Overview

This section describes how to use MindStudio to develop custom operators under different frameworks, including MindSpore-, PyTorch-, TensorFlow-, Caffe-, and ONNX-based TBE operator development and TensorFlow-, PyTorch-, MindSpore-, and ONNX-based AI CPU operator development. Operators of the MindSpore framework can only be used in Atlas training productss.The following mainly focuses on MindStudio operations but also touches on important code implementations. For details about the programming, see the corresponding guide listed in the following table.

MindStudio provides operator development samples. After the Ascend-CANN-Toolkit configuration is complete, you can obtain the samples from /MindStudio/plugins/ascend-foundation/samples/operator_demo_projects/quick_start under the MindStudio installation directory. For details, see Custom Operator Samples.

The Ascend-CANN-Toolkit installation directory/ascend-toolkit/latest/opp/built-in/op_impl/ai_core/tbe/impl directory stores the implementation files of the built-in AI Core operators supported by the Ascend AI Processor. You can refer to these files when implementing the operator code.