HYPRE_Int
main( HYPRE_Int argc,
char *argv[] )
{
hypre_ParCSRBooleanMatrix *A;
hypre_ParCSRBooleanMatrix *C;
hypre_CSRBooleanMatrix *As;
HYPRE_Int *row_starts, *col_starts;
HYPRE_Int num_procs, my_id;
/* 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);
row_starts = NULL;
col_starts = NULL;
if (my_id == 0)
{
As = hypre_CSRBooleanMatrixRead("inpr");
hypre_printf(" read input A\n");
}
A = hypre_CSRBooleanMatrixToParCSRBooleanMatrix(hypre_MPI_COMM_WORLD, As, row_starts,
col_starts);
row_starts = hypre_ParCSRBooleanMatrix_Get_RowStarts(A);
col_starts = hypre_ParCSRBooleanMatrix_Get_ColStarts(A);
hypre_ParCSRBooleanMatrixPrint(A, "echo_A" );
hypre_ParCSRBooleanMatrixPrintIJ(A, "echo_AIJ" );
C = hypre_ParBooleanAAt( A );
hypre_ParCSRBooleanMatrixPrint(C, "result");
hypre_ParCSRBooleanMatrixPrintIJ(C, "resultIJ");
if (my_id == 0)
{
hypre_CSRBooleanMatrixDestroy(As);
}
hypre_ParCSRBooleanMatrixDestroy(A);
hypre_ParCSRBooleanMatrixDestroy(C);
hypre_MPI_Finalize();
return 0;
}
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_Int
main( HYPRE_Int argc,
char *argv[] )
{
HYPRE_Int num_procs, myid;
HYPRE_Int verbose = 0, build_matrix_type = 1;
HYPRE_Int index, matrix_arg_index, commpkg_flag=3;
HYPRE_Int i, k, ierr=0;
HYPRE_Int row_start, row_end;
HYPRE_Int col_start, col_end, global_num_rows;
HYPRE_Int *row_part, *col_part;
char *csrfilename;
HYPRE_Int preload = 0, loop = 0, loop2 = LOOP2;
HYPRE_Int bcast_rows[2], *info;
hypre_ParCSRMatrix *parcsr_A, *small_A;
HYPRE_ParCSRMatrix A_temp, A_temp_small;
hypre_CSRMatrix *A_CSR;
hypre_ParCSRCommPkg *comm_pkg;
HYPRE_Int nx, ny, nz;
HYPRE_Int P, Q, R;
HYPRE_Int p, q, r;
HYPRE_Real values[4];
hypre_ParVector *x_new;
hypre_ParVector *y_new, *y;
HYPRE_Int *row_starts;
HYPRE_Real ans;
HYPRE_Real start_time, end_time, total_time, *loop_times;
HYPRE_Real T_avg, T_std;
HYPRE_Int noparmprint = 0;
#if mydebug
HYPRE_Int j, tmp_int;
#endif
/*-----------------------------------------------------------
* 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, &myid );
/*-----------------------------------------------------------
* default - is 27pt laplace
*-----------------------------------------------------------*/
build_matrix_type = 2;
matrix_arg_index = argc;
/*-----------------------------------------------------------
* Parse command line
*-----------------------------------------------------------*/
index = 1;
while ( index < argc)
{
if ( strcmp(argv[index], "-verbose") == 0 )
{
index++;
verbose = 1;
}
else if ( strcmp(argv[index], "-fromonecsrfile") == 0 )
{
index++;
build_matrix_type = 1;
matrix_arg_index = index; /*this tells where the name is*/
}
else if ( strcmp(argv[index], "-commpkg") == 0 )
{
index++;
commpkg_flag = atoi(argv[index++]);
}
else if ( strcmp(argv[index], "-laplacian") == 0 )
{
index++;
build_matrix_type = 2;
matrix_arg_index = index;
}
else if ( strcmp(argv[index], "-27pt") == 0 )
{
index++;
build_matrix_type = 4;
matrix_arg_index = index;
}
/*
else if ( strcmp(argv[index], "-nopreload") == 0 )
{
index++;
preload = 0;
}
*/
else if ( strcmp(argv[index], "-loop") == 0 )
{
index++;
loop = atoi(argv[index++]);
}
else if ( strcmp(argv[index], "-noparmprint") == 0 )
{
index++;
noparmprint = 1;
}
else
{
index++;
/*hypre_printf("Warning: Unrecogized option '%s'\n",argv[index++] );*/
}
}
/*-----------------------------------------------------------
* Setup the Matrix problem
*-----------------------------------------------------------*/
/*-----------------------------------------------------------
* Get actual partitioning-
* read in an actual csr matrix.
*-----------------------------------------------------------*/
if (build_matrix_type ==1) /*read in a csr matrix from one file */
{
if (matrix_arg_index < argc)
{
csrfilename = argv[matrix_arg_index];
}
else
{
hypre_printf("Error: No filename specified \n");
exit(1);
}
if (myid == 0)
{
/*hypre_printf(" FromFile: %s\n", csrfilename);*/
A_CSR = hypre_CSRMatrixRead(csrfilename);
}
row_part = NULL;
col_part = NULL;
parcsr_A = hypre_CSRMatrixToParCSRMatrix(hypre_MPI_COMM_WORLD, A_CSR,
row_part, col_part);
if (myid == 0) hypre_CSRMatrixDestroy(A_CSR);
}
else if (build_matrix_type ==2)
{
myBuildParLaplacian(argc, argv, matrix_arg_index, &A_temp, !noparmprint);
parcsr_A = (hypre_ParCSRMatrix *) A_temp;
}
else if (build_matrix_type ==4)
{
myBuildParLaplacian27pt(argc, argv, matrix_arg_index, &A_temp, !noparmprint);
parcsr_A = (hypre_ParCSRMatrix *) A_temp;
}
/*-----------------------------------------------------------
* create a small problem so that timings are more accurate -
* code gets run twice (small laplace)
*-----------------------------------------------------------*/
/*this is no longer being used - preload = 0 is set at the beginning */
if (preload == 1)
{
/*hypre_printf("preload!\n");*/
values[1] = -1;
values[2] = -1;
values[3] = -1;
values[0] = - 6.0 ;
nx = 2;
ny = num_procs;
nz = 2;
P = 1;
Q = num_procs;
R = 1;
p = myid % P;
q = (( myid - p)/P) % Q;
r = ( myid - p - P*q)/( P*Q );
A_temp_small = (HYPRE_ParCSRMatrix) GenerateLaplacian(hypre_MPI_COMM_WORLD, nx, ny, nz,
P, Q, R, p, q, r, values);
small_A = (hypre_ParCSRMatrix *) A_temp_small;
/*do comm packages*/
hypre_NewCommPkgCreate(small_A);
hypre_NewCommPkgDestroy(small_A);
hypre_MatvecCommPkgCreate(small_A);
hypre_ParCSRMatrixDestroy(small_A);
}
/*-----------------------------------------------------------
* Prepare for timing
*-----------------------------------------------------------*/
/* instead of preloading, let's not time the first one if more than one*/
if (!loop)
{
loop = 1;
/* and don't do any timings */
}
else
{
loop +=1;
if (loop < 2) loop = 2;
}
loop_times = hypre_CTAlloc(HYPRE_Real, loop);
/******************************************************************************************/
hypre_MPI_Barrier(hypre_MPI_COMM_WORLD);
if (commpkg_flag == 1 || commpkg_flag ==3 )
{
/*-----------------------------------------------------------
* Create new comm package
*-----------------------------------------------------------*/
if (!myid) hypre_printf("********************************************************\n" );
/*do loop times*/
for (i=0; i< loop; i++)
{
loop_times[i] = 0.0;
for (k=0; k< loop2; k++)
{
hypre_MPI_Barrier(hypre_MPI_COMM_WORLD);
start_time = hypre_MPI_Wtime();
#if mpip_on
if (i==(loop-1)) hypre_MPI_Pcontrol(1);
#endif
hypre_NewCommPkgCreate(parcsr_A);
#if mpip_on
if (i==(loop-1)) hypre_MPI_Pcontrol(0);
#endif
end_time = hypre_MPI_Wtime();
end_time = end_time - start_time;
hypre_MPI_Allreduce(&end_time, &total_time, 1,
HYPRE_MPI_REAL, hypre_MPI_MAX, hypre_MPI_COMM_WORLD);
loop_times[i] += total_time;
if ( !((i+1)== loop && (k+1) == loop2)) hypre_NewCommPkgDestroy(parcsr_A);
}/*end of loop2 */
} /*end of loop*/
/* calculate the avg and std. */
if (loop > 1)
{
/* calculate the avg and std. */
stats_mo(loop_times, loop, &T_avg, &T_std);
if (!myid) hypre_printf(" NewCommPkgCreate: AVG. wall clock time = %f seconds\n", T_avg);
if (!myid) hypre_printf(" STD. for %d runs = %f\n", loop-1, T_std);
if (!myid) hypre_printf(" (Note: avg./std. timings exclude run 0.)\n");
if (!myid) hypre_printf("********************************************************\n" );
for (i=0; i< loop; i++)
{
if (!myid) hypre_printf(" run %d = %f sec.\n", i, loop_times[i]);
}
if (!myid) hypre_printf("********************************************************\n" );
}
else
{
if (!myid) hypre_printf("********************************************************\n" );
if (!myid) hypre_printf(" NewCommPkgCreate:\n");
if (!myid) hypre_printf(" run time = %f sec.\n", loop_times[0]);
if (!myid) hypre_printf("********************************************************\n" );
}
/*-----------------------------------------------------------
* Verbose printing
*-----------------------------------------------------------*/
/*some verification*/
global_num_rows = hypre_ParCSRMatrixGlobalNumRows(parcsr_A);
if (verbose)
{
ierr = hypre_ParCSRMatrixGetLocalRange( parcsr_A,
&row_start, &row_end ,
&col_start, &col_end );
comm_pkg = hypre_ParCSRMatrixCommPkg(parcsr_A);
hypre_printf("myid = %i, my ACTUAL local range: [%i, %i]\n", myid,
row_start, row_end);
ierr = hypre_GetAssumedPartitionRowRange( myid, global_num_rows, &row_start,
&row_end);
hypre_printf("myid = %i, my assumed local range: [%i, %i]\n", myid,
row_start, row_end);
hypre_printf("myid = %d, num_recvs = %d\n", myid,
hypre_ParCSRCommPkgNumRecvs(comm_pkg) );
#if mydebug
for (i=0; i < hypre_ParCSRCommPkgNumRecvs(comm_pkg); i++)
{
hypre_printf("myid = %d, recv proc = %d, vec_starts = [%d : %d]\n",
myid, hypre_ParCSRCommPkgRecvProcs(comm_pkg)[i],
