The Parallel Universe Magazine

Intel’s quarterly magazine helps you take your software development into the future with the latest tools, tips, and training to expand your expertise.
 

Subscribe
Issue 47, January 2022

Feature: Vectorization in LLVM and GCC* for Intel® CPUs and GPUs

This issue focuses on evolving methods for optimizing parallel programming using oneAPI and other tools.

Download the current issue to find:

  • Editor in chief and senior Intel engineer Henry Gabb’s take on the momentum around oneAPI going into 2022, including a new specification and another award.
  • The feature article explores three methods for exploiting SIMD parallelism, including which one gets the best results.
  • Three data science articles introducing tools and methods to help you increase efficiency, reduce code rewrites, and scale workflows.
  • Best practices to keep the CPU and GPU working at the same time, an introduction to Level Zero programming, and more.
Download the PDF (15.7 MB)
Contents
Vectorization in LLVM and GCC for Intel CPUs and GPUs

SIMD Support Is Evolving Rapidly in Modern Compilers

 

Explore three methods for exploiting SIMD parallelism: automatic vectorization, programmer-guided vectorization, and a data parallel library approach.

ArrayFire Interoperability with oneAPI, Libraries, and OpenCL™ Code

Take Advantage of oneAPI to Avoid Code Rewrites

 

Explore how to integrate the oneAPI Deep Neural Network Library (oneDNN) and Data Parallel C++ (DPC++) into existing code bases to save time and minimize rewrites.

Hyperparameter Optimization with SigOpt for MLPerf Training on Habana Gaudi*

Build End-to-End AI Use Cases with BigDL on Top of Ray

 

Learn about the core components of BigDL and how BigDL leverages Ray to help you build AI applications using Project Chronos.

Scale Your pandas Workflow with Modin*

Scalable Data Analytics with No Rewrite Required

 

Discover the Intel® Distribution of Modin*, a powerful pandas replacement that enables you to speed up insights without rewriting your code.

Efficient Heterogeneous Parallel Programming Using OpenMP*

Best Practices to Keep the CPU and GPU Working at the Same Time

 

Learn how to express true asynchronous, heterogeneous parallelism using standard OpenMP* directives.

Use the oneAPI Level Zero Interface

A Brief Introduction to the Level Zero API

 

Explore a basic example of Level Zero programming to understand the fine-grained, explicit control oneAPI Level Zero provides over device discovery, memory management, and more.

Hyperparameter Optimization with SigOpt for MLPerf Training on Habana Gaudi*

Achieve Faster Convergence with Higher Accuracy in AI Training

 

See why hyperparameter optimization is an effective method for optimizing performance of large-scale, deep-learning training workloads.

From Ray to Chronos

Build End-to-End AI Use Cases with BigDL on Top of Ray

 

Learn about the core components of BigDL and how BigDL leverages Ray to help you build AI applications using Project Chronos.