2*dim+jjj;
ierr = VecGetValues(vectort,1,&ivar,&var_tmp);CHKERRQ(ierr);
density_tmp +=pow(PetscAbsComplex(var_tmp),2);
}
gsl_matrix_set(density1,gsl_matrix_get(basis1,jpar,jjj)-1,itime,gsl_matrix_get(density1,gsl_matrix_get(basis1,jpar,jjj)-1,itime)+density_tmp);
}
}
}
// correlation between impurity and majority fermion
if(rank==0) {
int ivar;
double corr_tmp=0;
for (int jimp=0; jimp<dim2; ++jimp) {
for (int jmaj=0; jmaj<dim; ++jmaj) {
for (int jpar=0; jpar<N; ++jpar) {
if (gsl_matrix_get(basis1,jpar,jmaj)==jimp+1){
ivar = jimp*dim+jmaj;
ierr = VecGetValues(vectort,1,&ivar,&var_tmp);CHKERRQ(ierr);
corr_tmp+=pow(PetscAbsComplex(var_tmp),2);
}
}
}
}
gsl_vector_set(corr12,itime,corr_tmp);
}// end of correlation
}
if (rank == 0) {
char filename[50];
sprintf(filename,"measurement.data");
output.open(filename);
output.is_open();
output.precision(16);
for (int itime = 0; itime < Nt; ++itime) {
if (itime==0) {
// cout << "time t[1] " << '\t' << "departure[2] " << '\t' << "entropy[3]" << '\t' << "density of majority [L]" <<'\t' << "density of impurity [L]" << endl;
}
output << dt*itime-3 << '\t' << ALLdepart[itime] << '\t' << ALLentropy[itime] << '\t';
for (int jpar = 0; jpar < L; ++jpar) {
output << gsl_matrix_get(density1,jpar,itime) << '\t';
}
for (int jpar = 0; jpar < L; ++jpar) {
output << gsl_matrix_get(density2,jpar,itime) << '\t';
}
output << gsl_vector_get(corr12,itime) << '\t';
output << endl;
}
output.close();
}
// CopyFile(source,destination,FALSE);
delete[] ALLdepart;
VecScatterDestroy(&ctx);
VecDestroy(&vectort);
gsl_matrix_complex_free(RDM);
gsl_vector_free(eval_RDM);
gsl_eigen_herm_free(w_RDM);
gsl_matrix_free(density1);
gsl_matrix_free(density2);
gsl_vector_free(corr12);
// CopyFile(source,destination,FALSE);
return ierr;
}
PetscErrorCode BearQuery(PetscInt s, PetscScalar c, Mat invL1, Mat invU1, Mat invL2, Mat invU2, Mat H12, Mat H21, Vec order, Vec or){
PetscErrorCode err;
PetscInt n1, n2, n;
PetscInt oseed;
PetscScalar val, one = 1.0;
Vec r = NULL;
Vec r1 = NULL, q1 = NULL, t1_1 = NULL, t1_2 = NULL, t1_3 = NULL, t1_4 = NULL, t1_5 = NULL; // dimension: n1
Vec r2 = NULL, q2 = NULL, q_tilda = NULL, t2_1 = NULL, t2_2 = NULL, t2_3 = NULL; // dimension: n2
err = MatGetSize(H12, &n1, &n2); CHKERRQ(err);
n = n1 + n2;
// err = PetscPrintf(PETSC_COMM_WORLD, "n1: %d, n2: %d\n", n1, n2); CHKERRQ(err);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n, &r); CHKERRQ(err);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n1, &q1); CHKERRQ(err);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n2, &q2); CHKERRQ(err);
err = VecSet(q1, 0); CHKERRQ(err);
err = VecSet(q2, 0); CHKERRQ(err);
s = s - 1; // shift -1 for zero-based index
err = VecGetValues(order, 1, &s, &val); CHKERRQ(err);
oseed = (PetscInt) val;
// err = PetscPrintf(PETSC_COMM_WORLD, "Given seed: %d, Reorered seed: %d (0 ~ n-1)\n", s, oseed); CHKERRQ(err);
if(oseed < n1){
err = VecSetValues(q1, 1, &oseed, &one, INSERT_VALUES); CHKERRQ(err);
}else{
oseed = oseed - n1;
err = VecSetValues(q2, 1, &oseed, &one, INSERT_VALUES); CHKERRQ(err);
//err = printVecSum(q2);
}
err = VecAssemblyBegin(q1); CHKERRQ(err);
err = VecAssemblyBegin(q2); CHKERRQ(err);
err = VecAssemblyEnd(q1); CHKERRQ(err);
err = VecAssemblyEnd(q2); CHKERRQ(err);
err = VecDuplicate(q1, &r1); CHKERRQ(err);
err = VecDuplicate(q1, &t1_1); CHKERRQ(err);
err = VecDuplicate(q1, &t1_2); CHKERRQ(err);
err = VecDuplicate(q1, &t1_3); CHKERRQ(err);
