LIS_INT lis_output(LIS_MATRIX A, LIS_VECTOR b, LIS_VECTOR x, LIS_INT format, char *path)
{
LIS_INT err;
LIS_MATRIX B;
LIS_DEBUG_FUNC_IN;
#ifdef USE_MPI
MPI_Barrier(A->comm);
#endif
err = lis_matrix_check(A,LIS_MATRIX_CHECK_ALL);
if( err ) return err;
if( format==LIS_FMT_MM || format==LIS_FMT_MMB )
{
switch( A->matrix_type )
{
case LIS_MATRIX_CSR:
err = lis_output_mm_csr(A,b,x,format,path);
break;
default:
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_type(B,LIS_MATRIX_CSR);
err = lis_matrix_convert(A,B);
if( err ) return err;
err = lis_output_mm_csr(B,b,x,format,path);
lis_matrix_destroy(B);
break;
}
}
LIS_DEBUG_FUNC_OUT;
return err;
}
LIS_INT lis_precon_create_ilut(LIS_SOLVER solver, LIS_PRECON precon)
{
LIS_INT err;
LIS_MATRIX A,B;
LIS_DEBUG_FUNC_IN;
switch( solver->A->matrix_type )
{
case LIS_MATRIX_CSR:
err = lis_precon_create_ilut_csr(solver,precon);
lis_psolve_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolve_ilut_csr;
lis_psolvet_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolvet_ilut_csr;
break;
default:
A = solver->A;
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_type(B,LIS_MATRIX_CSR);
err = lis_matrix_convert(A,B);
if( err ) return err;
solver->A = B;
err = lis_precon_create_ilut_csr(solver,precon);
lis_psolve_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolve_ilut_csr;
lis_psolvet_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolvet_ilut_csr;
lis_matrix_destroy(B);
solver->A = A;
break;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_precon_create_ilut(LIS_SOLVER solver, LIS_PRECON precon)
{
#ifdef ENABLE_BSR
LIS_INT storage,block;
#endif
LIS_INT err;
LIS_MATRIX A,B;
LIS_DEBUG_FUNC_IN;
#ifdef ENABLE_BSR
storage = solver->options[LIS_OPTIONS_STORAGE];
block = solver->options[LIS_OPTIONS_STORAGE_BLOCK];
if( solver->A->matrix_type!=LIS_MATRIX_BSR && storage==LIS_MATRIX_BSR )
{
err = lis_matrix_convert_self(solver);
if( err ) return err;
}
#endif
switch( solver->A->matrix_type )
{
case LIS_MATRIX_CSR:
err = lis_precon_create_ilut_csr(solver,precon);
lis_psolve_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolve_ilut_csr;
lis_psolvet_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolvet_ilut_csr;
break;
#ifdef ENABLE_BSR
case LIS_MATRIX_BSR:
err = lis_precon_create_ilut_bsr(solver,precon);
lis_psolve_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolve_ilut_bsr;
lis_psolvet_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolvet_ilut_bsr;
break;
#endif
default:
A = solver->A;
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_type(B,LIS_MATRIX_CSR);
err = lis_matrix_convert(A,B);
if( err ) return err;
solver->A = B;
err = lis_precon_create_ilut_csr(solver,precon);
lis_psolve_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolve_ilut_csr;
lis_psolvet_xxx[LIS_PRECON_TYPE_ILUT] = lis_psolvet_ilut_csr;
lis_matrix_destroy(B);
solver->A = A;
break;
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
void lis_matrix_convert_f(LIS_MATRIX_F *Ain, LIS_MATRIX_F *Aout, LIS_INT *ierr)
{
LIS_MATRIX AAin,AAout;
LIS_DEBUG_FUNC_IN;
AAin = (LIS_MATRIX)LIS_V2P(Ain);
AAout = (LIS_MATRIX)LIS_V2P(Aout);
*ierr = lis_matrix_convert(AAin,AAout);
*Aout = LIS_P2V(AAout);
LIS_DEBUG_FUNC_OUT;
return;
}
LIS_INT lis_precon_create_sainv(LIS_SOLVER solver, LIS_PRECON precon)
{
LIS_INT err;
LIS_MATRIX A,B;
LIS_DEBUG_FUNC_IN;
switch( solver->A->matrix_type )
{
case LIS_MATRIX_CSR:
err = lis_precon_create_sainv_csr(solver,precon);
break;
default:
A = solver->A;
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_type(B,LIS_MATRIX_CSR);
err = lis_matrix_convert(A,B);
if( err ) return err;
solver->A = B;
err = lis_precon_create_sainv_csr(solver,precon);
lis_matrix_destroy(B);
solver->A = A;
break;
}
#ifndef USE_QUAD_PRECISION
err = lis_vector_duplicate(solver->A,&precon->temp);
#else
if( solver->precision==LIS_PRECISION_DEFAULT )
{
err = lis_vector_duplicate(solver->A,&precon->temp);
}
else
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,solver->A,&precon->temp);
}
#endif
if( err ) return err;
precon->A = solver->A;
precon->is_copy = LIS_FALSE;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_input_hb(LIS_MATRIX A, LIS_VECTOR b, LIS_VECTOR x, FILE *file)
{
LIS_INT err;
LIS_INT matrix_type;
LIS_MATRIX B;
LIS_DEBUG_FUNC_IN;
matrix_type = A->matrix_type;
err = lis_input_hb_csr(A,b,x,file);
if( err ) return err;
