[concurrency-interest] Matrix multiply with parallelized inner product

Joe Bowbeer joe.bowbeer at gmail.com
Mon Feb 4 19:27:36 EST 2008


On Feb 4, 2008 4:15 PM, Tim Peierls wrote:
> On Feb 4, 2008 5:43 PM, Joe Bowbeer wrote:
>
> > If you have a bunch of PAs using the same executor, I think the number
> > of available processors is effectively reduced, and therefore the
> > effectiveness of this heuristic is reduced.  It would be better to
> > have each thread operate on larger chunks than to divide the work up
> > into more tasks that will only have to wait for threads to become
> > available.
> >
>
> How about, "Avoid nesting PA calls unless the number of processors available
> dominates the parallelism opportunities in the outer level" ?
>

I'm not really sure what that means.

Won't the outer layer always try to load all the processors?  To be
most efficient, I think the outer layer should try to save some
processors for the inner layers.

Doug??


More information about the Concurrency-interest mailing list