rework default_householder_driver

Author: lkdvos

ext/MatrixAlgebraKitCUDAExt/MatrixAlgebraKitCUDAExt.jl

@@ -29,7 +29,6 @@ end
 # include for block sector support
 const BlockView{T, A} = Base.ReshapedArray{T, 2, SubArray{T, 1, A, Tuple{UnitRange{Int}}, true}}
 
-MatrixAlgebraKit.default_householder_driver(::BlockView{T, A}) where {T <: BlasFloat, A <: CuVecOrMat{T}} = CUSOLVER()
 function MatrixAlgebraKit.default_svd_algorithm(::Type{BlockView{T, A}}; kwargs...) where {T <: BlasFloat, A <: CuVecOrMat{T}}
     return CUSOLVER_Jacobi(; kwargs...)
 end

src/interface/decompositions.jl

@@ -81,8 +81,14 @@ Depending on the driver, various other keywords may be (un)available to customiz
 @algdef Householder
 
 default_householder_driver(A) = Native()
+
 default_householder_driver(::YALAPACK.MaybeBlasMat) = LAPACK()
 
+# note: StridedVector fallback is needed for handling reshaped parent types
+default_householder_driver(::StridedVector{<:BlasFloat}) = LAPACK()
+default_householder_driver(A::Union{SubArray, Base.ReshapedArray}) = default_householder_driver(parent(A))
+
+
 # General Eigenvalue Decomposition
 # -------------------------------
 """