Results inconsistent with LinearAlgebra.mul!

[GiggleLiu]

MWE

julia> using LinearAlgebra

julia> LinearAlgebra.mul!(fill(NaN, 2, 2), randn(2,2), randn(2,2), 1, 0)
2×2 Matrix{Float64}:
  0.416567   0.0524288
 -0.851354  -3.77905

julia> using Octavian

julia> Octavian.matmul!(fill(NaN, 2, 2), randn(2,2), randn(2,2), 1, 0)
2×2 Matrix{Float64}:
 NaN  NaN
 NaN  NaN

It is not an issue in most cases, however, due to the fact that similar sometimes can create an array with NaNs inside. Octavian.matmul!(similar(a, size(a, 1), size(b, 2)), a, b, true, false) does not correspond to matrix multiplication. Related issue is https://github.com/TensorBFS/TropicalGEMM.jl/issues/16

Since I bind (pirate) the Octavian.matmul! to matrix multiplication in TropicalGEMM, so I need to fix it anyway. Wondering if I should fix it in Octavian or TropicalGEMM?

[chriselrod]

Octavian uses static(0) as the multiplier, but it could add a check.

Might be worth finding out at which point false gets promoted to 0.0.

[GiggleLiu]

Thanks, it makes sense.