[petsc-users] Obtaining bytes per second

[Matthew Knepley]

On Thu, May 7, 2015 at 1:25 PM, Jed Brown <jed at jedbrown.org> wrote:

> Matthew Knepley <knepley at gmail.com> writes:
>
> > On Thu, May 7, 2015 at 9:23 AM, Justin Chang <jychang48 at gmail.com>
> wrote:
> >
> >> So to summarize, if I understand everything, I should do the following:
> >>
> >> 1) calculate the flop/byte ratio for various problem sizes and solver
> >> methods on one process and:
>
> It is impossible to define "arithmetic intensity" (flops/byte) without
> selecting a cache model.  And my "cache model", I include register
> reuse.  Assuming no reuse at all is not even close to representative
> (off by orders of magnitude).


You are making assumptions about the algorithm here. Pure streaming
computations,
like VecAXPY do not depend on the cache model.

  Matt


>
> > Should be roughly invariant to problem size.
>
> Except (a) insofar as it depends on cache sizes and (b) the algorithm
> characteristics depend on data size (e.g., the cost of supernode
> factorization for sparse direct solvers scales superlinearly, so
> arithmetic intensity will drift as you increase the problem size).
>
>
> You can see how we compare "pessimal" versus "perfect" caching here:
>
> http://59A2.org/na/MayBrownLePourhiet-pTatin3d-2014.pdf
>



-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150507/569d0fc7/attachment.html>