You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It would be good (and I would hope not too difficult) to support muting LinearAlgebra methods that are related to taking linear combinations of arrays: mul!(y, x, scalar) axpy!(scalar, x, y) axpby!(scalar, x, scalar, y)
where scalar and also x and y can have different types.
Currently these operators fall back to their generic implementation in Base, which of course uses scalar indexing and is thus forbidden.
Not sure if this needs to go into CuArrays.jl or in GPUArrays.jl.
I encounter this in KrylovKit.jl, where complex valued linear combinations of the real-valued Schur vectors are taken in order to build the eigenvectors: Jutho/KrylovKit.jl#15
The text was updated successfully, but these errors were encountered:
I think many of these are implemented now, with scalar input types having been widened to Number. If there's any missing, feel free to open another issue.
It would be good (and I would hope not too difficult) to support muting
LinearAlgebra
methods that are related to taking linear combinations of arrays:mul!(y, x, scalar)
axpy!(scalar, x, y)
axpby!(scalar, x, scalar, y)
where
scalar
and alsox
andy
can have different types.Currently these operators fall back to their generic implementation in Base, which of course uses scalar indexing and is thus forbidden.
Not sure if this needs to go into CuArrays.jl or in GPUArrays.jl.
I encounter this in KrylovKit.jl, where complex valued linear combinations of the real-valued Schur vectors are taken in order to build the eigenvectors: Jutho/KrylovKit.jl#15
The text was updated successfully, but these errors were encountered: