Intel VTune Performance Analyzer can measure your application's performance, but not for (Frames per Second) FPS.
In general VTune Analyzer captures samples in your code when application is running.Depends on total samples in different processes, threads, modules, functions- we know hot functions in your code.Roughlywe can estimate execution timeofhot function(s)= total_events / CPU_Frequency (seconds) = (samples * SAV) / CPU_Frequency (seconds)
1. SAV means SampleAfter Value, for interval - how many ticks elapsed then capture a sample
2. We eliminated interruption handlingtime when VTune Analyzer captures a sample.I believe ISR routine take less time when interrupting, save a sample then return immediately.
Back to your question, VTune can estimate time spending in your critical (hot) function, but for video application specific.
PerhapsI can understand during specific test time - if time spending in your critical function is bigger, FPS is bigger. Am I right?
Also, this depends a lot on what you are using to render the frames! If you are using third-party software, look to see if they provide any performance counters of their own. In Microsoft Windows*, software counters are exposed via the performance counter subsystem, commonly referred to as perfmon. The VTune analyzer exposed these counters in the Counter Monitor and can correlate the software counters to processor counters in the sampling data.
Last week I came across a white paper on your site that described using the Windows VTune analyzer to capture FPS metrics. I was looking for it today and cannot find it. Can you provide a link. Also, is there a better recommendation for measuring FPS on a Linux system?
I'm not sure what paper you may remember. There's no feature in VTune analyzer that will directly measure frame rate (per second or per any other measure), but there may be hooks for a particular application that might be used to capture the frequency that some particular code, perhaps some frame-completion function, is executed. This could be accumulagted but would be at best an approximate measure if using VTune analyzer's (EBS) sampling collector, since the collector would also be sampling the data.
There are alternatives such as Intel's Graphics Performance Analyzer, which with their hook into the Microsoft DirectX interface can obtain frame completion information from that interface and therefore has the data to calculate a frame rate, and I've seen applications that measure and display frame rate on their own. GPA won't do you much good on Linux, though.
I think you pegged the methodology used in the paper. Your right GPA won't help in our case. So I am still looking for a way to get the job done on a Linux system. As I have it now the need is: "
measure the frames per second of a video running in Windows Media Player over a Rdesktop connection in Linux"
So any help would be greatly appreciated.