1 #ifndef VIENNACL_LINALG_ITERATIVE_OPERATIONS_HPP_ 
    2 #define VIENNACL_LINALG_ITERATIVE_OPERATIONS_HPP_ 
   37 #ifdef VIENNACL_WITH_OPENCL 
   41 #ifdef VIENNACL_WITH_CUDA 
   58 template<
typename NumericT>
 
   72 #ifdef VIENNACL_WITH_OPENCL 
   77 #ifdef VIENNACL_WITH_CUDA 
   96 template<
typename MatrixT, 
typename NumericT>
 
  107 #ifdef VIENNACL_WITH_OPENCL 
  112 #ifdef VIENNACL_WITH_CUDA 
  133 template<
typename NumericT>
 
  146 #ifdef VIENNACL_WITH_OPENCL 
  151 #ifdef VIENNACL_WITH_CUDA 
  170 template<
typename NumericT>
 
  181     viennacl::linalg::host_based::pipelined_bicgstab_vector_update(result, alpha, p, omega, s, residual, As, beta, Ap, r0star, inner_prod_buffer, buffer_chunk_size);
 
  183   #ifdef VIENNACL_WITH_OPENCL 
  185     viennacl::linalg::opencl::pipelined_bicgstab_vector_update(result, alpha, p, omega, s, residual, As, beta, Ap, r0star, inner_prod_buffer, buffer_chunk_size);
 
  188   #ifdef VIENNACL_WITH_CUDA 
  190     viennacl::linalg::cuda::pipelined_bicgstab_vector_update(result, alpha, p, omega, s, residual, As, beta, Ap, r0star, inner_prod_buffer, buffer_chunk_size);
 
  207 template<
typename MatrixT, 
typename NumericT>
 
  221 #ifdef VIENNACL_WITH_OPENCL 
  226 #ifdef VIENNACL_WITH_CUDA 
  247 template <
typename T>
 
  262 #ifdef VIENNACL_WITH_OPENCL 
  267 #ifdef VIENNACL_WITH_CUDA 
  285 template <
typename T>
 
  298 #ifdef VIENNACL_WITH_OPENCL 
  303 #ifdef VIENNACL_WITH_CUDA 
  320 template <
typename T>
 
  336 #ifdef VIENNACL_WITH_OPENCL 
  341 #ifdef VIENNACL_WITH_CUDA 
  355 template <
typename T>
 
  369 #ifdef VIENNACL_WITH_OPENCL 
  374 #ifdef VIENNACL_WITH_CUDA 
  392 template <
typename MatrixType, 
typename T>
 
  403 #ifdef VIENNACL_WITH_OPENCL 
  408 #ifdef VIENNACL_WITH_CUDA 
void 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. 
void pipelined_bicgstab_prod(compressed_matrix< NumericT > 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)
void 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 param_k, vector_base< T > &vi_in_vk_buffer, vcl_size_t buffer_chunk_size)
Exception class in case of memory errors. 
Generic size and resize functionality for different vector and matrix types. 
Extracts the underlying OpenCL start index handle from a vector, a matrix, an expression etc...
void 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 , i=0..k-1. 
void 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}. 
void 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 , i=0..k-1, then updates v_k -=  v_i and computes the first reduction stage for ||v_k||. 
void 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. 
void pipelined_cg_prod(compressed_matrix< NumericT > const &A, vector_base< NumericT > const &p, vector_base< NumericT > &Ap, vector_base< NumericT > &inner_prod_buffer)
void 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 , i=0..k-1, then updates v_k -=  v_i and computes the first reduction stage for ||v_k||. 
Implementations of specialized kernels for fast iterative solvers using OpenMP on the CPU...
void 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 param_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)
This file provides the forward declarations for the main types used within ViennaCL. 
void 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. 
Determines row and column increments for matrices and matrix proxies. 
void 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. 
void pipelined_bicgstab_prod(compressed_matrix< NumericT > 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)
void 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. 
void pipelined_bicgstab_prod(compressed_matrix< NumericT > 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 fused matrix-vector product with a compressed_matrix for an efficient pipelined BiCGStab a...
void 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. 
void 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...
void 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 param_k)
void 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...
Implementations of operations using sparse matrices using CUDA. 
void 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. 
Implementations of specialized kernels for fast iterative solvers using OpenCL. 
void 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)
void 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)
void 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. 
void 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)
void 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)
void 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 param_k, vector_base< T > &vi_in_vk_buffer, vcl_size_t buffer_chunk_size)
All the predicates used within ViennaCL. Checks for expressions to be vectors, etc. 
void 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)
void 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. 
void pipelined_cg_prod(compressed_matrix< NumericT > const &A, vector_base< NumericT > const &p, vector_base< NumericT > &Ap, vector_base< NumericT > &inner_prod_buffer)
Performs a fused matrix-vector product with a compressed_matrix for an efficient pipelined CG algorit...
void 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 param_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)
void 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 BiCGStab algorithm...
void 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)
void 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}. 
Implementation of a range object for use with proxy objects. 
void pipelined_cg_prod(compressed_matrix< NumericT > const &A, vector_base< NumericT > const &p, vector_base< NumericT > &Ap, vector_base< NumericT > &inner_prod_buffer)
void 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. 
Extracts the underlying OpenCL handle from a vector, a matrix, an expression etc. ...
viennacl::backend::mem_handle & handle(T &obj)
Returns the generic memory handle of an object. Non-const version. 
void 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 param_k)
void 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 first reduction stage for multiple inner products , i=0..k-1. 
Implementation of the ViennaCL scalar class. 
void pipelined_gmres_prod(compressed_matrix< T > const &A, vector_base< T > const &p, vector_base< T > &Ap, vector_base< T > &inner_prod_buffer)
void pipelined_gmres_prod(compressed_matrix< NumericT > const &A, vector_base< NumericT > const &p, vector_base< NumericT > &Ap, vector_base< NumericT > &inner_prod_buffer)
void pipelined_gmres_prod(MatrixType const &A, vector_base< T > const &p, vector_base< T > &Ap, vector_base< T > &inner_prod_buffer)
Simple enable-if variant that uses the SFINAE pattern.