Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Duration 14 hours
Course Outline
Introduction to Developing Custom Operators
- Rationale for building custom operators: use cases and constraints.
- CANN runtime structure and operator integration points.
- Overview of TBE, TIK, and TVM within the Huawei AI ecosystem.
Leveraging TIK for Low-Level Operator Programming
- Exploring the TIK programming model and supported APIs.
- Memory management and tiling strategies in TIK.
- Creating, compiling, and registering custom ops with CANN.
Testing and Validating Custom Operations
- Unit and integration testing of ops within the graph.
- Diagnosing kernel-level performance issues.
- Visualizing op execution and buffer behavior.
TVM-Based Scheduling and Optimization
- TVM as a compiler for tensor operations: an overview.
- Developing schedules for custom ops in TVM.
- TVM tuning, benchmarking, and code generation for Ascend.
Framework and Model Integration
- Registering custom ops for MindSpore and ONNX.
- Verifying model integrity and fallback mechanisms.
- Supporting multi-operator graphs with mixed precision.
Case Studies and Specialized Optimizations
- Case study: Efficient convolution for small input shapes.
- Case study: Optimizing memory-aware attention operators.
- Best practices for deploying custom ops across devices.
Summary and Future Directions
Requirements
- Profound understanding of AI model internals and operator-level computation.
- Proficiency with Python and Linux development environments.
- Familiarity with neural network compilers or graph-level optimizers.
Target Audience
- Compiler engineers specializing in AI toolchains.
- Systems developers focused on low-level AI optimization.
- Developers creating custom ops or addressing novel AI workloads.
