Enable the stack collection for the hardware event-based sampling analysis and configure the Intel® VTune™ Amplifier to statistically approximate the number of trip counts for sampled loops.
To enable the call count collection:
Click the
New Analysis button on the VTune Amplifier toolbar.
The New Amplifier Result tab opens with the Analysis Type window active.
In the Analysis Type window, do any of the following:
From the left pane, select Algorithm Analysis > Advanced Hotspots and choose the Hotspots, call counts, loop trip counts, stacks and context switches Collection level option in the configuration pane on the right.
From the left pane, select Custom Analysis > Hardware Event-based Sampling Analysis Type and select the Estimate trip counts option in the Custom Analysis dialog box.
Click the Start button on the right to run the analysis.
VTune Amplifier collects the data statistically estimating the loop properties and provides the following metrics in the Hotspots viewpoint: Total Iteration Count, Loop Entry Count, and Average Loop Trip Count.
In the Source view, loop metrics are not always associated with the first line of the loop due to compiler optimizations, which may be confusing in case of nested loops. In this case, refer to the grid for proper association of metrics with loops. You are recommended to select Loops and functions or Loops only options from the Loop mode drop-down list to see the metrics for individual loops. In the Functions only mode, loop metrics are aggregated over functions.
Loop metrics are estimated for a compiled binary file. They may not correspond to source code due to compiler optimizations (for example, loop unrolling or vectorization).
Trip count estimation is not available on Intel® Xeon Phi™ coprocessor and microprocessors with Intel microarchitecture code name Nehalem or Silvermont.