/*@
MatSchurComplementComputeExplicitOperator - Compute the Schur complement matrix explicitly
Collective on Mat
Input Parameter:
. M - the matrix obtained with MatCreateSchurComplement()
Output Parameter:
. S - the Schur complement matrix
Note: This can be expensive, so it is mainly for testing
Level: advanced
.seealso: MatCreateSchurComplement(), MatSchurComplementUpdate()
@*/
PetscErrorCode MatSchurComplementComputeExplicitOperator(Mat M, Mat *S)
{
Mat B, C, D;
KSP ksp;
PC pc;
PetscBool isLU, isILU;
PetscReal fill = 2.0;
PetscErrorCode ierr;
PetscFunctionBegin;
ierr = MatSchurComplementGetSubMatrices(M, NULL, NULL, &B, &C, &D);CHKERRQ(ierr);
ierr = MatSchurComplementGetKSP(M, &ksp);CHKERRQ(ierr);
ierr = KSPGetPC(ksp, &pc);CHKERRQ(ierr);
ierr = PetscObjectTypeCompare((PetscObject) pc, PCLU, &isLU);CHKERRQ(ierr);
ierr = PetscObjectTypeCompare((PetscObject) pc, PCILU, &isILU);CHKERRQ(ierr);
if (isLU || isILU) {
Mat fact, Bd, AinvB, AinvBd;
PetscReal eps = 1.0e-10;
/* This can be sped up for banded LU */
ierr = KSPSetUp(ksp);CHKERRQ(ierr);
ierr = PCFactorGetMatrix(pc, &fact);CHKERRQ(ierr);
ierr = MatConvert(B, MATDENSE, MAT_INITIAL_MATRIX, &Bd);CHKERRQ(ierr);
ierr = MatDuplicate(Bd, MAT_DO_NOT_COPY_VALUES, &AinvBd);CHKERRQ(ierr);
ierr = MatMatSolve(fact, Bd, AinvBd);CHKERRQ(ierr);
ierr = MatDestroy(&Bd);CHKERRQ(ierr);
ierr = MatChop(AinvBd, eps);CHKERRQ(ierr);
ierr = MatConvert(AinvBd, MATAIJ, MAT_INITIAL_MATRIX, &AinvB);CHKERRQ(ierr);
ierr = MatDestroy(&AinvBd);CHKERRQ(ierr);
ierr = MatMatMult(C, AinvB, MAT_INITIAL_MATRIX, fill, S);CHKERRQ(ierr);
ierr = MatDestroy(&AinvB);CHKERRQ(ierr);
} else {
Mat Ainvd, Ainv;
ierr = PCComputeExplicitOperator(pc, &Ainvd);CHKERRQ(ierr);
ierr = MatConvert(Ainvd, MATAIJ, MAT_INITIAL_MATRIX, &Ainv);CHKERRQ(ierr);
ierr = MatDestroy(&Ainvd);CHKERRQ(ierr);
#if 0
/* Symmetric version */
ierr = MatPtAP(Ainv, B, MAT_INITIAL_MATRIX, fill, S);CHKERRQ(ierr);
#else
/* Nonsymmetric version */
ierr = MatMatMatMult(C, Ainv, B, MAT_INITIAL_MATRIX, fill, S);CHKERRQ(ierr);
#endif
ierr = MatDestroy(&Ainv);CHKERRQ(ierr);
}
ierr = PetscViewerPushFormat(PETSC_VIEWER_STDOUT_WORLD, PETSC_VIEWER_ASCII_INFO);CHKERRQ(ierr);
ierr = MatView(*S, PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
ierr = PetscViewerPopFormat(PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
if (D) {
MatInfo info;
ierr = MatGetInfo(D, MAT_GLOBAL_SUM, &info);CHKERRQ(ierr);
if (info.nz_used) SETERRQ(PetscObjectComm((PetscObject) M), PETSC_ERR_SUP, "Not yet implemented");
}
PetscFunctionReturn(0);
}
/*@
MatSchurComplementComputeExplicitOperator - Compute the Schur complement matrix explicitly
Collective on Mat
Input Parameter:
. M - the matrix obtained with MatCreateSchurComplement()
Output Parameter:
. S - the Schur complement matrix
Note: This can be expensive, so it is mainly for testing
Level: advanced
.seealso: MatCreateSchurComplement(), MatSchurComplementUpdate()
@*/
PetscErrorCode MatSchurComplementComputeExplicitOperator(Mat M, Mat *S)
{
Mat B, C, D;
KSP ksp;
PC pc;
PetscBool isLU, isILU;
PetscReal fill = 2.0;
PetscErrorCode ierr;
PetscFunctionBegin;
ierr = MatSchurComplementGetSubMatrices(M, NULL, NULL, &B, &C, &D);CHKERRQ(ierr);
ierr = MatSchurComplementGetKSP(M, &ksp);CHKERRQ(ierr);
ierr = KSPGetPC(ksp, &pc);CHKERRQ(ierr);
ierr = PetscObjectTypeCompare((PetscObject) pc, PCLU, &isLU);CHKERRQ(ierr);
ierr = PetscObjectTypeCompare((PetscObject) pc, PCILU, &isILU);CHKERRQ(ierr);
if (isLU || isILU) {
Mat fact, Bd, AinvB, AinvBd;
PetscReal eps = 1.0e-10;
/* This can be sped up for banded LU */
ierr = KSPSetUp(ksp);CHKERRQ(ierr);
ierr = PCFactorGetMatrix(pc, &fact);CHKERRQ(ierr);
ierr = MatConvert(B, MATDENSE, MAT_INITIAL_MATRIX, &Bd);CHKERRQ(ierr);
ierr = MatDuplicate(Bd, MAT_DO_NOT_COPY_VALUES, &AinvBd);CHKERRQ(ierr);
ierr = MatMatSolve(fact, Bd, AinvBd);CHKERRQ(ierr);
ierr = MatDestroy(&Bd);CHKERRQ(ierr);
ierr = MatChop(AinvBd, eps);CHKERRQ(ierr);
ierr = MatConvert(AinvBd, MATAIJ, MAT_INITIAL_MATRIX, &AinvB);CHKERRQ(ierr);
ierr = MatDestroy(&AinvBd);CHKERRQ(ierr);
ierr = MatMatMult(C, AinvB, MAT_INITIAL_MATRIX, fill, S);CHKERRQ(ierr);
ierr = MatDestroy(&AinvB);CHKERRQ(ierr);
} else {
Mat Ainvd, Ainv;
ierr = PCComputeExplicitOperator(pc, &Ainvd);CHKERRQ(ierr);
ierr = MatConvert(Ainvd, MATAIJ, MAT_INITIAL_MATRIX, &Ainv);CHKERRQ(ierr);
ierr = MatDestroy(&Ainvd);CHKERRQ(ierr);
#if 0
/* Symmetric version */
ierr = MatPtAP(Ainv, B, MAT_INITIAL_MATRIX, fill, S);CHKERRQ(ierr);
#else
/* Nonsymmetric version */
ierr = MatMatMatMult(C, Ainv, B, MAT_INITIAL_MATRIX, fill, S);CHKERRQ(ierr);
#endif
ierr = MatDestroy(&Ainv);CHKERRQ(ierr);
}
if (D) {
ierr = MatAXPY(*S, -1.0, D, DIFFERENT_NONZERO_PATTERN);CHKERRQ(ierr);
}
ierr = MatScale(*S,-1.0);CHKERRQ(ierr);
PetscFunctionReturn(0);
}
static PetscErrorCode PCBDDCScalingSetUp_Deluxe_Private(PC pc)
{
PC_BDDC *pcbddc=(PC_BDDC*)pc->data;
PCBDDCDeluxeScaling deluxe_ctx=pcbddc->deluxe_ctx;
PCBDDCSubSchurs sub_schurs = pcbddc->sub_schurs;
PetscScalar *matdata,*matdata2;
PetscInt i,max_subset_size,cum,cum2;
const PetscInt *idxs;
PetscBool newsetup = PETSC_FALSE;
PetscErrorCode ierr;
PetscFunctionBegin;
if (!sub_schurs) SETERRQ(PetscObjectComm((PetscObject)pc),PETSC_ERR_PLIB,"Missing PCBDDCSubSchurs");
if (!sub_schurs->n_subs) PetscFunctionReturn(0);
/* Allocate arrays for subproblems */
if (!deluxe_ctx->seq_n) {
deluxe_ctx->seq_n = sub_schurs->n_subs;
ierr = PetscCalloc5(deluxe_ctx->seq_n,&deluxe_ctx->seq_scctx,deluxe_ctx->seq_n,&deluxe_ctx->seq_work1,deluxe_ctx->seq_n,&deluxe_ctx->seq_work2,deluxe_ctx->seq_n,&deluxe_ctx->seq_mat,deluxe_ctx->seq_n,&deluxe_ctx->seq_mat_inv_sum);CHKERRQ(ierr);
newsetup = PETSC_TRUE;
} else if (deluxe_ctx->seq_n != sub_schurs->n_subs) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Number of deluxe subproblems %D is different from the sub_schurs %D",deluxe_ctx->seq_n,sub_schurs->n_subs);
/* the change of basis is just a reference to sub_schurs->change (if any) */
deluxe_ctx->change = sub_schurs->change;
deluxe_ctx->change_with_qr = sub_schurs->change_with_qr;
/* Create objects for deluxe */
max_subset_size = 0;
for (i=0;i<sub_schurs->n_subs;i++) {
PetscInt subset_size;
ierr = ISGetLocalSize(sub_schurs->is_subs[i],&subset_size);CHKERRQ(ierr);
max_subset_size = PetscMax(subset_size,max_subset_size);
}
if (newsetup) {
ierr = PetscMalloc1(2*max_subset_size,&deluxe_ctx->workspace);CHKERRQ(ierr);
}
cum = cum2 = 0;
ierr = ISGetIndices(sub_schurs->is_Ej_all,&idxs);CHKERRQ(ierr);
ierr = MatSeqAIJGetArray(sub_schurs->S_Ej_all,&matdata);CHKERRQ(ierr);
ierr = MatSeqAIJGetArray(sub_schurs->sum_S_Ej_all,&matdata2);CHKERRQ(ierr);
for (i=0;i<deluxe_ctx->seq_n;i++) {
PetscInt subset_size;
ierr = ISGetLocalSize(sub_schurs->is_subs[i],&subset_size);CHKERRQ(ierr);
if (newsetup) {
IS sub;
/* work vectors */
ierr = VecCreateSeqWithArray(PETSC_COMM_SELF,1,subset_size,deluxe_ctx->workspace,&deluxe_ctx->seq_work1[i]);CHKERRQ(ierr);
ierr = VecCreateSeqWithArray(PETSC_COMM_SELF,1,subset_size,deluxe_ctx->workspace+subset_size,&deluxe_ctx->seq_work2[i]);CHKERRQ(ierr);
