From: Avinash S. (avishriv) <avi...@ci...> - 2009-04-27 23:10:20
Attachments:
benchmark_code.c
opreport_output.txt
|
Hi, I am trying to use oProfile for finding cache misses in an application code. I am running it on the board which has "Cavium Octeon 58xx" CPU. I am facing an issue with CIMISS and DMLDS event counters. I am allocating one buffer (worth one cache line) and accessing it multiple times. (PS: attached benchmark_code.c). I ran the benchmark_code multiple times, with different argv[1] (i.e. as 1, 100, 1000, 10000, 40000, 60000). I expected data cache misses (as seen in DMLDS) to reduce (from 100%), when the same element is accessed more number of times. But when I ran this program, I don't see any reduction in the misses (PS: attached oprreport_output.txt). Can someone please let me know:- -if I am missing something or if my understanding is not correct ? -what can I try differently, to find out the data cache misses in an application code ? As per http://oprofile.sourceforge.net/doc/index.html, I have configured the events as opcontrol --event=CIMISS:1000 --event=DMLDS:1000. Any pointers or answers would be of great help. Thanks, Avinash. |