By: anon (anon.delete@this.anon.com), October 31, 2008 2:57 am
Room: Moderated Discussions
Jouni Osmala (josmala@cc.hut.fi) on 10/31/08 wrote:
---------------------------
>Core2 has more than one Icache fetch per retired instruction.
>It means relatively high branch density in code, but it normally would indicate
>really catastrophicly bad branch predictor. Since in normal situations only way
>you get over 1 fetch per instruction is by branch missprediction.
>
>Amd figures except for the anomaly of farcry_high&X3 are what should be expected.
>As number of accesses per instruction.
>
>One possibility is that intel figures are from counting fetched instructions instead Icache fetches.
Fetching across cachelines might count as 2.
---------------------------
>Core2 has more than one Icache fetch per retired instruction.
>It means relatively high branch density in code, but it normally would indicate
>really catastrophicly bad branch predictor. Since in normal situations only way
>you get over 1 fetch per instruction is by branch missprediction.
>
>Amd figures except for the anomaly of farcry_high&X3 are what should be expected.
>As number of accesses per instruction.
>
>One possibility is that intel figures are from counting fetched instructions instead Icache fetches.
Fetching across cachelines might count as 2.
