hypre_ParMultiVector *
hypre_ParMultiVectorCreate(MPI_Comm comm, HYPRE_Int global_size, HYPRE_Int *partitioning,
HYPRE_Int num_vectors)
{
hypre_ParMultiVector *vector;
HYPRE_Int num_procs, my_id;
vector = hypre_CTAlloc(hypre_ParMultiVector, 1);
hypre_MPI_Comm_rank(comm, &my_id);
if (! partitioning)
{
hypre_MPI_Comm_size(comm, &num_procs);
hypre_GeneratePartitioning(global_size, num_procs, &partitioning);
}
hypre_ParMultiVectorComm(vector) = comm;
hypre_ParMultiVectorGlobalSize(vector) = global_size;
hypre_ParMultiVectorPartitioning(vector) = partitioning;
hypre_ParMultiVectorNumVectors(vector) = num_vectors;
hypre_ParMultiVectorLocalVector(vector) =
hypre_SeqMultivectorCreate((partitioning[my_id+1]-partitioning[my_id]), num_vectors);
hypre_ParMultiVectorFirstIndex(vector) = partitioning[my_id];
/* we set these 2 defaults exactly as in par_vector.c, although it's questionable */
hypre_ParMultiVectorOwnsData(vector) = 1;
hypre_ParMultiVectorOwnsPartitioning(vector) = 1;
return vector;
}
hypre_ParVector *
hypre_ParVectorCreate( MPI_Comm comm,
HYPRE_Int global_size,
HYPRE_Int *partitioning)
{
hypre_ParVector *vector;
HYPRE_Int num_procs, my_id;
if (global_size < 0)
{
hypre_error_in_arg(2);
return NULL;
}
vector = hypre_CTAlloc(hypre_ParVector, 1);
hypre_MPI_Comm_rank(comm,&my_id);
if (!partitioning)
{
hypre_MPI_Comm_size(comm,&num_procs);
#ifdef HYPRE_NO_GLOBAL_PARTITION
hypre_GenerateLocalPartitioning(global_size, num_procs, my_id, &partitioning);
#else
hypre_GeneratePartitioning(global_size, num_procs, &partitioning);
#endif
}
hypre_ParVectorAssumedPartition(vector) = NULL;
hypre_ParVectorComm(vector) = comm;
hypre_ParVectorGlobalSize(vector) = global_size;
#ifdef HYPRE_NO_GLOBAL_PARTITION
hypre_ParVectorFirstIndex(vector) = partitioning[0];
hypre_ParVectorLastIndex(vector) = partitioning[1]-1;
hypre_ParVectorPartitioning(vector) = partitioning;
hypre_ParVectorLocalVector(vector) =
hypre_SeqVectorCreate(partitioning[1]-partitioning[0]);
#else
hypre_ParVectorFirstIndex(vector) = partitioning[my_id];
hypre_ParVectorLastIndex(vector) = partitioning[my_id+1] -1;
hypre_ParVectorPartitioning(vector) = partitioning;
hypre_ParVectorLocalVector(vector) =
hypre_SeqVectorCreate(partitioning[my_id+1]-partitioning[my_id]);
#endif
/* set defaults */
hypre_ParVectorOwnsData(vector) = 1;
hypre_ParVectorOwnsPartitioning(vector) = 1;
return vector;
}
HYPRE_Int
main( HYPRE_Int argc,
char *argv[] )
{
hypre_ParVector *vector1;
hypre_ParVector *vector2;
hypre_ParVector *tmp_vector;
HYPRE_Int num_procs, my_id;
HYPRE_Int global_size = 20;
HYPRE_Int local_size;
HYPRE_Int first_index;
HYPRE_Int num_vectors, vecstride, idxstride;
HYPRE_Int i, j;
HYPRE_Int *partitioning;
double prod;
double *data, *data2;
hypre_Vector *vector;
hypre_Vector *local_vector;
hypre_Vector *local_vector2;
/* Initialize MPI */
hypre_MPI_Init(&argc, &argv);
hypre_MPI_Comm_size(hypre_MPI_COMM_WORLD, &num_procs );
hypre_MPI_Comm_rank(hypre_MPI_COMM_WORLD, &my_id );
hypre_printf(" my_id: %d num_procs: %d\n", my_id, num_procs);
partitioning = NULL;
num_vectors = 3;
vector1 = hypre_ParMultiVectorCreate
( hypre_MPI_COMM_WORLD, global_size, partitioning, num_vectors );
partitioning = hypre_ParVectorPartitioning(vector1);
hypre_ParVectorInitialize(vector1);
local_vector = hypre_ParVectorLocalVector(vector1);
data = hypre_VectorData(local_vector);
local_size = hypre_VectorSize(local_vector);
vecstride = hypre_VectorVectorStride(local_vector);
idxstride = hypre_VectorIndexStride(local_vector);
first_index = partitioning[my_id];
hypre_printf("vecstride=%i idxstride=%i local_size=%i num_vectors=%i",
vecstride, idxstride, local_size, num_vectors );
for (j=0; j<num_vectors; ++j )
for (i=0; i < local_size; i++)
data[ j*vecstride + i*idxstride ] = first_index+i + 100*j;
hypre_ParVectorPrint(vector1, "Vector");
local_vector2 = hypre_SeqMultiVectorCreate( global_size, num_vectors );
hypre_SeqVectorInitialize(local_vector2);
data2 = hypre_VectorData(local_vector2);
vecstride = hypre_VectorVectorStride(local_vector2);
idxstride = hypre_VectorIndexStride(local_vector2);
for (j=0; j<num_vectors; ++j )
for (i=0; i < global_size; i++)
data2[ j*vecstride + i*idxstride ] = i + 100*j;
/* partitioning = hypre_CTAlloc(HYPRE_Int,4);
partitioning[0] = 0;
partitioning[1] = 10;
partitioning[2] = 10;
partitioning[3] = 20;
*/
partitioning = hypre_CTAlloc(HYPRE_Int,1+num_procs);
hypre_GeneratePartitioning( global_size, num_procs, &partitioning );
vector2 = hypre_VectorToParVector(hypre_MPI_COMM_WORLD,local_vector2,partitioning);
hypre_ParVectorSetPartitioningOwner(vector2,0);
hypre_ParVectorPrint(vector2, "Convert");
vector = hypre_ParVectorToVectorAll(vector2);
/*-----------------------------------------------------------
* Copy the vector into tmp_vector
*-----------------------------------------------------------*/
/* Read doesn't work for multivectors yet...
tmp_vector = hypre_ParVectorRead(hypre_MPI_COMM_WORLD, "Convert");*/
tmp_vector = hypre_ParMultiVectorCreate
( hypre_MPI_COMM_WORLD, global_size, partitioning, num_vectors );
hypre_ParVectorInitialize( tmp_vector );
hypre_ParVectorCopy( vector2, tmp_vector );
/*
tmp_vector = hypre_ParVectorCreate(hypre_MPI_COMM_WORLD,global_size,partitioning);
hypre_ParVectorSetPartitioningOwner(tmp_vector,0);
hypre_ParVectorInitialize(tmp_vector);
hypre_ParVectorCopy(vector1, tmp_vector);
hypre_ParVectorPrint(tmp_vector,"Copy");
*/
/*-----------------------------------------------------------
* Scale tmp_vector
*-----------------------------------------------------------*/
hypre_ParVectorScale(2.0, tmp_vector);
hypre_ParVectorPrint(tmp_vector,"Scale");
/*-----------------------------------------------------------
* Do an Axpy (2*vector - vector) = vector
*-----------------------------------------------------------*/
hypre_ParVectorAxpy(-1.0, vector1, tmp_vector);
hypre_ParVectorPrint(tmp_vector,"Axpy");
/*-----------------------------------------------------------
* Do an inner product vector* tmp_vector
*-----------------------------------------------------------*/
prod = hypre_ParVectorInnerProd(vector1, tmp_vector);
hypre_printf (" prod: %8.2f \n", prod);
/*-----------------------------------------------------------
* Finalize things
*-----------------------------------------------------------*/
hypre_ParVectorDestroy(vector1);
hypre_ParVectorDestroy(vector2);
hypre_ParVectorDestroy(tmp_vector);
hypre_SeqVectorDestroy(local_vector2);
if (vector) hypre_SeqVectorDestroy(vector);
/* Finalize MPI */
hypre_MPI_Finalize();
return 0;
}
HYPRE_ParCSRMatrix
GenerateRotate7pt( MPI_Comm comm,
HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int p,
HYPRE_Int q,
HYPRE_Real alpha,