hypre_ParCSRCommPkgRecvVecStarts(comm_pkg)[i],
hypre_ParCSRCommPkgRecvVecStarts(comm_pkg)[i+1]-1);
}
#endif
hypre_printf("myid = %d, num_sends = %d\n", myid,
hypre_ParCSRCommPkgNumSends(comm_pkg) );
#if mydebug
for (i=0; i <hypre_ParCSRCommPkgNumSends(comm_pkg) ; i++)
{
tmp_int = hypre_ParCSRCommPkgSendMapStarts(comm_pkg)[i+1] -
hypre_ParCSRCommPkgSendMapStarts(comm_pkg)[i];
index = hypre_ParCSRCommPkgSendMapStarts(comm_pkg)[i];
for (j=0; j< tmp_int; j++)
{
hypre_printf("myid = %d, send proc = %d, send element = %d\n",myid,
hypre_ParCSRCommPkgSendProcs(comm_pkg)[i],
hypre_ParCSRCommPkgSendMapElmts(comm_pkg)[index+j]);
}
}
#endif
}
/*-----------------------------------------------------------
* To verify correctness (if commpkg_flag = 3)
*-----------------------------------------------------------*/
if (commpkg_flag == 3 )
{
/*do a matvec - we are assuming a square matrix */
row_starts = hypre_ParCSRMatrixRowStarts(parcsr_A);
x_new = hypre_ParVectorCreate(hypre_MPI_COMM_WORLD, global_num_rows, row_starts);
hypre_ParVectorSetPartitioningOwner(x_new, 0);
hypre_ParVectorInitialize(x_new);
hypre_ParVectorSetRandomValues(x_new, 1);
y_new = hypre_ParVectorCreate(hypre_MPI_COMM_WORLD, global_num_rows, row_starts);
hypre_ParVectorSetPartitioningOwner(y_new, 0);
hypre_ParVectorInitialize(y_new);
hypre_ParVectorSetConstantValues(y_new, 0.0);
/*y = 1.0*A*x+1.0*y */
hypre_ParCSRMatrixMatvec (1.0, parcsr_A, x_new, 1.0, y_new);
}
/*-----------------------------------------------------------
* Clean up after MyComm
*-----------------------------------------------------------*/
hypre_NewCommPkgDestroy(parcsr_A);
}
/******************************************************************************************/
/******************************************************************************************/
hypre_MPI_Barrier(hypre_MPI_COMM_WORLD);
if (commpkg_flag > 1 )
{
/*-----------------------------------------------------------
* Set up standard comm package
*-----------------------------------------------------------*/
bcast_rows[0] = 23;
bcast_rows[1] = 1789;
if (!myid) hypre_printf("********************************************************\n" );
/*do loop times*/
for (i=0; i< loop; i++)
{
loop_times[i] = 0.0;
for (k=0; k< loop2; k++)
{
hypre_MPI_Barrier(hypre_MPI_COMM_WORLD);
start_time = hypre_MPI_Wtime();
#if time_gather
info = hypre_CTAlloc(HYPRE_Int, num_procs);
hypre_MPI_Allgather(bcast_rows, 1, HYPRE_MPI_INT, info, 1, HYPRE_MPI_INT, hypre_MPI_COMM_WORLD);
#endif
hypre_MatvecCommPkgCreate(parcsr_A);
end_time = hypre_MPI_Wtime();
end_time = end_time - start_time;
hypre_MPI_Allreduce(&end_time, &total_time, 1,
HYPRE_MPI_REAL, hypre_MPI_MAX, hypre_MPI_COMM_WORLD);
loop_times[i] += total_time;
if ( !((i+1)== loop && (k+1) == loop2)) hypre_MatvecCommPkgDestroy(hypre_ParCSRMatrixCommPkg(parcsr_A));
}/* end of loop 2*/
} /*end of loop*/
/* calculate the avg and std. */
if (loop > 1)
{
stats_mo(loop_times, loop, &T_avg, &T_std);
if (!myid) hypre_printf("Current CommPkgCreate: AVG. wall clock time = %f seconds\n", T_avg);
if (!myid) hypre_printf(" STD. for %d runs = %f\n", loop-1, T_std);
if (!myid) hypre_printf(" (Note: avg./std. timings exclude run 0.)\n");
if (!myid) hypre_printf("********************************************************\n" );
for (i=0; i< loop; i++)
{
if (!myid) hypre_printf(" run %d = %f sec.\n", i, loop_times[i]);
}
if (!myid) hypre_printf("********************************************************\n" );
}
else
{
if (!myid) hypre_printf("********************************************************\n" );
if (!myid) hypre_printf(" Current CommPkgCreate:\n");
if (!myid) hypre_printf(" run time = %f sec.\n", loop_times[0]);
if (!myid) hypre_printf("********************************************************\n" );
}
/*-----------------------------------------------------------
* Verbose printing
*-----------------------------------------------------------*/
/*some verification*/
if (verbose)
{
ierr = hypre_ParCSRMatrixGetLocalRange( parcsr_A,
&row_start, &row_end ,
&col_start, &col_end );
comm_pkg = hypre_ParCSRMatrixCommPkg(parcsr_A);
hypre_printf("myid = %i, std - my local range: [%i, %i]\n", myid,
row_start, row_end);
ierr = hypre_ParCSRMatrixGetLocalRange( parcsr_A,
&row_start, &row_end ,
&col_start, &col_end );
hypre_printf("myid = %d, std - num_recvs = %d\n", myid,
hypre_ParCSRCommPkgNumRecvs(comm_pkg) );
#if mydebug
for (i=0; i < hypre_ParCSRCommPkgNumRecvs(comm_pkg); i++)
{
hypre_printf("myid = %d, std - recv proc = %d, vec_starts = [%d : %d]\n",
myid, hypre_ParCSRCommPkgRecvProcs(comm_pkg)[i],
hypre_ParCSRCommPkgRecvVecStarts(comm_pkg)[i],
hypre_ParCSRCommPkgRecvVecStarts(comm_pkg)[i+1]-1);
}
#endif
hypre_printf("myid = %d, std - num_sends = %d\n", myid,
hypre_ParCSRCommPkgNumSends(comm_pkg));
#if mydebug
for (i=0; i <hypre_ParCSRCommPkgNumSends(comm_pkg) ; i++)
{
tmp_int = hypre_ParCSRCommPkgSendMapStarts(comm_pkg)[i+1] -
hypre_ParCSRCommPkgSendMapStarts(comm_pkg)[i];
index = hypre_ParCSRCommPkgSendMapStarts(comm_pkg)[i];
for (j=0; j< tmp_int; j++)
{
hypre_printf("myid = %d, std - send proc = %d, send element = %d\n",myid,
hypre_ParCSRCommPkgSendProcs(comm_pkg)[i],
hypre_ParCSRCommPkgSendMapElmts(comm_pkg)[index+j]);
}
}
#endif
}
/*-----------------------------------------------------------
* Verify correctness
*-----------------------------------------------------------*/
if (commpkg_flag == 3 )
{
global_num_rows = hypre_ParCSRMatrixGlobalNumRows(parcsr_A);
row_starts = hypre_ParCSRMatrixRowStarts(parcsr_A);
y = hypre_ParVectorCreate(hypre_MPI_COMM_WORLD, global_num_rows,row_starts);
hypre_ParVectorSetPartitioningOwner(y, 0);
hypre_ParVectorInitialize(y);
hypre_ParVectorSetConstantValues(y, 0.0);
hypre_ParCSRMatrixMatvec (1.0, parcsr_A, x_new, 1.0, y);
}
}
/*-----------------------------------------------------------
* Compare matvecs for both comm packages (3)
*-----------------------------------------------------------*/
if (commpkg_flag == 3 )
{
/*make sure that y and y_new are the same - now y_new should=0*/
hypre_ParVectorAxpy( -1.0, y, y_new );
hypre_ParVectorSetRandomValues(y, 1);
ans = hypre_ParVectorInnerProd( y, y_new );
if (!myid)
{
if ( fabs(ans) > 1e-8 )
{
hypre_printf("!!!!! WARNING !!!!! should be zero if correct = %6.10f\n",
ans);
}
else
{
hypre_printf("Matvecs match ( should be zero = %6.10f )\n",
ans);
}
}
}
/*-----------------------------------------------------------
* Clean up
*-----------------------------------------------------------*/
hypre_ParCSRMatrixDestroy(parcsr_A); /*this calls the standard comm
package destroy - but we'll destroy
ours separately until it is
incorporated */
if (commpkg_flag == 3 )
{
hypre_ParVectorDestroy(x_new);
hypre_ParVectorDestroy(y);
hypre_ParVectorDestroy(y_new);
}
hypre_MPI_Finalize();
return(ierr);
}
HYPRE_Int main( HYPRE_Int argc, char *argv[] )
{
hypre_ParCSRMatrix *par_matrix, *g_matrix, **submatrices;
hypre_CSRMatrix *A_diag, *A_offd;
hypre_CSRBlockMatrix *diag;
hypre_CSRBlockMatrix *offd;
hypre_ParCSRBlockMatrix *par_blk_matrix, *par_blk_matrixT, *rap_matrix;
hypre_Vector *x_local;
hypre_Vector *y_local;
hypre_ParVector *x;
hypre_ParVector *y;
HYPRE_Solver gmres_solver, precon;
HYPRE_Int *diag_i, *diag_j, *offd_i, *offd_j;
HYPRE_Int *diag_i2, *diag_j2, *offd_i2, *offd_j2;
double *diag_d, *diag_d2, *offd_d, *offd_d2;
HYPRE_Int mypid, local_size, nprocs;
HYPRE_Int global_num_rows, global_num_cols, num_cols_offd;
HYPRE_Int num_nonzeros_diag, num_nonzeros_offd, *colMap;
HYPRE_Int ii, jj, kk, row, col, nnz, *indices, *colMap2;
double *data, ddata, *y_data;
HYPRE_Int *row_starts, *col_starts, *rstarts, *cstarts;
HYPRE_Int *row_starts2, *col_starts2;
HYPRE_Int block_size=2, bnnz=4, *index_set;
FILE *fp;
/* --------------------------------------------- */
/* Initialize MPI */
/* --------------------------------------------- */
hypre_MPI_Init(&argc, &argv);
hypre_MPI_Comm_rank(hypre_MPI_COMM_WORLD, &mypid);
hypre_MPI_Comm_size(hypre_MPI_COMM_WORLD, &nprocs);
/* build and fetch matrix */
MyBuildParLaplacian9pt((HYPRE_ParCSRMatrix *) &par_matrix);
global_num_rows = hypre_ParCSRMatrixGlobalNumRows(par_matrix);
global_num_cols = hypre_ParCSRMatrixGlobalNumCols(par_matrix);
row_starts = hypre_ParCSRMatrixRowStarts(par_matrix);
col_starts = hypre_ParCSRMatrixColStarts(par_matrix);
A_diag = hypre_ParCSRMatrixDiag(par_matrix);
A_offd = hypre_ParCSRMatrixOffd(par_matrix);
num_cols_offd = hypre_CSRMatrixNumCols(A_offd);
num_nonzeros_diag = hypre_CSRMatrixNumNonzeros(A_diag);
num_nonzeros_offd = hypre_CSRMatrixNumNonzeros(A_offd);
