int main(int argc,char **args)
{
Mat C;
PetscErrorCode ierr;
PetscInt i,m = 2,N,M,idx[4],Nsub1,Nsub2,ol=1,x1,x2;
PetscScalar Ke[16];
PetscReal x,y,h;
IS *is1,*is2;
PetscBool flg;
PetscInitialize(&argc,&args,(char *)0,help);
ierr = PetscOptionsGetInt(PETSC_NULL,"-m",&m,PETSC_NULL);
CHKERRQ(ierr);
N = (m+1)*(m+1); /* dimension of matrix */
M = m*m; /* number of elements */
h = 1.0/m; /* mesh width */
x1= (m+1)/2;
x2= x1;
ierr = PetscOptionsGetInt(PETSC_NULL,"-x1",&x1,PETSC_NULL);
CHKERRQ(ierr);
ierr = PetscOptionsGetInt(PETSC_NULL,"-x2",&x2,PETSC_NULL);
CHKERRQ(ierr);
/* create stiffness matrix */
ierr = MatCreateSeqAIJ(PETSC_COMM_SELF,N,N,9,PETSC_NULL,&C);
CHKERRQ(ierr);
/* forms the element stiffness for the Laplacian */
ierr = FormElementStiffness(h*h,Ke);
CHKERRQ(ierr);
for (i=0; i<M; i++) {
/* location of lower left corner of element */
x = h*(i % m);
y = h*(i/m);
/* node numbers for the four corners of element */
idx[0] = (m+1)*(i/m) + (i % m);
idx[1] = idx[0]+1;
idx[2] = idx[1] + m + 1;
idx[3] = idx[2] - 1;
ierr = MatSetValues(C,4,idx,4,idx,Ke,ADD_VALUES);
CHKERRQ(ierr);
}
ierr = MatAssemblyBegin(C,MAT_FINAL_ASSEMBLY);
CHKERRQ(ierr);
ierr = MatAssemblyEnd(C,MAT_FINAL_ASSEMBLY);
CHKERRQ(ierr);
for (ol=0; ol<m+2; ++ol) {
ierr = PCASMCreateSubdomains2D(m+1,m+1,x1,x2,1,0,&Nsub1,&is1);
CHKERRQ(ierr);
ierr = MatIncreaseOverlap(C,Nsub1,is1,ol);
CHKERRQ(ierr);
ierr = PCASMCreateSubdomains2D(m+1,m+1,x1,x2,1,ol,&Nsub2,&is2);
CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_SELF,"flg == 1 => both index sets are same\n");
CHKERRQ(ierr);
if(Nsub1 != Nsub2) {
ierr = PetscPrintf(PETSC_COMM_SELF,"Error: No of indes sets don't match\n");
CHKERRQ(ierr);
}
for (i=0; i<Nsub1; ++i) {
ierr = ISEqual(is1[i],is2[i],&flg);
CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_SELF,"i = %D,flg = %d \n",i,(int)flg);
CHKERRQ(ierr);
}
for (i=0; i<Nsub1; ++i) {
ierr = ISDestroy(&&is1[i]);
CHKERRQ(ierr);
}
for (i=0; i<Nsub2; ++i) {
ierr = ISDestroy(&&is2[i]);
CHKERRQ(ierr);
}
ierr = PetscFree(is1);
CHKERRQ(ierr);
ierr = PetscFree(is2);
CHKERRQ(ierr);
}
ierr = MatDestroy(&C);
CHKERRQ(ierr);
ierr = PetscFinalize();
return 0;
}
int main(int argc,char **args)
{
Mat A,B,*submatA,*submatB;
PetscInt bs=1,m=11,ov=1,i,j,k,*rows,*cols,nd=5,*idx,rstart,rend,sz,mm,nn,M,N,Mbs;
PetscErrorCode ierr;
PetscMPIInt size,rank;
PetscScalar *vals,rval;
IS *is1,*is2;
PetscRandom rdm;
Vec xx,s1,s2;
PetscReal s1norm,s2norm,rnorm,tol = 100*PETSC_SMALL;
PetscBool flg,test_nd0=PETSC_FALSE;
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 = PetscOptionsGetInt(NULL,NULL,"-mat_block_size",&bs,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-mat_size",&m,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-ov",&ov,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-nd",&nd,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetBool(NULL,NULL,"-test_nd0",&test_nd0,NULL);CHKERRQ(ierr);
/* Create a AIJ matrix A */
ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr);
ierr = MatSetSizes(A,m*bs,m*bs,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(A,MATAIJ);CHKERRQ(ierr);
ierr = MatSeqAIJSetPreallocation(A,PETSC_DEFAULT,NULL);CHKERRQ(ierr);
ierr = MatMPIAIJSetPreallocation(A,PETSC_DEFAULT,NULL,PETSC_DEFAULT,NULL);CHKERRQ(ierr);
ierr = MatSetFromOptions(A);CHKERRQ(ierr);
ierr = MatSetOption(A,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_FALSE);CHKERRQ(ierr);
/* Create a BAIJ matrix B */
ierr = MatCreate(PETSC_COMM_WORLD,&B);CHKERRQ(ierr);
ierr = MatSetSizes(B,m*bs,m*bs,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(B,MATBAIJ);CHKERRQ(ierr);
ierr = MatSeqBAIJSetPreallocation(B,bs,PETSC_DEFAULT,NULL);CHKERRQ(ierr);
ierr = MatMPIBAIJSetPreallocation(B,bs,PETSC_DEFAULT,NULL,PETSC_DEFAULT,NULL);CHKERRQ(ierr);
ierr = MatSetFromOptions(B);CHKERRQ(ierr);
ierr = MatSetOption(B,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_FALSE);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_WORLD,&rdm);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rdm);CHKERRQ(ierr);
ierr = MatGetOwnershipRange(A,&rstart,&rend);CHKERRQ(ierr);
ierr = MatGetSize(A,&M,&N);CHKERRQ(ierr);
Mbs = M/bs;
ierr = PetscMalloc1(bs,&rows);CHKERRQ(ierr);
ierr = PetscMalloc1(bs,&cols);CHKERRQ(ierr);
ierr = PetscMalloc1(bs*bs,&vals);CHKERRQ(ierr);
ierr = PetscMalloc1(M,&idx);CHKERRQ(ierr);
/* Now set blocks of values */
for (i=0; i<40*bs; i++) {
ierr = PetscRandomGetValue(rdm,&rval);CHKERRQ(ierr);
cols[0] = bs*(int)(PetscRealPart(rval)*Mbs);
ierr = PetscRandomGetValue(rdm,&rval);CHKERRQ(ierr);
rows[0] = rstart + bs*(int)(PetscRealPart(rval)*m);
for (j=1; j<bs; j++) {
rows[j] = rows[j-1]+1;
cols[j] = cols[j-1]+1;
}
for (j=0; j<bs*bs; j++) {
ierr = PetscRandomGetValue(rdm,&rval);CHKERRQ(ierr);
vals[j] = rval;
}
ierr = MatSetValues(A,bs,rows,bs,cols,vals,ADD_VALUES);CHKERRQ(ierr);
ierr = MatSetValues(B,bs,rows,bs,cols,vals,ADD_VALUES);CHKERRQ(ierr);
}
ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
/* Test MatIncreaseOverlap() */
ierr = PetscMalloc1(nd,&is1);CHKERRQ(ierr);
ierr = PetscMalloc1(nd,&is2);CHKERRQ(ierr);
if (!rank && test_nd0) nd = 0; /* test case */
for (i=0; i<nd; i++) {
