I'm using tprofile in linux system and use gui mode in windows. In my application, the overhead for each parallel region appears frequently with small values. I didn't see any bars or so what to sum up the total overhead, such that I can know how the application can be improvemented by reducing the overhead. Is there any feature I can use to sum up the total overhead in the app?
Thanks a lot
If I understand your request correctly - there is no indicator for overheads caused by Thread Profiler. Thus, the user can measure overheads caused by Thread Porfiler, by running application with tprofiler, and without tprofiler - compare their execution time.
Thisis the fact - tprofile can measure target application performance, but not for tool itself:(
Thanks for the reply. So, tprofiler will measure the performance accurately even the system loading is high? For instance, if we have 16 cores, and 5 cores are executing other jobs and the application being profiled only occupied 8 cores, then the profile measurement is believable? I'm interested in this because when I use elapse time to profile my tool, the system loading greatly affect the result and I'm crazy about that sometimes. So if tprofiler will overcome this issue, that will be so great.
1. Intel Thread Profiler will not collect the performance data on other applications (which consume your CPU resource), only collect data for target application.
2. Ifother application(s)isrunning on background, it will impact on results (e.g. ElapsedTime)oftarget application.
3. Sometime such impacting is little. For example, your CPU utilization is lower (<20% - no high computing work on background), then use Intel Thread Profiler withtarget application.