Migrates legacy PyTorch AT_DISPATCH macros to the modern Dispatch_v2 format in ATen C++ codebases.
This skill automates the complex refactoring process of converting PyTorch's legacy AT_DISPATCH macros to the more flexible and extensible V2 API. It handles the critical steps of the migration including header inclusion, argument reordering, lambda wrapping with AT_WRAP to prevent parsing errors, and the transition from implicit arity-based macros (such as AND3 or AND4) to explicit type group expansion using AT_EXPAND. It is specifically designed for developers maintaining ATen kernel implementations, CUDA kernels, or native operator code who need to modernize their C++ codebase according to current PyTorch standards.
Key Features
01Automatic header injection for Dispatch_v2.h while preserving existing includes
02Support for multi-group expansions including complex and integral type combinations
03Intelligent mapping of legacy macros (ALL_TYPES, FLOATING_TYPES, etc.) to V2 type groups
0481 GitHub stars
05Automated argument reordering for the AT_DISPATCH_V2 syntax
06Safe lambda wrapping with AT_WRAP to handle internal commas and complex expressions
Use Cases
01Refactoring C++ source files to reduce macro arity dependency and improve code clarity
02Porting legacy ATen kernels to the newer, more extensible Dispatch_v2 API
03Updating CUDA kernel implementations in PyTorch native operator directories