ierr = PetscRandomGetValue(rdm,&rval);CHKERRQ(ierr);
sz = (int)(PetscRealPart(rval)*m);
for (j=0; j<sz; j++) {
ierr = PetscRandomGetValue(rdm,&rval);CHKERRQ(ierr);
idx[j*bs] = bs*(int)(PetscRealPart(rval)*Mbs);
for (k=1; k<bs; k++) idx[j*bs+k] = idx[j*bs]+k;
}
ierr = ISCreateGeneral(PETSC_COMM_SELF,sz*bs,idx,PETSC_COPY_VALUES,is1+i);CHKERRQ(ierr);
ierr = ISCreateGeneral(PETSC_COMM_SELF,sz*bs,idx,PETSC_COPY_VALUES,is2+i);CHKERRQ(ierr);
}
ierr = MatIncreaseOverlap(A,nd,is1,ov);CHKERRQ(ierr);
ierr = MatIncreaseOverlap(B,nd,is2,ov);CHKERRQ(ierr);
for (i=0; i<nd; ++i) {
ierr = ISEqual(is1[i],is2[i],&flg);CHKERRQ(ierr);
if (!flg) {
ierr = PetscPrintf(PETSC_COMM_SELF,"i=%D, flg=%d :bs=%D m=%D ov=%D nd=%D np=%D\n",i,flg,bs,m,ov,nd,size);CHKERRQ(ierr);
}
}
for (i=0; i<nd; ++i) {
ierr = ISSort(is1[i]);CHKERRQ(ierr);
ierr = ISSort(is2[i]);CHKERRQ(ierr);
}
ierr = MatCreateSubMatrices(B,nd,is2,is2,MAT_INITIAL_MATRIX,&submatB);CHKERRQ(ierr);
ierr = MatCreateSubMatrices(A,nd,is1,is1,MAT_INITIAL_MATRIX,&submatA);CHKERRQ(ierr);
/* Test MatMult() */
for (i=0; i<nd; i++) {
ierr = MatGetSize(submatA[i],&mm,&nn);CHKERRQ(ierr);
ierr = VecCreateSeq(PETSC_COMM_SELF,mm,&xx);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s1);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s2);CHKERRQ(ierr);
for (j=0; j<3; j++) {
ierr = VecSetRandom(xx,rdm);CHKERRQ(ierr);
ierr = MatMult(submatA[i],xx,s1);CHKERRQ(ierr);
ierr = MatMult(submatB[i],xx,s2);CHKERRQ(ierr);
ierr = VecNorm(s1,NORM_2,&s1norm);CHKERRQ(ierr);
ierr = VecNorm(s2,NORM_2,&s2norm);CHKERRQ(ierr);
rnorm = s2norm-s1norm;
if (rnorm<-tol || rnorm>tol) {
ierr = PetscPrintf(PETSC_COMM_SELF,"[%d]Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",rank,s1norm,s2norm);CHKERRQ(ierr);
}
}
ierr = VecDestroy(&xx);CHKERRQ(ierr);
ierr = VecDestroy(&s1);CHKERRQ(ierr);
ierr = VecDestroy(&s2);CHKERRQ(ierr);
}
/* Now test MatCreateSubmatrices with MAT_REUSE_MATRIX option */
ierr = MatCreateSubMatrices(A,nd,is1,is1,MAT_REUSE_MATRIX,&submatA);CHKERRQ(ierr);
ierr = MatCreateSubMatrices(B,nd,is2,is2,MAT_REUSE_MATRIX,&submatB);CHKERRQ(ierr);
/* Test MatMult() */
for (i=0; i<nd; i++) {
ierr = MatGetSize(submatA[i],&mm,&nn);CHKERRQ(ierr);
ierr = VecCreateSeq(PETSC_COMM_SELF,mm,&xx);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s1);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s2);CHKERRQ(ierr);
for (j=0; j<3; j++) {
ierr = VecSetRandom(xx,rdm);CHKERRQ(ierr);
ierr = MatMult(submatA[i],xx,s1);CHKERRQ(ierr);
ierr = MatMult(submatB[i],xx,s2);CHKERRQ(ierr);
ierr = VecNorm(s1,NORM_2,&s1norm);CHKERRQ(ierr);
ierr = VecNorm(s2,NORM_2,&s2norm);CHKERRQ(ierr);
rnorm = s2norm-s1norm;
if (rnorm<-tol || rnorm>tol) {
ierr = PetscPrintf(PETSC_COMM_SELF,"[%d]Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",rank,s1norm,s2norm);CHKERRQ(ierr);
}
}
ierr = VecDestroy(&xx);CHKERRQ(ierr);
ierr = VecDestroy(&s1);CHKERRQ(ierr);
ierr = VecDestroy(&s2);CHKERRQ(ierr);
}
/* Free allocated memory */
for (i=0; i<nd; ++i) {
ierr = ISDestroy(&is1[i]);CHKERRQ(ierr);
ierr = ISDestroy(&is2[i]);CHKERRQ(ierr);
}
ierr = MatDestroySubMatrices(nd,&submatA);CHKERRQ(ierr);
ierr = MatDestroySubMatrices(nd,&submatB);CHKERRQ(ierr);
ierr = PetscFree(is1);CHKERRQ(ierr);
ierr = PetscFree(is2);CHKERRQ(ierr);
ierr = PetscFree(idx);CHKERRQ(ierr);
ierr = PetscFree(rows);CHKERRQ(ierr);
ierr = PetscFree(cols);CHKERRQ(ierr);
ierr = PetscFree(vals);CHKERRQ(ierr);
ierr = MatDestroy(&A);CHKERRQ(ierr);
ierr = MatDestroy(&B);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&rdm);CHKERRQ(ierr);
ierr = PetscFinalize();
return ierr;
}
int main(int argc,char **args)
{
Mat A,Atrans,sA,*submatA,*submatsA;
PetscErrorCode ierr;
PetscMPIInt size,rank;
PetscInt bs=1,mbs=10,ov=1,i,j,k,*rows,*cols,nd=2,*idx,rstart,rend,sz,M,N,Mbs;
PetscScalar *vals,rval,one=1.0;
IS *is1,*is2;
PetscRandom rand;
PetscBool flg,TestOverlap,TestSubMat,TestAllcols,test_sorted=PETSC_FALSE;
PetscInt vid = -1;
#if defined(PETSC_USE_LOG)
PetscLogStage stages[2];
#endif
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 = PetscOptionsGetInt(NULL,NULL,"-mat_block_size",&bs,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-mat_mbs",&mbs,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-ov",&ov,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-nd",&nd,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-view_id",&vid,NULL);CHKERRQ(ierr);
ierr = PetscOptionsHasName(NULL,NULL, "-test_overlap", &TestOverlap);CHKERRQ(ierr);
ierr = PetscOptionsHasName(NULL,NULL, "-test_submat", &TestSubMat);CHKERRQ(ierr);
ierr = PetscOptionsHasName(NULL,NULL, "-test_allcols", &TestAllcols);CHKERRQ(ierr);
ierr = PetscOptionsGetBool(NULL,NULL,"-test_sorted",&test_sorted,NULL);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr);
ierr = MatSetSizes(A,mbs*bs,mbs*bs,PETSC_DECIDE,PETSC_DECIDE);CHKERRQ(ierr);
ierr = MatSetType(A,MATBAIJ);CHKERRQ(ierr);
ierr = MatSeqBAIJSetPreallocation(A,bs,PETSC_DEFAULT,NULL);CHKERRQ(ierr);
ierr = MatMPIBAIJSetPreallocation(A,bs,PETSC_DEFAULT,NULL,PETSC_DEFAULT,NULL);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_WORLD,&rand);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rand);CHKERRQ(ierr);
