Can you compile a Windows dll with nvcc CUDA?

Accelerated Computing CUDA CUDA Programming and Performance
#1

(warning, newbie here! :"> )

I wonder whether you could compile CUDA applications into windows dlls. The reason being, I would like to share some given function (say a bitonic sort) among various callers or among various modules in a system without having to recompile it with the rest of the code everytime.

If it is possible what is the way to do it? (I noticed a “lib” option in nvcc command line, but that is not really a dll, is it?)

#2

Yes you can. But you’ll need to redistribute cudart.dll if you’re using Runtime API.

#3

Thanks.

Is there any example or template available anywhere?

My specific issue is how to compile the dll project using nvcc. What compile/link options should one use?

#4

Nevermind, I got it.

For the record, install the CUDA SDK and study project “template”.

Change the project configuration to generate a .lib output.

Not quite a dll but is enough for what I desired.