err = VecDuplicate(q1, &t1_4); CHKERRQ(err);
err = VecDuplicate(q1, &t1_5); CHKERRQ(err);
err = VecDuplicate(q2, &r2); CHKERRQ(err);
err = VecDuplicate(q2, &q_tilda); CHKERRQ(err);
err = VecDuplicate(q2, &t2_1); CHKERRQ(err);
err = VecDuplicate(q2, &t2_2); CHKERRQ(err);
err = VecDuplicate(q2, &t2_3); CHKERRQ(err);
// Start matrix-vec multiplications
err = MatMult(invL1, q1, t1_1); CHKERRQ(err);
err = MatMult(invU1, t1_1, t1_2); CHKERRQ(err);
err = MatMult(H21, t1_2, t2_1); CHKERRQ(err);
err = VecAXPBYPCZ(q_tilda, 1.0, -1.0, 0.0, q2, t2_1); CHKERRQ(err);
err = MatMult(invL2, q_tilda, t2_2); CHKERRQ(err);
err = MatMult(invU2, t2_2, r2); CHKERRQ(err);
err = MatMult(H12, r2, t1_3); CHKERRQ(err);
err = VecAXPBYPCZ(t1_4, 1.0, -1.0, 0.0, q1, t1_3); CHKERRQ(err);
err = MatMult(invL1, t1_4, t1_5); CHKERRQ(err);
err = MatMult(invU1, t1_5, r1); CHKERRQ(err);
//err = printVecSum(r1);
//err = VecView(r2, PETSC_VIEWER_STDOUT_WORLD);
// Concatenate r1 and r2
err = VecMerge(r1, r2, r); CHKERRQ(err);
err = VecScale(r, c); CHKERRQ(err);
//err = VecView(r, PETSC_VIEWER_STDOUT_WORLD);
//err = VecDuplicate(r, &or); CHKERRQ(err);
err = VecReorder(r, order, or); CHKERRQ(err);
//err = VecView(or, PETSC_VIEWER_STDOUT_WORLD);
err = VecDestroy(&r); CHKERRQ(err);
err = VecDestroy(&r1); CHKERRQ(err);
err = VecDestroy(&q1); CHKERRQ(err);
err = VecDestroy(&t1_1); CHKERRQ(err);
err = VecDestroy(&t1_2); CHKERRQ(err);
err = VecDestroy(&t1_3); CHKERRQ(err);
err = VecDestroy(&t1_4); CHKERRQ(err);
err = VecDestroy(&t1_5); CHKERRQ(err);
err = VecDestroy(&r2); CHKERRQ(err);
err = VecDestroy(&q2); CHKERRQ(err);
err = VecDestroy(&q_tilda); CHKERRQ(err);
err = VecDestroy(&t2_1); CHKERRQ(err);
err = VecDestroy(&t2_2); CHKERRQ(err);
err = VecDestroy(&t2_3); CHKERRQ(err);
return err;
}
PetscErrorCode BearQueryMat(PetscInt s, PetscScalar c, Mat invL1, Mat invU1, Mat invL2, Mat invU2, Mat H12, Mat H21, Vec order){
PetscErrorCode err;
PetscInt n1, n2, n, M, N;
PetscInt oseed;
PetscScalar val, one = 1.0;
PetscMPIInt size;
PetscLogDouble tic, toc;
Mat r = NULL;
Mat r1 = NULL, q1 = NULL, t1_1 = NULL, t1_2 = NULL, t1_3 = NULL, t1_4 = NULL, t1_5 = NULL; // dimension: n1
Mat r2 = NULL, q2 = NULL, q_tilda = NULL, t2_1 = NULL, t2_2 = NULL, t2_3 = NULL; // dimension: n2_idx
Vec vr=NULL, vr1=NULL, vr2=NULL;
PetscInt col = 0;
err = MPI_Comm_size(PETSC_COMM_WORLD, &size); CHKERRQ(err);
err = MatGetSize(H12, &n1, &n2); CHKERRQ(err);
n = n1 + n2;
err = PetscPrintf(PETSC_COMM_WORLD, "n1: %d, n2: %d\n", n1, n2); CHKERRQ(err);
err = MatCreateAIJ(PETSC_COMM_WORLD, PETSC_DECIDE, 1, n, size, 1, NULL, 1, NULL, &r); CHKERRQ(err);
err = MatCreateAIJ(PETSC_COMM_WORLD, PETSC_DECIDE, 1, n1, size, 1, NULL, 1, NULL, &q1); CHKERRQ(err);
err = MatCreateAIJ(PETSC_COMM_WORLD, PETSC_DECIDE, 1, n2, size, 1, NULL, 1, NULL, &q2); CHKERRQ(err);
// err = MatCreate(PETSC_COMM_WORLD, &q2); CHKERRQ(err);
// err = MatSetSizes(q2, PETSC_DECIDE, PETSC_DECIDE, n2, 1); CHKERRQ(err);
// err = MatSetType(q2, MATAIJ); CHKERRQ(err);
// err = MatSetUp(q2);
s = s - 1; // shift -1 for zero-based index
err = VecGetValues(order, 1, &s, &val); CHKERRQ(err);
oseed = (PetscInt) val;
// err = PetscPrintf(PETSC_COMM_WORLD, "Given seed: %d, Reorered seed: %d (0 ~ n-1)\n", s, oseed); CHKERRQ(err);
if(oseed < n1){
//err = MatSetValues(q1, 1, &oseed, 1, &col, &one, INSERT_VALUES); CHKERRQ(err);
err = MatSetValue(q1, oseed, col, one, INSERT_VALUES); CHKERRQ(err);
}else{
oseed = oseed - n1;