ierr = MatGetOwnershipRange(A,&rstart,&rend);CHKERRQ(ierr);
ierr = MatGetSize(A,&M,&N);CHKERRQ(ierr);
Mbs = M/bs;
ierr = PetscMalloc1(bs,&rows);CHKERRQ(ierr);
ierr = PetscMalloc1(bs,&cols);CHKERRQ(ierr);
ierr = PetscMalloc1(bs*bs,&vals);CHKERRQ(ierr);
ierr = PetscMalloc1(M,&idx);CHKERRQ(ierr);
/* Now set blocks of values */
for (j=0; j<bs*bs; j++) vals[j] = 0.0;
for (i=0; i<Mbs; i++) {
cols[0] = i*bs; rows[0] = i*bs;
for (j=1; j<bs; j++) {
rows[j] = rows[j-1]+1;
cols[j] = cols[j-1]+1;
}
ierr = MatSetValues(A,bs,rows,bs,cols,vals,ADD_VALUES);CHKERRQ(ierr);
}
/* second, add random blocks */
for (i=0; i<20*bs; i++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
cols[0] = bs*(PetscInt)(PetscRealPart(rval)*Mbs);
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
rows[0] = rstart + bs*(PetscInt)(PetscRealPart(rval)*mbs);
for (j=1; j<bs; j++) {
rows[j] = rows[j-1]+1;
cols[j] = cols[j-1]+1;
}
for (j=0; j<bs*bs; j++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
vals[j] = rval;
}
ierr = MatSetValues(A,bs,rows,bs,cols,vals,ADD_VALUES);CHKERRQ(ierr);
}
ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
/* make A a symmetric matrix: A <- A^T + A */
ierr = MatTranspose(A,MAT_INITIAL_MATRIX, &Atrans);CHKERRQ(ierr);
ierr = MatAXPY(A,one,Atrans,DIFFERENT_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = MatDestroy(&Atrans);CHKERRQ(ierr);
ierr = MatTranspose(A,MAT_INITIAL_MATRIX, &Atrans);CHKERRQ(ierr);
ierr = MatEqual(A, Atrans, &flg);CHKERRQ(ierr);
if (flg) {
ierr = MatSetOption(A,MAT_SYMMETRIC,PETSC_TRUE);CHKERRQ(ierr);
} else SETERRQ(PETSC_COMM_SELF,1,"A+A^T is non-symmetric");
ierr = MatDestroy(&Atrans);CHKERRQ(ierr);
/* create a SeqSBAIJ matrix sA (= A) */
ierr = MatConvert(A,MATSBAIJ,MAT_INITIAL_MATRIX,&sA);CHKERRQ(ierr);
if (vid >= 0 && vid < size) {
if (!rank) printf("A: \n");
ierr = MatView(A,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
if (!rank) printf("sA: \n");
ierr = MatView(sA,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
}
/* Test sA==A through MatMult() */
ierr = MatMultEqual(A,sA,10,&flg);CHKERRQ(ierr);
if (!flg) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONG,"Error in MatConvert(): A != sA");
/* Test MatIncreaseOverlap() */
ierr = PetscMalloc1(nd,&is1);CHKERRQ(ierr);
ierr = PetscMalloc1(nd,&is2);CHKERRQ(ierr);
for (i=0; i<nd; i++) {
if (!TestAllcols) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
sz = (PetscInt)((0.5+0.2*PetscRealPart(rval))*mbs); /* 0.5*mbs < sz < 0.7*mbs */
for (j=0; j<sz; j++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
idx[j*bs] = bs*(PetscInt)(PetscRealPart(rval)*Mbs);
for (k=1; k<bs; k++) idx[j*bs+k] = idx[j*bs]+k;
}
ierr = ISCreateGeneral(PETSC_COMM_SELF,sz*bs,idx,PETSC_COPY_VALUES,is1+i);CHKERRQ(ierr);
ierr = ISCreateGeneral(PETSC_COMM_SELF,sz*bs,idx,PETSC_COPY_VALUES,is2+i);CHKERRQ(ierr);
if (rank == vid) {
ierr = PetscPrintf(PETSC_COMM_SELF," [%d] IS sz[%d]: %d\n",rank,i,sz);CHKERRQ(ierr);
ierr = ISView(is2[i],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
}
} else { /* Test all rows and colums */
sz = M;
ierr = ISCreateStride(PETSC_COMM_SELF,sz,0,1,is1+i);CHKERRQ(ierr);
ierr = ISCreateStride(PETSC_COMM_SELF,sz,0,1,is2+i);CHKERRQ(ierr);
if (rank == vid) {
PetscBool colflag;
ierr = ISIdentity(is2[i],&colflag);CHKERRQ(ierr);
printf("[%d] is2[%d], colflag %d\n",rank,(int)i,(int)colflag);
ierr = ISView(is2[i],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
}
}
}
ierr = PetscLogStageRegister("MatOv_SBAIJ",&stages[0]);CHKERRQ(ierr);
ierr = PetscLogStageRegister("MatOv_BAIJ",&stages[1]);CHKERRQ(ierr);
/* Test MatIncreaseOverlap */
if (TestOverlap) {
ierr = PetscLogStagePush(stages[0]);CHKERRQ(ierr);
ierr = MatIncreaseOverlap(sA,nd,is2,ov);CHKERRQ(ierr);
ierr = PetscLogStagePop();CHKERRQ(ierr);
ierr = PetscLogStagePush(stages[1]);CHKERRQ(ierr);
ierr = MatIncreaseOverlap(A,nd,is1,ov);CHKERRQ(ierr);
ierr = PetscLogStagePop();CHKERRQ(ierr);
if (rank == vid) {
printf("\n[%d] IS from BAIJ:\n",rank);
ierr = ISView(is1[0],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
printf("\n[%d] IS from SBAIJ:\n",rank);
ierr = ISView(is2[0],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
}
for (i=0; i<nd; ++i) {
ierr = ISEqual(is1[i],is2[i],&flg);CHKERRQ(ierr);
if (!flg) {
if (!rank) {
ierr = ISSort(is1[i]);CHKERRQ(ierr);
/* ISView(is1[i],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr); */
ierr = ISSort(is2[i]);CHKERRQ(ierr);
/* ISView(is2[i],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr); */
}
SETERRQ1(PETSC_COMM_SELF,1,"i=%D, is1 != is2",i);
}
}
}
/* Test MatCreateSubmatrices */
if (TestSubMat) {
if (test_sorted) {
for (i = 0; i < nd; ++i) {
ierr = ISSort(is1[i]);CHKERRQ(ierr);
}
}
ierr = MatCreateSubMatrices(A,nd,is1,is1,MAT_INITIAL_MATRIX,&submatA);CHKERRQ(ierr);
ierr = MatCreateSubMatrices(sA,nd,is1,is1,MAT_INITIAL_MATRIX,&submatsA);CHKERRQ(ierr);
ierr = MatMultEqual(A,sA,10,&flg);CHKERRQ(ierr);
if (!flg) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"A != sA");
/* Now test MatCreateSubmatrices with MAT_REUSE_MATRIX option */
ierr = MatCreateSubMatrices(A,nd,is1,is1,MAT_REUSE_MATRIX,&submatA);CHKERRQ(ierr);
ierr = MatCreateSubMatrices(sA,nd,is1,is1,MAT_REUSE_MATRIX,&submatsA);CHKERRQ(ierr);