/* scatters */
ierr = ISCreateGeneral(PETSC_COMM_SELF,subset_size,idxs+cum,PETSC_COPY_VALUES,&sub);CHKERRQ(ierr);
ierr = VecScatterCreate(pcbddc->work_scaling,sub,deluxe_ctx->seq_work1[i],NULL,&deluxe_ctx->seq_scctx[i]);CHKERRQ(ierr);
ierr = ISDestroy(&sub);CHKERRQ(ierr);
}
/* S_E_j */
ierr = MatDestroy(&deluxe_ctx->seq_mat[i]);CHKERRQ(ierr);
ierr = MatCreateSeqDense(PETSC_COMM_SELF,subset_size,subset_size,matdata+cum2,&deluxe_ctx->seq_mat[i]);CHKERRQ(ierr);
/* \sum_k S^k_E_j */
ierr = MatDestroy(&deluxe_ctx->seq_mat_inv_sum[i]);CHKERRQ(ierr);
ierr = MatCreateSeqDense(PETSC_COMM_SELF,subset_size,subset_size,matdata2+cum2,&deluxe_ctx->seq_mat_inv_sum[i]);CHKERRQ(ierr);
ierr = MatSetOption(deluxe_ctx->seq_mat_inv_sum[i],MAT_SPD,sub_schurs->is_posdef);CHKERRQ(ierr);
ierr = MatSetOption(deluxe_ctx->seq_mat_inv_sum[i],MAT_HERMITIAN,sub_schurs->is_hermitian);CHKERRQ(ierr);
if (sub_schurs->is_hermitian) {
ierr = MatCholeskyFactor(deluxe_ctx->seq_mat_inv_sum[i],NULL,NULL);CHKERRQ(ierr);
} else {
ierr = MatLUFactor(deluxe_ctx->seq_mat_inv_sum[i],NULL,NULL,NULL);CHKERRQ(ierr);
}
if (pcbddc->deluxe_singlemat) {
Mat X,Y;
if (!sub_schurs->is_hermitian) {
ierr = MatTranspose(deluxe_ctx->seq_mat[i],MAT_INITIAL_MATRIX,&X);CHKERRQ(ierr);
} else {
ierr = PetscObjectReference((PetscObject)deluxe_ctx->seq_mat[i]);CHKERRQ(ierr);
X = deluxe_ctx->seq_mat[i];
}
ierr = MatDuplicate(X,MAT_DO_NOT_COPY_VALUES,&Y);CHKERRQ(ierr);
if (!sub_schurs->is_hermitian) {
ierr = PCBDDCMatTransposeMatSolve_SeqDense(deluxe_ctx->seq_mat_inv_sum[i],X,Y);CHKERRQ(ierr);
} else {
ierr = MatMatSolve(deluxe_ctx->seq_mat_inv_sum[i],X,Y);CHKERRQ(ierr);
}
ierr = MatDestroy(&deluxe_ctx->seq_mat_inv_sum[i]);CHKERRQ(ierr);
ierr = MatDestroy(&deluxe_ctx->seq_mat[i]);CHKERRQ(ierr);
ierr = MatDestroy(&X);CHKERRQ(ierr);
if (deluxe_ctx->change) {
Mat C,CY;
if (!deluxe_ctx->change_with_qr) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Only QR based change of basis");
ierr = KSPGetOperators(deluxe_ctx->change[i],&C,NULL);CHKERRQ(ierr);
ierr = MatMatMult(C,Y,MAT_INITIAL_MATRIX,PETSC_DEFAULT,&CY);CHKERRQ(ierr);
ierr = MatMatTransposeMult(CY,C,MAT_REUSE_MATRIX,PETSC_DEFAULT,&Y);CHKERRQ(ierr);
ierr = MatDestroy(&CY);CHKERRQ(ierr);
}
ierr = MatTranspose(Y,MAT_INPLACE_MATRIX,&Y);CHKERRQ(ierr);
deluxe_ctx->seq_mat[i] = Y;
}
cum += subset_size;
cum2 += subset_size*subset_size;
}
ierr = ISRestoreIndices(sub_schurs->is_Ej_all,&idxs);CHKERRQ(ierr);
ierr = MatSeqAIJRestoreArray(sub_schurs->S_Ej_all,&matdata);CHKERRQ(ierr);
ierr = MatSeqAIJRestoreArray(sub_schurs->sum_S_Ej_all,&matdata2);CHKERRQ(ierr);
if (pcbddc->deluxe_singlemat) {
deluxe_ctx->change = NULL;
deluxe_ctx->change_with_qr = PETSC_FALSE;
}
if (deluxe_ctx->change && !deluxe_ctx->change_with_qr) {
for (i=0;i<deluxe_ctx->seq_n;i++) {
if (newsetup) {
PC pc;
ierr = KSPGetPC(deluxe_ctx->change[i],&pc);CHKERRQ(ierr);
ierr = PCSetType(pc,PCLU);CHKERRQ(ierr);
ierr = KSPSetFromOptions(deluxe_ctx->change[i]);CHKERRQ(ierr);
}
ierr = KSPSetUp(deluxe_ctx->change[i]);CHKERRQ(ierr);
}
}
PetscFunctionReturn(0);
}
/*
DMDAGetFaceInterpolation - Gets the interpolation for a face based coarse space
*/
PetscErrorCode DMDAGetFaceInterpolation(DM da,PC_Exotic *exotic,Mat Aglobal,MatReuse reuse,Mat *P)
{
PetscErrorCode ierr;
PetscInt dim,i,j,k,m,n,p,dof,Nint,Nface,Nwire,Nsurf,*Iint,*Isurf,cint = 0,csurf = 0,istart,jstart,kstart,*II,N,c = 0;
PetscInt mwidth,nwidth,pwidth,cnt,mp,np,pp,Ntotal,gl[6],*globals,Ng,*IIint,*IIsurf,Nt;
Mat Xint, Xsurf,Xint_tmp;
IS isint,issurf,is,row,col;
ISLocalToGlobalMapping ltg;
MPI_Comm comm;
Mat A,Aii,Ais,Asi,*Aholder,iAii;
MatFactorInfo info;
PetscScalar *xsurf,*xint;
#if defined(PETSC_USE_DEBUG_foo)
PetscScalar tmp;
#endif
PetscTable ht;
PetscFunctionBegin;
ierr = DMDAGetInfo(da,&dim,0,0,0,&mp,&np,&pp,&dof,0,0,0,0,0);
CHKERRQ(ierr);
if (dof != 1) SETERRQ(PetscObjectComm((PetscObject)da),PETSC_ERR_SUP,"Only for single field problems");
if (dim != 3) SETERRQ(PetscObjectComm((PetscObject)da),PETSC_ERR_SUP,"Only coded for 3d problems");
ierr = DMDAGetCorners(da,0,0,0,&m,&n,&p);
CHKERRQ(ierr);
ierr = DMDAGetGhostCorners(da,&istart,&jstart,&kstart,&mwidth,&nwidth,&pwidth);
CHKERRQ(ierr);
istart = istart ? -1 : 0;
jstart = jstart ? -1 : 0;
kstart = kstart ? -1 : 0;
/*
the columns of P are the interpolation of each coarse grid point (one for each vertex and edge)
to all the local degrees of freedom (this includes the vertices, edges and faces).
Xint are the subset of the interpolation into the interior
Xface are the interpolation onto faces but not into the interior
Xsurf are the interpolation onto the vertices and edges (the surfbasket)
Xint
Symbolically one could write P = (Xface) after interchanging the rows to match the natural ordering on the domain
Xsurf
*/
N = (m - istart)*(n - jstart)*(p - kstart);
Nint = (m-2-istart)*(n-2-jstart)*(p-2-kstart);
Nface = 2*((m-2-istart)*(n-2-jstart) + (m-2-istart)*(p-2-kstart) + (n-2-jstart)*(p-2-kstart));
Nwire = 4*((m-2-istart) + (n-2-jstart) + (p-2-kstart)) + 8;
Nsurf = Nface + Nwire;
ierr = MatCreateSeqDense(MPI_COMM_SELF,Nint,6,NULL,&Xint);
CHKERRQ(ierr);
ierr = MatCreateSeqDense(MPI_COMM_SELF,Nsurf,6,NULL,&Xsurf);
CHKERRQ(ierr);
ierr = MatDenseGetArray(Xsurf,&xsurf);
CHKERRQ(ierr);
/*
Require that all 12 edges and 6 faces have at least one grid point. Otherwise some of the columns of
Xsurf will be all zero (thus making the coarse matrix singular).
*/
if (m-istart < 3) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Number of grid points per process in X direction must be at least 3");
if (n-jstart < 3) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Number of grid points per process in Y direction must be at least 3");
if (p-kstart < 3) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_SUP,"Number of grid points per process in Z direction must be at least 3");
cnt = 0;
for (j=1; j<n-1-jstart; j++) {
for (i=1; i<m-istart-1; i++) xsurf[cnt++ + 0*Nsurf] = 1;
}
for (k=1; k<p-1-kstart; k++) {
for (i=1; i<m-istart-1; i++) xsurf[cnt++ + 1*Nsurf] = 1;
for (j=1; j<n-1-jstart; j++) {
xsurf[cnt++ + 2*Nsurf] = 1;
/* these are the interior nodes */
xsurf[cnt++ + 3*Nsurf] = 1;
}
for (i=1; i<m-istart-1; i++) xsurf[cnt++ + 4*Nsurf] = 1;
}
for (j=1; j<n-1-jstart; j++) {
for (i=1; i<m-istart-1; i++) xsurf[cnt++ + 5*Nsurf] = 1;
}
#if defined(PETSC_USE_DEBUG_foo)
for (i=0; i<Nsurf; i++) {
tmp = 0.0;
for (j=0; j<6; j++) tmp += xsurf[i+j*Nsurf];
if (PetscAbsScalar(tmp-1.0) > 1.e-10) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Wrong Xsurf interpolation at i %D value %g",i,(double)PetscAbsScalar(tmp));
}
#endif
ierr = MatDenseRestoreArray(Xsurf,&xsurf);
CHKERRQ(ierr);
/* ierr = MatView(Xsurf,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);*/
/*
I are the indices for all the needed vertices (in global numbering)
Iint are the indices for the interior values, I surf for the surface values
(This is just for the part of the global matrix obtained with MatGetSubMatrix(), it
is NOT the local DMDA ordering.)
