| Re: [eigen] Re: tests taking forever when run with -j8 and OpenMP is enabled |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
On Sat, 3 Jul 2010 18:59:21 +0300, Aron Ahmadia <aja2111@xxxxxxxxxxxx>
wrote:
> In the world of scientific computing, your code is usually assumed to be
> the
> only process on the system, so this is normally a fairly safe assumption
to
> make. In fact, the OpenMP library has an API which allows you to ask
how
> many cores are on the physical CPU you are running on, which usually
allows
> you to leverage the entire machine. I think for 99% of the users you
> encounter who are using Eigen and care about this sort of thing, they
will
> want you to utilize the memory hierarchy as if you have the whole
machine
> cache (L1-L3) at your disposal.
>
> A
Unfortunately, I think the optimal cores & caches usage is somewhat more
complicated that
giving all the ressources (i.e. cores/caches ) to Eigen operations because
of the following
usecases even when the running process is the only "important" one (as in
"performance constrained"):
1°) you can have n concurrent Eigen operations in a n threads
2°) other threads in the process might also be important
So I think it would be interesting to be able to tune at runtime the nb of
cores allocated to various
Eigen operations. Of course it is a difficult problem because for maximal
efficiency it
would be best to select amongst near of far (i.e. sharing or not L2 cache)
core at runtime
and OpenMP did not allow that last time I checked ☹:.
my .02€
Best Regards,
B.
[0] and a single global setNbThreads() is unsatisfactory because it is not
threadsafe itself.