Collecting Accurate Event Data

Intel® VTune™ Amplifier runs the advanced hardware-level analysis to collect data based on the events defined for the selected analysis type. The number of events it can monitor during a single run is limited by the number of performance counters in your processor. You may choose to enable multiple runs of the data collection to get the most precise results. In this case, the VTune Amplifier runs the hardware event-based sampling data collector as many times as required to collect data on all the events specified for the analysis type. If you specified an application to launch as an analysis target, the VTune Amplifier launches your application each time the hardware event-based sampling collector runs.

VTune Amplifier allows to avoid multiple runs of the data collection by multiplexing the use of physical counters within a single sampling run. Event multiplexing removes the need for multiple runs of the application, thereby reducing the time needed to complete sampling collection at the cost of lower precision of the result data. Event sample counts collected in the multiplexed mode are extrapolated to the total collection runtime.

Event multiplexing is also useful if the application does not have a long steady state or takes a long time to get to steady state. On the other hand, if application initialization is short and it gets to steady state quickly, then you can do multiple short runs and will not need to do event multiplexing.

To enable/disable multiple runs of the data collection:

  1. Click the Project Properties button on the VTune Amplifier toolbar.

    The Project Properties: Target tab opens with the Application to Launch pane active.

    Note

    Collecting data in multiple runs is only possible if an application to launch is specified.

  2. Scroll down to the Advanced section and select the Allow multiple runs option to enable more precise event data collection or deselect the option to use event multiplexing.
For more complete information about compiler optimizations, see our Optimization Notice.