Open boegel opened 11 years ago
Updating status as of v2.2.0:
Application name and remarks follow:
There is also an apparent need for an extended Python module, covering
Finally, the PRACE CPE is the compatibility layer centers strive to be compliant against:
As per the following, some aliasing work will be required (or it can be done via alt. namespace):
[1] provide an alias to the recommended, highest performance, library or utility, supporting the platform recommended compiler. In case of a library, this might be fully contained in another product. If a recommended vendor library does not fully contain the module file library (including its API), separate libraries must be provided (e.g. IBM ESSL). [2] provide an alias to the recommended compiler, capable of compiling parallel (MPI) programs, with or without OpenMP.
Mockup -untested- versions for some of the missing small items above can be found at: https://github.com/fgeorgatos/easybuild.experimental/blob/master/contrib/pkgsrc/20121226/e/emacs-22.3.eb https://github.com/fgeorgatos/easybuild.experimental/blob/master/contrib/pkgsrc/20121226/n/nedit--src.eb
The PRACE module should automatically bring up as dependencies (ie. behave as an extended toolchain):
c c++ fortran mpi blas lapack blacs scalapack fftw/3 netcdf hdf5
and ensure it adjusts variables (ie. define modextravars
in relation to $EBPRACEDEVEL?):
$PRACE_{FFLAGS,CFLAGS,LDFLAGS,STORE,SCRATCH,ARCH}
Finally, these apps have to be available as modules, too:
For completeness -but not a prerequisite to resolve this issue-, here are the applications selected for the PRACE application benchmark suite:
Ref: http://www.hpcwire.com/hpcwire/2010-02-15/prace_benchmark_suite_finalised.html
Hoi *,
Getting ready to contribute the first iteration of the PRACE environment during the hackathon @ Cyprus: https://github.com/fgeorgatos/easybuild.experimental/tree/master/users/fgeorgatos/PRACE
Toolchains tested: goalf/1.1.0
, ictce/4.1.13
, ictce/5.2.0
We still have some way to go as regards minor tools and also variables, as defined here: http://www.prace-ri.eu/PRACE-Common-Production
Overall, it's looking good but gives some errors with the $EBDEVELPRACE
(it has to be tuned, because the whole point of the PRACE CPE is to eliminate such issues)
The upcoming hackathon in Nicosia next week is a prime opportunity to follow up on the issues, before we hit the trigger for the grand pull request.
Don't be dismayed to give it a try, its status is already pretty good for the majority of users;
fgeorgatos@gaia-10:~/easybuild.experimental/users/fgeorgatos/PRACE$ module load PRACE
fgeorgatos@gaia-10:~/easybuild.experimental/users/fgeorgatos/PRACE$ module list
Currently Loaded Modulefiles:
1) EasyBuild/1.2.0 5) imkl/11.0.2.146 9) Bash/4.2-ictce-5.2.0 13) zlib/1.2.7-ictce-5.2.0 17) Bison/2.5-ictce-5.2.0
2) icc/2013.2.146 6) ictce/5.2.0 10) tcsh/6.18.01-ictce-5.2.0 14) Szip/2.1-ictce-5.2.0 18) Doxygen/1.8.1.1-ictce-5.2.0
3) ifort/2013.2.146 7) make/3.82-ictce-5.2.0 11) Tcl/8.5.12-ictce-5.2.0 15) HDF5/1.8.9-ictce-5.2.0 19) netCDF/4.1.3-ictce-5.2.0
4) impi/4.1.0.030 8) Java/1.7.0_10 12) Tk/8.5.12-ictce-5.2.0 16) flex/2.5.35-ictce-5.2.0 20) PRACE/20130306-ictce-5.2.0
fgeorgatos@gaia-10:~/easybuild.experimental/users/fgeorgatos/PRACE$
fgeorgatos@gaia-10:~/easybuild.experimental/users/fgeorgatos/PRACE$ echo $EBDEVELPRACE
/home/users/fgeorgatos/.local/easybuild/software/PRACE/20130306-ictce-5.2.0/easybuild/PRACE-20130306-ictce-5.2.0-easybuild-devel
enjoy, Fotis
Progress report: We now have 2 PRACE toolchains, which cover both recent Intel and GNU compiler toolkits: https://github.com/fgeorgatos/easybuild.experimental/tree/master/users/fgeorgatos/PRACE PrgEnv business requires more work but the structure I think is relatively correct.
this feels like necroposting, nevertheless here is some newer information in relation to $subject: https://repository.prace-ri.eu/git/UEABS/ueabs#the-application-codes-that-constitute-the-ueabs-are
It basically has explicit instructions/info about the PRACE benchmark suite's components:
Oh they even added TensorFlow
recently to it, see https://repository.prace-ri.eu/git/UEABS/ueabs#tensorflow .
Now it's getting interesting... I've been looking for a decent TensorFlow benchmark for a while now.
(originally by @fgeorgatos)
most of the content on this has been shamelessly copied^H^H^H^H^H^H borrowed from: https://openwiki.uninett.no/metacenter:reference:noturcpe