/* --------------------------------------------- */
/* build vector and apply matvec */
/* --------------------------------------------- */
x = hypre_ParVectorCreate(hypre_MPI_COMM_WORLD,global_num_cols,col_starts);
hypre_ParVectorSetPartitioningOwner(x,0);
hypre_ParVectorInitialize(x);
x_local = hypre_ParVectorLocalVector(x);
data = hypre_VectorData(x_local);
local_size = col_starts[mypid+1] - col_starts[mypid];
for (ii = 0; ii < local_size; ii++) data[ii] = 1.0;
y = hypre_ParVectorCreate(hypre_MPI_COMM_WORLD,global_num_rows,row_starts);
hypre_ParVectorSetPartitioningOwner(y,0);
hypre_ParVectorInitialize(y);
hypre_ParCSRMatrixMatvec (1.0, par_matrix, x, 0.0, y);
ddata = hypre_ParVectorInnerProd(y, y);
if (mypid == 0) hypre_printf("y inner product = %e\n", ddata);
hypre_ParVectorDestroy(x);
hypre_ParVectorDestroy(y);
/* --------------------------------------------- */
/* build block matrix */
/* --------------------------------------------- */
rstarts = hypre_CTAlloc(HYPRE_Int, nprocs+1);
for (ii = 0; ii <= nprocs; ii++) rstarts[ii] = row_starts[ii];
cstarts = hypre_CTAlloc(HYPRE_Int, nprocs+1);
for (ii = 0; ii <= nprocs; ii++) cstarts[ii] = col_starts[ii];
par_blk_matrix = hypre_ParCSRBlockMatrixCreate(hypre_MPI_COMM_WORLD,block_size,
global_num_rows, global_num_cols, rstarts,
cstarts, num_cols_offd, num_nonzeros_diag,
num_nonzeros_offd);
colMap = hypre_ParCSRMatrixColMapOffd(par_matrix);
if (num_cols_offd > 0) colMap2 = hypre_CTAlloc(HYPRE_Int, num_cols_offd);
else colMap2 = NULL;
for (ii = 0; ii < num_cols_offd; ii++) colMap2[ii] = colMap[ii];
hypre_ParCSRBlockMatrixColMapOffd(par_blk_matrix) = colMap2;
diag_i = hypre_CSRMatrixI(hypre_ParCSRMatrixDiag(par_matrix));
diag_j = hypre_CSRMatrixJ(hypre_ParCSRMatrixDiag(par_matrix));
diag_d = hypre_CSRMatrixData(hypre_ParCSRMatrixDiag(par_matrix));
diag = hypre_ParCSRBlockMatrixDiag(par_blk_matrix);
diag_i2 = hypre_CTAlloc(HYPRE_Int, local_size+1);
diag_j2 = hypre_CTAlloc(HYPRE_Int, num_nonzeros_diag);
diag_d2 = hypre_CTAlloc(double, num_nonzeros_diag*bnnz);
for (ii = 0; ii <= local_size; ii++) diag_i2[ii] = diag_i[ii];
for (ii = 0; ii < num_nonzeros_diag; ii++) diag_j2[ii] = diag_j[ii];
hypre_CSRBlockMatrixI(diag) = diag_i2;
hypre_CSRBlockMatrixJ(diag) = diag_j2;
for (ii = 0; ii < num_nonzeros_diag; ii++)
{
for (jj = 0; jj < block_size; jj++)
for (kk = 0; kk < block_size; kk++)
{
if (jj <= kk)
diag_d2[ii*bnnz+jj*block_size+kk] = diag_d[ii];
else
diag_d2[ii*bnnz+jj*block_size+kk] = 0.0;
}
}
hypre_CSRBlockMatrixData(diag) = diag_d2;
offd_i = hypre_CSRMatrixI(hypre_ParCSRMatrixOffd(par_matrix));
offd_j = hypre_CSRMatrixJ(hypre_ParCSRMatrixOffd(par_matrix));
offd_d = hypre_CSRMatrixData(hypre_ParCSRMatrixOffd(par_matrix));
offd = hypre_ParCSRBlockMatrixOffd(par_blk_matrix);
offd_i2 = hypre_CTAlloc(HYPRE_Int, local_size+1);
for (ii = 0; ii <= local_size; ii++) offd_i2[ii] = offd_i[ii];
hypre_CSRBlockMatrixI(offd) = offd_i2;
if (num_cols_offd)
{
offd_j2 = hypre_CTAlloc(HYPRE_Int, num_nonzeros_offd);
for (ii = 0; ii < num_nonzeros_offd; ii++) offd_j2[ii] = offd_j[ii];
hypre_CSRBlockMatrixJ(offd) = offd_j2;
offd_d2 = hypre_CTAlloc(double, num_nonzeros_offd*bnnz);
for (ii = 0; ii < num_nonzeros_offd; ii++)
{
for (jj = 0; jj < block_size; jj++)
for (kk = 0; kk < block_size; kk++)
{
if (jj <= kk)
offd_d2[ii*bnnz+jj*block_size+kk] = offd_d[ii];
else
offd_d2[ii*bnnz+jj*block_size+kk] = 0.0;
}
}
hypre_CSRBlockMatrixData(offd) = offd_d2;
}
else
{
HYPRE_Int main( HYPRE_Int argc, char *argv[] )
{
HYPRE_Int arg_index;
HYPRE_Int print_usage;
HYPRE_Int build_matrix_arg_index;
HYPRE_Int solver_id;
HYPRE_Int ierr,i,j;
HYPRE_Int num_iterations;
HYPRE_ParCSRMatrix parcsr_A;
HYPRE_Int num_procs, myid;
HYPRE_Int local_row;
HYPRE_Int time_index;
MPI_Comm comm;
HYPRE_Int M, N;
HYPRE_Int first_local_row, last_local_row;
HYPRE_Int first_local_col, last_local_col;
HYPRE_Int size, *col_ind;
HYPRE_Real *values;
/* parameters for BoomerAMG */
HYPRE_Real strong_threshold;
HYPRE_Int num_grid_sweeps;
HYPRE_Real relax_weight;
/* parameters for GMRES */
HYPRE_Int k_dim;
char *paramString = new char[100];
/*-----------------------------------------------------------
* Initialize some stuff
*-----------------------------------------------------------*/
hypre_MPI_Init(&argc, &argv);
hypre_MPI_Comm_size(hypre_MPI_COMM_WORLD, &num_procs );
hypre_MPI_Comm_rank(hypre_MPI_COMM_WORLD, &myid );
/*-----------------------------------------------------------
* Set defaults
*-----------------------------------------------------------*/
build_matrix_arg_index = argc;
solver_id = 0;
strong_threshold = 0.25;
num_grid_sweeps = 2;
relax_weight = 0.5;
k_dim = 20;
/*-----------------------------------------------------------
* Parse command line
*-----------------------------------------------------------*/
print_usage = 0;
arg_index = 1;
while ( (arg_index < argc) && (!print_usage) )
{
if ( strcmp(argv[arg_index], "-solver") == 0 )
{
arg_index++;
solver_id = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-dbg") == 0 )
{
arg_index++;
atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-help") == 0 )
{
print_usage = 1;
}
else
{
arg_index++;
}
}
/*-----------------------------------------------------------
* Print usage info
*-----------------------------------------------------------*/
if ( (print_usage) && (myid == 0) )
{
hypre_printf("\n");
hypre_printf("Usage: %s [<options>]\n", argv[0]);
hypre_printf("\n");
hypre_printf(" -solver <ID> : solver ID\n");
hypre_printf(" 0=DS-PCG 1=ParaSails-PCG \n");
hypre_printf(" 2=AMG-PCG 3=DS-GMRES \n");
hypre_printf(" 4=PILUT-GMRES 5=AMG-GMRES \n");
hypre_printf("\n");
hypre_printf(" -rlx <val> : relaxation type\n");
hypre_printf(" 0=Weighted Jacobi \n");
hypre_printf(" 1=Gauss-Seidel (very slow!) \n");
hypre_printf(" 3=Hybrid Jacobi/Gauss-Seidel \n");
hypre_printf("\n");
exit(1);
}
/*-----------------------------------------------------------
* Print driver parameters
*-----------------------------------------------------------*/
if (myid == 0)
{
hypre_printf("Running with these driver parameters:\n");
hypre_printf(" solver ID = %d\n", solver_id);
}
/*-----------------------------------------------------------
* Set up matrix
*-----------------------------------------------------------*/
strcpy(paramString, "LS Interface");
time_index = hypre_InitializeTiming(paramString);
hypre_BeginTiming(time_index);
BuildParLaplacian27pt(argc, argv, build_matrix_arg_index, &parcsr_A);
/*-----------------------------------------------------------
* Copy the parcsr matrix into the LSI through interface calls
*-----------------------------------------------------------*/
ierr = HYPRE_ParCSRMatrixGetComm( parcsr_A, &comm );
ierr += HYPRE_ParCSRMatrixGetDims( parcsr_A, &M, &N );
ierr = HYPRE_ParCSRMatrixGetLocalRange( parcsr_A,
&first_local_row, &last_local_row ,
&first_local_col, &last_local_col );
HYPRE_LinSysCore H(hypre_MPI_COMM_WORLD);
HYPRE_Int numLocalEqns = last_local_row - first_local_row + 1;
H.createMatricesAndVectors(M,first_local_row+1,numLocalEqns);
HYPRE_Int index;
HYPRE_Int *rowLengths = new HYPRE_Int[numLocalEqns];
HYPRE_Int **colIndices = new HYPRE_Int*[numLocalEqns];
local_row = 0;
for (i=first_local_row; i<= last_local_row; i++)
{
ierr += HYPRE_ParCSRMatrixGetRow(parcsr_A,i,&size,&col_ind,&values );
rowLengths[local_row] = size;
colIndices[local_row] = new HYPRE_Int[size];
for (j=0; j<size; j++) colIndices[local_row][j] = col_ind[j] + 1;
local_row++;
HYPRE_ParCSRMatrixRestoreRow(parcsr_A,i,&size,&col_ind,&values);
}
H.allocateMatrix(colIndices, rowLengths);
delete [] rowLengths;
for (i=0; i< numLocalEqns; i++) delete [] colIndices[i];
delete [] colIndices;
HYPRE_Int *newColInd;
for (i=first_local_row; i<= last_local_row; i++)
{
ierr += HYPRE_ParCSRMatrixGetRow(parcsr_A,i,&size,&col_ind,&values );
newColInd = new HYPRE_Int[size];
for (j=0; j<size; j++) newColInd[j] = col_ind[j] + 1;
H.sumIntoSystemMatrix(i+1,size,(const HYPRE_Real*)values,
(const HYPRE_Int*)newColInd);
delete [] newColInd;
ierr += HYPRE_ParCSRMatrixRestoreRow(parcsr_A,i,&size,&col_ind,&values);
}
H.matrixLoadComplete();
HYPRE_ParCSRMatrixDestroy(parcsr_A);
/*-----------------------------------------------------------
* Set up the RHS and initial guess
*-----------------------------------------------------------*/
HYPRE_Real ddata=1.0;
HYPRE_Int status;
for (i=first_local_row; i<= last_local_row; i++)
{
index = i + 1;
H.sumIntoRHSVector(1,(const HYPRE_Real*) &ddata, (const HYPRE_Int*) &index);
}
hypre_EndTiming(time_index);
strcpy(paramString, "LS Interface");
