int lis_precon_create_jacobi(LIS_SOLVER solver, LIS_PRECON precon)
{
int err;
LIS_DEBUG_FUNC_IN;
if( solver->precision==LIS_PRECISION_DEFAULT )
{
err = lis_vector_duplicate(solver->A, &precon->D);
}
else
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,solver->A, &precon->D);
}
if( err )
{
return err;
}
lis_matrix_get_diagonal(solver->A, precon->D);
lis_vector_reciprocal(precon->D);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
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_precon_create_adds(LIS_SOLVER solver, LIS_PRECON precon)
{
LIS_INT i,j;
LIS_INT precon_type,worklen;
LIS_INT err;
LIS_VECTOR *work;
LIS_DEBUG_FUNC_IN;
precon_type = solver->options[LIS_OPTIONS_PRECON];
worklen = 2;
work = (LIS_VECTOR *)lis_malloc( worklen*sizeof(LIS_VECTOR),"lis_precon_create_adds::work" );
if( work==NULL )
{
LIS_SETERR_MEM(worklen*sizeof(LIS_VECTOR));
return LIS_OUT_OF_MEMORY;
}
if( solver->precision==LIS_PRECISION_DEFAULT )
{
for(i=0;i<worklen;i++)
{
err = lis_vector_duplicate(solver->A,&work[i]);
if( err ) break;
}
}
else
{
for(i=0;i<worklen;i++)
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,solver->A,&work[i]);
if( err ) break;
}
}
if( i<worklen )
{
for(j=0;j<i;j++) lis_vector_destroy(work[j]);
lis_free(work);
return err;
}
precon->worklen = worklen;
precon->work = work;
err = lis_precon_create_xxx[precon_type](solver,precon);
if( err )
{
lis_precon_destroy(precon);
return err;
}
precon->A = solver->A;
precon->is_copy = LIS_FALSE;
LIS_DEBUG_FUNC_OUT;
return err;
}
LIS_INT lis_gmres_malloc_work(LIS_SOLVER solver)
{
LIS_VECTOR *work;
LIS_INT i,j,restart,worklen,err;
LIS_DEBUG_FUNC_IN;
restart = solver->options[LIS_OPTIONS_RESTART];
worklen = NWORK + (restart+1);
work = (LIS_VECTOR *)lis_malloc( worklen*sizeof(LIS_VECTOR),"lis_gmres_malloc_work::work" );
if( work==NULL )
{
LIS_SETERR_MEM(worklen*sizeof(LIS_VECTOR));
return LIS_ERR_OUT_OF_MEMORY;
}
if( solver->precision==LIS_PRECISION_DEFAULT )
{
for(i=1;i<worklen;i++)
{
err = lis_vector_duplicate(solver->A,&work[i]);
if( err ) break;
}
}
else
{
for(i=1;i<worklen;i++)
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,solver->A,&work[i]);
if( err ) break;
memset(work[i]->value_lo,0,solver->A->np*sizeof(LIS_SCALAR));
}
}
if( i<worklen )
{
for(j=1;j<i;j++) lis_vector_destroy(work[j]);
lis_free(work);
return err;
}
if( solver->precision==LIS_PRECISION_DEFAULT )
{
lis_vector_create(solver->A->comm,&work[0]);
}
else
{
lis_vector_createex(LIS_PRECISION_QUAD,solver->A->comm,&work[0]);
}
lis_vector_set_size(work[0],restart+1,0);
solver->worklen = worklen;
solver->work = work;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LisMatrix::LisMatrix(std::size_t n_rows, LisOption::MatrixType mat_type)
: _n_rows(n_rows), _mat_type(mat_type), _is_assembled(false)
{
int ierr = lis_matrix_create(0, &_AA);
checkLisError(ierr);
ierr = lis_matrix_set_size(_AA, 0, n_rows);
checkLisError(ierr);
lis_matrix_get_range(_AA, &_is, &_ie);
ierr = lis_vector_duplicate(_AA, &_diag);
checkLisError(ierr);
}
LIS_INT lis_idrs_malloc_work(LIS_SOLVER solver)
{
LIS_VECTOR *work;
LIS_INT i,j,s,worklen,err;
LIS_DEBUG_FUNC_IN;
/*
err = lis_matrix_convert(solver->A,&solver->At,LIS_MATRIX_CCS);
if( err ) return err;
*/
s = solver->options[LIS_OPTIONS_IDRS_RESTART];
worklen = NWORK + 3*s;
work = (LIS_VECTOR *)lis_malloc(
worklen*sizeof(LIS_VECTOR),"lis_idrs_malloc_work::work" );
if( work==NULL )
{
LIS_SETERR_MEM(worklen*sizeof(LIS_VECTOR));
return LIS_ERR_OUT_OF_MEMORY;
}
if( solver->precision==LIS_PRECISION_DEFAULT )
{
for(i=0;i<worklen;i++)
{
err = lis_vector_duplicate(solver->A,&work[i]);
if( err ) break;
}
}
else
{
for(i=0;i<worklen;i++)
{
err =
lis_vector_duplicateex(LIS_PRECISION_QUAD,solver->A,&work[i]);
if( err ) break;
memset(work[i]->value_lo,0,solver->A->np*sizeof(LIS_SCALAR));
}
}
if( i<worklen )
{
for(j=0;j<i;j++) lis_vector_destroy(work[j]);
lis_free(work);
return err;
}
solver->worklen = worklen;
solver->work = work;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
void lis_vector_duplicate_f(LIS_VECTOR_F *vin, LIS_VECTOR_F *vout, LIS_INT *ierr)
{
LIS_VECTOR v;
LIS_DEBUG_FUNC_IN;
*ierr = lis_vector_duplicate((LIS_VECTOR)LIS_V2P(vin),&v);
if( *ierr ) return;
*vout = LIS_P2V(v);
LIS_DEBUG_FUNC_OUT;
return;
}
LIS_INT lis_esi_malloc_work(LIS_ESOLVER esolver)
{
LIS_VECTOR *work;
LIS_INT i,j,worklen,err,ss;
LIS_DEBUG_FUNC_IN;
ss = esolver->options[LIS_EOPTIONS_SUBSPACE];
worklen = NWORK + ss;
work = (LIS_VECTOR *)lis_malloc( worklen*sizeof(LIS_VECTOR),"lis_esi_malloc_work::work" );
if( work==NULL )
{
LIS_SETERR_MEM(worklen*sizeof(LIS_VECTOR));
return LIS_ERR_OUT_OF_MEMORY;
}
if( esolver->eprecision==LIS_PRECISION_DEFAULT )
{
for(i=0;i<worklen;i++)
{
err = lis_vector_duplicate(esolver->A,&work[i]);
if( err ) break;
}
}
else
{
for(i=0;i<worklen;i++)
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,esolver->A,&work[i]);
if( err ) break;
}
}
if( i<worklen )
