search

Jamy-L / Handheld-Multi-Frame-Super-Resolution

Handheld Multi-image Super-resolution [Wronski et al., SIGGRAPH19]. Non-official GPU-supported Python implementation.
MIT License
127 stars 20 forks source link

How to debug @cuda.jit functions? #19

Closed liuhuaer closed 1 year ago

liuhuaer commented 1 year ago

I use vscode remote ubuntu server to debug, but can't debug @cuda.jit function.

Jamy-L commented 1 year ago

@cuda.jit functions are executed on GPU and can therefore be tricky to debug, especially because CUDA errors are not very explicit. Numba has a built-in debugging feature but I have never really tried it. Bear in mind that during runtime, jitted functions are parsed by an interpreter, then converted to CUDA code which is then executed, which is why regular python debuggers may struggle.