Course Code: canntiktvm
Duration: 14 hours
Prerequisites:
  • Strong knowledge of AI model internals and operator-level computation
  • Experience with Python and Linux development environments
  • Familiarity with neural network compilers or graph-level optimizers

Audience

  • Compiler engineers working on AI toolchains
  • Systems developers focused on low-level AI optimization
  • Developers building custom ops or targeting novel AI workloads
Overview:

CANN TIK (Tensor Instruction Kernel) and Apache TVM enable advanced optimization and customization of AI model operators for Huawei Ascend hardware.

This instructor-led, live training (online or onsite) is aimed at advanced-level system developers who wish to build, deploy, and tune custom operators for AI models using CANN’s TIK programming model and TVM compiler integration.

By the end of this training, participants will be able to:

  • Write and test custom AI operators using the TIK DSL for Ascend processors.
  • Integrate custom ops into the CANN runtime and execution graph.
  • Use TVM for operator scheduling, auto-tuning, and benchmarking.
  • Debug and optimize instruction-level performance for custom computation patterns.

Format of the Course

  • Interactive lecture and demonstration.
  • Hands-on coding of operators using TIK and TVM pipelines.
  • Testing and tuning on Ascend hardware or simulators.

Course Customization Options

  • To request a customized training for this course, please contact us to arrange.
Course Outline:

Introduction to Custom Operator Development

  • Why build custom operators? Use cases and constraints
  • CANN runtime structure and operator integration points
  • Overview of TBE, TIK, and TVM in the Huawei AI ecosystem

Using TIK for Low-Level Operator Programming

  • Understanding the TIK programming model and supported APIs
  • Memory management and tiling strategy in TIK
  • Creating, compiling, and registering a custom op with CANN

Testing and Validating Custom Ops

  • Unit testing and integration testing of ops in the graph
  • Debugging kernel-level performance issues
  • Visualizing op execution and buffer behavior

TVM-Based Scheduling and Optimization

  • Overview of TVM as a compiler for tensor ops
  • Writing a schedule for a custom op in TVM
  • TVM tuning, benchmarking, and code generation for Ascend

Integration with Frameworks and Models

  • Registering custom ops for MindSpore and ONNX
  • Verifying model integrity and fallback behavior
  • Supporting multi-operator graphs with mixed precision

Case Studies and Specialized Optimizations

  • Case study: high-efficiency convolution for small input shapes
  • Case study: memory-aware attention operator optimization
  • Best practices in custom op deployment across devices

Summary and Next Steps

Sites Published:

