[llvm-branch-commits] [clang] [AMDGPU][Clang] add __builtin_amdgcn_av_(load|store)_b128 (PR #199176)

[Juan Manuel Martinez Caamaño via llvm-branch-commits]

================
@@ -0,0 +1,22 @@
+// RUN: %clang_cc1 -cl-std=CL2.0 -triple amdgcn-unknown-unknown -target-cpu gfx950         -S -verify -o - %s
----------------
jmmartinez wrote:

Can you add a test in `SemaHIP` where we test that the proper usage of these builtins validate for host and device?

Something like:

```cpp
// RUN: %clang_cc1 -fsyntax-only -triple amdgcn -target-cpu gfx90a -verify %s -fcuda-is-device
// RUN: %clang_cc1 -fsyntax-only -triple x86_64 -aux-triple amdgcn -verify %s
```

In the past we had some builtins that failed when building for the host (even though the calls are on the device side).

https://github.com/llvm/llvm-project/pull/199176