Author: Robert Hyatt
Date: 12:23:40 06/21/02
Go up one level in this thread
On June 20, 2002 at 18:21:47, Dann Corbit wrote: >ftp://cap.connx.com/pub/beowulf.jpg > >The hot spot is quiesce, which has as its stumbling blocks the long long shift >right routine and hashprobe. Hashprobe is held up by long long modulo. > >I have no idea if a switch to 64 bits will help. why not replace that long long module with address = (int) (hash_signature) % table_size That is effectively what us 2^N guys are doing with the AND function. No reason the upper end of the hash signature has to influence the table position, since the upper end is hopefully stored in the entry for verification. That will cull the long long modulo calculation. Of course AND is _still_ much faster than a native integer divide with remainder..
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.