Debug conv transpose kernel

I’m trying to debug conv transpose kernel implementation in pytorch. I have built pytorch from source and in develop mode. With a sample application I tried to put breakpoints using gdb but am not able to locate the kernel and the file in which it exists, any help?

The native implementations can be found here for the CPU and here for the GPU. You won’t be able to debug lib implementations, if they are closed source (e.g. such as cuDNN).

1 Like