diff --git a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
index 62cd9d48d399..3e66ccee4ebe 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
@@ -535,9 +535,16 @@ static CIRGenCallee emitDirectCallee(CIRGenModule &CGM, GlobalDecl GD) {
 
   mlir::Operation *CalleePtr = emitFunctionDeclPointer(CGM, GD);
 
-  if (CGM.getLangOpts().CUDA && !CGM.getLangOpts().CUDAIsDevice &&
-      FD->hasAttr<CUDAGlobalAttr>())
-    CalleePtr = CGM.getCUDARuntime().getKernelStub(CalleePtr);
+  if ((CGM.getLangOpts().HIP || CGM.getLangOpts().CUDA) &&
+      !CGM.getLangOpts().CUDAIsDevice && FD->hasAttr<CUDAGlobalAttr>()) {
+
+    // Ensure the handle is created and use it as the lookup key.
+    auto *Handle = CGM.getCUDARuntime().getKernelHandle(
+        llvm::cast<cir::FuncOp>(CalleePtr), GD);
+
+    // Now look up the stub via the handle
+    CalleePtr = CGM.getCUDARuntime().getKernelStub(Handle);
+  }
 
   return CIRGenCallee::forDirect(CalleePtr, GD);
 }
@@ -1560,8 +1567,6 @@ RValue CIRGenFunction::emitCall(clang::QualType CalleeType,
     Callee.setFunctionPointer(Fn);
   }
 
-  assert(!CGM.getLangOpts().HIP && "HIP NYI");
-
   assert(!MustTailCall && "Must tail NYI");
   cir::CIRCallOpInterface callOP;
   RValue Call = emitCall(FnInfo, Callee, ReturnValue, Args, &callOP,
diff --git a/clang/lib/CIR/CodeGen/CIRGenModule.cpp b/clang/lib/CIR/CodeGen/CIRGenModule.cpp
index d80f8b70964e..362ec7828265 100644
--- a/clang/lib/CIR/CodeGen/CIRGenModule.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenModule.cpp
@@ -2548,20 +2548,9 @@ cir::FuncOp CIRGenModule::GetAddrOfFunction(clang::GlobalDecl gd, mlir::Type ty,
   auto f = GetOrCreateCIRFunction(mangledName, ty, gd, forVTable, dontDefer,
                                   /*IsThunk=*/false, isForDefinition);
 
-  // As __global__ functions (kernels) always reside on device,
-  // when we access them from host, we must refer to the kernel handle.
-  // For HIP, we should never directly access the host device addr, but
-  // instead the Global Variable of that stub. For CUDA, it's just the device
-  // stub. For HIP, it's something different.
   if ((langOpts.HIP || langOpts.CUDA) && !langOpts.CUDAIsDevice &&
-      cast<FunctionDecl>(gd.getDecl())->hasAttr<CUDAGlobalAttr>()) {
+      cast<FunctionDecl>(gd.getDecl())->hasAttr<CUDAGlobalAttr>())
     (void)getCUDARuntime().getKernelHandle(f, gd);
-    if (isForDefinition)
-      return f;
-
-    if (langOpts.HIP)
-      llvm_unreachable("NYI");
-  }
 
   return f;
 }
diff --git a/clang/test/CIR/CodeGen/HIP/simple.cpp b/clang/test/CIR/CodeGen/HIP/simple.cpp
index 9c00b149e1c7..27e6e5a4cd67 100644
--- a/clang/test/CIR/CodeGen/HIP/simple.cpp
+++ b/clang/test/CIR/CodeGen/HIP/simple.cpp
@@ -32,3 +32,21 @@ __global__ void global_fn(int a) {}
 // The stub has the mangled name of the function
 // CIR-HOST: cir.get_global @_Z9global_fni
 // CIR-HOST: cir.call @hipLaunchKernel
+
+int main() {
+  global_fn<<<1, 1>>>(1);
+}
+// CIR-DEVICE-NOT: cir.func dso_local @main()
+
+// CIR-HOST: cir.func dso_local @main()
+// CIR-HOST: cir.call @_ZN4dim3C1Ejjj
+// CIR-HOST: cir.call @_ZN4dim3C1Ejjj
+// CIR-HOST: [[Push:%[0-9]+]] = cir.call @__hipPushCallConfiguration
+// CIR-HOST: [[ConfigOK:%[0-9]+]] = cir.cast int_to_bool [[Push]]
+// CIR-HOST: cir.if [[ConfigOK]] {
+// CIR-HOST: } else {
+// CIR-HOST:   [[Arg:%[0-9]+]] = cir.const #cir.int<1>
+// CIR-HOST:   cir.call @_Z24__device_stub__global_fni([[Arg]])
+// CIR-HOST: }
+
+