- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
hello
how can we estimate the number of miss L1 and miss L2 only for code within loop.
for exemple
for(i=1;i++;i<10000)
{
begin count miss L1
code;
end count miss L1
print(missL1)
}
i knew that we can do that with PAPI http://icl.cs.utk.edu/papi/
my question how can we instrument my code by function for exemple to estimate miss in a bloc of instruction
how can we estimate the number of miss L1 and miss L2 only for code within loop.
for exemple
for(i=1;i++;i<10000)
{
begin count miss L1
code;
end count miss L1
print(missL1)
}
i knew that we can do that with PAPI http://icl.cs.utk.edu/papi/
my question how can we instrument my code by function for exemple to estimate miss in a bloc of instruction
Link Copied
1 Reply
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
VTune Performance Analyzer's sampling data collectorcounts L1$ misses and L2$ missesin system-wide,I meanto collect for ALL active programs in the system - including yourrunning program.
For example,on Intel Core 2 processors- use event named L1D_REPL to count L1 misses.
It is a hardware interruption mechanism, and not instrumentuser's code. VTune Performance Analyzer does post-analyzing performance data and report results on product's GUI, so you can print missL1 in your code.
You may want to profile within your interest of code, thus, you can use "pause" at samplingcollector's startinganduse "resume" of VTuneAPI in your code. See more detail atpage 8 of http://cache-www.intel.com/cd/00/00/21/93/219345_sampling_vtune.pdf
Thanks, Peter

Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page