[clang] [AMDGPU][clang] provide device implementation for __builtin_logb and … (PR #129347)

[Joseph Huber via cfe-commits]

================
@@ -259,6 +259,26 @@ Value *readX18AsPtr(CodeGenFunction &CGF) {
   return CGF.Builder.CreateIntToPtr(X18, CGF.Int8PtrTy);
 }
 
+llvm::Constant *CodeGenModule::getDeviceLibFunction(const FunctionDecl *FD,
+                                                    unsigned BuiltinID) {
+  GlobalDecl D(FD);
+  llvm::SmallString<64> Name;
+  if (getTarget().getTriple().isAMDGCN()) {
+    switch (BuiltinID) {
+    default: return nullptr;
+    case Builtin::BIlogb:
+    case Builtin::BI__builtin_logb:
+      Name = "__ocml_logb_f64";
----------------
jhuber6 wrote:

I don't think the solution is to start hard-coding DeivceRTL names in the compiler. Coincidentally I actually created an RFC about these kinds of problems just a few minutes ago https://discourse.llvm.org/t/rfc-lto-handling-math-libcalls-with-lto/84884.

https://github.com/llvm/llvm-project/pull/129347