Am Dienstag, den 30.10.2007, 10:33 -0700 schrieb Nicholas Breen: > > as a member of the OpenMPI maintainer group, I'm very interested in OpenMPI > > support for GROMACS. Please find the attached patch to add it. It compiles > > fine in single and double precision modes on my maschine. > > Great! I'd been leaving it out because it didn't compile the last time > I'd checked it (which was admittedly back on openmpi 1.0.2) and wasn't > expecting to be able to add support until GROMACS 4.0. I'll be happy to > add support sooner.
I had trouble compiling GROMACS with OpenMPI for quite a while. I'm quite happy as well that it finally works. :) > I think debian/rules will require some further modification to keep it > from failing when calling the build target on the architectures where > OpenMPI is unavailable. It'll be a few days before I have time to look > at it. I didn't think about that, so I missed it. Anyway, there's no hurry. > Out of personal curiosity, are the remaining architectures likely to be > supported by the release of lenny? (I'm perhaps the only person that > tries running GROMACS on a MIPS box occasionally, mostly to check for > glaringly obvious non-i386 problems.) I don't think so. They're not supported by upstream and fail because there are no atomic primitives provided. See bugs #376833, #389306 and #405929, as well as the mailing list discussion on that. (Those are essentially the same bugs.) Upstream has no possibility to provide and test those, so we're kind of stuck until someone steps up and hacks some assembler routines. As an aside, you may have noticed that MPICH was given away for adoption. AFAIK it's removed from testing for a while now as well. This might affect the GROMCAS package in the (near?) future. Best regards Manuel
signature.asc
Description: Dies ist ein digital signierter Nachrichtenteil
