Search found 17076 matches

by bob
Wed Sep 05, 2007 3:29 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: How do I test the quality of the Zobrist keys?
Replies: 41
Views: 18046

Re: How do I test the quality of the Zobrist keys?

Well, to investigate if Hamming distance has any bearing whatsoever on this problem, I tested independence of Zobrist keys through the following simple program: #define SIZE (1<<27) long long int Zob[12*64]; int lock[SIZE]; int cnt; test(int d, int n, long long int k) { int i, ...
by bob
Tue Sep 04, 2007 7:56 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: SMP thread goes here
Replies: 39
Views: 16890

Re: SMP thread goes here

True, but you could put it in the same cache line with the other (private) variables that are used every node, so you would be sure it always is an L1 hit. (Or if it would be a miss due to an interrupt / task switch, you would have had to process the miss anyway on behalf of the other variables). F...
by bob
Tue Sep 04, 2007 12:52 am
Forum: Computer Chess Club: General Topics
Topic: Win or draw?
Replies: 5
Views: 2219

Re: Win or draw?

I don't understand what all of that is. Two engines playing against each other? or an engine playing against itself? Neither is worth a flip at deciding whether the original position is a win or draw (or worse)...
by bob
Mon Sep 03, 2007 9:49 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: SMP thread goes here
Replies: 39
Views: 16890

Re: SMP thread goes here

But is it really necessary to poll at every node? If you have only a thousand or so aborts per search, and each CPU does 1M nps, checking once every 1000 nodes makes you discover aborts on the average 0.5 ms late, so that a CPU that participated in all 1000 aborts still only uses 0.5 sec (out of 30...
by bob
Mon Sep 03, 2007 6:17 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: SMP thread goes here
Replies: 39
Views: 16890

Re: SMP thread goes here

Well, so it is all approximately equally trivial, and if there are performance differences they are on the order of 0.1% or so. I don't see that as a reason to abandon the first ideas that come into my mind, and copy others in stead. For that, my own ideas should be wildly inefficient (say, result ...
by bob
Mon Sep 03, 2007 6:04 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: How do I test the quality of the Zobrist keys?
Replies: 41
Views: 18046

Re: How do I test the quality of the Zobrist keys?

One simple test is to measure the hamming distance between any two Zobrist numbers (hamming distance is the number of bits that are different between two binary values). The minimum hamming distance between pairs of numbers is an interesting value. If you have several pairs with very small hamming ...
by bob
Mon Sep 03, 2007 2:23 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: How do I test the quality of the Zobrist keys?
Replies: 41
Views: 18046

Re: How do I test the quality of the Zobrist keys?

One simple test is to measure the hamming distance between any two Zobrist numbers (hamming distance is the number of bits that are different between two binary values). The minimum hamming distance between pairs of numbers is an interesting value. If you have several pairs with very small hamming d...
by bob
Mon Sep 03, 2007 12:53 am
Forum: Computer Chess Club: General Topics
Topic: Crafty.
Replies: 7
Views: 2680

Re: Crafty.

Has there been a new crafty lately. What is the latest version. Thanks, Gerold. We are still working on the final 21.6 to be released. We are running thousands of games to test, and adjust. I hope it won't be much longer, but I have said that previously. Some changes work well, some flop. A new ver...
by bob
Mon Sep 03, 2007 12:47 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: CS Theory and MS tuning
Replies: 3
Views: 2156

Re: CS Theory and MS tuning

Ok, saw something on an overclocking forum. It was a list of tweaks to speed up MS OS. One of them made me say huh? Basically, it claimed that the OS initially doesn't know to access the CPU's L2 cache and that you should put the size of the L2 cache (in megabytes) into a registry location. From my...
by bob
Mon Sep 03, 2007 12:46 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: SMP thread goes here
Replies: 39
Views: 16890

Re: SMP thread goes here

The same story all over again. So operating systems are crappy... It doesn't seem so difficult to program decent signals. Oh well, let's not go into that... Yes, your polling is ultra-simple code. But of course you need lots of code lines to set up the linked lists of split blocks. In my scheme tha...