ViennaCL - The Vienna Computing Library  1.7.0
Free open-source GPU-accelerated linear algebra and solver library.
iterative.cpp

This tutorial explains the use of iterative solvers in ViennaCL.

Note
iterative.cpp and iterative.cu are identical, the latter being required for compilation using CUDA nvcc

We start with including the necessary system headers:

//
// include necessary system headers
//
#include <iostream>
//
// Necessary to obtain a suitable performance in ublas
#ifndef NDEBUG
#define BOOST_UBLAS_NDEBUG
#endif
//
// ublas includes
//
#include <boost/numeric/ublas/io.hpp>
#include <boost/numeric/ublas/triangular.hpp>
#include <boost/numeric/ublas/matrix_sparse.hpp>
#include <boost/numeric/ublas/matrix.hpp>
#include <boost/numeric/ublas/matrix_proxy.hpp>
#include <boost/numeric/ublas/operation.hpp>
#include <boost/numeric/ublas/operation_sparse.hpp>
#include <boost/numeric/ublas/io.hpp>
#include <boost/numeric/ublas/lu.hpp>
// Must be set if you want to use ViennaCL algorithms on ublas objects
#define VIENNACL_WITH_UBLAS 1
//
// ViennaCL includes
//
#include "viennacl/scalar.hpp"
#include "viennacl/vector.hpp"
#include "viennacl/compressed_matrix.hpp"
#include "viennacl/coordinate_matrix.hpp"
#include "viennacl/linalg/prod.hpp"
#include "viennacl/linalg/ilu.hpp"
#include "viennacl/linalg/jacobi_precond.hpp"
#include "viennacl/linalg/cg.hpp"
#include "viennacl/linalg/bicgstab.hpp"
#include "viennacl/linalg/gmres.hpp"
#include "viennacl/io/matrix_market.hpp"
// Some helper functions for this tutorial:
#include "vector-io.hpp"

A shortcut for convience (use ublas:: instead of boost::numeric::ublas::)

using namespace boost::numeric;

In the main() routine we create matrices and vectors using Boost.uBLAS types and fill them with data. Then, ViennaCL types are created and the data is copied from the respective Boost.uBLAS objects. Various preconditioners are set up and finally the iterative solvers get called.

int main()
{
typedef float ScalarType;

Set up some ublas objects

ublas::vector<ScalarType> rhs;
ublas::vector<ScalarType> ref_result;
ublas::vector<ScalarType> result;
ublas::compressed_matrix<ScalarType> ublas_matrix;

Read system matrix from a matrix-market file

if (!viennacl::io::read_matrix_market_file(ublas_matrix, "../examples/testdata/mat65k.mtx"))
{
std::cout << "Error reading Matrix file" << std::endl;
return EXIT_FAILURE;
}
if (!readVectorFromFile("../examples/testdata/rhs65025.txt", rhs))
{
std::cout << "Error reading RHS file" << std::endl;
return EXIT_FAILURE;
}
if (!readVectorFromFile("../examples/testdata/result65025.txt", ref_result))
{
std::cout << "Error reading Result file" << std::endl;
return EXIT_FAILURE;
}

Set up some ViennaCL objects

std::size_t vcl_size = rhs.size();
viennacl::compressed_matrix<ScalarType> vcl_compressed_matrix;
viennacl::coordinate_matrix<ScalarType> vcl_coordinate_matrix;
viennacl::vector<ScalarType> vcl_rhs(vcl_size);
viennacl::vector<ScalarType> vcl_result(vcl_size);
viennacl::vector<ScalarType> vcl_ref_result(vcl_size);
viennacl::copy(rhs.begin(), rhs.end(), vcl_rhs.begin());
viennacl::copy(ref_result.begin(), ref_result.end(), vcl_ref_result.begin());

Transfer ublas-matrix to GPU:

viennacl::copy(ublas_matrix, vcl_compressed_matrix);

An alternative way is to use the C++ STL. In such case, the sparse matrix is represented as a std::vector< std::map< unsigned int, ScalarType> > After the instantiation, the data from the uBLAS matrix is copied over to the STL-matrix, which is then copied over to a ViennaCL matrix. This step is just for demonstration purposes.

