Hey, my application ships with OpenCV with the CUDA modules enabled. There are a lot of CUDA kernels included in OpenCV, so if JIT occurs (due to a new GPU model coming out), the JIT takes a very long time (a few minutes).
I would like the improve the user experience here. Is there any way to get some feedback that JIT compilation is occurring, so I can give a dialog/popup to the user telling them to be patient and the app hasn’t crashed/hung.
It seems like when JIT will occur isn’t consistent either. If opencv’s .dll is loaded but CUDA hasn’t been initialized yet, then a JIT will occur on one of my calls to cudaStreamCreate(). If CUDA has already been initialized and then opencv’s .dll is loaded later, then the JIT occurs on a call to v::cuda::Stream().
Is there a way to for JIT to occur for a .dll as another option, so I can put that into a thread and give feedback if it takes more than a few seconds?