Open daviehh opened 6 months ago
I think it should, as it's a common footgun. Also, matrix-matrix multiplication doesn't protect against this either other than the 2x2 and 3x3 cases.
... and in gemm_wrapper!
. Maybe those aliasing tests got lost a bit over the recent rearrangements?
As per the documentation for
mul!
,https://github.com/JuliaLang/julia/blob/263928f9ad450a28d601c3f185040987e0bc45d5/stdlib/LinearAlgebra/src/matmul.jl#L266-L268
The mwe
is 'protected' against wrong inputs for matrix-matrix product and errors w/
https://github.com/JuliaLang/julia/blob/263928f9ad450a28d601c3f185040987e0bc45d5/stdlib/LinearAlgebra/src/matmul.jl#L653-L655
however, the matrix-vector product a * v is not 'protected' against
C == A || C == B
should a similar alias check be applied for matrix-vector
mul!
at thegemv!
orgeneric_matvecmul!
methods?