if( matrix_type!=LIS_MATRIX_CSR && matrix_type!=LIS_MATRIX_CSC )
{
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_type(B,matrix_type);
err = lis_matrix_convert(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);
if( A->matrix_type==LIS_MATRIX_JAD )
{
A->work = (LIS_SCALAR *)lis_malloc(A->n*sizeof(LIS_SCALAR),"lis_input_hb::A->work");
if( A->work==NULL )
{
LIS_SETERR_MEM(A->n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
}
}
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_MATRIX A0,A,B;
LIS_VECTOR x,b,u;
LIS_SOLVER solver;
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nsol,rhs,len,i,j,k,jj,kk,p,nrow_p,l,n;
LIS_INT err,iter,iter_double,iter_quad;
LIS_INT *iw,*nrow,*index,*ptr;
LIS_SCALAR *s,t,*value;
double times,itimes,ptimes,p_c_times,p_i_times;
LIS_REAL resid;
char solvername[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 < 5 )
{
if( my_rank==0 )
{
printf("Usage: %s matrix_filename rhs_setting solution_filename residual_filename [options]\n", argv[0]);
}
CHKERR(1);
}
len = (LIS_INT)strlen(argv[2]);
if( len==1 )
{
if( argv[2][0]=='0' || argv[2][0]=='1' || argv[2][0]=='2' )
{
rhs = atoi(argv[2]);
}
else
{
rhs = -1;
}
}
else
{
rhs = -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
/* read matrix and vectors from file */
err = lis_matrix_create(LIS_COMM_WORLD,&A); CHKERR(err);
err = lis_vector_create(LIS_COMM_WORLD,&b); CHKERR(err);
err = lis_vector_create(LIS_COMM_WORLD,&x); CHKERR(err);
err = lis_input(A,b,x,argv[1]);
CHKERR(err);
err = lis_matrix_duplicate(A,&A0);
CHKERR(err);
lis_matrix_set_type(A0,LIS_MATRIX_CSR);
err = lis_matrix_convert(A,A0);
CHKERR(err);
lis_matrix_destroy(A);
A = A0;
err = lis_vector_duplicate(A,&u);
CHKERR(err);
if( lis_vector_is_null(b) )
{
lis_vector_destroy(b);
lis_vector_duplicate(A,&b);
CHKERR(err);
if( rhs==0 )
{
CHKERR(1);
}
else if( rhs==1 )
{
err = lis_vector_set_all(1.0,b);
}
else
{
err = lis_vector_set_all(1.0,u);
lis_matvec(A,u,b);
}
}
if( rhs==-1 )
{
lis_input_vector(b,argv[2]);
}
if( lis_vector_is_null(x) )
{
lis_vector_destroy(x);
err = lis_vector_duplicate(A,&x);
CHKERR(err);
}
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_itersex(solver,&iter,&iter_double,&iter_quad);
lis_solver_get_timeex(solver,×,&itimes,&ptimes,&p_c_times,&p_i_times);
lis_solver_get_residualnorm(solver,&resid);
lis_solver_get_solver(solver,&nsol);
lis_solver_get_solvername(nsol,solvername);
/* write results */
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,times);
printf("%s: preconditioner = %e sec.\n",solvername, ptimes);
printf("%s: matrix creation = %e sec.\n",solvername, p_c_times);
printf("%s: linear solver = %e sec.\n",solvername, itimes);
#ifdef _LONG__DOUBLE
printf("%s: relative residual 2-norm = %Le\n\n",solvername,resid);
#else
printf("%s: relative residual 2-norm = %e\n\n",solvername,resid);
#endif
}
/* write solution */
lis_output_vector(x,LIS_FMT_MM,argv[3]);
/* write residual */
lis_solver_output_rhistory(solver, argv[4]);
lis_solver_destroy(solver);
lis_vector_destroy(x);
lis_vector_destroy(u);
lis_vector_destroy(b);
lis_matrix_destroy(A);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
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_solve_kernel(LIS_MATRIX A, LIS_VECTOR b, LIS_VECTOR x, LIS_SOLVER solver, LIS_PRECON precon)
{
LIS_INT nsolver, precon_type, maxiter;
LIS_INT err;
LIS_SCALAR *residual;
LIS_VECTOR xx;
LIS_INT output;
LIS_INT scale;
LIS_INT conv_cond;
LIS_INT precision,is_use_at,storage,block;
LIS_INT i,n,np;
double p_c_times, p_i_times,itimes;
LIS_SCALAR nrm2,tol,tol_w;
LIS_VECTOR t;
LIS_VECTOR bb;
LIS_MATRIX AA,B;
LIS_MATRIX At;
char buf[64];
LIS_DEBUG_FUNC_IN;
nsolver = solver->options[LIS_OPTIONS_SOLVER];
precon_type = solver->options[LIS_OPTIONS_PRECON];
maxiter = solver->options[LIS_OPTIONS_MAXITER];
output = solver->options[LIS_OPTIONS_OUTPUT];
scale = solver->options[LIS_OPTIONS_SCALE];
precision = solver->options[LIS_OPTIONS_PRECISION];
is_use_at = solver->options[LIS_OPTIONS_USE_AT];
storage = solver->options[LIS_OPTIONS_STORAGE];
block = solver->options[LIS_OPTIONS_STORAGE_BLOCK];
conv_cond = solver->options[LIS_OPTIONS_CONV_COND];
tol = solver->params[LIS_PARAMS_RESID-LIS_OPTIONS_LEN];
tol_w = solver->params[LIS_PARAMS_RESID_WEIGHT-LIS_OPTIONS_LEN];
solver->precision = precision;
if( nsolver < 1 || nsolver > LIS_SOLVERS_LEN )
{