//err = MatSetValues(q2, 1, &oseed, 1, &col, &one, INSERT_VALUES); CHKERRQ(err);
err = MatSetValue(q2, oseed, col, one, INSERT_VALUES); CHKERRQ(err);
//err = printVecSum(q2);
}
err = MatAssemblyBegin(q1, MAT_FINAL_ASSEMBLY); CHKERRQ(err);
err = MatAssemblyEnd(q1, MAT_FINAL_ASSEMBLY); CHKERRQ(err);
err = MatAssemblyBegin(q2, MAT_FINAL_ASSEMBLY); CHKERRQ(err);
err = MatAssemblyEnd(q2, MAT_FINAL_ASSEMBLY); CHKERRQ(err);
err = printMatInfo("q1", q1);
err = printMatInfo("q2", q2);
//err = MatView(q1, PETSC_VIEWER_STDOUT_WORLD);
//err = MatView(q2, PETSC_VIEWER_STDOUT_WORLD);
err = MatDuplicate(q1, MAT_DO_NOT_COPY_VALUES, &r1); CHKERRQ(err);
err = MatDuplicate(q1, MAT_DO_NOT_COPY_VALUES, &t1_1); CHKERRQ(err);
err = MatDuplicate(q1, MAT_DO_NOT_COPY_VALUES, &t1_2); CHKERRQ(err);
err = MatDuplicate(q1, MAT_DO_NOT_COPY_VALUES, &t1_3); CHKERRQ(err);
err = MatDuplicate(q1, MAT_DO_NOT_COPY_VALUES, &t1_4); CHKERRQ(err);
err = MatDuplicate(q1, MAT_DO_NOT_COPY_VALUES, &t1_5); CHKERRQ(err);
err = MatDuplicate(q2, MAT_DO_NOT_COPY_VALUES, &r2); CHKERRQ(err);
err = MatDuplicate(q2, MAT_DO_NOT_COPY_VALUES, &q_tilda); CHKERRQ(err);
err = MatDuplicate(q2, MAT_DO_NOT_COPY_VALUES, &t2_1); CHKERRQ(err);
err = MatDuplicate(q2, MAT_DO_NOT_COPY_VALUES, &t2_2); CHKERRQ(err);
err = MatDuplicate(q2, MAT_DO_NOT_COPY_VALUES, &t2_3); CHKERRQ(err);
err = MatMatMult(invL1, q1, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &t1_1); CHKERRQ(err);
err = MatMatMult(invU1, t1_1, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &t1_2); CHKERRQ(err);
err = MatMatMult(H21, t1_2, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &t2_1); CHKERRQ(err);
err = MatScale(t2_1, -1.0); CHKERRQ(err);
err = MatAXPY(t2_1, 1.0, q2, DIFFERENT_NONZERO_PATTERN); CHKERRQ(err);
//MatView(t1_1, PETSC_VIEWER_STDOUT_WORLD);
err = PetscTime(&tic);
err = MatMatMult(invL2, t2_1, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &t2_2); CHKERRQ(err);
err = MatMatMult(invU2, t2_2, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &r2); CHKERRQ(err);
err = PetscTime(&toc);
err = PetscPrintf(PETSC_COMM_WORLD, "running time: %f sec\n", toc-tic); CHKERRQ(err);
err = MatMatMult(H12, r2, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &t1_3); CHKERRQ(err);
err = MatScale(t1_3, -1.0); CHKERRQ(err);
err = MatAXPY(t1_3, 1.0, q1, DIFFERENT_NONZERO_PATTERN); CHKERRQ(err);
err = MatMatMult(invL1, t1_3, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &t1_5); CHKERRQ(err);
err = MatMatMult(invU1, t1_5, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &r1); CHKERRQ(err);
//MatView(r1, PETSC_VIEWER_STDOUT_WORLD);
MatGetSize(r1, &M, &N);
PetscPrintf(PETSC_COMM_WORLD, "%d %d\n", M, N);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n1, &vr1);
err = MatGetColumnVector(r1, vr1, 0);
err = VecCreateMPI(PETSC_COMM_WORLD, PETSC_DECIDE, n2, &vr2);
err = MatGetColumnVector(r2, vr2, 0);
err = printMatInfo("r2", r2);
/*
// Start matrix-vec multiplications
err = MatMult(invU2, t2_2, r2); CHKERRQ(err);
err = MatMult(H12, r2, t1_3); CHKERRQ(err);
err = VecAXPBYPCZ(t1_4, 1.0, -1.0, 0.0, q1, t1_3); CHKERRQ(err);
err = MatMult(invL1, t1_4, t1_5); CHKERRQ(err);
err = MatMult(invU1, t1_5, r1); CHKERRQ(err);
//err = printVecSum(r1);
//err = VecView(r2, PETSC_VIEWER_STDOUT_WORLD);
// Concatenate r1 and r2
err = VecMerge(r1, r2, r); CHKERRQ(err);
err = VecScale(r, c); CHKERRQ(err);
//err = VecView(r, PETSC_VIEWER_STDOUT_WORLD);
//err = VecDuplicate(r, &or); CHKERRQ(err);
