Updates to Isambard-3 spack env and purify benchmark #356
Conversation
| py-numpy: | ||
| externals: | ||
| - spec: [email protected] | ||
| modules: | ||
| - cray-python | ||
| prefix: /opt/cray/pe/python/3.11.7 | ||
| buildable: false | ||
| py-scipy: | ||
| externals: | ||
| - spec: [email protected] | ||
| modules: | ||
| - cray-python | ||
| prefix: /opt/cray/pe/python/3.11.7 | ||
| buildable: false | ||
| py-mpi4py: | ||
| externals: | ||
| - spec: [email protected] | ||
| modules: | ||
| - cray-python | ||
| prefix: /opt/cray/pe/python/3.11.7 | ||
| buildable: false |
There was a problem hiding this comment.
Any reason for using buildable: false for this packages? Also, I believe you want also to append ^ [email protected] to all these specs, to avoid making it match with other versions of python.
There was a problem hiding this comment.
I thought buildable: false guarantees that the external will always get used. Is it too strong requirement? Happy to remove it if you think that's better. Adding the python version to the spec
There was a problem hiding this comment.
buildable: false complicates environment resolution quite a bit whenever a package appears in the environment. I'd expect troubles for py-numpy and py-scipy because they're somewhat common, py-mpi4py less so and I'd hope spack links it to the right MPI library anyway.
There was a problem hiding this comment.
👍 thanks for explaining. My builds seem fine after removing it.
There was a problem hiding this comment.
For example, one problem could be the following: for some reason a package in the environment requires a specific version of python and that's not compatible with 3.11.7 (say 3.10 or 3.12), another package needs py-numpy, then you end up in the situation where you can't concretize the environment at all. We do use buildable: false for MPI, but we do really want to enforce using system MPI libraries only, I don't see that requirement being useful for python and python packages.
Updates purify benchmark to use the latest 5.0 version, adds new benchmarks for the ForwardBackward algorithm and add some sanity checks to parameters.
Adds python packages to isambard-3 spack env
I've commented out external cray-fftw because of the fftw bug that breaks cmake builds. The bug is patched in the spack package so building your own fftw works just fine.