ViennaCL - The Vienna Computing Library  1.7.0
Free open-source GPU-accelerated linear algebra and solver library.
iterative_operations.hpp File Reference

Implementations of specialized routines for the iterative solvers. More...

Go to the source code of this file.

Namespaces

 viennacl
 Main namespace in ViennaCL. Holds all the basic types such as vector, matrix, etc. and defines operations upon them.
 
 viennacl::linalg
 Provides all linear algebra operations which are not covered by operator overloads.
 

Functions

template<typename NumericT >
void viennacl::linalg::pipelined_cg_vector_update (vector_base< NumericT > &result, NumericT alpha, vector_base< NumericT > &p, vector_base< NumericT > &r, vector_base< NumericT > const &Ap, NumericT beta, vector_base< NumericT > &inner_prod_buffer)
 Performs a joint vector update operation needed for an efficient pipelined CG algorithm. More...
 
template<typename MatrixT , typename NumericT >
void viennacl::linalg::pipelined_cg_prod (MatrixT const &A, vector_base< NumericT > const &p, vector_base< NumericT > &Ap, vector_base< NumericT > &inner_prod_buffer)
 Performs a joint vector update operation needed for an efficient pipelined CG algorithm. More...
 
template<typename NumericT >
void viennacl::linalg::pipelined_bicgstab_update_s (vector_base< NumericT > &s, vector_base< NumericT > &r, vector_base< NumericT > const &Ap, vector_base< NumericT > &inner_prod_buffer, vcl_size_t buffer_chunk_size, vcl_size_t buffer_chunk_offset)
 Performs a joint vector update operation needed for an efficient pipelined CG algorithm. More...
 
template<typename NumericT >
void viennacl::linalg::pipelined_bicgstab_vector_update (vector_base< NumericT > &result, NumericT alpha, vector_base< NumericT > &p, NumericT omega, vector_base< NumericT > const &s, vector_base< NumericT > &residual, vector_base< NumericT > const &As, NumericT beta, vector_base< NumericT > const &Ap, vector_base< NumericT > const &r0star, vector_base< NumericT > &inner_prod_buffer, vcl_size_t buffer_chunk_size)
 Performs a joint vector update operation needed for an efficient pipelined BiCGStab algorithm. More...
 
template<typename MatrixT , typename NumericT >
void viennacl::linalg::pipelined_bicgstab_prod (MatrixT const &A, vector_base< NumericT > const &p, vector_base< NumericT > &Ap, vector_base< NumericT > const &r0star, vector_base< NumericT > &inner_prod_buffer, vcl_size_t buffer_chunk_size, vcl_size_t buffer_chunk_offset)
 Performs a joint vector update operation needed for an efficient pipelined CG algorithm. More...
 
template<typename T >
void viennacl::linalg::pipelined_gmres_normalize_vk (vector_base< T > &v_k, vector_base< T > const &residual, vector_base< T > &R_buffer, vcl_size_t offset_in_R, vector_base< T > const &inner_prod_buffer, vector_base< T > &r_dot_vk_buffer, vcl_size_t buffer_chunk_size, vcl_size_t buffer_chunk_offset)
 Performs a vector normalization needed for an efficient pipelined GMRES algorithm. More...
 
template<typename T >
void viennacl::linalg::pipelined_gmres_gram_schmidt_stage1 (vector_base< T > const &device_krylov_basis, vcl_size_t v_k_size, vcl_size_t v_k_internal_size, vcl_size_t k, vector_base< T > &vi_in_vk_buffer, vcl_size_t buffer_chunk_size)
 Computes the first reduction stage for multiple inner products <v_i, v_k>, i=0..k-1. More...
 
template<typename T >
void viennacl::linalg::pipelined_gmres_gram_schmidt_stage2 (vector_base< T > &device_krylov_basis, vcl_size_t v_k_size, vcl_size_t v_k_internal_size, vcl_size_t k, vector_base< T > const &vi_in_vk_buffer, vector_base< T > &R_buffer, vcl_size_t krylov_dim, vector_base< T > &inner_prod_buffer, vcl_size_t buffer_chunk_size)
 Computes the second reduction stage for multiple inner products <v_i, v_k>, i=0..k-1, then updates v_k -= <v_i, v_k> v_i and computes the first reduction stage for ||v_k||. More...
 
template<typename T >
void viennacl::linalg::pipelined_gmres_update_result (vector_base< T > &result, vector_base< T > const &residual, vector_base< T > const &krylov_basis, vcl_size_t v_k_size, vcl_size_t v_k_internal_size, vector_base< T > const &coefficients, vcl_size_t k)
 Computes x += eta_0 r + sum_{i=1}^{k-1} eta_i v_{i-1}. More...
 
template<typename MatrixType , typename T >
void viennacl::linalg::pipelined_gmres_prod (MatrixType const &A, vector_base< T > const &p, vector_base< T > &Ap, vector_base< T > &inner_prod_buffer)
 Performs a joint vector update operation needed for an efficient pipelined GMRES algorithm. More...
 

Detailed Description

Implementations of specialized routines for the iterative solvers.

Definition in file iterative_operations.hpp.