void lis_matrix_assemble_f(LIS_MATRIX_F *A, LIS_INT *ierr)
{
LIS_MATRIX AA;
LIS_DEBUG_FUNC_IN;
AA = (LIS_MATRIX)LIS_V2P(A);
*ierr = lis_matrix_assemble(AA);
*A = LIS_P2V(AA);
LIS_DEBUG_FUNC_OUT;
return;
}
bool finalizeMatrixAssembly(LisMatrix &mat)
{
LIS_MATRIX &A = mat.getRawMatrix();
if (!mat.isAssembled()) {
int ierr = lis_matrix_set_type(A, static_cast<int>(mat.getMatrixType()));
checkLisError(ierr);
ierr = lis_matrix_assemble(A);
checkLisError(ierr);
mat._is_assembled = true;
}
return true;
}
LisMatrix::LisMatrix(std::size_t n_rows, int nnz, IndexType *row_ptr,
IndexType *col_idx, double *data)
: _n_rows(n_rows),
_mat_type(MatrixType::CRS),
_is_assembled(false),
_use_external_arrays(true)
{
int ierr = lis_matrix_create(0, &_AA);
checkLisError(ierr);
ierr = lis_matrix_set_size(_AA, 0, n_rows);
checkLisError(ierr);
ierr = lis_matrix_set_csr(nnz, row_ptr, col_idx, data, _AA);
checkLisError(ierr);
ierr = lis_matrix_assemble(_AA);
checkLisError(ierr);
_is_assembled = true;
lis_matrix_get_range(_AA, &_is, &_ie);
ierr = lis_vector_duplicate(_AA, &_diag);
checkLisError(ierr);
}
LIS_INT lis_esolver_get_evectors(LIS_ESOLVER esolver, LIS_MATRIX M)
{
LIS_INT i,ii,j,jj,n,gn,is,ie,js;
LIS_INT ss,lis_esolver_evector_size;
LIS_DEBUG_FUNC_IN;
if ( esolver->options[LIS_EOPTIONS_ESOLVER] != LIS_ESOLVER_SI && esolver->options[LIS_EOPTIONS_ESOLVER] != LIS_ESOLVER_LI && esolver->options[LIS_EOPTIONS_ESOLVER] != LIS_ESOLVER_AI )
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_ESOLVER is %d (Set Subspace, Lanczos, or Arnoldi)\n", esolver->options[LIS_EOPTIONS_ESOLVER]);
return LIS_ERR_ILL_ARG;
}
ss = esolver->options[LIS_EOPTIONS_SUBSPACE];
lis_esolver_evector_size = esolver->evector[0]->gn;
lis_matrix_set_size(M,0,lis_esolver_evector_size);
lis_matrix_get_size(M,&n,&gn);
lis_matrix_get_range(M,&is,&ie);
js=0;
if( esolver->evector[0]->origin )
{
is++;
js++;
}
for(j=0;j<ss;j++)
{
for(i=0;i<n;i++)
{
ii=i+is;
jj=j+js;
lis_matrix_set_value(LIS_INS_VALUE,ii,jj,esolver->evector[j]->value[i],M);
}
}
lis_matrix_set_type(M,LIS_MATRIX_CSR);
lis_matrix_assemble(M);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_MATRIX A0,A;
LIS_VECTOR x,b,u;
LIS_SOLVER solver;
LIS_INT m,n,nn,nnz;
LIS_INT i,j,ii,jj,ctr;
LIS_INT is,ie;
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nsol;
LIS_INT err,iter,mtype,iter_double,iter_quad;
double time,itime,ptime,p_c_time,p_i_time;
LIS_REAL resid;
char solvername[128];
LIS_INT *ptr,*index;
LIS_SCALAR *value;
LIS_DEBUG_FUNC_IN;
lis_initialize(&argc, &argv);
#ifdef USE_MPI
MPI_Comm_size(MPI_COMM_WORLD,&int_nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&int_my_rank);
nprocs = int_nprocs;
my_rank = int_my_rank;
#else
nprocs = 1;
my_rank = 0;
#endif
if( argc < 6 )
{
if( my_rank==0 )
{
printf("Usage: %s m n matrix_type solution_filename rhistory_filename [options]\n", argv[0]);
}
CHKERR(1);
}
m = atoi(argv[1]);
n = atoi(argv[2]);
mtype = atoi(argv[3]);
if( m<=0 || n<=0 )
{
#ifdef _LONGLONG
if( my_rank==0 ) printf("m=%lld <=0 or n=%lld <=0\n",m,n);
#else
if( my_rank==0 ) printf("m=%d <=0 or n=%d <=0\n",m,n);
#endif
CHKERR(1);
}
if( my_rank==0 )
{
printf("\n");
#ifdef _LONGLONG
printf("number of processes = %lld\n",nprocs);
#else
printf("number of processes = %d\n",nprocs);
#endif
}
#ifdef _OPENMP
if( my_rank==0 )
{
#ifdef _LONGLONG
printf("max number of threads = %lld\n",omp_get_num_procs());
printf("number of threads = %lld\n",omp_get_max_threads());
#else
printf("max number of threads = %d\n",omp_get_num_procs());
printf("number of threads = %d\n",omp_get_max_threads());
#endif
}
#endif
/* create matrix and vectors */
nn = m*n;
err = lis_matrix_create(LIS_COMM_WORLD,&A);
err = lis_matrix_set_size(A,0,nn);
CHKERR(err);
ptr = (LIS_INT *)malloc((A->n+1)*sizeof(LIS_INT));
if( ptr==NULL ) CHKERR(1);
index = (LIS_INT *)malloc(5*A->n*sizeof(LIS_INT));
if( index==NULL ) CHKERR(1);
value = (LIS_SCALAR *)malloc(5*A->n*sizeof(LIS_SCALAR));
if( value==NULL ) CHKERR(1);
lis_matrix_get_range(A,&is,&ie);
ctr = 0;
for(ii=is;ii<ie;ii++)
{
i = ii/m;
j = ii - i*m;
if( i>0 ) { jj = ii - m; index[ctr] = jj; value[ctr++] = -1.0;}
if( i<n-1 ) { jj = ii + m; index[ctr] = jj; value[ctr++] = -1.0;}
if( j>0 ) { jj = ii - 1; index[ctr] = jj; value[ctr++] = -1.0;}
if( j<m-1 ) { jj = ii + 1; index[ctr] = jj; value[ctr++] = -1.0;}
index[ctr] = ii; value[ctr++] = 4.0;
ptr[ii-is+1] = ctr;
}
ptr[0] = 0;
err = lis_matrix_set_csr(ptr[ie-is],ptr,index,value,A);
CHKERR(err);
err = lis_matrix_assemble(A);
CHKERR(err);
nnz = A->nnz;
#ifdef USE_MPI
MPI_Allreduce(&nnz,&i,1,LIS_MPI_INT,MPI_SUM,A->comm);
nnz = i;
#endif
#ifdef _LONGLONG
if( my_rank==0 ) printf("matrix size = %lld x %lld (%lld nonzero entries)\n\n",nn,nn,nnz);
#else
if( my_rank==0 ) printf("matrix size = %d x %d (%d nonzero entries)\n\n",nn,nn,nnz);
#endif
err = lis_matrix_duplicate(A,&A0);
CHKERR(err);
lis_matrix_set_type(A0,mtype);
err = lis_matrix_convert(A,A0);
CHKERR(err);
lis_matrix_destroy(A);
A = A0;
err = lis_vector_duplicate(A,&u);
CHKERR(err);
err = lis_vector_duplicate(A,&b);
CHKERR(err);
err = lis_vector_duplicate(A,&x);
CHKERR(err);
err = lis_vector_set_all(1.0,u);
lis_matvec(A,u,b);
err = lis_solver_create(&solver); CHKERR(err);
lis_solver_set_option("-print mem",solver);
lis_solver_set_optionC(solver);
err = lis_solve(A,b,x,solver);
CHKERR(err);
lis_solver_get_iterex(solver,&iter,&iter_double,&iter_quad);
lis_solver_get_timeex(solver,&time,&itime,&ptime,&p_c_time,&p_i_time);
lis_solver_get_residualnorm(solver,&resid);
lis_solver_get_solver(solver,&nsol);
lis_solver_get_solvername(nsol,solvername);
if( my_rank==0 )
{
#ifdef _LONGLONG
#ifdef _LONG__DOUBLE
printf("%s: number of iterations = %lld \n",solvername, iter);
#else
printf("%s: number of iterations = %lld (double = %lld, quad = %lld)\n",solvername,iter, iter_double, iter_quad);
#endif
#else
#ifdef _LONG__DOUBLE
printf("%s: number of iterations = %d \n",solvername, iter);
#else
printf("%s: number of iterations = %d (double = %d, quad = %d)\n",solvername,iter, iter_double, iter_quad);
#endif
#endif
printf("%s: elapsed time = %e sec.\n",solvername,time);
printf("%s: preconditioner = %e sec.\n",solvername, ptime);
