Refactors legacy PyTorch AT_DISPATCH macros into the modern AT_DISPATCH_V2 format for ATen C++ development.
This skill automates the migration of legacy PyTorch dispatch macros to the updated V2 API found in ATen. It streamlines the process of porting ATen kernels and native operators by correctly reordering arguments, wrapping lambdas with the required AT_WRAP macro, and expanding type groups using AT_EXPAND. It is specifically designed for developers working on PyTorch core, CUDA kernels, or custom operators who need to maintain compatibility with the latest ATen dispatching standards while reducing boilerplate and macro arity issues.
主要功能
010 GitHub stars
02Automatic header injection for ATen/Dispatch_v2.h
03Support for complex multi-line lambdas with internal commas
04Automatic argument reordering for AT_DISPATCH_V2 compliance
05Conversion of implicit macro type sets to explicit AT_EXPAND type groups
06Safe lambda wrapping using AT_WRAP to prevent macro parsing errors
使用场景
01Refactoring ATen native operators to use composable type sets
02Porting legacy PyTorch C++ kernels to the modern Dispatch V2 API
03Upgrading custom CUDA kernels to follow current PyTorch coding standards
