Author: Robert Hyatt
Date: 08:14:22 01/18/04
Go up one level in this thread
On January 18, 2004 at 06:42:22, Alessandro Damiani wrote: >>>As Gerd said, the two additional instructions can be put into the SlideIndexXX. >>>I am going to test the smaller tables. >> >>While I am not doing it exactly as he does, I found the 75% memory reduction >>helped performance. >> > >Mmmh. Currently by switching to the smaller tables worsened performance by about >25%. I could improve it a bit (register calculation instead of table look-up), >but it is still about that margin slower. Maybe it has to do with my old >machine, a Pentium II 450MHz (512kByte L2 cache)?? > >puzzled > >Alessandro In my code, +all+ that changed was reducing the array from attacks[64][256] to attacks[64][64]. _nothing_ was added, no extra shifts, or anything. So all I changed was to reduce the L1/L2 cache footprint by 75%. I already had to do the shift/and, so I just added one to the shift amount, and reduced the and mask to 6 bits. If you have to add anything else to support this, then it might cost more than it saves, but for me it was a _pure_ 96kb cache reduction per table and there are 4 of them. I now need 128kb total for all the attack tables, rather than 512kb.
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.