LIS_SETERR2(LIS_ERR_ILL_ARG,"Parameter LIS_OPTIONS_SOLVER is %d (Set between 1 to %d)\n",nsolver, LIS_SOLVERS_LEN);
return LIS_ERR_ILL_ARG;
}
if( precon_type < 0 || precon_type > precon_register_type )
{
LIS_SETERR2(LIS_ERR_ILL_ARG,"Parameter LIS_OPTIONS_PRECON is %d (Set between 0 to %d)\n",precon_type, precon_register_type-1);
return LIS_ERR_ILL_ARG;
}
if( maxiter<0 )
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Parameter LIS_OPTIONS_MAXITER(=%d) is less than 0\n",maxiter);
return LIS_ERR_ILL_ARG;
}
#ifdef USE_MPI
if( precon_type == LIS_PRECON_TYPE_SAAMG && solver->A->nprocs < 2)
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Parameter A->nprocs (=%d) is less than 2 (Set more than 1 when using parallel version of SAAMG)\n",solver->A->nprocs);
return LIS_ERR_ILL_ARG;
}
#endif
#ifdef USE_QUAD_PRECISION
if( precision==LIS_PRECISION_QUAD && lis_solver_execute_quad[nsolver]==NULL )
{
LIS_SETERR1(LIS_ERR_NOT_IMPLEMENTED,"Quad precision solver %s is not implemented\n",lis_solvername[nsolver]);
return LIS_ERR_NOT_IMPLEMENTED;
}
else if( precision==LIS_PRECISION_SWITCH && lis_solver_execute_switch[nsolver]==NULL )
{
LIS_SETERR1(LIS_ERR_NOT_IMPLEMENTED,"Switch solver %s is not implemented\n",lis_solvername[nsolver]);
return LIS_ERR_NOT_IMPLEMENTED;
}
if( solver->options[LIS_OPTIONS_SWITCH_MAXITER]==-1 )
{
solver->options[LIS_OPTIONS_SWITCH_MAXITER] = maxiter;
}
#endif
err = lis_solver_check_params[nsolver](solver);
if( err )
{
solver->retcode = err;
return err;
}
/* end parameter check */
solver->A = A;
solver->b = b;
/* create initial vector */
#ifndef USE_QUAD_PRECISION
err = lis_vector_duplicate(A,&xx);
#else
if( precision==LIS_PRECISION_DOUBLE )
{
err = lis_vector_duplicate(A,&xx);
}
else
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,A,&xx);
}
#endif
if( err )
{
solver->retcode = err;
return err;
}
if( solver->options[LIS_OPTIONS_INITGUESS_ZEROS] )
{
if( output ) lis_printf(A->comm,"initial vector x = 0\n");
#ifndef USE_QUAD_PRECISION
lis_vector_set_all(0.0,xx);
#else
if( precision==LIS_PRECISION_DOUBLE )
{
lis_vector_set_all(0.0,xx);
}
else
{
lis_vector_set_allex_nm(0.0,xx);
}
#endif
}
else
{
if( output ) lis_printf(A->comm,"initial vector x = user defined\n");
#ifndef USE_QUAD_PRECISION
lis_vector_copy(x,xx);
#else
if( precision==LIS_PRECISION_DOUBLE )
{
lis_vector_copy(x,xx);
}
else
{
lis_vector_copyex_nm(x,xx);
}
#endif
}
/* create residual history vector */
if( solver->residual ) lis_free(solver->residual);
residual = (LIS_SCALAR *)lis_malloc((maxiter+2)*sizeof(LIS_SCALAR),"lis_solve::residual");
if( residual==NULL )
{
LIS_SETERR_MEM((maxiter+2)*sizeof(LIS_SCALAR));
lis_vector_destroy(xx);
solver->retcode = err;
return err;
}
residual[0] = 1.0;
n = A->n;
np = A->np;
t = NULL;
At = NULL;
p_c_times = lis_wtime();
if( precon_type==LIS_PRECON_TYPE_IS )
{
if( solver->d==NULL )
{
err = lis_vector_duplicate(A,&solver->d);
if( err )
{
return err;
}
}
if( !A->is_scaled )
{
lis_matrix_scaling(A,b,solver->d,LIS_SCALE_JACOBI);
}
else if( !b->is_scaled )
{
#ifdef _OPENMP
#pragma omp parallel for
#endif
for(i=0;i<n;i++)
{
b->value[i] = b->value[i]*solver->d->value[i];
}
}
if( nsolver >= LIS_SOLVER_JACOBI && nsolver <= LIS_SOLVER_SOR )
{
solver->options[LIS_OPTIONS_ISLEVEL] = 0;
}
}
else if( nsolver >= LIS_SOLVER_JACOBI && nsolver <= LIS_SOLVER_SOR && precon_type!=LIS_PRECON_TYPE_NONE )
{
if( solver->d==NULL )
{
err = lis_vector_duplicate(A,&solver->d);
if( err )
{
return err;
}
}
if( !A->is_scaled )
{
lis_matrix_scaling(A,b,solver->d,LIS_SCALE_JACOBI);
}
}
else if( scale )
{
if( storage==LIS_MATRIX_BSR && scale==LIS_SCALE_JACOBI )
{
if( A->matrix_type!=LIS_MATRIX_BSR )
{
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_blocksize(B,block,block,NULL,NULL);
lis_matrix_set_type(B,storage);
err = lis_matrix_convert(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);
}
err = lis_matrix_split(A);
if( err ) return err;
err = lis_matrix_diag_duplicate(A->D,&solver->WD);
if( err ) return err;
lis_matrix_diag_copy(A->D,solver->WD);
lis_matrix_diag_inverse(solver->WD);
lis_matrix_bscaling_bsr(A,solver->WD);
lis_vector_duplicate(A,&t);
lis_matrix_diag_matvec(solver->WD,b,t);
lis_vector_copy(t,b);
lis_vector_destroy(t);
t = NULL;
}
else
{
if( solver->d==NULL )
{
err = lis_vector_duplicate(A,&solver->d);
if( err )
{
return err;
}
}