printf("%s: matrix creation = %e sec.\n",solvername, p_c_time);
printf("%s: linear solver = %e sec.\n",solvername, itime);
#ifdef _LONG__DOUBLE
printf("%s: relative residual = %Le\n\n",solvername,resid);
#else
printf("%s: relative residual = %e\n\n",solvername,resid);
#endif
}
/* write solution */
lis_output_vector(x,LIS_FMT_MM,argv[4]);
/* write residual history */
lis_solver_output_rhistory(solver, argv[5]);
lis_solver_destroy(solver);
lis_matrix_destroy(A);
lis_vector_destroy(b);
lis_vector_destroy(x);
lis_vector_destroy(u);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
LIS_INT lis_matrix_copyDLU_csr(LIS_MATRIX Ain, LIS_MATRIX_DIAG *D, LIS_MATRIX *L, LIS_MATRIX *U)
{
LIS_INT err;
LIS_INT i,n,np,lnnz,unnz;
LIS_INT *lptr,*lindex;
LIS_INT *uptr,*uindex;
LIS_SCALAR *lvalue,*uvalue,*diag;
LIS_DEBUG_FUNC_IN;
*D = NULL;
*L = NULL;
*U = NULL;
err = lis_matrix_check(Ain,LIS_MATRIX_CHECK_ALL);
if( err ) return err;
n = Ain->n;
np = Ain->np;
err = lis_matrix_duplicate(Ain,L);
if( err )
{
return err;
}
err = lis_matrix_duplicate(Ain,U);
if( err )
{
lis_matrix_destroy(*L);
return err;
}
err = lis_matrix_diag_duplicateM(Ain,D);
if( err )
{
lis_matrix_destroy(*L);
lis_matrix_destroy(*U);
return err;
}
lis_free((*D)->value);
if( Ain->is_splited )
{
}
lnnz = Ain->L->nnz;
unnz = Ain->U->nnz;
lptr = NULL;
lindex = NULL;
uptr = NULL;
uindex = NULL;
diag = NULL;
err = lis_matrix_malloc_csr(n,lnnz,&lptr,&lindex,&lvalue);
if( err )
{
return err;
}
err = lis_matrix_malloc_csr(n,unnz,&uptr,&uindex,&uvalue);
if( err )
{
lis_free2(7,diag,uptr,lptr,uindex,lindex,uvalue,lvalue);
return err;
}
diag = (LIS_SCALAR *)lis_malloc(np*sizeof(LIS_SCALAR),"lis_matrix_copyDLU_csr::diag");
if( diag==NULL )
{
lis_free2(7,diag,uptr,lptr,uindex,lindex,uvalue,lvalue);
return err;
}
#ifdef _OPENMP
#pragma omp parallel for private(i)
#endif
for(i=0;i<n;i++)
{
diag[i] = Ain->D->value[i];
}
lis_matrix_elements_copy_csr(n,Ain->L->ptr,Ain->L->index,Ain->L->value,lptr,lindex,lvalue);
lis_matrix_elements_copy_csr(n,Ain->U->ptr,Ain->U->index,Ain->U->value,uptr,uindex,uvalue);
(*D)->value = diag;
err = lis_matrix_set_csr(lnnz,lptr,lindex,lvalue,*L);
if( err )
{
lis_free2(7,diag,uptr,lptr,uindex,lindex,uvalue,lvalue);
return err;
}
err = lis_matrix_set_csr(unnz,uptr,uindex,uvalue,*U);
if( err )
{
lis_free2(7,diag,uptr,lptr,uindex,lindex,uvalue,lvalue);
return err;
}
err = lis_matrix_assemble(*L);
if( err )
{
return err;
}
err = lis_matrix_assemble(*U);
if( err )
{
return err;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matrix_copy_csr(LIS_MATRIX Ain, LIS_MATRIX Aout)
{
LIS_INT err;
LIS_INT i,n,nnz,lnnz,unnz;
LIS_INT *ptr,*index;
LIS_INT *lptr,*lindex;
LIS_INT *uptr,*uindex;
LIS_SCALAR *value,*lvalue,*uvalue,*diag;
LIS_DEBUG_FUNC_IN;
n = Ain->n;
if( Ain->is_splited )
{
lnnz = Ain->L->nnz;
unnz = Ain->U->nnz;
lptr = NULL;
lindex = NULL;
uptr = NULL;
uindex = NULL;
diag = NULL;
err = lis_matrix_malloc_csr(n,lnnz,&lptr,&lindex,&lvalue);
if( err )
{
return err;
}
err = lis_matrix_malloc_csr(n,unnz,&uptr,&uindex,&uvalue);
if( err )
{
lis_free2(7,diag,uptr,lptr,uindex,lindex,uvalue,lvalue);
return err;
}
diag = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_matrix_copy_csr::diag");
if( diag==NULL )
{
lis_free2(7,diag,uptr,lptr,uindex,lindex,uvalue,lvalue);
return err;
}
#ifdef _OPENMP
#pragma omp parallel for private(i)
#endif
for(i=0;i<n;i++)
{
diag[i] = Ain->D->value[i];
}
lis_matrix_elements_copy_csr(n,Ain->L->ptr,Ain->L->index,Ain->L->value,lptr,lindex,lvalue);
lis_matrix_elements_copy_csr(n,Ain->U->ptr,Ain->U->index,Ain->U->value,uptr,uindex,uvalue);
err = lis_matrix_setDLU_csr(lnnz,unnz,diag,lptr,lindex,lvalue,uptr,uindex,uvalue,Aout);
if( err )
{
lis_free2(7,diag,uptr,lptr,uindex,lindex,uvalue,lvalue);
return err;
}
}
if( !Ain->is_splited || (Ain->is_splited && Ain->is_save) )
{
ptr = NULL;
index = NULL;
value = NULL;
nnz = Ain->nnz;
err = lis_matrix_malloc_csr(n,nnz,&ptr,&index,&value);
if( err )
{
return err;
}
lis_matrix_elements_copy_csr(n,Ain->ptr,Ain->index,Ain->value,ptr,index,value);
err = lis_matrix_set_csr(nnz,ptr,index,value,Aout);
if( err )
{
lis_free2(3,ptr,index,value);
return err;
}
}
if( Ain->matrix_type==LIS_MATRIX_CSC )
{
Aout->matrix_type = LIS_MATRIX_CSC;
Aout->status = -LIS_MATRIX_CSC;
err = lis_matrix_assemble(Aout);
}
else
{
err = lis_matrix_assemble(Aout);
}
if( err )
{
lis_matrix_storage_destroy(Aout);
return err;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_MATRIX A;
LIS_VECTOR x,b,u;
LIS_SOLVER solver;
LIS_INT k,n,gn,ii,jj;
LIS_INT is,ie;
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nsol;
LIS_INT err,iter,iter_double,iter_quad;
double time,itime,ptime,p_c_time,p_i_time;
LIS_REAL resid;
char solvername[128];
LIS_INT *ptr,*index;
LIS_SCALAR *value,gamma;
LIS_DEBUG_FUNC_IN;
lis_initialize(&argc, &argv);
#ifdef USE_MPI
MPI_Comm_size(MPI_COMM_WORLD,&int_nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&int_my_rank);
nprocs = int_nprocs;
my_rank = int_my_rank;
#else
nprocs = 1;
my_rank = 0;
#endif
if( argc < 3 )
{
if( my_rank==0 )
{
printf("Usage: %s n gamma [options]\n", argv[0]);
}
CHKERR(1);
}
gn = atoi(argv[1]);
gamma = atof(argv[2]);
if( gn<=0 )
{
#ifdef _LONGLONG
if( my_rank==0 ) printf("n=%lld <=0 \n",gn);
#else
if( my_rank==0 ) printf("n=%d <=0 \n",gn);
#endif
CHKERR(1);
}
if( my_rank==0 )
{
printf("\n");
#ifdef _LONGLONG
printf("number of processes = %lld\n",nprocs);
#else
printf("number of processes = %d\n",nprocs);
#endif
}
#ifdef _OPENMP
if( my_rank==0 )
{
#ifdef _LONGLONG
printf("max number of threads = %lld\n",omp_get_num_procs());
printf("number of threads = %lld\n",omp_get_max_threads());
#else
printf("max number of threads = %d\n",omp_get_num_procs());
printf("number of threads = %d\n",omp_get_max_threads());
#endif
}
#endif
if(my_rank==0)
{
#ifdef _LONGLONG
printf("n = %lld, gamma = %f\n\n",gn,gamma);
#else
printf("n = %d, gamma = %f\n\n",gn,gamma);
#endif
}
/* create matrix and vectors */
err = lis_matrix_create(LIS_COMM_WORLD,&A); CHKERR(err);
err = lis_matrix_set_size(A,0,gn); CHKERR(err);
err = lis_matrix_get_size(A,&n,&gn); CHKERR(err);