err = VecReorder(r, order, or); CHKERRQ(err);
//err = VecView(or, PETSC_VIEWER_STDOUT_WORLD);
*/
err = MatDestroy(&r); CHKERRQ(err);
err = MatDestroy(&r1); CHKERRQ(err);
err = MatDestroy(&q1); CHKERRQ(err);
err = MatDestroy(&t1_1); CHKERRQ(err);
err = MatDestroy(&t1_2); CHKERRQ(err);
err = MatDestroy(&t1_3); CHKERRQ(err);
err = MatDestroy(&t1_4); CHKERRQ(err);
err = MatDestroy(&t1_5); CHKERRQ(err);
err = MatDestroy(&r2); CHKERRQ(err);
err = MatDestroy(&q2); CHKERRQ(err);
err = MatDestroy(&q_tilda); CHKERRQ(err);
err = MatDestroy(&t2_1); CHKERRQ(err);
err = MatDestroy(&t2_2); CHKERRQ(err);
err = MatDestroy(&t2_3); CHKERRQ(err);
return err;
}
int
SparseGp_logLikeGrad (SparseGp *gp, HyperParam hp, int lengthInd, double *logLikeGrad)
{
PetscErrorCode ierr;
(void) ierr;
/* compute t' inv(K) (dKdt inv(K) t) */
/* 1. solve inv(K) t */
PetscInt N = gp->trainLabels->size;
Vec invKt;
ierr = petsc_util_createVec (&invKt, gp->nlocal, N);
SparseGp_solve (gp, gp->_trainLabels, &invKt);
/* 2. multiply dKdt by invKt */
Vec dKdtInvKt;
ierr = petsc_util_createVec (&dKdtInvKt, gp->nlocal, N);
SparseGp_KGradient (gp, hp, lengthInd);
MatMult (gp->_KGradient, invKt, dKdtInvKt);
/* 3. solve invK (vector from step 2.) */
Vec invKDkdtInvKt;
ierr = petsc_util_createVec (&invKDkdtInvKt, gp->nlocal, N);
SparseGp_solve (gp, dKdtInvKt, &invKDkdtInvKt);
/* 4. compute inner product */
double dotProd;
VecDot (gp->_trainLabels, invKDkdtInvKt, &dotProd);
/* compute trace (invK dKdt) */
/* 1. for each column in dKdt, solve */
double myTrace = 0;
Vec col, solution;
petsc_util_createVec (&col, gp->nlocal, N);
petsc_util_createVec (&solution, gp->nlocal, N);
for (int i=0; i<N; i++) {
/* IOU_ROOT_PRINT ("%d --- %d\n", i, N); */
int row = i;
ierr = MatGetColumnVector (gp->_KGradient, col, row);
/* IOU_ROOT_PRINT ("solving...\n"); */
SparseGp_solve (gp, col, &solution);
/* IOU_ROOT_PRINT ("solved\n"); */
if (row < gp->rend && row >= gp->rstart) {
double ii;
VecGetValues (solution, 1, &row, &ii);
myTrace += ii;
}
}
/* gather all trace terms */
int numProcs;
MPI_Comm_size (PETSC_COMM_WORLD, &numProcs);
double diag[numProcs];
int ret = MPI_Allgather (&myTrace, 1, MPI_DOUBLE, diag, 1, MPI_DOUBLE, PETSC_COMM_WORLD);
(void) ret;
double trace = 0;
for (int i=0; i<numProcs; i++)
trace += diag[i];
/* IOU_ROOT_PRINT ("Cleaning up...\n"); */
/* clean up */
VecDestroy (&invKt);
VecDestroy (&dKdtInvKt);
VecDestroy (&invKDkdtInvKt);
VecDestroy (&col);
VecDestroy (&solution);
/* IOU_ROOT_PRINT ("Done cleaning up...\n"); */
*logLikeGrad = 0.5*dotProd + 0.5*trace;
return EXIT_SUCCESS;
}
int main(int argc,char **args)
{
const ptrdiff_t N0=2056,N1=2056;
fftw_plan bplan,fplan;
fftw_complex *out;
double *in1,*in2;
ptrdiff_t alloc_local,local_n0,local_0_start;
ptrdiff_t local_n1,local_1_start;
PetscInt i,j;
PetscMPIInt size,rank;
int n,N,N_factor,NM;
PetscScalar one=2.0,zero=0.5;
PetscScalar two=4.0,three=8.0,four=16.0;
PetscScalar a,*x_arr,*y_arr,*z_arr;
PetscReal enorm;
Vec fin,fout,fout1;
Vec ini,final;
PetscRandom rnd;
PetscErrorCode ierr;
PetscInt *indx3,tempindx,low,*indx4,tempindx1;
ierr = PetscInitialize(&argc,&args,(char*)0,help);if (ierr) return ierr;
ierr = MPI_Comm_size(PETSC_COMM_WORLD, &size);CHKERRQ(ierr);
ierr = MPI_Comm_rank(PETSC_COMM_WORLD, &rank);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_WORLD,&rnd);CHKERRQ(ierr);
alloc_local = fftw_mpi_local_size_2d_transposed(N0,N1/2+1,PETSC_COMM_WORLD,&local_n0,&local_0_start,&local_n1,&local_1_start);
