The ChipList, by Adrian Offerman; The Processor Portal

Processor Selector

View: show / edit

bookmark bookmark site
bookmark permalink
Mon 21 May 2012, 18:00

Two tips of comparing Hotspots Analysis result and Lightweight-Hotspot Analysis result

Two tips of comparing Hotspots Analysis result and Lightweight-Hotspot Analysis result

There are two kinds of hotspot collectors provided by Intel(R) VTune(TM) Amplifier XE, called Hotspots Analysis and Lightweight-hotspots Analysis. Hotspots Analysis is to interrupt a process, collect samples in all active modules of this process and do post-analyzing after running. With "Sample Interval" 10ms as a default, Hotspots Analysis has 2%-5% of overhead. Lightweight-hotspots uses device driver (CPU clockticks sample interval) to capture samples in system mode so it has less overhead.

Here is a simple example code to be profiled (test.cpp and binary test.exe), see attached test.cpp file.

a. Run application, launched by Microsoft Visual Studio*


b. Run application with Hotspots Analysis


c. Run application with Lightweight-Hotspots Analysis


You can see that Elapsed time in Lightweight-Hotspots report is less than execution time when program was launched by Microsoft Visual Studio.

Conclusion 1: if the user is sensitive to overhead, use lightweight-hotspots instead.


Also, the user may find there were different results:

                                                CPU time (threadfunc)

Hotspots                                     9.584

Lightweight-Hotspots                   6.896


Why did they have different results? Step into their Bottom-up reports. (see below)

There were four working threads in this example, we only observe one thread because it ...

Filed under: Intel® VTune™ Amplifier XE Knowledge Base
Also published by:

ChipList news channel Last Months News

ChipList developers news channel Last Months Developers News

Page viewed 1585 times since Thu 31 May 2012, 8:42.