Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
Beginner
26 Views

Sampling 1st level cache misses retired - Discrepancy between hotspot and source views

Hi,

When I view the hotspots results for 1st level cache misses retired events done by Event Based Sampling(EBS) in my program, a particular function: viscous_flux shows up with 202,571 samples. This was in GUI mode.

However I summed up the samples shown on each source line in the source view and got for the same function: 52,981. This was by using the command 'vtl view -code -mn SUmb_linux_intel -rva 0x3D8604', where the last is the address of the viscous_flux fn, and summing up the Ev1 column by a Perl script(which I've checked independently).

What's going on? Am I doing something wrong, I use the 'Calibrate value after sample for all the events' option and the sampling interval is set to 1ms.

Interestingly this discrepancy does not show up in a simpler test program.

I run VTune 3.0 on a Xeon Hyper threaded dual core. uname -a:
Linux localhost 2.6.9-5.ELsmp #1 SMP Wed Jan 5 19:30:39 EST 2005 i686 i686 i386 GNU/Linux

Thanks,
Aravind.
0 Kudos
1 Reply
Highlighted
26 Views

Aravind,

VERY interesting!

Can you clarify further: What vtl command line did you use before you ran the "vtl view ..."command that you listed below? (Something like "vtl activity -c sampling -app ... run")

In other words: below, it seems at least possible to me that you are examining the results of two different projects. One for vtlec, and a different one for vtl?

cheers

jdg

0 Kudos