err = lis_matrix_malloc_csr(n,3*n,&ptr,&index,&value); CHKERR(err);
err = lis_matrix_get_range(A,&is,&ie); CHKERR(err);
k = 0;
ptr[0] = 0;
for(ii=is;ii<ie;ii++)
{
if( ii>1 ) { jj = ii - 2; index[k] = jj; value[k++] = gamma;}
if( ii<gn-1 ) { jj = ii + 1; index[k] = jj; value[k++] = 1.0;}
index[k] = ii; value[k++] = 2.0;
ptr[ii-is+1] = k;
}
err = lis_matrix_set_csr(ptr[ie-is],ptr,index,value,A); CHKERR(err);
err = lis_matrix_assemble(A); CHKERR(err);
err = lis_vector_duplicate(A,&u); CHKERR(err);
err = lis_vector_duplicate(u,&b); CHKERR(err);
err = lis_vector_duplicate(u,&x); CHKERR(err);
err = lis_vector_set_all(1.0,u);
lis_matvec(A,u,b);
err = lis_solver_create(&solver); CHKERR(err);
lis_solver_set_option("-print mem",solver);
lis_solver_set_optionC(solver);
err = lis_solve(A,b,x,solver); CHKERR(err);
lis_solver_get_iterex(solver,&iter,&iter_double,&iter_quad);
lis_solver_get_timeex(solver,&time,&itime,&ptime,&p_c_time,&p_i_time);
lis_solver_get_residualnorm(solver,&resid);
lis_solver_get_solver(solver,&nsol);
lis_solver_get_solvername(nsol,solvername);
if( my_rank==0 )
{
#ifdef _LONGLONG
#ifdef _LONG__DOUBLE
printf("%s: number of iterations = %lld \n",solvername, iter);
#else
printf("%s: number of iterations = %lld (double = %lld, quad = %lld)\n",solvername,iter, iter_double, iter_quad);
#endif
#else
#ifdef _LONG__DOUBLE
printf("%s: number of iterations = %d \n",solvername, iter);
#else
printf("%s: number of iterations = %d (double = %d, quad = %d)\n",solvername,iter, iter_double, iter_quad);
#endif
#endif
printf("%s: elapsed time = %e sec.\n",solvername,time);
printf("%s: preconditioner = %e sec.\n",solvername, ptime);
printf("%s: matrix creation = %e sec.\n",solvername, p_c_time);
printf("%s: linear solver = %e sec.\n",solvername, itime);
#ifdef _LONG__DOUBLE
printf("%s: relative residual = %Le\n\n",solvername,resid);
#else
printf("%s: relative residual = %e\n\n",solvername,resid);
#endif
}
lis_solver_destroy(solver);
lis_matrix_destroy(A);
lis_vector_destroy(b);
lis_vector_destroy(x);
lis_vector_destroy(u);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_MATRIX A,A0;
LIS_VECTOR b,x;
LIS_SCALAR *value;
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nthreads,maxthreads;
LIS_INT gn,nnz,np;
LIS_INT i,j,k,si,sj,sk,ii,jj,ctr;
LIS_INT l,m,n,nn;
LIS_INT is,ie;
LIS_INT err,iter,matrix_type,storage,ss,se;
LIS_INT *ptr,*index;
double time,time2,nnzs,nnzap,nnzt;
LIS_SCALAR val;
double commtime,comptime,flops;
LIS_DEBUG_FUNC_IN;
lis_initialize(&argc, &argv);
#ifdef USE_MPI
MPI_Comm_size(MPI_COMM_WORLD,&int_nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&int_my_rank);
nprocs = int_nprocs;
my_rank = int_my_rank;
#else
nprocs = 1;
my_rank = 0;
#endif
if( argc < 5 )
{
if( my_rank==0 )
{
printf("Usage: %s l m n iter [matrix_type]\n", argv[0]);
}
CHKERR(1);
}
l = atoi(argv[1]);
m = atoi(argv[2]);
n = atoi(argv[3]);
iter = atoi(argv[4]);
if (argv[5] == NULL) {
storage = 0;
}
else {
storage = atoi(argv[5]);
}
if( iter<=0 )
{
#ifdef _LONG__LONG
if( my_rank==0 ) printf("iter=%lld <= 0\n",iter);
#else
if( my_rank==0 ) printf("iter=%d <= 0\n",iter);
#endif
CHKERR(1);
}
if( l<=0 || m<=0 || n<=0 )
{
#ifdef _LONG__LONG
if( my_rank==0 ) printf("l=%lld <=0, m=%lld <=0 or n=%lld <=0\n",l,m,n);
#else
if( my_rank==0 ) printf("l=%d <=0, m=%d <=0 or n=%d <=0\n",l,m,n);
#endif
CHKERR(1);
}
if( storage<0 || storage>11 )
{
#ifdef _LONG__LONG
if( my_rank==0 ) printf("matrix_type=%lld < 0 or matrix_type=%lld > 11\n",storage,storage);
#else
if( my_rank==0 ) printf("matrix_type=%d < 0 or matrix_type=%d > 11\n",storage,storage);
#endif
CHKERR(1);
}
if( my_rank==0 )
{
printf("\n");
#ifdef _LONG__LONG
printf("number of processes = %lld\n",nprocs);
#else
printf("number of processes = %d\n",nprocs);
#endif
}
#ifdef _OPENMP
nthreads = omp_get_num_procs();
maxthreads = omp_get_max_threads();
if( my_rank==0 )
{
#ifdef _LONG__LONG
printf("max number of threads = %lld\n", nthreads);
printf("number of threads = %lld\n", maxthreads);
#else
printf("max number of threads = %d\n", nthreads);
printf("number of threads = %d\n", maxthreads);
#endif
}
#else
nthreads = 1;
maxthreads = 1;
#endif
/* create matrix and vectors */
nn = l*m*n;
err = lis_matrix_create(LIS_COMM_WORLD,&A0);
err = lis_matrix_set_size(A0,0,nn);
CHKERR(err);
ptr = (LIS_INT *)malloc((A0->n+1)*sizeof(LIS_INT));
if( ptr==NULL ) CHKERR(1);
index = (LIS_INT *)malloc(27*A0->n*sizeof(LIS_INT));
if( index==NULL ) CHKERR(1);
value = (LIS_SCALAR *)malloc(27*A0->n*sizeof(LIS_SCALAR));
if( value==NULL ) CHKERR(1);
lis_matrix_get_range(A0,&is,&ie);
ctr = 0;
for(ii=is;ii<ie;ii++)
{
i = ii/(m*n);
j = (ii - i*m*n)/n;
k = ii - i*m*n - j*n;
for(si=-1;si<=1;si++) {
if( i+si>-1 && i+si<l ) {
for(sj=-1;sj<=1;sj++) {
if( j+sj>-1 && j+sj<m ) {
for(sk=-1;sk<=1;sk++) {
if( k+sk>-1 && k+sk<n ) {
jj = ii + si*m*n + sj*n + sk;
index[ctr] = jj;
if( jj==ii ) { value[ctr++] = 26.0;}
else { value[ctr++] = -1.0;}
}
}
}
}
}
}
ptr[ii-is+1] = ctr;
}
ptr[0] = 0;
err = lis_matrix_set_csr(ptr[ie-is],ptr,index,value,A0);
CHKERR(err);
err = lis_matrix_assemble(A0);
CHKERR(err);
n = A0->n;
gn = A0->gn;
nnz = A0->nnz;
np = A0->np-n;
#ifdef USE_MPI
MPI_Allreduce(&nnz,&i,1,LIS_MPI_INT,MPI_SUM,A0->comm);
nnzap = (double)i / (double)nprocs;
nnzt = ((double)nnz -nnzap)*((double)nnz -nnzap);
nnz = i;
MPI_Allreduce(&nnzt,&nnzs,1,MPI_DOUBLE,MPI_SUM,A0->comm);
nnzs = (nnzs / (double)nprocs)/nnzap;
MPI_Allreduce(&np,&i,1,LIS_MPI_INT,MPI_SUM,A0->comm);
np = i;
#endif
if( my_rank==0 )
{
#ifdef _LONG__LONG
printf("matrix size = %lld x %lld (%lld nonzero entries)\n",gn,gn,nnz);
printf("number of iterations = %lld\n\n",iter);
#else
printf("matrix size = %d x %d (%d nonzero entries)\n",gn,gn,nnz);
printf("number of iterations = %d\n\n",iter);
#endif
}
err = lis_vector_duplicate(A0,&x);
if( err ) CHKERR(err);
err = lis_vector_duplicate(A0,&b);
if( err ) CHKERR(err);
lis_matrix_get_range(A0,&is,&ie);
for(i=0;i<n;i++)
{
err = lis_vector_set_value(LIS_INS_VALUE,i+is,1.0,x);
}
for(i=0;i<n;i++)
{
lis_sort_id(A0->ptr[i],A0->ptr[i+1]-1,A0->index,A0->value);
}
/*
MPI version of VBR is not implemented.
DNS is also excluded to reduce memory usage.
