Parametrize vector in BlockDiagonal

[theogf]

Right now the block matrices are hard-coded into a Vector, this makes it too restrictive for some actions like passing a BlockDiagonal to the GPU.
Parametrizing it as

struct BlockDiagonal{T, V<:AbstractMatrix{T}, AV<:AbstractVector{V}} <: AbstractMatrix{T}
    blocks::AV

    function BlockDiagonal{T, V, AV}(blocks::AV) where {T, V<:AbstractMatrix{T}, AV<:AbstractVector{V}}
        return new{T, V, AV}(blocks)
    end
end

solve this issue.

[codecov]

Codecov Report

Merging #38 into master will decrease coverage by 0.20%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #38      +/-   ##
==========================================
- Coverage   95.58%   95.37%   -0.21%     
==========================================
  Files           4        4              
  Lines         181      173       -8     
==========================================
- Hits          173      165       -8     
  Misses          8        8              

Impacted Files	Coverage Δ
src/blockdiagonal.jl	82.92% <100.00%> (-0.80%)	⬇️
src/base_maths.jl	100.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 9fdcb28...b7da683. Read the comment docs.

[dkarrasch]

Another option is #33. Would that also solve your issue with the GPU?

[theogf]

Aren't Tuple very inefficient for a large number of elements?

[dkarrasch]

Aren't Tuple very inefficient for a large number of elements?

Could be. I simply don't know how that balances with possibly having an abstract eltype for the AbstractVector. Could be that the multiplication cost dominates anyway.

[theogf]

So the best approach would be to get a Vector of matrices. Check if the matrices are homogeneous, if yes use a vector, if not use a Tuple. I need to check if it would also work when passing it to the GPU (but I think yes)

[dkarrasch]

In fact, this would be relevant over in LinearMaps.jl as well, see JuliaLinearAlgebra/LinearMaps.jl#84 (comment). One could leave it to the user:

struct BlockDiagonal{T, V} <: AbstractMatrix{T}
     blocks::V
end

and then restrict V via available constructors?

[theogf]

Closing this one for a new one using @dkarrasch's comment