MI300X build script

[Heinrich-BR]

Added a build script that works for the MI300X partition on the CSD3 system. It builds with HIP-aware OpenMPI.

[Heinrich-BR]

@alexanderianblair the test that's failing has nothing to do with the PR, I believe.

[sean-baccas]

All good from my side, apart from some issues with headers for CEED's JIT backend. Henrique has come up with a fix, which I'm testing atm. But network problems on csd3 are making this very slow

[sean-baccas]

@Heinrich-BR @alexanderianblair Last time I had a go at this, I had an issue with the stdint header not being included in some part of CEED's just-in-time compilation stuff (I can fetch the exact location and error message when CSD3 is back online). The result is that it fails at runtime when running with CEED backend - it tries to do some JIT compilation, and fails with a compilation error. Its problem is something to do with int32_t not being correctly namespaced - the hip compiler was expecting something like hip::int32_t

My solution was to just copy #include <stdint> into the source file where the error occurs. This fixes everything, and does not require recompiling libCEED since it's only an issue at JIT compilation time. This works, but it isn't really something that can be added to this script.

What do you think is a good solution to this? The only thing I can think of is it to:

• Fork petsc
• Set this fork of petsc to pull a fork of libCEED (already made one here)
• Fork moose
• Set this fork of moose to pull the fork of petsc

But this is very ugly. Any ideas?

[sean-baccas]

@Heinrich-BR @alexanderianblair I've finally managed to get the build working on csd3 with these two minor changes. Please apply these changes and I'll be happy!