HYPRE_Real eps )
{
hypre_ParCSRMatrix *A;
hypre_CSRMatrix *diag;
hypre_CSRMatrix *offd;
HYPRE_Int *diag_i;
HYPRE_Int *diag_j;
HYPRE_Real *diag_data;
HYPRE_Int *offd_i;
HYPRE_Int *offd_j;
HYPRE_Real *offd_data;
HYPRE_Real *value;
HYPRE_Real ac, bc, cc, s, c, pi, x;
HYPRE_Int *global_part;
HYPRE_Int ix, iy;
HYPRE_Int cnt, o_cnt;
HYPRE_Int local_num_rows;
HYPRE_Int *col_map_offd;
HYPRE_Int *work;
HYPRE_Int row_index;
HYPRE_Int i,j;
HYPRE_Int nx_local, ny_local;
HYPRE_Int nx_size, ny_size;
HYPRE_Int num_cols_offd;
HYPRE_Int grid_size;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int num_procs, my_id;
HYPRE_Int P_busy, Q_busy;
hypre_MPI_Comm_size(comm,&num_procs);
hypre_MPI_Comm_rank(comm,&my_id);
grid_size = nx*ny;
value = hypre_CTAlloc(HYPRE_Real,4);
pi = 4.0*atan(1.0);
x = pi*alpha/180.0;
s = sin(x);
c = cos(x);
ac = -(c*c + eps*s*s);
bc = 2.0*(1.0 - eps)*s*c;
cc = -(s*s + eps*c*c);
value[0] = -2*(2*ac+bc+2*cc);
value[1] = 2*ac+bc;
value[2] = bc+2*cc;
value[3] = -bc;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q+1);
global_part[0] = 0;
cnt = 1;
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size;
}
}
nx_local = nx_part[p+1] - nx_part[p];
ny_local = ny_part[q+1] - ny_part[q];
my_id = q*P + p;
num_procs = P*Q;
local_num_rows = nx_local*ny_local;
diag_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
offd_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
P_busy = hypre_min(nx,P);
Q_busy = hypre_min(ny,Q);
num_cols_offd = 0;
if (p) num_cols_offd += ny_local;
if (p < P_busy-1) num_cols_offd += ny_local;
if (q) num_cols_offd += nx_local;
if (q < Q_busy-1) num_cols_offd += nx_local;
if (p && q) num_cols_offd++;
if (p && q < Q_busy-1 ) num_cols_offd++;
if (p < P_busy-1 && q ) num_cols_offd++;
if (p < P_busy-1 && q < Q_busy-1 ) num_cols_offd++;
if (!local_num_rows) num_cols_offd = 0;
col_map_offd = hypre_CTAlloc(HYPRE_Int, num_cols_offd);
cnt = 0;
o_cnt = 0;
diag_i[0] = 0;
offd_i[0] = 0;
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
cnt++;
o_cnt++;
diag_i[cnt] = diag_i[cnt-1];
offd_i[o_cnt] = offd_i[o_cnt-1];
diag_i[cnt]++;
if (iy > ny_part[q])
{
diag_i[cnt]++;
if (ix > nx_part[p])
{
diag_i[cnt]++;
}
else
{
if (ix)
offd_i[o_cnt]++;
}
}
else
{
if (iy)
{
offd_i[o_cnt]++;
if (ix > nx_part[p])
{
offd_i[o_cnt]++;
}
else if (ix)
{
offd_i[o_cnt]++;
}
}
}
if (ix > nx_part[p])
diag_i[cnt]++;
else
{
if (ix)
{
offd_i[o_cnt]++;
}
}
if (ix+1 < nx_part[p+1])
diag_i[cnt]++;
else
{
if (ix+1 < nx)
{
offd_i[o_cnt]++;
}
}
if (iy+1 < ny_part[q+1])
{
diag_i[cnt]++;
if (ix < nx_part[p+1]-1)
{
diag_i[cnt]++;
}
else
{
if (ix+1 < nx)
offd_i[o_cnt]++;
}
}
else
{
if (iy+1 < ny)
{
offd_i[o_cnt]++;
if (ix < nx_part[p+1]-1)
{
offd_i[o_cnt]++;
}
else if (ix < nx-1)
{
offd_i[o_cnt]++;
}
}
}
}
}
diag_j = hypre_CTAlloc(HYPRE_Int, diag_i[local_num_rows]);
diag_data = hypre_CTAlloc(HYPRE_Real, diag_i[local_num_rows]);
if (num_procs > 1)
{
offd_j = hypre_CTAlloc(HYPRE_Int, offd_i[local_num_rows]);
offd_data = hypre_CTAlloc(HYPRE_Real, offd_i[local_num_rows]);
}
row_index = 0;
cnt = 0;
o_cnt = 0;
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
diag_j[cnt] = row_index;
diag_data[cnt++] = value[0];
if (iy > ny_part[q])
{
if (ix > nx_part[p])
{
diag_j[cnt] = row_index-nx_local-1 ;
diag_data[cnt++] = value[3];
}
else
{
if (ix)
{
offd_j[o_cnt] = hypre_map2(ix-1,iy-1,p-1,q,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[3];
}
}
diag_j[cnt] = row_index-nx_local;
diag_data[cnt++] = value[2];
}
else
{
if (iy)
{
if (ix > nx_part[p])
{
offd_j[o_cnt] = hypre_map2(ix-1,iy-1,p,q-1,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[3];
}
else if (ix)
{
offd_j[o_cnt] = hypre_map2(ix-1,iy-1,p-1,q-1,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[3];
}
offd_j[o_cnt] = hypre_map2(ix,iy-1,p,q-1,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[2];
}
}
if (ix > nx_part[p])
{
diag_j[cnt] = row_index-1;
diag_data[cnt++] = value[1];
}
else
{
if (ix)
{
offd_j[o_cnt] = hypre_map2(ix-1,iy,p-1,q,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[1];
}
}
if (ix+1 < nx_part[p+1])
{
diag_j[cnt] = row_index+1;
diag_data[cnt++] = value[1];
}
else
{
if (ix+1 < nx)
{
offd_j[o_cnt] = hypre_map2(ix+1,iy,p+1,q,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[1];
}
}
if (iy+1 < ny_part[q+1])
{
diag_j[cnt] = row_index+nx_local;
diag_data[cnt++] = value[2];
if (ix < nx_part[p+1]-1)
{
diag_j[cnt] = row_index+nx_local+1 ;
diag_data[cnt++] = value[3];
}
else
{
if (ix+1 < nx)
{
offd_j[o_cnt] = hypre_map2(ix+1,iy+1,p+1,q,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[3];
}
}
}
else
{
if (iy+1 < ny)
{
offd_j[o_cnt] = hypre_map2(ix,iy+1,p,q+1,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[2];
if (ix < nx_part[p+1]-1)
{
offd_j[o_cnt] = hypre_map2(ix+1,iy+1,p,q+1,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[3];
}
else if (ix < nx-1)
{
offd_j[o_cnt] = hypre_map2(ix+1,iy+1,p+1,q+1,P,Q,
nx_part,ny_part,global_part);
offd_data[o_cnt++] = value[3];
}
}
}
row_index++;
}
}
if (num_procs > 1)
{
work = hypre_CTAlloc(HYPRE_Int,o_cnt);
for (i=0; i < o_cnt; i++)
work[i] = offd_j[i];
qsort0(work, 0, o_cnt-1);
col_map_offd[0] = work[0];
cnt = 0;
for (i=0; i < o_cnt; i++)
{
if (work[i] > col_map_offd[cnt])
{
cnt++;
col_map_offd[cnt] = work[i];
}
}
for (i=0; i < o_cnt; i++)
{
for (j=0; j < num_cols_offd; j++)
{
if (offd_j[i] == col_map_offd[j])
{
offd_j[i] = j;
break;
}
}
}
hypre_TFree(work);
}
A = hypre_ParCSRMatrixCreate(comm, grid_size, grid_size,
global_part, global_part, num_cols_offd,
diag_i[local_num_rows],
offd_i[local_num_rows]);
hypre_ParCSRMatrixColMapOffd(A) = col_map_offd;
diag = hypre_ParCSRMatrixDiag(A);
hypre_CSRMatrixI(diag) = diag_i;
hypre_CSRMatrixJ(diag) = diag_j;
hypre_CSRMatrixData(diag) = diag_data;
offd = hypre_ParCSRMatrixOffd(A);
hypre_CSRMatrixI(offd) = offd_i;
if (num_cols_offd)
{
hypre_CSRMatrixJ(offd) = offd_j;
hypre_CSRMatrixData(offd) = offd_data;
}
hypre_TFree(nx_part);
hypre_TFree(ny_part);
hypre_TFree(value);
return (HYPRE_ParCSRMatrix) A;
}
hypre_ParCSRBooleanMatrix *hypre_ParCSRBooleanMatrixCreate( MPI_Comm comm,
HYPRE_Int global_num_rows, HYPRE_Int global_num_cols,
HYPRE_Int *row_starts, HYPRE_Int *col_starts,
HYPRE_Int num_cols_offd, HYPRE_Int num_nonzeros_diag,
HYPRE_Int num_nonzeros_offd)
{
hypre_ParCSRBooleanMatrix *matrix;
HYPRE_Int num_procs, my_id;
HYPRE_Int local_num_rows, local_num_cols;
HYPRE_Int first_row_index, first_col_diag;
matrix = hypre_CTAlloc(hypre_ParCSRBooleanMatrix, 1);
hypre_MPI_Comm_rank(comm,&my_id);
hypre_MPI_Comm_size(comm,&num_procs);
if (!row_starts)
{
hypre_GeneratePartitioning(global_num_rows,num_procs,&row_starts);
}
if (!col_starts)
{
if (global_num_rows == global_num_cols)
{
