[Mlir-commits] [mlir] [mlir][mpi] Lowering Mpi To LLVM (PR #127053)

[Frank Schlimbach]

fschlimb wrote:

> > Any objections to addressing this in a separate PR?
> 
> IIUC, the current lowering pass changes behavior depending on the MPI version that was around during build time. If that is really the case, then the test should not work properly depending on the build setting, no?

The tests are written in a way, that the differences between OpenMPI and MPICH* are hidden. This is mostly about different constant values. @AntonLydike also found places where details of checked operations are missing (and only the operation name itself gets verified). 

https://github.com/llvm/llvm-project/pull/127053