{
for(j=0;j<i;j++) lis_vector_destroy(work[j]);
lis_free(work);
return err;
}
esolver->worklen = worklen;
esolver->work = work;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_cgs_malloc_work(LIS_SOLVER solver)
{
LIS_VECTOR *work;
LIS_INT i,j,worklen,err;
LIS_DEBUG_FUNC_IN;
worklen = NWORK;
work = (LIS_VECTOR *)lis_malloc( worklen*sizeof(LIS_VECTOR),"lis_cgs_malloc_work::work" );
if( work==NULL )
{
LIS_SETERR_MEM(worklen*sizeof(LIS_VECTOR));
return LIS_ERR_OUT_OF_MEMORY;
}
if( solver->precision==LIS_PRECISION_DEFAULT )
{
for(i=0;i<worklen;i++)
{
err = lis_vector_duplicate(solver->A,&work[i]);
if( err ) break;
}
}
else
{
for(i=0;i<worklen;i++)
{
err = lis_vector_duplicateex(LIS_PRECISION_QUAD,solver->A,&work[i]);
if( err ) break;
memset(work[i]->value_lo,0,solver->A->np*sizeof(LIS_SCALAR));
}
}
if( i<worklen )
{
for(j=0;j<i;j++) lis_vector_destroy(work[j]);
lis_free(work);
return err;
}
solver->worklen = worklen;
solver->work = work;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
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_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 lis_precon_create_sainv_csr(LIS_SOLVER solver, LIS_PRECON precon)
{
LIS_INT err;
LIS_INT i,j,k,ii,jj,ik,jk;
LIS_INT n,annz,cl,cu;
LIS_INT *ww,*il,*iu;
LIS_SCALAR t,dd;
LIS_REAL tol,nrm;
LIS_SCALAR *d,*l,*u;
LIS_MATRIX A,B;
LIS_MATRIX_ILU W,Z;
LIS_VECTOR D;
LIS_DEBUG_FUNC_IN;
A = solver->A;
n = A->n;
tol = solver->params[LIS_PARAMS_DROP-LIS_OPTIONS_LEN];
annz = A->n / 10;
W = NULL;
ww = NULL;
d = NULL;
l = NULL;
u = NULL;
il = NULL;
iu = NULL;
err = lis_matrix_ilu_create(n,1,&W);
if( err ) return err;
err = lis_matrix_ilu_create(n,1,&Z);
if( err ) return err;
err = lis_matrix_ilu_setCR(W);
if( err ) return err;
err = lis_matrix_ilu_setCR(Z);
if( err ) return err;
err = lis_vector_duplicate(A,&D);
if( err ) return err;
d = D->value;
err = lis_matrix_ilu_premalloc(annz,W);
if( err ) return err;
err = lis_matrix_ilu_premalloc(annz,Z);
if( err ) return err;
l = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_sainv_csr::l");
if( l==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
u = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_sainv_csr::u");
if( u==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
il = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::il");
if( il==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
iu = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::iu");
if( iu==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
ww = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::ww");
if( ww==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
err = lis_matrix_convert_csr2csc(A,B);
if( err )
{
return err;
}
for(i=0;i<n;i++) ww[i] = 0;
for(i=0;i<n;i++)
{
W->value[i][0] = 1.0;
W->index[i][0] = i;
W->nnz[i] = 1;
Z->value[i][0] = 1.0;
Z->index[i][0] = i;
Z->nnz[i] = 1;
}
for(i=0; i<n; i++)
{
/* nrm_inf(A[i,:]) */
nrm = 0.0;
for(j=A->ptr[i];j<A->ptr[i+1];j++)
{
nrm = _max(nrm,fabs(A->value[j]));
}
nrm = 1.0/nrm;
/* l = AZ_i */
cl = 0;
memset(l,0,n*sizeof(LIS_SCALAR));
for(k=0;k<Z->nnz[i];k++)
{
ii = Z->index[i][k];
for(j=B->ptr[ii];j<B->ptr[ii+1];j++)
{
jj = B->index[j];
if( jj>i )
{
l[jj] += B->value[j]*Z->value[i][k];
if( ww[jj]==0 )
{
ww[jj] = 1;
il[cl++] = jj;
}
}
}
}
for(k=0;k<cl;k++) ww[il[k]] = 0;
/* u = W_i'A */
cu = 0;
memset(u,0,n*sizeof(LIS_SCALAR));
for(k=0;k<W->nnz[i];k++)
{
ii = W->index[i][k];
for(j=A->ptr[ii];j<A->ptr[ii+1];j++)
{
jj = A->index[j];
#ifdef USE_MPI
if( jj>n-1 ) continue;
#endif
u[jj] += A->value[j]*W->value[i][k];
if( jj>i && ww[jj]==0 )
{
ww[jj] = 1;
iu[cu++] = jj;
}
}
}
for(k=0;k<cu;k++) ww[iu[k]] = 0;
/* d_ii = uZ_i or W_i'l */
t = 0.0;
for(k=0;k<Z->nnz[i];k++)
{
t += u[Z->index[i][k]]*Z->value[i][k];
}
d[i] = 1.0/t;
/* for j>i, l_j!=0 */
/* w_j = w_j - (l_j/d_ii)*w_i */
for(jj=0;jj<cl;jj++)
{
j = il[jj];
dd = l[j]*d[i];
for(k=0;k<W->nnz[j];k++)
{
ww[W->index[j][k]] = k+1;
}
for(ik=0;ik<W->nnz[i];ik++)
{
jk = ww[W->index[i][ik]];
if( jk!=0 )
{
t = dd*W->value[i][ik];
if( fabs(t)*nrm > tol )
{
W->value[j][jk-1] -= t;
}
}
else
{
t = dd*W->value[i][ik];
if( fabs(t)*nrm > tol )
{
if( W->nnz[j] == W->nnz_ma[j] )
{
W->nnz_ma[j] += annz;
err = lis_matrix_ilu_realloc(j,W->nnz_ma[j],W);
if( err ) return err;
}
jk = W->nnz[j];
W->index[j][jk] = W->index[i][ik];
W->value[j][jk] = -t;
W->nnz[j]++;
}
}
}
for(k=0;k<W->nnz[j];k++)
{
ww[W->index[j][k]] = 0;
}
}
/* for j>i, u_j!=0 */
/* z_j = z_j - (u_j/d_ii)*z_i */
for(jj=0;jj<cu;jj++)
{
j = iu[jj];
dd = u[j]*d[i];
for(k=0;k<Z->nnz[j];k++)
{
ww[Z->index[j][k]] = k+1;
}
for(ik=0;ik<Z->nnz[i];ik++)
{
jk = ww[Z->index[i][ik]];
if( jk!=0 )
{
t = dd*Z->value[i][ik];
if( fabs(t)*nrm > tol )
{
Z->value[j][jk-1] -= t;
}
}
else
{
t = dd*Z->value[i][ik];
if( fabs(t)*nrm > tol )
{
if( Z->nnz[j] == Z->nnz_ma[j] )
{
Z->nnz_ma[j] += annz;
err = lis_matrix_ilu_realloc(j,Z->nnz_ma[j],Z);
if( err ) return err;
}