std::vector< std::map< unsigned int, ScalarType> > stl_matrix(rhs.size());
for (ublas::compressed_matrix<ScalarType>::iterator1 iter1 = ublas_matrix.begin1();
iter1 != ublas_matrix.end1();
++iter1)
{
for (ublas::compressed_matrix<ScalarType>::iterator2 iter2 = iter1.begin();
iter2 != iter1.end();
++iter2)
stl_matrix[iter2.index1()][static_cast<unsigned int>(iter2.index2())] = *iter2;
}
std::vector<ScalarType> stl_rhs(rhs.size()), stl_result(result.size());
std::copy(rhs.begin(), rhs.end(), stl_rhs.begin());
viennacl::copy(stl_matrix, vcl_coordinate_matrix);
viennacl::copy(vcl_coordinate_matrix, stl_matrix);

Set up ILUT preconditioners for ublas and ViennaCL objects:

std::cout << "Setting up preconditioners for uBLAS-matrix..." << std::endl;
viennacl::linalg::ilut_precond< ublas::compressed_matrix<ScalarType> > ublas_ilut(ublas_matrix, viennacl::linalg::ilut_tag());
viennacl::linalg::ilu0_precond< ublas::compressed_matrix<ScalarType> > ublas_ilu0(ublas_matrix, viennacl::linalg::ilu0_tag());
viennacl::linalg::block_ilu_precond< ublas::compressed_matrix<ScalarType>,
viennacl::linalg::ilu0_tag> ublas_block_ilu0(ublas_matrix, viennacl::linalg::ilu0_tag());
std::cout << "Setting up preconditioners for ViennaCL-matrix..." << std::endl;
viennacl::linalg::ilut_precond< viennacl::compressed_matrix<ScalarType> > vcl_ilut(vcl_compressed_matrix, viennacl::linalg::ilut_tag());
viennacl::linalg::ilu0_precond< viennacl::compressed_matrix<ScalarType> > vcl_ilu0(vcl_compressed_matrix, viennacl::linalg::ilu0_tag());
viennacl::linalg::block_ilu_precond< viennacl::compressed_matrix<ScalarType>,
viennacl::linalg::ilu0_tag> vcl_block_ilu0(vcl_compressed_matrix, viennacl::linalg::ilu0_tag());

set up Jacobi preconditioners for ViennaCL and ublas objects:

viennacl::linalg::jacobi_precond< ublas::compressed_matrix<ScalarType> > ublas_jacobi(ublas_matrix, viennacl::linalg::jacobi_tag());
viennacl::linalg::jacobi_precond< viennacl::compressed_matrix<ScalarType> > vcl_jacobi(vcl_compressed_matrix, viennacl::linalg::jacobi_tag());

Conjugate Gradient Solver

std::cout << "----- CG Method -----" << std::endl;

Run the CG method with uBLAS objects. Use a relative tolerance of $ 10^{-6} $ and a maximum of 20 iterations when using ILUT or Jacobi preconditioners. You might need higher iteration counts for poorly conditioned systems.

result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::cg_tag());
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::cg_tag(1e-6, 20), ublas_ilut);
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::cg_tag(1e-6, 20), ublas_jacobi);

Run the CG method for ViennaCL objects. The results here should be the same as with uBLAS objects (at least up to round-off error).

vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::cg_tag());
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_ilut);
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_jacobi);

Convenience option: Run the CG method by passing STL types. This will use appropriate ViennaCL objects internally. You need to include viennacl/compressed_matrix.hpp and viennacl/vector.hpp before viennacl/linalg/cg.hpp for this to work!

stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::cg_tag());
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_ilut);
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_jacobi);

Stabilized BiConjugate Gradient Solver

std::cout << "----- BiCGStab Method -----" << std::endl;

Run BiCGStab for Boost.uBLAS objects. Parameters are the same as for the CG method.

result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::bicgstab_tag()); //without preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), ublas_ilut); //with preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), ublas_jacobi); //with preconditioner

Run the BiCGStab method for ViennaCL objects. The results here should be the same as with uBLAS objects (at least up to round-off error).

vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::bicgstab_tag()); //without preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_ilut); //with preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_jacobi); //with preconditioner

Convenience option: Run the BiCGStab method by passing STL types. This will use appropriate ViennaCL objects internally. You need to include viennacl/compressed_matrix.hpp and viennacl/vector.hpp before viennacl/linalg/bicgstab.hpp for this to work!

stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::bicgstab_tag());
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_ilut);
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_jacobi);

GMRES Solver

std::cout << "----- GMRES Method -----" << std::endl;

