add-uint-support

What the skill teaches

A contributor skill in the PyTorch repository for letting existing operator kernels accept 16, 32 and 64-bit unsigned integer tensors. The agent checks that the file already uses the newer AT_DISPATCH_V2 macros, converting it first with the companion at-dispatch-v2 skill if not, then either adds the unsigned type group to each dispatch call or swaps the integral group for the version that includes unsigned types. It updates every dispatch site in the file, CPU and CUDA alike, leaves floating-point-only operators alone, skips files that already have unsigned support, and shows the user the diff. Testing is left to the user, and it notes that some operators make no sense for unsigned types.

Capabilities

Details