I am actually don't think that there is any point trying implementing fast matrix operations though. There is Intel MKL and the likes for that, all native code