Analyzers
Support for Analyzers (Intel VTune™ Profiler, Intel Advisor, Intel Inspector)
Announcements
This community is designed for sharing of public information. Please do not share Intel or third-party confidential information here.

measure "false sharing"

softarts
Beginner
166 Views
hwo to measure the effect of "false sharing"?

any KPI in VTUNE?
0 Kudos
3 Replies
Vladimir_T_Intel
Moderator
166 Views
Quoting - softarts
hwo to measure the effect of "false sharing"?

any KPI in VTUNE?

On the Core2 platform we use EXT_SNOOP.ALL_AGENTS.HITM event to identify snoop responses to bus transactions.HITM occurs when the snooped address at the responder's cache is in a Modified state. So, it reflects the number of modified data sharing cases.
softarts
Beginner
166 Views

On the Core2 platform we use EXT_SNOOP.ALL_AGENTS.HITM event to identify snoop responses to bus transactions.HITM occurs when the snooped address at the responder's cache is in a Modified state. So, it reflects the number of modified data sharing cases.

which method should use?call graph or sampling?

thanks!
Thomas_W_Intel
Employee
166 Views
Quoting - softarts

which method should use?call graph or sampling?

thanks!

Sampling. Vladimir has already named the event that you should sample. If you are unfamiliar with sampling, it is recommended that you read some introduction to sampling and try some basic tutorial before jumping into measuring false sharing. You will need to understand the difference between event and sample and know what the "sample-after-value" means, or it will be really hard to interpret the results. You can find such information in the documentation or the VTune book, and there come some examples with VTune.

Call-graph essentially measures the time that is spent in functions and captures how tfunction callsrelate to each other. It is therefore unsuited to identify false sharing.

Kind regards
thomas
Reply