Main Contents

The ParkBench Benchmark Collection

Jack Dongarra and Tony Hey

• Introduction
  • Background and Objectives
  • Programming Models
  • Computer Terminology
  • How to get the PARKBENCH Report and Benchmarks
• Benchmarking Procedure and Code Optimisation
• Low-Level Benchmarks
  • Introduction
    • Most Reported Benchmark: LINPACKD (n=100)
    • Performance Range: The Livermore Loops
  • Single-Processor Benchmarks
    • Timer resolution: TICK1
    • Timer value: TICK2
    • Basic Arithmetic Operations: RINF1
    • Memory-Bottleneck Benchmarks: POLY1 and POLY2
  • Multi-Processor Benchmarks
    • Communication Benchmarks: COMMS1 and COMMS2
    • Total Saturation Bandwidth: COMMS3
    • Communication Bottleneck: POLY3
    • Synchronisation Benchmarks: SYNCH1
  • Summary of Benchmarks
• Kernel Benchmarks
  • Introduction and Rationale
  • The Kernel Benchmarks
    • Matrix benchmarks
    • Fourier Transforms
  • Benchmark Implementation
  • Concluding Remarks
• Compact Applications
  • Introduction
    • Self Checking Applications
    • Programming Languages
  • Proposed Compact Application Benchmarks
• HPF Compiler Benchmarks
  • Objectives
  • Summary
• Conclusions
  • Acknowledgements
• References
• Contents
• List of Figures
• List of Tables