is it possible to configure a combined Locks&Waits and CPU Hotspots view in VTune?
What I imagine is a view which shows the complete wall time structure of the callgraph profile data. The numbers show how long a given function resides on the stack. The colors of the bars (currently they show whether the CPU is over/undercommitted) would instead distinguish between CPU bound hotspot and Lock/Wait time. I.e.:
main | 10s | 8s CPU | 2s Locks&Waits
- bar | 8s | 7s CPU | 1s Locks&Waits
- foo | 2s | 1s CPU | 1s Locks&Waits
The advantages I see are that I don't have to switch between the two views. Furthermore, I could potentially find "hidden hotspots" which are due to a high combined count of wait and CPU time in a function.
So, is this somehow possible to configure in VTune now? If not, could this maybe integrated in future versions of the tool?
Have you tried "Hotspots by Thread Concurrency" viewpoint?
Click on the link "Change" on the top of result to choose it. Use "Function / Call Stack" to see hotspots with CPU and Wait time metrics.
Indeed, that comes very close to it - thanks. Its only lacking the combined column though, which one could use in the Caller/Callee view to sort such that for these contrived results:
function | cpu time | wait time | combined
foo | 5 | 1 | 6
bar | 1 | 6 | 6
asdf | 4 | 4 | 8
But its certainly very good already - thank you for the hint!