Intel® Advisor User Guide

User Guide

Contents

Analyze Loop Call Count

Accuracy Level

Medium

Enabled Analyses

Survey + Trip Counts (Characterization)
Collecting additional data may substantially increase report generation time. There is a variety of techniques available to minimize data collection, result size, and execution time. Check Minimize Analysis Overhead.

Result Interpretation

After you run the
Vectorization and Code Insights
 perspective with
medium
 accuracy and Trip Counts collection enabled,
Intel® Advisor
 dynamically identifies the number of times loops are invoked and execute and extends the basic vectorization report with the Trip Counts data. Use Trip Counts data to analyze parallelism granularity more deeply and fine-tune vector efficiency and capability.
Columns with Trip Count metrics of the Survey report (Vectorization perspective)
By default, the
Trip Counts
 column shows only
Average
 and
Call Count
 metrics. Look for the following to find good candidates for optimization:
  • Detect loops with too-small trip counts and trip counts that are not a multiple of vector length.
  • A high number in the
    Call Count
     column means there is an outer loop in the selected loop call chain with high trip count values.
  • If the loop has a low trip count value, the outer loop could be a better candidate for parallelization (threading/vectorization).
To optimize such loops, follow the
Intel® Advisor
Recommendations
 for the loop/function, for example, use specific recommended pragmas to provide the information about loop trip counts to a compiler.

Next Steps

For further investigation, you can run the
Vectorization and Code Insights
 perspective with a higher accuracy level or with different configurations:

Product and Performance Information

1

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.