col_starts = row_starts;
}
else
{
hypre_GeneratePartitioning(global_num_cols,num_procs,&col_starts);
}
}
first_row_index = row_starts[my_id];
local_num_rows = row_starts[my_id+1]-first_row_index;
first_col_diag = col_starts[my_id];
local_num_cols = col_starts[my_id+1]-first_col_diag;
hypre_ParCSRBooleanMatrix_Get_Comm(matrix) = comm;
hypre_ParCSRBooleanMatrix_Get_Diag(matrix) =
hypre_CSRBooleanMatrixCreate(local_num_rows, local_num_cols,
num_nonzeros_diag);
hypre_ParCSRBooleanMatrix_Get_Offd(matrix) =
hypre_CSRBooleanMatrixCreate(local_num_rows, num_cols_offd,
num_nonzeros_offd);
hypre_ParCSRBooleanMatrix_Get_GlobalNRows(matrix) = global_num_rows;
hypre_ParCSRBooleanMatrix_Get_GlobalNCols(matrix) = global_num_cols;
hypre_ParCSRBooleanMatrix_Get_StartRow(matrix) = first_row_index;
hypre_ParCSRBooleanMatrix_Get_FirstColDiag(matrix) = first_col_diag;
hypre_ParCSRBooleanMatrix_Get_ColMapOffd(matrix) = NULL;
hypre_ParCSRBooleanMatrix_Get_RowStarts(matrix) = row_starts;
hypre_ParCSRBooleanMatrix_Get_ColStarts(matrix) = col_starts;
hypre_ParCSRBooleanMatrix_Get_CommPkg(matrix) = NULL;
hypre_ParCSRBooleanMatrix_Get_OwnsData(matrix) = 1;
hypre_ParCSRBooleanMatrix_Get_OwnsRowStarts(matrix) = 1;
hypre_ParCSRBooleanMatrix_Get_OwnsColStarts(matrix) = 1;
if (row_starts == col_starts)
hypre_ParCSRBooleanMatrix_Get_OwnsColStarts(matrix) = 0;
hypre_ParCSRBooleanMatrix_Get_Rowindices(matrix) = NULL;
hypre_ParCSRBooleanMatrix_Get_Getrowactive(matrix) = 0;
return matrix;
}
HYPRE_ParCSRMatrix
GenerateLaplacian9pt( MPI_Comm comm,
HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int p,
HYPRE_Int q,
double *value )
{
hypre_ParCSRMatrix *A;
hypre_CSRMatrix *diag;
hypre_CSRMatrix *offd;
HYPRE_Int *diag_i;
HYPRE_Int *diag_j;
double *diag_data;
HYPRE_Int *offd_i;
HYPRE_Int *offd_j;
double *offd_data;
HYPRE_Int *global_part;
HYPRE_Int ix, iy;
HYPRE_Int cnt, o_cnt;
HYPRE_Int local_num_rows;
HYPRE_Int *col_map_offd;
HYPRE_Int *work;
HYPRE_Int row_index;
HYPRE_Int i,j;
HYPRE_Int nx_local, ny_local;
HYPRE_Int nx_size, ny_size;
HYPRE_Int num_cols_offd;
HYPRE_Int grid_size;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int num_procs, my_id;
HYPRE_Int P_busy, Q_busy;
hypre_MPI_Comm_size(comm,&num_procs);
hypre_MPI_Comm_rank(comm,&my_id);
grid_size = nx*ny;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q+1);
global_part[0] = 0;
cnt = 1;
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size;
}
}
nx_local = nx_part[p+1] - nx_part[p];
ny_local = ny_part[q+1] - ny_part[q];
my_id = q*P + p;
num_procs = P*Q;
local_num_rows = nx_local*ny_local;
diag_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
offd_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
P_busy = hypre_min(nx,P);
Q_busy = hypre_min(ny,Q);
num_cols_offd = 0;
if (p) num_cols_offd += ny_local;
if (p < P_busy-1) num_cols_offd += ny_local;
if (q) num_cols_offd += nx_local;
if (q < Q_busy-1) num_cols_offd += nx_local;
if (p && q) num_cols_offd++;
if (p && q < Q_busy-1 ) num_cols_offd++;
if (p < P_busy-1 && q ) num_cols_offd++;
if (p < P_busy-1 && q < Q_busy-1 ) num_cols_offd++;
if (!local_num_rows) num_cols_offd = 0;
col_map_offd = hypre_CTAlloc(HYPRE_Int, num_cols_offd);
cnt = 0;
o_cnt = 0;
diag_i[0] = 0;
offd_i[0] = 0;
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
cnt++;
o_cnt++;
diag_i[cnt] = diag_i[cnt-1];
offd_i[o_cnt] = offd_i[o_cnt-1];
diag_i[cnt]++;
if (iy > ny_part[q])
{
diag_i[cnt]++;
if (ix > nx_part[p])
{
diag_i[cnt]++;
}
else
{
if (ix)
offd_i[o_cnt]++;
}
if (ix < nx_part[p+1]-1)
{
diag_i[cnt]++;
}
else
{
if (ix+1 < nx)
offd_i[o_cnt]++;
}
}
else
{
if (iy)
{
offd_i[o_cnt]++;
if (ix > nx_part[p])
{
offd_i[o_cnt]++;
}
else if (ix)
{
offd_i[o_cnt]++;
}
if (ix < nx_part[p+1]-1)
{
offd_i[o_cnt]++;
}
else if (ix < nx-1)
{
offd_i[o_cnt]++;
}
}
}
if (ix > nx_part[p])
diag_i[cnt]++;
else
{
if (ix)
{
offd_i[o_cnt]++;
}
}
if (ix+1 < nx_part[p+1])
diag_i[cnt]++;
else
{
if (ix+1 < nx)
{
offd_i[o_cnt]++;
}
}
if (iy+1 < ny_part[q+1])
{
diag_i[cnt]++;
if (ix > nx_part[p])
{
diag_i[cnt]++;
}
else
{
if (ix)
offd_i[o_cnt]++;
}
if (ix < nx_part[p+1]-1)
{
diag_i[cnt]++;
}
else
{
if (ix+1 < nx)
offd_i[o_cnt]++;
}
}
else
{
if (iy+1 < ny)
{
offd_i[o_cnt]++;
if (ix > nx_part[p])
{
offd_i[o_cnt]++;
}
else if (ix)
{
offd_i[o_cnt]++;
}
if (ix < nx_part[p+1]-1)
{
offd_i[o_cnt]++;
}
else if (ix < nx-1)
{
offd_i[o_cnt]++;
}
}
}
}
}
diag_j = hypre_CTAlloc(HYPRE_Int, diag_i[local_num_rows]);
diag_data = hypre_CTAlloc(double, diag_i[local_num_rows]);
if (num_procs > 1)
{
offd_j = hypre_CTAlloc(HYPRE_Int, offd_i[local_num_rows]);
offd_data = hypre_CTAlloc(double, offd_i[local_num_rows]);
}
HYPRE_ParCSRMatrix
GenerateRotate7pt( MPI_Comm comm,
HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int p,
HYPRE_Int q,
double alpha,
double eps )
{
hypre_ParCSRMatrix *A;
hypre_CSRMatrix *diag;
hypre_CSRMatrix *offd;
HYPRE_Int *diag_i;
HYPRE_Int *diag_j;
double *diag_data;
HYPRE_Int *offd_i;
HYPRE_Int *offd_j;
double *offd_data;
double *value;
double ac, bc, cc, s, c, pi, x;
HYPRE_Int *global_part;
HYPRE_Int ix, iy;
HYPRE_Int cnt, o_cnt;
HYPRE_Int local_num_rows;
HYPRE_Int *col_map_offd;
HYPRE_Int *work;
HYPRE_Int row_index;
HYPRE_Int i,j;
HYPRE_Int nx_local, ny_local;
HYPRE_Int nx_size, ny_size;
HYPRE_Int num_cols_offd;
HYPRE_Int grid_size;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int num_procs, my_id;
HYPRE_Int P_busy, Q_busy;
hypre_MPI_Comm_size(comm,&num_procs);
hypre_MPI_Comm_rank(comm,&my_id);
grid_size = nx*ny;
value = hypre_CTAlloc(double,4);
pi = 4.0*atan(1.0);
x = pi*alpha/180.0;
s = sin(x);
c = cos(x);
ac = -(c*c + eps*s*s);
bc = 2.0*(1.0 - eps)*s*c;
cc = -(s*s + eps*c*c);
value[0] = -2*(2*ac+bc+2*cc);
value[1] = 2*ac+bc;
value[2] = bc+2*cc;
value[3] = -bc;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q+1);
global_part[0] = 0;
cnt = 1;
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size;
}
}
nx_local = nx_part[p+1] - nx_part[p];
ny_local = ny_part[q+1] - ny_part[q];
my_id = q*P + p;
num_procs = P*Q;
local_num_rows = nx_local*ny_local;
diag_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
offd_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
P_busy = hypre_min(nx,P);
Q_busy = hypre_min(ny,Q);
num_cols_offd = 0;
if (p) num_cols_offd += ny_local;
if (p < P_busy-1) num_cols_offd += ny_local;
if (q) num_cols_offd += nx_local;
if (q < Q_busy-1) num_cols_offd += nx_local;
if (p && q) num_cols_offd++;
if (p && q < Q_busy-1 ) num_cols_offd++;
if (p < P_busy-1 && q ) num_cols_offd++;
if (p < P_busy-1 && q < Q_busy-1 ) num_cols_offd++;
if (!local_num_rows) num_cols_offd = 0;
col_map_offd = hypre_CTAlloc(HYPRE_Int, num_cols_offd);
cnt = 0;
o_cnt = 0;
diag_i[0] = 0;
offd_i[0] = 0;
