[clang] [CIR][CUDA] Support built-in CUDA surface type (PR #196079)

[Andy Kaylor via cfe-commits]

================
@@ -0,0 +1,21 @@
+// REQUIRES: x86-registered-target
+// REQUIRES: nvptx-registered-target
+// RUN: %clang_cc1 -fclangir -std=c++11 -fcuda-is-device -triple nvptx64-nvidia-cuda -emit-cir -o - %s | FileCheck --check-prefix=DEVICE-CIR %s
+// RUN: %clang_cc1 -fclangir -std=c++11 -fcuda-is-device -triple nvptx64-nvidia-cuda -emit-llvm -o - %s | FileCheck --check-prefix=DEVICE-LLVM %s
----------------
andykaylor wrote:

```suggestion
// RUN: %clang_cc1 -std=c++11 -fcuda-is-device -triple nvptx64-nvidia-cuda -emit-llvm -o - %s | FileCheck --check-prefix=DEVICE-LLVM %s
```
We like to have tests without CIR to compare the output. If the LLVM IR is the same with and without CIR, you can use the same check prefix as I've suggested here. Otherwise, use DEVICE-OGCG and add comments explaining any differences.

https://github.com/llvm/llvm-project/pull/196079