esithole

Resolving profile results based on same scenario

Discussion created by esithole on Apr 20, 2010

Hi all,

  

I’ve run 3 measurement sessions for the same scenario using the CodeAnalyst standard profiles (Investigate Data Access & Investigate L2 Access) as well as a customised profile (that combines events for Data Cache Accesses, L2 Accesses, Cache refills from system memory, Cache Line evictions, DRAM Accesses, Retired Instructions and CPU Clocks).  For each of 3 measurement sessions I have set up, the duration of the profile is 100 seconds. However, as can be seen from the results I’m getting (see tables below)  the output measurements for DC accesses, DC refills L2 misses aren’t comparable at all.

 

Any suggestions as to how I can reconcile or account for the wide variation in the results from the three profiles?

 

Profile 1 – Investigate Data Access

DC accesses

DC misses

DTLB L1M L2H

DTLB L1M L2M

Misalign access

Ret inst

DC refills L2/sys

70453

21434

559

377

0

174032

21407

 

Profile 2 - Investigate L2 Cache Access

Ret inst

L2 fill/write

L2 requests

L2 misses

176396

47379

49555

37990

 

Profile 3 – Customized/Combined Profile

DC accesses

DC misses

CPU clocks

Ret inst

L2 fill/write

L2 requests

L2 misses

DRAM accesses

DC refills L2/sys

DC refills sys

DC evicted

47737

14344

45586

297980

8029

8124

6426

12075

14266

6302

13515

 

 

Regards,

 

esithole

 

Outcomes