[Mlir-commits] [mlir] Lower allreduce (PR #144716)

Frank Schlimbach llvmlistbot at llvm.org
Mon Jun 23 05:00:17 PDT 2025


================
@@ -584,11 +584,11 @@ def Mesh_AllReduceOp : Mesh_CollectiveCommunicationOpBase<"all_reduce", [
     ```
   }];
   let arguments = !con(commonArgs, (ins
-    AnyRankedTensor:$input,
+    AnyTypeOf<[AnyMemRef, AnyRankedTensor]>:$input,
----------------
fschlimb wrote:

MPI dialect is strictly buffer semantics. There is no room for tensors there.

https://github.com/llvm/llvm-project/pull/144716


More information about the Mlir-commits mailing list