Default to add '-wdir' arguments #351
Conversation
|
Is the MVIPICH wrapper still used? If so, I could revert the changes and only add ‘-wdir’ arguments if mpi_module does not start with ‘mvapich’ rather than previously which was checking mpi_module starts with |
|
Looks like MPICH supports https://www.mpich.org/static/docs/v3.4.x/www1/mpiexec.html See #350 |
payu/fsops.py
Outdated
| # NOTE: ldd <binary> | ||
|
|
||
| # Example 1: | ||
| # libmpi.so.40 => /apps/openmpi/4.0.2/lib/libmpi.so.40 (0x00007fa493665000) | ||
| # libmpi_usempif08_Intel.so.40 => /apps/openmpi/4.0.2/lib/libmpi_usempif08_Intel.so.40 (0x00007fa492a7c000) | ||
| # libmpi_usempi_ignore_tkr_Intel.so.40 => /apps/openmpi/4.0.2/lib/libmpi_usempi_ignore_tkr_Intel.so.40 (0x00007fa492863000) | ||
| # libmpi_mpifh_Intel.so.40 => /apps/openmpi/4.0.2/lib/libmpi_mpifh_Intel.so.40 (0x00007fa4925cb000) | ||
|
|
||
| # Example 2: | ||
| # libmpi_usempif08.so.40 => $SPACKDIR/opt/spack/linux-rocky8-cascadelake/intel-2019.5.281/openmpi-4.1.5-ooyg5wc7sa3tvmcpazqqb44pzip3wbyo/lib/libmpi_usempif08.so.40 (0x00007f12671c0000) | ||
| # libmpi_usempi_ignore_tkr.so.40 => $SPACKDIR/opt/spack/linux-rocky8-cascadelake/intel-2019.5.281/openmpi-4.1.5-ooyg5wc7sa3tvmcpazqqb44pzip3wbyo/lib/libmpi_usempi_ignore_tkr.so.40 (0x00007f1266fb4000) | ||
| # libmpi_mpifh.so.40 => $SPACKDIR/opt/spack/linux-rocky8-cascadelake/intel-2019.5.281/openmpi-4.1.5-ooyg5wc7sa3tvmcpazqqb44pzip3wbyo/lib/libmpi_mpifh.so.40 (0x00007f1266d44000) | ||
| # libmpi.so.40 => $SPACKDIR/opt/spack/linux-rocky8-cascadelake/intel-2019.5.281/openmpi-4.1.5-ooyg5wc7sa3tvmcpazqqb44pzip3wbyo/lib/libmpi.so.40 (0x00007f12667c3000) |
There was a problem hiding this comment.
In my opinion these examples should be moved into a test of required_libs.
There was a problem hiding this comment.
Ok, I've separating the parsing of ldd output into a separate file so its simpler to test. The other option could be to mock out the ldd call?
jo-basevi
force-pushed
the
341-missing-wdir-arg
branch
from
250303a
to
279fedb
Compare
There was a problem hiding this comment.
Agreed you could mock ldd as an alternative to storing the output in a file, but seems a lot of hassle and I kind of like the resulting design as it isolates functionality, which is not only good for your testing, but allows someone else to call a function like parse_ldd_output independently of running payu, which can help with checking for errors, or ensuring the code is doing what you expect.
| # Our MVAPICH wrapper does not support working directories | ||
| if mpi_module.startswith('mvapich'): | ||
| curdir = os.getcwd() | ||
| os.chdir(self.work_path) | ||
| else: | ||
| curdir = None | ||
|
|
There was a problem hiding this comment.
Just wanted to leave a comment to say deleting this relates directly to #350.
This just can't work with submodels and there is no attempt to test for that, nor do we want to support something that is so limited, so it is being removed.
Co-authored-by: Aidan Heerdegen <[email protected]>
Default to add '-wdir' arguments
Change logic in experiments to default to adding
-wdirto the arguments given to mpirun.Should close #341 and close #350