ierr = MatMultEqual(A,sA,10,&flg);CHKERRQ(ierr);
if (!flg) SETERRQ(PETSC_COMM_SELF,PETSC_ERR_ARG_WRONGSTATE,"MatCreateSubmatrices(): A != sA");
ierr = MatDestroySubMatrices(nd,&submatA);CHKERRQ(ierr);
ierr = MatDestroySubMatrices(nd,&submatsA);CHKERRQ(ierr);
}
/* Free allocated memory */
for (i=0; i<nd; ++i) {
ierr = ISDestroy(&is1[i]);CHKERRQ(ierr);
ierr = ISDestroy(&is2[i]);CHKERRQ(ierr);
}
ierr = PetscFree(is1);CHKERRQ(ierr);
ierr = PetscFree(is2);CHKERRQ(ierr);
ierr = PetscFree(idx);CHKERRQ(ierr);
ierr = PetscFree(rows);CHKERRQ(ierr);
ierr = PetscFree(cols);CHKERRQ(ierr);
ierr = PetscFree(vals);CHKERRQ(ierr);
ierr = MatDestroy(&A);CHKERRQ(ierr);
ierr = MatDestroy(&sA);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&rand);CHKERRQ(ierr);
ierr = PetscFinalize();
return ierr;
}
PetscErrorCode PCGAMGProlongator_Classical_Standard(PC pc, const Mat A, const Mat G, PetscCoarsenData *agg_lists,Mat *P)
{
PetscErrorCode ierr;
Mat *lA;
Vec lv,v,cv;
PetscScalar *lcid;
IS lis;
PetscInt fs,fe,cs,ce,nl,i,j,k,li,lni,ci;
VecScatter lscat;
PetscInt fn,cn,cid,c_indx;
PetscBool iscoarse;
PetscScalar c_scalar;
const PetscScalar *vcol;
const PetscInt *icol;
const PetscInt *gidx;
PetscInt ncols;
PetscInt *lsparse,*gsparse;
MatType mtype;
PetscInt maxcols;
PetscReal diag,jdiag,jwttotal;
PetscScalar *pvcol,vi;
PetscInt *picol;
PetscInt pncols;
PetscScalar *pcontrib,pentry,pjentry;
/* PC_MG *mg = (PC_MG*)pc->data; */
/* PC_GAMG *gamg = (PC_GAMG*)mg->innerctx; */
PetscFunctionBegin;
ierr = MatGetOwnershipRange(A,&fs,&fe);CHKERRQ(ierr);
fn = fe-fs;
ierr = MatGetVecs(A,NULL,&v);CHKERRQ(ierr);
ierr = ISCreateStride(PETSC_COMM_SELF,fe-fs,fs,1,&lis);CHKERRQ(ierr);
/* increase the overlap by two to get neighbors of neighbors */
ierr = MatIncreaseOverlap(A,1,&lis,2);CHKERRQ(ierr);
ierr = ISSort(lis);CHKERRQ(ierr);
/* get the local part of A */
ierr = MatGetSubMatrices(A,1,&lis,&lis,MAT_INITIAL_MATRIX,&lA);CHKERRQ(ierr);
/* build the scatter out of it */
ierr = ISGetLocalSize(lis,&nl);CHKERRQ(ierr);
ierr = VecCreateSeq(PETSC_COMM_SELF,nl,&lv);CHKERRQ(ierr);
ierr = VecScatterCreate(v,lis,lv,NULL,&lscat);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscInt)*fn,&lsparse);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscInt)*fn,&gsparse);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscScalar)*nl,&pcontrib);CHKERRQ(ierr);
/* create coarse vector */
cn = 0;
for (i=0;i<fn;i++) {
ierr = PetscCDEmptyAt(agg_lists,i,&iscoarse);CHKERRQ(ierr);
if (!iscoarse) {
cn++;
}
}
ierr = VecCreateMPI(PetscObjectComm((PetscObject)A),cn,PETSC_DECIDE,&cv);CHKERRQ(ierr);
ierr = VecGetOwnershipRange(cv,&cs,&ce);CHKERRQ(ierr);
cn = 0;
for (i=0;i<fn;i++) {
ierr = PetscCDEmptyAt(agg_lists,i,&iscoarse); CHKERRQ(ierr);
if (!iscoarse) {
cid = cs+cn;
cn++;
} else {
cid = -1;
}
*(PetscInt*)&c_scalar = cid;
c_indx = fs+i;
ierr = VecSetValues(v,1,&c_indx,&c_scalar,INSERT_VALUES);CHKERRQ(ierr);
}
ierr = VecScatterBegin(lscat,v,lv,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
ierr = VecScatterEnd(lscat,v,lv,INSERT_VALUES,SCATTER_FORWARD);CHKERRQ(ierr);
/* count to preallocate the prolongator */
ierr = ISGetIndices(lis,&gidx);CHKERRQ(ierr);
ierr = VecGetArray(lv,&lcid);CHKERRQ(ierr);
maxcols = 0;
/* count the number of unique contributing coarse cells for each fine */
for (i=0;i<nl;i++) {
pcontrib[i] = 0.;
ierr = MatGetRow(lA[0],i,&ncols,&icol,NULL);CHKERRQ(ierr);
if (gidx[i] >= fs && gidx[i] < fe) {
li = gidx[i] - fs;
lsparse[li] = 0;
gsparse[li] = 0;
cid = *(PetscInt*)&(lcid[i]);
if (cid >= 0) {
lsparse[li] = 1;
} else {
for (j=0;j<ncols;j++) {
if (*(PetscInt*)&(lcid[icol[j]]) >= 0) {
pcontrib[icol[j]] = 1.;
} else {
ci = icol[j];
ierr = MatRestoreRow(lA[0],i,&ncols,&icol,NULL);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],ci,&ncols,&icol,NULL);CHKERRQ(ierr);
for (k=0;k<ncols;k++) {
if (*(PetscInt*)&(lcid[icol[k]]) >= 0) {
pcontrib[icol[k]] = 1.;
}
}
ierr = MatRestoreRow(lA[0],ci,&ncols,&icol,NULL);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],i,&ncols,&icol,NULL);CHKERRQ(ierr);
}
}
for (j=0;j<ncols;j++) {
if (*(PetscInt*)&(lcid[icol[j]]) >= 0 && pcontrib[icol[j]] != 0.) {
lni = *(PetscInt*)&(lcid[icol[j]]);
if (lni >= cs && lni < ce) {
lsparse[li]++;
} else {
gsparse[li]++;
}
pcontrib[icol[j]] = 0.;
} else {
ci = icol[j];
ierr = MatRestoreRow(lA[0],i,&ncols,&icol,NULL);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],ci,&ncols,&icol,NULL);CHKERRQ(ierr);
for (k=0;k<ncols;k++) {
if (*(PetscInt*)&(lcid[icol[k]]) >= 0 && pcontrib[icol[k]] != 0.) {
lni = *(PetscInt*)&(lcid[icol[k]]);
if (lni >= cs && lni < ce) {
lsparse[li]++;
} else {
gsparse[li]++;
}
pcontrib[icol[k]] = 0.;
}
}
ierr = MatRestoreRow(lA[0],ci,&ncols,&icol,NULL);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],i,&ncols,&icol,NULL);CHKERRQ(ierr);
}
}
}
if (lsparse[li] + gsparse[li] > maxcols) maxcols = lsparse[li]+gsparse[li];
}
ierr = MatRestoreRow(lA[0],i,&ncols,&icol,&vcol);CHKERRQ(ierr);
}
ierr = PetscMalloc(sizeof(PetscInt)*maxcols,&picol);CHKERRQ(ierr);
ierr = PetscMalloc(sizeof(PetscScalar)*maxcols,&pvcol);CHKERRQ(ierr);
ierr = MatCreate(PetscObjectComm((PetscObject)A),P);CHKERRQ(ierr);
