Very nice write up. Those are the kind of matrix sizes that MKL is fairly good at, might be worth a comparison as well?
Also, if you were designing for smaller cases, say MNK=16 or 32, how would you approach it differently? I'm implementing neural ODEs and this is one point I've been considering.
Also, if you were designing for smaller cases, say MNK=16 or 32, how would you approach it differently? I'm implementing neural ODEs and this is one point I've been considering.