Run GMRES for Boost.uBLAS objects. Parameters are the same as for the CG method.

result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::gmres_tag()); //without preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::gmres_tag(1e-6, 20), ublas_ilut);//with preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::gmres_tag(1e-6, 20), ublas_jacobi);//with preconditioner

Run the GMRES method for ViennaCL objects. The results here should be the same as with uBLAS objects (at least up to round-off error).

vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::gmres_tag()); //without preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_ilut);//with preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_jacobi);//with preconditioner

Convenience option: Run the GMRES method by passing STL types. This will use appropriate ViennaCL objects internally. You need to include viennacl/compressed_matrix.hpp and viennacl/vector.hpp before viennacl/linalg/gmres.hpp for this to work!

stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::gmres_tag());
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_ilut);
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_jacobi);

That's it, the tutorial is completed.

std::cout << "!!!! TUTORIAL COMPLETED SUCCESSFULLY !!!!" << std::endl;
return EXIT_SUCCESS;
}

Full Example Code

/* =========================================================================
Copyright (c) 2010-2015, Institute for Microelectronics,
Institute for Analysis and Scientific Computing,
TU Wien.
Portions of this software are copyright by UChicago Argonne, LLC.
-----------------
ViennaCL - The Vienna Computing Library
-----------------
Project Head: Karl Rupp rupp@iue.tuwien.ac.at
(A list of authors and contributors can be found in the PDF manual)
License: MIT (X11), see file LICENSE in the base directory
============================================================================= */
//
// include necessary system headers
//
#include <iostream>
//
// Necessary to obtain a suitable performance in ublas
#ifndef NDEBUG
#define BOOST_UBLAS_NDEBUG
#endif
//
// ublas includes
//
#include <boost/numeric/ublas/io.hpp>
#include <boost/numeric/ublas/triangular.hpp>
#include <boost/numeric/ublas/matrix_sparse.hpp>
#include <boost/numeric/ublas/matrix.hpp>
#include <boost/numeric/ublas/matrix_proxy.hpp>
#include <boost/numeric/ublas/operation.hpp>
#include <boost/numeric/ublas/operation_sparse.hpp>
#include <boost/numeric/ublas/io.hpp>
#include <boost/numeric/ublas/lu.hpp>
// Must be set if you want to use ViennaCL algorithms on ublas objects
#define VIENNACL_WITH_UBLAS 1
//
// ViennaCL includes
//
#include "viennacl/scalar.hpp"
#include "viennacl/vector.hpp"
#include "viennacl/compressed_matrix.hpp"
#include "viennacl/coordinate_matrix.hpp"
#include "viennacl/linalg/prod.hpp"
#include "viennacl/linalg/ilu.hpp"
#include "viennacl/linalg/jacobi_precond.hpp"
#include "viennacl/linalg/cg.hpp"
#include "viennacl/linalg/bicgstab.hpp"
#include "viennacl/linalg/gmres.hpp"
#include "viennacl/io/matrix_market.hpp"
// Some helper functions for this tutorial:
#include "vector-io.hpp"
using namespace boost::numeric;
int main()
{
typedef float ScalarType;
ublas::vector<ScalarType> rhs;
ublas::vector<ScalarType> ref_result;
ublas::vector<ScalarType> result;
ublas::compressed_matrix<ScalarType> ublas_matrix;
if (!viennacl::io::read_matrix_market_file(ublas_matrix, "../examples/testdata/mat65k.mtx"))
{
std::cout << "Error reading Matrix file" << std::endl;
return EXIT_FAILURE;
}
if (!readVectorFromFile("../examples/testdata/rhs65025.txt", rhs))
{
std::cout << "Error reading RHS file" << std::endl;
return EXIT_FAILURE;
}
if (!readVectorFromFile("../examples/testdata/result65025.txt", ref_result))
{
std::cout << "Error reading Result file" << std::endl;
return EXIT_FAILURE;
}
std::size_t vcl_size = rhs.size();
viennacl::compressed_matrix<ScalarType> vcl_compressed_matrix;
viennacl::coordinate_matrix<ScalarType> vcl_coordinate_matrix;
viennacl::vector<ScalarType> vcl_rhs(vcl_size);
viennacl::vector<ScalarType> vcl_result(vcl_size);
viennacl::vector<ScalarType> vcl_ref_result(vcl_size);
viennacl::copy(rhs.begin(), rhs.end(), vcl_rhs.begin());
viennacl::copy(ref_result.begin(), ref_result.end(), vcl_ref_result.begin());
viennacl::copy(ublas_matrix, vcl_compressed_matrix);
