Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Playing the NPS game

Author: Pham Minh Tri

Date: 19:19:22 09/26/01

Go up one level in this thread


On September 26, 2001 at 20:57:53, Ren Wu wrote:

>Here is Initiative's log, two pos ecm 1582 and ecm 1527, searched up to 8 plies.
>
>     383.008  16.0      383.008  16.0  1809781 _Attacking (rcgen.obj)
>     269.220  11.2      269.220  11.2   169580 _GenCapMove (rcgen.obj)
>     256.254  10.7     2220.588  92.6   967692 _NegaScout (rcsech.obj)
>     226.982   9.5      226.982   9.5   967693 _HashTableLookup (rchash.obj)
>     213.103   8.9      213.103   8.9  1719790 _Update (rceng.obj)
>     206.721   8.6      206.721   8.6   849079 _NextMove (rcsech.obj)
>     177.683   7.4      189.571   7.9    78714 _GenAllMove (rcgen.obj)
>     117.972   4.9      117.972   4.9        2 _HashTableClear (rchash.obj)
>      80.751   3.4       86.419   3.6    42184 _ScorePawn (rceva.obj)
>      74.670   3.1      435.134  18.1    78714 _SEMoveGen (rcsech.obj)
>      74.464   3.1      209.035   8.7   782713 _RCEngScore (rcsech.obj)
>      53.791   2.2       53.791   2.2  1719790 _Dndate (rceng.obj)
>      48.165   2.0      134.571   5.6   782715 _EvaPawnStruc (rceva.obj)
>      40.769   1.7      469.323  19.6   227145 _CapSech (rcsech.obj)
>
>Compare this with yours, from first few lines, things looks ok. However, i saw
>this from yours
>
>[6]     66.5    0.43   21.46 1642284+928083  qsearch [6]
>                5.26    4.70 2570367/2570368     opn_eval [7]
>                2.48    3.01 1613239/2066384     gen [8]
>                1.03    3.94 1584298/2036996     order_moves [9]
>
>is this quiese search routine? it tells me that it took about 2/3 of the time, i
>think this is a problem, which position are you using here?

But from the line:

  1.31     29.95     0.43  1642284     0.00     0.01  qsearch

I know that his qsearch consumes only 0.43% of total time for itself. That is
fine. 30-60% are in good range.

BTW, time of qsearch = time of qsearch itself + time of eval. I think in
general, the higher percent of eval time, the better program. If his program
runs slowly because of heavy eval - no problem.

>
>Mine only took 20% of the total time, and i don't use SEE, and so in theory I
>will search more capture nodes than you. Maybe your pruning code at quiese
>search borken, and so you end up search more nodes there, and because you use
>SEE, those nodes are expensive to search.
>
>try to get rid hash probe at capture, and use MVV/LVA to sort moves and see if
>helps.
>
>I am not sure if this help in anyway, good luck.
>
>btw mine get about 500K nps on a PIII-733. So I think yours can be speed up
>quite a bit.

Your program is so fast. I have the same machine (PIII-733, 128M RAM), Crafty:
250-280kn, mine: 200kns.

Some months ago, mine was only 130-150 knps. I increased its speech
significantly in two weeks just by revising and rewriting / important functions
and changing programming style - using more pointers instead of tables,
implement some tricky techniques, following some otimization tips from AMD
manual and some sites.

>
>Ren.



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.