hypre_PrintTiming(paramString, hypre_MPI_COMM_WORLD);
hypre_FinalizeTiming(time_index);
hypre_ClearTiming();
/*-----------------------------------------------------------
* Solve the system using PCG
*-----------------------------------------------------------*/
if ( solver_id == 0 )
{
strcpy(paramString, "solver cg");
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: DS-PCG\n");
strcpy(paramString, "preconditioner diagonal");
H.parameters(1, ¶mString);
}
else if ( solver_id == 1 )
{
strcpy(paramString, "solver cg");
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: ParaSails-PCG\n");
strcpy(paramString, "preconditioner parasails");
H.parameters(1, ¶mString);
strcpy(paramString, "parasailsNlevels 1");
H.parameters(1, ¶mString);
strcpy(paramString, "parasailsThreshold 0.1");
H.parameters(1, ¶mString);
}
else if ( solver_id == 2 )
{
strcpy(paramString, "solver cg");
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: AMG-PCG\n");
strcpy(paramString, "preconditioner boomeramg");
H.parameters(1, ¶mString);
strcpy(paramString, "amgCoarsenType falgout");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "amgStrongThreshold %e", strong_threshold);
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "amgNumSweeps %d", num_grid_sweeps);
H.parameters(1, ¶mString);
strcpy(paramString, "amgRelaxType jacobi");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "amgRelaxWeight %e", relax_weight);
H.parameters(1, ¶mString);
}
else if ( solver_id == 3 )
{
strcpy(paramString, "solver cg");
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: Poly-PCG\n");
strcpy(paramString, "preconditioner poly");
H.parameters(1, ¶mString);
strcpy(paramString, "polyOrder 9");
H.parameters(1, ¶mString);
}
else if ( solver_id == 4 )
{
strcpy(paramString, "solver gmres");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "gmresDim %d", k_dim);
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: DS-GMRES\n");
strcpy(paramString, "preconditioner diagonal");
H.parameters(1, ¶mString);
}
else if ( solver_id == 5 )
{
strcpy(paramString, "solver gmres");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "gmresDim %d", k_dim);
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: PILUT-GMRES\n");
strcpy(paramString, "preconditioner pilut");
H.parameters(1, ¶mString);
strcpy(paramString, "pilutRowSize 0");
H.parameters(1, ¶mString);
strcpy(paramString, "pilutDropTol 0.0");
H.parameters(1, ¶mString);
}
else if ( solver_id == 6 )
{
strcpy(paramString, "solver gmres");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "gmresDim %d", k_dim);
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: AMG-GMRES\n");
strcpy(paramString, "preconditioner boomeramg");
H.parameters(1, ¶mString);
strcpy(paramString, "amgCoarsenType falgout");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "amgStrongThreshold %e", strong_threshold);
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "amgNumSweeps %d", num_grid_sweeps);
H.parameters(1, ¶mString);
strcpy(paramString, "amgRelaxType jacobi");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "amgRelaxWeight %e", relax_weight);
H.parameters(1, ¶mString);
}
else if ( solver_id == 7 )
{
strcpy(paramString, "solver gmres");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "gmresDim %d", k_dim);
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: DDILUT-GMRES\n");
strcpy(paramString, "preconditioner ddilut");
H.parameters(1, ¶mString);
strcpy(paramString, "ddilutFillin 5.0");
H.parameters(1, ¶mString);
strcpy(paramString, "ddilutDropTol 0.0");
H.parameters(1, ¶mString);
}
else if ( solver_id == 8 )
{
strcpy(paramString, "solver gmres");
H.parameters(1, ¶mString);
hypre_sprintf(paramString, "gmresDim %d", k_dim);
H.parameters(1, ¶mString);
if (myid == 0) hypre_printf("Solver: POLY-GMRES\n");
strcpy(paramString, "preconditioner poly");
H.parameters(1, ¶mString);
strcpy(paramString, "polyOrder 5");
H.parameters(1, ¶mString);
}
strcpy(paramString, "Krylov Solve");
time_index = hypre_InitializeTiming(paramString);
hypre_BeginTiming(time_index);
H.launchSolver(status, num_iterations);
hypre_EndTiming(time_index);
strcpy(paramString, "Solve phase times");
hypre_PrintTiming(paramString, hypre_MPI_COMM_WORLD);
hypre_FinalizeTiming(time_index);
hypre_ClearTiming();
if (myid == 0)
{
hypre_printf("\n Iterations = %d\n", num_iterations);
hypre_printf("\n");
}
/*-----------------------------------------------------------
* Finalize things
*-----------------------------------------------------------*/
delete [] paramString;
hypre_MPI_Finalize();
return (0);
}
hypre_int
main( hypre_int argc,
char *argv[] )
{
HYPRE_Int arg_index;
HYPRE_Int print_usage;
HYPRE_Int nx, ny, nz;
HYPRE_Int P, Q, R;
HYPRE_Int bx, by, bz;
HYPRE_StructGrid from_grid, to_grid;
HYPRE_StructVector from_vector, to_vector, check_vector;
HYPRE_CommPkg comm_pkg;
HYPRE_Int time_index;
HYPRE_Int num_procs, myid;
HYPRE_Int p, q, r;
HYPRE_Int dim;
HYPRE_Int nblocks ;
HYPRE_Int **ilower, **iupper, **iupper2;
HYPRE_Int istart[3];
HYPRE_Int i, ix, iy, iz, ib;
HYPRE_Int print_system = 0;
HYPRE_Real check;
/*-----------------------------------------------------------
* Initialize some stuff
*-----------------------------------------------------------*/
/* 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, &myid );
/*-----------------------------------------------------------
* Set defaults
*-----------------------------------------------------------*/
dim = 3;
nx = 2;
ny = 2;
nz = 2;
P = num_procs;
Q = 1;
R = 1;
bx = 1;
by = 1;
bz = 1;
istart[0] = 1;
istart[1] = 1;
istart[2] = 1;
/*-----------------------------------------------------------
* Parse command line
*-----------------------------------------------------------*/
print_usage = 0;
arg_index = 1;
while (arg_index < argc)
{
if ( strcmp(argv[arg_index], "-n") == 0 )
{
arg_index++;
nx = atoi(argv[arg_index++]);
ny = atoi(argv[arg_index++]);
nz = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-istart") == 0 )
{
arg_index++;
istart[0] = atoi(argv[arg_index++]);
istart[1] = atoi(argv[arg_index++]);
istart[2] = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-P") == 0 )
{
arg_index++;
P = atoi(argv[arg_index++]);
Q = atoi(argv[arg_index++]);
R = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-b") == 0 )
{
arg_index++;
bx = atoi(argv[arg_index++]);
by = atoi(argv[arg_index++]);
bz = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-d") == 0 )
{
arg_index++;
dim = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-print") == 0 )
{
arg_index++;
print_system = 1;
}
else if ( strcmp(argv[arg_index], "-help") == 0 )
{
print_usage = 1;
break;
}
else
{
arg_index++;
}
}
/*-----------------------------------------------------------
* Print usage info
*-----------------------------------------------------------*/
if ( (print_usage) && (myid == 0) )
{
hypre_printf("\n");
hypre_printf("Usage: %s [<options>]\n", argv[0]);
hypre_printf("\n");
hypre_printf(" -n <nx> <ny> <nz> : problem size per block\n");
hypre_printf(" -istart <ix> <iy> <iz> : start of box\n");
hypre_printf(" -P <Px> <Py> <Pz> : processor topology\n");
hypre_printf(" -b <bx> <by> <bz> : blocking per processor\n");
hypre_printf(" -d <dim> : problem dimension (2 or 3)\n");
hypre_printf(" -print : print vectors\n");
hypre_printf("\n");
}
if ( print_usage )
{
exit(1);
}
/*-----------------------------------------------------------
* Check a few things
*-----------------------------------------------------------*/
if ((P*Q*R) > num_procs)
{
if (myid == 0)
{
hypre_printf("Error: PxQxR is more than the number of processors\n");
}
exit(1);
}
else if ((P*Q*R) < num_procs)
{
if (myid == 0)
{
hypre_printf("Warning: PxQxR is less than the number of processors\n");
}
}
/*-----------------------------------------------------------
* Print driver parameters
*-----------------------------------------------------------*/
if (myid == 0)
{
hypre_printf("Running with these driver parameters:\n");
hypre_printf(" (nx, ny, nz) = (%d, %d, %d)\n", nx, ny, nz);
hypre_printf(" (ix, iy, iz) = (%d, %d, %d)\n",
istart[0],istart[1],istart[2]);
hypre_printf(" (Px, Py, Pz) = (%d, %d, %d)\n", P, Q, R);
hypre_printf(" (bx, by, bz) = (%d, %d, %d)\n", bx, by, bz);
hypre_printf(" dim = %d\n", dim);
}
/*-----------------------------------------------------------
* Set up the stencil structure (7 points) when matrix is NOT read from file
* Set up the grid structure used when NO files are read
*-----------------------------------------------------------*/
switch (dim)
{
case 1:
nblocks = bx;
p = myid % P;
break;
case 2:
nblocks = bx*by;
p = myid % P;