#if defined(DEBUGGING)
printf("The value alloc_local is %ld from process %d\n",alloc_local,rank);
printf("The value local_n0 is %ld from process %d\n",local_n0,rank);
printf("The value local_0_start is %ld from process %d\n",local_0_start,rank);
/* printf("The value local_n1 is %ld from process %d\n",local_n1,rank); */
/* printf("The value local_1_start is %ld from process %d\n",local_1_start,rank); */
/* printf("The value local_n0 is %ld from process %d\n",local_n0,rank); */
#endif
/* Allocate space for input and output arrays */
in1=(double*)fftw_malloc(sizeof(double)*alloc_local*2);
in2=(double*)fftw_malloc(sizeof(double)*alloc_local*2);
out=(fftw_complex*)fftw_malloc(sizeof(fftw_complex)*alloc_local);
N = 2*N0*(N1/2+1);
N_factor = N0*N1;
n = 2*local_n0*(N1/2+1);
/* printf("The value N is %d from process %d\n",N,rank); */
/* printf("The value n is %d from process %d\n",n,rank); */
/* printf("The value n1 is %d from process %d\n",n1,rank);*/
/* Creating data vector and accompanying array with VeccreateMPIWithArray */
ierr = VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)in1,&fin);CHKERRQ(ierr);
ierr = VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)out,&fout);CHKERRQ(ierr);
ierr = VecCreateMPIWithArray(PETSC_COMM_WORLD,1,n,N,(PetscScalar*)in2,&fout1);CHKERRQ(ierr);
/* Set the vector with random data */
ierr = VecSet(fin,zero);CHKERRQ(ierr);
/* for (i=0;i<N0*N1;i++) */
/* { */
/* VecSetValues(fin,1,&i,&one,INSERT_VALUES); */
/* } */
/* VecSet(fin,one); */
i =0;
ierr = VecSetValues(fin,1,&i,&one,INSERT_VALUES);CHKERRQ(ierr);
i =1;
ierr = VecSetValues(fin,1,&i,&two,INSERT_VALUES);CHKERRQ(ierr);
i =4;
ierr = VecSetValues(fin,1,&i,&three,INSERT_VALUES);CHKERRQ(ierr);
i =5;
ierr = VecSetValues(fin,1,&i,&four,INSERT_VALUES);CHKERRQ(ierr);
ierr = VecAssemblyBegin(fin);CHKERRQ(ierr);
ierr = VecAssemblyEnd(fin);CHKERRQ(ierr);
ierr = VecSet(fout,zero);CHKERRQ(ierr);
ierr = VecSet(fout1,zero);CHKERRQ(ierr);
/* Get the meaningful portion of array */
ierr = VecGetArray(fin,&x_arr);CHKERRQ(ierr);
ierr = VecGetArray(fout1,&z_arr);CHKERRQ(ierr);
ierr = VecGetArray(fout,&y_arr);CHKERRQ(ierr);
fplan=fftw_mpi_plan_dft_r2c_2d(N0,N1,(double*)x_arr,(fftw_complex*)y_arr,PETSC_COMM_WORLD,FFTW_ESTIMATE);
bplan=fftw_mpi_plan_dft_c2r_2d(N0,N1,(fftw_complex*)y_arr,(double*)z_arr,PETSC_COMM_WORLD,FFTW_ESTIMATE);
fftw_execute(fplan);
fftw_execute(bplan);
ierr = VecRestoreArray(fin,&x_arr);
ierr = VecRestoreArray(fout1,&z_arr);
ierr = VecRestoreArray(fout,&y_arr);
/* VecView(fin,PETSC_VIEWER_STDOUT_WORLD); */
ierr = VecCreate(PETSC_COMM_WORLD,&ini);CHKERRQ(ierr);
ierr = VecCreate(PETSC_COMM_WORLD,&final);CHKERRQ(ierr);
ierr = VecSetSizes(ini,local_n0*N1,N0*N1);CHKERRQ(ierr);
ierr = VecSetSizes(final,local_n0*N1,N0*N1);CHKERRQ(ierr);
ierr = VecSetFromOptions(ini);CHKERRQ(ierr);
ierr = VecSetFromOptions(final);CHKERRQ(ierr);
if (N1%2==0) {
NM = N1+2;
} else {
NM = N1+1;
}
/*printf("The Value of NM is %d",NM); */
ierr = VecGetOwnershipRange(fin,&low,NULL);
/*printf("The local index is %d from %d\n",low,rank); */
ierr = PetscMalloc1(local_n0*N1,&indx3);
ierr = PetscMalloc1(local_n0*N1,&indx4);
for (i=0;i<local_n0;i++) {
for (j=0;j<N1;j++) {
tempindx = i*N1 + j;
tempindx1 = i*NM + j;
indx3[tempindx]=local_0_start*N1+tempindx;
indx4[tempindx]=low+tempindx1;
/* printf("index3 %d from proc %d is \n",indx3[tempindx],rank); */
/* printf("index4 %d from proc %d is \n",indx4[tempindx],rank); */
