Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Failure on mips64el #2922

Closed
amckinstry opened this issue Feb 3, 2017 · 10 comments
Closed

Failure on mips64el #2922

amckinstry opened this issue Feb 3, 2017 · 10 comments

Comments

@amckinstry
Copy link

We have a major problem at the moment in Debian on mips64el:

https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=848574

Now its quite possible / likely this is a local toolchain issue, but my question: is openmpi 2.0.2 regression tested on mips64el ?
(Its ok on mips, mipsel and other 64-bit archs).

@jsquyres
Copy link
Member

jsquyres commented Feb 3, 2017

AFAIK we have no regresstion testing on MIPS at all. 😬

@ggouaillardet
Copy link
Contributor

@PHHargrove is there any mips64el in your computer zoo ?

fwiw, computer zoo is defined at http://insidehpc.com/2012/11/silicon-darwinism-how-the-computer-zoo-is-charting-the-future-of-cpus/

@PHHargrove
Copy link
Member

@ggouaillardet
Yes, I have both mips64 and mips64el and test thee ABIs (-mabi={32,n32,64}) on each.
All 6 passed my testing of 2.0.2rc4, which included running the ring examples (ring_{c,mpih,usempi,usempif08}) with two ranks on 1 node.
My tests are on Octeon II systems running Debian Wheezy, and are not mine to upgrade.

@amckinstry
Copy link
Author

@jsquyres This appears to work fine with 2.0.2.
However Debian Stretch (9.0) is now in a freeze, and version 2.0.2 has a library version change, requiring a "transition" in Debian (rebuild of everything that uses openmpi, on all platforms).
Which won't happen.

So my best option at the moment is to do 2.0.2 without the change that required the version number bump (just the public interfaces / libraries, not the internal ones). IIRC the Fortran one at least was a trivial API change , but I can't find the ref anymore, can you point me to the changes please?

@hppritcha
Copy link
Member

@amckinstry are you wanting changes between versions 2.0.1 and 2.0.2?

@amckinstry
Copy link
Author

amckinstry commented Feb 9, 2017

@hppritcha Yes
Just the changes that caused the version bumps, not all changes.
(I'm trying a git diff tag/2.0.1 tag/2.0.2 but it would help if someone knew)
.e.g. git diff 467ff3a 7dac5d3
bumps the MPI version but doesn't explain why it was bumped.

@jsquyres
Copy link
Member

jsquyres commented Feb 9, 2017

@amckinstry I'm not sure what you're asking.

Yes, we bumped libtool shared library versions in v2.0.2, but they're all in a backwards-compatible way. I.e., you shouldn't need to recompile applications/libraries dependent upon Open MPI (unless you have some policy that says so...?).

All the bumps in Libtool c values (from c:r:a) also bumped a, meaning that all the changes should be backwards compatible / not require any recompilation of apps or libraries.

Do you mean something else?

@amckinstry
Copy link
Author

Apologies, you are correct. I was confusing major and minor version changes. I can use 2.0.2 as is.

I've uploaded 2.0.2 into Debian for the Stretch release.

@hppritcha
Copy link
Member

Looks like #2280, #2541, #2752 should at least be considered.

@amckinstry
Copy link
Author

@hppritcha Thanks. I'd checked out #2280 already, rebuilding everything in Debian (that uses opepenmpi) against it and it all built as-is.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants