Author: Richard Pijl
Date: 10:09:22 06/18/02
Go up one level in this thread
> >>It is not only passing the list, also building and maintaining it. >>When I changed my pv array to retrieving from the hashtable, it saved >>me about 5%. > >5% seems incredibly high for just updating the PV array. I didn't measure it in >my engine (esp since I don't have HTs atm :) but still it seems way too much.. I >spend about 5% of the total time in makeMove() and this function surely is > >-called more often than updating the PV moves >-takes longer to execute > >So, in short, I'm puzzled by your 5%. :) > >Note: I could imagine to lose a bit when the arrays for the potential PVs are >created on the stack but someone who uses the 'triangle array approach' >shouldn't lose that much. > >On the other hand, I could be completely wrong. :) Do other people have numbers >to compare? > The gain was mostly in more efficient move ordering as described next. > >>I also simplified my moveordering (as I first checked for a PV move before >>probing the hashtable), but that could have been done anyway. > >Yes, that will be much easier. I think I will also do that as soon as I have >re-implemented HTs. > > >>I keep the rootnode move in a separate variable, to prevent it from being >>overwritten in the hashtable. > >Aha! A special case! :) I'm not sure but don't some people store the PV in the >hashtable after a search, in case the PV nodes were overwritten in the >meanwhile? Of course you'd have to keep the PV moves separately when doing >this... > I do store the rootnode move in the hashtable (in fact forcing it to overwrite anything that might be on that entry) for move ordering purposes. The problem is that when interrupting search (move now, out-of-time) the incomplete search may have overwritten the hashentry containing the root node move. Goldbar MMII at the latest CSVN tournament suffered from something related, causing simple losses against Holmes and some others I don't remember. After storing the rootmove in a separate variable the problems were over. >>On the other hand I have a full PV-line on a 1 ply search depth when reusing >>hash entries from a previous search, also on unexpected moves by the opponent > >That's a nice feature. :) > >Sargon
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.