KLU

A linear solver for sparse circuit simulation problems. This package wraps the SuiteSparse solver KLU.

Installation

using Pkg
Pkg.add("KLU")

Usage

julia> using KLU
julia> using SparseArrays
julia> A = sprand(15, 15, 0.45);
julia> factor = klu(A);
julia> b = rand(15);
julia> x = factor \ b
15-element Vector{Float64}:
  22.501711574339556
  27.80357152530039
 -37.08488119597194
  61.470604418791346
  -0.49109038415655193
  38.03163985997178
   1.2417021657549396
 -51.154984106483155
   0.9488554483842341
   1.587572050319512
 -31.251127382496147
   1.4508846663863602
 -25.866070230462796
  -5.46661465074706
  10.688731342850245

Thread safety

KLU's C library keeps scratch workspace inside each numeric factorization, so two simultaneous solves (or a solve and a refactorization) on the same KLUFactorization would corrupt each other. KLU.jl therefore guards every libklu call with a lock stored in the factorization object: sharing one factorization between threads is safe, but those calls are serialized. Distinct factorizations share no state and run fully in parallel. To solve many right-hand sides against one matrix in parallel, batch them as the columns of a single B in one solve!/ldiv! call, or give each task its own klu(A). The lock is reentrant and can be held by the user across a sequence of calls with Base.@lock F ....

Citing this package

Please cite both SuiteSparse as well as KLU.