OpenACC Programming and Best Practices Guide

• Introduction
  • Writing Portable Code
  • What is OpenACC?
• Accelerating an Application with OpenACC
  • OpenACC Directive Syntax
  • Porting Cycle
  • Heterogenous Computing Best Practices
  • Case Study - Jacobi Iteration
• Assess Application Performance
  • Baseline Profiling
  • Additional Profiling
  • Case Study - Analysis
• Parallelize Loops
  • The Kernels Construct
  • The Parallel Construct
  • Differences Between Parallel and Kernels
  • The Loop Construct
  • Routine Directive
  • Atomic Operations
  • Case Study - Parallelize
• Optimize Data Locality
  • Data Regions
  • Data Clauses
  • Unstructured Data Lifetimes
  • Update Directive
  • Best Practice: Offload Inefficient Operations to Maintain Data Locality
  • Case Study - Optimize Data Locality
• Optimize Loops
  • Efficient Loop Ordering
  • OpenACC’s 3 Levels of Parallelism
  • Mapping Parallelism to the Hardware
  • Collapse Clause
  • Routine Parallelism
  • Case Study - Optimize Loops
• OpenACC Interoperability
  • The Host Data Region
  • Using Device Pointers
  • Obtaining Device and Host Pointer Addresses
  • Additional Vendor-Specific Interoperability Features
• Advanced OpenACC Features
  • Asynchronous Operation
  • Multi-device Programming