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
cnt++;
o_cnt++;
diag_i[cnt] = diag_i[cnt-1];
offd_i[o_cnt] = offd_i[o_cnt-1];
diag_i[cnt]++;
if (iy > ny_part[q])
{
diag_i[cnt]++;
if (ix > nx_part[p])
{
diag_i[cnt]++;
}
else
{
if (ix)
offd_i[o_cnt]++;
}
}
else
{
if (iy)
{
offd_i[o_cnt]++;
if (ix > nx_part[p])
{
offd_i[o_cnt]++;
}
else if (ix)
{
offd_i[o_cnt]++;
}
}
}
if (ix > nx_part[p])
diag_i[cnt]++;
else
{
if (ix)
{
offd_i[o_cnt]++;
}
}
if (ix+1 < nx_part[p+1])
diag_i[cnt]++;
else
{
if (ix+1 < nx)
{
offd_i[o_cnt]++;
}
}
if (iy+1 < ny_part[q+1])
{
diag_i[cnt]++;
if (ix < nx_part[p+1]-1)
{
diag_i[cnt]++;
}
else
{
if (ix+1 < nx)
offd_i[o_cnt]++;
}
}
else
{
if (iy+1 < ny)
{
offd_i[o_cnt]++;
if (ix < nx_part[p+1]-1)
{
offd_i[o_cnt]++;
}
else if (ix < nx-1)
{
offd_i[o_cnt]++;
}
}
}
}
}
diag_j = hypre_CTAlloc(HYPRE_Int, diag_i[local_num_rows]);
diag_data = hypre_CTAlloc(double, diag_i[local_num_rows]);
if (num_procs > 1)
{
offd_j = hypre_CTAlloc(HYPRE_Int, offd_i[local_num_rows]);
offd_data = hypre_CTAlloc(double, offd_i[local_num_rows]);
}
hypre_ParCSRBlockMatrix *
hypre_ParCSRBlockMatrixCreate( MPI_Comm comm,
HYPRE_Int block_size,
HYPRE_Int global_num_rows,
HYPRE_Int global_num_cols,
HYPRE_Int *row_starts,
HYPRE_Int *col_starts,
HYPRE_Int num_cols_offd,
HYPRE_Int num_nonzeros_diag,
HYPRE_Int num_nonzeros_offd )
{
hypre_ParCSRBlockMatrix *matrix;
HYPRE_Int num_procs, my_id;
HYPRE_Int local_num_rows, local_num_cols;
HYPRE_Int first_row_index, first_col_diag;
matrix = hypre_CTAlloc(hypre_ParCSRBlockMatrix, 1);
hypre_MPI_Comm_rank(comm,&my_id);
hypre_MPI_Comm_size(comm,&num_procs);
if (!row_starts)
{
#ifdef HYPRE_NO_GLOBAL_PARTITION
hypre_GenerateLocalPartitioning(global_num_rows, num_procs, my_id, &row_starts);
#else
hypre_GeneratePartitioning(global_num_rows,num_procs,&row_starts);
#endif
}
if (!col_starts)
{
if (global_num_rows == global_num_cols)
{
col_starts = row_starts;
}
else
{
#ifdef HYPRE_NO_GLOBAL_PARTITION
hypre_GenerateLocalPartitioning(global_num_cols, num_procs, my_id, &col_starts);
#else
hypre_GeneratePartitioning(global_num_cols,num_procs,&col_starts);
#endif
}
}
#ifdef HYPRE_NO_GLOBAL_PARTITION
/* row_starts[0] is start of local rows. row_starts[1] is start of next
processor's rows */
first_row_index = row_starts[0];
local_num_rows = row_starts[1]-first_row_index ;
first_col_diag = col_starts[0];
local_num_cols = col_starts[1]-first_col_diag;
#else
first_row_index = row_starts[my_id];
local_num_rows = row_starts[my_id+1]-first_row_index;
first_col_diag = col_starts[my_id];
local_num_cols = col_starts[my_id+1]-first_col_diag;
#endif
hypre_ParCSRBlockMatrixComm(matrix) = comm;
hypre_ParCSRBlockMatrixDiag(matrix) = hypre_CSRBlockMatrixCreate(block_size,
local_num_rows,local_num_cols,num_nonzeros_diag);
hypre_ParCSRBlockMatrixOffd(matrix) = hypre_CSRBlockMatrixCreate(block_size,
local_num_rows, num_cols_offd, num_nonzeros_offd);
hypre_ParCSRBlockMatrixBlockSize(matrix) = block_size;
hypre_ParCSRBlockMatrixGlobalNumRows(matrix) = global_num_rows;
hypre_ParCSRBlockMatrixGlobalNumCols(matrix) = global_num_cols;
hypre_ParCSRBlockMatrixFirstRowIndex(matrix) = first_row_index;
hypre_ParCSRBlockMatrixFirstColDiag(matrix) = first_col_diag;
hypre_ParCSRBlockMatrixLastRowIndex(matrix) = first_row_index + local_num_rows - 1;
hypre_ParCSRBlockMatrixLastColDiag(matrix) = first_col_diag + local_num_cols - 1;
hypre_ParCSRBlockMatrixColMapOffd(matrix) = NULL;
hypre_ParCSRBlockMatrixAssumedPartition(matrix) = NULL;
/* When NO_GLOBAL_PARTITION is set we could make these null, instead
of leaving the range. If that change is made, then when this create
is called from functions like the matrix-matrix multiply, be careful
not to generate a new partition */
hypre_ParCSRBlockMatrixRowStarts(matrix) = row_starts;
hypre_ParCSRBlockMatrixColStarts(matrix) = col_starts;
hypre_ParCSRBlockMatrixCommPkg(matrix) = NULL;
hypre_ParCSRBlockMatrixCommPkgT(matrix) = NULL;
/* set defaults */
hypre_ParCSRBlockMatrixOwnsData(matrix) = 1;
hypre_ParCSRBlockMatrixOwnsRowStarts(matrix) = 1;
hypre_ParCSRBlockMatrixOwnsColStarts(matrix) = 1;
if (row_starts == col_starts)
hypre_ParCSRBlockMatrixOwnsColStarts(matrix) = 0;
return matrix;
}
hypre_ParVector *
hypre_ParVectorCreateFromBlock( MPI_Comm comm,
HYPRE_Int p_global_size,
HYPRE_Int *p_partitioning, HYPRE_Int block_size)
{
hypre_ParVector *vector;
HYPRE_Int num_procs, my_id, i;
HYPRE_Int global_size;
HYPRE_Int *new_partitioning; /* need to create a new partitioning - son't want to write over
what is passed in */
global_size = p_global_size*block_size;
vector = hypre_CTAlloc(hypre_ParVector, 1);
hypre_MPI_Comm_rank(comm,&my_id);
hypre_MPI_Comm_size(comm,&num_procs);
if (!p_partitioning)
{
#ifdef HYPRE_NO_GLOBAL_PARTITION
hypre_GenerateLocalPartitioning(global_size, num_procs, my_id, &new_partitioning);
#else
hypre_GeneratePartitioning(global_size, num_procs, &new_partitioning);
#endif
}
else /* adjust for block_size */
{
#ifdef HYPRE_NO_GLOBAL_PARTITION
new_partitioning = hypre_CTAlloc(HYPRE_Int, 2);
for(i = 0; i < 2; i++)
{
new_partitioning[i] = p_partitioning[i]*block_size;
}
#else
new_partitioning = hypre_CTAlloc(HYPRE_Int, num_procs + 1);
for(i = 0; i < num_procs + 1; i++)
{
new_partitioning[i] = p_partitioning[i]*block_size;
}
#endif
}
hypre_ParVectorComm(vector) = comm;
hypre_ParVectorGlobalSize(vector) = global_size;
#ifdef HYPRE_NO_GLOBAL_PARTITION
hypre_ParVectorFirstIndex(vector) = new_partitioning[0];
hypre_ParVectorLastIndex(vector) = new_partitioning[1]-1;
hypre_ParVectorPartitioning(vector) = new_partitioning;
hypre_ParVectorLocalVector(vector) =
hypre_SeqVectorCreate(new_partitioning[1]-new_partitioning[0]);
#else
hypre_ParVectorFirstIndex(vector) = new_partitioning[my_id];
hypre_ParVectorLastIndex(vector) = new_partitioning[my_id+1] -1;
hypre_ParVectorPartitioning(vector) = new_partitioning;
hypre_ParVectorLocalVector(vector) =
hypre_SeqVectorCreate(new_partitioning[my_id+1]-new_partitioning[my_id]);
#endif
/* set defaults */
hypre_ParVectorOwnsData(vector) = 1;
hypre_ParVectorOwnsPartitioning(vector) = 1;
return vector;
}
HYPRE_ParCSRMatrix
GenerateDifConv( MPI_Comm comm,
HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int nz,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int R,
HYPRE_Int p,
HYPRE_Int q,
HYPRE_Int r,
double *value )
{
hypre_ParCSRMatrix *A;
hypre_CSRMatrix *diag;
hypre_CSRMatrix *offd;
HYPRE_Int *diag_i;
HYPRE_Int *diag_j;
double *diag_data;
HYPRE_Int *offd_i;
HYPRE_Int *offd_j;
double *offd_data;
HYPRE_Int *global_part;
HYPRE_Int ix, iy, iz;
HYPRE_Int cnt, o_cnt;
HYPRE_Int local_num_rows;
HYPRE_Int *col_map_offd;
HYPRE_Int row_index;
HYPRE_Int i,j;
HYPRE_Int nx_local, ny_local, nz_local;
