[Pkg-openmpi-maintainers] Release 1.2.3-4

Manuel Prinz debian at pinguinkiste.de
Fri Aug 24 20:28:20 UTC 2007


Hello Craig, Grant, Jens, Thiemo and OpenMPI Maintainers!

You recieve this email because you offered some help with OpenMPI on
Alpha, MIPS and HPPA in an earlier discussion. We hope that your offer
still lasts and you can help us to get OpenMPI build on the archs since
we lack of experience here.

I summarized the discussion in a previous email. You'll find it below.
It would be nice if you could give a status update on that. Thanks in
advance!

Best regards
Manuel


Am Dienstag, den 21.08.2007, 14:18 -0500 schrieb Dirk Eddelbuettel:
> Manuel,
> 
> Thanks *again* for doing all the legwork here. Much appreciated.

Always welcome, it's part of maintaining, after all! ;)

Sorry for not replying earlier, I was kinda busy during the last days...

> On 21 August 2007 at 13:59, Manuel Prinz wrote:
> | Am Montag, den 20.08.2007, 20:46 -0500 schrieb Dirk Eddelbuettel:
> | > We are now down to three open bugs, all on the 'lack of atomistic ops' for
> | > m68k, s390 and hppa as the kfreebsd* variants should build and run.
> | > 
> | > Concerning the 'lack of atomistic ops', we had started a thread on the
> | > openmpi-devel list. I recall that some folks had volunteered to try to get
> | > some of this going where possible.  Would anybody have a moment to collect
> | > that information and follow-up here, and/or on openmpi-devel?
> | 
> | A little bit of history:
> | 
> | According to Brain Barret's mail [1] there has to be some implementation
> | work of atomic wordsize compare and swap and of the memory barrier
> | functions.
> | 
> | George Bosilca [2] proposed disabling threading features on platforms
> | that lack atomic support. Brain agreed to that and stated that compare
> | and swap is used in the LIFO used for free list. Gleb Natapov added that
> | proper memory barrier is needed "for openib BTL eager RDMA support".
> | 
> | In [3] Brain states that it's possible to determine if support atomic
> | operations is available or not. He also says that there's code for Alpha
> | and MIPS, and Jens Seidel offered help on testing on MIPS and MISPEL
> | [4]. Craig Prescott offered help for Alpha [5].
> | 
> | For HPPA, there some issues where discussed. Thiemo Seufer proposed the
> | use of libatomicops [6], Grant Grundler proposed to talk to HP's MPI
> | team.
> 
> So I guess we need to get back to 
> 	Jens Seidel for MIPS
> 	Craig for Alpha
> 	Thiemo/Grant for Hppa
> 
> Do you want to follow-up, maybe with CCs here? 

Done so, as you can see. ;)

> | This is where threads end. I can't see if this lead to some code in the
> | end.
[...]
> | 
> | Footnotes:
> |  1. http://lists.alioth.debian.org/pipermail/pkg-openmpi-maintainers/2007-July/000102.html
> |  2. http://lists.alioth.debian.org/pipermail/pkg-openmpi-maintainers/2007-July/000104.html
> |  3. http://lists.alioth.debian.org/pipermail/pkg-openmpi-maintainers/2007-July/000112.html
> |  4. http://lists.alioth.debian.org/pipermail/pkg-openmpi-maintainers/2007-July/000106.html
> |  5. http://lists.alioth.debian.org/pipermail/pkg-openmpi-maintainers/2007-July/000120.html
> |  6. http://lists.alioth.debian.org/pipermail/pkg-openmpi-maintainers/2007-July/000122.html





More information about the Pkg-openmpi-maintainers mailing list