Eelco
Posts : Join date : Ed, small, single line, change for Rebel: I have not really tested the change in plain Rebel but I think the line is bad. It seems to be some micro optimization; if you reach a mate score jump out of aspiration search because Mate will win (probably true) especially if the game is adjudicated, and spending more time might be flag.
But it seems to break any Mate search, strange that nobody seems to have noted ever.
It's line 440-441 in search.cpp
Code:
Maybe this alone does not make much difference but other changes I have made in search.cpp are not tested much yet.
Here some Rebel 15.1 MOD results with Norm Pollock's double Mate in six position, the White to move version. Rebel 14.1 MOD had no problems, I don't think Rebel 14.1 without modifications has either.
Code:
Likewise, I just checked line 555 for SMP (threads 2,3 4 etc). I have not yet thrown out, have not tested any multithreaded version on Norm Pollock's position yet, but this is probably just as bad:
Code: if (value_is_mate(SearchBest[ThreadId]->value)) break;
Ghppn likes this post
Like 1
Admin Admin
Posts : Join date : Location : PVS search is the best for a chess engine, in use by almost everybody. However Aspiration search is a little bit faster in the early plies and thereafter engines switch to PVS. We are talking about micro seconds.
Ghppn likes this post
Like 1
Eelco
Posts : Join date : Hi Ed, sorry, I will try to improve my code. The removal of the two lines on their own will be practically invisible because that code only comes into effect if there is a Mate to be found. But also with the present code, if you remove this line (two lines) in Toga it is not possible to find any real Mate distance unless you do umpteen iterations with very narrow windows (That is what you see in Norm Pollock's example, one of the tries is with Vanilla Rebel 15.1) To actually find the Mate with accuracy you have to change the width of the search window faster (code that comes just after the two offending lines). That is pretty straight forward I just wanted to give a small hint. Sorry But it is not certain that the scaling will be (very) different if you are going to do that always, change the width of the aspiration windows I mean (it is not really Aspiration search but that is what I call it when you use such small search windows, they are almost null window searches at first)
Ghppn likes this post
Like 1
Sponsored content
Home 