*/
if (storage==0)
{
ss = 1;
se = 11;
}
else
{
ss = storage;
se = storage+1;
}
for (matrix_type=ss;matrix_type<se;matrix_type++)
{
if ( nprocs>1 && matrix_type==9 ) continue;
lis_matrix_duplicate(A0,&A);
lis_matrix_set_type(A,matrix_type);
err = lis_matrix_convert(A0,A);
if( err ) CHKERR(err);
comptime = 0.0;
commtime = 0.0;
for(i=0;i<iter;i++)
{
#ifdef USE_MPI
MPI_Barrier(A->comm);
time = lis_wtime();
lis_send_recv(A->commtable,x->value);
commtime += lis_wtime() - time;
#endif
time2 = lis_wtime();
lis_matvec(A,x,b);
comptime += lis_wtime() - time2;
}
lis_vector_nrm2(b,&val);
if( my_rank==0 )
{
flops = 2.0*nnz*iter*1.0e-6 / comptime;
#ifdef USE_MPI
#ifdef _LONG__DOUBLE
#ifdef _LONG__LONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#endif
#else
#ifdef _LONG__LONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#endif
#endif
#else
#ifdef _LONG__DOUBLE
#ifdef _LONG__LONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#endif
#else
#ifdef _LONG__LONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#endif
#endif
#endif
}
lis_matrix_destroy(A);
}
lis_matrix_destroy(A0);
lis_vector_destroy(b);
lis_vector_destroy(x);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
LIS_INT lis_matrix_copy_dia(LIS_MATRIX Ain, LIS_MATRIX Aout)
{
LIS_INT err;
LIS_INT i,n,nnd,lnnd,unnd;
LIS_INT *index;
LIS_INT *lindex;
LIS_INT *uindex;
LIS_SCALAR *value,*lvalue,*uvalue,*diag;
LIS_DEBUG_FUNC_IN;
n = Ain->n;
if( Ain->is_splited )
{
lnnd = Ain->L->nnd;
unnd = Ain->U->nnd;
lindex = NULL;
uindex = NULL;
diag = NULL;
err = lis_matrix_malloc_dia(n,lnnd,&lindex,&lvalue);
if( err )
{
return err;
}
err = lis_matrix_malloc_dia(n,unnd,&uindex,&uvalue);
if( err )
{
lis_free2(5,diag,uindex,lindex,uvalue,lvalue);
return err;
}
diag = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_matrix_copy_dia::diag");
if( diag==NULL )
{
lis_free2(5,diag,uindex,lindex,uvalue,lvalue);
return err;
}
#ifdef _OPENMP
#pragma omp parallel for private(i)
#endif
for(i=0;i<n;i++)
{
diag[i] = Ain->D->value[i];
}
lis_matrix_elements_copy_dia(n,lnnd,Ain->L->index,Ain->L->value,lindex,lvalue);
lis_matrix_elements_copy_dia(n,unnd,Ain->U->index,Ain->U->value,uindex,uvalue);
err = lis_matrix_setDLU_dia(lnnd,unnd,diag,lindex,lvalue,uindex,uvalue,Aout);
if( err )
{
lis_free2(5,diag,uindex,lindex,uvalue,lvalue);
return err;
}
}
if( !Ain->is_splited || (Ain->is_splited && Ain->is_save) )
{
index = NULL;
value = NULL;
nnd = Ain->nnd;
err = lis_matrix_malloc_dia(n,nnd,&index,&value);
if( err )
{
return err;
}
lis_matrix_elements_copy_dia(n,nnd,Ain->index,Ain->value,index,value);
err = lis_matrix_set_dia(nnd,index,value,Aout);
if( err )
{
lis_free2(2,index,value);
return err;
}
}
err = lis_matrix_assemble(Aout);
if( err )
{
lis_matrix_storage_destroy(Aout);
return err;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matrix_convert_rco2csr(LIS_MATRIX Ain, LIS_MATRIX Aout)
{
LIS_INT i,j,k,n,nnz,err;
LIS_INT *ptr,*index;
LIS_SCALAR *value;
LIS_DEBUG_FUNC_IN;
ptr = NULL;
index = NULL;
value = NULL;
n = Ain->n;
nnz = 0;
#ifdef _OPENMP
#pragma omp parallel for reduction(+:nnz) private(i)
#endif
for(i=0;i<n;i++)
{
nnz += Ain->w_row[i];
}
err = lis_matrix_malloc_csr(n,nnz,&ptr,&index,&value);
if( err )
{
return err;
}
#ifdef _NUMA
#pragma omp parallel for private(i)
for(i=0;i<n+1;i++) ptr[i] = 0;
#else
ptr[0] = 0;
#endif
for(i=0;i<n;i++)
{
ptr[i+1] = ptr[i] + Ain->w_row[i];
}
#ifdef _OPENMP
#pragma omp parallel for private(i,j,k)
#endif
for(i=0;i<n;i++)
{
k = ptr[i];
for(j=0;j<Ain->w_row[i];j++)
{
index[k] = Ain->w_index[i][j];
value[k] = Ain->w_value[i][j];
k++;
}
}
err = lis_matrix_set_csr(nnz,ptr,index,value,Aout);
if( err )
{
lis_free2(3,ptr,index,value);
return err;
}
err = lis_matrix_assemble(Aout);
if( err )
{
lis_matrix_storage_destroy(Aout);
return err;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_input_hb_csr(LIS_MATRIX A, LIS_VECTOR b, LIS_VECTOR x, FILE *file)
{
char buf[BUFSIZE];
char title[128], key[128], mtx[64], dat[128];
char *p;
char MXTYPE_F,MXTYPE_S,MXTYPE_T;
char RHSTYP_F,RHSTYP_S,RHSTYP_T;
LIS_INT TOTCRD,PTRCRD,INDCRD,VALCRD,RHSCRD;
LIS_INT NROW,NCOL,NNZERO,NELTVL;
LIS_INT NRHS,NRHSIX;
LIS_INT iptr,iind,ival,irhs;
LIS_INT wptr,wind,wval,wrhs;
LIS_INT i,k,j,my_rank;
LIS_INT err;
LIS_INT n,is,ie;
LIS_INT *ptr, *index;
LIS_INT matrix_type;
LIS_SCALAR *value;
LIS_MATRIX B;
#ifdef USE_MPI
MPI_Comm_rank(A->comm,&my_rank);
#else
my_rank = 0;
#endif
matrix_type = A->matrix_type;
/* Line 1 */
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
strncpy(title, buf ,72); title[72] = '\0';
strncpy(key ,&buf[72], 8); key[8] = '\0';
printf("title: %s\n",title);
printf("key : %s\n",key);
/* Line 2 */
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
#ifdef _LONGLONG
if( sscanf(buf, "%14lld%14lld%14lld%14lld%14lld", &TOTCRD, &PTRCRD, &INDCRD, &VALCRD, &RHSCRD) != 5 )
#else
if( sscanf(buf, "%14d%14d%14d%14d%14d", &TOTCRD, &PTRCRD, &INDCRD, &VALCRD, &RHSCRD) != 5 )
#endif
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
#ifdef _LONGLONG
printf("%14lld%14lld%14lld%14lld%14lld\n",TOTCRD, PTRCRD, INDCRD, VALCRD, RHSCRD);
#else
printf("%14d%14d%14d%14d%14d\n",TOTCRD, PTRCRD, INDCRD, VALCRD, RHSCRD);
#endif
/* Line 3 */
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
#ifdef _LONGLONG
if( sscanf(buf, "%s %lld %lld %lld %lld", mtx, &NROW, &NCOL, &NNZERO, &NELTVL) != 5 )
#else
if( sscanf(buf, "%s %d %d %d %d", mtx, &NROW, &NCOL, &NNZERO, &NELTVL) != 5 )
#endif
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
for(p=mtx;*p!='\0';p++) *p = (char)tolower(*p);
MXTYPE_F = mtx[0];
MXTYPE_S = mtx[1];
MXTYPE_T = mtx[2];
if( mtx[0]!='r' )
{
LIS_SETERR(LIS_ERR_FILE_IO,"Not real\n");
return LIS_ERR_FILE_IO;
}
/*
if( mtx[1]!='u' )
{
LIS_SETERR(LIS_ERR_FILE_IO,"Not unsymmetric\n");
return LIS_ERR_FILE_IO;
}
*/
if( mtx[2]!='a' )
{
LIS_SETERR(LIS_ERR_FILE_IO,"Not assembled\n");
return LIS_ERR_FILE_IO;
}
if( NROW!=NCOL )
{
LIS_SETERR(LIS_ERR_FILE_IO,"matrix is not square\n");
return LIS_ERR_FILE_IO;
}
#ifdef _LONGLONG
printf("%c%c%c %lld %lld %lld %lld\n",MXTYPE_F, MXTYPE_S, MXTYPE_T, NROW, NCOL, NNZERO, NELTVL);
#else
printf("%c%c%c %d %d %d %d\n",MXTYPE_F, MXTYPE_S, MXTYPE_T, NROW, NCOL, NNZERO, NELTVL);
#endif
/* Line 4 */
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
lis_input_hb_get_fmt( buf ,16,&iptr,&wptr);
lis_input_hb_get_fmt(&buf[16],16,&iind,&wind);
lis_input_hb_get_fmt(&buf[32],20,&ival,&wval);
lis_input_hb_get_fmt(&buf[52],20,&irhs,&wrhs);
#ifdef _LONGLONG
printf("%lld %lld %lld %lld\n",iptr,iind,ival,irhs);
printf("%lld %lld %lld %lld\n",wptr,wind,wval,wrhs);
#else
printf("%d %d %d %d\n",iptr,iind,ival,irhs);
printf("%d %d %d %d\n",wptr,wind,wval,wrhs);
#endif
/* Line 5 */
if( RHSCRD!=0 )
{
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
#ifdef _LONGLONG
sscanf(buf, "%s %lld %lld", mtx, &NRHS, &NRHSIX);
#else
sscanf(buf, "%s %d %d", mtx, &NRHS, &NRHSIX);
#endif
/*
#ifdef _LONGLONG
if( sscanf(buf, "%s %lld %lld", mtx, &NRHS, &NRHSIX) != 3 )
#else
if( sscanf(buf, "%s %d %d", mtx, &NRHS, &NRHSIX) != 3 )
