Skip to main content
Ctrl+K
onnx-light-cpu  documentation - Home onnx-light-cpu  documentation - Home
  • Getting Started
  • Design
  • ByOp
  • API reference
  • Examples
  • Next Steps
  • GitHub
  • Getting Started
  • Design
  • ByOp
  • API reference
  • Examples
  • Next Steps
  • GitHub

Section Navigation

  • Registering kernels
  • Parallel kernel execution
  • Kernel and implementation selection
  • Benchmark Methodology
  • Kernels
    • Unary Kernel Design
    • Binary Kernel Design
    • Gemm Kernel Design
    • MatMul Kernel Design
    • Attention Kernel Design
  • Design

Design#

onnx-light-cpu is an extension of onnx-light: it installs fast CPU kernels into onnx-light’s shared C++ dispatch table. It accelerates standard ONNX operators in the ai.onnx domain and operators in the com.microsoft domain. The pages below explain registration, runtime parallel execution, implementation selection, benchmarks, and the main kernel families.

  • Registering kernels
  • Parallel kernel execution
  • Kernel and implementation selection
  • Benchmark Methodology
  • Kernels

previous

Getting Started

next

Registering kernels

Show Source

Last updated on Sep 07, 2026.