Case Study: Performance Tuning of ISV Application

Index

• 1. Introduction
  • 1.1. Purpose of this document
  • 1.2. Application software to be tuned
• 2. Target performance and procedure of tuning
  • 2.1. Target performance and conditions
  • 2.2. Procedure of tuning
• 3. Tuning details and results
  • 3.1. Evaluation of the performance
  • 3.2. Cost of each function
  • 3.3. Tuning of the Application
    • 3.3.1. Tuning items
    • 3.3.2. Tuning results
• 4. Tuning items
  • 4.1. SIMDization of division operations and suppression of SIMDization for loops with small iteration counts
    • 4.1.1. Target for this tuning
    • 4.1.2. Analysis
    • 4.1.3. Tuning
    • 4.1.4. Evaluation of the performance
  • 4.2. Reducing load and store operations of data by loop unrolling
    • 4.2.1. Target for this tuning
    • 4.2.2. Analysis
    • 4.2.3. Tuning
    • 4.2.4. Evaluation of the performance
  • 4.3. SIMDization by loop collapse
    • 4.3.1. Target for this tuning
    • 4.3.2. Analysis
    • 4.3.3. Tuning
    • 4.3.4. Evaluation of the performance
  • 4.4. Changing the access direction of arrays
    • 4.4.1. Target for this tuning
    • 4.4.2. Analysis
    • 4.4.3. Tuning
    • 4.4.4. Evaluation of the performance
  • 4.5. SIMDization by SVE ACLE
    • 4.5.1. Target for this tuning
    • 4.5.2. Analysis
    • 4.5.3. Tuning
    • 4.5.4. Evaluation of the performance
  • 4.6. Built-in prefetch
    • 4.6.1. Target for this tuning
    • 4.6.2. Analysis
    • 4.6.3. Tuning
    • 4.6.4. Evaluation of the performance
  • 4.7. Moving division operations to outside of the loop, and applying SIMDization to the division operations
    • 4.7.1. Target for this tuning
    • 4.7.2. Analysis
    • 4.7.3. Tuning
    • 4.7.4. Evaluation of the performance
  • 4.8. Moving invariant expressions to outside of the loop
    • 4.8.1. Target for this tuning
    • 4.8.2. Analysis
    • 4.8.3. Tuning
    • 4.8.4. Evaluation of the performance
  • 4.9. Loop unrolling manually instead of OCLs
    • 4.9.1. Target for this tuning
    • 4.9.2. Analysis
    • 4.9.3. Tuning
    • 4.9.4. Evaluation of the performance
  • 4.10. Improving the memory placement of two-dimensional arrays for sequential access
    • 4.10.1. Target for this tuning
    • 4.10.2. Analysis
    • 4.10.3. Tuning
    • 4.10.4. Evaluation of the performance
• 5. Summary