On Wed, Nov 23, 2011 at 9:55 AM, Ross Walker <rosscwalker.gmail.com> wrote:
> This is also, I believe an issue with using git and cherry picking to make
> bugfixes rather than the old fashioned way of making them manually. Git
> commits do not tend to be atomic, building on previous commits that added
> new features etc and not just fixing bugs. Thus for fairly complex bugfixes
> it is necessary to go through and manually back port the fix to amber 11,
> working around all the new features etc that have been added to amber 12 in
> the meantime.
>
> The current bugfix.19 scares me since clearly the bugfixes amber11 was not
> even built, let alone having the test cases run to make sure the fix was
> good and did not break a ton of stuff in the older code.
>
It was actually tested. I must have just left out the CUDA tests. I have
the bug fix ready, I just have to test to make sure that the tests pass
(but now it at least compiles). However, this is an issue with
cherry-picking. I was under the impression that all of the changes were
stored in that single commit, hence the error. It's a 1-line fix that I
was already planning to replace the patch in-place.
As I told Ross, though, patch_amber.py will not appreciate in-place patch
changes (since it doesn't compare the online patches with the applied ones
to make sure they haven't changed), so this shouldn't be done in the future.
Also, I have an AmberTools 1.5 + Amber11 with bug fixes repository that I
created and am storing locally on our git server (which is what I've been
using to test builds). If desired, I can just go ahead and push this
branch to git.ambermd.org if the powers that be create an amber11.git
repository in gitosis.
I'll get the patch ready to go later today.
All the best,
Jason
> :-(
>
> On Nov 23, 2011, at 5:59, Mark Williamson <mjw.mjw.name> wrote:
>
> > Dear all,
> >
> > Seeing this on a CUDA compile with the latest bug fixes (23/11/2011):
> >
> > mpicc -ip -O3 -no-prec-div -xHost -DMPICH_IGNORE_CXX_SEEK
> > -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -DBINTRAJ -DMPI -DCUDA -DMPI
> > -DMPICH_IGNORE_CXX_SEEK -I/usr/local/Cluster-Apps/cuda/3.2/cuda/include
> > -IB40C -IB40C/KernelCommon
> > -I/usr/local/Cluster-Apps.nehalem/mvapich2/1.6/ofa/include -c gpu.cpp
> > gpu.cpp(445): error: class "_gpuContext" has no member "totalMemory"
> > if (gpu->totalMemory < 2 * 1024 * 1024 * 1024)
> > ^
> >
> > gpu.cpp(445): warning #61: integer operation result is out of range
> > if (gpu->totalMemory < 2 * 1024 * 1024 * 1024)
> >
> > I think this is incorrect; totalMemory does not exist, it should either
> > be: totalGPUMemory or totalCPUMemory. Is this something local to me or
> > have these patches not been tested before being released? Secondly, the
> > subsequent warning sounds a bit worrying as well.
> >
> > Seeing that the cruise control framework is now in place, it might be an
> > idea to introduce a test that automatically generates a patch from the
> > amber11-with-patches and at15-with-patches git branches and attempts to
> > apply them to the tagged amber11/at15 release branch and then builds and
> > tests etc. This should hopefully prevent the above from happening.
> >
> > Regards,
> >
> > Mark
> >
> > _______________________________________________
> > AMBER-Developers mailing list
> > AMBER-Developers.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber-developers
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Nov 23 2011 - 07:30:03 PST
