Author: Vincent Diepeveen
Date: 11:30:36 12/07/99
Go up one level in this thread
On December 07, 1999 at 10:32:38, Brian Richardson wrote: >On December 07, 1999 at 08:28:53, Vincent Diepeveen wrote: > >>On December 06, 1999 at 15:20:39, Brian Richardson wrote: >> >>>On December 06, 1999 at 11:29:56, Bas Hamstra wrote: >>> >>>>Hi Bob and other bb experts, >>>> >>>>What do you use the rotated bb's for, exactly: only for movegeneration, or also >>>>for attack detection? >>>> >>>>I ask this because I saw in Crafty some code to detect blocking of sliding >>>>attacks, like I do it. Whith a BlockMask[File][Rank]. I thought the essence of >>>>the rotated bb's is that you can avoid this block checking, so why this >>>>block-code in Crafty? >>>> >>>> >>>> >>>>Regards, >>>>Bas Hamstra. >>> >>>I had done some informal testing of rotated vs non for move generation with >>>Crafty and found non to be slightly faster (<10%) for move generation (I suspect >>>due to less overhead updating). However, for attack detection and eval >>>functions, I think rotated more than makes up for the additional complexity... >>> >>>Nonetheless, I do not use rotated in my program for simplicity (at this stage of >>>development, I prefer KISS to maximum performance). >>> >>>An easy test would be rotated vs non in Crafty's Attacked() function. >>>Differences probably vary significantly by game stage and mobility. >> >>Can you please post at what outdated hardware you tested that? > >Tested on PII 266, 512K cache, 64MB back in Sept 98 under Win95, MSVC 5. Cut'n >paste Crafty move generation, stripped hashing updates, but left incremental >material updates. Tried with and without rotated bitmaps. This was before >vcinline.h ! Certainly not a "fair" comparison, but I was only looking for >"big" differences (turned out much faster than gnuchess pointer method, and my >much older method, which I abandonded). Tested with perft timing using starting >position, which is another "problem". I am not a bitmap guru. Currently my >program only does about 1M per second (with vcinline.h), so perft to depth 6 >takes about 130 secs, which is about the same as Crafty 16.18 Again, for my >purposes this informal testing was adequate, since my program spends about 50% >of the time in eval(). I am sure well-done rotated bitmaps are a significant >improvement, but I am postponing rotated bitmaps until after the eval() gets >smarter, and pondering, books, learning and EGTBs get implemented. this is weird. please retest with a new crafty version. it's faster here than with the SMALL thing.
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.