numba-dpex

Core Features

  • Code-generation based on a device
  • Automatic offload of NumPy expressions

User Guides

  • Getting Started
  • Programming SYCL Kernels
    • Writing SYCL Kernels
    • Memory Management
    • Synchronization Functions
    • Writing Device Functions
    • Supported Atomic Operations
    • Defining the execution queue for a kernel function
    • Supported Address Space Qualifiers
    • Reduction on SYCL-supported Devices
    • Universal Functions
    • Supported Python Features in Kernel
  • Debugging with GDB
  • For numba.cuda Programmers

Developer Guides

  • DPNP integration
  • Debugging the compilation pipeline
numba-dpex
  • »
  • Programming SYCL Kernels Using kernel()
  • View page source

Programming SYCL Kernels Using kernel()

  • Writing SYCL Kernels
    • Introduction
    • Kernel declaration
    • Kernel invocation
    • Indexing functions
  • Memory Management
    • SYCL USM Array Interface
    • Device-only memory and explicit data transfer
    • Local memory
    • Private and Constant memory
  • Synchronization Functions
  • Writing Device Functions
  • Supported Atomic Operations
    • Example
    • Generating Native FP Atomics
    • Full examples
  • Defining the execution queue for a kernel function
    • Example
  • Supported Address Space Qualifiers
  • Reduction on SYCL-supported Devices
    • Examples
    • Full examples
  • Universal Functions
    • Example 1: Basic Example
    • Example 2: Calling numba.vectorize inside a numba_dpex.kernel
    • Full Examples
  • Supported Python Features in Kernel
    • Built-in types
    • Built-in functions
    • Standard library modules
    • Unsupported Constructs
    • NumPy support
Previous Next

© Copyright 2021, Intel.

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