}
}
ierr = PetscMalloc2(local_n0*N1,&x_arr,local_n0*N1,&y_arr);CHKERRQ(ierr); /* arr must be allocated for VecGetValues() */
ierr = VecGetValues(fin,local_n0*N1,indx4,(PetscScalar*)x_arr);CHKERRQ(ierr);
ierr = VecSetValues(ini,local_n0*N1,indx3,x_arr,INSERT_VALUES);CHKERRQ(ierr);
ierr = VecAssemblyBegin(ini);CHKERRQ(ierr);
ierr = VecAssemblyEnd(ini);CHKERRQ(ierr);
ierr = VecGetValues(fout1,local_n0*N1,indx4,y_arr);
ierr = VecSetValues(final,local_n0*N1,indx3,y_arr,INSERT_VALUES);
ierr = VecAssemblyBegin(final);
ierr = VecAssemblyEnd(final);
ierr = PetscFree2(x_arr,y_arr);CHKERRQ(ierr);
/*
VecScatter vecscat;
IS indx1,indx2;
for (i=0;i<N0;i++) {
indx = i*NM;
ISCreateStride(PETSC_COMM_WORLD,N1,indx,1,&indx1);
indx = i*N1;
ISCreateStride(PETSC_COMM_WORLD,N1,indx,1,&indx2);
VecScatterCreate(fin,indx1,ini,indx2,&vecscat);
VecScatterBegin(vecscat,fin,ini,INSERT_VALUES,SCATTER_FORWARD);
VecScatterEnd(vecscat,fin,ini,INSERT_VALUES,SCATTER_FORWARD);
VecScatterBegin(vecscat,fout1,final,INSERT_VALUES,SCATTER_FORWARD);
VecScatterEnd(vecscat,fout1,final,INSERT_VALUES,SCATTER_FORWARD);
}
*/
a = 1.0/(PetscReal)N_factor;
ierr = VecScale(fout1,a);CHKERRQ(ierr);
ierr = VecScale(final,a);CHKERRQ(ierr);
/* VecView(ini,PETSC_VIEWER_STDOUT_WORLD); */
/* VecView(final,PETSC_VIEWER_STDOUT_WORLD); */
ierr = VecAXPY(final,-1.0,ini);CHKERRQ(ierr);
ierr = VecNorm(final,NORM_1,&enorm);CHKERRQ(ierr);
if (enorm > 1.e-10) {
ierr = PetscPrintf(PETSC_COMM_WORLD," Error norm of |x - z| = %e\n",enorm);CHKERRQ(ierr);
}
/* Execute fftw with function fftw_execute and destory it after execution */
fftw_destroy_plan(fplan);
fftw_destroy_plan(bplan);
fftw_free(in1); ierr = VecDestroy(&fin);CHKERRQ(ierr);
fftw_free(out); ierr = VecDestroy(&fout);CHKERRQ(ierr);
fftw_free(in2); ierr = VecDestroy(&fout1);CHKERRQ(ierr);
ierr = VecDestroy(&ini);CHKERRQ(ierr);
ierr = VecDestroy(&final);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&rnd);CHKERRQ(ierr);
ierr = PetscFree(indx3);CHKERRQ(ierr);
ierr = PetscFree(indx4);CHKERRQ(ierr);
ierr = PetscFinalize();
return ierr;
}
PetscErrorCode PCGAMGProlongator_Classical_Direct(PC pc, const Mat A, const Mat G, PetscCoarsenData *agg_lists,Mat *P)
{
PetscErrorCode ierr;
MPI_Comm comm;
PetscReal *Amax_pos,*Amax_neg;
Mat lA,gA; /* on and off diagonal matrices */
PetscInt fn; /* fine local blocked sizes */
PetscInt cn; /* coarse local blocked sizes */
PetscInt gn; /* size of the off-diagonal fine vector */
PetscInt fs,fe; /* fine (row) ownership range*/
PetscInt cs,ce; /* coarse (column) ownership range */
PetscInt i,j; /* indices! */
PetscBool iscoarse; /* flag for determining if a node is coarse */
PetscInt *lcid,*gcid; /* on and off-processor coarse unknown IDs */
PetscInt *lsparse,*gsparse; /* on and off-processor sparsity patterns for prolongator */
PetscScalar pij;
const PetscScalar *rval;
const PetscInt *rcol;
PetscScalar g_pos,g_neg,a_pos,a_neg,diag,invdiag,alpha,beta;
Vec F; /* vec of coarse size */
Vec C; /* vec of fine size */
Vec gF; /* vec of off-diagonal fine size */
MatType mtype;
PetscInt c_indx;
PetscScalar c_scalar;
PetscInt ncols,col;
PetscInt row_f,row_c;
PetscInt cmax=0,idx;
PetscScalar *pvals;
PetscInt *pcols;
PC_MG *mg = (PC_MG*)pc->data;
PC_GAMG *gamg = (PC_GAMG*)mg->innerctx;
PetscFunctionBegin;
comm = ((PetscObject)pc)->comm;
ierr = MatGetOwnershipRange(A,&fs,&fe); CHKERRQ(ierr);
fn = (fe - fs);
ierr = MatGetVecs(A,&F,NULL);CHKERRQ(ierr);
