A simple question regarding Vtune usage

A simple question regarding Vtune usage

The question is really elementary. I am running Suse linux on P4. All I want to try now is to gather L1 data miss/reference on a simple triple loop (say a Matrix-Matrix-Multiplication), only the reference & miss regarding the loop. Is there a simple way to do it?
Thanks a million!!


2 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hey Joy,

Great question! (And nice to see you posting here.)

Since you seem like the kind of person who is both curious AND adventurous, here's your (QUICK!) homework:

  1. find out the supported CPU events for your server (which for Intel always translates into "for your processor") and then
  2. check out the reference manual for some pretty spot-on descriptions of how to work with them.

Continuing to stay general for a minute:

1) find out the events you can sample for on your precise system:

$ vtl -help -c sampling | less
$ vtl -help -c sampling > readme
$ vi readme

2) Bring up the Reference Manual (examples below are for the default installation tree):

$ cd /opt/intel/vtune/doc/reference
$ netscape index.htm &

(If netscape isn't your favorite browser, go with your personal fave...)

If you take a few minutes and track this stuff down, Joy, I think you'll be on the track to some fast answers.

NOW, while I'm not sure the following will answer your specific questions, as it happens I have a test Pentium 4 system right here at my desk. I list out the L1 sampling events, and here's they are:

L1 Lines Allocated
L1 M-state Data Cache Lines Allocated
L1 M-state Data Cache Lines Evicted
L1 Misses outstanding (weighted)

If you scan those out in the Reference Manual, I'm guessing it may give you what you need.

Either way, let us know if this posting helped!



Message Edited by jdgallag on 02-24-2004 10:03 AM

Leave a Comment

Please sign in to add a comment. Not a member? Join today