#endif
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
*/
for(p=mtx;*p!='\0';p++) *p = (char)tolower(*p);
RHSTYP_F = mtx[0];
RHSTYP_S = mtx[1];
RHSTYP_T = mtx[2];
#ifdef _LONGLONG
printf("%c%c%c %lld %lld\n",RHSTYP_F, RHSTYP_S, RHSTYP_T, NRHS, NRHSIX);
#else
printf("%c%c%c %d %d\n",RHSTYP_F, RHSTYP_S, RHSTYP_T, NRHS, NRHSIX);
#endif
}
err = lis_matrix_set_size(A,0,NROW);
if( err )
{
return err;
}
n = A->n;
lis_matrix_get_range(A,&is,&ie);
err = lis_matrix_malloc_csr(n,NNZERO,&ptr,&index,&value);
if( err )
{
return err;
}
/* read data */
k = 0;
for( i=0; i<PTRCRD; i++ )
{
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
p = buf;
for(j=0;j<iptr&&k<n+1;j++)
{
strncpy(dat, p, wptr); dat[wptr] = '\0';
ptr[k] = atoi(dat) - 1;
p += wptr;
k++;
}
}
k = 0;
for( i=0; i<INDCRD; i++ )
{
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
p = buf;
for(j=0;j<iind&&k<NNZERO;j++)
{
strncpy(dat, p, wind); dat[wind] = '\0';
index[k] = atoi(dat) - 1;
p += wind;
k++;
}
}
k = 0;
for( i=0; i<VALCRD; i++ )
{
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
p = buf;
for(j=0;j<ival&&k<NNZERO;j++)
{
strncpy(dat, p, wval); dat[wval] = '\0';
value[k] = atof(dat);
p += wval;
k++;
}
}
if( RHSCRD>0 )
{
/*
k = 0;
for( i=0; i<RHSCRD; i++ )
{
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
return LIS_ERR_FILE_IO;
}
p = buf;
for(j=0;j<ival&&k<NNZERO;j++)
{
strncpy(dat, p, wval); dat[wval] = '\0';
value[k] = atof(dat);
p += wval;
printf("%e ",value[k]);
k++;
}
printf("\n");
}
*/
}
err = lis_matrix_set_csc(NNZERO,ptr,index,value,A);
if( err )
{
return err;
}
err = lis_matrix_assemble(A);
if( err ) return err;
if( matrix_type!=LIS_MATRIX_CSC )
{
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_type(B,LIS_MATRIX_CSR);
err = lis_matrix_convert_csc2csr(A,B);
if( err ) return err;
lis_matrix_storage_destroy(A);
lis_matrix_DLU_destroy(A);
lis_matrix_diag_destroy(A->WD);
if( A->l2g_map ) lis_free( A->l2g_map );
if( A->commtable ) lis_commtable_destroy( A->commtable );
if( A->ranges ) lis_free( A->ranges );
err = lis_matrix_copy_struct(B,A);
if( err ) return err;
lis_free(B);
}
return LIS_SUCCESS;
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_MATRIX A,A0;
LIS_VECTOR b,x,v;
LIS_SCALAR ntimes,nmflops,nnrm2;
LIS_SCALAR *value;
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nthreads, maxthreads;
LIS_INT gn,nnz,mode;
LIS_INT i,ii,j,jj,j0,j1,l,k,n,np,h,ih;
LIS_INT rn,rmin,rmax,rb;
LIS_INT is,ie,clsize,ci,*iw;
LIS_INT err,iter,matrix_type,s,ss,se;
LIS_INT *ptr,*index;
double mem,ra,rs,ri,ria,ca,time,time2,convtime,nnzs,nnzap,nnzt;
LIS_SCALAR val;
double commtime,comptime,flops;
FILE *file;
char path[1024];
LIS_DEBUG_FUNC_IN;
lis_initialize(&argc, &argv);
#ifdef USE_MPI
MPI_Comm_size(MPI_COMM_WORLD,&int_nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&int_my_rank);
nprocs = int_nprocs;
my_rank = int_my_rank;
#else
nprocs = 1;
my_rank = 0;
#endif
if( argc < 3 )
{
if( my_rank==0 )
{
printf("Usage: %s n iter [matrix_type]\n", argv[0]);
}
CHKERR(1);
}
n = atoi(argv[1]);
iter = atoi(argv[2]);
if (argv[3] == NULL) {
s = 0;
}
else {
s = atoi(argv[3]);
}
if( n<=0 )
{
#ifdef _LONGLONG
if( my_rank==0 ) printf("n=%lld <=0\n",n);
#else
if( my_rank==0 ) printf("n=%d <=0\n",n);
#endif
CHKERR(1);
}
if( iter<=0 )
{
#ifdef _LONGLONG
if( my_rank==0 ) printf("iter=%lld <= 0\n",iter);
#else
if( my_rank==0 ) printf("iter=%d <= 0\n",iter);
#endif
CHKERR(1);
}
if( s<0 || s>11 )
{
#ifdef _LONGLONG
if( my_rank==0 ) printf("matrix_type=%lld < 0 or matrix_type=%lld > 11\n",s,s);
#else
if( my_rank==0 ) printf("matrix_type=%d < 0 or matrix_type=%d > 11\n",s,s);
#endif
CHKERR(1);
}
if( my_rank==0 )
{
printf("\n");
#ifdef _LONGLONG
printf("number of processes = %lld\n",nprocs);
#else
printf("number of processes = %d\n",nprocs);
#endif
}
#ifdef _OPENMP
nthreads = omp_get_num_procs();
maxthreads = omp_get_max_threads();
if( my_rank==0 )
{
#ifdef _LONGLONG
printf("max number of threads = %lld\n", nthreads);
printf("number of threads = %lld\n", maxthreads);
#else
printf("max number of threads = %d\n", nthreads);
printf("number of threads = %d\n", maxthreads);
#endif
}
#else
nthreads = 1;
maxthreads = 1;
#endif
/* create matrix and vectors */
lis_matrix_create(LIS_COMM_WORLD,&A0);
lis_matrix_set_size(A0,0,n);
lis_matrix_get_size(A0,&n,&gn);
lis_matrix_get_range(A0,&is,&ie);
k = 0;
for(i=is;i<ie;i++)
{
if( i>0 ) lis_matrix_set_value(LIS_INS_VALUE,i,i-1,-1.0,A0);
if( i<gn-1 ) lis_matrix_set_value(LIS_INS_VALUE,i,i+1,-1.0,A0);
lis_matrix_set_value(LIS_INS_VALUE,i,i,2.0,A0);
}
err = lis_matrix_assemble(A0);
CHKERR(err);
n = A0->n;
gn = A0->gn;
nnz = A0->nnz;
np = A0->np-n;
#ifdef USE_MPI
MPI_Allreduce(&nnz,&i,1,LIS_MPI_INT,MPI_SUM,A0->comm);
nnzap = (double)i / (double)nprocs;
nnzt = ((double)nnz -nnzap)*((double)nnz -nnzap);
nnz = i;
MPI_Allreduce(&nnzt,&nnzs,1,MPI_DOUBLE,MPI_SUM,A0->comm);
nnzs = (nnzs / (double)nprocs)/nnzap;
MPI_Allreduce(&np,&i,1,LIS_MPI_INT,MPI_SUM,A0->comm);
np = i;
#endif
if( my_rank==0 )
{
#ifdef _LONGLONG
printf("matrix size = %lld x %lld (%lld nonzero entries)\n",gn,gn,nnz);
printf("iteration count = %lld\n\n",iter);
#else
printf("matrix size = %d x %d (%d nonzero entries)\n",gn,gn,nnz);
printf("iteration count = %d\n\n",iter);
#endif
}
err = lis_vector_duplicate(A0,&x);
if( err ) CHKERR(err);
err = lis_vector_duplicate(A0,&b);
if( err ) CHKERR(err);
lis_matrix_get_range(A0,&is,&ie);
for(i=0;i<n;i++)
{
err = lis_vector_set_value(LIS_INS_VALUE,i+is,1.0,x);
}
/*
MPI version of VBR is not implemented.
DNS is also excluded to reduce memory usage.
*/
if (s==0)
{
ss = 1;
se = 11;
}
else
{
ss = s;
se = s+1;
}
for (matrix_type=ss;matrix_type<se;matrix_type++)
{
if ( nprocs>1 && matrix_type==9 ) continue;
lis_matrix_duplicate(A0,&A);
lis_matrix_set_type(A,matrix_type);
err = lis_matrix_convert(A0,A);
if( err ) CHKERR(err);
comptime = 0.0;
commtime = 0.0;
for(i=0;i<iter;i++)
{
#ifdef USE_MPI
MPI_Barrier(A->comm);
time = lis_wtime();
lis_send_recv(A->commtable,x->value);
commtime += lis_wtime() - time;
MPI_Barrier(A->comm);
#endif
time2 = lis_wtime();
lis_matvec(A,x,b);
comptime += lis_wtime() - time2;
}
lis_vector_nrm2(b,&val);
if( my_rank==0 )
{
flops = 2.0*nnz*iter*1.0e-6 / comptime;
#ifdef USE_MPI
#ifdef _LONG__DOUBLE
#ifdef _LONGLONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#endif
#else
#ifdef _LONGLONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, communication = %e sec, communication/computation = %3.3f %%, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,commtime,commtime/comptime*100,val);
#endif
#endif
#else
#ifdef _LONG__DOUBLE
#ifdef _LONGLONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %Le\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#endif
#else
#ifdef _LONGLONG
printf("matrix_type = %2lld (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#else
printf("matrix_type = %2d (%s), computation = %e sec, %8.3f MFLOPS, 2-norm = %e\n",matrix_type,lis_storagename2[matrix_type-1],comptime,flops,val);
#endif
#endif
#endif
}
lis_matrix_destroy(A);
