[Debichem-devel] OpenMM package for Debian/Ubuntu

[Peter Eastman]

Hi Andreas,

> The first and probably major problem is that the build depends on packages in non-free: nvidia-cuda-toolkit.

All of the code that relies on CUDA is in plugins that get loaded dynamically at runtime.  So CUDA is needed to build it, but not to run it.  If CUDA isn't available at runtime, those plugins just don't get loaded.

Peter


On Jan 23, 2014, at 5:15 AM, Andreas Tille <andreas at an3as.eu> wrote:

> Hi,
> 
> On Thu, Jan 23, 2014 at 10:35:48AM +0100, Andreas Tille wrote:
>>> Andreas meant some subset of your group to register on alioth.debian.org
>>> (caveat, your account name will have a suffix "-guest" auto-added and you
>>> need to add this manually for login) and then join Debian Med 
>>> on https://alioth.debian.org/projects/debian-med
>> 
>> s/debian-med/debichem/
>> 
>>> via that conate hyperlink. In a separate email (already deleted, sorry) I 
>>> recall to have read that he moved the OpenMM bits to the Debichem project,
>>> i.e. away from Debian Med, but I did not find it there, yet.
>> 
>> I'm in the middle of moving it to
>> 
>>   git://anonscm.debian.org/debichem/packages/openmm.git
>> 
>> I'll come back either today evening or tomorrow with results and comments.
> 
> Well, it is not really evening ... but here are my remarks:
> 
> I have commited the packaging to DebiChem Git.  It is advisable if those
> (upstream ?) developers who are interested in working on this package
> 
>  a) become a member of 
>       https://alioth.debian.org/projects/debichem
>  b) subscribe this mailing list for communication
>       https://lists.alioth.debian.org/cgi-bin/mailman/listinfo/debichem-devel
>     (we could spare longish CC lists - please confirm if you subscribed)
> 
> For the package I detected two sources of trouble where I need your
> input.  I tweaked your file debian/README.md to document these problems.
> Once they might be solved this file can go anyway because the
> instructions are redundant if the package is in official Debian.
> 
> The first and probably major problem is that the build depends on
> packages in non-free: nvidia-cuda-toolkit.  That means the package can
> only go to contrib.  In case there might be some options to drop the
> non-free accellerator code and provide some general package we might
> consider whether it makes sense to provide a free package without the
> nvidia dependency in main and one linked against nvidia libs in contrib.
> I personally do not have any experience with things like this.
> 
> Moreover currently the package does not even build and this is alsoe
> connected to the nvidia libraries.  I get:
> 
> 
> cd /tmp/buildd/openmm-5.2+dfsg/obj-x86_64-linux-gnu/platforms/cuda/sharedTarget && /usr/bin/c++   -DOPENMM_AUTHORS=\"Peter.Eastman\" -DOPENMM_BUILD_VERSION=0 -DOPENMM_COPYRIGHT_YEARS=\"2008\" -DOPENMM_LIBRARY_NAME=OpenMM -DOPENMM_MAJOR_VERSION=5 -DOPENMM_MINOR_VERSION=2 -DOpenMMCUDA_EXPORTS -g -O2 -fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -fPIC -I/tmp/buildd/openmm-5.2+dfsg/obj-x86_64-linux-gnu/platforms/cuda/src -I/tmp/buildd/openmm-5.2+dfsg/platforms/cuda/src -I/tmp/buildd/openmm-5.2+dfsg/platforms/cuda/./include -I/tmp/buildd/openmm-5.2+dfsg/src -I/tmp/buildd/openmm-5.2+dfsg/libraries/validate/include -I/tmp/buildd/openmm-5.2+dfsg/platforms/reference/include -I/tmp/buildd/openmm-5.2+dfsg/libraries/csha1/include -I/tmp/buildd/openmm-5.2+dfsg/libraries/hilbert/include -I/tmp/buildd/openmm-5.2+dfsg/libraries/lbfgs/include -I/tmp/buildd/openmm-5.2+dfsg/libraries/sfmt/include -I/tmp/buildd/openmm-5.2+dfsg/libraries/lepton/include -I/tmp/buildd/openmm-5.2+dfsg/libraries/quern/include -I/tmp/buildd/openmm-5.2+dfsg/libraries/jama/include -I/tmp/buildd/openmm-5.2+dfsg/olla/include -I/tmp/buildd/openmm-5.2+dfsg/openmmapi/include -I/tmp/buildd/openmm-5.2+dfsg/./include    -DOPENMM_CUDA_BUILDING_SHARED_LIBRARY -o CMakeFiles/OpenMMCUDA.dir/__/src/CudaKernelSources.cpp.o -c /tmp/buildd/openmm-5.2+dfsg/obj-x86_64-linux-gnu/platforms/cuda/src/CudaKernelSources.cpp
> make[3]: *** No rule to make target `/libcuda.so', needed by `libOpenMMCUDA.so'.  Stop.
> make[3]: Leaving directory `/tmp/buildd/openmm-5.2+dfsg/obj-x86_64-linux-gnu'
> make[2]: *** [platforms/cuda/sharedTarget/CMakeFiles/OpenMMCUDA.dir/all] Error 2
> make[2]: Leaving directory `/tmp/buildd/openmm-5.2+dfsg/obj-x86_64-linux-gnu'
> make[1]: *** [all] Error 2
> 
> 
> Any idea how to fix this?
> 
> 
> Finally I dropped some remark about the binary code copy of saxon jars
> inside the code.  I think we can deal with this since saxon is packaged
> for Debian and we just might need to tweak the build system to use the
> Debian packaged library.
> 
> 
> The repository is ready to be used with git-buildpackage.  If you have
> trouble with this you might either consult the Debian Med policy
> document which should sufficiently describe its usage - or for sure
> come back here on this list with your question where I'd be happy to
> patiently guide you through the usage of the needed tools.
> 
> Kind regards
> 
>      Andreas.
> 
> -- 
> http://fam-tille.de