if( scale==LIS_SCALE_JACOBI && nsolver==LIS_SOLVER_CG )
{
scale = LIS_SCALE_SYMM_DIAG;
}
if( !A->is_scaled )
{
lis_matrix_scaling(A,b,solver->d,scale);
}
else if( !b->is_scaled )
{
#ifdef _OPENMP
#pragma omp parallel for
#endif
for(i=0;i<n;i++)
{
b->value[i] = b->value[i]*solver->d->value[i];
}
}
}
}
/* precon_type = precon->precon_type;*/
if( precon_type==LIS_PRECON_TYPE_IS )
{
if( nsolver < LIS_SOLVER_JACOBI || nsolver > LIS_SOLVER_SOR )
{
AA = solver->A;
bb = solver->b;
}
else
{
AA = precon->A;
bb = precon->Pb;
}
}
else
{
AA = A;
bb = b;
}
p_c_times = lis_wtime() - p_c_times;
itimes = lis_wtime();
/* Matrix Convert */
solver->A = AA;
solver->b = bb;
err = lis_matrix_convert_self(solver);
if( err )
{
lis_vector_destroy(xx);
lis_solver_work_destroy(solver);
lis_free(residual);
solver->retcode = err;
return err;
}
block = solver->A->bnr;
if( A->my_rank==0 )
{
if( output ) printf("precision : %s\n", lis_precisionname[precision]);
if( output ) printf("solver : %s %d\n", lis_solvername[nsolver],nsolver);
switch( precon_type )
{
case LIS_PRECON_TYPE_ILU:
i = solver->options[LIS_OPTIONS_FILL];
if( A->matrix_type==LIS_MATRIX_BSR || A->matrix_type==LIS_MATRIX_VBR )
{
if( output ) sprintf(buf,"Block %s(%d)",lis_preconname[precon_type],i);
}
else
{
if( output ) sprintf(buf,"%s(%d)",lis_preconname[precon_type],i);
}
break;
default:
if( output ) sprintf(buf,"%s",lis_preconname[precon_type]);
break;
}
if( solver->options[LIS_OPTIONS_ADDS] && precon_type )
{
if( output ) printf("precon : %s + additive schwarz\n", buf);
}
else
{
if( output ) printf("precon : %s\n", buf);
}
}
switch(conv_cond)
{
case LIS_CONV_COND_NRM2_R:
case LIS_CONV_COND_NRM2_B:
if( A->my_rank==0 )
{
if( output ) ("CONV_COND : ||r||_2 <= %6.1e*||r_0||_2\n", tol);
}
break;
case LIS_CONV_COND_NRM1_B:
lis_vector_nrm1(b,&nrm2);
nrm2 = nrm2*tol_w + tol;
if( A->my_rank==0 )
{
if( output ) printf("conv_cond : ||r||_1 <= %6.1e*||b||_1 + %6.1e = %6.1e\n", tol_w,tol,nrm2);
}
break;
}
if( A->my_rank==0 )
{
if( AA->matrix_type==LIS_MATRIX_BSR || AA->matrix_type==LIS_MATRIX_BSC )
{
if( output ) printf("storage : %s(%d x %d)\n", lis_storagename[AA->matrix_type-1],block,block);
}
else
{
if( output ) printf("storage : %s\n", lis_storagename[AA->matrix_type-1]);
}
}
/* create work vector */
err = lis_solver_malloc_work[nsolver](solver);
if( err )
{
lis_vector_destroy(xx);
lis_precon_destroy(precon);
solver->retcode = err;
return err;
}
if( nsolver==LIS_SOLVER_BICG && is_use_at )
{
if( output ) lis_printf(A->comm,"Use At\n");
lis_matrix_duplicate(AA,&At);
lis_matrix_set_type(At,LIS_USE_AT_TYPE[AA->matrix_type]);
lis_matrix_convert(AA,At);
solver->At = At;
}
solver->x = xx;
solver->xx = x;
solver->precon = precon;
solver->residual = residual;
/* execute solver */
#ifndef USE_QUAD_PRECISION
err = lis_solver_execute[nsolver](solver);
#else
if( precision==LIS_PRECISION_DOUBLE )
{
err = lis_solver_execute[nsolver](solver);
}
else if( precision==LIS_PRECISION_QUAD )
{
err = lis_solver_execute_quad[nsolver](solver);
}
else if( precision==LIS_PRECISION_SWITCH )
{
err = lis_solver_execute_switch[nsolver](solver);
}
#endif
solver->retcode = err;
if( scale==LIS_SCALE_SYMM_DIAG && precon_type!=LIS_PRECON_TYPE_IS)
{
#ifdef _OPENMP
#pragma omp parallel for
#endif
for(i=0;i<n;i++)
{
x->value[i] = xx->value[i]*solver->d->value[i];
}
}
else
{
#ifndef USE_QUAD_PRECISION
lis_vector_copy(xx,x);
#else
if( precision==LIS_PRECISION_DOUBLE )
{
lis_vector_copy(xx,x);
}
else
{
lis_vector_copyex_mn(xx,x);
}
#endif
}
itimes = lis_wtime() - itimes - solver->ptimes;
p_i_times = solver->ptimes;
solver->ptimes = p_c_times + p_i_times;
solver->p_c_times = p_c_times;
solver->p_i_times = p_i_times;
solver->times = solver->ptimes + itimes;
solver->itimes = itimes;
lis_solver_work_destroy(solver);
lis_vector_duplicate(A,&t);
xx->precision = LIS_PRECISION_DEFAULT;
lis_matvec(A,xx,t);
lis_vector_xpay(b,-1.0,t);
if( scale==LIS_SCALE_SYMM_DIAG && precon_type!=LIS_PRECON_TYPE_IS)
{
#ifdef _OPENMP
#pragma omp parallel for
#endif
for(i=0;i<n;i++)
{
t->value[i] = t->value[i]/solver->d->value[i];
}
}
lis_vector_nrm2(t,&nrm2);
/*
solver->resid = nrm2;
*/
if( A->my_rank==0 )
{
if( err )
{
if( output ) printf("lis_solve : %s(code=%d)\n\n",lis_returncode[err],err);
}
else
{
if( output ) printf("lis_solve : normal end\n\n");
}
}
if( precision==LIS_PRECISION_DOUBLE )
{
solver->iter2 = solver->iter;
