[Mlir-commits] [mlir] [MLIR] Add `InParallelOpInterface` for parallel combining operations (PR #157736)

[Mehdi Amini]

================
@@ -52,8 +53,60 @@ def ParallelCombiningOpInterface : OpInterface<"ParallelCombiningOpInterface"> {
   ];
   // TODO: Single region single block interface on interfaces ?
   let verify = [{
-    return verifyParallelCombiningOpInterface($_op);
+    return verifyInParallelOpInterface($_op);
+  }];
+}
+
+def ParallelCombiningOpInterface : OpInterface<"ParallelCombiningOpInterface"> {
+  let description = [{
+    A parallel combining op is an operation performs parallel updates to
+    destination tensors within the context of a parent iterating operation.
+    
+    This interface is designed for operations that need to coordinate parallel
+    insertions or updates to tensors that are being constructed or modified
----------------
joker-eph wrote:

Nit: tensors cannot be modified as they are immutable.
We should formulate this in terms of a loop where iteration are producing pieces of data that the enclosing operation is aggregating in a resulting tensor, or something like that.

https://github.com/llvm/llvm-project/pull/157736