[clang] [llvm] [AMDGPU] Enable OpenCL hostcall printf (WIP) (PR #72556)

[Sameer Sahasrabuddhe via cfe-commits]

================
@@ -3616,6 +3617,12 @@ unsigned FunctionDecl::getBuiltinID(bool ConsiderWrapperFunctions) const {
   if (!ConsiderWrapperFunctions && getStorageClass() == SC_Static)
     return 0;
 
+  // AMDGCN implementation supports printf as a builtin
+  // for OpenCL
+  if (Context.getTargetInfo().getTriple().isAMDGCN() &&
+      Context.getLangOpts().OpenCL && BuiltinID == AMDGPU::BIprintf)
+    return BuiltinID;
----------------
ssahasra wrote:

I thought this had been clarified earlier too. It's quite imprecise to just say that "signatures differ". Perhaps the following detailed explanation might move the conversatino forward. The problem is that the OpenCL printf expects a format string in the constant address space, which has no representation in Clang builtin. What we do have is the ability to specify an address-space number in the builtin declaration. But this number is target-specific, which makes the whole builtin target-specific. Is there a way around that magic number 4?


https://github.com/llvm/llvm-project/pull/72556