Reading Paths

Use this page when you do not want to read the repository in directory order.

Path 1: First contact

Best for readers who want a fast orientation before opening code.

1. Getting Started
2. 01-SGEMM Tutorial
3. System Architecture
4. Roadmap

Path 2: Kernel optimization

Best for readers chasing performance mechanics first.

1. 01-SGEMM Tutorial
2. SGEMM Optimization
3. 03-HPC Advanced
4. Benchmarks

Path 3: Reusable CUDA library design

Best for readers more interested in interfaces, library shape, and reuse.

1. 02-TensorCraft Core
2. TensorCraft Design
3. Architecture
4. API Reference

Path 4: Inference system integration

Best for readers who care about how kernels become part of a runnable system.

1. 04-Inference Engine
2. Inference Engine Design
3. Quick Start
4. Performance Tuning

Verification reality

• Docs and Pages builds are CPU-safe.
• CUDA build and runtime validation still belong on a real GPU machine.
• Use Workflow for the repo change loop and verification commands.