[clang] [llvm] [AMDGPU] Add OpenCL-specific fence address space masks (PR #78572)

[Sameer Sahasrabuddhe via cfe-commits]

================
@@ -18365,6 +18366,30 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
   return nullptr;
 }
 
+void CodeGenFunction::AddAMDGCNAddressSpaceMMRA(llvm::Instruction *Inst,
+                                                const CallExpr *E,
+                                                unsigned FirstASNameIdx) {
+  constexpr const char *Tag = "opencl-fence-mem";
+
+  LLVMContext &Ctx = Inst->getContext();
+  SmallVector<MMRAMetadata::TagT, 3> MMRAs;
+  for (unsigned K = FirstASNameIdx; K < E->getNumArgs(); ++K) {
+    llvm::Value *V = EmitScalarExpr(E->getArg(K));
+    StringRef AS;
+    if (llvm::getConstantStringInfo(V, AS) &&
+        (AS == "local" || AS == "global" || AS == "image")) {
----------------
ssahasra wrote:

Can these magic strings be declared somewhere? Ideally some sort of target-specific mechanism that Clang will use to find such sets of valid strings?

https://github.com/llvm/llvm-project/pull/78572