[clang] [CIR] Implement device-wide nounwind attribute (PR #227016)

[Steffen Larsen via cfe-commits]

================
@@ -0,0 +1,68 @@
+#include "Inputs/cuda.h"
+
+// REQUIRES: nvptx-registered-target
+// RUN: %clang_cc1 -triple nvptx64-nvidia-cuda -x cuda -fcuda-is-device \
----------------
steffenlarsen wrote:

That's a fair point. I don't think they do, as is probably the case for most of the CIR tests that have them. I don't know the history, whether it might just be "just in case" or if it's to have the structural separation of disabling tests for targets you have not enabled. That said, there also seems to be examples of combined testing, so I'll follow that precedence instead! The tests have been unified.

https://github.com/llvm/llvm-project/pull/227016