metal-kernel - PyTorch Metal Kernels for Apple Silicon
Implement Metal/MPS kernels for PyTorch operators using native c10/metal infrastructure
Tags
Updated: 2026-06-30Capabilities
Typical Inputs
Typical Outputs
What this skill does
- write Metal kernel
- update native_functions.yaml
- implement host-side stub
- register dispatch
- compile Metal code
- remove MPSGraph implementation
- enable operator testing
Inputs
- PyTorch operator
- CUDA kernel
- native_functions.yaml
- Tensor data
Outputs
- Metal kernel file
- host-side stub
- updated native_functions.yaml
- compiled binary
Requirements
- Apple Silicon hardware
- macOS
- Xcode
- Clang
- PyTorch source tree