}
lis_matrix_destroy(A0);
lis_vector_destroy(b);
lis_vector_destroy(x);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
LIS_INT lis_input_mm_csr(LIS_MATRIX A, LIS_VECTOR b, LIS_VECTOR x, FILE *file)
{
char buf[BUFSIZE];
LIS_INT nr,nc,nnz;
LIS_INT i,j,my_rank;
LIS_INT err;
LIS_INT mmtype,mode;
LIS_INT n,is,ie;
LIS_INT ridx,cidx;
LIS_INT *ptr, *index;
LIS_INT *work;
LIS_INT isb,isx,isbin;
LIS_SCALAR val;
LIS_SCALAR *value;
LIS_MM_MATFMT matfmt;
LIS_DEBUG_FUNC_IN;
#ifdef USE_MPI
my_rank = A->my_rank;
#else
my_rank = 0;
#endif
/* check banner */
err = lis_input_mm_banner(file,&mmtype);
if( err ) return err;
/* check size */
err = lis_input_mm_size(file,&nr,&nc,&nnz,&isb,&isx,&isbin);
if( err ) return err;
err = lis_matrix_set_size(A,0,nr);
if( err ) return err;
#ifdef _LONGLONG
if( my_rank==0 ) printf("matrix size = %lld x %lld (%lld nonzero entries)\n\n",nr,nc,nnz);
#else
if( my_rank==0 ) printf("matrix size = %d x %d (%d nonzero entries)\n\n",nr,nc,nnz);
#endif
n = A->n;
ptr = NULL;
index = NULL;
value = NULL;
work = NULL;
lis_matrix_get_range(A,&is,&ie);
ptr = (LIS_INT *)lis_malloc( (n+1)*sizeof(LIS_INT),"lis_input_mm_csr::ptr" );
if( ptr==NULL )
{
LIS_SETERR_MEM((n+1)*sizeof(LIS_INT));
lis_free2(4,ptr,index,value,work);
return LIS_OUT_OF_MEMORY;
}
work = (LIS_INT *)lis_malloc( (n+1)*sizeof(LIS_INT),"lis_input_mm_csr::work" );
if( work==NULL )
{
LIS_SETERR_MEM((n+1)*sizeof(LIS_INT));
lis_free2(4,ptr,index,value,work);
return LIS_OUT_OF_MEMORY;
}
#ifdef _OPENMP
#pragma omp parallel for private(i)
#endif
for(i=0;i<n+1;i++)
{
ptr[i] = 0;
work[i] = 0;
}
/* read data */
mode = 1;
mode = *(char *)&mode;
if( mode!=(isbin-1) )
{
mode = LIS_TRUE;
}
else
{
mode = LIS_FALSE;
}
for( i=0; i<nnz; i++ )
{
if( isbin )
{
if( fread(&matfmt, sizeof(matfmt), 1, file)!=1 )
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
ridx = matfmt.i;
cidx = matfmt.j;
if( mode )
{
lis_bswap_int(1,&ridx);
lis_bswap_int(1,&cidx);
}
}
else
{
if( fgets(buf, BUFSIZE, file)==NULL )
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
#ifdef _LONGLONG
#ifdef _LONG__DOUBLE
if( sscanf(buf, "%lld %lld %Lg", &ridx, &cidx, &val) != 3 )
#else
if( sscanf(buf, "%lld %lld %lg", &ridx, &cidx, &val) != 3 )
#endif
#else
#ifdef _LONG__DOUBLE
if( sscanf(buf, "%d %d %Lg", &ridx, &cidx, &val) != 3 )
#else
if( sscanf(buf, "%d %d %lg", &ridx, &cidx, &val) != 3 )
#endif
#endif
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
}
/* if( val!=0.0 )*/
{
if( mmtype==MM_SYMM && ridx!=cidx )
{
if( cidx>is && cidx<=ie ) work[cidx-is-1]++;
}
if( ridx>is && ridx<=ie )
{
ptr[ridx-is]++;
}
}
}
ptr[0] = 0;
for( i=0; i<n; i++ )
{
if( mmtype==MM_SYMM )
{
ptr[i+1] += ptr[i] + work[i];
}
else
{
ptr[i+1] += ptr[i];
}
work[i] = 0;
}
index = (LIS_INT *)lis_malloc( ptr[n]*sizeof(LIS_INT),"lis_input_mm_csr::index" );
if( index==NULL )
{
LIS_SETERR_MEM(ptr[n]*sizeof(LIS_INT));
lis_free2(4,ptr,index,value,work);
return LIS_OUT_OF_MEMORY;
}
value = (LIS_SCALAR *)lis_malloc( ptr[n]*sizeof(LIS_SCALAR),"lis_input_mm_csr::value" );
if( value==NULL )
{
LIS_SETERR_MEM(ptr[n]*sizeof(LIS_SCALAR));
lis_free2(4,ptr,index,value,work);
return LIS_OUT_OF_MEMORY;
}
#ifdef _OPENMP
#pragma omp parallel for private(i,j)
#endif
for(i=0;i<n;i++)
{
for(j=ptr[i];j<ptr[i+1];j++)
{
index[j] = 0;
value[j] = 0.0;
}
}
rewind(file);
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
do
{
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
}while( buf[0]=='%' );
for( i=0; i<nnz; i++ )
{
if( isbin )
{
if( fread(&matfmt, sizeof(matfmt), 1, file)!=1 )
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
ridx = matfmt.i;
cidx = matfmt.j;
val = matfmt.value;
if( mode )
{
lis_bswap_int(1,&ridx);
lis_bswap_int(1,&cidx);
lis_bswap_scalar(1,&val);
}
}
else
{
if( fgets(buf, BUFSIZE, file) == NULL )
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
#ifdef _LONGLONG
#ifdef _LONG__DOUBLE
if( sscanf(buf, "%lld %lld %Lg", &ridx, &cidx, &val) != 3 )
#else
if( sscanf(buf, "%lld %lld %lg", &ridx, &cidx, &val) != 3 )
#endif
#else
#ifdef _LONG__DOUBLE
if( sscanf(buf, "%d %d %Lg", &ridx, &cidx, &val) != 3 )
#else
if( sscanf(buf, "%d %d %lg", &ridx, &cidx, &val) != 3 )
#endif
#endif
{
LIS_SETERR_FIO;
lis_free2(4,ptr,index,value,work);
return LIS_ERR_FILE_IO;
}
}
ridx--;
cidx--;
if( ridx==cidx && val==0.0 )
{
#ifdef _LONGLONG
printf("diagonal element is zero (i=%lld)\n",ridx);
#else
printf("diagonal element is zero (i=%d)\n",ridx);
#endif
}
/* if( val!=0.0 )*/
{
if( mmtype==MM_SYMM && ridx!=cidx )
{
if( cidx>=is && cidx<ie )
{
value[ptr[cidx-is]+work[cidx-is]] = val;
index[ptr[cidx-is]+work[cidx-is]] = ridx;
work[cidx-is]++;
}
}
if( ridx>=is && ridx<ie )
{
value[ptr[ridx-is]+work[ridx-is]] = val;
index[ptr[ridx-is]+work[ridx-is]] = cidx;
work[ridx-is]++;
}
}
}
#ifdef USE_MPI
MPI_Barrier(A->comm);
#endif
err = lis_matrix_set_csr(ptr[n],ptr,index,value,A);
if( err )
{
lis_free2(4,ptr,index,value,work);
return err;
}
err = lis_matrix_assemble(A);
if( err )
{
lis_matrix_storage_destroy(A);
lis_free(work);
return err;
}
if( b!=NULL && x!=NULL )
{
err = lis_input_mm_vec(A,b,x,file,isb,isx,isbin);
if( err )
{
lis_matrix_storage_destroy(A);
lis_free(work);
}
}
lis_free(work);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matrix_convert_csr2msr(LIS_MATRIX Ain, LIS_MATRIX Aout)
{
LIS_INT i,j,k,jj;
LIS_INT err;
LIS_INT n,nnz,ndz;
LIS_INT count;
LIS_INT *iw;
LIS_INT *index;
LIS_SCALAR *value;
LIS_DEBUG_FUNC_IN;
n = Ain->n;
nnz = Ain->nnz;
iw = NULL;
index = NULL;
value = NULL;
iw = (LIS_INT *)lis_malloc( (n+1)*sizeof(LIS_INT),"lis_matrix_convert_csr2msr::iw" );
if( iw==NULL )
{
LIS_SETERR_MEM((n+1)*sizeof(LIS_INT));
return LIS_ERR_OUT_OF_MEMORY;
}
/* check ndz */
for(i=0;i<n+1;i++) iw[i] = 0;
count = 0;
#ifdef _OPENMP
#pragma omp parallel private(i,j)
#endif
{
#ifdef _OPENMP
#pragma omp for
#endif
for(i=0;i<n;i++)
{
iw[i+1] = 0;
for(j=Ain->ptr[i];j<Ain->ptr[i+1];j++)
{
if( i==Ain->index[j] )
{
iw[i+1] = 1;
}
}
}
#ifdef _OPENMP
#pragma omp for reduction(+:count)
#endif
for(i=0;i<n;i++)
{
count += iw[i+1];
}
#ifdef _OPENMP
#pragma omp for
#endif
for(i=0;i<n;i++)
{
iw[i+1] = Ain->ptr[i+1]-Ain->ptr[i]-iw[i+1];
}
}
ndz = n - count;
err = lis_matrix_malloc_msr(n,nnz,ndz,&index,&value);
if( err )
{
lis_free2(3,index,value,iw);
return err;
}
/* convert msr */
iw[0] = n+1;
for(i=0;i<n;i++)
{
iw[i+1] = iw[i+1] + iw[i];
}
#ifdef _OPENMP
#pragma omp parallel private(i,j,k)
#endif
{
#ifdef _OPENMP
#pragma omp for
#endif
for(i=0;i<n+1;i++)
{
index[i] = iw[i];
}
#ifdef _OPENMP
#pragma omp for
#endif
for(i=0;i<n;i++)
{
k = index[i];
for(j=Ain->ptr[i];j<Ain->ptr[i+1];j++)
{
jj = Ain->index[j];
if( jj==i )
{
value[i] = Ain->value[j];
}
else
{
value[k] = Ain->value[j];
index[k] = Ain->index[j];
k++;
}
}
}
}
err = lis_matrix_set_msr(nnz,ndz,index,value,Aout);
if( err )
{
lis_free2(3,index,value,iw);
return err;
}
err = lis_matrix_assemble(Aout);
if( err )
{
lis_free(iw);
lis_matrix_storage_destroy(Aout);
return err;
}
lis_free(iw);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matrix_convert_msr2csr(LIS_MATRIX Ain, LIS_MATRIX Aout)
