Window: Vectorization Summary

After running the Vectorization and Code Insights perspective , consider reviewing a results summary that includes the most important information about your code. Click the Summary tab after running an analysis to view results.

Program Metrics Pane

View the main performance metrics of your program, such as execution time statistics, vector instruction set (and whether extensions, such as VNNI, are used), and number of CPU threads utilized. The section is broken down into several sub-sections:

• Performance characteristics: View execution time details, such as total CPU time and time spent in vectorized and scalar code.

  If your application uses Intel® oneAPI Math Kernel Library (oneMKL), you will see the MKL detail button in the Performance characteristics section, which toggles two additional columns: the User column, which reports time spent in your code and corresponding compute metrics, and the MKL column, which reports time spent in the oneMKL code and corresponding compute metrics.

• Vectorization Gain/Efficiency: View average estimated speedup of vectorized loops and total estimated program speedup.

  Note

  The vectorization efficiency data is available only for vectorized loops.

• OP/S and Bandwidth: View GFLOPS and GINTOPS usage and cache bandwidth metrics compared to hardware peak. Hover the mouse over the Utilization column and click the button to select single-core or multicore benchmarks utilization metrics.

  Note

  The OP/S and bandwidth metrics are available after you run the Trip Counts and FLOP or the Roofline analysis.

Per Program Recommendations Pane

View suggested changes for your program that you might want to apply to achieve better performance.

Top Time-consuming Loops Pane

View top five time-consuming loops sorted by self time with performance metrics, such as execution time statistics and vectorization efficiency with comparison to original scalar loop efficiency.

Refinement Analysis Data Pane

View details about found dependencies and memory access patterns.

The Dependencies column summarizes the predicted data sharing problems collected by the Dependencies tool. To display the Dependencies Report window at the corresponding parallel site location, click a function link in the Site Location column.

The Strides Distribution column reports the memory access stride distribution within a loop in the ratio format in %: unit strides, constant strides, and variable strides.

Recommendations Pane

View suggested changes with high confidence level for first five loops in the code that you might want to apply to achieve better performance. Click a recommendation link to access the recommendations texts.

Collection Details Pane

View execution statistics for each of the collectors, as well as the Collection Log, Application Output, and Collection Command Line links that lead to the corresponding report logs, command line and output details.

Platform Information Pane

View the system information including software and hardware summary.