Skip to content
Snippets Groups Projects
Select Git revision
  • 6.x protected
  • master default protected
  • madysson
  • 5.x protected
  • asm
  • time-varying-information-set
  • 4.6 protected
  • dynare_minreal
  • dragonfly
  • various_fixes
  • 4.5 protected
  • clang+openmp
  • exo_steady_state
  • declare_vars_in_model_block
  • julia
  • error_msg_undeclared_model_vars
  • static_aux_vars
  • slice
  • aux_func
  • penalty
  • 6.3 protected
  • 6.2 protected
  • 6.1 protected
  • 6.0 protected
  • 6-beta2 protected
  • 6-beta1 protected
  • 5.5 protected
  • 5.4 protected
  • 5.3 protected
  • 5.2 protected
  • 5.1 protected
  • 5.0 protected
  • 5.0-rc1 protected
  • 4.7-beta3 protected
  • 4.7-beta2 protected
  • 4.7-beta1 protected
  • 4.6.4 protected
  • 4.6.3 protected
  • 4.6.2 protected
  • 4.6.1 protected
40 results

bytecode.cc

  • Sébastien Villemot's avatar
    7387dfb9
    Better preconditioners for stack_solve_algo={2,3} (GMRES + BiCGStab solvers) · 7387dfb9
    Sébastien Villemot authored
    Implement preconditioners as in UMFITER and ITERSTACK:
    – For UMFITER, on first iteration, do a full LU decomposition, and use it as a
      preconditioner
      in further iterations (similar to UMFITER in TROLL).
    – For ITERSTACK, build a preconditioner based on a small LU decomposition
      repeated several times across the stack
    
    By the way, uniformize the preconditioners for solve_algo={7,8} across the
    different implementations.
    Verified
    7387dfb9
    History
    Better preconditioners for stack_solve_algo={2,3} (GMRES + BiCGStab solvers)
    Sébastien Villemot authored
    Implement preconditioners as in UMFITER and ITERSTACK:
    – For UMFITER, on first iteration, do a full LU decomposition, and use it as a
      preconditioner
      in further iterations (similar to UMFITER in TROLL).
    – For ITERSTACK, build a preconditioner based on a small LU decomposition
      repeated several times across the stack
    
    By the way, uniformize the preconditioners for solve_algo={7,8} across the
    different implementations.