High Performance Computing: Programming GPU using Open ACC

Prerequisites

This course aims to provide a basic knowledge of GPU programming using OpenACC directives. The course is very hands-on oriented, aiming to give to you the opportunity to practice and experiment from the very beginning.

• All current Cambridge University members (departments and colleges)
• Especially University alumni (PhD) and researchers who already have access to HPCS computing facilities
• Further details regarding eligibility criteria are available

• It is mandatory to have knowledge of C or FORTRAN similar to Fortran: Introduction to Modern Fortran
  
• Knowledge of the Unix environment similar to the Unix: Introduction to the Command Line Interface (Self-paced)
  
• No specific knowledge of CUDA is needed (but appreciated)
  

• General introduction about accelerated computing
• APOD: Assess, Parallelize, Optimize, Deploy
• OpenACC basic syntax:
  • parallel loop directive
  • define data regions
  • data clause
  • kernel construct
  • reduction clause
• Managing parallelism: gang, worker, vector
• Understand compiler output
• Mapping OpenACC to CUDA
  • Optimization techniques:
  • kernel fusion
  • async & wait clauses
  • pipelining
  • advanced data layouts

• Presentations and practicals

• OpenACC (PGI compiler installed on HPCS systems) and PuTTY on MCS Windows

• HPCS users will have the opportunity to perform experiments directly to Wilkes
• Participants are encouraged to bring their own scientific applications and discuss with the trainers about parallelization strategies and opportunities

• Two sessions of half a day each

• High Performance Computing: An Introduction
• Unix: Building, Installing and Running Software
• Unix: Introduction to the Command Line Interface (Self-paced)
• Unix: Simple Shell Scripting for Scientists

• High Performance Computing
• Programming and Scripting
• Scientific Computing
• UIS Systems