ierr = MatGetType(A,&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);
for (i=0;i<nl;i++) {
diag = 0.;
if (gidx[i] >= fs && gidx[i] < fe) {
li = gidx[i] - fs;
pncols=0;
cid = *(PetscInt*)&(lcid[i]);
if (cid >= 0) {
pncols = 1;
picol[0] = cid;
pvcol[0] = 1.;
} else {
ierr = MatGetRow(lA[0],i,&ncols,&icol,&vcol);CHKERRQ(ierr);
for (j=0;j<ncols;j++) {
pentry = vcol[j];
if (*(PetscInt*)&(lcid[icol[j]]) >= 0) {
/* coarse neighbor */
pcontrib[icol[j]] += pentry;
} else if (icol[j] != i) {
/* the neighbor is a strongly connected fine node */
ci = icol[j];
vi = vcol[j];
ierr = MatRestoreRow(lA[0],i,&ncols,&icol,&vcol);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],ci,&ncols,&icol,&vcol);CHKERRQ(ierr);
jwttotal=0.;
jdiag = 0.;
for (k=0;k<ncols;k++) {
if (ci == icol[k]) {
jdiag = PetscRealPart(vcol[k]);
}
}
for (k=0;k<ncols;k++) {
if (*(PetscInt*)&(lcid[icol[k]]) >= 0 && jdiag*PetscRealPart(vcol[k]) < 0.) {
pjentry = vcol[k];
jwttotal += PetscRealPart(pjentry);
}
}
if (jwttotal != 0.) {
jwttotal = PetscRealPart(vi)/jwttotal;
for (k=0;k<ncols;k++) {
if (*(PetscInt*)&(lcid[icol[k]]) >= 0 && jdiag*PetscRealPart(vcol[k]) < 0.) {
pjentry = vcol[k]*jwttotal;
pcontrib[icol[k]] += pjentry;
}
}
} else {
diag += PetscRealPart(vi);
}
ierr = MatRestoreRow(lA[0],ci,&ncols,&icol,&vcol);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],i,&ncols,&icol,&vcol);CHKERRQ(ierr);
} else {
diag += PetscRealPart(vcol[j]);
}
}
if (diag != 0.) {
diag = 1./diag;
for (j=0;j<ncols;j++) {
if (*(PetscInt*)&(lcid[icol[j]]) >= 0 && pcontrib[icol[j]] != 0.) {
/* the neighbor is a coarse node */
if (PetscAbsScalar(pcontrib[icol[j]]) > 0.0) {
lni = *(PetscInt*)&(lcid[icol[j]]);
pvcol[pncols] = -pcontrib[icol[j]]*diag;
picol[pncols] = lni;
pncols++;
}
pcontrib[icol[j]] = 0.;
} else {
/* the neighbor is a strongly connected fine node */
ci = icol[j];
ierr = MatRestoreRow(lA[0],i,&ncols,&icol,&vcol);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],ci,&ncols,&icol,&vcol);CHKERRQ(ierr);
for (k=0;k<ncols;k++) {
if (*(PetscInt*)&(lcid[icol[k]]) >= 0 && pcontrib[icol[k]] != 0.) {
if (PetscAbsScalar(pcontrib[icol[k]]) > 0.0) {
lni = *(PetscInt*)&(lcid[icol[k]]);
pvcol[pncols] = -pcontrib[icol[k]]*diag;
picol[pncols] = lni;
pncols++;
}
pcontrib[icol[k]] = 0.;
}
}
ierr = MatRestoreRow(lA[0],ci,&ncols,&icol,&vcol);CHKERRQ(ierr);
ierr = MatGetRow(lA[0],i,&ncols,&icol,&vcol);CHKERRQ(ierr);
}
pcontrib[icol[j]] = 0.;
}
ierr = MatRestoreRow(lA[0],i,&ncols,&icol,&vcol);CHKERRQ(ierr);
}
}
ci = gidx[i];
li = gidx[i] - fs;
if (pncols > 0) {
ierr = MatSetValues(*P,1,&ci,pncols,picol,pvcol,INSERT_VALUES);CHKERRQ(ierr);
}
}
}
ierr = ISRestoreIndices(lis,&gidx);CHKERRQ(ierr);
ierr = VecRestoreArray(lv,&lcid);CHKERRQ(ierr);
ierr = PetscFree(pcontrib);CHKERRQ(ierr);
ierr = PetscFree(picol);CHKERRQ(ierr);
ierr = PetscFree(pvcol);CHKERRQ(ierr);
ierr = PetscFree(lsparse);CHKERRQ(ierr);
ierr = PetscFree(gsparse);CHKERRQ(ierr);
ierr = ISDestroy(&lis);CHKERRQ(ierr);
ierr = MatDestroyMatrices(1,&lA);CHKERRQ(ierr);
ierr = VecDestroy(&lv);CHKERRQ(ierr);
ierr = VecDestroy(&cv);CHKERRQ(ierr);
ierr = VecDestroy(&v);CHKERRQ(ierr);
ierr = VecScatterDestroy(&lscat);CHKERRQ(ierr);
ierr = MatAssemblyBegin(*P, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(*P, MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
/*
Mat Pold;
ierr = PCGAMGProlongator_Classical(pc,A,G,agg_lists,&Pold);CHKERRQ(ierr);
ierr = MatView(Pold,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
ierr = MatView(*P,PETSC_VIEWER_STDOUT_WORLD);CHKERRQ(ierr);
ierr = MatDestroy(&Pold);CHKERRQ(ierr);
*/
PetscFunctionReturn(0);
}
int main(int argc,char **args)
{
Mat A,Atrans,sA,*submatA,*submatsA;
PetscInt bs=1,m=43,ov=1,i,j,k,*rows,*cols,M,nd=5,*idx,mm,nn;
PetscErrorCode ierr;
PetscMPIInt size;
PetscScalar *vals,rval,one=1.0;
IS *is1,*is2;
PetscRandom rand;
Vec xx,s1,s2;
PetscReal s1norm,s2norm,rnorm,tol = 1.e-10;
PetscBool flg;
PetscInitialize(&argc,&args,(char *)0,help);
ierr = PetscOptionsGetInt(PETSC_NULL,"-mat_block_size",&bs,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(PETSC_NULL,"-mat_size",&m,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(PETSC_NULL,"-ov",&ov,PETSC_NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(PETSC_NULL,"-nd",&nd,PETSC_NULL);CHKERRQ(ierr);
/* create a SeqBAIJ matrix A */
M = m*bs;
ierr = MatCreateSeqBAIJ(PETSC_COMM_SELF,bs,M,M,1,PETSC_NULL,&A);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_SELF,&rand);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rand);CHKERRQ(ierr);
ierr = PetscMalloc(bs*sizeof(PetscInt),&rows);CHKERRQ(ierr);
ierr = PetscMalloc(bs*sizeof(PetscInt),&cols);CHKERRQ(ierr);
ierr = PetscMalloc(bs*bs*sizeof(PetscScalar),&vals);CHKERRQ(ierr);
ierr = PetscMalloc(M*sizeof(PetscScalar),&idx);CHKERRQ(ierr);
/* Now set blocks of random values */
/* first, set diagonal blocks as zero */
for (j=0; j<bs*bs; j++) vals[j] = 0.0;
for (i=0; i<m; i++){
cols[0] = i*bs; rows[0] = i*bs;
for (j=1; j<bs; j++) {
rows[j] = rows[j-1]+1;
cols[j] = cols[j-1]+1;
}
ierr = MatSetValues(A,bs,rows,bs,cols,vals,ADD_VALUES);CHKERRQ(ierr);
}
/* second, add random blocks */
