I could not get accurate description that aids to my understanding of inner blocking size parameter that is used in many LAPACK routines(such as DGEQRT, DLARFB).


  • $\begingroup$ Could you elaborate a little more? What don't you understand about the inner blocking size parameter? Right now, it seems like you posed a statement, rather than a question. $\endgroup$ – Geoff Oxberry Dec 5 '12 at 2:36
  • $\begingroup$ My question in particular is about, how is inner block size different from Block size? $\endgroup$ – piyush_sao Dec 5 '12 at 14:56

You achieve ideal blocking, if your block fits in your cache. If it does not, it will cost you way more performance. So you have to fit the block size to cache size.

  • $\begingroup$ Can you elaborate a bit more? $\endgroup$ – piyush_sao Dec 4 '12 at 14:18
  • $\begingroup$ Have a look at stackoverflow.com/questions/5200338/… $\endgroup$ – vanCompute Dec 4 '12 at 20:08
  • $\begingroup$ This paper has a short introduction: suif.stanford.edu/papers/lam-asplos91.pdf $\endgroup$ – vanCompute Dec 4 '12 at 20:10
  • $\begingroup$ How is it different from Block size? Why dgemm doesn't use the inner blocking parameter while dgeqrt , dlarfb uses? $\endgroup$ – piyush_sao Dec 5 '12 at 18:16
  • $\begingroup$ This is a very specific question. I am sorry, but I don't have time to dig deeper. I have a link for you: math-atlas.sourceforge.net This is ATLAS a library for automatic tuning, just in case you find that interesting ;-) $\endgroup$ – vanCompute Dec 8 '12 at 10:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.