IIint and IIsurf are the same as the Iint, Isurf except they are in the global numbering
*/
#define Endpoint(a,start,b) (a == 0 || a == (b-1-start))
ierr = PetscMalloc3(N,&II,Nint,&Iint,Nsurf,&Isurf);
CHKERRQ(ierr);
ierr = PetscMalloc2(Nint,&IIint,Nsurf,&IIsurf);
CHKERRQ(ierr);
for (k=0; k<p-kstart; k++) {
for (j=0; j<n-jstart; j++) {
for (i=0; i<m-istart; i++) {
II[c++] = i + j*mwidth + k*mwidth*nwidth;
if (!Endpoint(i,istart,m) && !Endpoint(j,jstart,n) && !Endpoint(k,kstart,p)) {
IIint[cint] = i + j*mwidth + k*mwidth*nwidth;
Iint[cint++] = i + j*(m-istart) + k*(m-istart)*(n-jstart);
} else {
IIsurf[csurf] = i + j*mwidth + k*mwidth*nwidth;
Isurf[csurf++] = i + j*(m-istart) + k*(m-istart)*(n-jstart);
}
}
}
}
if (c != N) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"c != N");
if (cint != Nint) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"cint != Nint");
if (csurf != Nsurf) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_PLIB,"csurf != Nsurf");
ierr = DMGetLocalToGlobalMapping(da,<g);
CHKERRQ(ierr);
ierr = ISLocalToGlobalMappingApply(ltg,N,II,II);
CHKERRQ(ierr);
ierr = ISLocalToGlobalMappingApply(ltg,Nint,IIint,IIint);
CHKERRQ(ierr);
ierr = ISLocalToGlobalMappingApply(ltg,Nsurf,IIsurf,IIsurf);
CHKERRQ(ierr);
ierr = PetscObjectGetComm((PetscObject)da,&comm);
CHKERRQ(ierr);
ierr = ISCreateGeneral(comm,N,II,PETSC_COPY_VALUES,&is);
CHKERRQ(ierr);
ierr = ISCreateGeneral(PETSC_COMM_SELF,Nint,Iint,PETSC_COPY_VALUES,&isint);
CHKERRQ(ierr);
ierr = ISCreateGeneral(PETSC_COMM_SELF,Nsurf,Isurf,PETSC_COPY_VALUES,&issurf);
CHKERRQ(ierr);
ierr = PetscFree3(II,Iint,Isurf);
CHKERRQ(ierr);
ierr = ISSort(is);
CHKERRQ(ierr);
ierr = MatGetSubMatrices(Aglobal,1,&is,&is,MAT_INITIAL_MATRIX,&Aholder);
CHKERRQ(ierr);
A = *Aholder;
ierr = PetscFree(Aholder);
CHKERRQ(ierr);
ierr = MatGetSubMatrix(A,isint,isint,MAT_INITIAL_MATRIX,&Aii);
CHKERRQ(ierr);
ierr = MatGetSubMatrix(A,isint,issurf,MAT_INITIAL_MATRIX,&Ais);
CHKERRQ(ierr);
ierr = MatGetSubMatrix(A,issurf,isint,MAT_INITIAL_MATRIX,&Asi);
CHKERRQ(ierr);
/*
Solve for the interpolation onto the interior Xint
*/
ierr = MatMatMult(Ais,Xsurf,MAT_INITIAL_MATRIX,PETSC_DETERMINE,&Xint_tmp);
CHKERRQ(ierr);
ierr = MatScale(Xint_tmp,-1.0);
CHKERRQ(ierr);
if (exotic->directSolve) {
ierr = MatGetFactor(Aii,MATSOLVERPETSC,MAT_FACTOR_LU,&iAii);
CHKERRQ(ierr);
ierr = MatFactorInfoInitialize(&info);
CHKERRQ(ierr);
ierr = MatGetOrdering(Aii,MATORDERINGND,&row,&col);
CHKERRQ(ierr);
ierr = MatLUFactorSymbolic(iAii,Aii,row,col,&info);
CHKERRQ(ierr);
ierr = ISDestroy(&row);
CHKERRQ(ierr);
ierr = ISDestroy(&col);
CHKERRQ(ierr);
ierr = MatLUFactorNumeric(iAii,Aii,&info);
CHKERRQ(ierr);
ierr = MatMatSolve(iAii,Xint_tmp,Xint);
CHKERRQ(ierr);
ierr = MatDestroy(&iAii);
CHKERRQ(ierr);
} else {
Vec b,x;
PetscScalar *xint_tmp;
ierr = MatDenseGetArray(Xint,&xint);
CHKERRQ(ierr);
ierr = VecCreateSeqWithArray(PETSC_COMM_SELF,1,Nint,0,&x);
CHKERRQ(ierr);
ierr = MatDenseGetArray(Xint_tmp,&xint_tmp);
CHKERRQ(ierr);
ierr = VecCreateSeqWithArray(PETSC_COMM_SELF,1,Nint,0,&b);
CHKERRQ(ierr);
ierr = KSPSetOperators(exotic->ksp,Aii,Aii);
CHKERRQ(ierr);
for (i=0; i<6; i++) {
ierr = VecPlaceArray(x,xint+i*Nint);
CHKERRQ(ierr);
ierr = VecPlaceArray(b,xint_tmp+i*Nint);
CHKERRQ(ierr);
ierr = KSPSolve(exotic->ksp,b,x);
CHKERRQ(ierr);
ierr = VecResetArray(x);
CHKERRQ(ierr);
ierr = VecResetArray(b);
CHKERRQ(ierr);
}
ierr = MatDenseRestoreArray(Xint,&xint);
CHKERRQ(ierr);
ierr = MatDenseRestoreArray(Xint_tmp,&xint_tmp);
CHKERRQ(ierr);
ierr = VecDestroy(&x);
CHKERRQ(ierr);
ierr = VecDestroy(&b);
CHKERRQ(ierr);
}
ierr = MatDestroy(&Xint_tmp);
CHKERRQ(ierr);
#if defined(PETSC_USE_DEBUG_foo)
ierr = MatDenseGetArray(Xint,&xint);
CHKERRQ(ierr);
for (i=0; i<Nint; i++) {
tmp = 0.0;
for (j=0; j<6; j++) tmp += xint[i+j*Nint];
if (PetscAbsScalar(tmp-1.0) > 1.e-10) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Wrong Xint interpolation at i %D value %g",i,(double)PetscAbsScalar(tmp));
}
ierr = MatDenseRestoreArray(Xint,&xint);
CHKERRQ(ierr);
/* ierr =MatView(Xint,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr); */
#endif
/* total faces */
Ntotal = mp*np*(pp+1) + mp*pp*(np+1) + np*pp*(mp+1);
/*
For each vertex, edge, face on process (in the same orderings as used above) determine its local number including ghost points
*/
cnt = 0;
{
gl[cnt++] = mwidth+1;
}
{
{
gl[cnt++] = mwidth*nwidth+1;
}
{
gl[cnt++] = mwidth*nwidth + mwidth; /* these are the interior nodes */ gl[cnt++] = mwidth*nwidth + mwidth+m-istart-1;
}
{
gl[cnt++] = mwidth*nwidth+mwidth*(n-jstart-1)+1;
}
}
{
gl[cnt++] = mwidth*nwidth*(p-kstart-1) + mwidth+1;
}
/* PetscIntView(6,gl,PETSC_VIEWER_STDOUT_WORLD); */
/* convert that to global numbering and get them on all processes */
ierr = ISLocalToGlobalMappingApply(ltg,6,gl,gl);
CHKERRQ(ierr);
/* PetscIntView(6,gl,PETSC_VIEWER_STDOUT_WORLD); */
ierr = PetscMalloc1(6*mp*np*pp,&globals);
CHKERRQ(ierr);
ierr = MPI_Allgather(gl,6,MPIU_INT,globals,6,MPIU_INT,PetscObjectComm((PetscObject)da));
CHKERRQ(ierr);
/* Number the coarse grid points from 0 to Ntotal */
ierr = MatGetSize(Aglobal,&Nt,NULL);
CHKERRQ(ierr);
ierr = PetscTableCreate(Ntotal/3,Nt+1,&ht);
CHKERRQ(ierr);
for (i=0; i<6*mp*np*pp; i++) {
ierr = PetscTableAddCount(ht,globals[i]+1);
CHKERRQ(ierr);
}
ierr = PetscTableGetCount(ht,&cnt);
CHKERRQ(ierr);
if (cnt != Ntotal) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Hash table size %D not equal to total number coarse grid points %D",cnt,Ntotal);
ierr = PetscFree(globals);
CHKERRQ(ierr);
for (i=0; i<6; i++) {
ierr = PetscTableFind(ht,gl[i]+1,&gl[i]);
CHKERRQ(ierr);
gl[i]--;
}
ierr = PetscTableDestroy(&ht);
CHKERRQ(ierr);
/* PetscIntView(6,gl,PETSC_VIEWER_STDOUT_WORLD); */
/* construct global interpolation matrix */
ierr = MatGetLocalSize(Aglobal,&Ng,NULL);
CHKERRQ(ierr);
if (reuse == MAT_INITIAL_MATRIX) {
ierr = MatCreateAIJ(PetscObjectComm((PetscObject)da),Ng,PETSC_DECIDE,PETSC_DECIDE,Ntotal,Nint+Nsurf,NULL,Nint,NULL,P);
CHKERRQ(ierr);
} else {
ierr = MatZeroEntries(*P);
CHKERRQ(ierr);
}
ierr = MatSetOption(*P,MAT_ROW_ORIENTED,PETSC_FALSE);
CHKERRQ(ierr);
ierr = MatDenseGetArray(Xint,&xint);
CHKERRQ(ierr);
ierr = MatSetValues(*P,Nint,IIint,6,gl,xint,INSERT_VALUES);
CHKERRQ(ierr);
ierr = MatDenseRestoreArray(Xint,&xint);
CHKERRQ(ierr);
ierr = MatDenseGetArray(Xsurf,&xsurf);
CHKERRQ(ierr);
ierr = MatSetValues(*P,Nsurf,IIsurf,6,gl,xsurf,INSERT_VALUES);
CHKERRQ(ierr);
ierr = MatDenseRestoreArray(Xsurf,&xsurf);
CHKERRQ(ierr);
ierr = MatAssemblyBegin(*P,MAT_FINAL_ASSEMBLY);
CHKERRQ(ierr);
ierr = MatAssemblyEnd(*P,MAT_FINAL_ASSEMBLY);
CHKERRQ(ierr);
ierr = PetscFree2(IIint,IIsurf);
CHKERRQ(ierr);
#if defined(PETSC_USE_DEBUG_foo)
{
Vec x,y;
PetscScalar *yy;
ierr = VecCreateMPI(PetscObjectComm((PetscObject)da),Ng,PETSC_DETERMINE,&y);
CHKERRQ(ierr);
ierr = VecCreateMPI(PetscObjectComm((PetscObject)da),PETSC_DETERMINE,Ntotal,&x);
CHKERRQ(ierr);
ierr = VecSet(x,1.0);
CHKERRQ(ierr);
ierr = MatMult(*P,x,y);
CHKERRQ(ierr);
ierr = VecGetArray(y,&yy);
CHKERRQ(ierr);
for (i=0; i<Ng; i++) {
if (PetscAbsScalar(yy[i]-1.0) > 1.e-10) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_PLIB,"Wrong p interpolation at i %D value %g",i,(double)PetscAbsScalar(yy[i]));
}
ierr = VecRestoreArray(y,&yy);
CHKERRQ(ierr);
ierr = VecDestroy(x);
CHKERRQ(ierr);
ierr = VecDestroy(y);
CHKERRQ(ierr);
}
#endif
ierr = MatDestroy(&Aii);
CHKERRQ(ierr);
ierr = MatDestroy(&Ais);
CHKERRQ(ierr);
ierr = MatDestroy(&Asi);
CHKERRQ(ierr);
ierr = MatDestroy(&A);
CHKERRQ(ierr);
ierr = ISDestroy(&is);
CHKERRQ(ierr);
ierr = ISDestroy(&isint);
CHKERRQ(ierr);
ierr = ISDestroy(&issurf);
CHKERRQ(ierr);
ierr = MatDestroy(&Xint);
CHKERRQ(ierr);
ierr = MatDestroy(&Xsurf);
CHKERRQ(ierr);
PetscFunctionReturn(0);
}
int main(int argc,char **args)
{
Mat A,RHS,C,F,X,S;
Vec u,x,b;
Vec xschur,bschur,uschur;
IS is_schur;
PetscErrorCode ierr;
PetscMPIInt size;
PetscInt isolver=0,size_schur,m,n,nfact,nsolve,nrhs;
PetscReal norm,tol=PETSC_SQRT_MACHINE_EPSILON;
PetscRandom rand;
PetscBool data_provided,herm,symm,use_lu;
PetscReal sratio = 5.1/12.;
PetscViewer fd; /* viewer */
char solver[256];
char file[PETSC_MAX_PATH_LEN]; /* input file name */
PetscInitialize(&argc,&args,(char*)0,help);
ierr = MPI_Comm_size(PETSC_COMM_WORLD, &size);CHKERRQ(ierr);
if (size > 1) SETERRQ(PETSC_COMM_WORLD,1,"This is a uniprocessor test");