HYPRE_Int nx_size, ny_size, nz_size;
HYPRE_Int num_cols_offd;
HYPRE_Int grid_size;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int *nz_part;
HYPRE_Int num_procs, my_id;
HYPRE_Int P_busy, Q_busy, R_busy;
hypre_MPI_Comm_size(comm,&num_procs);
hypre_MPI_Comm_rank(comm,&my_id);
grid_size = nx*ny*nz;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
hypre_GeneratePartitioning(nz,R,&nz_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q*R+1);
global_part[0] = 0;
cnt = 1;
for (iz = 0; iz < R; iz++)
{
nz_size = nz_part[iz+1]-nz_part[iz];
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size*nz_size;
}
}
}
nx_local = nx_part[p+1] - nx_part[p];
ny_local = ny_part[q+1] - ny_part[q];
nz_local = nz_part[r+1] - nz_part[r];
my_id = r*(P*Q) + q*P + p;
num_procs = P*Q*R;
local_num_rows = nx_local*ny_local*nz_local;
diag_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
offd_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
P_busy = hypre_min(nx,P);
Q_busy = hypre_min(ny,Q);
R_busy = hypre_min(nz,R);
num_cols_offd = 0;
if (p) num_cols_offd += ny_local*nz_local;
if (p < P_busy-1) num_cols_offd += ny_local*nz_local;
if (q) num_cols_offd += nx_local*nz_local;
if (q < Q_busy-1) num_cols_offd += nx_local*nz_local;
if (r) num_cols_offd += nx_local*ny_local;
if (r < R_busy-1) num_cols_offd += nx_local*ny_local;
if (!local_num_rows) num_cols_offd = 0;
col_map_offd = hypre_CTAlloc(HYPRE_Int, num_cols_offd);
cnt = 1;
o_cnt = 1;
diag_i[0] = 0;
offd_i[0] = 0;
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
diag_i[cnt] = diag_i[cnt-1];
offd_i[o_cnt] = offd_i[o_cnt-1];
diag_i[cnt]++;
if (iz > nz_part[r])
diag_i[cnt]++;
else
{
if (iz)
{
offd_i[o_cnt]++;
}
}
if (iy > ny_part[q])
diag_i[cnt]++;
else
{
if (iy)
{
offd_i[o_cnt]++;
}
}
if (ix > nx_part[p])
diag_i[cnt]++;
else
{
if (ix)
{
offd_i[o_cnt]++;
}
}
if (ix+1 < nx_part[p+1])
diag_i[cnt]++;
else
{
if (ix+1 < nx)
{
offd_i[o_cnt]++;
}
}
if (iy+1 < ny_part[q+1])
diag_i[cnt]++;
else
{
if (iy+1 < ny)
{
offd_i[o_cnt]++;
}
}
if (iz+1 < nz_part[r+1])
diag_i[cnt]++;
else
{
if (iz+1 < nz)
{
offd_i[o_cnt]++;
}
}
cnt++;
o_cnt++;
}
}
}
diag_j = hypre_CTAlloc(HYPRE_Int, diag_i[local_num_rows]);
diag_data = hypre_CTAlloc(double, diag_i[local_num_rows]);
if (num_procs > 1)
{
offd_j = hypre_CTAlloc(HYPRE_Int, offd_i[local_num_rows]);
offd_data = hypre_CTAlloc(double, offd_i[local_num_rows]);
}
hypre_CSRMatrix *
hypre_GenerateDifConv( HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int nz,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int R,
HYPRE_Real *value )
{
hypre_CSRMatrix *A;
HYPRE_Int *A_i;
HYPRE_Int *A_j;
HYPRE_Real *A_data;
HYPRE_Int *global_part;
HYPRE_Int ix, iy, iz;
HYPRE_Int p, q, r;
HYPRE_Int cnt;
HYPRE_Int num_rows;
HYPRE_Int row_index;
HYPRE_Int nx_size, ny_size, nz_size;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int *nz_part;
num_rows = nx*ny*nz;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
hypre_GeneratePartitioning(nz,R,&nz_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q*R+1);
global_part[0] = 0;
cnt = 1;
for (iz = 0; iz < R; iz++)
{
nz_size = nz_part[iz+1]-nz_part[iz];
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size*nz_size;
}
}
}
A_i = hypre_CTAlloc(HYPRE_Int, num_rows+1);
cnt = 1;
A_i[0] = 0;
for (r = 0; r < R; r++)
{
for (q = 0; q < Q; q++)
{
for (p = 0; p < P; p++)
{
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
A_i[cnt] = A_i[cnt-1];
A_i[cnt]++;
if (iz > nz_part[r])
A_i[cnt]++;
else
{
if (iz)
{
A_i[cnt]++;
}
}
if (iy > ny_part[q])
A_i[cnt]++;
else
{
if (iy)
{
A_i[cnt]++;
}
}
if (ix > nx_part[p])
A_i[cnt]++;
else
{
if (ix)
{
A_i[cnt]++;
}
}
if (ix+1 < nx_part[p+1])
A_i[cnt]++;
else
{
if (ix+1 < nx)
{
A_i[cnt]++;
}
}
if (iy+1 < ny_part[q+1])
A_i[cnt]++;
else
{
if (iy+1 < ny)
{
A_i[cnt]++;
}
}
if (iz+1 < nz_part[r+1])
A_i[cnt]++;
else
{
if (iz+1 < nz)
{
A_i[cnt]++;
}
}
cnt++;
}
}
}
}
}
}
A_j = hypre_CTAlloc(HYPRE_Int, A_i[num_rows]);
A_data = hypre_CTAlloc(HYPRE_Real, A_i[num_rows]);
row_index = 0;
cnt = 0;
for (r = 0; r < R; r++)
{
for (q = 0; q < Q; q++)
{
ny_size = ny_part[q+1]-ny_part[q];
for (p = 0; p < P; p++)
{
nx_size = nx_part[p+1] - nx_part[p];
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
A_j[cnt] = row_index;
A_data[cnt++] = value[0];
if (iz > nz_part[r])
{
A_j[cnt] = row_index-nx_size*ny_size;
A_data[cnt++] = value[3];
}
else
{
if (iz)
{
A_j[cnt] = map(ix,iy,iz-1,p,q,r-1,P,Q,R,
nx_part,ny_part,nz_part,global_part);
A_data[cnt++] = value[3];
}
}
if (iy > ny_part[q])
{
A_j[cnt] = row_index-nx_size;
A_data[cnt++] = value[2];
}
else
{
if (iy)
{
A_j[cnt] = map(ix,iy-1,iz,p,q-1,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
A_data[cnt++] = value[2];
}
}
if (ix > nx_part[p])
{
A_j[cnt] = row_index-1;
A_data[cnt++] = value[1];
}
else
{
if (ix)
{
A_j[cnt] = map(ix-1,iy,iz,p-1,q,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
A_data[cnt++] = value[1];
}
}
if (ix+1 < nx_part[p+1])
{
A_j[cnt] = row_index+1;
A_data[cnt++] = value[4];
}
else
{
if (ix+1 < nx)
{
A_j[cnt] = map(ix+1,iy,iz,p+1,q,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
A_data[cnt++] = value[4];
}
}
if (iy+1 < ny_part[q+1])
{
A_j[cnt] = row_index+nx_size;
A_data[cnt++] = value[5];
}
else
{
if (iy+1 < ny)
{
A_j[cnt] = map(ix,iy+1,iz,p,q+1,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
A_data[cnt++] = value[5];
}
}
if (iz+1 < nz_part[r+1])
{
A_j[cnt] = row_index+nx_size*ny_size;
A_data[cnt++] = value[6];
}
else
{
if (iz+1 < nz)
{
A_j[cnt] = map(ix,iy,iz+1,p,q,r+1,P,Q,R,
nx_part,ny_part,nz_part,global_part);
A_data[cnt++] = value[6];
}
}
row_index++;
}
}
}
}
}
}
A = hypre_CSRMatrixCreate(num_rows, num_rows, A_i[num_rows]);
hypre_CSRMatrixI(A) = A_i;
hypre_CSRMatrixJ(A) = A_j;
hypre_CSRMatrixData(A) = A_data;
hypre_TFree(nx_part);
hypre_TFree(ny_part);
hypre_TFree(nz_part);
hypre_TFree(global_part);
return A;
}
hypre_CSRMatrix *
hypre_GenerateLaplacian9pt( HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int P,
HYPRE_Int Q,
double *value )
{
hypre_CSRMatrix *A;
HYPRE_Int *A_i;
HYPRE_Int *A_j;
double *A_data;
HYPRE_Int *global_part;
HYPRE_Int ix, iy;
HYPRE_Int p, q;
HYPRE_Int cnt;
HYPRE_Int num_rows;
HYPRE_Int row_index;
HYPRE_Int nx_size, ny_size;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
num_rows = nx*ny;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q+1);
global_part[0] = 0;
cnt = 1;
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size;
}
}
A_i = hypre_CTAlloc(HYPRE_Int,num_rows+1);
cnt = 0;
A_i[0] = 0;
for (q = 0; q < Q; q++)
{
for (p=0; p < P; p++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
cnt++;
A_i[cnt] = A_i[cnt-1];
A_i[cnt]++;
if (iy > ny_part[q])
{
A_i[cnt]++;