q = (( myid - p)/P) % Q;
break;
case 3:
nblocks = bx*by*bz;
p = myid % P;
q = (( myid - p)/P) % Q;
r = ( myid - p - P*q)/( P*Q );
break;
}
if (myid >= (P*Q*R))
{
/* My processor has no data on it */
nblocks = bx = by = bz = 0;
}
/*-----------------------------------------------------------
* prepare space for the extents
*-----------------------------------------------------------*/
ilower = hypre_CTAlloc(HYPRE_Int*, nblocks);
iupper = hypre_CTAlloc(HYPRE_Int*, nblocks);
iupper2 = hypre_CTAlloc(HYPRE_Int*, nblocks);
for (i = 0; i < nblocks; i++)
{
ilower[i] = hypre_CTAlloc(HYPRE_Int, dim);
iupper[i] = hypre_CTAlloc(HYPRE_Int, dim);
iupper2[i] = hypre_CTAlloc(HYPRE_Int, dim);
}
ib = 0;
switch (dim)
{
case 1:
for (ix = 0; ix < bx; ix++)
{
ilower[ib][0] = istart[0]+ nx*(bx*p+ix);
iupper[ib][0] = istart[0]+ nx*(bx*p+ix+1) - 1;
iupper2[ib][0] = iupper[ib][0];
if ( (ix == (bx-1)) && (p < (P-1)) )
iupper2[ib][0] = iupper[ib][0] + 1;
ib++;
}
break;
case 2:
for (iy = 0; iy < by; iy++)
for (ix = 0; ix < bx; ix++)
{
ilower[ib][0] = istart[0]+ nx*(bx*p+ix);
iupper[ib][0] = istart[0]+ nx*(bx*p+ix+1) - 1;
ilower[ib][1] = istart[1]+ ny*(by*q+iy);
iupper[ib][1] = istart[1]+ ny*(by*q+iy+1) - 1;
iupper2[ib][0] = iupper[ib][0];
iupper2[ib][1] = iupper[ib][1];
if ( (ix == (bx-1)) && (p < (P-1)) )
iupper2[ib][0] = iupper[ib][0] + 1;
if ( (iy == (by-1)) && (q < (Q-1)) )
iupper2[ib][1] = iupper[ib][1] + 1;
ib++;
}
break;
case 3:
for (iz = 0; iz < bz; iz++)
for (iy = 0; iy < by; iy++)
for (ix = 0; ix < bx; ix++)
{
ilower[ib][0] = istart[0]+ nx*(bx*p+ix);
iupper[ib][0] = istart[0]+ nx*(bx*p+ix+1) - 1;
ilower[ib][1] = istart[1]+ ny*(by*q+iy);
iupper[ib][1] = istart[1]+ ny*(by*q+iy+1) - 1;
ilower[ib][2] = istart[2]+ nz*(bz*r+iz);
iupper[ib][2] = istart[2]+ nz*(bz*r+iz+1) - 1;
iupper2[ib][0] = iupper[ib][0];
iupper2[ib][1] = iupper[ib][1];
iupper2[ib][2] = iupper[ib][2];
if ( (ix == (bx-1)) && (p < (P-1)) )
iupper2[ib][0] = iupper[ib][0] + 1;
if ( (iy == (by-1)) && (q < (Q-1)) )
iupper2[ib][1] = iupper[ib][1] + 1;
if ( (iz == (bz-1)) && (r < (R-1)) )
iupper2[ib][2] = iupper[ib][2] + 1;
ib++;
}
break;
}
HYPRE_StructGridCreate(hypre_MPI_COMM_WORLD, dim, &from_grid);
HYPRE_StructGridCreate(hypre_MPI_COMM_WORLD, dim, &to_grid);
for (ib = 0; ib < nblocks; ib++)
{
HYPRE_StructGridSetExtents(from_grid, ilower[ib], iupper[ib]);
HYPRE_StructGridSetExtents(to_grid, ilower[ib], iupper2[ib]);
}
HYPRE_StructGridAssemble(from_grid);
HYPRE_StructGridAssemble(to_grid);
/*-----------------------------------------------------------
* Set up the vectors
*-----------------------------------------------------------*/
HYPRE_StructVectorCreate(hypre_MPI_COMM_WORLD, from_grid, &from_vector);
HYPRE_StructVectorInitialize(from_vector);
AddValuesVector(from_grid, from_vector, 1.0);
HYPRE_StructVectorAssemble(from_vector);
HYPRE_StructVectorCreate(hypre_MPI_COMM_WORLD, to_grid, &to_vector);
HYPRE_StructVectorInitialize(to_vector);
AddValuesVector(to_grid, to_vector, 0.0);
HYPRE_StructVectorAssemble(to_vector);
/* Vector used to check the migration */
HYPRE_StructVectorCreate(hypre_MPI_COMM_WORLD, to_grid, &check_vector);
HYPRE_StructVectorInitialize(check_vector);
AddValuesVector(to_grid, check_vector, 1.0);
HYPRE_StructVectorAssemble(check_vector);
/*-----------------------------------------------------------
* Migrate
*-----------------------------------------------------------*/
time_index = hypre_InitializeTiming("Struct Migrate");
hypre_BeginTiming(time_index);
HYPRE_StructVectorGetMigrateCommPkg(from_vector, to_vector, &comm_pkg);
HYPRE_StructVectorMigrate(comm_pkg, from_vector, to_vector);
HYPRE_CommPkgDestroy(comm_pkg);
hypre_EndTiming(time_index);
hypre_PrintTiming("Struct Migrate", hypre_MPI_COMM_WORLD);
hypre_FinalizeTiming(time_index);
/*-----------------------------------------------------------
* Check the migration and print the result
*-----------------------------------------------------------*/
hypre_StructAxpy(-1.0, to_vector, check_vector);
check = hypre_StructInnerProd (check_vector, check_vector);
if (myid == 0)
{
printf("\nCheck = %1.0f (success = 0)\n\n", check);
}
/*-----------------------------------------------------------
* Print out the vectors
*-----------------------------------------------------------*/
if (print_system)
{
HYPRE_StructVectorPrint("struct_migrate.out.xfr", from_vector, 0);
HYPRE_StructVectorPrint("struct_migrate.out.xto", to_vector, 0);
}
/*-----------------------------------------------------------
* Finalize things
*-----------------------------------------------------------*/
HYPRE_StructGridDestroy(from_grid);
HYPRE_StructGridDestroy(to_grid);
for (i = 0; i < nblocks; i++)
{
hypre_TFree(ilower[i]);
hypre_TFree(iupper[i]);
hypre_TFree(iupper2[i]);
}
hypre_TFree(ilower);
hypre_TFree(iupper);
hypre_TFree(iupper2);
HYPRE_StructVectorDestroy(from_vector);
HYPRE_StructVectorDestroy(to_vector);
HYPRE_StructVectorDestroy(check_vector);
/* Finalize MPI */
hypre_MPI_Finalize();
return (0);
}
HYPRE_Int main (HYPRE_Int argc, char *argv[])
{
HYPRE_Int i;
HYPRE_Int myid, num_procs;
HYPRE_Int N, n;
HYPRE_Int ilower, iupper;
HYPRE_Int local_size, extra;
HYPRE_Int solver_id;
HYPRE_Int print_solution;
double h, h2;
#ifdef HYPRE_FORTRAN
hypre_F90_Obj A;
hypre_F90_Obj parcsr_A;
hypre_F90_Obj b;
hypre_F90_Obj par_b;
hypre_F90_Obj x;
hypre_F90_Obj par_x;
hypre_F90_Obj solver, precond;
hypre_F90_Obj long_temp_COMM;
HYPRE_Int temp_COMM;
HYPRE_Int precond_id;
HYPRE_Int one = 1;
HYPRE_Int two = 2;
HYPRE_Int three = 3;
HYPRE_Int six = 6;
HYPRE_Int twenty = 20;
HYPRE_Int thousand = 1000;
HYPRE_Int hypre_type = HYPRE_PARCSR;
double oo1 = 1.e-3;
double tol = 1.e-7;
#else
HYPRE_IJMatrix A;
HYPRE_ParCSRMatrix parcsr_A;
HYPRE_IJVector b;
HYPRE_ParVector par_b;
HYPRE_IJVector x;
HYPRE_ParVector par_x;
HYPRE_Solver solver, precond;
#endif
/* Initialize MPI */
hypre_MPI_Init(&argc, &argv);
hypre_MPI_Comm_rank(hypre_MPI_COMM_WORLD, &myid);
hypre_MPI_Comm_size(hypre_MPI_COMM_WORLD, &num_procs);
/* Default problem parameters */
n = 33;
solver_id = 0;
print_solution = 0;
/* Parse command line */
{
HYPRE_Int arg_index = 0;
HYPRE_Int print_usage = 0;
while (arg_index < argc)
{
if ( strcmp(argv[arg_index], "-n") == 0 )
{
arg_index++;
n = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-solver") == 0 )
{
arg_index++;
solver_id = atoi(argv[arg_index++]);
}
else if ( strcmp(argv[arg_index], "-print_solution") == 0 )
{
arg_index++;
print_solution = 1;
}
else if ( strcmp(argv[arg_index], "-help") == 0 )
{
print_usage = 1;
break;
}
else
{
arg_index++;
}
}
if ((print_usage) && (myid == 0))
{
hypre_printf("\n");
hypre_printf("Usage: %s [<options>]\n", argv[0]);
hypre_printf("\n");
hypre_printf(" -n <n> : problem size in each direction (default: 33)\n");
hypre_printf(" -solver <ID> : solver ID\n");
hypre_printf(" 0 - AMG (default) \n");
hypre_printf(" 1 - AMG-PCG\n");
hypre_printf(" 8 - ParaSails-PCG\n");
hypre_printf(" 50 - PCG\n");
hypre_printf(" -print_solution : print the solution vector\n");
hypre_printf("\n");
}
if (print_usage)
{
hypre_MPI_Finalize();
return (0);
}
}
/* Preliminaries: want at least one processor per row */
if (n*n < num_procs) n = sqrt(num_procs) + 1;
N = n*n; /* global number of rows */
h = 1.0/(n+1); /* mesh size*/
h2 = h*h;
/* Each processor knows only of its own rows - the range is denoted by ilower
and upper. Here we partition the rows. We account for the fact that
N may not divide evenly by the number of processors. */
local_size = N/num_procs;
extra = N - local_size*num_procs;
ilower = local_size*myid;
ilower += hypre_min(myid, extra);
iupper = local_size*(myid+1);
iupper += hypre_min(myid+1, extra);
iupper = iupper - 1;
/* How many rows do I have? */
local_size = iupper - ilower + 1;
/* Create the matrix.
Note that this is a square matrix, so we indicate the row partition
size twice (since number of rows = number of cols) */
#ifdef HYPRE_FORTRAN
long_temp_COMM = (hypre_F90_Obj) hypre_MPI_COMM_WORLD;
temp_COMM = (HYPRE_Int) hypre_MPI_COMM_WORLD;
HYPRE_IJMatrixCreate(&long_temp_COMM, &ilower, &iupper, &ilower, &iupper, &A);
#else
HYPRE_IJMatrixCreate(hypre_MPI_COMM_WORLD, ilower, iupper, ilower, iupper, &A);
#endif
/* Choose a parallel csr format storage (see the User's Manual) */
#ifdef HYPRE_FORTRAN
HYPRE_IJMatrixSetObjectType(&A, &hypre_type);
#else
HYPRE_IJMatrixSetObjectType(A, HYPRE_PARCSR);
#endif
/* Initialize before setting coefficients */
#ifdef HYPRE_FORTRAN
HYPRE_IJMatrixInitialize(&A);
#else
HYPRE_IJMatrixInitialize(A);
#endif
/* Now go through my local rows and set the matrix entries.