}
else if( precision==LIS_PRECISION_QUAD )
{
solver->iter2 = 0;
}
lis_vector_destroy(t);
/* lis_vector_destroy(d);*/
lis_vector_destroy(xx);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
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_precon_create_saamg(LIS_SOLVER solver, LIS_PRECON precon)
{
#if defined(USE_SAAMG)
LIS_MATRIX A,B;
LIS_COMMTABLE table;
LIS_INT i;
LIS_INT unsym,sol;
LIS_INT err;
LIS_REAL theta;
#ifdef USE_MPI
LIS_MPI_Fint comm;
#endif
LIS_DEBUG_FUNC_IN;
if( solver->A->matrix_type!=LIS_MATRIX_CRS )
{
A = solver->A;
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_type(B,LIS_MATRIX_CRS);
err = lis_matrix_convert(A,B);
if( err ) return err;
solver->A = B;
lis_matrix_destroy(B);
solver->A = A;
}
precon->A = solver->A;
precon->is_copy = LIS_FALSE;
A = precon->A;
sol = solver->options[LIS_OPTIONS_SOLVER];
unsym = solver->options[LIS_OPTIONS_SAAMG_UNSYM];
theta = solver->params[LIS_PARAMS_SAAMG_THETA - LIS_OPTIONS_LEN];
#if 0
if( sol!=LIS_SOLVER_CG && !unsym )
{
unsym = LIS_TRUE;
}
#endif
err = lis_vector_duplicate(A,&precon->temp);
if( err )
{
return err;
}
F77_FUNC_(finit_data_creation,FINIT_DATA_CREATION)(c_data_creation_ptr_bar);
F77_FUNC_(finit_data_creation_unsym,FINIT_DATA_CREATION_UNSYM)(c_data_creation_unsym_ptr_bar);
F77_FUNC_(finit_v_cycle,FINIT_V_CYCLE)(c_v_cycle_ptr_bar);
F77_FUNC_(finit_clear_matrix,FINIT_CLEAR_MATRIX)(c_clear_matrix_ptr_bar);
lis_matrix_split(A);
#ifdef USE_MPI
comm = MPI_Comm_c2f(A->comm);
lis_send_recv(A->commtable,A->D->value);
table = A->commtable;
if( !unsym )
{
(*(void (*)())f_data_creation_ptr)(&A->n,&A->np,&A->L->nnz,&A->U->nnz,
A->D->value,A->L->value,A->L->ptr,A->L->index,
A->U->value, A->U->ptr, A->U->index,
&table->neibpetot, table->neibpe, table->import_ptr,
table->import_index, table->export_ptr, table->export_index,
&table->imnnz,&table->exnnz,
&comm, &precon->level_num,&precon->wsize, &theta);
}
else
{
(*(void (*)())f_data_creation_unsym_ptr)(&A->n,&A->np,&A->L->nnz,&A->U->nnz,
A->D->value,A->L->value,A->L->ptr,A->L->index,
A->U->value, A->U->ptr, A->U->index,
&table->neibpetot, table->neibpe, table->import_ptr,
table->import_index, table->export_ptr, table->export_index,
&table->imnnz,&table->exnnz,
&comm, &precon->level_num,&precon->wsize, &theta);
}
#else
if( !unsym )
{
(*(void (*)())f_data_creation_ptr)(&A->n,&A->L->nnz,&A->U->nnz,
A->D->value,A->L->value,A->L->ptr,A->L->index,
A->U->value, A->U->ptr, A->U->index, &precon->level_num, &theta);
}
else
{
(*(void (*)())f_data_creation_unsym_ptr)(&A->n,&A->L->nnz,&A->U->nnz,
A->D->value,A->L->value,A->L->ptr,A->L->index,
A->U->value, A->U->ptr, A->U->index, &precon->level_num, &theta);
}
#endif
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
#else
LIS_DEBUG_FUNC_IN;
precon->A = solver->A;
precon->is_copy = LIS_FALSE;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
#endif
}
LIS_INT main(LIS_INT argc, char* argv[])
{
LIS_MATRIX A,A0;
LIS_VECTOR b,x;
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nthreads, maxthreads;
LIS_INT nnz;
LIS_INT i,n,np;
LIS_INT block;
LIS_INT is,ie;
LIS_INT err,iter,matrix_type;
double time,time2,nnzs,nnzap,nnzt;
LIS_SCALAR val;
double commtime,comptime,flops;
char path[1024];
FILE *file;
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 matrix_filename_list iter [block] \n", argv[0]);
}
lis_finalize();
exit(0);
}
file = fopen(argv[1], "r");
if( file==NULL ) CHKERR(1);
iter = atoi(argv[2]);
if (argv[3] == NULL) {
block = 2;
}
else {
block = atoi(argv[3]);
}
if( iter<=0 )
{
#ifdef _LONG__LONG
printf("iter=%lld <= 0\n",iter);
#else
printf("iter=%d <= 0\n",iter);
#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
if( my_rank==0 )
{
nthreads = omp_get_num_procs();
maxthreads = omp_get_max_threads();
#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 */
while( fscanf(file, "%s\n", path)==1 )
{
if( my_rank==0 )
{
printf("matrix_filename = %s\n", path);
}
lis_matrix_create(LIS_COMM_WORLD,&A0);
err = lis_input(A0,NULL,NULL,path);
if( err ) CHKERR(err);
n = A0->n;
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("block size of BSR and BSC = %lld x %lld\n",block,block);
printf("number of iterations = %lld\n\n",iter);
#else
printf("block size of BSR and BSC = %d x %d\n",block,block);
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);
}
/*
MPI version of VBR is not implemented.
