Use CUDA graph with CUmodule

Accelerated Computing CUDA CUDA Programming and Performance
1

I would like to add a kernel node to CUDA graph using API cudaGraphAddKernelNode. The cudaKernelNodeParams (CUDA Runtime API :: CUDA Toolkit Documentation) has a func field, we need to set params.func to a function pointer to a kernel.
However, my kernel is JIT-generated and I have a CUmodule. I can use cuModuleGetFunction to get a CUfunction, but how should I set params.func?