std::vector< std::map< unsigned int, ScalarType> > stl_matrix(rhs.size());
for (ublas::compressed_matrix<ScalarType>::iterator1 iter1 = ublas_matrix.begin1();
iter1 != ublas_matrix.end1();
++iter1)
{
for (ublas::compressed_matrix<ScalarType>::iterator2 iter2 = iter1.begin();
iter2 != iter1.end();
++iter2)
stl_matrix[iter2.index1()][static_cast<unsigned int>(iter2.index2())] = *iter2;
}
std::vector<ScalarType> stl_rhs(rhs.size()), stl_result(result.size());
std::copy(rhs.begin(), rhs.end(), stl_rhs.begin());
viennacl::copy(stl_matrix, vcl_coordinate_matrix);
viennacl::copy(vcl_coordinate_matrix, stl_matrix);
std::cout << "Setting up preconditioners for uBLAS-matrix..." << std::endl;
viennacl::linalg::ilut_precond< ublas::compressed_matrix<ScalarType> > ublas_ilut(ublas_matrix, viennacl::linalg::ilut_tag());
viennacl::linalg::ilu0_precond< ublas::compressed_matrix<ScalarType> > ublas_ilu0(ublas_matrix, viennacl::linalg::ilu0_tag());
viennacl::linalg::block_ilu_precond< ublas::compressed_matrix<ScalarType>,
viennacl::linalg::ilu0_tag> ublas_block_ilu0(ublas_matrix, viennacl::linalg::ilu0_tag());
std::cout << "Setting up preconditioners for ViennaCL-matrix..." << std::endl;
viennacl::linalg::ilut_precond< viennacl::compressed_matrix<ScalarType> > vcl_ilut(vcl_compressed_matrix, viennacl::linalg::ilut_tag());
viennacl::linalg::ilu0_precond< viennacl::compressed_matrix<ScalarType> > vcl_ilu0(vcl_compressed_matrix, viennacl::linalg::ilu0_tag());
viennacl::linalg::block_ilu_precond< viennacl::compressed_matrix<ScalarType>,
viennacl::linalg::ilu0_tag> vcl_block_ilu0(vcl_compressed_matrix, viennacl::linalg::ilu0_tag());
viennacl::linalg::jacobi_precond< ublas::compressed_matrix<ScalarType> > ublas_jacobi(ublas_matrix, viennacl::linalg::jacobi_tag());
viennacl::linalg::jacobi_precond< viennacl::compressed_matrix<ScalarType> > vcl_jacobi(vcl_compressed_matrix, viennacl::linalg::jacobi_tag());
std::cout << "----- CG Method -----" << std::endl;
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::cg_tag());
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::cg_tag(1e-6, 20), ublas_ilut);
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::cg_tag(1e-6, 20), ublas_jacobi);
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::cg_tag());
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_ilut);
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_jacobi);
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::cg_tag());
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_ilut);
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::cg_tag(1e-6, 20), vcl_jacobi);
std::cout << "----- BiCGStab Method -----" << std::endl;
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::bicgstab_tag()); //without preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), ublas_ilut); //with preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), ublas_jacobi); //with preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::bicgstab_tag()); //without preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_ilut); //with preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_jacobi); //with preconditioner
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::bicgstab_tag());
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_ilut);
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::bicgstab_tag(1e-6, 20), vcl_jacobi);
std::cout << "----- GMRES Method -----" << std::endl;
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::gmres_tag()); //without preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::gmres_tag(1e-6, 20), ublas_ilut);//with preconditioner
result = viennacl::linalg::solve(ublas_matrix, rhs, viennacl::linalg::gmres_tag(1e-6, 20), ublas_jacobi);//with preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::gmres_tag()); //without preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_ilut);//with preconditioner
vcl_result = viennacl::linalg::solve(vcl_compressed_matrix, vcl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_jacobi);//with preconditioner
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::gmres_tag());
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_ilut);
stl_result = viennacl::linalg::solve(stl_matrix, stl_rhs, viennacl::linalg::gmres_tag(1e-6, 20), vcl_jacobi);
std::cout << "!!!! TUTORIAL COMPLETED SUCCESSFULLY !!!!" << std::endl;
return EXIT_SUCCESS;
}