Question

I was wondering if the two following scenarios have the same performance impact on native C++ code (if there is any performance impact at all).

Let’s assume I have the function cpp_calc() that is doing some calculation stuff and is written in native C++. Also, there is cs_show_gui_stuff(), which is written in C#.

Now, which of the following scenarios will worsen the native c++ performance (if there is any performance penalty at all)?

1. Creating a .Net (C#) application that runs cs_show_gui_stuff() and calls cpp_calc() in the native C++ dll using DllImport or turning C++ into a COM DLL.

2. Creating a C++ application that implements cpp_calc() in C++ and runs cs_show_guid_stuff() by placing the C# code in a .Net COM DLL.

Thanks 🙂

Answer 1

It really depends on what the other parts of the system are mainly written in. From a performance-only perspective, one PInvoke (via DllImport attribute) call will probably be faster than one COM call if the method arguments do not need any special marshaling.

A third, and probably the best alternative, is to create a managed C++/CLI library that calls the unmanaged C++ method with nearly no performance impact and add a reference to the C++/CLI library in the C# application. The C# application can then make managed method calls to the C++/CLI application, which in turn can make unmanaged method calls. While this adds one level of indirection, it will provide way better performance than the methods you mentioned.