jk = Z->nnz[j];
Z->index[j][jk] = Z->index[i][ik];
Z->value[j][jk] = -t;
Z->nnz[j]++;
}
}
}
for(k=0;k<Z->nnz[j];k++)
{
ww[Z->index[j][k]] = 0;
}
}
}
lis_matrix_destroy(B);
lis_free2(5,l,u,ww,il,iu);
precon->L = W;
precon->U = Z;
precon->D = D;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
LIS_INT lis_precon_create_sainv_csr(LIS_SOLVER solver, LIS_PRECON precon)
{
LIS_INT err;
LIS_INT i,j,k,ii,jj,len,lfil;
LIS_INT n,nnz,annz,cl,cu,cc,m;
LIS_INT *wu,*wl,*il,*iu,*ic,*pc;
LIS_SCALAR t,v;
LIS_REAL tol,tol_dd,nrm;
LIS_SCALAR *d,*r,*c,*l,*u,*tmp;
LIS_MATRIX A,B;
LIS_MATRIX_ILU W,Z;
LIS_VECTOR D;
LIS_DEBUG_FUNC_IN;
A = solver->A;
n = A->n;
nnz = A->nnz;
tol = solver->params[LIS_PARAMS_DROP-LIS_OPTIONS_LEN];
m = solver->params[LIS_PARAMS_RATE-LIS_OPTIONS_LEN];
annz = 10+A->nnz / A->n;
lfil = (LIS_INT)((double)A->nnz/(2.0*n))*m;
W = NULL;
Z = NULL;
wu = NULL;
wl = NULL;
d = NULL;
l = NULL;
u = NULL;
il = NULL;
iu = NULL;
err = lis_matrix_ilu_create(n,1,&W);
if( err ) return err;
err = lis_matrix_ilu_create(n,1,&Z);
if( err ) return err;
err = lis_matrix_ilu_setCR(W);
if( err ) return err;
err = lis_matrix_ilu_setCR(Z);
if( err ) return err;
err = lis_vector_duplicate(A,&D);
if( err ) return err;
d = D->value;
tmp = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_sainv_csr::l");
if( tmp==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
r = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_sainv_csr::l");
if( r==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
c = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_sainv_csr::u");
if( c==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
l = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_sainv_csr::l");
if( l==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
u = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_sainv_csr::u");
if( u==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
il = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::il");
if( il==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
iu = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::iu");
if( iu==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
ic = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::iu");
if( ic==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
wu = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::ww");
if( wu==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
wl = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::ww");
if( wl==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
pc = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_sainv_csr::iu");
if( pc==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
lis_matrix_sort_csr(A);
err = lis_matrix_duplicate(A,&B);
if( err ) return err;
err = lis_matrix_convert_csr2csc(A,B);
if( err ) return err;
for(i=0;i<n;i++)
{
wu[i] = 0;
wl[i] = 0;
pc[i] = A->ptr[i];
}
for(i=0; i<n; i++)
{
/* nrm_inf(A[i,:]) */
nrm = 0.0;
for(j=A->ptr[i];j<A->ptr[i+1];j++)
{
nrm = _max(nrm,fabs(A->value[j]));
}
tol_dd = nrm * tol;
/* l = e_i */
/* u = e_i */
l[i] = 1.0;
u[i] = 1.0;
il[0] = i;
iu[0] = i;
cl = 1;
cu = 1;
wu[i] = 1;
wl[i] = 1;
cc = 0;
/* r = e_i^T*A */
for(j=A->ptr[i];j<A->ptr[i+1];j++)
{
jj = A->index[j];
r[jj] = A->value[j];
}
/* c = A_i = A*e_i */
for(j=B->ptr[i];j<B->ptr[i+1];j++)
{
jj = B->index[j];
c[jj] = B->value[j];
}
/* W_i = W_i - (r*Z_j/D_jj)*W_j */
for(j=0;j<i;j++)
{
t = 0.0;
for(k=0;k<Z->nnz[j];k++)
{
t += r[Z->index[j][k]]*Z->value[j][k];
}
t = t * d[j];
if( fabs(t) > tol_dd )
{
for(k=0;k<W->nnz[j];k++)
{
v = t * W->value[j][k];
if( fabs(v) > tol_dd )
{
jj = W->index[j][k];
if( wl[jj]==1 )
{
l[jj] -= v;
}
else
{
l[jj] = -v;
il[cl++] = jj;
wl[jj] = 1;
}
}
}
}
}
/* Z_i = Z_i - (W_j^T*c/D_jj)*Z_j */
for(j=0;j<i;j++)
{
t = 0.0;
for(k=0;k<W->nnz[j];k++)
{
t += c[W->index[j][k]]*W->value[j][k];
}
t = t * d[j];
if( fabs(t) > tol_dd )
{
for(k=0;k<Z->nnz[j];k++)
{
v = t * Z->value[j][k];
if( fabs(v) > tol_dd )
{
jj = Z->index[j][k];
if( wu[jj]==1 )
{
u[jj] -= v;
}
else
{
u[jj] = -v;
iu[cu++] = jj;
wu[jj] = 1;
}
}
}
}
}
/*
len = _min(lfil,cl);
for(j=0;j<cl;j++) tmp[j] = fabs(l[il[j]]);
lis_sort_di(0,cl-1,tmp,il);
lis_sort_i(0,len-1,il);
cl = len;
*/
/*
k = cl;
for(j=0;j<cl;j++)
{
if( fabs(l[il[j]])<= tol_dd )
{
wl[il[j]] = 0;
il[j] = n;
k--;
}
}
lis_sort_i(0,cl-1,il);
cl = k;
k = cu;
for(j=0;j<cu;j++)
{
if( fabs(u[iu[j]])<= tol_dd )
{
wu[iu[j]] = 0;
iu[j] = n;
k--;
}
}
lis_sort_i(0,cu-1,iu);
cu = k;
*/
W->nnz[i] = cl;
if( cl > 0 )
{
W->index[i] = (LIS_INT *)malloc(cl*sizeof(LIS_INT));
W->value[i] = (LIS_SCALAR *)malloc(cl*sizeof(LIS_SCALAR));
memcpy(W->index[i],il,cl*sizeof(LIS_INT));
for(j=0;j<cl;j++)
{
W->value[i][j] = l[il[j]];
}
}
Z->nnz[i] = cu;
if( cu > 0 )
{
Z->index[i] = (LIS_INT *)malloc(cu*sizeof(LIS_INT));
Z->value[i] = (LIS_SCALAR *)malloc(cu*sizeof(LIS_SCALAR));
memcpy(Z->index[i],iu,cu*sizeof(LIS_INT));
for(j=0;j<cu;j++)
{
