By: David Kanter (dkanter.delete@this.realworldtech.com), October 31, 2008 6:19 pm
Room: Moderated Discussions
anon.moose (anon,moose@r00t.org) on 10/31/08 wrote:
---------------------------
>anon (anon@anon.com) on 10/31/08 wrote:
>
>>Fetching across cachelines might count as 2.
>
>Cachelines are fairly large compared to instructions, it wouldn't affect the >numbers greatly.
So my initial inclination is that Jouni's hypothesis is probably correct.
The differences in the results were pretty large, about a factor of 3-8X. This suggests to me that one is probably measuring instruction cache accesses, while the other may be measuring instruction fetches from the small pre-decode buffer.
David
---------------------------
>anon (anon@anon.com) on 10/31/08 wrote:
>
>>Fetching across cachelines might count as 2.
>
>Cachelines are fairly large compared to instructions, it wouldn't affect the >numbers greatly.
So my initial inclination is that Jouni's hypothesis is probably correct.
The differences in the results were pretty large, about a factor of 3-8X. This suggests to me that one is probably measuring instruction cache accesses, while the other may be measuring instruction fetches from the small pre-decode buffer.
David
