Link Copied
Was it possible that the program went different paths for Hotspots and chronometer?
Did youcompare outpus of program for these two sessions?
It's hard image that the elapsed time of running application withHotspots is half of the original.
Can you please provide a simple test case to reproduce this issue? Or submit a ticket tohttps://premier.intel.com
Regards, Peter
Hi,
Here is the new (and weirdest) chapter on my VTune experience
Yesterday I developed a very simple Win32 Console application that executed a loop 100 times. Within the loop I loaded a JPG image using OpenCVs highgui and then called several image processing functions on that image to finally save the resultant image to disk. The total execution time was about 12 seconds.
I run the software using the VTune Hotspots analysis and everything went well. It took about 12 seconds to run and the elapsed time as shown in the Summary window was about 12 seconds. Great!
I also tried some of the sample programs included with the suite and everything went fine. After that I spent the rest of the day doing something else (recompiling the OpenCV dlls so they can use IPP 7.0).
Well, here is when things get really weird:
This morning I tried again my MFC application that executes calls to my dll (this dll also uses OpenCV functions).
Yesterday the 26 seconds test execution was producing an elapsed time of 13 seconds in the Summary window of VTune. All the times in the bottom-up analysis were consistent with the 13 seconds total elapsed time (this means that they were also about half what they really are).
Well, this morning the same test with the same MFC application and the same DLL is giving me the correct results (about 26 seconds elapsed time for a 26 seconds execution, and so on) I restored the previous OpenCV dlls (the ones that dont use IPP) just in case this was making a difference, but still getting the correct results.
I know how this sounds but I swear to all the gods that yesterday I was getting different times. I dont know if anybody ever encountered this kind of thing before. It looks like the typical joke about fixing problems by switching off and on the computer ;-)
Anyway, Ill keep an eye open in case I find out anything else and Ill post here about any other poltergeist that I might encounter.
Tulgas
For more complete information about compiler optimizations, see our Optimization Notice.