Z->value[i][j] = u[iu[j]];
}
}
for(j=A->ptr[i];j<A->ptr[i+1];j++) r[A->index[j]] = 0.0;
for(j=B->ptr[i];j<B->ptr[i+1];j++) c[B->index[j]] = 0.0;
for(j=0;j<cl;j++)
{
wl[il[j]] = 0;
l[il[j]] = 0.0;
}
for(j=0;j<cu;j++)
{
wu[iu[j]] = 0;
}
/* D_ii = W_i^T * A * Z_i */
cl = 0;
for(k=0;k<Z->nnz[i];k++)
{
ii = Z->index[i][k];
for(j=B->ptr[ii];j<B->ptr[ii+1];j++)
{
jj = B->index[j];
if( wl[jj]==0 )
{
l[jj] = B->value[j]*Z->value[i][k];
wl[jj] = 1;
il[cl++] = jj;
}
else
{
l[jj] += B->value[j]*Z->value[i][k];
}
}
}
t = 0.0;
for(j=0;j<W->nnz[i];j++)
{
k = W->index[i][j];
t += W->value[i][j] * l[k];
}
d[i] = 1.0 / t;
for(j=0;j<cl;j++) wl[il[j]] = 0;
}
lis_matrix_destroy(B);
lis_free2(11,r,c,il,l,wl,iu,u,wu,ic,pc,tmp);
precon->L = W;
precon->U = Z;
precon->D = D;
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
}
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;
}
LisVector::LisVector(LisVector const& src)
{
lis_vector_duplicate(src._vec, &_vec);
lis_vector_copy(src._vec, _vec);
}
int main(int argc, char* argv[])
{
int err;
int nprocs,mtype,my_rank;
int nsol;
LIS_MATRIX A,A0;
LIS_VECTOR x;
LIS_REAL evalue0;
LIS_SCALAR shift, *tmpa;
LIS_ESOLVER esolver;
LIS_SOLVER solver;
LIS_REAL residual;
int iters;
double times;
double itimes,ptimes,p_c_times,p_i_times;
LIS_PRECON precon;
int *ptr,*index;
LIS_SCALAR *value;
int nsolver;
char esolvername[128];
int mode;
LIS_DEBUG_FUNC_IN;
lis_initialize(&argc, &argv);
#ifdef USE_MPI
MPI_Comm_size(MPI_COMM_WORLD,&nprocs);
MPI_Comm_rank(MPI_COMM_WORLD,&my_rank);
#else
nprocs = 1;
my_rank = 0;
#endif
if( argc < 4 )
{
if( my_rank==0 ) printf("Usage: etest1 matrix_filename solution_filename residual_filename [options]\n");
lis_finalize();
exit(0);
}
if( my_rank==0 )
{
printf("\n");
printf("number of processes = %d\n",nprocs);
}
#ifdef _OPENMP
if( my_rank==0 )
{
printf("max number of threads = %d\n",omp_get_num_procs());
printf("number of threads = %d\n",omp_get_max_threads());
}
#endif
/* create matrix and vectors */
lis_matrix_create(LIS_COMM_WORLD,&A);
lis_input_matrix(A,argv[1]);
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,&nsol);
lis_get_esolvername(nsol,esolvername);
lis_esolver_get_residualnorm(esolver, &residual);
lis_esolver_get_iters(esolver, &iters);
lis_esolver_get_timeex(esolver,×,&itimes,&ptimes,&p_c_times,&p_i_times);
if( my_rank==0 ) {
printf("%s: mode number = %d\n", esolvername, esolver->options[LIS_EOPTIONS_MODE]);
printf("%s: eigenvalue = %e\n", esolvername, evalue0);
printf("%s: number of iterations = %d\n",esolvername, iters);
printf("%s: elapsed time = %e sec.\n", esolvername, times);
printf("%s: preconditioner = %e sec.\n", esolvername, ptimes);
printf("%s: matrix creation = %e sec.\n", esolvername, p_c_times);
printf("%s: linear solver = %e sec.\n", esolvername, itimes);
printf("%s: relative residual 2-norm = %e\n\n",esolvername, residual);
}
/* write solution */
lis_output_vector(x,LIS_FMT_MM,argv[2]);
/* write residual */
lis_esolver_output_rhistory(esolver, argv[3]);
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;
}
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 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_precon_create_ilut_csr(LIS_SOLVER solver, LIS_PRECON precon)
{
#ifdef _OPENMP
LIS_INT err;
LIS_INT i,j,k,ii,jj,kk;
LIS_INT is,ie,my_rank,nprocs;
LIS_INT n,nr,nnz,lfil,len;
LIS_SCALAR gamma,t,tol,toldd,m;
LIS_MATRIX A;
LIS_MATRIX_ILU L,U;
LIS_VECTOR D;
LIS_SCALAR tnorm, tolnorm;
LIS_SCALAR fact,lxu,*wn,*w;
LIS_INT lenu,lenl,col,jpos,jrow,upos,para;
LIS_INT *jbuf,*iw;
LIS_DEBUG_FUNC_IN;
A = solver->A;
n = A->n;
tol = solver->params[LIS_PARAMS_DROP-LIS_OPTIONS_LEN];
m = solver->params[LIS_PARAMS_RATE-LIS_OPTIONS_LEN];
gamma = solver->params[LIS_PARAMS_GAMMA-LIS_OPTIONS_LEN];
lfil = (LIS_INT)((double)A->nnz/(2.0*n))*m;
nprocs = omp_get_max_threads();
L = NULL;
U = NULL;
err = lis_matrix_ilu_create(n,1,&L);
if( err ) return err;
err = lis_matrix_ilu_create(n,1,&U);
if( err ) return err;
err = lis_matrix_ilu_setCR(L);
if( err ) return err;
err = lis_matrix_ilu_setCR(U);
if( err ) return err;
err = lis_vector_duplicate(A,&D);
if( err )
{
return err;
}
w = (LIS_SCALAR *)lis_malloc(nprocs*(n+1)*sizeof(LIS_SCALAR),"lis_precon_create_ilut_csr::w");
if( w==NULL )
{
LIS_SETERR_MEM(nprocs*(n+1)*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
wn = (LIS_SCALAR *)lis_malloc(nprocs*n*sizeof(LIS_SCALAR),"lis_precon_create_ilut_csr::w");
if( wn==NULL )
{
LIS_SETERR_MEM(nprocs*n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
jbuf = (LIS_INT *)lis_malloc(nprocs*n*sizeof(LIS_INT),"lis_precon_create_ilut_csr::iw");
if( jbuf==NULL )
{
LIS_SETERR_MEM(nprocs*n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
iw = (LIS_INT *)lis_malloc(nprocs*n*sizeof(LIS_INT),"lis_precon_create_ilut_csr::iw");
if( iw==NULL )
{
LIS_SETERR_MEM(nprocs*n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
