[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: SSE gemm question


>The latest cleanup (not released yet) does considerably better (see
>earlier email), but no matter how you slice it, the cleanup is going
>to take a sizeable hit, and not because of m and n runtime.  The chief
>single factor is that you no longer have any data alignment with odd
>sized ldb/lda, at least as far as I can tell.  Is there *anything* we
>would know about the alignment in cleanup cases?

So, not knowing alignment is only true for K-cleanup (where you know nothing).
For M & N cleanup, lda=ldb=NB as usual, so things are golden . . .