(Driver API) JIT compiling direct from source code

I’m using the Driver API (Win XP, VC++, new to CUDA).

Rather than loading cubin/PTX files using cuModuleLoad/LoadData, I’d like to be able to compile and load straight from source code held in a character array. What are my options ? Do I have to save the code as a text file, and then use something like ShellExecute with nvcc.exe ? I’m really hoping for something like cuCompile(…).


Currently, there’s not any way to do this via a function call that we provide.

shame. It would be nice to be able to generate say 100 variants of a kernel, like you can with shaders using shader-based GPGPU. Something for CUDA 2.4 ? (cuCompile)