[clang] [CIR][HIP] Skip global init for device variables (PR #231971)

[Steffen Larsen via cfe-commits]

================
@@ -0,0 +1,33 @@
+// REQUIRES: amdgpu-registered-target
----------------
steffenlarsen wrote:

I don't think we need this requirement. Also, could we move this to CodeGenCUDA and test it there as well?

https://github.com/llvm/llvm-project/pull/231971