for (i=0; i<20*bs; i++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
cols[0] = bs*(int)(PetscRealPart(rval)*m);
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
rows[0] = bs*(int)(PetscRealPart(rval)*m);
for (j=1; j<bs; j++) {
rows[j] = rows[j-1]+1;
cols[j] = cols[j-1]+1;
}
for (j=0; j<bs*bs; j++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
vals[j] = rval;
}
ierr = MatSetValues(A,bs,rows,bs,cols,vals,ADD_VALUES);CHKERRQ(ierr);
}
ierr = MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
ierr = MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr);
/* make A a symmetric matrix: A <- A^T + A */
ierr = MatTranspose(A,MAT_INITIAL_MATRIX, &Atrans);CHKERRQ(ierr);
ierr = MatAXPY(A,one,Atrans,DIFFERENT_NONZERO_PATTERN);CHKERRQ(ierr);
ierr = MatDestroy(&Atrans);CHKERRQ(ierr);
ierr = MatTranspose(A,MAT_INITIAL_MATRIX, &Atrans);
ierr = MatEqual(A, Atrans, &flg);
if (!flg) {
SETERRQ(PETSC_COMM_SELF,1,"A+A^T is non-symmetric");
}
ierr = MatDestroy(&Atrans);CHKERRQ(ierr);
/* create a SeqSBAIJ matrix sA (= A) */
ierr = MatConvert(A,MATSEQSBAIJ,MAT_INITIAL_MATRIX,&sA);CHKERRQ(ierr);
/* Test sA==A through MatMult() */
for (i=0; i<nd; i++) {
ierr = MatGetSize(A,&mm,&nn);CHKERRQ(ierr);
ierr = VecCreateSeq(PETSC_COMM_SELF,mm,&xx);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s1);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s2);CHKERRQ(ierr);
for (j=0; j<3; j++) {
ierr = VecSetRandom(xx,rand);CHKERRQ(ierr);
ierr = MatMult(A,xx,s1);CHKERRQ(ierr);
ierr = MatMult(sA,xx,s2);CHKERRQ(ierr);
ierr = VecNorm(s1,NORM_2,&s1norm);CHKERRQ(ierr);
ierr = VecNorm(s2,NORM_2,&s2norm);CHKERRQ(ierr);
rnorm = s2norm-s1norm;
if (rnorm<-tol || rnorm>tol) {
ierr = PetscPrintf(PETSC_COMM_SELF,"Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",s1norm,s2norm);CHKERRQ(ierr);
}
}
ierr = VecDestroy(&xx);CHKERRQ(ierr);
ierr = VecDestroy(&s1);CHKERRQ(ierr);
ierr = VecDestroy(&s2);CHKERRQ(ierr);
}
/* Test MatIncreaseOverlap() */
ierr = PetscMalloc(nd*sizeof(IS **),&is1);CHKERRQ(ierr);
ierr = PetscMalloc(nd*sizeof(IS **),&is2);CHKERRQ(ierr);
for (i=0; i<nd; i++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
size = (int)(PetscRealPart(rval)*m);
for (j=0; j<size; j++) {
ierr = PetscRandomGetValue(rand,&rval);CHKERRQ(ierr);
idx[j*bs] = bs*(int)(PetscRealPart(rval)*m);
for (k=1; k<bs; k++) idx[j*bs+k] = idx[j*bs]+k;
}
ierr = ISCreateGeneral(PETSC_COMM_SELF,size*bs,idx,PETSC_COPY_VALUES,is1+i);CHKERRQ(ierr);
ierr = ISCreateGeneral(PETSC_COMM_SELF,size*bs,idx,PETSC_COPY_VALUES,is2+i);CHKERRQ(ierr);
}
/* for debugging */
/*
ierr = MatView(A,PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
ierr = MatView(sA,PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
*/
ierr = MatIncreaseOverlap(A,nd,is1,ov);CHKERRQ(ierr);
ierr = MatIncreaseOverlap(sA,nd,is2,ov);CHKERRQ(ierr);
for (i=0; i<nd; ++i) {
ierr = ISSort(is1[i]);CHKERRQ(ierr);
ierr = ISSort(is2[i]);CHKERRQ(ierr);
}
for (i=0; i<nd; ++i) {
ierr = ISEqual(is1[i],is2[i],&flg);CHKERRQ(ierr);
if (!flg){
/* ISView(is1[i],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr);
ISView(is2[i],PETSC_VIEWER_STDOUT_SELF);CHKERRQ(ierr); */
SETERRQ1(PETSC_COMM_SELF,1,"i=%d, is1 != is2",i);
}
}
ierr = MatGetSubMatrices(A,nd,is1,is1,MAT_INITIAL_MATRIX,&submatA);CHKERRQ(ierr);
ierr = MatGetSubMatrices(sA,nd,is2,is2,MAT_INITIAL_MATRIX,&submatsA);CHKERRQ(ierr);
/* Test MatMult() */
for (i=0; i<nd; i++) {
ierr = MatGetSize(submatA[i],&mm,&nn);CHKERRQ(ierr);
ierr = VecCreateSeq(PETSC_COMM_SELF,mm,&xx);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s1);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s2);CHKERRQ(ierr);
for (j=0; j<3; j++) {
ierr = VecSetRandom(xx,rand);CHKERRQ(ierr);
ierr = MatMult(submatA[i],xx,s1);CHKERRQ(ierr);
ierr = MatMult(submatsA[i],xx,s2);CHKERRQ(ierr);
ierr = VecNorm(s1,NORM_2,&s1norm);CHKERRQ(ierr);
ierr = VecNorm(s2,NORM_2,&s2norm);CHKERRQ(ierr);
rnorm = s2norm-s1norm;
if (rnorm<-tol || rnorm>tol) {
ierr = PetscPrintf(PETSC_COMM_SELF,"Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",s1norm,s2norm);CHKERRQ(ierr);
}
}
ierr = VecDestroy(&xx);CHKERRQ(ierr);
ierr = VecDestroy(&s1);CHKERRQ(ierr);
ierr = VecDestroy(&s2);CHKERRQ(ierr);
}
/* Now test MatGetSubmatrices with MAT_REUSE_MATRIX option */
ierr = MatGetSubMatrices(A,nd,is1,is1,MAT_REUSE_MATRIX,&submatA);CHKERRQ(ierr);
ierr = MatGetSubMatrices(sA,nd,is2,is2,MAT_REUSE_MATRIX,&submatsA);CHKERRQ(ierr);
/* Test MatMult() */
for (i=0; i<nd; i++) {
ierr = MatGetSize(submatA[i],&mm,&nn);CHKERRQ(ierr);
ierr = VecCreateSeq(PETSC_COMM_SELF,mm,&xx);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s1);CHKERRQ(ierr);
ierr = VecDuplicate(xx,&s2);CHKERRQ(ierr);
for (j=0; j<3; j++) {
ierr = VecSetRandom(xx,rand);CHKERRQ(ierr);
ierr = MatMult(submatA[i],xx,s1);CHKERRQ(ierr);
ierr = MatMult(submatsA[i],xx,s2);CHKERRQ(ierr);
ierr = VecNorm(s1,NORM_2,&s1norm);CHKERRQ(ierr);
ierr = VecNorm(s2,NORM_2,&s2norm);CHKERRQ(ierr);
rnorm = s2norm-s1norm;
if (rnorm<-tol || rnorm>tol) {
ierr = PetscPrintf(PETSC_COMM_SELF,"Error:MatMult - Norm1=%16.14e Norm2=%16.14e\n",s1norm,s2norm);CHKERRQ(ierr);
}
}
ierr = VecDestroy(&xx);CHKERRQ(ierr);
ierr = VecDestroy(&s1);CHKERRQ(ierr);
