Intel® Xeon Phi™ Coprocessor Analysis Configuration

For Intel® Xeon Phi™ coprocessor (code name: Knights Corner), configure the Intel® VTune™ Amplifier either to run the predefined Advanced Hotspots, General Exploration, or Memory Access analysis types, or create a custom analysis type with an arbitrary set of hardware events.

1. Click the New Analysis button on the VTune Amplifier toolbar.

   The New Amplifier Result tab opens with the Analysis Type window active. The Analysis Type tab displays only analysis types applicable to the Intel Xeon Phi coprocessor.

2. In the Analysis Type tab, select the required analysis type from the left pane.

   The corresponding configuration pane opens on the right.

3. Configure the selected analysis type, if required.

   Note

   For Intel Xeon Phi coprocessor analysis, the call stack collection is supported only for the Intel Xeon Phi coprocessor (native) > Launch Application target types.

4. Click the Start button on the right to start the analysis.

   VTune Amplifier collects samples for a list of events predefined in the analysis configuration. You may change this list by creating a new custom configuration: New Knights Corner Hardware Event-based Sampling Analysis.

   The number of hardware events that can be collected simultaneously is limited by CPU capabilities. It is no more than two events on an Intel Xeon Phi coprocessor card. To overcome this limitation, the VTune Amplifier splits the event list into several event groups. Each group consists of events that can be collected simultaneously. VTune Amplifier uses one of the following techniques:

   • Runs an application only once and multiplexes the event groups in a round-robin fashion during the run (default mode).

   • Runs an application several times collecting one event group during each run to achieve more precise results. To enable this mode in the GUI, set the Allow multiple runs option in the Advanced section of Analysis Type tab.

   Note

   In this version of the product, the Allow multiple runs option is not available for the system-wide collection. So, you should manually distribute events by multiple runs if you do not want to use the event multiplexing mode.