ONE(On-device Neural Engine)
  • Overview
  • How To
  • Runtime
    • API
    • Core
    • Controlflow Operations
    • Executors
    • Heterogeneous Execution
    • Backend API
    • Compute
    • Supported Operations and backend
  • Compiler
  • Common IR
  • Package
  • Platform
  • Devices
  • Test & Benchmarks
  • Release
ONE(On-device Neural Engine)
  • Runtime
  • View page source

Runtime

  • API
    • Runtime Layered Architecture
    • Frontend API
      • NN API
      • NNFW API
    • Backend API
  • Core
    • Modules
      • Module ir
      • Module compiler
        • 1. Lowering
        • 2. Tensor Registration
          • Q. What are the differences between ‘operand’ and ‘tensor’?
        • 3. Linearization (Linear Executor Only)
        • 4. Kernel Generation
        • 5. Create Executor
      • Module exec
      • Module backend
  • Controlflow Operations
    • Defining controlflow operations
    • Implementation
      • Graph representation
      • Execution
        • Kernel Implementation
  • Executors
    • Understanding models
    • Linear Executor
    • Dataflow Executor (experimental)
    • Parallel Executor (experimental)
  • Heterogeneous Execution
    • Graph Transformation
    • Automatic Scheduling
  • Backend API
    • How backends are loaded
    • C and C++ API
      • C API
      • C++ API
    • Provided Backend Implementations
    • builtin Backend
    • Bundle Backends
      • cpu
      • acl_neon
      • acl_cl
  • Compute
    • ARMComputeEx
    • cker
  • Supported Operations and backend
    • Raw-data format (float32, int32, boolean, etc)
    • Quantization format (uint8 asymmetric)
    • Quantization format (int8)
Previous Next

© Copyright 2020, Samsung Research & contributors.

Built with Sphinx using a theme provided by Read the Docs.