This is primarily a deployment problem, just have your installer copy the right DLL based on the Windows version on the target machine.
But nobody ever likes to do that. Dynamically pinvoking the correct DLL's function is enormously painfully, you have to write delegate types for every exported function and use LoadLibrary + GetProcAddress + Marshal.GetDelegateForFunctionPointer to create the delegate object.
But nobody ever likes to do that. The less painful tack is to declare the function twice, giving it different names and using the EntryPoint property in the [DllImport] attribute to specify the real name. Then test at runtime which you want to call.
But nobody ever likes to do that. The most effective trick is steer Windows into loading the correct DLL for you. First thing you have to do is copy the DLL into a directory where Windows will not look for it. Best way is to create an "x86" and an "x64" subdirectory in your build directory and copy the appropriate DLL into each. Do so by writing a post-build event that creates the directories and copies the DLLs.
Then tell Windows about it by pinvoking SetDllDirectory(). The path you specify will be added to the directories that Windows searches for a DLL. Like this:
using System;
using System.Runtime.InteropServices;
using System.Reflection;
using System.IO;
class Program {
static void Main(string[] args) {
var path = Path.GetDirectoryName(Assembly.GetEntryAssembly().Location);
path = Path.Combine(path, IntPtr.Size == 8 ? "x64" : "x86");
bool ok = SetDllDirectory(path);
if (!ok) throw new System.ComponentModel.Win32Exception();
//etc..
}
[DllImport("kernel32.dll", CharSet = CharSet.Auto, SetLastError = true)]
private static extern bool SetDllDirectory(string path);
}
Do consider if having the code run in 64-bit mode is actually useful to you. It is pretty rare to need the giant virtual memory address space you get from it, the only real benefit. You still need to support the 32-bit version that needs to operate correctly in the 2 gigabyte case.