{
LIS_INT i,j,k;
LIS_INT err;
LIS_INT n,nnz,is;
LIS_INT *ptr,*index;
LIS_SCALAR *value;
LIS_DEBUG_FUNC_IN;
n = Ain->n;
nnz = Ain->nnz;
is = Ain->is;
ptr = NULL;
index = NULL;
value = NULL;
err = lis_matrix_malloc_csr(n,nnz,&ptr,&index,&value);
if( err )
{
return err;
}
/* convert csr */
#ifdef _OPENMP
#pragma omp parallel for private(i)
#endif
for(i=0;i<n;i++)
{
ptr[i+1] = Ain->index[i+1] - Ain->index[i];
if( Ain->value[i]!=0.0 )
{
ptr[i+1]++;
}
}
ptr[0] = 0;
for(i=0;i<n;i++)
{
ptr[i+1] += ptr[i];
}
#ifdef _OPENMP
#pragma omp parallel for private(i,j,k)
#endif
for(i=0;i<n;i++)
{
k = ptr[i];
if( Ain->value[i]!=(LIS_SCALAR)0.0 )
{
value[k] = Ain->value[i];
index[k] = i;
k++;
}
for(j=Ain->index[i];j<Ain->index[i+1];j++)
{
value[k] = Ain->value[j];
index[k] = Ain->index[j];
k++;
}
}
err = lis_matrix_set_csr(nnz,ptr,index,value,Aout);
if( err )
{
lis_free2(3,ptr,index,value);
return err;
}
err = lis_matrix_assemble(Aout);
if( err )
{
lis_matrix_storage_destroy(Aout);
return err;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matrix_convert_rco2csc(LIS_MATRIX Ain, LIS_MATRIX Aout)
{
LIS_INT i,j,k,l,n,nnz,err;
LIS_INT *ptr,*index,*iw;
LIS_SCALAR *value;
LIS_DEBUG_FUNC_IN;
ptr = NULL;
index = NULL;
value = NULL;
iw = NULL;
n = Ain->n;
iw = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_matrix_convert_rco2csc::iw");
if( iw==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
lis_free2(4,ptr,index,value,iw);
return LIS_OUT_OF_MEMORY;
}
ptr = (LIS_INT *)lis_malloc((n+1)*sizeof(LIS_INT),"lis_matrix_convert_rco2csc::ptr");
if( ptr==NULL )
{
LIS_SETERR_MEM((n+1)*sizeof(LIS_INT));
lis_free2(4,ptr,index,value,iw);
return LIS_OUT_OF_MEMORY;
}
for(i=0;i<n;i++) iw[i] = 0;
for(i=0;i<n;i++)
{
for(j=0;j<Ain->w_row[i];j++)
{
iw[Ain->w_index[i][j]]++;
}
}
ptr[0] = 0;
for(i=0;i<n;i++)
{
ptr[i+1] = ptr[i] + iw[i];
iw[i] = ptr[i];
}
nnz = ptr[n];
index = (LIS_INT *)lis_malloc( nnz*sizeof(LIS_INT),"lis_matrix_convert_rco2csc::index" );
if( index==NULL )
{
LIS_SETERR_MEM(nnz*sizeof(LIS_INT));
lis_free2(4,ptr,index,value,iw);
return LIS_OUT_OF_MEMORY;
}
value = (LIS_SCALAR *)lis_malloc( nnz*sizeof(LIS_SCALAR),"lis_matrix_convert_rco2csc::value" );
if( value==NULL )
{
LIS_SETERR_MEM(nnz*sizeof(LIS_SCALAR));
lis_free2(4,ptr,index,value,iw);
return LIS_OUT_OF_MEMORY;
}
for(i=0;i<n;i++)
{
for(j=0;j<Ain->w_row[i];j++)
{
k = Ain->w_index[i][j];
l = iw[k];
value[l] = Ain->w_value[i][j];
index[l] = i;
iw[k]++;
}
}
err = lis_matrix_set_csc(nnz,ptr,index,value,Aout);
if( err )
{
lis_free2(4,ptr,index,value,iw);
return err;
}
err = lis_matrix_assemble(Aout);
if( err )
{
lis_matrix_storage_destroy(Aout);
return err;
}
lis_free(iw);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_matrix_convert_rco2bsr(LIS_MATRIX Ain, LIS_MATRIX Aout)
{
LIS_INT i,j,k,n,gn,nnz,bnnz,nr,nc,bnr,bnc,err;
LIS_INT ii,jj,kk,bj,jpos,ij,kv,bi;
LIS_INT *iw,*iw2;
LIS_INT *bptr,*bindex;
LIS_SCALAR *value;
LIS_DEBUG_FUNC_IN;
bnr = Ain->conv_bnr;
bnc = Ain->conv_bnc;
n = Ain->n;
gn = Ain->gn;
nr = 1 + (n-1)/bnr;
nc = 1 + (gn-1)/bnc;
bptr = NULL;
bindex = NULL;
value = NULL;
iw = NULL;
iw2 = NULL;
bptr = (LIS_INT *)lis_malloc( (nr+1)*sizeof(LIS_INT),"lis_matrix_convert_rco2bsr::bptr" );
if( bptr==NULL )
{
LIS_SETERR_MEM((nr+1)*sizeof(LIS_INT));
lis_free2(5,bptr,bindex,value,iw,iw2);
return LIS_OUT_OF_MEMORY;
}
#ifdef _OPENMP
#pragma omp parallel private(i,k,ii,j,bj,kk,ij,jj,iw,iw2,kv,jpos)
#endif
{
iw = (LIS_INT *)lis_malloc( nc*sizeof(LIS_INT),"lis_matrix_convert_rco2bsr::iw" );
iw2 = (LIS_INT *)lis_malloc( nc*sizeof(LIS_INT),"lis_matrix_convert_rco2bsr::iw2" );
memset(iw,0,nc*sizeof(LIS_INT));
#ifdef _OPENMP
#pragma omp for
#endif
for(i=0;i<nr;i++)
{
k = 0;
kk = bnr*i;
jj = 0;
for(ii=0;ii+kk<n&&ii<bnr;ii++)
{
for(j=0;j<Ain->w_row[kk+ii];j++)
{
bj = Ain->w_index[kk+ii][j]/bnc;
jpos = iw[bj];
if( jpos==0 )
{
iw[bj] = 1;
iw2[jj] = bj;
jj++;
}
}
}
for(bj=0;bj<jj;bj++)
{
k++;
ii = iw2[bj];
iw[ii]=0;
}
bptr[i+1] = k;
}
lis_free(iw);
lis_free(iw2);
}
bptr[0] = 0;
for(i=0;i<nr;i++)
{
bptr[i+1] += bptr[i];
}
bnnz = bptr[nr];
nnz = bnnz*bnr*bnc;
bindex = (LIS_INT *)lis_malloc( bnnz*sizeof(LIS_INT),"lis_matrix_convert_rco2bsr::bindex" );
if( bindex==NULL )
{
LIS_SETERR_MEM((nr+1)*sizeof(LIS_INT));
lis_free2(3,bptr,bindex,value);
return LIS_OUT_OF_MEMORY;
}
value = (LIS_SCALAR *)lis_malloc( nnz*sizeof(LIS_SCALAR),"lis_matrix_convert_rco2bsr::value" );
if( value==NULL )
{
LIS_SETERR_MEM(nnz*sizeof(LIS_SCALAR));
lis_free2(3,bptr,bindex,value);
return LIS_OUT_OF_MEMORY;
}
/* convert bsr */
#ifdef _OPENMP
#pragma omp parallel private(bi,i,ii,k,j,bj,jpos,kv,kk,ij,jj,iw)
#endif
{
iw = (LIS_INT *)lis_malloc( nc*sizeof(LIS_INT),"lis_matrix_convert_rco2bsr::iw" );
memset(iw,0,nc*sizeof(LIS_INT));
#ifdef _OPENMP
#pragma omp for
#endif
for(bi=0;bi<nr;bi++)
{
i = bi*bnr;
ii = 0;
kk = bptr[bi];
while( i+ii<n && ii<=bnr-1 )
{
for( k=0;k<Ain->w_row[i+ii];k++)
{
j = Ain->w_index[i+ii][k];
bj = j/bnc;
j = j%bnc;
jpos = iw[bj];
if( jpos==0 )
{
kv = kk * bnr * bnc;
iw[bj] = kv+1;
bindex[kk] = bj;
for(jj=0;jj<bnr*bnc;jj++) value[kv+jj] = 0.0;
ij = j*bnr + ii;
value[kv+ij] = Ain->w_value[i+ii][k];
kk = kk+1;
}
else
{
ij = j*bnr + ii;
value[jpos+ij-1] = Ain->w_value[i+ii][k];
}
}
ii = ii+1;
}
for(j=bptr[bi];j<bptr[bi+1];j++)
{
iw[bindex[j]] = 0;
}
}
lis_free(iw);
}
err = lis_matrix_set_bsr(bnr,bnc,bnnz,bptr,bindex,value,Aout);
if( err )
{
lis_free2(3,bptr,bindex,value);
return err;
}
err = lis_matrix_assemble(Aout);
if( err )
{
lis_matrix_storage_destroy(Aout);
return err;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_INT i,n,gn,is,ie;
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nesol;
LIS_MATRIX A;
LIS_VECTOR x;
LIS_REAL evalue0;
LIS_ESOLVER esolver;
LIS_REAL residual;
LIS_INT iter;
double time;
double itime,ptime,p_c_time,p_i_time;
char esolvername[128];
LIS_DEBUG_FUNC_IN;
lis_initialize(&argc, &argv);
#ifdef USE_MPI
MPI_Comm_size(MPI_COMM_WORLD,&int_nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&int_my_rank);
nprocs = int_nprocs;
my_rank = int_my_rank;
#else
nprocs = 1;
my_rank = 0;
#endif
if( argc < 2 )
{
if( my_rank==0 )
{
printf("Usage: %s n [eoptions]\n", argv[0]);
}
CHKERR(1);
}
if( my_rank==0 )
{
printf("\n");
printf("number of processes = %d\n",nprocs);
}
#ifdef _OPENMP
if( my_rank==0 )
{
#ifdef _LONG__LONG
printf("max number of threads = %lld\n",omp_get_num_procs());
printf("number of threads = %lld\n",omp_get_max_threads());
#else
printf("max number of threads = %d\n",omp_get_num_procs());
printf("number of threads = %d\n",omp_get_max_threads());
#endif
}
#endif
/* generate coefficient matrix for one dimensional Poisson equation */
n = atoi(argv[1]);
lis_matrix_create(LIS_COMM_WORLD,&A);
lis_matrix_set_size(A,0,n);
lis_matrix_get_size(A,&n,&gn);
lis_matrix_get_range(A,&is,&ie);
for(i=is;i<ie;i++)
{
if( i>0 ) lis_matrix_set_value(LIS_INS_VALUE,i,i-1,-1.0,A);
if( i<gn-1 ) lis_matrix_set_value(LIS_INS_VALUE,i,i+1,-1.0,A);
lis_matrix_set_value(LIS_INS_VALUE,i,i,2.0,A);
}