/* Determine which type of solver we want to test for */
herm = PETSC_FALSE;
symm = PETSC_FALSE;
ierr = PetscOptionsGetBool(NULL,NULL,"-symmetric_solve",&symm,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetBool(NULL,NULL,"-hermitian_solve",&herm,NULL);CHKERRQ(ierr);
if (herm) symm = PETSC_TRUE;
/* Determine file from which we read the matrix A */
ierr = PetscOptionsGetString(NULL,NULL,"-f",file,PETSC_MAX_PATH_LEN,&data_provided);CHKERRQ(ierr);
if (!data_provided) { /* get matrices from PETSc distribution */
sprintf(file,PETSC_DIR);
ierr = PetscStrcat(file,"/share/petsc/datafiles/matrices/");CHKERRQ(ierr);
if (symm) {
#if defined (PETSC_USE_COMPLEX)
ierr = PetscStrcat(file,"hpd-complex-");CHKERRQ(ierr);
#else
ierr = PetscStrcat(file,"spd-real-");CHKERRQ(ierr);
#endif
} else {
#if defined (PETSC_USE_COMPLEX)
ierr = PetscStrcat(file,"nh-complex-");CHKERRQ(ierr);
#else
ierr = PetscStrcat(file,"ns-real-");CHKERRQ(ierr);
#endif
}
#if defined(PETSC_USE_64BIT_INDICES)
ierr = PetscStrcat(file,"int64-");CHKERRQ(ierr);
#else
ierr = PetscStrcat(file,"int32-");CHKERRQ(ierr);
#endif
#if defined (PETSC_USE_REAL_SINGLE)
ierr = PetscStrcat(file,"float32");CHKERRQ(ierr);
#else
ierr = PetscStrcat(file,"float64");CHKERRQ(ierr);
#endif
}
/* Load matrix A */
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file,FILE_MODE_READ,&fd);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr);
ierr = MatLoad(A,fd);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd);CHKERRQ(ierr);
ierr = MatGetSize(A,&m,&n);CHKERRQ(ierr);
if (m != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ, "This example is not intended for rectangular matrices (%d, %d)", m, n);
/* Create dense matrix C and X; C holds true solution with identical colums */
nrhs = 2;
ierr = PetscOptionsGetInt(NULL,NULL,"-nrhs",&nrhs,NULL);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&C);CHKERRQ(ierr);
ierr = MatSetSizes(C,m,PETSC_DECIDE,PETSC_DECIDE,nrhs);CHKERRQ(ierr);
ierr = MatSetType(C,MATDENSE);CHKERRQ(ierr);
ierr = MatSetFromOptions(C);CHKERRQ(ierr);
ierr = MatSetUp(C);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_WORLD,&rand);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rand);CHKERRQ(ierr);
ierr = MatSetRandom(C,rand);CHKERRQ(ierr);
ierr = MatDuplicate(C,MAT_DO_NOT_COPY_VALUES,&X);CHKERRQ(ierr);
/* Create vectors */
ierr = VecCreate(PETSC_COMM_WORLD,&x);CHKERRQ(ierr);
ierr = VecSetSizes(x,n,PETSC_DECIDE);CHKERRQ(ierr);
ierr = VecSetFromOptions(x);CHKERRQ(ierr);
ierr = VecDuplicate(x,&b);CHKERRQ(ierr);
ierr = VecDuplicate(x,&u);CHKERRQ(ierr); /* save the true solution */
ierr = PetscOptionsGetInt(NULL,NULL,"-solver",&isolver,NULL);CHKERRQ(ierr);
switch (isolver) {
#if defined(PETSC_HAVE_MUMPS)
case 0:
ierr = PetscStrcpy(solver,MATSOLVERMUMPS);CHKERRQ(ierr);
break;
#endif
#if defined(PETSC_HAVE_MKL_PARDISO)
case 1:
ierr = PetscStrcpy(solver,MATSOLVERMKL_PARDISO);CHKERRQ(ierr);
break;
#endif
default:
ierr = PetscStrcpy(solver,MATSOLVERPETSC);CHKERRQ(ierr);
break;
}
#if defined (PETSC_USE_COMPLEX)
if (isolver == 0 && symm && !data_provided) { /* MUMPS (5.0.0) does not have support for hermitian matrices, so make them symmetric */
PetscScalar im = PetscSqrtScalar((PetscScalar)-1.);
PetscScalar val = -1.0;
val = val + im;
ierr = MatSetValue(A,1,0,val,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
}
#endif
ierr = PetscOptionsGetReal(NULL,NULL,"-schur_ratio",&sratio,NULL);CHKERRQ(ierr);
if (sratio < 0. || sratio > 1.) {
SETERRQ1(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ, "Invalid ratio for schur degrees of freedom %f", sratio);
}
size_schur = (PetscInt)(sratio*m);
ierr = PetscPrintf(PETSC_COMM_SELF,"Solving with %s: nrhs %d, sym %d, herm %d, size schur %d, size mat %d\n",solver,nrhs,symm,herm,size_schur,m);CHKERRQ(ierr);
/* Test LU/Cholesky Factorization */
use_lu = PETSC_FALSE;
if (!symm) use_lu = PETSC_TRUE;
#if defined (PETSC_USE_COMPLEX)
if (isolver == 1) use_lu = PETSC_TRUE;
#endif
if (herm && !use_lu) { /* test also conversion routines inside the solver packages */
ierr = MatSetOption(A,MAT_SYMMETRIC,PETSC_TRUE);CHKERRQ(ierr);
ierr = MatConvert(A,MATSEQSBAIJ,MAT_INPLACE_MATRIX,&A);CHKERRQ(ierr);
}
if (use_lu) {
ierr = MatGetFactor(A,solver,MAT_FACTOR_LU,&F);CHKERRQ(ierr);
} else {
if (herm) {
ierr = MatSetOption(A,MAT_SYMMETRIC,PETSC_TRUE);CHKERRQ(ierr);
ierr = MatSetOption(A,MAT_SPD,PETSC_TRUE);CHKERRQ(ierr);
} else {
ierr = MatSetOption(A,MAT_SYMMETRIC,PETSC_TRUE);CHKERRQ(ierr);
ierr = MatSetOption(A,MAT_SPD,PETSC_FALSE);CHKERRQ(ierr);
}
ierr = MatGetFactor(A,solver,MAT_FACTOR_CHOLESKY,&F);CHKERRQ(ierr);
}
ierr = ISCreateStride(PETSC_COMM_SELF,size_schur,m-size_schur,1,&is_schur);CHKERRQ(ierr);
ierr = MatFactorSetSchurIS(F,is_schur);CHKERRQ(ierr);
ierr = ISDestroy(&is_schur);CHKERRQ(ierr);
if (use_lu) {
ierr = MatLUFactorSymbolic(F,A,NULL,NULL,NULL);CHKERRQ(ierr);
} else {
ierr = MatCholeskyFactorSymbolic(F,A,NULL,NULL);CHKERRQ(ierr);
}
for (nfact = 0; nfact < 3; nfact++) {
Mat AD;
if (!nfact) {
ierr = VecSetRandom(x,rand);CHKERRQ(ierr);
if (symm && herm) {
ierr = VecAbs(x);CHKERRQ(ierr);
}
ierr = MatDiagonalSet(A,x,ADD_VALUES);CHKERRQ(ierr);
}
if (use_lu) {
ierr = MatLUFactorNumeric(F,A,NULL);CHKERRQ(ierr);
} else {
ierr = MatCholeskyFactorNumeric(F,A,NULL);CHKERRQ(ierr);
}
ierr = MatFactorCreateSchurComplement(F,&S);CHKERRQ(ierr);
ierr = MatCreateVecs(S,&xschur,&bschur);CHKERRQ(ierr);
ierr = VecDuplicate(xschur,&uschur);CHKERRQ(ierr);
if (nfact == 1) {
ierr = MatFactorInvertSchurComplement(F);CHKERRQ(ierr);
}
for (nsolve = 0; nsolve < 2; nsolve++) {
ierr = VecSetRandom(x,rand);CHKERRQ(ierr);
ierr = VecCopy(x,u);CHKERRQ(ierr);
if (nsolve) {
ierr = MatMult(A,x,b);CHKERRQ(ierr);
ierr = MatSolve(F,b,x);CHKERRQ(ierr);
} else {
ierr = MatMultTranspose(A,x,b);CHKERRQ(ierr);
ierr = MatSolveTranspose(F,b,x);CHKERRQ(ierr);
}
/* Check the error */
ierr = VecAXPY(u,-1.0,x);CHKERRQ(ierr); /* u <- (-1.0)x + u */
ierr = VecNorm(u,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
PetscReal resi;
if (nsolve) {
ierr = MatMult(A,x,u);CHKERRQ(ierr); /* u = A*x */
} else {
ierr = MatMultTranspose(A,x,u);CHKERRQ(ierr); /* u = A*x */
}
ierr = VecAXPY(u,-1.0,b);CHKERRQ(ierr); /* u <- (-1.0)b + u */
ierr = VecNorm(u,NORM_2,&resi);CHKERRQ(ierr);
if (nsolve) {
ierr = PetscPrintf(PETSC_COMM_SELF,"(f %d, s %d) MatSolve error: Norm of error %g, residual %f\n",nfact,nsolve,norm,resi);CHKERRQ(ierr);
} else {
ierr = PetscPrintf(PETSC_COMM_SELF,"(f %d, s %d) MatSolveTranspose error: Norm of error %g, residual %f\n",nfact,nsolve,norm,resi);CHKERRQ(ierr);
}
}
ierr = VecSetRandom(xschur,rand);CHKERRQ(ierr);
ierr = VecCopy(xschur,uschur);CHKERRQ(ierr);
if (nsolve) {
ierr = MatMult(S,xschur,bschur);CHKERRQ(ierr);
ierr = MatFactorSolveSchurComplement(F,bschur,xschur);CHKERRQ(ierr);
} else {
ierr = MatMultTranspose(S,xschur,bschur);CHKERRQ(ierr);
ierr = MatFactorSolveSchurComplementTranspose(F,bschur,xschur);CHKERRQ(ierr);
}
/* Check the error */
ierr = VecAXPY(uschur,-1.0,xschur);CHKERRQ(ierr); /* u <- (-1.0)x + u */
ierr = VecNorm(uschur,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
PetscReal resi;
if (nsolve) {
ierr = MatMult(S,xschur,uschur);CHKERRQ(ierr); /* u = A*x */
} else {
ierr = MatMultTranspose(S,xschur,uschur);CHKERRQ(ierr); /* u = A*x */
}
ierr = VecAXPY(uschur,-1.0,bschur);CHKERRQ(ierr); /* u <- (-1.0)b + u */
ierr = VecNorm(uschur,NORM_2,&resi);CHKERRQ(ierr);
if (nsolve) {
ierr = PetscPrintf(PETSC_COMM_SELF,"(f %d, s %d) MatFactorSolveSchurComplement error: Norm of error %g, residual %f\n",nfact,nsolve,norm,resi);CHKERRQ(ierr);
} else {
ierr = PetscPrintf(PETSC_COMM_SELF,"(f %d, s %d) MatFactorSolveSchurComplementTranspose error: Norm of error %g, residual %f\n",nfact,nsolve,norm,resi);CHKERRQ(ierr);
}
}
}
ierr = MatConvert(A,MATSEQAIJ,MAT_INITIAL_MATRIX,&AD);
if (!nfact) {
ierr = MatMatMult(AD,C,MAT_INITIAL_MATRIX,2.0,&RHS);CHKERRQ(ierr);
} else {
ierr = MatMatMult(AD,C,MAT_REUSE_MATRIX,2.0,&RHS);CHKERRQ(ierr);
}
ierr = MatDestroy(&AD);CHKERRQ(ierr);
for (nsolve = 0; nsolve < 2; nsolve++) {
ierr = MatMatSolve(F,RHS,X);CHKERRQ(ierr);
/* Check the error */
ierr = MatAXPY(X,-1.0,C,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = MatNorm(X,NORM_FROBENIUS,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_SELF,"(f %D, s %D) MatMatSolve: Norm of error %g\n",nfact,nsolve,norm);CHKERRQ(ierr);
