[SciPy-user] weave.inline with mpi
Mon Mar 16 16:09:39 CDT 2009
I am using mpi4py on a set of nodes, some which share a local file
system and some that don't. They all share a slower, global
filesystem. My script uses several pieces of weave.inline'd code.
Do I need to worry about weave code generation running into problems
in this setting? Is there a way to specify the location of the
generated extension and have the root node generate it and force the
other nodes to use that extension as well?
The code generation is insignificant compared to the run time of the
application. It would be ok if the code was generated on each node
each time I ran the application. Is there a way to have weave generate
a unique extension and ignore what is currently on the local or global
filesystem, once per mpi process?
More information about the SciPy-user