I would really like this too and have looked through base but haven't found anything. For example, the gemm_wrapper <https://github.com/JuliaLang/julia/blob/8f4d5868e17b4ac40e09d58f96ec91d0f8a001df/base/linalg/matmul.jl#L304> hardcodes alpha = 1 and beta = 0. What I want it for is to extend Simon's nice package InplaceOps.jl <https://github.com/simonbyrne/InplaceOps.jl> to not only support @into! A = B * C but also accumulation like @into! A += alpha * B * C. However, since there is no A_mul_B with multipliers I need to go straight for the Blas calls and basically implement the whole A_mul_B dispatch all over again.
I guess it would be possible to go through base and add optional alpha and beta parameters to whatever A_mul_B methods that could support it... On Monday, September 14, 2015 at 8:28:07 PM UTC+2, Douglas Bates wrote: > > I want to perform an operation like the BLAS.gemm! operation with > multipliers α and β but for my own custom matrix type. The name gemm! is > unsuitable because the ge part indicates general matrices. I can, of > course, choose my own name for the generic but I thought I would ask if > there was already a name for such an operation. > > In more detail, the BLAS.gemm! function returns the target, C, as C := > α*A*B + β*C where α and β are scalars. (You can also specify transpose or > not on A and/or B). The most common case is C := A*B, corresponding to α = > 1 and β = 0 but other cases are possible and often convenient. > > >
