Author: Uri Blass
Date: 08:43:17 02/20/03
Go up one level in this thread
On February 20, 2003 at 11:06:28, Filip Tvrzsky wrote: >On February 20, 2003 at 09:59:13, Uri Blass wrote: > >> >>I see that it does not help me. >> > >It's a pity, because this approach was really usefull for me, but you know, each >program is a different case and what works in one doesn't work in other. It is >hard to say why it doesn't help you. There are so many dependencies ... >Probably this approach claims some deeper changes of your implementation in >order to gain from it. The main idea here is IMO to restrict the memory access >... > >>I had to replace in hundreds of places in my code and after all the job I have 2 >>problems from the opening position: >>1)The code is slightly slower in the first ply(1% slower) >>2)The bigger problem:after more than 40 seconds I start to get different number >>of nodes(maybe I did a mistake in one of my replaces). >> >>Uri > >Of course such a change in implementation basics is very dangerous and work >intensive operation. I hope you have some handy testing and debugging code and a >backup of your original code too ... > >Filip I have backup of my original code so I guess that I will use it. I guess that it did not help me because of the fact that I use bits in a lot of places in my code for example I have some if (bits(m)&) in my evaluation code in order to evaluate change in the evaluation from moves. Maybe the problem is that I have now if ((bits(gen_dat[i].m)&33) in my qsearch I check for every legal move if it is a capture in order to decide if to extend it. I guess that all these & did my code slower. bits(gen_dat[i].m) was already & by definition(I could save one & by having gen_dat[i].m&(33<<24) but in that case I do the code less clear. I also use in a lot of cases expressions like to(gen_dat[i].m) Maybe the previous code gen_dat[i].m.b.to was faster. Uri
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.