Each row has at most 5 entries. For example, if n=3:
A = [M -I 0; -I M -I; 0 -I M]
M = [4 -1 0; -1 4 -1; 0 -1 4]
Note that here we are setting one row at a time, though
one could set all the rows together (see the User's Manual).
*/
{
HYPRE_Int nnz;
double values[5];
HYPRE_Int cols[5];
for (i = ilower; i <= iupper; i++)
{
nnz = 0;
/* The left identity block:position i-n */
if ((i-n)>=0)
{
cols[nnz] = i-n;
values[nnz] = -1.0;
nnz++;
}
/* The left -1: position i-1 */
if (i%n)
{
cols[nnz] = i-1;
values[nnz] = -1.0;
nnz++;
}
/* Set the diagonal: position i */
cols[nnz] = i;
values[nnz] = 4.0;
nnz++;
/* The right -1: position i+1 */
if ((i+1)%n)
{
cols[nnz] = i+1;
values[nnz] = -1.0;
nnz++;
}
/* The right identity block:position i+n */
if ((i+n)< N)
{
cols[nnz] = i+n;
values[nnz] = -1.0;
nnz++;
}
/* Set the values for row i */
#ifdef HYPRE_FORTRAN
HYPRE_IJMatrixSetValues(&A, &one, &nnz, &i, &cols[0], &values[0]);
#else
HYPRE_IJMatrixSetValues(A, 1, &nnz, &i, cols, values);
#endif
}
}
/* Assemble after setting the coefficients */
#ifdef HYPRE_FORTRAN
HYPRE_IJMatrixAssemble(&A);
#else
HYPRE_IJMatrixAssemble(A);
#endif
/* Get the parcsr matrix object to use */
#ifdef HYPRE_FORTRAN
HYPRE_IJMatrixGetObject(&A, &parcsr_A);
HYPRE_IJMatrixGetObject(&A, &parcsr_A);
#else
HYPRE_IJMatrixGetObject(A, (void**) &parcsr_A);
HYPRE_IJMatrixGetObject(A, (void**) &parcsr_A);
#endif
/* Create the rhs and solution */
#ifdef HYPRE_FORTRAN
HYPRE_IJVectorCreate(&temp_COMM, &ilower, &iupper, &b);
HYPRE_IJVectorSetObjectType(&b, &hypre_type);
HYPRE_IJVectorInitialize(&b);
#else
HYPRE_IJVectorCreate(hypre_MPI_COMM_WORLD, ilower, iupper,&b);
HYPRE_IJVectorSetObjectType(b, HYPRE_PARCSR);
HYPRE_IJVectorInitialize(b);
#endif
#ifdef HYPRE_FORTRAN
HYPRE_IJVectorCreate(&temp_COMM, &ilower, &iupper, &x);
HYPRE_IJVectorSetObjectType(&x, &hypre_type);
HYPRE_IJVectorInitialize(&x);
#else
HYPRE_IJVectorCreate(hypre_MPI_COMM_WORLD, ilower, iupper,&x);
HYPRE_IJVectorSetObjectType(x, HYPRE_PARCSR);
HYPRE_IJVectorInitialize(x);
#endif
/* Set the rhs values to h^2 and the solution to zero */
{
double *rhs_values, *x_values;
HYPRE_Int *rows;
rhs_values = calloc(local_size, sizeof(double));
x_values = calloc(local_size, sizeof(double));
rows = calloc(local_size, sizeof(HYPRE_Int));
for (i=0; i<local_size; i++)
{
rhs_values[i] = h2;
x_values[i] = 0.0;
rows[i] = ilower + i;
}
#ifdef HYPRE_FORTRAN
HYPRE_IJVectorSetValues(&b, &local_size, &rows[0], &rhs_values[0]);
HYPRE_IJVectorSetValues(&x, &local_size, &rows[0], &x_values[0]);
#else
HYPRE_IJVectorSetValues(b, local_size, rows, rhs_values);
HYPRE_IJVectorSetValues(x, local_size, rows, x_values);
#endif
free(x_values);
free(rhs_values);
free(rows);
}
#ifdef HYPRE_FORTRAN
HYPRE_IJVectorAssemble(&b);
HYPRE_IJVectorGetObject(&b, &par_b);
#else
HYPRE_IJVectorAssemble(b);
HYPRE_IJVectorGetObject(b, (void **) &par_b);
#endif
#ifdef HYPRE_FORTRAN
HYPRE_IJVectorAssemble(&x);
HYPRE_IJVectorGetObject(&x, &par_x);
#else
HYPRE_IJVectorAssemble(x);
HYPRE_IJVectorGetObject(x, (void **) &par_x);
#endif
/* Choose a solver and solve the system */
/* AMG */
if (solver_id == 0)
{
HYPRE_Int num_iterations;
double final_res_norm;
/* Create solver */
#ifdef HYPRE_FORTRAN
HYPRE_BoomerAMGCreate(&solver);
#else
HYPRE_BoomerAMGCreate(&solver);
#endif
/* Set some parameters (See Reference Manual for more parameters) */
#ifdef HYPRE_FORTRAN
HYPRE_BoomerAMGSetPrintLevel(&solver, &three); /* print solve info + parameters */
HYPRE_BoomerAMGSetCoarsenType(&solver, &six); /* Falgout coarsening */
HYPRE_BoomerAMGSetRelaxType(&solver, &three); /* G-S/Jacobi hybrid relaxation */
HYPRE_BoomerAMGSetNumSweeps(&solver, &one); /* Sweeeps on each level */
HYPRE_BoomerAMGSetMaxLevels(&solver, &twenty); /* maximum number of levels */
HYPRE_BoomerAMGSetTol(&solver, &tol); /* conv. tolerance */
#else
HYPRE_BoomerAMGSetPrintLevel(solver, 3); /* print solve info + parameters */
HYPRE_BoomerAMGSetCoarsenType(solver, 6); /* Falgout coarsening */
HYPRE_BoomerAMGSetRelaxType(solver, 3); /* G-S/Jacobi hybrid relaxation */
HYPRE_BoomerAMGSetNumSweeps(solver, 1); /* Sweeeps on each level */
HYPRE_BoomerAMGSetMaxLevels(solver, 20); /* maximum number of levels */
HYPRE_BoomerAMGSetTol(solver, 1e-7); /* conv. tolerance */
#endif
/* Now setup and solve! */
#ifdef HYPRE_FORTRAN
HYPRE_BoomerAMGSetup(&solver, &parcsr_A, &par_b, &par_x);
HYPRE_BoomerAMGSolve(&solver, &parcsr_A, &par_b, &par_x);
#else
HYPRE_BoomerAMGSetup(solver, parcsr_A, par_b, par_x);
HYPRE_BoomerAMGSolve(solver, parcsr_A, par_b, par_x);
#endif
/* Run info - needed logging turned on */
#ifdef HYPRE_FORTRAN
HYPRE_BoomerAMGGetNumIterations(&solver, &num_iterations);
HYPRE_BoomerAMGGetFinalRelativeResidualNorm(&solver, &final_res_norm);
#else
HYPRE_BoomerAMGGetNumIterations(solver, &num_iterations);
HYPRE_BoomerAMGGetFinalRelativeResidualNorm(solver, &final_res_norm);
#endif
if (myid == 0)
{
hypre_printf("\n");
hypre_printf("Iterations = %d\n", num_iterations);
hypre_printf("Final Relative Residual Norm = %e\n", final_res_norm);
hypre_printf("\n");
}
/* Destroy solver */
#ifdef HYPRE_FORTRAN
HYPRE_BoomerAMGDestroy(&solver);
#else
HYPRE_BoomerAMGDestroy(solver);
#endif
}
/* PCG */
else if (solver_id == 50)
{
HYPRE_Int num_iterations;
double final_res_norm;
/* Create solver */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGCreate(&temp_COMM, &solver);
#else
HYPRE_ParCSRPCGCreate(hypre_MPI_COMM_WORLD, &solver);
#endif
/* Set some parameters (See Reference Manual for more parameters) */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGSetMaxIter(&solver, &thousand); /* max iterations */
HYPRE_ParCSRPCGSetTol(&solver, &tol); /* conv. tolerance */
HYPRE_ParCSRPCGSetTwoNorm(&solver, &one); /* use the two norm as the stopping criteria */
HYPRE_ParCSRPCGSetPrintLevel(&solver, &two); /* prints out the iteration info */
#else
HYPRE_PCGSetMaxIter(solver, 1000); /* max iterations */
HYPRE_PCGSetTol(solver, 1e-7); /* conv. tolerance */
HYPRE_PCGSetTwoNorm(solver, 1); /* use the two norm as the stopping criteria */
HYPRE_PCGSetPrintLevel(solver, 2); /* prints out the iteration info */
HYPRE_PCGSetLogging(solver, 1); /* needed to get run info later */
#endif
/* Now setup and solve! */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGSetup(&solver, &parcsr_A, &par_b, &par_x);
HYPRE_ParCSRPCGSolve(&solver, &parcsr_A, &par_b, &par_x);
#else
HYPRE_ParCSRPCGSetup(solver, parcsr_A, par_b, par_x);
HYPRE_ParCSRPCGSolve(solver, parcsr_A, par_b, par_x);
#endif
/* Run info - needed logging turned on */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGGetNumIterations(&solver, &num_iterations);
HYPRE_ParCSRPCGGetFinalRelativeResidualNorm(&solver, &final_res_norm);
#else
HYPRE_PCGGetNumIterations(solver, &num_iterations);
HYPRE_PCGGetFinalRelativeResidualNorm(solver, &final_res_norm);
#endif
if (myid == 0)
{
hypre_printf("\n");
hypre_printf("Iterations = %d\n", num_iterations);
hypre_printf("Final Relative Residual Norm = %e\n", final_res_norm);
hypre_printf("\n");
}
/* Destroy solver */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGDestroy(&solver);
#else
HYPRE_ParCSRPCGDestroy(solver);
#endif
}
/* PCG with AMG preconditioner */
else if (solver_id == 1)
{
HYPRE_Int num_iterations;
double final_res_norm;
/* Create solver */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGCreate(&temp_COMM, &solver);
#else
HYPRE_ParCSRPCGCreate(hypre_MPI_COMM_WORLD, &solver);
#endif
/* Set some parameters (See Reference Manual for more parameters) */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGSetMaxIter(&solver, &thousand); /* max iterations */
HYPRE_ParCSRPCGSetTol(&solver, &tol); /* conv. tolerance */
HYPRE_ParCSRPCGSetTwoNorm(&solver, &one); /* use the two norm as the stopping criteria */
HYPRE_ParCSRPCGSetPrintLevel(&solver, &two); /* print solve info */
