I dont want people to run the profiler on my program and see the kernel names because the decomposition of the problem can give too much insight in my solution.
Is there a way to hide the cuda symbol names from showing in the profiler?
I know that the profiler itself has a switch to hide the names, but can I prevent it at the compilation/linking stage itself? Is there a nvcc switch that can help me out?
So the answer seems to be - Cant do it. The profiler is going to spill your beans.
If you use Cuda to build your app, the whole world can see your app’s workflow and use it to their advantage. And to do all this you dont need any 3rd party reverse engineering tool, rather one of the ‘official tools’ from nVidia itself!