United Arab Emirates - Building Custom AI Operators with CANN TIK and TVM

Qatar - Building Custom AI Operators with CANN TIK and TVM

Egypt - Building Custom AI Operators with CANN TIK and TVM

Saudi Arabia - Building Custom AI Operators with CANN TIK and TVM

South Africa - Building Custom AI Operators with CANN TIK and TVM

Brasil - Building Custom AI Operators with CANN TIK and TVM

Canada - Building Custom AI Operators with CANN TIK and TVM

中国 - Building Custom AI Operators with CANN TIK and TVM

香港 - Building Custom AI Operators with CANN TIK and TVM

澳門 - Building Custom AI Operators with CANN TIK and TVM

台灣 - Building Custom AI Operators with CANN TIK and TVM

USA - Building Custom AI Operators with CANN TIK and TVM

Österreich - Building Custom AI Operators with CANN TIK and TVM

Schweiz - Building Custom AI Operators with CANN TIK and TVM

Deutschland - Building Custom AI Operators with CANN TIK and TVM

Czech Republic - Building Custom AI Operators with CANN TIK and TVM

Denmark - Building Custom AI Operators with CANN TIK and TVM

Estonia - Building Custom AI Operators with CANN TIK and TVM

Finland - Building Custom AI Operators with CANN TIK and TVM

Greece - Building Custom AI Operators with CANN TIK and TVM

Magyarország - Building Custom AI Operators with CANN TIK and TVM

Ireland - Building Custom AI Operators with CANN TIK and TVM

Luxembourg - Building Custom AI Operators with CANN TIK and TVM

Latvia - Building Custom AI Operators with CANN TIK and TVM

España - Building Custom AI Operators with CANN TIK and TVM

Italia - Building Custom AI Operators with CANN TIK and TVM

Lithuania - Building Custom AI Operators with CANN TIK and TVM

Nederland - Building Custom AI Operators with CANN TIK and TVM

Norway - Building Custom AI Operators with CANN TIK and TVM

Portugal - Building Custom AI Operators with CANN TIK and TVM

România - Building Custom AI Operators with CANN TIK and TVM

Sverige - Building Custom AI Operators with CANN TIK and TVM

Türkiye - Building Custom AI Operators with CANN TIK and TVM

Malta - Building Custom AI Operators with CANN TIK and TVM

Belgique - Building Custom AI Operators with CANN TIK and TVM

France - Building Custom AI Operators with CANN TIK and TVM

日本 - Building Custom AI Operators with CANN TIK and TVM

Australia - Building Custom AI Operators with CANN TIK and TVM

Malaysia - Building Custom AI Operators with CANN TIK and TVM

New Zealand - Building Custom AI Operators with CANN TIK and TVM

Philippines - Building Custom AI Operators with CANN TIK and TVM

Singapore - Building Custom AI Operators with CANN TIK and TVM

Thailand - Building Custom AI Operators with CANN TIK and TVM

Vietnam - Building Custom AI Operators with CANN TIK and TVM

India - Building Custom AI Operators with CANN TIK and TVM

Argentina - Building Custom AI Operators with CANN TIK and TVM

Chile - Building Custom AI Operators with CANN TIK and TVM

Costa Rica - Building Custom AI Operators with CANN TIK and TVM

Ecuador - Building Custom AI Operators with CANN TIK and TVM

Guatemala - Building Custom AI Operators with CANN TIK and TVM

Colombia - Building Custom AI Operators with CANN TIK and TVM

México - Building Custom AI Operators with CANN TIK and TVM

Panama - Building Custom AI Operators with CANN TIK and TVM

Peru - Building Custom AI Operators with CANN TIK and TVM

Uruguay - Building Custom AI Operators with CANN TIK and TVM

Venezuela - Building Custom AI Operators with CANN TIK and TVM

Polska - Building Custom AI Operators with CANN TIK and TVM

United Kingdom - Building Custom AI Operators with CANN TIK and TVM

South Korea - Building Custom AI Operators with CANN TIK and TVM

Pakistan - Building Custom AI Operators with CANN TIK and TVM

Sri Lanka - Building Custom AI Operators with CANN TIK and TVM

Bulgaria - Building Custom AI Operators with CANN TIK and TVM

Bolivia - Building Custom AI Operators with CANN TIK and TVM

Indonesia - Building Custom AI Operators with CANN TIK and TVM

Kazakhstan - Building Custom AI Operators with CANN TIK and TVM

Moldova - Building Custom AI Operators with CANN TIK and TVM

Morocco - Building Custom AI Operators with CANN TIK and TVM

Tunisia - Building Custom AI Operators with CANN TIK and TVM

Kuwait - Building Custom AI Operators with CANN TIK and TVM

Oman - Building Custom AI Operators with CANN TIK and TVM

Slovakia - Building Custom AI Operators with CANN TIK and TVM

Kenya - Building Custom AI Operators with CANN TIK and TVM

Nigeria - Building Custom AI Operators with CANN TIK and TVM

Botswana - Building Custom AI Operators with CANN TIK and TVM

Slovenia - Building Custom AI Operators with CANN TIK and TVM

Croatia - Building Custom AI Operators with CANN TIK and TVM

Serbia - Building Custom AI Operators with CANN TIK and TVM

Bhutan - Building Custom AI Operators with CANN TIK and TVM

Nepal - Building Custom AI Operators with CANN TIK and TVM

Uzbekistan - Building Custom AI Operators with CANN TIK and TVM