Open wsshin opened 5 years ago
The allocations are (still) due to (internal) views. For the transpose/adjoint case, we fall back to some 3-arg matrix ldiv!
instead of to the 3-arg ldiv!(y, F::Factorization, x)
. We should perhaps widen the signature of the factorization-based method to handle non-square cases correctly.
ldiv!
with transposed LU factors uses less allocations andldiv!
with the original LU factors. For example,Is this expected behavior?