Skip to contentSkip to conversation

perf not measuring LLC events?

DISCUSSION Dervin Thunk 4 participants 5 messages Dec 15, 2012 – Jan 22, 2013

lore
Share & tools
lore

Dervin Thunk

author
#0 lore raw

Dear all,

I'm trying to measure the LLC events of a program. I'm running linux-tools-3.2.0- 34 on a fairly recent 4 core Intel machine. For a certain program that reads a 64 MB file into memory, I get

Performance counter stats for './algo data/E.dna':

                 0 LLC-loads                                                   
                 0 LLC-load-misses
        26,538,349 cache-misses                                                
     4,465,644,541 L1-dcache-misses                                            

      35.116615511 seconds time elapsed

I have tried this in another machine, and it does give me some results for LLC. Also Valgrind gives me results (different from 0, that is)

Any idea what I'm doing wrong? Anyone else experiencing this problem with perf right now? Please let me know if I need to provide more information.

Thanks.

#1 lore raw
Quoted 11 lines · Dervin Thunk <dervin.thunk <at> gmail.com> writes:

Dear all,

I'm trying to measure the LLC events of a program. I'm running linux-tools-3.2.0- 34 on a fairly recent 4 core Intel machine. For a certain program that reads a 64 MB file into memory, I get

Performance counter stats for './algo data/E.dna':

0 LLC-loads
0 LLC-load-misses

Dervin,
I am experiencing the same issue. I am running perf on an i5 machine. Have you gotten any feedback regarding this?
-Jeff

#2 lore raw
Quoted 11 lines · Jeff Johnson
Quoted 9 lines · Dervin Thunk <dervin.thunk <at> gmail.com> writes:

Dear all,

I'm trying to measure the LLC events of a program. I'm running linux-tools-3.2.0- 34 on a fairly recent 4 core Intel machine. For a certain program that reads a 64 MB file into memory, I get

Performance counter stats for './algo data/E.dna':

Quoted 7 lines
Quoted 2 lines

0 LLC-loads
0 LLC-load-misses

Dervin,
I am experiencing the same issue. I am running perf on an i5 machine. Have you gotten any feedback regarding this?

I'm getting some 'sane' numbers on 2 cores i7 SandyBridge on 3.6.11-1.fc17.x86_64

not sure if this was broken back in 3.2.0

please be more specific on what system you're testing on and provide whole perf stat command line

jirka

#3 lore raw

Specs:
i5 Sandy Bridge 2400
Model 42
Stepping 7

Ubuntu 12.10 64 bit.
Dell Optiplex 790

Quoted 20 lines · Jiri Olsa
Quoted 11 lines · Jeff Johnson
Quoted 9 lines · Dervin Thunk <dervin.thunk <at> gmail.com> writes:

Dear all,

I'm trying to measure the LLC events of a program. I'm running linux-tools-3.2.0- 34 on a fairly recent 4 core Intel machine. For a certain program that reads a 64 MB file into memory, I get

Performance counter stats for './algo data/E.dna':

Quoted 7 lines
Quoted 2 lines

0 LLC-loads
0 LLC-load-misses

Dervin,
I am experiencing the same issue. I am running perf on an i5 machine. Have you gotten any feedback regarding this?

Quoted 8 lines

I'm getting some 'sane' numbers on 2 cores i7 SandyBridge on 3.6.11-1.fc17.x86_64

not sure if this was broken back in 3.2.0

please be more specific on what system you're testing on and provide whole perf stat command line

jirka

Specs:
i5 Sandy Bridge 2400
Model 42
Stepping 7

Ubuntu 12.10 64 bit.
Dell Optiplex 790

Perf command: perf stat -e LLC-loads -r 3 <my binary under test>

Thanks,
Jeff

#4 lore raw
Quoted 44 lines · Jeff

Specs:
i5 Sandy Bridge 2400
Model 42
Stepping 7

Ubuntu 12.10 64 bit.
Dell Optiplex 790

Quoted 20 lines · Jiri Olsa
Quoted 11 lines · Jeff Johnson
Quoted 9 lines · Dervin Thunk <dervin.thunk <at> gmail.com> writes:

Dear all,

I'm trying to measure the LLC events of a program. I'm running linux-tools-3.2.0- 34 on a fairly recent 4 core Intel machine. For a certain program that reads a 64 MB file into memory, I get

Performance counter stats for './algo data/E.dna':

Quoted 7 lines
Quoted 2 lines

0 LLC-loads
0 LLC-load-misses

Dervin,
I am experiencing the same issue. I am running perf on an i5 machine. Have you gotten any feedback regarding this?

Quoted 8 lines

I'm getting some 'sane' numbers on 2 cores i7 SandyBridge on 3.6.11-1.fc17.x86_64

not sure if this was broken back in 3.2.0

please be more specific on what system you're testing on and provide whole perf stat command line

jirka

Specs:
i5 Sandy Bridge 2400
Model 42
Stepping 7

Quoted 4 lines

Ubuntu 12.10 64 bit.
Dell Optiplex 790

Perf command: perf stat -e LLC-loads -r 3 <my binary under test>

I found Ubuntu 12.10 is like 3.5 kernel, right?

There are some perf tool cache event fixies after 3.5, but non of the tags I tried showed the error.

Any chance you could try with latest upstream - kernel/perf?

jirka