DNS is also excluded to reduce memory usage.
*/
for (matrix_type=1;matrix_type<11;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);
if( my_rank==0 )
{
if( A->matrix_type==LIS_MATRIX_BSR || A->matrix_type==LIS_MATRIX_BSC )
{
A->bnr = block;
A->bnc = block;
}
}
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);
}
fclose(file);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
LIS_INT lis_esolve(LIS_MATRIX A, LIS_VECTOR x, LIS_SCALAR *evalue0, LIS_ESOLVER esolver)
{
LIS_INT nesolver,niesolver,emaxiter;
LIS_SCALAR *evalue;
LIS_VECTOR *evector;
LIS_SCALAR *resid;
LIS_SCALAR *rhistory;
LIS_INT *iter,*iter2;
LIS_INT err;
LIS_INT output;
LIS_INT ss, mode;
double time;
double gshift;
LIS_INT estorage,eblock;
LIS_MATRIX B;
LIS_INT eprecision;
LIS_VECTOR xx;
LIS_DEBUG_FUNC_IN;
/* begin parameter check */
err = lis_matrix_check(A,LIS_MATRIX_CHECK_ALL);
if( err ) return err;
if( x==NULL )
{
LIS_SETERR(LIS_ERR_ILL_ARG,"vector x is undefined\n");
return LIS_ERR_ILL_ARG;
}
if( A->n!=x->n )
{
return LIS_ERR_ILL_ARG;
}
if( A->gn<=0 )
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Size n(=%d) of matrix A is less than 0\n",A->gn);
return LIS_ERR_ILL_ARG;
}
nesolver = esolver->options[LIS_EOPTIONS_ESOLVER];
niesolver = esolver->options[LIS_EOPTIONS_INNER_ESOLVER];
ss = esolver->options[LIS_EOPTIONS_SUBSPACE];
mode = esolver->options[LIS_EOPTIONS_MODE];
emaxiter = esolver->options[LIS_EOPTIONS_MAXITER];
gshift = esolver->params[LIS_EPARAMS_SHIFT - LIS_EOPTIONS_LEN];
output = esolver->options[LIS_EOPTIONS_OUTPUT];
estorage = esolver->options[LIS_EOPTIONS_STORAGE];
eblock = esolver->options[LIS_EOPTIONS_STORAGE_BLOCK];
eprecision = esolver->options[LIS_EOPTIONS_PRECISION];
esolver->eprecision = eprecision;
if( nesolver < 1 || nesolver > LIS_ESOLVER_LEN )
{
LIS_SETERR2(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_ESOLVER is %d (Set between 1 to %d)\n",nesolver, LIS_ESOLVER_LEN);
return LIS_ERR_ILL_ARG;
}
if( niesolver < 1 || niesolver > 7 )
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_INNER_ESOLVER is %d (Set between 1 to 7)\n", niesolver);
return LIS_ERR_ILL_ARG;
}
if ( esolver->options[LIS_EOPTIONS_ESOLVER] == LIS_ESOLVER_SI && niesolver > 4 )
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_INNER_ESOLVER is %d (Set between 1 to 4 for Subspace)\n", niesolver);
return LIS_ERR_ILL_ARG;
}
if ( esolver->options[LIS_EOPTIONS_ESOLVER] == LIS_ESOLVER_LI && niesolver == LIS_ESOLVER_PI )
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_INNER_ESOLVER is %d (Set between 2 to 7 for Lanczos)\n", niesolver);
return LIS_ERR_ILL_ARG;
}
if ( esolver->options[LIS_EOPTIONS_ESOLVER] == LIS_ESOLVER_AI && (( niesolver == LIS_ESOLVER_PI ) || ( niesolver == LIS_ESOLVER_CG) || ( niesolver == LIS_ESOLVER_JD)) )
{
LIS_SETERR1(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_INNER_ESOLVER is %d (Set between 2 to 4 or 6 for Arnoldi)\n", niesolver);
return LIS_ERR_ILL_ARG;
}
if ( esolver->options[LIS_EOPTIONS_ESOLVER] == LIS_ESOLVER_SI && ss > A->gn )
{
LIS_SETERR2(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_SUBSPACE is %d (Set less than or equal to matrix size %d for Subspace)\n", ss, A->gn);
return LIS_ERR_ILL_ARG;
}
if (( esolver->options[LIS_EOPTIONS_ESOLVER] == LIS_ESOLVER_LI || esolver->options[LIS_EOPTIONS_ESOLVER] == LIS_ESOLVER_AI ) && ss > A->gn )
{
LIS_SETERR2(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_SUBSPACE is %d (Set less than or equal to matrix size %d for Lanczos and Arnoldi)\n", ss, A->gn);
return LIS_ERR_ILL_ARG;
}
if ( esolver->options[LIS_EOPTIONS_ESOLVER] == LIS_ESOLVER_SI && mode >= ss )
{
LIS_SETERR2(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_MODE is %d (Set less than subspace size %d for Subspace)\n", mode, ss);
return LIS_ERR_ILL_ARG;
}
if ( esolver->options[LIS_EOPTIONS_ESOLVER] == ( LIS_ESOLVER_LI || LIS_ESOLVER_AI ) && mode >= ss )
{
LIS_SETERR2(LIS_ERR_ILL_ARG,"Parameter LIS_EOPTIONS_MODE is %d (Set less than subspace size %d for Lanczos or Arnoldi)\n", mode, ss);
return LIS_ERR_ILL_ARG;
}
#ifdef USE_QUAD_PRECISION
if( eprecision==LIS_PRECISION_QUAD && lis_esolver_execute_quad[nesolver]==NULL )
