Cache lines and branch prediction.

Scott A Crosby crosby at qwes.math.cmu.edu
Mon Oct 1 07:54:58 UTC 2001


On Sun, 30 Sep 2001, Scott A Crosby wrote:

>
> Note that this scheme still wastes 1/3 of the slots, but it will probably


I hate to reply to myself, but this is a fatal flaw: If 1/3 of the slots
are wasted, in that they're empty or likely overwritten, this means that
our hit rate is likewise going to be reduced to a max of 2/3 per probe.
This also means that the existing cache can, at *max* have a hit rate of
66% on ANY probe, whether first or last, unless it has been made so large
that we never wipe entries.

Sorry for barking up the wrong tree.

Scott





More information about the Squeak-dev mailing list