#pragma omp parallel private(is,ie,my_rank,i,j,k,jj,tnorm,tolnorm,len,lenu,lenl,col,t,jpos,jrow,fact,lxu,upos)
{
my_rank = omp_get_thread_num();
LIS_GET_ISIE(my_rank,nprocs,n,is,ie);
for(i=is;i<ie;i++) iw[my_rank*n+i] = -1;
for(i=is;i<ie;i++)
{
tnorm = 0;
k = 0;
for(j=A->ptr[i];j<A->ptr[i+1];j++)
{
jj = A->index[j];
if( jj<is || jj>=ie ) continue;
tnorm += fabs(A->value[j]);
k++;
}
tnorm = tnorm / (double)k;
tolnorm = tol * tnorm;
lenu = 0;
lenl = 0;
jbuf[my_rank*n+i] = i;
w[my_rank*n+i] = 0;
iw[my_rank*n+i] = i;
for(j=A->ptr[i];j<A->ptr[i+1];j++)
{
col = A->index[j];
if( col<is || col>=ie ) continue;
t = A->value[j];
if( col < i )
{
jbuf[my_rank*n+lenl] = col;
iw[my_rank*n+col] = lenl;
w[my_rank*n+lenl] = t;
lenl++;
}
else if( col == i )
{
w[my_rank*n+i] = t;
}
else
{
lenu++;
jpos = i + lenu;
jbuf[my_rank*n+jpos] = col;
iw[my_rank*n+col] = jpos;
w[my_rank*n+jpos] = t;
}
}
j = -1;
len = 0;
while( ++j < lenl )
{
jrow = jbuf[my_rank*n+j];
jpos = j;
for(k=j+1;k<lenl;k++)
{
if( jbuf[my_rank*n+k]<jrow )
{
jrow = jbuf[my_rank*n+k];
jpos = k;
}
}
if( jpos!=j )
{
col = jbuf[my_rank*n+j];
jbuf[my_rank*n+j] = jbuf[my_rank*n+jpos];
jbuf[my_rank*n+jpos] = col;
iw[my_rank*n+jrow] = j;
iw[my_rank*n+col] = jpos;
t = w[my_rank*n+j];
w[my_rank*n+j] = w[my_rank*n+jpos];
w[my_rank*n+jpos] = t;
}
fact = w[my_rank*n+j] * D->value[jrow];
w[my_rank*n+j] = fact;
iw[my_rank*n+jrow] = -1;
for(k=0;k<U->nnz[jrow];k++)
{
col = U->index[jrow][k];
jpos = iw[my_rank*n+col];
lxu = -fact * U->value[jrow][k];
if( fabs(lxu) < tolnorm && jpos==-1 ) continue;
if( col >= i )
{
if( jpos == -1 )
{
lenu++;
upos = i + lenu;
jbuf[my_rank*n+upos] = col;
iw[my_rank*n+col] = upos;
w[my_rank*n+upos] = lxu;
}
else
{
w[my_rank*n+jpos] += lxu;
}
}
else
{
if( jpos == -1 )
{
jbuf[my_rank*n+lenl] = col;
iw[my_rank*n+col] = lenl;
w[my_rank*n+lenl] = lxu;
lenl++;
}
else
{
w[my_rank*n+jpos] += lxu;
}
}
}
}
iw[my_rank*n+i] = -1;
for(j=0;j<lenu;j++)
{
iw[ my_rank*n+jbuf[my_rank*n+i+j+1] ] = -1;
}
D->value[i] = 1.0 / w[my_rank*n+i];
len = _min(lfil,lenl);
for(j=0;j<lenl;j++)
{
wn[my_rank*n+j] = fabs(w[my_rank*n+j]);
iw[my_rank*n+j] = j;
}
lis_sort_di(0,lenl-1,&wn[my_rank*n],&iw[my_rank*n]);
lis_sort_i(0,len-1,&iw[my_rank*n]);
L->nnz[i] = len;
if( len>0 )
{
L->index[i] = (LIS_INT *)malloc(len*sizeof(LIS_INT));
L->value[i] = (LIS_SCALAR *)malloc(len*sizeof(LIS_SCALAR));
}
for(j=0;j<len;j++)
{
jpos = iw[my_rank*n+j];
L->index[i][j] = jbuf[my_rank*n+jpos];
L->value[i][j] = w[my_rank*n+jpos];
}
for(j=0;j<lenl;j++) iw[my_rank*n+j] = -1;
len = _min(lfil,lenu);
for(j=0;j<lenu;j++)
{
wn[my_rank*n+j] = fabs(w[my_rank*n+i+j+1]);
iw[my_rank*n+j] = i+j+1;
}
lis_sort_di(0,lenu-1,&wn[my_rank*n],&iw[my_rank*n]);
lis_sort_i(0,len-1,&iw[my_rank*n]);
U->nnz[i] = len;
if( len>0 )
{
U->index[i] = (LIS_INT *)malloc(len*sizeof(LIS_INT));
U->value[i] = (LIS_SCALAR *)malloc(len*sizeof(LIS_SCALAR));
}
for(j=0;j<len;j++)
{
jpos = iw[my_rank*n+j];
U->index[i][j] = jbuf[my_rank*n+jpos];
U->value[i][j] = w[my_rank*n+jpos];
}
for(j=0;j<lenu;j++) iw[my_rank*n+j] = -1;
}
}
precon->L = L;
precon->U = U;
precon->D = D;
lis_free2(4,w,iw,wn,jbuf);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
#else
LIS_INT err;
LIS_INT i,j,k;
LIS_INT n,lfil,len;
LIS_SCALAR gamma,t,tol,m;
LIS_MATRIX A;
LIS_MATRIX_ILU L,U;
LIS_VECTOR D;
LIS_SCALAR tnorm, tolnorm;
LIS_SCALAR fact,lxu,*wn,*w;
LIS_INT lenu,lenl,col,jpos,jrow,upos;
LIS_INT *jbuf,*iw;
LIS_DEBUG_FUNC_IN;
A = solver->A;
n = A->n;
tol = solver->params[LIS_PARAMS_DROP-LIS_OPTIONS_LEN];
m = solver->params[LIS_PARAMS_RATE-LIS_OPTIONS_LEN];
gamma = solver->params[LIS_PARAMS_GAMMA-LIS_OPTIONS_LEN];
lfil = (LIS_INT)(((double)A->nnz/(2.0*n))*m);
L = NULL;
U = NULL;
err = lis_matrix_ilu_create(n,1,&L);
if( err ) return err;
err = lis_matrix_ilu_create(n,1,&U);
if( err ) return err;
err = lis_matrix_ilu_setCR(L);
if( err ) return err;
err = lis_matrix_ilu_setCR(U);
if( err ) return err;
err = lis_vector_duplicate(A,&D);
if( err )
{
return err;
}
w = (LIS_SCALAR *)lis_malloc((n+1)*sizeof(LIS_SCALAR),"lis_precon_create_ilut_csr::w");
if( w==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
wn = (LIS_SCALAR *)lis_malloc(n*sizeof(LIS_SCALAR),"lis_precon_create_ilut_csr::w");
if( wn==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_SCALAR));
return LIS_OUT_OF_MEMORY;
}
jbuf = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_ilut_csr::iw");
if( jbuf==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
iw = (LIS_INT *)lis_malloc(n*sizeof(LIS_INT),"lis_precon_create_ilut_csr::iw");
if( iw==NULL )
{
LIS_SETERR_MEM(n*sizeof(LIS_INT));
return LIS_OUT_OF_MEMORY;
}
for(i=0;i<n;i++) iw[i] = -1;
for(i=0;i<n;i++)
{
tnorm = 0;
for(j=A->ptr[i];j<A->ptr[i+1];j++)
{
tnorm += fabs(A->value[j]);
}
tnorm = tnorm / (double)(A->ptr[i+1]-A->ptr[i]);
tolnorm = tol * tnorm;
lenu = 0;
lenl = 0;
jbuf[i] = i;
w[i] = 0;
iw[i] = i;
for(j=A->ptr[i];j<A->ptr[i+1];j++)
{
col = A->index[j];
#ifdef USE_MPI