{
LIS_SETERR1(LIS_ERR_NOT_IMPLEMENTED,"Quad precision eigensolver %s is not implemented\n",lis_esolvername[nesolver]);
return LIS_ERR_NOT_IMPLEMENTED;
}
else if( eprecision==LIS_PRECISION_SWITCH && lis_esolver_execute_switch[nesolver]==NULL )
{
LIS_SETERR1(LIS_ERR_NOT_IMPLEMENTED,"Switch esolver %s is not implemented\n",lis_esolvername[nesolver]);
return LIS_ERR_NOT_IMPLEMENTED;
}
if( esolver->options[LIS_EOPTIONS_SWITCH_MAXITER]==-1 )
{
esolver->options[LIS_EOPTIONS_SWITCH_MAXITER] = emaxiter;
}
#endif
/* create eigenvalue array */
if( esolver->evalue ) lis_free(esolver->evalue);
evalue = (LIS_SCALAR *)lis_malloc((ss+2)*sizeof(LIS_SCALAR),"lis_esolve::evalue");
if( evalue==NULL )
{
LIS_SETERR_MEM((ss+2)*sizeof(LIS_SCALAR));
esolver->retcode = err;
return err;
}
evalue[0] = 1.0;
evalue[ss-1] = 1.0;
/* create residual norm array */
if( esolver->resid ) lis_free(esolver->resid);
resid = (LIS_SCALAR *)lis_malloc((ss+2)*sizeof(LIS_SCALAR),"lis_esolve::resid");
if( resid==NULL )
{
LIS_SETERR_MEM((ss+2)*sizeof(LIS_SCALAR));
esolver->retcode = err;
return err;
}
/* create number of iterations array */
if( esolver->iter ) lis_free(esolver->iter);
iter = (LIS_INT *)lis_malloc((ss+2)*sizeof(LIS_SCALAR),"lis_esolve::iter");
if( iter==NULL )
{
LIS_SETERR_MEM((ss+2)*sizeof(LIS_SCALAR));
esolver->retcode = err;
return err;
}
/* create quad precision number of iterations array */
if( esolver->iter2 ) lis_free(esolver->iter2);
iter2 = (LIS_INT *)lis_malloc((ss+2)*sizeof(LIS_SCALAR),"lis_esolve::iter2");
if( iter2==NULL )
{
LIS_SETERR_MEM((ss+2)*sizeof(LIS_SCALAR));
esolver->retcode = err;
return err;
}
/* create initial vector */
#ifndef USE_QUAD_PRECISION
err = lis_vector_duplicate(A,&xx);
#else
if( eprecision==LIS_PRECISION_DOUBLE )
{
err = lis_vector_duplicate(A,&xx);
}
else
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,A,&xx);
}
#endif
if( err )
{
esolver->retcode = err;
return err;
}
if( esolver->options[LIS_EOPTIONS_INITGUESS_ONES] )
{
if( output ) lis_printf(A->comm,"initial vector x : 1\n");
#ifndef USE_QUAD_PRECISION
lis_vector_set_all(1.0,xx);
#else
if( eprecision==LIS_PRECISION_DOUBLE )
{
lis_vector_set_all(1.0,xx);
}
else
{
lis_vector_set_allex_nm(1.0,xx);
}
#endif
}
else
{
if( output ) lis_printf(A->comm,"initial vector x : user defined\n");
#ifndef USE_QUAD_PRECISION
lis_vector_copy(x,xx);
#else
if( eprecision==LIS_PRECISION_DOUBLE )
{
lis_vector_copy(x,xx);
}
else
{
lis_vector_copyex_nm(x,xx);
}
#endif
}
/* global shift */
if ( output ) if( A->my_rank==0 ) printf("shift : %e\n", gshift);
/* create eigenvector array */
if( esolver->evector ) lis_free(esolver->evector);
evector = (LIS_VECTOR *)lis_malloc((ss+2)*sizeof(LIS_VECTOR),"lis_esolve::evector");
if( evector==NULL )
{
LIS_SETERR_MEM((ss+2)*sizeof(LIS_VECTOR));
esolver->retcode = err;
return err;
}
/* create residual history array */
if( esolver->rhistory ) lis_free(esolver->rhistory);
rhistory = (LIS_SCALAR *)lis_malloc((emaxiter+2)*sizeof(LIS_SCALAR),"lis_esolve::rhistory");
if( rhistory==NULL )
{
LIS_SETERR_MEM((emaxiter+2)*sizeof(LIS_SCALAR));
lis_vector_destroy(xx);
esolver->retcode = err;
return err;
}
/* convert matrix */
if( estorage>0 && A->matrix_type!=estorage )
{
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
lis_matrix_set_blocksize(B,eblock,eblock,NULL,NULL);
lis_matrix_set_type(B,estorage);
err = lis_matrix_convert(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);
}
esolver->A = A;
esolver->evalue = evalue;
esolver->x = x;
esolver->evector = evector;
rhistory[0] = 1.0;
esolver->rhistory = rhistory;
esolver->resid = resid;
esolver->iter = iter;
esolver->iter2 = iter2;
if( A->my_rank==0 )
{
#ifdef _LONG__DOUBLE
if ( output ) printf("precision : long double\n");
#else
if ( output ) printf("precision : %s\n", lis_eprecisionname[eprecision]);
#endif
#ifdef _LONG__LONG
if ( output ) printf("eigensolver : %s\n", lis_esolvername[nesolver]);
#else
if ( output ) printf("eigensolver : %s\n", lis_esolvername[nesolver]);
#endif
}
if( A->my_rank==0 )
{
#ifdef _LONG__DOUBLE