if (ix > nx_part[p])
{
A_i[cnt]++;
}
else
{
if (ix)
A_i[cnt]++;
}
if (ix < nx_part[p+1]-1)
{
A_i[cnt]++;
}
else
{
if (ix+1 < nx)
A_i[cnt]++;
}
}
else
{
if (iy)
{
A_i[cnt]++;
if (ix > nx_part[p])
{
A_i[cnt]++;
}
else if (ix)
{
A_i[cnt]++;
}
if (ix < nx_part[p+1]-1)
{
A_i[cnt]++;
}
else if (ix < nx-1)
{
A_i[cnt]++;
}
}
}
if (ix > nx_part[p])
A_i[cnt]++;
else
{
if (ix)
{
A_i[cnt]++;
}
}
if (ix+1 < nx_part[p+1])
A_i[cnt]++;
else
{
if (ix+1 < nx)
{
A_i[cnt]++;
}
}
if (iy+1 < ny_part[q+1])
{
A_i[cnt]++;
if (ix > nx_part[p])
{
A_i[cnt]++;
}
else
{
if (ix)
A_i[cnt]++;
}
if (ix < nx_part[p+1]-1)
{
A_i[cnt]++;
}
else
{
if (ix+1 < nx)
A_i[cnt]++;
}
}
else
{
if (iy+1 < ny)
{
A_i[cnt]++;
if (ix > nx_part[p])
{
A_i[cnt]++;
}
else if (ix)
{
A_i[cnt]++;
}
if (ix < nx_part[p+1]-1)
{
A_i[cnt]++;
}
else if (ix < nx-1)
{
A_i[cnt]++;
}
}
}
}
}
}
}
A_j = hypre_CTAlloc(HYPRE_Int, A_i[num_rows]);
A_data = hypre_CTAlloc(double, A_i[num_rows]);
row_index = 0;
cnt = 0;
for (q=0; q < Q; q++)
{
for (p=0; p < P; p++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
nx_size = nx_part[p+1]-nx_part[p];
A_j[cnt] = row_index;
A_data[cnt++] = value[0];
if (iy > ny_part[q])
{
if (ix > nx_part[p])
{
A_j[cnt] = row_index-nx_size-1 ;
A_data[cnt++] = value[1];
}
else
{
if (ix)
{
A_j[cnt] = map2(ix-1,iy-1,p-1,q,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
A_j[cnt] = row_index-nx_size;
A_data[cnt++] = value[1];
if (ix < nx_part[p+1]-1)
{
A_j[cnt] = row_index-nx_size+1 ;
A_data[cnt++] = value[1];
}
else
{
if (ix+1 < nx)
{
A_j[cnt] = map2(ix+1,iy-1,p+1,q,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
}
else
{
if (iy)
{
if (ix > nx_part[p])
{
A_j[cnt] = map2(ix-1,iy-1,p,q-1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
else if (ix)
{
A_j[cnt] = map2(ix-1,iy-1,p-1,q-1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
A_j[cnt] = map2(ix,iy-1,p,q-1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
if (ix < nx_part[p+1]-1)
{
A_j[cnt] = map2(ix+1,iy-1,p,q-1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
else if (ix+1 < nx)
{
A_j[cnt] = map2(ix+1,iy-1,p+1,q-1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
}
if (ix > nx_part[p])
{
A_j[cnt] = row_index-1;
A_data[cnt++] = value[1];
}
else
{
if (ix)
{
A_j[cnt] = map2(ix-1,iy,p-1,q,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
if (ix+1 < nx_part[p+1])
{
A_j[cnt] = row_index+1;
A_data[cnt++] = value[1];
}
else
{
if (ix+1 < nx)
{
A_j[cnt] = map2(ix+1,iy,p+1,q,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
if (iy+1 < ny_part[q+1])
{
if (ix > nx_part[p])
{
A_j[cnt] = row_index+nx_size-1 ;
A_data[cnt++] = value[1];
}
else
{
if (ix)
{
A_j[cnt] = map2(ix-1,iy+1,p-1,q,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
A_j[cnt] = row_index+nx_size;
A_data[cnt++] = value[1];
if (ix < nx_part[p+1]-1)
{
A_j[cnt] = row_index+nx_size+1 ;
A_data[cnt++] = value[1];
}
else
{
if (ix+1 < nx)
{
A_j[cnt] = map2(ix+1,iy+1,p+1,q,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
}
else
{
if (iy+1 < ny)
{
if (ix > nx_part[p])
{
A_j[cnt] = map2(ix-1,iy+1,p,q+1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
else if (ix)
{
A_j[cnt] = map2(ix-1,iy+1,p-1,q+1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
A_j[cnt] = map2(ix,iy+1,p,q+1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
if (ix < nx_part[p+1]-1)
{
A_j[cnt] = map2(ix+1,iy+1,p,q+1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
else if (ix < nx-1)
{
A_j[cnt] = map2(ix+1,iy+1,p+1,q+1,P,Q,
nx_part,ny_part,global_part);
A_data[cnt++] = value[1];
}
}
}
row_index++;
}
}
}
}
A = hypre_CSRMatrixCreate(num_rows, num_rows, A_i[num_rows]);
hypre_CSRMatrixI(A) = A_i;
hypre_CSRMatrixJ(A) = A_j;
hypre_CSRMatrixData(A) = A_data;
hypre_TFree(nx_part);
hypre_TFree(ny_part);
hypre_TFree(global_part);
return A;
}
HYPRE_ParCSRMatrix
GenerateSysLaplacianVCoef( MPI_Comm comm,
HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int nz,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int R,
HYPRE_Int p,
HYPRE_Int q,
HYPRE_Int r,
HYPRE_Int num_fun,
HYPRE_Real *mtrx,
HYPRE_Real *value )
{
hypre_ParCSRMatrix *A;
hypre_CSRMatrix *diag;
hypre_CSRMatrix *offd;
HYPRE_Int *diag_i;
HYPRE_Int *diag_j;
HYPRE_Real *diag_data;
HYPRE_Int *offd_i;
HYPRE_Int *offd_j;
HYPRE_Real *offd_data;
HYPRE_Int *global_part;
HYPRE_Int ix, iy, iz;
HYPRE_Int cnt, o_cnt;
HYPRE_Int local_num_rows;
HYPRE_Int *col_map_offd;
HYPRE_Int row_index, row, col;
HYPRE_Int index, diag_index;
HYPRE_Int i,j;
HYPRE_Int nx_local, ny_local, nz_local;
HYPRE_Int nx_size, ny_size, nz_size;
HYPRE_Int num_cols_offd;
HYPRE_Int grid_size;
HYPRE_Int local_grid_size;
HYPRE_Int first_j, j_ind;
HYPRE_Int num_coeffs, num_offd_coeffs;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int *nz_part;
HYPRE_Int num_procs, my_id;
HYPRE_Int P_busy, Q_busy, R_busy;
HYPRE_Real val;
/* for indexing in values */
HYPRE_Int sz = num_fun*num_fun;
hypre_MPI_Comm_size(comm,&num_procs);
hypre_MPI_Comm_rank(comm,&my_id);
grid_size = nx*ny*nz;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
hypre_GeneratePartitioning(nz,R,&nz_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q*R+1);
global_part[0] = 0;
cnt = 1;
for (iz = 0; iz < R; iz++)
{
nz_size = nz_part[iz+1]-nz_part[iz];
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size*nz_size;
}
}
}
nx_local = nx_part[p+1] - nx_part[p];
ny_local = ny_part[q+1] - ny_part[q];
nz_local = nz_part[r+1] - nz_part[r];
my_id = r*(P*Q) + q*P + p;
num_procs = P*Q*R;
local_grid_size = nx_local*ny_local*nz_local;
local_num_rows = num_fun*local_grid_size;
diag_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
offd_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
P_busy = hypre_min(nx,P);
Q_busy = hypre_min(ny,Q);
R_busy = hypre_min(nz,R);
num_cols_offd = 0;
if (p) num_cols_offd += ny_local*nz_local;
if (p < P_busy-1) num_cols_offd += ny_local*nz_local;
if (q) num_cols_offd += nx_local*nz_local;
if (q < Q_busy-1) num_cols_offd += nx_local*nz_local;
if (r) num_cols_offd += nx_local*ny_local;
if (r < R_busy-1) num_cols_offd += nx_local*ny_local;
num_cols_offd *= num_fun;
if (!local_num_rows) num_cols_offd = 0;
col_map_offd = hypre_CTAlloc(HYPRE_Int, num_cols_offd);
cnt = 1;
diag_i[0] = 0;
offd_i[0] = 0;
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
diag_i[cnt] = diag_i[cnt-1];
offd_i[cnt] = offd_i[cnt-1];
diag_i[cnt] += num_fun;
if (iz > nz_part[r])
diag_i[cnt] += num_fun;
else
{
if (iz)
{
offd_i[cnt] += num_fun;
}
}
if (iy > ny_part[q])
diag_i[cnt] += num_fun;
else
{
if (iy)
{
offd_i[cnt] += num_fun;
}
}
if (ix > nx_part[p])
diag_i[cnt] += num_fun;
else
{
if (ix)
{
offd_i[cnt] += num_fun;
}
}
if (ix+1 < nx_part[p+1])
diag_i[cnt] += num_fun;
else
{
if (ix+1 < nx)
