[clang] [AMDGPU] Use generic builtins for `wave_reduce` ops (PR #179589)

[Shilei Tian via cfe-commits]

================
@@ -2126,9 +2126,9 @@ bool Sema::CheckTSBuiltinFunctionCall(const TargetInfo &TI, unsigned BuiltinID,
   case llvm::Triple::spirv:
   case llvm::Triple::spirv32:
   case llvm::Triple::spirv64:
-    if (TI.getTriple().getOS() != llvm::Triple::OSType::AMDHSA)
+    if (TI.getTriple().getVendor() != llvm::Triple::VendorType::AMD)
----------------
shiltian wrote:

Does this need canonical triple?

https://github.com/llvm/llvm-project/pull/179589