}
}
ierr = MatDestroy(&S);CHKERRQ(ierr);
ierr = VecDestroy(&xschur);CHKERRQ(ierr);
ierr = VecDestroy(&bschur);CHKERRQ(ierr);
ierr = VecDestroy(&uschur);CHKERRQ(ierr);
}
/* Free data structures */
ierr = MatDestroy(&A);CHKERRQ(ierr);
ierr = MatDestroy(&C);CHKERRQ(ierr);
ierr = MatDestroy(&F);CHKERRQ(ierr);
ierr = MatDestroy(&X);CHKERRQ(ierr);
ierr = MatDestroy(&RHS);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&rand);CHKERRQ(ierr);
ierr = VecDestroy(&x);CHKERRQ(ierr);
ierr = VecDestroy(&b);CHKERRQ(ierr);
ierr = VecDestroy(&u);CHKERRQ(ierr);
ierr = PetscFinalize();
return 0;
}
int main(int argc,char **argv)
{
Mat A,F,B,X,C,Aher,G;
Vec b,x,c,d,e;
PetscErrorCode ierr;
PetscInt m = 5,n,p,i,j,nrows,ncols;
PetscScalar *v,*barray,rval;
PetscReal norm,tol=1.e-12;
PetscMPIInt size,rank;
PetscRandom rand;
const PetscInt *rows,*cols;
IS isrows,iscols;
PetscBool mats_view=PETSC_FALSE;
MatFactorInfo finfo;
PetscInitialize(&argc,&argv,(char*) 0,help);
ierr = MPI_Comm_rank(PETSC_COMM_WORLD,&rank);CHKERRQ(ierr);
ierr = MPI_Comm_size(PETSC_COMM_WORLD,&size);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_WORLD,&rand);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rand);CHKERRQ(ierr);
/* Get local dimensions of matrices */
ierr = PetscOptionsGetInt(NULL,"-m",&m,NULL);CHKERRQ(ierr);
n = m;
ierr = PetscOptionsGetInt(NULL,"-n",&n,NULL);CHKERRQ(ierr);
p = m/2;
ierr = PetscOptionsGetInt(NULL,"-p",&p,NULL);CHKERRQ(ierr);
ierr = PetscOptionsHasName(NULL,"-mats_view",&mats_view);CHKERRQ(ierr);
/* Create matrix A */
ierr = PetscPrintf(PETSC_COMM_WORLD," Create Elemental matrix A\n");CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr);
ierr = MatSetSizes(A,m,n,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(A,MATELEMENTAL);CHKERRQ(ierr);
ierr = MatSetFromOptions(A);CHKERRQ(ierr);
ierr = MatSetUp(A);CHKERRQ(ierr);
/* Set local matrix entries */
ierr = MatGetOwnershipIS(A,&isrows,&iscols);CHKERRQ(ierr);
ierr = ISGetLocalSize(isrows,&nrows);CHKERRQ(ierr);
ierr = ISGetIndices(isrows,&rows);CHKERRQ(ierr);
ierr = ISGetLocalSize(iscols,&ncols);CHKERRQ(ierr);
ierr = ISGetIndices(iscols,&cols);CHKERRQ(ierr);
ierr = PetscMalloc1(nrows*ncols,&v);CHKERRQ(ierr);
for (i=0; i<nrows; i++) {
for (j=0; j<ncols; j++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
v[i*ncols+j] = rval;
}
}
ierr = MatSetValues(A,nrows,rows,ncols,cols,v,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = ISRestoreIndices(isrows,&rows);CHKERRQ(ierr);
ierr = ISRestoreIndices(iscols,&cols);CHKERRQ(ierr);
ierr = ISDestroy(&isrows);CHKERRQ(ierr);
ierr = ISDestroy(&iscols);CHKERRQ(ierr);
ierr = PetscFree(v);CHKERRQ(ierr);
if (mats_view) {
ierr = PetscPrintf(PETSC_COMM_WORLD, "A: nrows %d, m %d; ncols %d, n %d\n",nrows,m,ncols,n);CHKERRQ(ierr);
ierr = MatView(A,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
/* Create rhs matrix B */
ierr = PetscPrintf(PETSC_COMM_WORLD," Create rhs matrix B\n");CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&B);CHKERRQ(ierr);
ierr = MatSetSizes(B,m,p,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(B,MATELEMENTAL);CHKERRQ(ierr);
ierr = MatSetFromOptions(B);CHKERRQ(ierr);
ierr = MatSetUp(B);CHKERRQ(ierr);
ierr = MatGetOwnershipIS(B,&isrows,&iscols);CHKERRQ(ierr);
ierr = ISGetLocalSize(isrows,&nrows);CHKERRQ(ierr);
ierr = ISGetIndices(isrows,&rows);CHKERRQ(ierr);
ierr = ISGetLocalSize(iscols,&ncols);CHKERRQ(ierr);
ierr = ISGetIndices(iscols,&cols);CHKERRQ(ierr);
ierr = PetscMalloc1(nrows*ncols,&v);CHKERRQ(ierr);
for (i=0; i<nrows; i++) {
for (j=0; j<ncols; j++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
v[i*ncols+j] = rval;
}
}
ierr = MatSetValues(B,nrows,rows,ncols,cols,v,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = ISRestoreIndices(isrows,&rows);CHKERRQ(ierr);
ierr = ISRestoreIndices(iscols,&cols);CHKERRQ(ierr);
ierr = ISDestroy(&isrows);CHKERRQ(ierr);
ierr = ISDestroy(&iscols);CHKERRQ(ierr);
ierr = PetscFree(v);CHKERRQ(ierr);
if (mats_view) {
ierr = PetscPrintf(PETSC_COMM_WORLD, "B: nrows %d, m %d; ncols %d, p %d\n",nrows,m,ncols,p);CHKERRQ(ierr);
ierr = MatView(B,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
/* Create rhs vector b and solution x (same size as b) */
ierr = VecCreate(PETSC_COMM_WORLD,&b);CHKERRQ(ierr);
ierr = VecSetSizes(b,m,PETSC_DECIDE);CHKERRQ(ierr);
ierr = VecSetFromOptions(b);CHKERRQ(ierr);
ierr = VecGetArray(b,&barray);CHKERRQ(ierr);
for (j=0; j<m; j++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
barray[j] = rval;
}
ierr = VecRestoreArray(b,&barray);CHKERRQ(ierr);
ierr = VecAssemblyBegin(b);CHKERRQ(ierr);
ierr = VecAssemblyEnd(b);CHKERRQ(ierr);
if (mats_view) {
ierr = PetscSynchronizedPrintf(PETSC_COMM_WORLD, "[%d] b: m %d\n",rank,m);CHKERRQ(ierr);
ierr = PetscSynchronizedFlush(PETSC_COMM_WORLD,PETSC_STDOUT);CHKERRQ(ierr);
ierr = VecView(b,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
ierr = VecDuplicate(b,&x);CHKERRQ(ierr);
/* Create matrix X - same size as B */
ierr = PetscPrintf(PETSC_COMM_WORLD," Create solution matrix X\n");CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&X);CHKERRQ(ierr);
ierr = MatSetSizes(X,m,p,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(X,MATELEMENTAL);CHKERRQ(ierr);
ierr = MatSetFromOptions(X);CHKERRQ(ierr);
ierr = MatSetUp(X);CHKERRQ(ierr);
ierr = MatAssemblyBegin(X,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(X,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
/* Cholesky factorization */
/*------------------------*/
ierr = PetscPrintf(PETSC_COMM_WORLD," Create Elemental matrix Aher\n");CHKERRQ(ierr);
ierr = MatHermitianTranspose(A,MAT_INITIAL_MATRIX,&Aher);CHKERRQ(ierr);
ierr = MatAXPY(Aher,1.0,A,SAME_NONZERO_PATTERN);CHKERRQ(ierr); /* Aher = A + A^T */
if (!rank) { /* add 100.0 to diagonals of Aher to make it spd */
PetscInt M,N;
ierr = MatGetSize(Aher,&M,&N);CHKERRQ(ierr);
for (i=0; i<M; i++) {
rval = 100.0;
ierr = MatSetValues(Aher,1,&i,1,&i,&rval,ADD_VALUES);CHKERRQ(ierr);
}
}
ierr = MatAssemblyBegin(Aher,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(Aher,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
if (mats_view) {
ierr = PetscPrintf(PETSC_COMM_WORLD, "Aher:\n");CHKERRQ(ierr);
ierr = MatView(Aher,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
/* Cholesky factorization */
/*------------------------*/
ierr = PetscPrintf(PETSC_COMM_WORLD," Test Cholesky Solver \n");CHKERRQ(ierr);
/* In-place Cholesky */
/* Create matrix factor G, then copy Aher to G */
ierr = MatCreate(PETSC_COMM_WORLD,&G);CHKERRQ(ierr);
ierr = MatSetSizes(G,m,n,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(G,MATELEMENTAL);CHKERRQ(ierr);
ierr = MatSetFromOptions(G);CHKERRQ(ierr);
ierr = MatSetUp(G);CHKERRQ(ierr);
ierr = MatAssemblyBegin(G,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(G,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatCopy(Aher,G,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
/* Only G = U^T * U is implemented for now */
ierr = MatCholeskyFactor(G,0,0);CHKERRQ(ierr);
if (mats_view) {
ierr = PetscPrintf(PETSC_COMM_WORLD, "Cholesky Factor G:\n");CHKERRQ(ierr);
ierr = MatView(G,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
/* Solve U^T * U x = b and U^T * U X = B */
ierr = MatSolve(G,b,x);CHKERRQ(ierr);
ierr = MatMatSolve(G,B,X);CHKERRQ(ierr);
ierr = MatDestroy(&G);CHKERRQ(ierr);
/* Out-place Cholesky */
ierr = MatGetFactor(Aher,MATSOLVERELEMENTAL,MAT_FACTOR_CHOLESKY,&G);CHKERRQ(ierr);
ierr = MatCholeskyFactorSymbolic(G,Aher,0,&finfo);CHKERRQ(ierr);
ierr = MatCholeskyFactorNumeric(G,Aher,&finfo);CHKERRQ(ierr);
if (mats_view) {
ierr = MatView(G,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
ierr = MatSolve(G,b,x);CHKERRQ(ierr);
ierr = MatMatSolve(G,B,X);CHKERRQ(ierr);
ierr = MatDestroy(&G);CHKERRQ(ierr);
/* Check norm(Aher*x - b) */
ierr = VecCreate(PETSC_COMM_WORLD,&c);CHKERRQ(ierr);
ierr = VecSetSizes(c,m,PETSC_DECIDE);CHKERRQ(ierr);
ierr = VecSetFromOptions(c);CHKERRQ(ierr);
ierr = MatMult(Aher,x,c);CHKERRQ(ierr);
ierr = VecAXPY(c,-1.0,b);CHKERRQ(ierr);
ierr = VecNorm(c,NORM_1,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: |Aher*x - b| for Cholesky %g\n",(double)norm);CHKERRQ(ierr);
}
/* Check norm(Aher*X - B) */
ierr = MatMatMult(Aher,X,MAT_INITIAL_MATRIX,PETSC_DEFAULT,&C);CHKERRQ(ierr);