/* get the number of local unknowns and the indices of the local unknowns */
ierr = PetscMalloc(sizeof(PetscInt)*fn,&lsparse);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscInt)*fn,&gsparse);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscInt)*fn,&lcid);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscReal)*fn,&Amax_pos);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscReal)*fn,&Amax_neg);CHKERRQ(ierr);
/* count the number of coarse unknowns */
cn = 0;
for (i=0;i<fn;i++) {
/* filter out singletons */
ierr = PetscCDEmptyAt(agg_lists,i,&iscoarse); CHKERRQ(ierr);
lcid[i] = -1;
if (!iscoarse) {
cn++;
}
}
/* create the coarse vector */
ierr = VecCreateMPI(comm,cn,PETSC_DECIDE,&C);CHKERRQ(ierr);
ierr = VecGetOwnershipRange(C,&cs,&ce);CHKERRQ(ierr);
/* construct a global vector indicating the global indices of the coarse unknowns */
cn = 0;
for (i=0;i<fn;i++) {
ierr = PetscCDEmptyAt(agg_lists,i,&iscoarse); CHKERRQ(ierr);
if (!iscoarse) {
lcid[i] = cs+cn;
cn++;
} else {
lcid[i] = -1;
}
*((PetscInt *)&c_scalar) = lcid[i];
c_indx = fs+i;
ierr = VecSetValues(F,1,&c_indx,&c_scalar,INSERT_VALUES);CHKERRQ(ierr);
}
ierr = VecAssemblyBegin(F);CHKERRQ(ierr);
ierr = VecAssemblyEnd(F);CHKERRQ(ierr);
/* determine the biggest off-diagonal entries in each row */
for (i=fs;i<fe;i++) {
Amax_pos[i-fs] = 0.;
Amax_neg[i-fs] = 0.;
ierr = MatGetRow(A,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
for(j=0;j<ncols;j++){
if ((PetscRealPart(-rval[j]) > Amax_neg[i-fs]) && i != rcol[j]) Amax_neg[i-fs] = PetscAbsScalar(rval[j]);
if ((PetscRealPart(rval[j]) > Amax_pos[i-fs]) && i != rcol[j]) Amax_pos[i-fs] = PetscAbsScalar(rval[j]);
}
if (ncols > cmax) cmax = ncols;
ierr = MatRestoreRow(A,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
}
ierr = PetscMalloc(sizeof(PetscInt)*cmax,&pcols);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscScalar)*cmax,&pvals);CHKERRQ(ierr);
/* split the operator into two */
ierr = PCGAMGClassicalGraphSplitting_Private(A,&lA,&gA);CHKERRQ(ierr);
/* scatter to the ghost vector */
ierr = PCGAMGClassicalCreateGhostVector_Private(A,&gF,NULL);CHKERRQ(ierr);
ierr = PCGAMGClassicalGhost_Private(A,F,gF);CHKERRQ(ierr);
if (gA) {
ierr = VecGetSize(gF,&gn);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscInt)*gn,&gcid);CHKERRQ(ierr);
for (i=0;i<gn;i++) {
ierr = VecGetValues(gF,1,&i,&c_scalar);CHKERRQ(ierr);
gcid[i] = *((PetscInt *)&c_scalar);
}
}
ierr = VecDestroy(&F);CHKERRQ(ierr);
ierr = VecDestroy(&gF);CHKERRQ(ierr);
ierr = VecDestroy(&C);CHKERRQ(ierr);
/* count the on and off processor sparsity patterns for the prolongator */
for (i=0;i<fn;i++) {
/* on */
lsparse[i] = 0;
gsparse[i] = 0;
if (lcid[i] >= 0) {
lsparse[i] = 1;
gsparse[i] = 0;
} else {
ierr = MatGetRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
for (j = 0;j < ncols;j++) {
col = rcol[j];
if (lcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) {
lsparse[i] += 1;
}
}
ierr = MatRestoreRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
/* off */
if (gA) {
ierr = MatGetRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
for (j = 0; j < ncols; j++) {
col = rcol[j];
if (gcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) {
gsparse[i] += 1;
}
}
ierr = MatRestoreRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
}
}
}
/* preallocate and create the prolongator */
ierr = MatCreate(comm,P); CHKERRQ(ierr);
ierr = MatGetType(G,&mtype);CHKERRQ(ierr);
ierr = MatSetType(*P,mtype);CHKERRQ(ierr);