if ( output ) printf("convergence condition : ||lx-Ax||_2 <= %6.1Le * ||lx||_2\n", esolver->params[LIS_EPARAMS_RESID - LIS_EOPTIONS_LEN]);
#else
if ( output ) printf("convergence condition : ||lx-Ax||_2 <= %6.1e * ||lx||_2\n", esolver->params[LIS_EPARAMS_RESID - LIS_EOPTIONS_LEN]);
#endif
}
if( A->my_rank==0 )
{
if( A->matrix_type==LIS_MATRIX_BSR || A->matrix_type==LIS_MATRIX_BSC )
{
#ifdef _LONG__LONG
if ( output ) printf("matrix storage format : %s(%lld x %lld)\n", lis_estoragename[A->matrix_type-1],eblock,eblock);
#else
if ( output ) printf("matrix storage format : %s(%d x %d)\n", lis_estoragename[A->matrix_type-1],eblock,eblock);
#endif
}
else
{
if ( output ) printf("matrix storage format : %s\n", lis_estoragename[A->matrix_type-1]);
}
}
time = lis_wtime();
esolver->ptime = 0;
esolver->itime = 0;
esolver->p_c_time = 0;
esolver->p_i_time = 0;
if (gshift != 0.0) lis_matrix_shift_diagonal(A, gshift);
/* create work vector */
err = lis_esolver_malloc_work[nesolver](esolver);
if( err )
{
lis_vector_destroy(xx);
esolver->retcode = err;
return err;
}
esolver->x = xx;
esolver->xx = x;
/* execute esolver */
#ifndef USE_QUAD_PRECISION
err = lis_esolver_execute[nesolver](esolver);
#else
if( eprecision==LIS_PRECISION_DOUBLE )
{
err = lis_esolver_execute[nesolver](esolver);
}
else if( eprecision==LIS_PRECISION_QUAD )
{
err = lis_esolver_execute_quad[nesolver](esolver);
}
else if( eprecision==LIS_PRECISION_SWITCH )
{
err = lis_esolver_execute_switch[nesolver](esolver);
}
#endif
esolver->retcode = err;
*evalue0 = esolver->evalue[0];
lis_vector_copy(esolver->x, x);
esolver->time = lis_wtime() - time;
lis_matrix_shift_diagonal(A, -gshift);
if( A->my_rank==0 )
{
if( err )
{
#ifdef _LONG__LONG
if ( output ) printf("eigensolver status : %s(code=%lld)\n\n",lis_ereturncode[err],err);
#else
if ( output ) printf("eigensolver status : %s(code=%d)\n\n",lis_ereturncode[err],err);
#endif
}
else
{
if ( output ) printf("eigensolver status : normal end\n\n");
}
}
if( eprecision==LIS_PRECISION_DOUBLE )
{
esolver->iter2[mode] = esolver->iter[mode];
}
else if( eprecision==LIS_PRECISION_QUAD )
{
esolver->iter2[mode] = 0;
}
lis_vector_destroy(xx);
LIS_DEBUG_FUNC_OUT;
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,j,jj,j0,j1,l,k,n,np,h,ih;
LIS_INT m,nn,ii;
LIS_INT block;
LIS_INT rn,rmin,rmax,rb;
LIS_INT is,ie,clsize,ci,*iw;
LIS_INT err,iter,matrix_type;
LIS_INT *ptr,*index;
double mem,val,ra,rs,ri,ria,ca,time,time2,convtime,val2,nnzs,nnzap,nnzt;
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 < 4 )
{
if( my_rank==0 ) printf("Usage: spmvtest5 matrix_filename matrix_type iter [block] \n");
lis_finalize();
exit(0);
}
file = fopen(argv[1], "r");
if( file==NULL ) CHKERR(1);
matrix_type = atoi(argv[2]);
iter = atoi(argv[3]);
if (argv[4] == NULL) {
block = 2;
}
else {
block = atoi(argv[4]);
}
if( matrix_type<1 || matrix_type>11 )
{
if( my_rank==0 ) printf("matrix_type=%d <1 or matrix_type=%d >11\n",matrix_type,matrix_type);
CHKERR(1);
}
if( iter<=0 )
{
if( my_rank==0 ) printf("iter=%d <= 0\n",iter);
CHKERR(1);
}
if( my_rank==0 )
{
printf("\n");
printf("number of processes = %d\n",nprocs);
}
#ifdef _OPENMP
if( my_rank==0 )
{
nthreads = omp_get_num_procs();
maxthreads = omp_get_max_threads();
printf("max number of threads = %d\n", nthreads);
printf("number of threads = %d\n", maxthreads);
}
#else
nthreads = 1;
maxthreads = 1;
#endif
/* create matrix and vectors */
lis_matrix_create(LIS_COMM_WORLD,&A0);
err = lis_input(A0,NULL,NULL,argv[1]);
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 )
{
printf("block size of BSR and BSC = %d x %d\n",block,block);
printf("iteration count = %d\n\n",iter);
}
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);
}
lis_matrix_duplicate(A0,&A);
lis_matrix_set_type(A,matrix_type);
err = lis_matrix_convert(A0,A);
if( err ) CHKERR(err);
if( A->matrix_type==LIS_MATRIX_BSR || A->matrix_type==LIS_MATRIX_BSC )
{
A->bnr = block;
A->bnc = block;
}
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;
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);
}
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 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;
}