ierr = MatAXPY(C,-1.0,B,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = MatNorm(C,NORM_1,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: |Aher*X - B| for Cholesky %g\n",(double)norm);CHKERRQ(ierr);
}
/* LU factorization */
/*------------------*/
ierr = PetscPrintf(PETSC_COMM_WORLD," Test LU Solver \n");CHKERRQ(ierr);
/* In-place LU */
/* Create matrix factor F, then copy A to F */
ierr = MatCreate(PETSC_COMM_WORLD,&F);CHKERRQ(ierr);
ierr = MatSetSizes(F,m,n,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(F,MATELEMENTAL);CHKERRQ(ierr);
ierr = MatSetFromOptions(F);CHKERRQ(ierr);
ierr = MatSetUp(F);CHKERRQ(ierr);
ierr = MatAssemblyBegin(F,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(F,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatCopy(A,F,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
/* Create vector d to test MatSolveAdd() */
ierr = VecDuplicate(x,&d);CHKERRQ(ierr);
ierr = VecCopy(x,d);CHKERRQ(ierr);
/* PF=LU or F=LU factorization - perms is ignored by Elemental;
set finfo.dtcol !0 or 0 to enable/disable partial pivoting */
finfo.dtcol = 0.1;
ierr = MatLUFactor(F,0,0,&finfo);CHKERRQ(ierr);
/* Solve LUX = PB or LUX = B */
ierr = MatSolveAdd(F,b,d,x);CHKERRQ(ierr);
ierr = MatMatSolve(F,B,X);CHKERRQ(ierr);
ierr = MatDestroy(&F);CHKERRQ(ierr);
/* Check norm(A*X - B) */
ierr = VecCreate(PETSC_COMM_WORLD,&e);CHKERRQ(ierr);
ierr = VecSetSizes(e,m,PETSC_DECIDE);CHKERRQ(ierr);
ierr = VecSetFromOptions(e);CHKERRQ(ierr);
ierr = MatMult(A,x,c);CHKERRQ(ierr);
ierr = MatMult(A,d,e);CHKERRQ(ierr);
ierr = VecAXPY(c,-1.0,e);CHKERRQ(ierr);
ierr = VecAXPY(c,-1.0,b);CHKERRQ(ierr);
ierr = VecNorm(c,NORM_1,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: |A*x - b| for LU %g\n",(double)norm);CHKERRQ(ierr);
}
ierr = MatMatMult(A,X,MAT_REUSE_MATRIX,PETSC_DEFAULT,&C);CHKERRQ(ierr);
ierr = MatAXPY(C,-1.0,B,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = MatNorm(C,NORM_1,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: |A*X - B| for LU %g\n",(double)norm);CHKERRQ(ierr);
}
/* Out-place LU */
ierr = MatGetFactor(A,MATSOLVERELEMENTAL,MAT_FACTOR_LU,&F);CHKERRQ(ierr);
ierr = MatLUFactorSymbolic(F,A,0,0,&finfo);CHKERRQ(ierr);
ierr = MatLUFactorNumeric(F,A,&finfo);CHKERRQ(ierr);
if (mats_view) {
ierr = MatView(F,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
ierr = MatSolve(F,b,x);CHKERRQ(ierr);
ierr = MatMatSolve(F,B,X);CHKERRQ(ierr);
ierr = MatDestroy(&F);CHKERRQ(ierr);
/* Free space */
ierr = MatDestroy(&A);CHKERRQ(ierr);
ierr = MatDestroy(&Aher);CHKERRQ(ierr);
ierr = MatDestroy(&B);CHKERRQ(ierr);
ierr = MatDestroy(&C);CHKERRQ(ierr);
ierr = MatDestroy(&X);CHKERRQ(ierr);
ierr = VecDestroy(&b);CHKERRQ(ierr);
ierr = VecDestroy(&c);CHKERRQ(ierr);
ierr = VecDestroy(&d);CHKERRQ(ierr);
ierr = VecDestroy(&e);CHKERRQ(ierr);
ierr = VecDestroy(&x);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&rand);CHKERRQ(ierr);
ierr = PetscFinalize();
return 0;
}
For MATSEQDENSE matrix, the factorization is just a thin wrapper to LAPACK \n\n";
#include <petscmat.h>
#undef __FUNCT__
#define __FUNCT__ "main"
int main(int argc,char **argv)
{
Mat mat,F,RHS,SOLU;
MatInfo info;
PetscErrorCode ierr;
PetscInt m = 10,n = 10,i,j,rstart,rend,nrhs=2;
PetscScalar value = 1.0;
Vec x,y,b,ytmp;
PetscReal norm,tol=1.e-15;
PetscMPIInt size;
PetscScalar *rhs_array,*solu_array;
PetscRandom rand;
PetscScalar *array,rval;
PetscInitialize(&argc,&argv,(char*) 0,help);
ierr = MPI_Comm_size(PETSC_COMM_WORLD,&size);CHKERRQ(ierr);
if (size != 1) SETERRQ(PETSC_COMM_WORLD,PETSC_ERR_SUP,"This is a uniprocessor example only!");
/* create single vectors */
ierr = VecCreate(PETSC_COMM_WORLD,&y);CHKERRQ(ierr);
ierr = VecSetSizes(y,PETSC_DECIDE,m);CHKERRQ(ierr);
ierr = VecSetFromOptions(y);CHKERRQ(ierr);
ierr = VecDuplicate(y,&x);CHKERRQ(ierr);
ierr = VecDuplicate(y,&ytmp);CHKERRQ(ierr);
ierr = VecSet(x,value);CHKERRQ(ierr);
ierr = VecCreate(PETSC_COMM_WORLD,&b);CHKERRQ(ierr);
ierr = VecSetSizes(b,PETSC_DECIDE,n);CHKERRQ(ierr);
ierr = VecSetFromOptions(b);CHKERRQ(ierr);
/* create multiple vectors RHS and SOLU */
ierr = MatCreate(PETSC_COMM_WORLD,&RHS);CHKERRQ(ierr);
ierr = MatSetSizes(RHS,PETSC_DECIDE,PETSC_DECIDE,n,nrhs);CHKERRQ(ierr);
ierr = MatSetType(RHS,MATDENSE);CHKERRQ(ierr);
ierr = MatSetFromOptions(RHS);CHKERRQ(ierr);
ierr = MatSeqDenseSetPreallocation(RHS,NULL);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_WORLD,&rand);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rand);CHKERRQ(ierr);
ierr = MatDenseGetArray(RHS,&array);CHKERRQ(ierr);
for (j=0; j<nrhs; j++) {
for (i=0; i<n; i++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
array[n*j+i] = rval;
}
}
ierr = MatDenseRestoreArray(RHS,&array);CHKERRQ(ierr);
ierr = MatDuplicate(RHS,MAT_DO_NOT_COPY_VALUES,&SOLU);CHKERRQ(ierr);
/* create matrix */
ierr = MatCreateSeqDense(PETSC_COMM_WORLD,m,n,NULL,&mat);CHKERRQ(ierr);
ierr = MatGetOwnershipRange(mat,&rstart,&rend);CHKERRQ(ierr);
for (i=rstart; i<rend; i++) {
value = (PetscReal)i+1;
ierr = MatSetValues(mat,1,&i,1,&i,&value,INSERT_VALUES);CHKERRQ(ierr);
}
ierr = MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatGetInfo(mat,MAT_LOCAL,&info);CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,"matrix nonzeros = %D, allocated nonzeros = %D\n",
(PetscInt)info.nz_used,(PetscInt)info.nz_allocated);CHKERRQ(ierr);
/* Cholesky factorization - perm and factinfo are ignored by LAPACK */
/* in-place Cholesky */
ierr = MatMult(mat,x,b);CHKERRQ(ierr);
ierr = MatDuplicate(mat,MAT_COPY_VALUES,&F);CHKERRQ(ierr);
ierr = MatCholeskyFactor(F,0,0);CHKERRQ(ierr);
ierr = MatSolve(F,b,y);CHKERRQ(ierr);
ierr = MatDestroy(&F);CHKERRQ(ierr);
value = -1.0; ierr = VecAXPY(y,value,x);CHKERRQ(ierr);
ierr = VecNorm(y,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: Norm of error for Cholesky %G\n",norm);CHKERRQ(ierr);
}
/* out-place Cholesky */
ierr = MatGetFactor(mat,MATSOLVERPETSC,MAT_FACTOR_CHOLESKY,&F);CHKERRQ(ierr);
ierr = MatCholeskyFactorSymbolic(F,mat,0,0);CHKERRQ(ierr);
ierr = MatCholeskyFactorNumeric(F,mat,0);CHKERRQ(ierr);
ierr = MatSolve(F,b,y);CHKERRQ(ierr);
value = -1.0; ierr = VecAXPY(y,value,x);CHKERRQ(ierr);
ierr = VecNorm(y,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: Norm of error for Cholesky %G\n",norm);CHKERRQ(ierr);
}
ierr = MatDestroy(&F);CHKERRQ(ierr);
/* LU factorization - perms and factinfo are ignored by LAPACK */
i = m-1; value = 1.0;
ierr = MatSetValues(mat,1,&i,1,&i,&value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(mat,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatMult(mat,x,b);CHKERRQ(ierr);
ierr = MatDuplicate(mat,MAT_COPY_VALUES,&F);CHKERRQ(ierr);
/* in-place LU */
ierr = MatLUFactor(F,0,0,0);CHKERRQ(ierr);
ierr = MatSolve(F,b,y);CHKERRQ(ierr);
value = -1.0; ierr = VecAXPY(y,value,x);CHKERRQ(ierr);
ierr = VecNorm(y,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: Norm of error for LU %G\n",norm);CHKERRQ(ierr);
}
ierr = MatMatSolve(F,RHS,SOLU);CHKERRQ(ierr);
ierr = MatDenseGetArray(SOLU,&solu_array);CHKERRQ(ierr);
ierr = MatDenseGetArray(RHS,&rhs_array);CHKERRQ(ierr);
for (j=0; j<nrhs; j++) {
ierr = VecPlaceArray(y,solu_array+j*m);CHKERRQ(ierr);
ierr = VecPlaceArray(b,rhs_array+j*m);CHKERRQ(ierr);
ierr = MatMult(mat,y,ytmp);CHKERRQ(ierr);
ierr = VecAXPY(ytmp,-1.0,b);CHKERRQ(ierr); /* ytmp = mat*SOLU[:,j] - RHS[:,j] */
ierr = VecNorm(ytmp,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Error: Norm of residual for LU %G\n",norm);CHKERRQ(ierr);
}
ierr = VecResetArray(b);CHKERRQ(ierr);
ierr = VecResetArray(y);CHKERRQ(ierr);
}
ierr = MatDenseRestoreArray(RHS,&rhs_array);CHKERRQ(ierr);
ierr = MatDenseRestoreArray(SOLU,&solu_array);CHKERRQ(ierr);
ierr = MatDestroy(&F);CHKERRQ(ierr);
/* out-place LU */
ierr = MatGetFactor(mat,MATSOLVERPETSC,MAT_FACTOR_LU,&F);CHKERRQ(ierr);
ierr = MatLUFactorSymbolic(F,mat,0,0,0);CHKERRQ(ierr);
ierr = MatLUFactorNumeric(F,mat,0);CHKERRQ(ierr);
ierr = MatSolve(F,b,y);CHKERRQ(ierr);
value = -1.0; ierr = VecAXPY(y,value,x);CHKERRQ(ierr);
ierr = VecNorm(y,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = PetscPrintf(PETSC_COMM_WORLD,"Warning: Norm of error for LU %G\n",norm);CHKERRQ(ierr);
}
/* free space */
ierr = MatDestroy(&F);CHKERRQ(ierr);
ierr = MatDestroy(&mat);CHKERRQ(ierr);
ierr = MatDestroy(&RHS);CHKERRQ(ierr);
