Migrating CUDA Applications to Chinese GPU Architectures Training Course

Overview of the Chinese AI GPU Ecosystem

• Comparison of Huawei Ascend, Biren, and Cambricon MLU
• CUDA vs CANN, Biren SDK, and BANGPy programming models
• Industry trends and vendor ecosystem developments

Preparing for Migration

• Evaluating your existing CUDA codebase
• Identifying target platforms and compatible SDK versions
• Installing the toolchain and setting up the development environment

Code Translation Techniques

• Translating CUDA memory access patterns and kernel logic
• Mapping compute grid and thread models
• Exploring automated versus manual translation approaches

Platform-Specific Implementations

• Leveraging Huawei CANN operators and custom kernels
• Utilising the Biren SDK conversion pipeline
• Rebuilding models using BANGPy (Cambricon)

Cross-Platform Testing and Optimization

• Profiling execution on each target platform
• Tuning memory usage and comparing parallel execution performance
• Monitoring performance metrics and iterative refinement

Managing Mixed GPU Environments

• Implementing hybrid deployments across multiple architectures
• Developing fallback strategies and device detection mechanisms
• Establishing abstraction layers to ensure code maintainability

Case Studies and Best Practices

• Porting vision and NLP models to Ascend or Cambricon
• Retrofitting inference pipelines on Biren clusters
• Mitigating version mismatches and API discrepancies

Summary and Next Steps