if( col>n-1 ) continue;
#endif
t = A->value[j];
if( col < i )
{
jbuf[lenl] = col;
iw[col] = lenl;
w[lenl] = t;
lenl++;
}
else if( col == i )
{
w[i] = t;
}
else
{
lenu++;
jpos = i + lenu;
jbuf[jpos] = col;
iw[col] = jpos;
w[jpos] = t;
}
}
j = -1;
len = 0;
while( ++j < lenl )
{
jrow = jbuf[j];
jpos = j;
for(k=j+1;k<lenl;k++)
{
if( jbuf[k]<jrow )
{
jrow = jbuf[k];
jpos = k;
}
}
if( jpos!=j )
{
col = jbuf[j];
jbuf[j] = jbuf[jpos];
jbuf[jpos] = col;
iw[jrow] = j;
iw[col] = jpos;
t = w[j];
w[j] = w[jpos];
w[jpos] = t;
}
fact = w[j] * D->value[jrow];
w[j] = fact;
iw[jrow] = -1;
for(k=0;k<U->nnz[jrow];k++)
{
col = U->index[jrow][k];
jpos = iw[col];
lxu = -fact * U->value[jrow][k];
if( fabs(lxu) < tolnorm && jpos==-1 ) continue;
if( col >= i )
{
if( jpos == -1 )
{
lenu++;
upos = i + lenu;
jbuf[upos] = col;
iw[col] = upos;
w[upos] = lxu;
}
else
{
w[jpos] += lxu;
}
}
else
{
if( jpos == -1 )
{
jbuf[lenl] = col;
iw[col] = lenl;
w[lenl] = lxu;
lenl++;
}
else
{
w[jpos] += lxu;
}
}
}
/* for(kk=0;kk<bs;kk++)
{
w[bs*len+kk] = -buf_fact[kk];
}
jbuf[len] = jrow;
len++;*/
}
iw[i] = -1;
for(j=0;j<lenu;j++)
{
iw[ jbuf[i+j+1] ] = -1;
}
D->value[i] = 1.0 / w[i];
len = _min(lfil,lenl);
for(j=0;j<lenl;j++)
{
wn[j] = fabs(w[j]);
iw[j] = j;
}
lis_sort_di(0,lenl-1,wn,iw);
lis_sort_i(0,len-1,iw);
L->nnz[i] = len;
if( len>0 )
{
L->index[i] = (LIS_INT *)malloc(len*sizeof(LIS_INT));
L->value[i] = (LIS_SCALAR *)malloc(len*sizeof(LIS_SCALAR));
}
for(j=0;j<len;j++)
{
jpos = iw[j];
L->index[i][j] = jbuf[jpos];
L->value[i][j] = w[jpos];
}
for(j=0;j<lenl;j++) iw[j] = -1;
len = _min(lfil,lenu);
for(j=0;j<lenu;j++)
{
wn[j] = fabs(w[i+j+1]);
iw[j] = i+j+1;
}
lis_sort_di(0,lenu-1,wn,iw);
lis_sort_i(0,len-1,iw);
U->nnz[i] = len;
if( len>0 )
{
U->index[i] = (LIS_INT *)malloc(len*sizeof(LIS_INT));
U->value[i] = (LIS_SCALAR *)malloc(len*sizeof(LIS_SCALAR));
}
for(j=0;j<len;j++)
{
jpos = iw[j];
U->index[i][j] = jbuf[jpos];
U->value[i][j] = w[jpos];
}
for(j=0;j<lenu;j++) iw[j] = -1;
}
precon->L = L;
precon->U = U;
precon->D = D;
lis_free2(4,w,iw,wn,jbuf);
LIS_DEBUG_FUNC_OUT;
return LIS_SUCCESS;
#endif
}
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(int argc, char* argv[])
{
LIS_Comm comm;
LIS_INT err;
int nprocs,my_rank;
LIS_INT nesol;
LIS_MATRIX A,B;
LIS_VECTOR x;
LIS_SCALAR 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);
comm = LIS_COMM_WORLD;
#ifdef USE_MPI
MPI_Comm_size(comm,&nprocs);
MPI_Comm_rank(comm,&my_rank);
#else
nprocs = 1;
my_rank = 0;
#endif
if( argc < 5 )
{
lis_printf(comm,"Usage: %s matrix_a_filename matrix_b_filename evector_filename rhistory_filename [options]\n", argv[0]);
CHKERR(1);
}
lis_printf(comm,"\n");
lis_printf(comm,"number of processes = %d\n",nprocs);
#ifdef _OPENMP
lis_printf(comm,"max number of threads = %d\n",omp_get_num_procs());
lis_printf(comm,"number of threads = %d\n",omp_get_max_threads());
#endif
/* create matrix and vectors */
lis_matrix_create(comm,&A);
lis_matrix_create(comm,&B);
lis_printf(comm,"\nmatrix A:\n");
lis_input_matrix(A,argv[1]);
lis_printf(comm,"matrix B:\n");
lis_input_matrix(B,argv[2]);
lis_vector_duplicate(A,&x);
lis_esolver_create(&esolver);
lis_esolver_set_option("-e gii -eprint mem",esolver);
err = lis_esolver_set_optionC(esolver);
CHKERR(err);
err = lis_gesolve(A,B,x,&evalue0,esolver);
CHKERR(err);
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);
lis_printf(comm,"%s: mode number = %d\n", esolvername, 0);
#ifdef _COMPLEX
lis_printf(comm,"%s: eigenvalue = (%e, %e)\n", esolvername, (double)creal(evalue0), (double)cimag(evalue0));
#else
lis_printf(comm,"%s: eigenvalue = %e\n", esolvername, (double)evalue0);
#endif
lis_printf(comm,"%s: number of iterations = %D\n",esolvername, iter);
lis_printf(comm,"%s: elapsed time = %e sec.\n", esolvername, time);
lis_printf(comm,"%s: preconditioner = %e sec.\n", esolvername, ptime);
lis_printf(comm,"%s: matrix creation = %e sec.\n", esolvername, p_c_time);
lis_printf(comm,"%s: linear solver = %e sec.\n", esolvername, itime);
lis_printf(comm,"%s: relative residual = %e\n\n",esolvername, (double)residual);
/* write eigenvector */
lis_output_vector(x,LIS_FMT_MM,argv[3]);
/* write residual history */
lis_esolver_output_rhistory(esolver,argv[4]);
lis_esolver_destroy(esolver);
lis_matrix_destroy(A);
lis_matrix_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 lis_esi_quad(LIS_ESOLVER esolver)
{
LIS_MATRIX A;
LIS_VECTOR x, Ax;
LIS_SCALAR xAx, xx, mu, lshift;
LIS_INT ss;
LIS_INT emaxiter;
LIS_REAL tol;
LIS_INT i,j,k;
LIS_SCALAR evalue,dotvr;
LIS_INT iter,giter,output,niesolver;
LIS_INT nprocs,my_rank;
LIS_REAL nrm2,dot,resid,resid0;
LIS_QUAD_PTR qdot_vv, qdot_vr;
LIS_VECTOR *v,r,q;
LIS_SOLVER solver;
LIS_PRECON precon;
double times,itimes,ptimes,p_c_times,p_i_times;
LIS_INT err;
LIS_INT nsol, precon_type;
char solvername[128], preconname[128];
LIS_DEBUG_FUNC_IN;
A = esolver->A;
x = esolver->x;
ss = esolver->options[LIS_EOPTIONS_SUBSPACE];
