Use f-ordered result array and a more comparable gemm invocation for ND4J #46
Thank you for this update. I'll add gemm in addition to mmul, instead of replacing it. Neanderthal's mm does much more than gemm. Being polymorphic not only regarding the first matrix, but also the second, it does lots of things. gemm is, on the other hand, a static method that straightforwardly does gemm. Nevermind, Neanderthal still comes out ahead, so why not accept gemm? :)
I'll add your changes as addition to the existing code outside this PR. When can I expect the release of the beta2 of ND4J, so I can update the project to use ND4J for the recent bugfixes related to this? Snapshot does not work for me, and will probably not work for other users, since it picks up OpenBLAS instead of MKL.
With snapshots, and probably with any upcoming versions you've got to explicitly tell ND4J (or rather JavaCPP which can pickup any BLAS Library via the Openblas Bindings) what you'd like to use:
For my own tests I used
Concerning the next release, all that I'm aware of is that it isn't going to be very soon.