lis_matrix_set_type(A,LIS_MATRIX_CSR);
lis_matrix_assemble(A);
lis_vector_duplicate(A,&x);
lis_esolver_create(&esolver);
lis_esolver_set_option("-eprint mem",esolver);
lis_esolver_set_optionC(esolver);
lis_esolve(A, x, &evalue0, esolver);
lis_esolver_get_esolver(esolver,&nesol);
lis_esolver_get_esolvername(nesol,esolvername);
lis_esolver_get_residualnorm(esolver, &residual);
lis_esolver_get_iter(esolver, &iter);
lis_esolver_get_timeex(esolver,&time,&itime,&ptime,&p_c_time,&p_i_time);
if( my_rank==0 ) {
printf("%s: mode number = %d\n", esolvername, 0);
#ifdef _LONG__DOUBLE
printf("%s: eigenvalue = %Le\n", esolvername, evalue0);
#else
printf("%s: eigenvalue = %e\n", esolvername, evalue0);
#endif
#ifdef _LONG__LONG
printf("%s: number of iterations = %lld\n",esolvername, iter);
#else
printf("%s: number of iterations = %d\n",esolvername, iter);
#endif
printf("%s: elapsed time = %e sec.\n", esolvername, time);
printf("%s: preconditioner = %e sec.\n", esolvername, ptime);
printf("%s: matrix creation = %e sec.\n", esolvername, p_c_time);
printf("%s: linear solver = %e sec.\n", esolvername, itime);
#ifdef _LONG__DOUBLE
printf("%s: relative residual = %Le\n\n",esolvername, residual);
#else
printf("%s: relative residual = %e\n\n",esolvername, residual);
#endif
}
/*
lis_vector_nrm2(x, &xnrm2);
lis_vector_scale((1/xnrm2*sqrt(n)), x);
lis_vector_print(x);
*/
/*
lis_vector_create(LIS_COMM_WORLD,&y);
lis_matrix_create(LIS_COMM_WORLD,&B);
lis_esolver_get_evalues(esolver,y);
lis_esolver_get_evectors(esolver,B);
lis_output_vector(y,LIS_FMT_MM,"evalues.out");
lis_output_matrix(B,LIS_FMT_MM,"evectors.out");
lis_vector_destroy(y);
lis_matrix_destroy(B);
*/
lis_esolver_destroy(esolver);
lis_matrix_destroy(A);
lis_vector_destroy(x);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_MATRIX A;
LIS_VECTOR b,x,u;
LIS_SOLVER solver;
LIS_INT my_rank;
#ifdef USE_MPI
int int_nprocs,int_my_rank;
#endif
LIS_INT err,i,n,gn,is,ie,iter;
n = 12;
lis_initialize(&argc, &argv);
#ifdef USE_MPI
MPI_Comm_size(MPI_COMM_WORLD,&int_nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&int_my_rank);
my_rank = int_my_rank;
#else
my_rank = 0;
#endif
lis_matrix_create(LIS_COMM_WORLD,&A);
err = lis_matrix_set_size(A,0,n);
CHKERR(err);
lis_matrix_get_size(A,&n,&gn);
lis_matrix_get_range(A,&is,&ie);
for(i=is;i<ie;i++)
{
if( i>0 ) lis_matrix_set_value(LIS_INS_VALUE,i,i-1,-1.0,A);
if( i<gn-1 ) lis_matrix_set_value(LIS_INS_VALUE,i,i+1,-1.0,A);
lis_matrix_set_value(LIS_INS_VALUE,i,i,2.0,A);
}
lis_matrix_set_type(A,LIS_MATRIX_CSR);
lis_matrix_assemble(A);
lis_vector_duplicate(A,&u);
lis_vector_duplicate(A,&b);
lis_vector_duplicate(A,&x);
lis_vector_set_all(1.0,u);
lis_matvec(A,u,b);
lis_solver_create(&solver);
lis_solver_set_option("-print mem",solver);
lis_solver_set_optionC(solver);
lis_solve(A,b,x,solver);
lis_solver_get_iter(solver,&iter);
if (my_rank==0)
{
#ifdef _LONG__LONG
printf("number of iterations = %lld\n",iter);
#else
printf("number of iterations = %d\n",iter);
#endif
printf("\n");
}
lis_vector_print(x);
lis_matrix_destroy(A);
lis_vector_destroy(b);
lis_vector_destroy(x);
lis_vector_destroy(u);
lis_solver_destroy(solver);
lis_finalize();
return 0;
}
int
solveLis(DATA *data, threadData_t *threadData, int sysNumber)
{
void *dataAndThreadData[2] = {data, threadData};
LINEAR_SYSTEM_DATA* systemData = &(data->simulationInfo->linearSystemData[sysNumber]);
DATA_LIS* solverData = (DATA_LIS*)systemData->solverData;
int i, ret, success = 1, ni, iflag = 1, n = systemData->size, eqSystemNumber = systemData->equationIndex;
char *lis_returncode[] = {"LIS_SUCCESS", "LIS_ILL_OPTION", "LIS_BREAKDOWN", "LIS_OUT_OF_MEMORY", "LIS_MAXITER", "LIS_NOT_IMPLEMENTED", "LIS_ERR_FILE_IO"};
LIS_INT err;
int indexes[2] = {1,eqSystemNumber};
infoStreamPrintWithEquationIndexes(LOG_LS, 0, indexes, "Start solving Linear System %d (size %d) at time %g with Lis Solver",
eqSystemNumber, (int) systemData->size,
data->localData[0]->timeValue);
/* set old values as start value for the iteration */
for(i=0; i<n; i++) {
err = lis_vector_set_value(LIS_INS_VALUE, i, systemData->x[i], solverData->x);
}
rt_ext_tp_tick(&(solverData->timeClock));
if (0 == systemData->method)
{
lis_matrix_set_size(solverData->A, solverData->n_row, 0);
/* set A matrix */
systemData->setA(data, threadData, systemData);
lis_matrix_assemble(solverData->A);
/* set b vector */
systemData->setb(data, threadData, systemData);
} else {
lis_matrix_set_size(solverData->A, solverData->n_row, 0);
/* calculate jacobian -> matrix A*/
if(systemData->jacobianIndex != -1) {
getAnalyticalJacobianLis(data, threadData, sysNumber);
} else {
assertStreamPrint(threadData, 1, "jacobian function pointer is invalid" );
}
lis_matrix_assemble(solverData->A);
/* calculate vector b (rhs) */
memcpy(solverData->work, systemData->x, sizeof(double)*solverData->n_row);
wrapper_fvec_lis(solverData->work, systemData->b, dataAndThreadData, sysNumber);
/* set b vector */
for(i=0; i<n; i++) {
err = lis_vector_set_value(LIS_INS_VALUE, i, systemData->b[i], solverData->b);
}
}
infoStreamPrint(LOG_LS, 0, "### %f time to set Matrix A and vector b.", rt_ext_tp_tock(&(solverData->timeClock)));
rt_ext_tp_tick(&(solverData->timeClock));
err = lis_solve(solverData->A,solverData->b,solverData->x,solverData->solver);
infoStreamPrint(LOG_LS, 0, "Solve System: %f", rt_ext_tp_tock(&(solverData->timeClock)));
if (err) {
warningStreamPrint(LOG_LS_V, 0, "lis_solve : %s(code=%d)\n\n ", lis_returncode[err], err);
printLisMatrixCSR(solverData->A, solverData->n_row);
success = 0;
}
/* Log A*x=b */
if(ACTIVE_STREAM(LOG_LS_V))
{
char *buffer = (char*)malloc(sizeof(char)*n*25);
printLisMatrixCSR(solverData->A, n);
/* b vector */
infoStreamPrint(LOG_LS_V, 1, "b vector [%d]", n);
for(i=0; i<n; i++)
{
buffer[0] = 0;
sprintf(buffer, "%s%20.12g ", buffer, solverData->b->value[i]);
infoStreamPrint(LOG_LS_V, 0, "%s", buffer);
}
messageClose(LOG_LS_V);
free(buffer);
}
/* print solution */
if (1 == success) {
if (1 == systemData->method) {
/* take the solution */
lis_vector_get_values(solverData->x, 0, solverData->n_row, systemData->x);
for(i = 0; i < solverData->n_row; ++i)
systemData->x[i] += solverData->work[i];
/* update inner equations */
wrapper_fvec_lis(systemData->x, solverData->work, dataAndThreadData, sysNumber);
} else {
/* write solution */
lis_vector_get_values(solverData->x, 0, solverData->n_row, systemData->x);
}
if (ACTIVE_STREAM(LOG_LS_V))
{
infoStreamPrint(LOG_LS_V, 1, "Solution x:");
infoStreamPrint(LOG_LS_V, 0, "System %d numVars %d.", eqSystemNumber, modelInfoGetEquation(&data->modelData->modelDataXml,eqSystemNumber).numVar);
for(i = 0; i < systemData->size; ++i)
infoStreamPrint(LOG_LS_V, 0, "[%d] %s = %g", i+1, modelInfoGetEquation(&data->modelData->modelDataXml,eqSystemNumber).vars[i], systemData->x[i]);
messageClose(LOG_LS_V);
}
}
else
{
warningStreamPrint(LOG_STDOUT, 0,
"Failed to solve linear system of equations (no. %d) at time %f, system status %d.",
(int)systemData->equationIndex, data->localData[0]->timeValue, err);
}
return success;
}