emaxiter = esolver->options[LIS_EOPTIONS_MAXITER];
tol = esolver->params[LIS_EPARAMS_RESID - LIS_EOPTIONS_LEN];
lshift = esolver->lshift;
output = esolver->options[LIS_EOPTIONS_OUTPUT];
niesolver = esolver->options[LIS_EOPTIONS_INNER_ESOLVER];
r = esolver->work[0];
q = esolver->work[1];
v = &esolver->work[2];
Ax = esolver->work[3];
LIS_QUAD_SCALAR_MALLOC(qdot_vv,0,1);
LIS_QUAD_SCALAR_MALLOC(qdot_vr,1,1);
lis_vector_set_all(1.0,r);
lis_vector_nrm2(r, &nrm2);
lis_vector_scale(1/nrm2,r);
switch ( niesolver )
{
case LIS_ESOLVER_II:
lis_solver_create(&solver);
lis_solver_set_option("-i bicg -p ilu -precision quad",solver);
lis_solver_set_optionC(solver);
lis_solver_get_solver(solver, &nsol);
lis_solver_get_precon(solver, &precon_type);
lis_get_solvername(nsol, solvername);
lis_get_preconname(precon_type, preconname);
printf("solver : %s %d\n", solvername, nsol);
printf("precon : %s %d\n", preconname, precon_type);
if( A->my_rank==0 ) printf("local shift = %e\n", lshift);
if (lshift != 0) lis_matrix_shift_diagonal(A, lshift);
break;
case LIS_ESOLVER_AII:
lis_solver_create(&solver);
lis_solver_set_option("-i bicg -p ilu -precision quad",solver);
lis_solver_set_optionC(solver);
lis_solver_get_solver(solver, &nsol);
lis_solver_get_precon(solver, &precon_type);
lis_get_solvername(nsol, solvername);
lis_get_preconname(precon_type, preconname);
printf("solver : %s %d\n", solvername, nsol);
printf("precon : %s %d\n", preconname, precon_type);
if( A->my_rank==0 ) printf("local shift = %e\n", lshift);
if (lshift != 0) lis_matrix_shift_diagonal(A, lshift);
lis_vector_set_all(1.0,q);
lis_solve(A, q, x, solver);
lis_precon_create(solver, &precon);
solver->precon = precon;
break;
case LIS_ESOLVER_RQI:
lis_solver_create(&solver);
lis_solver_set_option("-p ilu -precision quad -maxiter 10",solver);
lis_solver_set_optionC(solver);
lis_solver_get_solver(solver, &nsol);
lis_solver_get_precon(solver, &precon_type);
lis_get_solvername(nsol, solvername);
lis_get_preconname(precon_type, preconname);
printf("solver : %s %d\n", solvername, nsol);
printf("precon : %s %d\n", preconname, precon_type);
if( A->my_rank==0 ) printf("local shift = %e\n", lshift);
if (lshift != 0) lis_matrix_shift_diagonal(A, lshift);
break;
}
giter=0;
j=0;
while (j<ss)
{
lis_vector_duplicate(A,&esolver->evector[j]);
j = j+1;
lis_vector_copy(r, v[j]);
if (niesolver==LIS_ESOLVER_II || niesolver==LIS_ESOLVER_RQI)
{
/* create preconditioner */
solver->A = A;
err = lis_precon_create(solver, &precon);
if( err )
{
lis_solver_work_destroy(solver);
solver->retcode = err;
return err;
}
}
if (niesolver==LIS_ESOLVER_RQI)
{
lis_vector_nrm2(x, &nrm2);
lis_vector_scale(1/nrm2, x);
lis_matvec(A, x, Ax);
lis_vector_dot(x, Ax, &xAx);
lis_vector_dot(x, x, &xx);
mu = xAx / xx;
}
iter = 0;
while (iter<emaxiter)
{
/* diagonalization */
iter = iter+1;
giter = giter+1;
for (k=1;k<j;k++)
{
lis_vector_dotex_mmm(v[j], v[k], &qdot_vv);
lis_quad_minus((LIS_QUAD *)qdot_vv.hi);
lis_vector_axpyex_mmm(qdot_vv,v[k],v[j]);
}
switch( niesolver )
{
case LIS_ESOLVER_PI:
lis_matvec(A,v[j],r);
break;
case LIS_ESOLVER_II:
lis_solve_kernel(A, v[j], r, solver, precon);
break;
case LIS_ESOLVER_AII:
lis_psolve(solver, v[j], r);
break;
case LIS_ESOLVER_RQI:
lis_vector_nrm2(v[j], &nrm2);
lis_vector_scale(1/nrm2, v[j]);
lis_matrix_shift_diagonal(A, -mu);
lis_solve_kernel(A, v[j], r, solver, precon);
lis_matrix_shift_diagonal(A, mu);
break;
}
if ( j==1 && ( niesolver==LIS_ESOLVER_II || niesolver==LIS_ESOLVER_AII || niesolver==LIS_ESOLVER_RQI ))
{
lis_solver_get_timeex(solver,×,&itimes,&ptimes,&p_c_times,&p_i_times);
esolver->ptimes += solver->ptimes;
esolver->itimes += solver->itimes;
esolver->p_c_times += solver->p_c_times;
esolver->p_i_times += solver->p_i_times;
}
lis_vector_nrm2(r, &nrm2);
lis_vector_dotex_mmm(v[j], r, &qdot_vr);
lis_quad_minus((LIS_QUAD *)qdot_vr.hi);
lis_vector_axpyzex_mmmm(qdot_vr,v[j],r,q);
lis_quad_minus((LIS_QUAD *)qdot_vr.hi);
dotvr = qdot_vr.hi[0];
mu = mu + 1/dotvr;
lis_vector_nrm2(q, &resid);
resid = fabs(resid / dotvr);
lis_vector_scale(1/nrm2,r);
lis_vector_copy(r, v[j]);
if ( j==1 )
{
if( output & LIS_PRINT_MEM ) esolver->residual[iter] = resid;
if( output & LIS_PRINT_OUT ) printf("iter: %5d residual = %e\n", iter, resid);
esolver->iter = iter;
esolver->resid = resid;
}
if (tol>resid) break;
}
if (niesolver==LIS_ESOLVER_II || niesolver==LIS_ESOLVER_RQI)
{
lis_precon_destroy(precon);
}
switch ( niesolver )
{
case LIS_ESOLVER_PI:
esolver->evalue[j-1] = dotvr;
break;
case LIS_ESOLVER_II:
esolver->evalue[j-1] = 1/dotvr;
break;
case LIS_ESOLVER_AII:
esolver->evalue[j-1] = 1/dotvr;
break;
case LIS_ESOLVER_RQI:
esolver->evalue[j-1] = mu;
break;
}
lis_vector_copy(v[j], esolver->evector[j-1]);
if (A->my_rank==0 && ss>1)
{
#ifdef _LONGLONG
printf("Subspace: mode number = %lld\n", j-1);
#else
printf("Subspace: mode number = %d\n", j-1);
#endif
printf("Subspace: eigenvalue = %e\n", esolver->evalue[j-1]);
