CUDA.jl
CUDA.jl copied to clipboard
\ does not work for least squares
Currently \
only works for square matrices. If I try to solve an overdetemined system (i.e. using least squares), it gives an error.
Example:
A = randn(Float32, 6, 5)
y = randn(Float32, 6)
A \ y # works
gpu(A) \ gpu(y) # gives DimensionMismatch("LU factored matrix A must be square!")
Ideally this should "just work" like it does for normal arrays.
Any movement on this issue?
Nobody has fixed this, no.
Running into this too. Any hints as to where in the source code should one look to attempt to fix this?
Running into this too. Any hints as to where in the source code should one look to attempt to fix this?
I think that the culprit is here.
This seems to work now.