[Mlir-commits] [mlir] [mlir][mesh, mpi] More on MeshToMPI (PR #129048)

[Christian Ulmann]

================
@@ -390,8 +698,25 @@ struct ConvertUpdateHaloOp
         offsets[dim] = orgOffset;
       };
 
-      genSendRecv(false);
-      genSendRecv(true);
+      auto get_i32val = [&](OpFoldResult &v) {
+        return isa<Value>(v)
+                   ? cast<Value>(v)
+                   : rewriter.create<arith::ConstantOp>(
+                         loc,
+                         rewriter.getI32IntegerAttr(
+                             cast<IntegerAttr>(cast<Attribute>(v)).getInt()));
+      };
----------------
Dinistro wrote:

Nit: Why not wrap the entire loop body in this lambda and just call it with 0 and 1?

https://github.com/llvm/llvm-project/pull/129048