Index: clang/lib/CodeGen/CGCUDANV.cpp
===================================================================
--- clang/lib/CodeGen/CGCUDANV.cpp
+++ clang/lib/CodeGen/CGCUDANV.cpp
@@ -472,6 +472,18 @@
     CtorBuilder.CreateCall(RegisterLinkedBinaryFunc, Args);
   }
 
+  // Create destructor and register it with atexit() the way NVCC does it. Doing
+  // it during regular destructor phase worked in CUDA before 9.2 but results in
+  // double-free in 9.2.
+  llvm::Function *CleanupFn = makeModuleDtorFunction();
+  // extern "C" int atexit(void (*f)(void));
+  llvm::FunctionType *AtExitTy =
+      llvm::FunctionType::get(IntTy, CleanupFn->getType(), false);
+  llvm::Constant *AtExitFunc =
+      CGM.CreateRuntimeFunction(AtExitTy, "atexit", llvm::AttributeList(),
+                                /*Local=*/true);
+  CtorBuilder.CreateCall(AtExitFunc, CleanupFn);
+
   CtorBuilder.CreateRetVoid();
   return ModuleCtorFunc;
 }
Index: clang/lib/CodeGen/CodeGenModule.cpp
===================================================================
--- clang/lib/CodeGen/CodeGenModule.cpp
+++ clang/lib/CodeGen/CodeGenModule.cpp
@@ -404,10 +404,9 @@
       AddGlobalCtor(ObjCInitFunction);
   if (Context.getLangOpts().CUDA && !Context.getLangOpts().CUDAIsDevice &&
       CUDARuntime) {
-    if (llvm::Function *CudaCtorFunction = CUDARuntime->makeModuleCtorFunction())
+    if (llvm::Function *CudaCtorFunction =
+            CUDARuntime->makeModuleCtorFunction())
       AddGlobalCtor(CudaCtorFunction);
-    if (llvm::Function *CudaDtorFunction = CUDARuntime->makeModuleDtorFunction())
-      AddGlobalDtor(CudaDtorFunction);
   }
   if (OpenMPRuntime) {
     if (llvm::Function *OpenMPRegistrationFunction =