Morphology benchmarks to torch (CPU), for example kornia

kornia has [torch-native morphology ops](https://kornia.readthedocs.io/en/latest/morphology.html). Benchmarks to that would be great. I guess accounting numpy to torch and back to numpy conversions to have apples to apples. All on CPU obviously.