PyTorch performance testing mechanism

PyTorch supports many aten operators, and Pytroch has a huge amount of code. How does the community do performance testing?