{
offd_i[cnt] += num_fun;
}
}
if (iy+1 < ny_part[q+1])
diag_i[cnt] += num_fun;
else
{
if (iy+1 < ny)
{
offd_i[cnt] += num_fun;
}
}
if (iz+1 < nz_part[r+1])
diag_i[cnt] += num_fun;
else
{
if (iz+1 < nz)
{
offd_i[cnt] += num_fun;
}
}
num_coeffs = diag_i[cnt]-diag_i[cnt-1];
num_offd_coeffs = offd_i[cnt]-offd_i[cnt-1];
cnt++;
for (i=1; i < num_fun; i++)
{
diag_i[cnt] = diag_i[cnt-1]+num_coeffs;
offd_i[cnt] = offd_i[cnt-1]+num_offd_coeffs;
cnt++;
}
}
}
}
diag_j = hypre_CTAlloc(HYPRE_Int, diag_i[local_num_rows]);
diag_data = hypre_CTAlloc(HYPRE_Real, diag_i[local_num_rows]);
if (num_procs > 1)
{
offd_j = hypre_CTAlloc(HYPRE_Int, offd_i[local_num_rows]);
offd_data = hypre_CTAlloc(HYPRE_Real, offd_i[local_num_rows]);
}
row_index = 0;
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
cnt = diag_i[row_index];;
o_cnt = offd_i[row_index];;
num_coeffs = diag_i[row_index+1]-diag_i[row_index];
num_offd_coeffs = offd_i[row_index+1]-offd_i[row_index];
first_j = row_index;
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = cnt+i*num_coeffs+j;
diag_j[j_ind] = first_j+j;
diag_data[j_ind] = value[0*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
cnt += num_fun;
if (iz > nz_part[r])
{
first_j = row_index-nx_local*ny_local*num_fun;
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = cnt+i*num_coeffs+j;
diag_j[j_ind] = first_j+j;
diag_data[j_ind] = value[3*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
cnt += num_fun;
}
else
{
if (iz)
{
first_j = num_fun*hypre_map(ix,iy,iz-1,p,q,r-1,P,Q,R,
nx_part,ny_part,nz_part,global_part);
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = o_cnt+i*num_offd_coeffs+j;
offd_j[j_ind] = first_j+j;
offd_data[j_ind] = value[3*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
o_cnt += num_fun;
}
}
if (iy > ny_part[q])
{
first_j = row_index-nx_local*num_fun;
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = cnt+i*num_coeffs+j;
diag_j[j_ind] = first_j+j;
diag_data[j_ind] = value[2*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
cnt += num_fun;
}
else
{
if (iy)
{
first_j = num_fun*hypre_map(ix,iy-1,iz,p,q-1,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = o_cnt+i*num_offd_coeffs+j;
offd_j[j_ind] = first_j+j;
offd_data[j_ind] = value[2*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
o_cnt += num_fun;
}
}
if (ix > nx_part[p])
{
first_j = row_index-num_fun;
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = cnt+i*num_coeffs+j;
diag_j[j_ind] = first_j+j;
diag_data[j_ind] = value[1*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
cnt += num_fun;
}
else
{
if (ix)
{
first_j = num_fun*hypre_map(ix-1,iy,iz,p-1,q,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = o_cnt+i*num_offd_coeffs+j;
offd_j[j_ind] = first_j+j;
offd_data[j_ind] = value[1*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
o_cnt += num_fun;
}
}
if (ix+1 < nx_part[p+1])
{
first_j = row_index+num_fun;
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = cnt+i*num_coeffs+j;
diag_j[j_ind] = first_j+j;
diag_data[j_ind] = value[1*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
cnt += num_fun;
}
else
{
if (ix+1 < nx)
{
first_j = num_fun*hypre_map(ix+1,iy,iz,p+1,q,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = o_cnt+i*num_offd_coeffs+j;
offd_j[j_ind] = first_j+j;
offd_data[j_ind] = value[1*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
o_cnt += num_fun;
}
}
if (iy+1 < ny_part[q+1])
{
first_j = row_index+nx_local*num_fun;
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = cnt+i*num_coeffs+j;
diag_j[j_ind] = first_j+j;
diag_data[j_ind] = value[2*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
cnt += num_fun;
}
else
{
if (iy+1 < ny)
{
first_j = num_fun*hypre_map(ix,iy+1,iz,p,q+1,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = o_cnt+i*num_offd_coeffs+j;
offd_j[j_ind] = first_j+j;
offd_data[j_ind] = value[2*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
o_cnt += num_fun;
}
}
if (iz+1 < nz_part[r+1])
{
first_j = row_index+nx_local*ny_local*num_fun;
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = cnt+i*num_coeffs+j;
diag_j[j_ind] = first_j+j;
diag_data[j_ind] = value[3*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
cnt += num_fun;
}
else
{
if (iz+1 < nz)
{
first_j = num_fun*hypre_map(ix,iy,iz+1,p,q,r+1,P,Q,R,
nx_part,ny_part,nz_part,global_part);
for (i=0; i < num_fun; i++)
{
for (j=0; j < num_fun; j++)
{
j_ind = o_cnt+i*num_offd_coeffs+j;
offd_j[j_ind] = first_j+j;
offd_data[j_ind] = value[3*sz + i*num_fun+j]*mtrx[i*num_fun+j];
}
}
o_cnt += num_fun;
}
}
row_index += num_fun;
}
}
}
if (num_procs > 1)
{
cnt = 0;
for (i=0; i < local_num_rows; i+=num_fun)
{
for (j=offd_i[i]; j < offd_i[i+1]; j++)
{
col_map_offd[cnt++] = offd_j[j];
}
}
hypre_qsort0(col_map_offd, 0, num_cols_offd-1);
for (i=0; i < num_fun*num_cols_offd; i++)
for (j=hypre_min(0,abs(i-num_fun)); j < num_cols_offd; j++)
if (offd_j[i] == col_map_offd[j])
{
offd_j[i] = j;
break;
}
}
for (i=0; i < num_procs+1; i++)
global_part[i] *= num_fun;
for (j=1; j< num_fun; j++)
{
for (i=0; i<local_grid_size; i++)
{
row = i*num_fun+j;
diag_index = diag_i[row];
index = diag_index+j;
val = diag_data[diag_index];
col = diag_j[diag_index];
diag_data[diag_index] = diag_data[index];
diag_j[diag_index] = diag_j[index];
diag_data[index] = val;
diag_j[index] = col;
}
}
#ifdef HYPRE_NO_GLOBAL_PARTITION
/* ideally we would use less storage earlier in this function, but this is fine
for testing */
{
HYPRE_Int tmp1, tmp2;
tmp1 = global_part[my_id];
tmp2 = global_part[my_id + 1];
hypre_TFree(global_part);
global_part = hypre_CTAlloc(HYPRE_Int, 2);
global_part[0] = tmp1;
global_part[1] = tmp2;
}
#endif
A = hypre_ParCSRMatrixCreate(comm, num_fun*grid_size, num_fun*grid_size,
global_part, global_part, num_cols_offd,
diag_i[local_num_rows],
offd_i[local_num_rows]);
hypre_ParCSRMatrixColMapOffd(A) = col_map_offd;
diag = hypre_ParCSRMatrixDiag(A);
hypre_CSRMatrixI(diag) = diag_i;
hypre_CSRMatrixJ(diag) = diag_j;
hypre_CSRMatrixData(diag) = diag_data;
offd = hypre_ParCSRMatrixOffd(A);
hypre_CSRMatrixI(offd) = offd_i;
if (num_cols_offd)
{
hypre_CSRMatrixJ(offd) = offd_j;
hypre_CSRMatrixData(offd) = offd_data;
}
hypre_TFree(nx_part);
hypre_TFree(ny_part);
hypre_TFree(nz_part);
return (HYPRE_ParCSRMatrix) A;
}
HYPRE_ParCSRMatrix
GenerateLaplacian( MPI_Comm comm,
HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int nz,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int R,
HYPRE_Int p,
HYPRE_Int q,
HYPRE_Int r,
HYPRE_Real *value )
{
hypre_ParCSRMatrix *A;
hypre_CSRMatrix *diag;
hypre_CSRMatrix *offd;
HYPRE_Int *diag_i;
HYPRE_Int *diag_j;
HYPRE_Real *diag_data;
HYPRE_Int *offd_i;
HYPRE_Int *offd_j = NULL;
HYPRE_Real *offd_data = NULL;
