[mlir] [llvm] [mlir][mesh] Add spmdization pass (PR #80518)

[Chengji Yao via llvm-commits]

================
@@ -0,0 +1,131 @@
+// RUN: mlir-opt --mesh-spmdization --test-constant-fold %s | FileCheck %s
----------------
yaochengji wrote:

I would suggest to add some canonicalization pattern on the mesh dialect later.

E.g. `mesh.all_reduce` + `mesh.local_split` might be optimized to `mesh.reduce_scatter`

https://github.com/llvm/llvm-project/pull/80518