On Thu, 9 Jul 2020, Litvinov, Vass wrote:
`a.out --dataParTasksPerLocale=... .....` controls the degree of data
parallelism for standard data structures. It sets the same restriction
for each locale.
Thanks Vass. And Thanks Greg.
https://chapel-lang.org/docs/language/spec/data-parallelism.html?#configuration-constants-for-default-data-parallelism
Thanks. I was fishing in Task parallelism.
Exactly what I was looking for.
Or are you concerned with restricting the hardware parallelism
available?
No. I just want a way to alter the number of tasks being run during any
forall loop in my program doing an SVD matrix decomposition.
I have run tests in serial mode on a 6-core, 36-core and 52-core systems.
And I have parallel results using all cores on each of those systems. I
was curious if the parallelism overhead was a function of the number of
cores. Mind you, it might just be the version of Chapel. The 36-core and
52-core systems have a high overheads. The 6-core has next to none. But
then the 6-core runs 1.20 and the others 1.22.
Thanks again - Damian
Pacific Engineering Systems International, 277-279 Broadway, Glebe NSW 2037
Ph:+61-2-8571-0847 .. Fx:+61-2-9692-9623 | unsolicited email not wanted here
Views & opinions here are mine and not those of any past or present employer
_______________________________________________
Chapel-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/chapel-developers