HYPRE_Int *global_part;
HYPRE_Int ix, iy, iz;
HYPRE_Int cnt, o_cnt;
HYPRE_Int local_num_rows;
HYPRE_Int *col_map_offd;
HYPRE_Int row_index;
HYPRE_Int i,j;
HYPRE_Int nx_local, ny_local, nz_local;
HYPRE_Int nx_size, ny_size, nz_size;
HYPRE_Int num_cols_offd;
HYPRE_Int grid_size;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int *nz_part;
HYPRE_Int num_procs, my_id;
HYPRE_Int P_busy, Q_busy, R_busy;
hypre_MPI_Comm_size(comm,&num_procs);
hypre_MPI_Comm_rank(comm,&my_id);
grid_size = nx*ny*nz;
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
hypre_GeneratePartitioning(nz,R,&nz_part);
global_part = hypre_CTAlloc(HYPRE_Int,P*Q*R+1);
global_part[0] = 0;
cnt = 1;
for (iz = 0; iz < R; iz++)
{
nz_size = nz_part[iz+1]-nz_part[iz];
for (iy = 0; iy < Q; iy++)
{
ny_size = ny_part[iy+1]-ny_part[iy];
for (ix = 0; ix < P; ix++)
{
nx_size = nx_part[ix+1] - nx_part[ix];
global_part[cnt] = global_part[cnt-1];
global_part[cnt++] += nx_size*ny_size*nz_size;
}
}
}
nx_local = nx_part[p+1] - nx_part[p];
ny_local = ny_part[q+1] - ny_part[q];
nz_local = nz_part[r+1] - nz_part[r];
my_id = r*(P*Q) + q*P + p;
num_procs = P*Q*R;
local_num_rows = nx_local*ny_local*nz_local;
diag_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
offd_i = hypre_CTAlloc(HYPRE_Int, local_num_rows+1);
P_busy = hypre_min(nx,P);
Q_busy = hypre_min(ny,Q);
R_busy = hypre_min(nz,R);
num_cols_offd = 0;
if (p) num_cols_offd += ny_local*nz_local;
if (p < P_busy-1) num_cols_offd += ny_local*nz_local;
if (q) num_cols_offd += nx_local*nz_local;
if (q < Q_busy-1) num_cols_offd += nx_local*nz_local;
if (r) num_cols_offd += nx_local*ny_local;
if (r < R_busy-1) num_cols_offd += nx_local*ny_local;
if (!local_num_rows) num_cols_offd = 0;
col_map_offd = hypre_CTAlloc(HYPRE_Int, num_cols_offd);
cnt = 1;
o_cnt = 1;
diag_i[0] = 0;
offd_i[0] = 0;
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
diag_i[cnt] = diag_i[cnt-1];
offd_i[o_cnt] = offd_i[o_cnt-1];
diag_i[cnt]++;
if (iz > nz_part[r])
diag_i[cnt]++;
else
{
if (iz)
{
offd_i[o_cnt]++;
}
}
if (iy > ny_part[q])
diag_i[cnt]++;
else
{
if (iy)
{
offd_i[o_cnt]++;
}
}
if (ix > nx_part[p])
diag_i[cnt]++;
else
{
if (ix)
{
offd_i[o_cnt]++;
}
}
if (ix+1 < nx_part[p+1])
diag_i[cnt]++;
else
{
if (ix+1 < nx)
{
offd_i[o_cnt]++;
}
}
if (iy+1 < ny_part[q+1])
diag_i[cnt]++;
else
{
if (iy+1 < ny)
{
offd_i[o_cnt]++;
}
}
if (iz+1 < nz_part[r+1])
diag_i[cnt]++;
else
{
if (iz+1 < nz)
{
offd_i[o_cnt]++;
}
}
cnt++;
o_cnt++;
}
}
}
diag_j = hypre_CTAlloc(HYPRE_Int, diag_i[local_num_rows]);
diag_data = hypre_CTAlloc(HYPRE_Real, diag_i[local_num_rows]);
if (num_procs > 1)
{
offd_j = hypre_CTAlloc(HYPRE_Int, offd_i[local_num_rows]);
offd_data = hypre_CTAlloc(HYPRE_Real, offd_i[local_num_rows]);
}
row_index = 0;
cnt = 0;
o_cnt = 0;
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
diag_j[cnt] = row_index;
diag_data[cnt++] = value[0];
if (iz > nz_part[r])
{
diag_j[cnt] = row_index-nx_local*ny_local;
diag_data[cnt++] = value[3];
}
else
{
if (iz)
{
offd_j[o_cnt] = hypre_map(ix,iy,iz-1,p,q,r-1,P,Q,R,
nx_part,ny_part,nz_part,global_part);
offd_data[o_cnt++] = value[3];
}
}
if (iy > ny_part[q])
{
diag_j[cnt] = row_index-nx_local;
diag_data[cnt++] = value[2];
}
else
{
if (iy)
{
offd_j[o_cnt] = hypre_map(ix,iy-1,iz,p,q-1,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
offd_data[o_cnt++] = value[2];
}
}
if (ix > nx_part[p])
{
diag_j[cnt] = row_index-1;
diag_data[cnt++] = value[1];
}
else
{
if (ix)
{
offd_j[o_cnt] = hypre_map(ix-1,iy,iz,p-1,q,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
offd_data[o_cnt++] = value[1];
}
}
if (ix+1 < nx_part[p+1])
{
diag_j[cnt] = row_index+1;
diag_data[cnt++] = value[1];
}
else
{
if (ix+1 < nx)
{
offd_j[o_cnt] = hypre_map(ix+1,iy,iz,p+1,q,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
offd_data[o_cnt++] = value[1];
}
}
if (iy+1 < ny_part[q+1])
{
diag_j[cnt] = row_index+nx_local;
diag_data[cnt++] = value[2];
}
else
{
if (iy+1 < ny)
{
offd_j[o_cnt] = hypre_map(ix,iy+1,iz,p,q+1,r,P,Q,R,
nx_part,ny_part,nz_part,global_part);
offd_data[o_cnt++] = value[2];
}
}
if (iz+1 < nz_part[r+1])
{
diag_j[cnt] = row_index+nx_local*ny_local;
diag_data[cnt++] = value[3];
}
else
{
if (iz+1 < nz)
{
offd_j[o_cnt] = hypre_map(ix,iy,iz+1,p,q,r+1,P,Q,R,
nx_part,ny_part,nz_part,global_part);
offd_data[o_cnt++] = value[3];
}
}
row_index++;
}
}
}
if (num_procs > 1)
{
for (i=0; i < num_cols_offd; i++)
col_map_offd[i] = offd_j[i];
hypre_qsort0(col_map_offd, 0, num_cols_offd-1);
for (i=0; i < num_cols_offd; i++)
for (j=0; j < num_cols_offd; j++)
if (offd_j[i] == col_map_offd[j])
{
offd_j[i] = j;
break;
}
}
#ifdef HYPRE_NO_GLOBAL_PARTITION
/* ideally we would use less storage earlier in this function, but this is fine
for testing */
{
HYPRE_Int tmp1, tmp2;
tmp1 = global_part[my_id];
tmp2 = global_part[my_id + 1];
hypre_TFree(global_part);
global_part = hypre_CTAlloc(HYPRE_Int, 2);
global_part[0] = tmp1;
global_part[1] = tmp2;
}
#endif
A = hypre_ParCSRMatrixCreate(comm, grid_size, grid_size,
global_part, global_part, num_cols_offd,
diag_i[local_num_rows],
offd_i[local_num_rows]);
hypre_ParCSRMatrixColMapOffd(A) = col_map_offd;
diag = hypre_ParCSRMatrixDiag(A);
hypre_CSRMatrixI(diag) = diag_i;
hypre_CSRMatrixJ(diag) = diag_j;
hypre_CSRMatrixData(diag) = diag_data;
offd = hypre_ParCSRMatrixOffd(A);
hypre_CSRMatrixI(offd) = offd_i;
if (num_cols_offd)
{
hypre_CSRMatrixJ(offd) = offd_j;
hypre_CSRMatrixData(offd) = offd_data;
}
hypre_TFree(nx_part);
hypre_TFree(ny_part);
hypre_TFree(nz_part);
return (HYPRE_ParCSRMatrix) A;
}
float *
GenerateCoordinates( MPI_Comm comm,
HYPRE_Int nx,
HYPRE_Int ny,
HYPRE_Int nz,
HYPRE_Int P,
HYPRE_Int Q,
HYPRE_Int R,
HYPRE_Int p,
HYPRE_Int q,
HYPRE_Int r,
HYPRE_Int coorddim)
{
HYPRE_Int ix, iy, iz;
HYPRE_Int cnt;
HYPRE_Int nx_local, ny_local, nz_local;
HYPRE_Int local_num_rows;
HYPRE_Int *nx_part;
HYPRE_Int *ny_part;
HYPRE_Int *nz_part;
float *coord=NULL;
if (coorddim<1 || coorddim>3) {
return NULL;
}
hypre_GeneratePartitioning(nx,P,&nx_part);
hypre_GeneratePartitioning(ny,Q,&ny_part);
hypre_GeneratePartitioning(nz,R,&nz_part);
nx_local = nx_part[p+1] - nx_part[p];
ny_local = ny_part[q+1] - ny_part[q];
nz_local = nz_part[r+1] - nz_part[r];
local_num_rows = nx_local*ny_local*nz_local;
coord = hypre_CTAlloc(float, coorddim*local_num_rows);
cnt = 0;
for (iz = nz_part[r]; iz < nz_part[r+1]; iz++)
{
for (iy = ny_part[q]; iy < ny_part[q+1]; iy++)
{
for (ix = nx_part[p]; ix < nx_part[p+1]; ix++)
{
/* set coordinates BM Oct 17, 2006 */
if (coord) {
if (nx>1) coord[cnt++] = ix;
if (ny>1) coord[cnt++] = iy;
if (nz>1) coord[cnt++] = iz;
}
}
}
}
hypre_TFree(nx_part);
hypre_TFree(ny_part);
hypre_TFree(nz_part);
return coord;
}
