[clang] [CIR][AMDGPU] Route printf in device code through the OpenCL runtime (PR #226435)

[Steffen Larsen via cfe-commits]

================
@@ -65,8 +65,12 @@ lowerFromCIRToLLVMIR(mlir::ModuleOp MLIRModule, llvm::LLVMContext &LLVMCtx,
                      bool EnableOpenMP,
                      llvm::StringRef mlirSaveTempsOutFile = {},
                      llvm::vfs::FileSystem *fs = nullptr) {
-  return direct::lowerDirectlyFromCIRToLLVMIR(MLIRModule, LLVMCtx, EnableOpenMP,
-                                              mlirSaveTempsOutFile, fs);
+  std::unique_ptr<llvm::Module> LLVMModule =
+      direct::lowerDirectlyFromCIRToLLVMIR(MLIRModule, LLVMCtx, EnableOpenMP,
+                                           mlirSaveTempsOutFile, fs);
+  if (LLVMModule)
+    direct::expandAMDGPUDevicePrintf(*LLVMModule);
----------------
steffenlarsen wrote:

See https://github.com/llvm/llvm-project/pull/226435#discussion_r4120789158 - I was of two minds. I'm generally not much for the post-translation cleanup solution, but without an op I don't see much alternative. It has been moved to it though, so let's discuss the option of an Op in the other thread.

https://github.com/llvm/llvm-project/pull/226435