ierr = MatDestroy(&SOLU);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&rand);CHKERRQ(ierr);
ierr = VecDestroy(&x);CHKERRQ(ierr);
ierr = VecDestroy(&b);CHKERRQ(ierr);
ierr = VecDestroy(&y);CHKERRQ(ierr);
ierr = VecDestroy(&ytmp);CHKERRQ(ierr);
ierr = PetscFinalize();
return 0;
}
int main(int argc,char **args)
{
Mat A,RHS,C,F,X;
Vec u,x,b;
PetscErrorCode ierr;
PetscMPIInt rank,nproc;
PetscInt i,m,n,nfact,nsolve,nrhs,ipack=0;
PetscScalar *array,rval;
PetscReal norm,tol=1.e-12;
IS perm,iperm;
MatFactorInfo info;
PetscRandom rand;
PetscBool flg,testMatSolve=PETSC_TRUE,testMatMatSolve=PETSC_TRUE;
PetscViewer fd; /* viewer */
char file[PETSC_MAX_PATH_LEN]; /* input file name */
PetscInitialize(&argc,&args,(char*)0,help);
ierr = MPI_Comm_rank(PETSC_COMM_WORLD, &rank);CHKERRQ(ierr);
ierr = MPI_Comm_size(PETSC_COMM_WORLD, &nproc);CHKERRQ(ierr);
/* Determine file from which we read the matrix A */
ierr = PetscOptionsGetString(NULL,"-f",file,PETSC_MAX_PATH_LEN,&flg);CHKERRQ(ierr);
if (!flg) SETERRQ(PETSC_COMM_WORLD,1,"Must indicate binary file with the -f option");
/* Load matrix A */
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file,FILE_MODE_READ,&fd);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr);
ierr = MatLoad(A,fd);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd);CHKERRQ(ierr);
ierr = MatGetLocalSize(A,&m,&n);CHKERRQ(ierr);
if (m != n) SETERRQ2(PETSC_COMM_SELF,PETSC_ERR_ARG_SIZ, "This example is not intended for rectangular matrices (%d, %d)", m, n);
/* Create dense matrix C and X; C holds true solution with identical colums */
nrhs = 2;
ierr = PetscOptionsGetInt(NULL,"-nrhs",&nrhs,NULL);CHKERRQ(ierr);
if (!rank) printf("ex125: nrhs %d\n",nrhs);
ierr = MatCreate(PETSC_COMM_WORLD,&C);CHKERRQ(ierr);
ierr = MatSetSizes(C,m,PETSC_DECIDE,PETSC_DECIDE,nrhs);CHKERRQ(ierr);
ierr = MatSetType(C,MATDENSE);CHKERRQ(ierr);
ierr = MatSetFromOptions(C);CHKERRQ(ierr);
ierr = MatSetUp(C);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_WORLD,&rand);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rand);CHKERRQ(ierr);
ierr = MatSetRandom(C,rand);CHKERRQ(ierr);
ierr = MatDuplicate(C,MAT_DO_NOT_COPY_VALUES,&X);CHKERRQ(ierr);
/* Create vectors */
ierr = VecCreate(PETSC_COMM_WORLD,&x);CHKERRQ(ierr);
ierr = VecSetSizes(x,n,PETSC_DECIDE);CHKERRQ(ierr);
ierr = VecSetFromOptions(x);CHKERRQ(ierr);
ierr = VecDuplicate(x,&b);CHKERRQ(ierr);
ierr = VecDuplicate(x,&u);CHKERRQ(ierr); /* save the true solution */
/* Test LU Factorization */
ierr = MatGetOrdering(A,MATORDERINGND,&perm,&iperm);CHKERRQ(ierr);
/*ierr = ISView(perm,PETSC_VIEWER_STDOUT_WORLD);*/
/*ierr = ISView(perm,PETSC_VIEWER_STDOUT_SELF);*/
ierr = PetscOptionsGetInt(NULL,"-mat_solver_package",&ipack,NULL);CHKERRQ(ierr);
switch (ipack) {
case 0:
#if defined(PETSC_HAVE_SUPERLU)
if (!rank) printf(" SUPERLU LU:\n");
ierr = MatGetFactor(A,MATSOLVERSUPERLU,MAT_FACTOR_LU,&F);CHKERRQ(ierr);
break;
#endif
case 1:
#if defined(PETSC_HAVE_SUPERLU_DIST)
if (!rank) printf(" SUPERLU_DIST LU:\n");
ierr = MatGetFactor(A,MATSOLVERSUPERLU_DIST,MAT_FACTOR_LU,&F);CHKERRQ(ierr);
break;
#endif
case 2:
#if defined(PETSC_HAVE_MUMPS)
if (!rank) printf(" MUMPS LU:\n");
ierr = MatGetFactor(A,MATSOLVERMUMPS,MAT_FACTOR_LU,&F);CHKERRQ(ierr);
{
/* test mumps options */
PetscInt icntl_7 = 5;
ierr = MatMumpsSetIcntl(F,7,icntl_7);CHKERRQ(ierr);
}
break;
#endif
default:
if (!rank) printf(" PETSC LU:\n");
ierr = MatGetFactor(A,MATSOLVERPETSC,MAT_FACTOR_LU,&F);CHKERRQ(ierr);
}
info.fill = 5.0;
ierr = MatLUFactorSymbolic(F,A,perm,iperm,&info);CHKERRQ(ierr);
for (nfact = 0; nfact < 2; nfact++) {
if (!rank) printf(" %d-the LU numfactorization \n",nfact);
ierr = MatLUFactorNumeric(F,A,&info);CHKERRQ(ierr);
/* Test MatMatSolve() */
/*
if ((ipack == 0 || ipack == 2) && testMatMatSolve) {
printf(" MatMatSolve() is not implemented for this package. Skip the testing.\n");
testMatMatSolve = PETSC_FALSE;
}
*/
if (testMatMatSolve) {
if (!nfact) {
ierr = MatMatMult(A,C,MAT_INITIAL_MATRIX,2.0,&RHS);CHKERRQ(ierr);
} else {
ierr = MatMatMult(A,C,MAT_REUSE_MATRIX,2.0,&RHS);CHKERRQ(ierr);
}
for (nsolve = 0; nsolve < 2; nsolve++) {
if (!rank) printf(" %d-the MatMatSolve \n",nsolve);
ierr = MatMatSolve(F,RHS,X);CHKERRQ(ierr);
/* Check the error */
ierr = MatAXPY(X,-1.0,C,SAME_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = MatNorm(X,NORM_FROBENIUS,&norm);CHKERRQ(ierr);
if (norm > tol) {
if (!rank) {
ierr = PetscPrintf(PETSC_COMM_SELF,"1st MatMatSolve: Norm of error %g, nsolve %d\n",norm,nsolve);CHKERRQ(ierr);
}
}
}
}
/* Test MatSolve() */
if (testMatSolve) {
for (nsolve = 0; nsolve < 2; nsolve++) {
ierr = VecGetArray(x,&array);CHKERRQ(ierr);
for (i=0; i<m; i++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
array[i] = rval;
}
ierr = VecRestoreArray(x,&array);CHKERRQ(ierr);
ierr = VecCopy(x,u);CHKERRQ(ierr);
ierr = MatMult(A,x,b);CHKERRQ(ierr);
if (!rank) printf(" %d-the MatSolve \n",nsolve);
ierr = MatSolve(F,b,x);CHKERRQ(ierr);
/* Check the error */
ierr = VecAXPY(u,-1.0,x);CHKERRQ(ierr); /* u <- (-1.0)x + u */
ierr = VecNorm(u,NORM_2,&norm);CHKERRQ(ierr);
if (norm > tol) {
ierr = MatMult(A,x,u);CHKERRQ(ierr); /* u = A*x */
PetscReal resi;
ierr = VecAXPY(u,-1.0,b);CHKERRQ(ierr); /* u <- (-1.0)b + u */
ierr = VecNorm(u,NORM_2,&resi);CHKERRQ(ierr);
if (!rank) {
ierr = PetscPrintf(PETSC_COMM_SELF,"MatSolve: Norm of error %g, resi %g, LU numfact %d\n",norm,resi,nfact);CHKERRQ(ierr);
}
}
}
}
}
/* Free data structures */
ierr = MatDestroy(&A);CHKERRQ(ierr);
ierr = MatDestroy(&C);CHKERRQ(ierr);
ierr = MatDestroy(&F);CHKERRQ(ierr);
ierr = MatDestroy(&X);CHKERRQ(ierr);
if (testMatMatSolve) {
ierr = MatDestroy(&RHS);CHKERRQ(ierr);
}
ierr = PetscRandomDestroy(&rand);CHKERRQ(ierr);
ierr = ISDestroy(&perm);CHKERRQ(ierr);
ierr = ISDestroy(&iperm);CHKERRQ(ierr);
ierr = VecDestroy(&x);CHKERRQ(ierr);
ierr = VecDestroy(&b);CHKERRQ(ierr);
ierr = VecDestroy(&u);CHKERRQ(ierr);
ierr = PetscFinalize();
return 0;
}
int main(int argc,char **args)
{
const int iM11=0, iM12=1, iL11=2, iL22=3, iL21=4, ix=5;
Mat M11,M12,L11,L22,L21; /* matrix */
Vec x,y; /* input and output vectors */
Vec omg1,omg2,omg3,omg4; /* temporary vectors for the operation y=Ax */
KSP ksp; /* linear solver context */
PetscViewer fd[5]; /* viewer */
char file[6][PETSC_MAX_PATH_LEN]; /* input file name */
PetscErrorCode ierr;
PetscInt M, N; /* number of rows and columns of the GLOBAL matrices (they should be the same) */
PetscInt m, n; /* number of rows and columns of the LOCAL matrices */
PetscInt istart, iend; /* ownership row range of the process using GLOBAL indexes */
PetscScalar one=1.0;
PetscMPIInt rank, size;
PetscBool flg[6];
PetscInitialize(&argc,&args,(char *)0,help);
ierr = MPI_Comm_rank(PETSC_COMM_WORLD,&rank);CHKERRQ(ierr);
ierr = MPI_Comm_size(PETSC_COMM_WORLD,&size);CHKERRQ(ierr);
PetscPrintf(PETSC_COMM_WORLD, "Number of processes size=%d\n", size);
// ************ READ MATRICES AND VECTOR x FROM INPUT *****
// M11 and M12
ierr = PetscOptionsGetString(PETSC_NULL,"-m11",file[iM11],PETSC_MAX_PATH_LEN,&flg[iM11]);CHKERRQ(ierr);
ierr = PetscOptionsGetString(PETSC_NULL,"-m12",file[iM12],PETSC_MAX_PATH_LEN,&flg[iM12]);CHKERRQ(ierr);
// L11 L22 and L21
ierr = PetscOptionsGetString(PETSC_NULL,"-l11",file[iL11],PETSC_MAX_PATH_LEN,&flg[iL11]);CHKERRQ(ierr);
ierr = PetscOptionsGetString(PETSC_NULL,"-l22",file[iL22],PETSC_MAX_PATH_LEN,&flg[iL22]);CHKERRQ(ierr);
ierr = PetscOptionsGetString(PETSC_NULL,"-l21",file[iL21],PETSC_MAX_PATH_LEN,&flg[iL21]);CHKERRQ(ierr);
// All of the matrix have to be defined by the user.
// If the user don't specify none of them, it will generate laplacian matrices.
if (flg[iM11] && flg[iM12] && flg[iL11] && flg[iL22] && flg[iL21]){
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file[iM11],FILE_MODE_READ,\
&fd[iM11]);CHKERRQ(ierr);
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file[iM12],FILE_MODE_READ,\
&fd[iM12]);CHKERRQ(ierr);
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file[iL11],FILE_MODE_READ,\
&fd[iL11]);CHKERRQ(ierr);
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file[iL22],FILE_MODE_READ,\
&fd[iL22]);CHKERRQ(ierr);
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file[iL21],FILE_MODE_READ,\
&fd[iL21]);CHKERRQ(ierr);
// Load the matrix and vector; then destroy the viewer.