#else
HYPRE_PCGSetMaxIter(solver, 1000); /* max iterations */
HYPRE_PCGSetTol(solver, 1e-7); /* conv. tolerance */
HYPRE_PCGSetTwoNorm(solver, 1); /* use the two norm as the stopping criteria */
HYPRE_PCGSetPrintLevel(solver, 2); /* print solve info */
HYPRE_PCGSetLogging(solver, 1); /* needed to get run info later */
#endif
/* Now set up the AMG preconditioner and specify any parameters */
#ifdef HYPRE_FORTRAN
HYPRE_BoomerAMGCreate(&precond);
HYPRE_BoomerAMGSetPrintLevel(&precond, &one); /* print amg solution info*/
HYPRE_BoomerAMGSetCoarsenType(&precond, &six);
HYPRE_BoomerAMGSetRelaxType(&precond, &three);
HYPRE_BoomerAMGSetNumSweeps(&precond, &one);
HYPRE_BoomerAMGSetTol(&precond, &oo1);
#else
HYPRE_BoomerAMGCreate(&precond);
HYPRE_BoomerAMGSetPrintLevel(precond, 1); /* print amg solution info*/
HYPRE_BoomerAMGSetCoarsenType(precond, 6);
HYPRE_BoomerAMGSetRelaxType(precond, 3);
HYPRE_BoomerAMGSetNumSweeps(precond, 1);
HYPRE_BoomerAMGSetTol(precond, 1e-3);
#endif
/* Set the PCG preconditioner */
#ifdef HYPRE_FORTRAN
precond_id = 2;
HYPRE_ParCSRPCGSetPrecond(&solver, &precond_id, &precond);
#else
HYPRE_PCGSetPrecond(solver, (HYPRE_PtrToSolverFcn) HYPRE_BoomerAMGSolve,
(HYPRE_PtrToSolverFcn) HYPRE_BoomerAMGSetup, precond);
#endif
/* Now setup and solve! */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGSetup(&solver, &parcsr_A, &par_b, &par_x);
HYPRE_ParCSRPCGSolve(&solver, &parcsr_A, &par_b, &par_x);
#else
HYPRE_ParCSRPCGSetup(solver, parcsr_A, par_b, par_x);
HYPRE_ParCSRPCGSolve(solver, parcsr_A, par_b, par_x);
#endif
/* Run info - needed logging turned on */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGGetNumIterations(&solver, &num_iterations);
HYPRE_ParCSRPCGGetFinalRelativeResidualNorm(&solver, &final_res_norm);
#else
HYPRE_PCGGetNumIterations(solver, &num_iterations);
HYPRE_PCGGetFinalRelativeResidualNorm(solver, &final_res_norm);
#endif
if (myid == 0)
{
hypre_printf("\n");
hypre_printf("Iterations = %d\n", num_iterations);
hypre_printf("Final Relative Residual Norm = %e\n", final_res_norm);
hypre_printf("\n");
}
/* Destroy solver and preconditioner */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGDestroy(&solver);
HYPRE_BoomerAMGDestroy(&precond);
#else
HYPRE_ParCSRPCGDestroy(solver);
HYPRE_BoomerAMGDestroy(precond);
#endif
}
/* PCG with Parasails Preconditioner */
else if (solver_id == 8)
{
HYPRE_Int num_iterations;
double final_res_norm;
HYPRE_Int sai_max_levels = 1;
double sai_threshold = 0.1;
double sai_filter = 0.05;
HYPRE_Int sai_sym = 1;
/* Create solver */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGCreate(&temp_COMM, &solver);
#else
HYPRE_ParCSRPCGCreate(hypre_MPI_COMM_WORLD, &solver);
#endif
/* Set some parameters (See Reference Manual for more parameters) */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGSetMaxIter(&solver, &thousand); /* max iterations */
HYPRE_ParCSRPCGSetTol(&solver, &tol); /* conv. tolerance */
HYPRE_ParCSRPCGSetTwoNorm(&solver, &one); /* use the two norm as the stopping criteria */
HYPRE_ParCSRPCGSetPrintLevel(&solver, &two); /* print solve info */
#else
HYPRE_PCGSetMaxIter(solver, 1000); /* max iterations */
HYPRE_PCGSetTol(solver, 1e-7); /* conv. tolerance */
HYPRE_PCGSetTwoNorm(solver, 1); /* use the two norm as the stopping criteria */
HYPRE_PCGSetPrintLevel(solver, 2); /* print solve info */
HYPRE_PCGSetLogging(solver, 1); /* needed to get run info later */
#endif
/* Now set up the ParaSails preconditioner and specify any parameters */
#ifdef HYPRE_FORTRAN
HYPRE_ParaSailsCreate(&temp_COMM, &precond);
#else
HYPRE_ParaSailsCreate(hypre_MPI_COMM_WORLD, &precond);
#endif
/* Set some parameters (See Reference Manual for more parameters) */
#ifdef HYPRE_FORTRAN
HYPRE_ParaSailsSetParams(&precond, &sai_threshold, &sai_max_levels);
HYPRE_ParaSailsSetFilter(&precond, &sai_filter);
HYPRE_ParaSailsSetSym(&precond, &sai_sym);
HYPRE_ParaSailsSetLogging(&precond, &three);
#else
HYPRE_ParaSailsSetParams(precond, sai_threshold, sai_max_levels);
HYPRE_ParaSailsSetFilter(precond, sai_filter);
HYPRE_ParaSailsSetSym(precond, sai_sym);
HYPRE_ParaSailsSetLogging(precond, 3);
#endif
/* Set the PCG preconditioner */
#ifdef HYPRE_FORTRAN
precond_id = 4;
HYPRE_ParCSRPCGSetPrecond(&solver, &precond_id, &precond);
#else
HYPRE_PCGSetPrecond(solver, (HYPRE_PtrToSolverFcn) HYPRE_ParaSailsSolve,
(HYPRE_PtrToSolverFcn) HYPRE_ParaSailsSetup, precond);
#endif
/* Now setup and solve! */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGSetup(&solver, &parcsr_A, &par_b, &par_x);
HYPRE_ParCSRPCGSolve(&solver, &parcsr_A, &par_b, &par_x);
#else
HYPRE_ParCSRPCGSetup(solver, parcsr_A, par_b, par_x);
HYPRE_ParCSRPCGSolve(solver, parcsr_A, par_b, par_x);
#endif
/* Run info - needed logging turned on */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGGetNumIterations(&solver, &num_iterations);
HYPRE_ParCSRPCGGetFinalRelativeResidualNorm(&solver, &final_res_norm);
#else
HYPRE_PCGGetNumIterations(solver, &num_iterations);
HYPRE_PCGGetFinalRelativeResidualNorm(solver, &final_res_norm);
#endif
if (myid == 0)
{
hypre_printf("\n");
hypre_printf("Iterations = %d\n", num_iterations);
hypre_printf("Final Relative Residual Norm = %e\n", final_res_norm);
hypre_printf("\n");
}
/* Destory solver and preconditioner */
#ifdef HYPRE_FORTRAN
HYPRE_ParCSRPCGDestroy(&solver);
HYPRE_ParaSailsDestroy(&precond);
#else
HYPRE_ParCSRPCGDestroy(solver);
HYPRE_ParaSailsDestroy(precond);
#endif
}
else
{
if (myid ==0) hypre_printf("Invalid solver id specified.\n");
}
/* Print the solution */
#ifdef HYPRE_FORTRAN
if (print_solution)
HYPRE_IJVectorPrint(&x, "ij.out.x");
#else
if (print_solution)
HYPRE_IJVectorPrint(x, "ij.out.x");
#endif
/* Clean up */
#ifdef HYPRE_FORTRAN
HYPRE_IJMatrixDestroy(&A);
HYPRE_IJVectorDestroy(&b);
HYPRE_IJVectorDestroy(&x);
#else
HYPRE_IJMatrixDestroy(A);
HYPRE_IJVectorDestroy(b);
HYPRE_IJVectorDestroy(x);
#endif
/* Finalize MPI*/
hypre_MPI_Finalize();
return(0);
}
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 i;
HYPRE_Int *partitioning;
HYPRE_Complex prod;
HYPRE_Complex *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;
vector1 = hypre_ParVectorCreate(hypre_MPI_COMM_WORLD,global_size,partitioning);
partitioning = hypre_ParVectorPartitioning(vector1);
hypre_ParVectorInitialize(vector1);
local_vector = hypre_ParVectorLocalVector(vector1);
data = hypre_VectorData(local_vector);
local_size = hypre_VectorSize(local_vector);
first_index = partitioning[my_id];
for (i=0; i < local_size; i++)
data[i] = first_index+i;
/*
hypre_ParVectorPrint(vector1, "Vector");
*/
local_vector2 = hypre_SeqVectorCreate(global_size);
hypre_SeqVectorInitialize(local_vector2);
data2 = hypre_VectorData(local_vector2);
for (i=0; i < global_size; i++)
data2[i] = i+1;
/* partitioning = hypre_CTAlloc(HYPRE_Int,4);
partitioning[0] = 0;
partitioning[1] = 10;
partitioning[2] = 10;
partitioning[3] = 20;
*/
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
*-----------------------------------------------------------*/
tmp_vector = hypre_ParVectorRead(hypre_MPI_COMM_WORLD, "Convert");
/*
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_Int
main( HYPRE_Int argc,
char *argv[] )
{
hypre_CSRMatrix *matrix;
hypre_CSRMatrix *matrix1;
hypre_ParCSRMatrix *par_matrix;
hypre_Vector *x_local;
hypre_Vector *y_local;
hypre_Vector *y2_local;
hypre_ParVector *x;
hypre_ParVector *x2;
hypre_ParVector *y;
hypre_ParVector *y2;
HYPRE_Int vecstride_x, idxstride_x, vecstride_y, idxstride_y;
HYPRE_Int num_procs, my_id;
HYPRE_Int local_size;
HYPRE_Int num_vectors;
HYPRE_Int global_num_rows, global_num_cols;
HYPRE_Int first_index;
HYPRE_Int i, j, ierr=0;
double *data, *data2;
HYPRE_Int *row_starts, *col_starts;
char file_name[80];
/* 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);
if (my_id == 0)
{
matrix = hypre_CSRMatrixRead("input");
