topic OpenMP load imbalance in Analyzers

OpenMP load imbalance

mriedmann — Fri, 17 Jan 2025 08:35:45 GMT

In the past I used Oprofile for investigations on load imbalance because it has this nice table output with a column for each thread, see below. That make it easy to localize load imbalance. Is there a way to get a similar output formatting with VTune ?

Re: OpenMP load imbalance

yuzhang3_intel — Sat, 18 Jan 2025 13:23:01 GMT

VTune provides a similar format report.

You need to run the analysis type you need first, and then get report command line from the GUI, this can generate the report.

This is a hotspot result; you can press the >_ button to get the command line to generate the report.

Re: OpenMP load imbalance

mriedmann — Mon, 20 Jan 2025 08:47:40 GMT

Thanks, that helps a bit. However my main question is how to get the top hotspots listing side by side for each thread, not accumulated across all threads ?

Re: OpenMP load imbalance

yuzhang3_intel — Wed, 22 Jan 2025 09:04:31 GMT

Set Grouping to 'Thread/Function/Call Stack' and get the command line to generate the report by pressing the button '>_ ' on the right side.

vtune -report exec-query -rep-knob row-by="/GenericThread/Function/ParentCallStack" -sort-desc "CPU Time:Self" -rep-knob column-by="ViewpointGUIandCLIColumns" -r <result_dir>