Closed shanedsnyder closed 3 years ago
In GitLab by @jenkins on Sep 29, 2016, 10:40
Added 1 commit:
In GitLab by @shanedsnyder on Oct 3, 2016, 16:06
One question after looking through the code some: Is there a particular reason the margo initialization isn't handled in ssg_init_mpi()
like it is in ssg_init_config()
? I always assumed that a user could just pick whichever initialization method they prefer, and that upon return internal SSG state would more or less be the same (other than possibly different mappings between addresses and ranks). Comparing the 2 ssg_init
implementations, they appear to be doing more or less the same thing, except for some subtleties in the setting up of the output ssg_t
data structure (including the margo initialization I mention). There might be a good reason for this that I'm missing, but if not, we should make sure both calls leave SSG in a similar state?
Other than that, I think the API looks fine. If you want to merge it over, I can make the changes to the swim repo to get it back in sync.
In GitLab by @jenkins on Oct 3, 2016, 16:23
Added 1 commit:
In GitLab by @jenkins on Oct 3, 2016, 16:24
Thanks for the catch! The lack of margo initialization was an oversight, I've pushed in a fix.
I'll merge it momentarily.
In GitLab by @jenkins on Oct 3, 2016, 16:25
Mentioned in commit 5edcac2d9d0fbc42eddf85c82812a2628a67cd0f
In GitLab by @shanedsnyder on Oct 3, 2016, 16:28
Great-- thanks!
In GitLab by @jenkins on Sep 28, 2016, 13:54
Merges margo-additions -> master
I beefed up the margo-related part of ssg, including a "barrier" call (with naive implementation). I also tweaked the example programs. @shanedsnyder, mind taking a quick look at the API and see if anything's goofy?
Also, I changed the name of ssg to "Simple Static Group" here - it will go into master one way or another.