ierr = VecDestroy(&s2);CHKERRQ(ierr);
}
/* Free allocated memory */
for (i=0; i<nd; ++i) {
ierr = ISDestroy(&is1[i]);CHKERRQ(ierr);
ierr = ISDestroy(&is2[i]);CHKERRQ(ierr);
ierr = MatDestroy(&submatA[i]);CHKERRQ(ierr);
ierr = MatDestroy(&submatsA[i]);CHKERRQ(ierr);
}
ierr = PetscFree(submatA);CHKERRQ(ierr);
ierr = PetscFree(submatsA);CHKERRQ(ierr);
ierr = PetscFree(is1);CHKERRQ(ierr);
ierr = PetscFree(is2);CHKERRQ(ierr);
ierr = PetscFree(idx);CHKERRQ(ierr);
ierr = PetscFree(rows);CHKERRQ(ierr);
ierr = PetscFree(cols);CHKERRQ(ierr);
ierr = PetscFree(vals);CHKERRQ(ierr);
ierr = MatDestroy(&A);CHKERRQ(ierr);
ierr = MatDestroy(&sA);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&rand);CHKERRQ(ierr);
ierr = PetscFinalize();
return 0;
}
-ov <overlap> : >=0 amount of overlap between domains\n\n";
#include <petscmat.h>
int main(int argc,char **args)
{
PetscErrorCode ierr;
PetscInt nd = 2,ov=1,i,start,m,n,end,lsize;
PetscMPIInt rank;
PetscBool flg;
Mat A,B;
char file[PETSC_MAX_PATH_LEN];
PetscViewer fd;
IS *is1,*is2;
PetscRandom r;
PetscScalar rand;
ierr = PetscInitialize(&argc,&args,(char*)0,help);if (ierr) return ierr;
#if defined(PETSC_USE_COMPLEX)
SETERRQ(PETSC_COMM_WORLD,1,"This example does not work with complex numbers");
#else
ierr = MPI_Comm_rank(PETSC_COMM_WORLD,&rank);CHKERRQ(ierr);
ierr = PetscOptionsGetString(NULL,NULL,"-f",file,PETSC_MAX_PATH_LEN,&flg);CHKERRQ(ierr);
if (!flg) SETERRQ(PETSC_COMM_WORLD,1,"Must use -f filename to indicate a file containing a PETSc binary matrix");
ierr = PetscOptionsGetInt(NULL,NULL,"-nd",&nd,NULL);CHKERRQ(ierr);
ierr = PetscOptionsGetInt(NULL,NULL,"-ov",&ov,NULL);CHKERRQ(ierr);
/* Read matrix and RHS */
ierr = PetscViewerBinaryOpen(PETSC_COMM_WORLD,file,FILE_MODE_READ,&fd);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_WORLD,&A);CHKERRQ(ierr);
ierr = MatSetType(A,MATMPIAIJ);CHKERRQ(ierr);
ierr = MatLoad(A,fd);CHKERRQ(ierr);
ierr = MatSetFromOptions(A);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd);CHKERRQ(ierr);
/* Read the matrix again as a sequential matrix */
ierr = PetscViewerBinaryOpen(PETSC_COMM_SELF,file,FILE_MODE_READ,&fd);CHKERRQ(ierr);
ierr = MatCreate(PETSC_COMM_SELF,&B);CHKERRQ(ierr);
ierr = MatSetType(B,MATSEQAIJ);CHKERRQ(ierr);
ierr = MatLoad(B,fd);CHKERRQ(ierr);
ierr = MatSetFromOptions(B);CHKERRQ(ierr);
ierr = PetscViewerDestroy(&fd);CHKERRQ(ierr);
/* Create the IS corresponding to subdomains */
ierr = PetscMalloc1(nd,&is1);CHKERRQ(ierr);
ierr = PetscMalloc1(nd,&is2);CHKERRQ(ierr);
/* Create the random Index Sets */
ierr = MatGetSize(A,&m,&n);CHKERRQ(ierr);
ierr = PetscRandomCreate(PETSC_COMM_SELF,&r);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(r);CHKERRQ(ierr);
for (i=0; i<nd; i++) {
ierr = PetscRandomGetValue(r,&rand);CHKERRQ(ierr);
start = (PetscInt)(rand*m);
ierr = PetscRandomGetValue(r,&rand);CHKERRQ(ierr);
end = (PetscInt)(rand*m);
lsize = end - start;
if (start > end) { start = end; lsize = -lsize;}
ierr = ISCreateStride(PETSC_COMM_SELF,lsize,start,1,is1+i);CHKERRQ(ierr);
ierr = ISCreateStride(PETSC_COMM_SELF,lsize,start,1,is2+i);CHKERRQ(ierr);
}
ierr = MatIncreaseOverlap(A,nd,is1,ov);CHKERRQ(ierr);
ierr = MatIncreaseOverlap(B,nd,is2,ov);CHKERRQ(ierr);
/* Now see if the serial and parallel case have the same answers */
for (i=0; i<nd; ++i) {
ierr = ISEqual(is1[i],is2[i],&flg);CHKERRQ(ierr);
ierr = PetscPrintf(PETSC_COMM_SELF,"proc:[%d], i=%D, flg =%d\n",rank,i,(int)flg);CHKERRQ(ierr);
}
/* Free allocated memory */
for (i=0; i<nd; ++i) {
ierr = ISDestroy(&is1[i]);CHKERRQ(ierr);
ierr = ISDestroy(&is2[i]);CHKERRQ(ierr);
}
ierr = PetscFree(is1);CHKERRQ(ierr);
ierr = PetscFree(is2);CHKERRQ(ierr);
ierr = PetscRandomDestroy(&r);CHKERRQ(ierr);
ierr = MatDestroy(&A);CHKERRQ(ierr);
ierr = MatDestroy(&B);CHKERRQ(ierr);
#endif
ierr = PetscFinalize();
return ierr;
}
PetscErrorCode MatColoringGetDegrees(Mat G,PetscInt distance,PetscInt *degrees)
{
PetscInt j,i,s,e,n,ln,lm,degree,bidx,idx,dist;
Mat lG,*lGs;
IS ris;
PetscErrorCode ierr;
PetscInt *seen;
const PetscInt *gidx;
PetscInt *idxbuf;
PetscInt *distbuf;
PetscInt ncols;
const PetscInt *cols;
PetscBool isSEQAIJ;
Mat_SeqAIJ *aij;
PetscInt *Gi,*Gj;
PetscFunctionBegin;
ierr = MatGetOwnershipRange(G,&s,&e);CHKERRQ(ierr);
n=e-s;
ierr = ISCreateStride(PetscObjectComm((PetscObject)G),n,s,1,&ris);CHKERRQ(ierr);
ierr = MatIncreaseOverlap(G,1,&ris,distance);CHKERRQ(ierr);
ierr = ISSort(ris);CHKERRQ(ierr);
ierr = MatGetSubMatrices(G,1,&ris,&ris,MAT_INITIAL_MATRIX,&lGs);CHKERRQ(ierr);
lG = lGs[0];
ierr = PetscObjectTypeCompare((PetscObject)lG,MATSEQAIJ,&isSEQAIJ);CHKERRQ(ierr);
if (!isSEQAIJ) SETERRQ(PetscObjectComm((PetscObject)G),PETSC_ERR_SUP,"Requires an MPI/SEQAIJ Matrix");
ierr = MatGetSize(lG,&ln,&lm);CHKERRQ(ierr);
aij = (Mat_SeqAIJ*)lG->data;
Gi = aij->i;
Gj = aij->j;
ierr = PetscMalloc3(lm,&seen,lm,&idxbuf,lm,&distbuf);CHKERRQ(ierr);
for (i=0;i<ln;i++) {
seen[i]=-1;
}
ierr = ISGetIndices(ris,&gidx);CHKERRQ(ierr);
for (i=0;i<ln;i++) {
if (gidx[i] >= e || gidx[i] < s) continue;
bidx=-1;
ncols = Gi[i+1]-Gi[i];
cols = &(Gj[Gi[i]]);
degree = 0;
/* place the distance-one neighbors on the queue */
for (j=0;j<ncols;j++) {
