Data Parallel Extensions for Python

Data Parallel Extensions for Python* extend numerical Python capabilities beyond CPU and allow even higher performance gains on data parallel devices, such as GPUs. It consists of three foundational packages:

• dpnp - Data Parallel Extensions for Numpy* - a library that implements a subset of Numpy that can be executed on any data parallel device. The subset is a drop-in replacement of core Numpy functions and numerical data types.

• numba_dpex - Data Parallel Extensions for Numba* - an extension for Numba compiler that lets you program data-parallel devices as you program CPU with Numba.

• dpctl - Data Parallel Control library that provides utilities for device selection, allocation of data on devices, tensor data structure along with Python* Array API Standard implementation, and support for creation of user-defined data-parallel extensions.

Table of Contents

• Prerequisites and Installation
  • 1. Device Drivers
  • 2. Python Interpreter
  • 3. Data Parallel Extensions for Python
• Parallelism in Modern Data-Parallel Architectures
• Heterogeneous Computing
  • Device Offload
  • Unified Shared Memory
  • Compute-Follows-Data
  • Copying Data Between Devices and Queues
  • Creating Additional Queues
• Programming with Data Parallel Extensions for Python
  • Scalars vs. 0-Dimensional Arrays
  • Data Parallel Extension for Numpy - dpnp
  • Data Parallel Extension for Numba - numba-dpex
  • Data Parallel Control - dpctl
  • Debugging and Profiling Data Parallel Extensions for Python
  • Writing Robust Numerical Codes for Heterogeneous Computing
• Examples
• Tutorials
  • Getting Started
  • Controlling `dpnp` fallback to `numpy`
• Benchmarks
• Demos
• Useful links
• To-Do