// M11 and M12
ierr = MatCreate(PETSC_COMM_WORLD,&M11);CHKERRQ(ierr);
ierr = MatSetType(M11,MATAIJ);CHKERRQ(ierr);
ierr = MatSetFromOptions(M11);CHKERRQ(ierr);
ierr = MatLoad(M11,fd[iM11]);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&M12);CHKERRQ(ierr);
ierr = MatSetType(M12,MATAIJ);CHKERRQ(ierr);
ierr = MatSetFromOptions(M12);CHKERRQ(ierr);
ierr = MatLoad(M12,fd[iM12]);CHKERRQ(ierr);
// L11 L22 and L21
ierr = MatCreate(PETSC_COMM_WORLD,&L11);CHKERRQ(ierr);
ierr = MatSetType(L11,MATAIJ);CHKERRQ(ierr);
ierr = MatSetFromOptions(L11);CHKERRQ(ierr);
ierr = MatLoad(L11,fd[iL11]);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&L22);CHKERRQ(ierr);
ierr = MatSetType(L22,MATAIJ);CHKERRQ(ierr);
ierr = MatSetFromOptions(L22);CHKERRQ(ierr);
ierr = MatLoad(L22,fd[iL22]);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&L21);CHKERRQ(ierr);
ierr = MatSetType(L21,MATAIJ);CHKERRQ(ierr);
ierr = MatSetFromOptions(L21);CHKERRQ(ierr);
ierr = MatLoad(L21,fd[iL21]);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd[iM11]);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd[iM12]);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd[iL11]);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd[iL22]);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd[iL21]);CHKERRQ(ierr);
}
else if(!flg[iM11] && !flg[iM12] && !flg[iL11] && !flg[iL22] && !flg[iL21]){
// ******************* CREATING FAKE MATRICES *****************
PetscInt i,col[3];
M = N = 100;
PetscScalar value[3];
ierr = MatCreate(PETSC_COMM_WORLD,&M11);CHKERRQ(ierr);
ierr = MatSetSizes(M11,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(M11);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&M12);CHKERRQ(ierr);
ierr = MatSetSizes(M12,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(M12);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&L11);CHKERRQ(ierr);
ierr = MatSetSizes(L11,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(L11);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&L22);CHKERRQ(ierr);
ierr = MatSetSizes(L22,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(L22);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&L21);CHKERRQ(ierr);
ierr = MatSetSizes(L21,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(L21);CHKERRQ(ierr);
// Set values for them
value[0] = -1.0; value[1] = 2.0; value[2] = -1.0;
for (i=1; i<M-1; i++) {
col[0] = i-1; col[1] = i; col[2] = i+1;
ierr = MatSetValues(M11,1,&i,3,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(M12,1,&i,3,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L11,1,&i,3,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L22,1,&i,3,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L21,1,&i,3,col,value,INSERT_VALUES);CHKERRQ(ierr);
}
i = N - 1; col[0] = N - 2; col[1] = N - 1;
ierr = MatSetValues(M11,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(M12,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L11,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L22,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L21,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
i = 0; col[0] = 0; col[1] = 1; value[0] = 2.0; value[1] = -1.0;
ierr = MatSetValues(M11,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(M12,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L11,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L22,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(L21,1,&i,2,col,value,INSERT_VALUES);CHKERRQ(ierr);
ierr = MatAssemblyBegin(M11,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(M11,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyBegin(M12,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(M12,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyBegin(L11,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(L11,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyBegin(L22,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(L22,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyBegin(L21,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(L21,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
// ******************** END CREATING FAKE MATRICES *************
}
else{
SETERRQ(PETSC_COMM_WORLD,1,"You must either indicate the ascii file for each matrix M11 M12 L11 L22 L21 using the option -m11 -m12 .... or without any of these options.");
PetscFinalize();
return 1;
}
// Get some information about the partitioning of the matrix
ierr = MatGetSize(M11,&M,&N);CHKERRQ(ierr);
printf("Global dimension of the matrix M11 M=%d N=%d\n",M,N);
ierr = MatGetLocalSize(M11,&m,&n);
printf("Local dimension of the matrix M11 m=%d n=%d\n",m,n);
ierr = MatGetOwnershipRange(M11,&istart,&iend);
printf("Ownership range of the rows for process %d istart=%d iend=%d\n",rank,istart,iend);
// Read vector x from input. If it's not specified by the user, the vector x will be a unitary vector.
ierr = PetscOptionsGetString(PETSC_NULL,"-x",file[ix],PETSC_MAX_PATH_LEN,&flg[ix]);CHKERRQ(ierr);
if(!flg[ix]){
ierr = VecCreate(PETSC_COMM_WORLD,&x);CHKERRQ(ierr);
ierr = VecSetSizes(x,PETSC_DECIDE,M);CHKERRQ(ierr);
ierr = VecSetFromOptions(x);CHKERRQ(ierr);
ierr = VecSet(x,one);CHKERRQ(ierr);
/*ierr = VecAssemblyBegin(x);CHKERRQ(ierr);*/
/*ierr = VecAssemblyEnd(x);CHKERRQ(ierr);*/
}
else{
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file[ix],FILE_MODE_READ,&fd[ix]);CHKERRQ(ierr);
ierr = VecLoad(x,fd[ix]);CHKERRQ(ierr);
}
ierr = PetscObjectSetName((PetscObject) x, "The input vector");CHKERRQ(ierr);
// ************ END READ MATRICES AND VECTOR x FROM INPUT *****
// Create the temporary vectors and y
ierr = VecDuplicate(x,&y);CHKERRQ(ierr);
ierr = VecDuplicate(x,&omg1);CHKERRQ(ierr);
ierr = VecDuplicate(x,&omg2);CHKERRQ(ierr);
ierr = VecDuplicate(x,&omg3);CHKERRQ(ierr);
ierr = VecDuplicate(x,&omg4);CHKERRQ(ierr);
// ****************** COMPUTE y=Ax *******************
// Set the Krylov object
ierr = KSPCreate(PETSC_COMM_WORLD,&ksp);CHKERRQ(ierr);
// Set operators. Here the matrix that defines the linear system
// also serves as the preconditioning matrix.
ierr = KSPSetOperators(ksp,L22,L22,DIFFERENT_NONZERO_PATTERN);CHKERRQ(ierr);
// Set runtime options, e.g.,
// -ksp_type <type> -pc_type <type> -ksp_monitor -ksp_rtol <rtol>
// These options will override those specified above as long as
// KSPSetFromOptions() is called _after_ any other customization
// routines.
ierr = KSPSetFromOptions(ksp);CHKERRQ(ierr);
// Multiplication y = A*x
// omg1 = M11*x
// omg2 = L21*x
// L22*omg3 = omg2
// omg4 = omg1 + M12*omg3
// L11*y = omg4
ierr = MatMult(M11, x, omg1);CHKERRQ(ierr);
ierr = MatMult(L21, x, omg2);CHKERRQ(ierr);
ierr = KSPSolve(ksp,omg2,omg3);CHKERRQ(ierr);
ierr = MatMult(M12, omg3, omg4);CHKERRQ(ierr);
ierr = VecAXPY(omg4, one, omg1);CHKERRQ(ierr);
ierr = KSPSetOperators(ksp,L11,L11,DIFFERENT_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = KSPSolve(ksp,omg4,y);CHKERRQ(ierr);
// ****************** END COMPUTE y=Ax **************************
/*ierr = VecView(y, PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);*/
PetscBool test = PETSC_FALSE;
ierr = PetscOptionsGetBool(PETSC_NULL,"-test",&test,PETSC_NULL);CHKERRQ(ierr);
if(test){
// ******************** TESTING *********************************
// the testing doesn't work if the number of process are more than one
// because the type of the matrices must be different from matmpiaij. Let's try matseqaij
Mat L11_inv, L22_inv, I;
Mat A;
Mat M11_d;
Vec y2;
PetscInt i;
PetscScalar val;
// Create identity matrix
ierr = MatCreate(PETSC_COMM_WORLD,&I);CHKERRQ(ierr);
ierr = MatSetType(I, MATDENSE);
ierr = MatSetSizes(I,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(I);CHKERRQ(ierr);
val = 1.0;
for (i=0; i<M; i++) {
ierr = MatSetValues(I,1,&i,1,&i,&val,INSERT_VALUES);CHKERRQ(ierr);
}
ierr = MatAssemblyBegin(I,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(I,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
// Create L11_inv
ierr = MatCreate(PETSC_COMM_WORLD,&L11_inv);CHKERRQ(ierr);
ierr = MatSetType(L11_inv, MATDENSE);
ierr = MatSetSizes(L11_inv,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(L11_inv);CHKERRQ(ierr);
ierr = MatAssemblyBegin(L11_inv,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(L11_inv,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
// Create L22_inv
ierr = MatCreate(PETSC_COMM_WORLD,&L22_inv);CHKERRQ(ierr);
ierr = MatSetType(L22_inv, MATDENSE);
ierr = MatSetSizes(L22_inv,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(L22_inv);CHKERRQ(ierr);
ierr = MatAssemblyBegin(L22_inv,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(L22_inv,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
// Create A
ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr);
ierr = MatSetSizes(A,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(A);CHKERRQ(ierr);
ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
// Define M11_d
ierr = MatCreate(PETSC_COMM_WORLD,&M11_d);CHKERRQ(ierr);
ierr = MatSetType(M11_d, MATDENSE);
ierr = MatSetSizes(M11_d,PETSC_DECIDE,PETSC_DECIDE,M,N);CHKERRQ(ierr);
ierr = MatSetFromOptions(M11_d);CHKERRQ(ierr);
ierr = MatAssemblyBegin(M11_d,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(M11_d,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
// Calculate A = L11^{-1}*(M11 + M12*L22^{-1}*L21)
IS perm, iperm;
MatFactorInfo info;
ierr = MatGetOrdering(L11,MATORDERINGNATURAL,&perm,&iperm);CHKERRQ(ierr);
ierr = MatFactorInfoInitialize(&info); CHKERRQ(ierr);
ierr = MatLUFactor(L11, perm, iperm, &info); CHKERRQ(ierr);
ierr = MatMatSolve(L11, I, L11_inv);CHKERRQ(ierr);
// TODO try to convert L11_inv to be sparse such as matseqaij
// ierr = MatView(L11_inv, PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
ierr = MatGetOrdering(L22,MATORDERINGNATURAL,&perm,&iperm);CHKERRQ(ierr);
ierr = MatFactorInfoInitialize(&info); CHKERRQ(ierr);
ierr = MatLUFactor(L22, perm, iperm, &info); CHKERRQ(ierr);
ierr = MatMatSolve(L22, I, L22_inv);CHKERRQ(ierr);
ierr = MatMatMult(M12, L22_inv, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &A);CHKERRQ(ierr);
ierr = MatMatMult(A, L21, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &A);CHKERRQ(ierr);
ierr = MatConvert(M11, MATDENSE, MAT_INITIAL_MATRIX, &M11_d);
ierr = MatAXPY(A,1.0,M11_d,DIFFERENT_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = MatMatMult(L11_inv, A, MAT_INITIAL_MATRIX, PETSC_DEFAULT, &A);CHKERRQ(ierr);
ierr = VecDuplicate(x,&y2);CHKERRQ(ierr);
ierr = MatMult(A, x, y2);CHKERRQ(ierr);
/*ierr = VecView(y2, PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);*/
// Check the error
PetscReal norm;
ierr = VecAXPY(y2,-1.0,y);CHKERRQ(ierr);
ierr = VecNorm(y2,NORM_2,&norm);CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_WORLD,"Norm of error %A\n",
norm);CHKERRQ(ierr);
// ******************** END TESTING *****************************
}
PetscFinalize();
return 0;
}