HAL ML API Overview

Introduction

The HAL ML API provides an interface for Tizen's higher-level Machine Learning API to interact with various hardware acceleration devices (e.g., NPUs, GPUs, DSPs) via HAL ML Accelerator Backends. Users would not use this directly, it is used in implementation of NNStreamer's tensor_filter sub-plugin which is backend of Tizen ML API.

Architecture

graph TD
    subgraph TF["Tizen Framework"]
        A["Tizen ML API"]
    end

    subgraph HM["HAL API ML"]
        D["Tizen HAL API for ML"]
    end

    subgraph HBMA["HAL Backend ML Accelerator"]
        E["backend for each vendor"]
    end

    subgraph MA["Device"]
        CPU
        GPU
        NPU
        DSP
    end

    TF <--> HM <--> HBMA <--> MA

    style TF fill:#68b,color:#000,stroke:#000
    style HM fill:#68b,color:#000,stroke:#000
    style HBMA fill:#68b,color:#000,stroke:#000
    style MA fill:#68b,color:#000,stroke:#000

    style A fill:#fff,color:#000,stroke:#000
    style D fill:#fff,color:#000,stroke:#000
    style E fill:#fff,color:#000,stroke:#000
    style CPU fill:#fff,color:#000,stroke:#000
    style GPU fill:#fff,color:#000,stroke:#000
    style NPU fill:#fff,color:#000,stroke:#000
    style DSP fill:#fff,color:#000,stroke:#000

Loading

License

This project is licensed under the Apache License, Version 2.0. See the LICENSE file for details.