Support for non-Vector v

[CorbinFoucart]

Is it the case that while the linear operator may be a custom type, v must be a vector? I am implementing a shared-memory code that would be much more efficient if I were able to iterate between a custom type and a v::SharedArray rather than a v::Vector.

If this is not possible, is it something that the developers are interested in implementing? It is possible that I could work on it.

[andreasnoack]

I don't think the solvers are restricted to Vector. Have you tried with a SharedVector?

[lopezm94]

Some solvers do restrict, not explicitly but by using KrylovSubspace.init!. For example cg!:

function init!{T}(K::KrylovSubspace{T}, v::Vector{T})
#    K.v = Vector{T}[all(v.==zero(T)) ? v : v/norm(v)]
    K.v = Vector{T}[copy(v)]
end

Trying to use SharedArray throws an error:

julia> s = SharedArray(Float64,(10); init=true);

julia> cg!(s,rand(10,10),rand(10))
ERROR: MethodError: no method matching init!(::IterativeSolvers.KrylovSubspace{Float64,Array{Float64,2}}, ::SharedArray{Float64,1})
Closest candidates are:
  init!{T}(::IterativeSolvers.KrylovSubspace{T,OpT}, ::Array{T,1}) at /home/juan/.julia/v0.5/IterativeSolvers/src/krylov.jl:172
 in #cg!#38(::Array{Any,1}, ::Function, ::SharedArray{Float64,1}, ::Array{Float64,2}, ::Array{Float64,1}) at /home/juan/.julia/v0.5/IterativeSolvers/src/cg.jl:11
 in cg!(::SharedArray{Float64,1}, ::Array{Float64,2}, ::Array{Float64,1}) at /home/juan/.julia/v0.5/IterativeSolvers/src/cg.jl:10

This is not the case for gmres!:

julia> gmres!(s,rand(10,10),rand(10))
10-element SharedArray{Float64,1}:
  31.7301 
 -68.2552 
  22.5307 
 -20.5499 
   3.4108 
   6.11011
 -28.9851 
   4.63121
  62.258  
  13.8198

In this case KrylovSubspace is initialized by wrapping the vector in a function.