ierr = MatSetSizes(*P,fn,cn,PETSC_DETERMINE,PETSC_DETERMINE);CHKERRQ(ierr);
ierr = MatMPIAIJSetPreallocation(*P,0,lsparse,0,gsparse);CHKERRQ(ierr);
ierr = MatSeqAIJSetPreallocation(*P,0,lsparse);CHKERRQ(ierr);
/* loop over local fine nodes -- get the diagonal, the sum of positive and negative strong and weak weights, and set up the row */
for (i = 0;i < fn;i++) {
/* determine on or off */
row_f = i + fs;
row_c = lcid[i];
if (row_c >= 0) {
pij = 1.;
ierr = MatSetValues(*P,1,&row_f,1,&row_c,&pij,INSERT_VALUES);CHKERRQ(ierr);
} else {
g_pos = 0.;
g_neg = 0.;
a_pos = 0.;
a_neg = 0.;
diag = 0.;
/* local connections */
ierr = MatGetRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
for (j = 0; j < ncols; j++) {
col = rcol[j];
if (lcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) {
if (PetscRealPart(rval[j]) > 0.) {
g_pos += rval[j];
} else {
g_neg += rval[j];
}
}
if (col != i) {
if (PetscRealPart(rval[j]) > 0.) {
a_pos += rval[j];
} else {
a_neg += rval[j];
}
} else {
diag = rval[j];
}
}
ierr = MatRestoreRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
/* ghosted connections */
if (gA) {
ierr = MatGetRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
for (j = 0; j < ncols; j++) {
col = rcol[j];
if (gcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) {
if (PetscRealPart(rval[j]) > 0.) {
g_pos += rval[j];
} else {
g_neg += rval[j];
}
}
if (PetscRealPart(rval[j]) > 0.) {
a_pos += rval[j];
} else {
a_neg += rval[j];
}
}
ierr = MatRestoreRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
}
if (g_neg == 0.) {
alpha = 0.;
} else {
alpha = -a_neg/g_neg;
}
if (g_pos == 0.) {
diag += a_pos;
beta = 0.;
} else {
beta = -a_pos/g_pos;
}
if (diag == 0.) {
invdiag = 0.;
} else invdiag = 1. / diag;
/* on */
ierr = MatGetRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
idx = 0;
for (j = 0;j < ncols;j++) {
col = rcol[j];
if (lcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) {
row_f = i + fs;
row_c = lcid[col];
/* set the values for on-processor ones */
if (PetscRealPart(rval[j]) < 0.) {
pij = rval[j]*alpha*invdiag;
} else {
pij = rval[j]*beta*invdiag;
}
if (PetscAbsScalar(pij) != 0.) {
pvals[idx] = pij;
pcols[idx] = row_c;
idx++;
}
}
}
ierr = MatRestoreRow(lA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
/* off */
if (gA) {
ierr = MatGetRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
for (j = 0; j < ncols; j++) {
col = rcol[j];
if (gcid[col] >= 0 && (PetscRealPart(rval[j]) > gamg->threshold*Amax_pos[i] || PetscRealPart(-rval[j]) > gamg->threshold*Amax_neg[i])) {
row_f = i + fs;
row_c = gcid[col];
/* set the values for on-processor ones */
if (PetscRealPart(rval[j]) < 0.) {
pij = rval[j]*alpha*invdiag;
} else {
pij = rval[j]*beta*invdiag;
}
if (PetscAbsScalar(pij) != 0.) {
pvals[idx] = pij;
pcols[idx] = row_c;
idx++;
}
}
}
ierr = MatRestoreRow(gA,i,&ncols,&rcol,&rval);CHKERRQ(ierr);
}
ierr = MatSetValues(*P,1,&row_f,idx,pcols,pvals,INSERT_VALUES);CHKERRQ(ierr);
}
}
ierr = MatAssemblyBegin(*P, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(*P, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = PetscFree(lsparse);CHKERRQ(ierr);
ierr = PetscFree(gsparse);CHKERRQ(ierr);
ierr = PetscFree(pcols);CHKERRQ(ierr);
ierr = PetscFree(pvals);CHKERRQ(ierr);
ierr = PetscFree(Amax_pos);CHKERRQ(ierr);
ierr = PetscFree(Amax_neg);CHKERRQ(ierr);
ierr = PetscFree(lcid);CHKERRQ(ierr);
if (gA) {ierr = PetscFree(gcid);CHKERRQ(ierr);}
PetscFunctionReturn(0);
}