hypre_printf(" read input\n");
}
row_starts = NULL;
col_starts = NULL;
par_matrix = hypre_CSRMatrixToParCSRMatrix(hypre_MPI_COMM_WORLD, matrix,
row_starts, col_starts);
hypre_printf(" converted\n");
matrix1 = hypre_ParCSRMatrixToCSRMatrixAll(par_matrix);
hypre_sprintf(file_name,"matrix1.%d",my_id);
if (matrix1) hypre_CSRMatrixPrint(matrix1, file_name);
hypre_ParCSRMatrixPrint(par_matrix,"matrix");
hypre_ParCSRMatrixPrintIJ(par_matrix,0,0,"matrixIJ");
par_matrix = hypre_ParCSRMatrixRead(hypre_MPI_COMM_WORLD,"matrix");
global_num_cols = hypre_ParCSRMatrixGlobalNumCols(par_matrix);
hypre_printf(" global_num_cols %d\n", global_num_cols);
global_num_rows = hypre_ParCSRMatrixGlobalNumRows(par_matrix);
col_starts = hypre_ParCSRMatrixColStarts(par_matrix);
first_index = col_starts[my_id];
local_size = col_starts[my_id+1] - first_index;
num_vectors = 3;
x = hypre_ParMultiVectorCreate( hypre_MPI_COMM_WORLD, global_num_cols,
col_starts, num_vectors );
hypre_ParVectorSetPartitioningOwner(x,0);
hypre_ParVectorInitialize(x);
x_local = hypre_ParVectorLocalVector(x);
data = hypre_VectorData(x_local);
vecstride_x = hypre_VectorVectorStride(x_local);
idxstride_x = hypre_VectorIndexStride(x_local);
for ( j=0; j<num_vectors; ++j )
for (i=0; i < local_size; i++)
data[i*idxstride_x + j*vecstride_x] = first_index+i+1 + 100*j;
x2 = hypre_ParMultiVectorCreate( hypre_MPI_COMM_WORLD, global_num_cols,
col_starts, num_vectors );
hypre_ParVectorSetPartitioningOwner(x2,0);
hypre_ParVectorInitialize(x2);
hypre_ParVectorSetConstantValues(x2,2.0);
row_starts = hypre_ParCSRMatrixRowStarts(par_matrix);
first_index = row_starts[my_id];
local_size = row_starts[my_id+1] - first_index;
y = hypre_ParMultiVectorCreate( hypre_MPI_COMM_WORLD, global_num_rows,
row_starts, num_vectors );
hypre_ParVectorSetPartitioningOwner(y,0);
hypre_ParVectorInitialize(y);
y_local = hypre_ParVectorLocalVector(y);
y2 = hypre_ParMultiVectorCreate( hypre_MPI_COMM_WORLD, global_num_rows,
row_starts, num_vectors );
hypre_ParVectorSetPartitioningOwner(y2,0);
hypre_ParVectorInitialize(y2);
y2_local = hypre_ParVectorLocalVector(y2);
data2 = hypre_VectorData(y2_local);
vecstride_y = hypre_VectorVectorStride(y2_local);
idxstride_y = hypre_VectorIndexStride(y2_local);
for ( j=0; j<num_vectors; ++j )
for (i=0; i < local_size; i++)
data2[i*idxstride_y+j*vecstride_y] = first_index+i+1 + 100*j;
hypre_ParVectorSetConstantValues(y,1.0);
hypre_printf(" initialized vectors, first_index=%i\n", first_index);
hypre_ParVectorPrint(x, "vectorx");
hypre_ParVectorPrint(y, "vectory");
hypre_MatvecCommPkgCreate(par_matrix);
hypre_ParCSRMatrixMatvec ( 1.0, par_matrix, x, 1.0, y);
hypre_printf(" did matvec\n");
hypre_ParVectorPrint(y, "result");
ierr = hypre_ParCSRMatrixMatvecT ( 1.0, par_matrix, y2, 1.0, x2);
hypre_printf(" did matvecT %d\n", ierr);
hypre_ParVectorPrint(x2, "transp");
hypre_ParCSRMatrixDestroy(par_matrix);
hypre_ParVectorDestroy(x);
hypre_ParVectorDestroy(x2);
hypre_ParVectorDestroy(y);
hypre_ParVectorDestroy(y2);
if (my_id == 0) hypre_CSRMatrixDestroy(matrix);
if (matrix1) hypre_CSRMatrixDestroy(matrix1);
/* Finalize MPI */
hypre_MPI_Finalize();
return 0;
}
HYPRE_Int main(HYPRE_Int argc, char *argv[])
{
HYPRE_Int mype, npes;
HYPRE_Int symmetric;
HYPRE_Int num_runs;
Matrix *A;
ParaSails *ps;
FILE *file;
HYPRE_Int n, beg_row, end_row;
HYPRE_Real time0, time1;
HYPRE_Real setup_time, solve_time;
HYPRE_Real max_setup_time, max_solve_time;
HYPRE_Real cost;
HYPRE_Real *x, *b;
HYPRE_Int i, niter;
HYPRE_Real thresh;
HYPRE_Real threshg;
HYPRE_Int nlevels;
HYPRE_Real filter;
HYPRE_Real loadbal;
hypre_MPI_Init(&argc, &argv);
hypre_MPI_Comm_rank(hypre_MPI_COMM_WORLD, &mype);
hypre_MPI_Comm_size(hypre_MPI_COMM_WORLD, &npes);
/* Read number of rows in matrix */
symmetric = atoi(argv[1]);
num_runs = atoi(argv[2]);
file = fopen(argv[3], "r");
assert(file != NULL);
#ifdef EMSOLVE
hypre_fscanf(file, "%*d %d\n", &n);
#else
hypre_fscanf(file, "%d\n", &n);
#endif
fclose(file);
assert(n >= npes);
beg_row = (HYPRE_Int) ((HYPRE_Real)(mype*n) / npes) + 1; /* assumes 1-based */
end_row = (HYPRE_Int) ((HYPRE_Real)((mype+1)* n) / npes);
if (mype == 0)
assert(beg_row == 1);
if (mype == npes-1)
assert(end_row == n);
#ifdef EMSOLVE
beg_row--;
end_row--;
#endif
x = (HYPRE_Real *) malloc((end_row-beg_row+1) * sizeof(HYPRE_Real));
b = (HYPRE_Real *) malloc((end_row-beg_row+1) * sizeof(HYPRE_Real));
A = MatrixCreate(hypre_MPI_COMM_WORLD, beg_row, end_row);
MatrixRead(A, argv[3]);
if (mype == 0)
hypre_printf("%s\n", argv[3]);
/* MatrixPrint(A, "A"); */
/* Right-hand side */
if (argc > 4)
{
RhsRead(b, A, argv[4]);
if (mype == 0)
hypre_printf("Using rhs from %s\n", argv[4]);
}
else
{
for (i=0; i<end_row-beg_row+1; i++)
b[i] = (HYPRE_Real) (2*rand()) / (HYPRE_Real) RAND_MAX - 1.0;
}
while (num_runs && num_runs >= -1)
{
/* Initial guess */
for (i=0; i<end_row-beg_row+1; i++)
x[i] = 0.0;
if (num_runs == -1)
{
thresh = 0.0;
nlevels = 0;
filter = 0.0;
loadbal = 0.0;
}
else
{
if (mype == 0)
{
#if PARASAILS_EXT_PATTERN
hypre_printf("Enter parameters threshg, thresh, nlevels, "
"filter, beta:\n");
fflush(stdout);
hypre_scanf("%lf %lf %d %lf %lf", &threshg, &thresh, &nlevels,
&filter, &loadbal);
#else
hypre_printf("Enter parameters thresh, nlevels, "
"filter, beta:\n");
fflush(stdout);
hypre_scanf("%lf %d %lf %lf", &thresh, &nlevels,
&filter, &loadbal);
#endif
}
hypre_MPI_Bcast(&threshg, 1, hypre_MPI_DOUBLE, 0, hypre_MPI_COMM_WORLD);
hypre_MPI_Bcast(&thresh, 1, hypre_MPI_DOUBLE, 0, hypre_MPI_COMM_WORLD);
hypre_MPI_Bcast(&nlevels, 1, HYPRE_MPI_INT, 0, hypre_MPI_COMM_WORLD);
hypre_MPI_Bcast(&filter, 1, hypre_MPI_DOUBLE, 0, hypre_MPI_COMM_WORLD);
hypre_MPI_Bcast(&loadbal, 1, hypre_MPI_DOUBLE, 0, hypre_MPI_COMM_WORLD);
if (nlevels < 0)
break;
}
/**************
* Setup phase
**************/
hypre_MPI_Barrier(hypre_MPI_COMM_WORLD);
time0 = hypre_MPI_Wtime();
ps = ParaSailsCreate(hypre_MPI_COMM_WORLD, beg_row, end_row, symmetric);
ps->loadbal_beta = loadbal;
#if PARASAILS_EXT_PATTERN
ParaSailsSetupPatternExt(ps, A, threshg, thresh, nlevels);
#else
ParaSailsSetupPattern(ps, A, thresh, nlevels);
#endif
time1 = hypre_MPI_Wtime();
setup_time = time1-time0;
cost = ParaSailsStatsPattern(ps, A);
if (cost > 5.e11)
{
hypre_printf("Aborting setup and solve due to high cost.\n");
goto cleanup;
}
hypre_MPI_Barrier(hypre_MPI_COMM_WORLD);
time0 = hypre_MPI_Wtime();
err = ParaSailsSetupValues(ps, A, filter);
if (err != 0)
{
hypre_printf("ParaSailsSetupValues returned error.\n");
goto cleanup;
}
time1 = hypre_MPI_Wtime();
setup_time += (time1-time0);
ParaSailsStatsValues(ps, A);
if (!strncmp(argv[3], "testpsmat", 8))
MatrixPrint(ps->M, "M");
#if 0
if (mype == 0)
hypre_printf("SETTING UP VALUES AGAIN WITH FILTERED PATTERN\n");
ps->loadbal_beta = 0;
ParaSailsSetupValues(ps, A, 0.0);
#endif
/*****************
* Solution phase
*****************/
niter = 3000;
if (MatrixNnz(ps->M) == n) /* if diagonal preconditioner */
niter = 5000;
hypre_MPI_Barrier(hypre_MPI_COMM_WORLD);
time0 = hypre_MPI_Wtime();
if (symmetric == 1)
PCG_ParaSails(A, ps, b, x, 1.e-8, niter);
else
FGMRES_ParaSails(A, ps, b, x, 50, 1.e-8, niter);
time1 = hypre_MPI_Wtime();
solve_time = time1-time0;
hypre_MPI_Reduce(&setup_time, &max_setup_time, 1, hypre_MPI_DOUBLE, hypre_MPI_MAX, 0,
hypre_MPI_COMM_WORLD);
hypre_MPI_Reduce(&solve_time, &max_solve_time, 1, hypre_MPI_DOUBLE, hypre_MPI_MAX, 0,
hypre_MPI_COMM_WORLD);
if (mype == 0)
{
hypre_printf("**********************************************\n");
hypre_printf("*** Setup Solve Total\n");
hypre_printf("III %8.1f %8.1f %8.1f\n", max_setup_time, max_solve_time,
max_setup_time+max_solve_time);
hypre_printf("**********************************************\n");
}
cleanup:
ParaSailsDestroy(ps);
num_runs--;
}
free(x);
free(b);
MatrixDestroy(A);
hypre_MPI_Finalize();
return 0;
}
