[Mlir-commits] [mlir] [mlir] Initial patch to add an MPI dialect (PR #68892)

[Kiran Chandramohan]

kiranchandramohan wrote:

Could you summarise the answers to the questions in the guidelines on contributing a new dialect in the RFC? Apologies, if this has been done already, then a link will be useful.

https://mlir.llvm.org/getting_started/DeveloperGuide/#guidelines-on-contributing-a-new-dialect-or-important-components


> One concern I'm having is the restriction to only work with memref types. While I understand that this is easier to work with, some users might not be able to work with memrefs and thus cannot use this dialect. AFAIK, flang doesn't use memrefs, and neither does code that will come from some lower level input (or the LLVM import).

Yes, a memref only approach will make it not immediately useful in Flang. But that need not be a blocker. Fortran has co-arrays which can potentially lower to MPI.


> The failure in `Lower/OpenMP/parallel-private-clause-fixes.f90` seems unrelated and looks just like bad test (`CHECK-DAG` instead of `CHECK`).

Is this still failing? I thought I had fixed in https://github.com/llvm/llvm-project/commit/a4deb14e353cd8c543f78979ce986f98023d36dd. 

https://github.com/llvm/llvm-project/pull/68892