#ifdef _LONGLONG
printf("Subspace: number of iterations = %lld\n",iter);
#else
printf("Subspace: number of iterations = %d\n",iter);
#endif
printf("Subspace: relative residual 2-norm = %e\n",resid);
}
}
lis_vector_copy(esolver->evector[esolver->options[LIS_EOPTIONS_MODE]], esolver->x);
switch ( niesolver )
{
case LIS_ESOLVER_II:
if (lshift != 0) lis_matrix_shift_diagonal(A, -lshift);
lis_solver_destroy(solver);
break;
case LIS_ESOLVER_AII:
if (lshift != 0) lis_matrix_shift_diagonal(A, -lshift);
lis_precon_destroy(precon);
lis_solver_destroy(solver);
break;
case LIS_ESOLVER_RQI:
if (lshift != 0) lis_matrix_shift_diagonal(A, -lshift);
lis_solver_destroy(solver);
break;
}
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 lis_precon_create_hybrid(LIS_SOLVER solver, LIS_PRECON precon)
{
LIS_INT nsolver, maxiter, precision;
LIS_INT err;
LIS_SCALAR *rhistory;
LIS_VECTOR xx;
LIS_SOLVER psolver;
LIS_MATRIX A;
LIS_PRECON pprecon;
LIS_DEBUG_FUNC_IN;
A = solver->A;
err = lis_solver_create(&psolver);
if( err )
{
return err;
}
psolver->params[LIS_PARAMS_RESID-LIS_OPTIONS_LEN] = solver->params[LIS_PARAMS_PRESID-LIS_OPTIONS_LEN];
psolver->params[LIS_PARAMS_W-LIS_OPTIONS_LEN] = solver->params[LIS_PARAMS_PW-LIS_OPTIONS_LEN];
psolver->options[LIS_OPTIONS_MAXITER] = solver->options[LIS_OPTIONS_PMAXITER];
psolver->options[LIS_OPTIONS_ELL] = solver->options[LIS_OPTIONS_PELL];
psolver->options[LIS_OPTIONS_RESTART] = solver->options[LIS_OPTIONS_PRESTART];
psolver->options[LIS_OPTIONS_OUTPUT] = 0;
psolver->options[LIS_OPTIONS_SOLVER] = solver->options[LIS_OPTIONS_PSOLVER];
psolver->options[LIS_OPTIONS_PRECON] = solver->options[LIS_OPTIONS_PPRECON];
psolver->options[LIS_OPTIONS_INITGUESS_ZEROS] = solver->options[LIS_OPTIONS_INITGUESS_ZEROS];
psolver->options[LIS_OPTIONS_PRECISION] = solver->options[LIS_OPTIONS_PRECISION];
psolver->A = solver->A;
psolver->At = solver->At;
psolver->precision = solver->precision;
nsolver = psolver->options[LIS_OPTIONS_SOLVER];
maxiter = psolver->options[LIS_OPTIONS_MAXITER];
precision = psolver->options[LIS_OPTIONS_PRECISION];
A = psolver->A;
/* 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;
}
/* create residual history vector */
rhistory = (LIS_SCALAR *)lis_malloc((maxiter+2)*sizeof(LIS_SCALAR),"lis_precon_create_hybrid::rhistory");
if( rhistory==NULL )
{
LIS_SETERR_MEM((maxiter+2)*sizeof(LIS_SCALAR));
lis_vector_destroy(xx);
solver->retcode = err;
return err;
}
/* create preconditioner */
err = lis_precon_create(psolver, &pprecon);
if( err )
{
lis_vector_destroy(xx);
lis_solver_work_destroy(psolver);
lis_free(rhistory);
solver->retcode = err;
return err;
}
/* create work vector */
err = lis_solver_malloc_work[nsolver](psolver);
if( err )
{
lis_vector_destroy(xx);
lis_precon_destroy(pprecon);
solver->retcode = err;
return err;
}
psolver->x = xx;
psolver->precon = pprecon;
psolver->rhistory = rhistory;
precon->solver = psolver;
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_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,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 main(LIS_INT argc, char* argv[])
{
LIS_INT nprocs,my_rank;
int int_nprocs,int_my_rank;
LIS_INT nsol;
LIS_MATRIX A,B;
LIS_VECTOR x,y,z,w;
LIS_SCALAR 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 < 6 )
{
if( my_rank==0 )
{
printf("Usage: %s matrix_filename evalues_filename evectors_filename residuals_filename iters_filename [options]\n", argv[0]);
}
CHKERR(1);
}
if( my_rank==0 )
{
printf("\n");
printf("number of processes = %d\n",nprocs);
}
#ifdef _OPENMP
if( my_rank==0 )
{
printf("max number of threads = %d\n",omp_get_num_procs());
printf("number of threads = %d\n",omp_get_max_threads());
}
#endif
/* create matrix and vectors */
lis_matrix_create(LIS_COMM_WORLD,&A);
lis_input_matrix(A,argv[1]);
lis_vector_duplicate(A,&x);
lis_esolver_create(&esolver);
lis_esolver_set_option("-e si -ss 1 -eprint mem",esolver);
lis_esolver_set_optionC(esolver);
lis_esolve(A, x, &evalue0, esolver);
lis_esolver_get_esolver(esolver,&nsol);
lis_esolver_get_esolvername(nsol,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_create(LIS_COMM_WORLD,&y);
lis_vector_create(LIS_COMM_WORLD,&z);
lis_vector_create(LIS_COMM_WORLD,&w);
lis_matrix_create(LIS_COMM_WORLD,&B);
lis_esolver_get_evalues(esolver,y);
lis_esolver_get_residualnorms(esolver,z);
lis_esolver_get_iters(esolver,w);
lis_esolver_get_evectors(esolver,B);
/* write eigenvalues */
lis_output_vector(y,LIS_FMT_MM,argv[2]);
/* write eigenvectors */
lis_output_matrix(B,LIS_FMT_MM,argv[3]);
/* write residual norms */
lis_output_vector(z,LIS_FMT_MM,argv[4]);
/* write numbers of iterations */
lis_output_vector(w,LIS_FMT_MM,argv[5]);
lis_esolver_destroy(esolver);
lis_matrix_destroy(A);
lis_vector_destroy(x);
lis_matrix_destroy(B);
lis_vector_destroy(y);
lis_vector_destroy(z);
lis_vector_destroy(w);
lis_finalize();
LIS_DEBUG_FUNC_OUT;
return 0;
}
