[clang] [CIR] Upstream CIR Dialect TryOp with Catch Attrs (PR #162897)

[Andy Kaylor via cfe-commits]

================
@@ -4296,6 +4296,81 @@ def CIR_AllocExceptionOp : CIR_Op<"alloc.exception"> {
   }];
 }
 
+//===----------------------------------------------------------------------===//
+// TryOp
+//===----------------------------------------------------------------------===//
+
+def CIR_TryOp : CIR_Op<"try",[
+  DeclareOpInterfaceMethods<RegionBranchOpInterface>,
+  RecursivelySpeculatable, AutomaticAllocationScope, NoRegionArguments
+]> {
+  let summary = "C++ try block";
+  let description = [{
+    Holds the lexical scope of `try {}`. Note that resources used on catch
+    clauses are usually allocated in the same parent as `cir.try`.
+
+    `synthetic`: use `cir.try` to represent try/catches not originally
+    present in the source code (e.g. `g = new Class` under `-fexceptions`).
+
+    `cleanup`: signal to targets (LLVM for now) that this try/catch, needs
+    to specially tag their landing pads as needing "cleanup".
+
+    Example:
+
+    ```mlir
+    %0 = cir.alloc.exception 16 -> !cir.ptr<!some_record>
+    %1 = cir.get_global @d2 : !cir.ptr<!some_record>
+    cir.try synthetic cleanup {
+      cir.call exception @_ZN7test2_DC1ERKS_(%0, %1)
+            : (!cir.ptr<!some_record>, !cir.ptr<!some_record>) -> () cleanup {
+        %2 = cir.cast bitcast %0 : !cir.ptr<!some_record> -> !cir.ptr<!void>
+        cir.free.exception %2
+        cir.yield
+      }
----------------
andykaylor wrote:

Shouldn't there be an unwind region here?

https://github.com/llvm/llvm-project/pull/162897