bidx++;
seen[cols[j]] = i;
distbuf[bidx] = 1;
idxbuf[bidx] = cols[j];
}
while (bidx >= 0) {
/* pop */
idx = idxbuf[bidx];
dist = distbuf[bidx];
bidx--;
degree++;
if (dist < distance) {
ncols = Gi[idx+1]-Gi[idx];
cols = &(Gj[Gi[idx]]);
for (j=0;j<ncols;j++) {
if (seen[cols[j]] != i) {
bidx++;
seen[cols[j]] = i;
idxbuf[bidx] = cols[j];
distbuf[bidx] = dist+1;
}
}
}
}
degrees[gidx[i]-s] = degree;
}
ierr = ISRestoreIndices(ris,&gidx);CHKERRQ(ierr);
ierr = ISDestroy(&ris);CHKERRQ(ierr);
ierr = PetscFree3(seen,idxbuf,distbuf);CHKERRQ(ierr);
ierr = MatDestroyMatrices(1,&lGs);CHKERRQ(ierr);
PetscFunctionReturn(0);
}
PetscErrorCode MatColoringCreateSmallestLastWeights(MatColoring mc,PetscReal *weights)
{
PetscInt *degrees,*degb,*llprev,*llnext;
PetscInt j,i,s,e,n,nin,ln,lm,degree,maxdegree=0,bidx,idx,dist,distance=mc->dist;
Mat lG,*lGs;
IS ris;
PetscErrorCode ierr;
PetscInt *seen;
const PetscInt *gidx;
PetscInt *idxbuf;
PetscInt *distbuf;
PetscInt ncols,nxt,prv,cur;
const PetscInt *cols;
PetscBool isSEQAIJ;
Mat_SeqAIJ *aij;
PetscInt *Gi,*Gj,*rperm;
Mat G = mc->mat;
PetscReal *lweights,r;
PetscRandom rand;
PetscFunctionBegin;
ierr = MatGetOwnershipRange(G,&s,&e);CHKERRQ(ierr);
n=e-s;
ierr = ISCreateStride(PetscObjectComm((PetscObject)G),n,s,1,&ris);CHKERRQ(ierr);
ierr = MatIncreaseOverlap(G,1,&ris,distance+1);CHKERRQ(ierr);
ierr = ISSort(ris);CHKERRQ(ierr);
ierr = MatGetSubMatrices(G,1,&ris,&ris,MAT_INITIAL_MATRIX,&lGs);CHKERRQ(ierr);
lG = lGs[0];
ierr = PetscObjectTypeCompare((PetscObject)lG,MATSEQAIJ,&isSEQAIJ);CHKERRQ(ierr);
if (!isSEQAIJ) SETERRQ(PetscObjectComm((PetscObject)G),PETSC_ERR_ARG_WRONGSTATE,"Requires an MPI/SEQAIJ Matrix");
ierr = MatGetSize(lG,&ln,&lm);CHKERRQ(ierr);
aij = (Mat_SeqAIJ*)lG->data;
Gi = aij->i;
Gj = aij->j;
ierr = PetscMalloc3(lm,&seen,lm,&idxbuf,lm,&distbuf);CHKERRQ(ierr);
ierr = PetscMalloc1(lm,°rees);CHKERRQ(ierr);
ierr = PetscMalloc1(lm,&lweights);CHKERRQ(ierr);
for (i=0;i<ln;i++) {
seen[i]=-1;
lweights[i] = 1.;
}
ierr = ISGetIndices(ris,&gidx);CHKERRQ(ierr);
for (i=0;i<ln;i++) {
bidx=-1;
ncols = Gi[i+1]-Gi[i];
cols = &(Gj[Gi[i]]);
degree = 0;
/* place the distance-one neighbors on the queue */
for (j=0;j<ncols;j++) {
bidx++;
seen[cols[j]] = i;
distbuf[bidx] = 1;
idxbuf[bidx] = cols[j];
}
while (bidx >= 0) {
/* pop */
idx = idxbuf[bidx];
dist = distbuf[bidx];
bidx--;
degree++;
if (dist < distance) {
ncols = Gi[idx+1]-Gi[idx];
cols = &(Gj[Gi[idx]]);
for (j=0;j<ncols;j++) {
if (seen[cols[j]] != i) {
bidx++;
seen[cols[j]] = i;
idxbuf[bidx] = cols[j];
distbuf[bidx] = dist+1;
}
}
}
}
degrees[i] = degree;
if (degree > maxdegree) maxdegree = degree;
}
/* bucket by degree by some random permutation */
ierr = PetscRandomCreate(PetscObjectComm((PetscObject)mc),&rand);CHKERRQ(ierr);
ierr = PetscRandomSetFromOptions(rand);CHKERRQ(ierr);
ierr = PetscMalloc1(ln,&rperm);CHKERRQ(ierr);
for (i=0;i<ln;i++) {
ierr = PetscRandomGetValueReal(rand,&r);CHKERRQ(ierr);
lweights[i] = r;
rperm[i]=i;
}
ierr = PetscSortRealWithPermutation(lm,lweights,rperm);CHKERRQ(ierr);
ierr = PetscMalloc1(maxdegree+1,°b);CHKERRQ(ierr);
ierr = PetscMalloc2(ln,&llnext,ln,&llprev);CHKERRQ(ierr);
for (i=0;i<maxdegree+1;i++) {
degb[i] = -1;
}
for (i=0;i<ln;i++) {
llnext[i] = -1;
llprev[i] = -1;
seen[i] = -1;
}
for (i=0;i<ln;i++) {
idx = rperm[i];
llnext[idx] = degb[degrees[idx]];
if (degb[degrees[idx]] > 0) llprev[degb[degrees[idx]]] = idx;
degb[degrees[idx]] = idx;
}
ierr = PetscFree(rperm);CHKERRQ(ierr);
/* remove the lowest degree one */
i=0;
nin=0;
while (i != maxdegree+1) {
for (i=1;i<maxdegree+1; i++) {
if (degb[i] > 0) {
cur = degb[i];
nin++;
degrees[cur] = 0;
degb[i] = llnext[cur];
bidx=-1;
ncols = Gi[cur+1]-Gi[cur];
cols = &(Gj[Gi[cur]]);
/* place the distance-one neighbors on the queue */
for (j=0;j<ncols;j++) {
if (cols[j] != cur) {
bidx++;
seen[cols[j]] = i;
distbuf[bidx] = 1;
idxbuf[bidx] = cols[j];
}
}
while (bidx >= 0) {
/* pop */
idx = idxbuf[bidx];
dist = distbuf[bidx];
bidx--;
nxt=llnext[idx];
prv=llprev[idx];
if (degrees[idx] > 0) {
/* change up the degree of the neighbors still in the graph */
if (lweights[idx] <= lweights[cur]) lweights[idx] = lweights[cur]+1;
if (nxt > 0) {
llprev[nxt] = prv;
}
if (prv > 0) {
llnext[prv] = nxt;
} else {
degb[degrees[idx]] = nxt;
}
degrees[idx]--;
llnext[idx] = degb[degrees[idx]];
llprev[idx] = -1;
if (degb[degrees[idx]] >= 0) {
llprev[degb[degrees[idx]]] = idx;
}
degb[degrees[idx]] = idx;
if (dist < distance) {
ncols = Gi[idx+1]-Gi[idx];
cols = &(Gj[Gi[idx]]);
for (j=0;j<ncols;j++) {
if (seen[cols[j]] != i) {
bidx++;
seen[cols[j]] = i;
idxbuf[bidx] = cols[j];
distbuf[bidx] = dist+1;
}
}
}
}
}
break;
}
}
}
for (i=0;i<lm;i++) {
if (gidx[i] >= s && gidx[i] < e) {
weights[gidx[i]-s] = lweights[i];
}
}
ierr = PetscRandomDestroy(&rand);CHKERRQ(ierr);
ierr = PetscFree(degb);CHKERRQ(ierr);
ierr = PetscFree2(llnext,llprev);CHKERRQ(ierr);
ierr = PetscFree(degrees);CHKERRQ(ierr);
ierr = PetscFree(lweights);CHKERRQ(ierr);
ierr = ISRestoreIndices(ris,&gidx);CHKERRQ(ierr);
ierr = ISDestroy(&ris);CHKERRQ(ierr);
ierr = PetscFree3(seen,idxbuf,distbuf);CHKERRQ(ierr);
ierr = MatDestroyMatrices(1,&lGs);CHKERRQ(ierr);
PetscFunctionReturn(0);
}
