CCRL Discussion Board

I tried running the code from your website and after changing SIZE and alignment: #define SIZE (1<<6*MEN) for(tb=p;(int)tb&4095;++tb); /* align with cash line */ it didn't cause a segmentation fault but gave the result: 79fbb010 79fbc000 KBB_KN 0. 0 0 0 0 0.000 0.000 0.000 1. 0 0 0 0 0.000 0.000...

I have been watching youtube videos and found Disk-Based Parallel Computation, Rubik's Cube, and Checkpointing is about computations somewhat similar to what we want to do.

So it seems that leapfrog degenerates into no-communicating parallel lines, (N-1->N+1; N->N+2), (N->N+2; N+1->N+3), ..., the numbers giving the dtz in ply of lost btm positions. The second cycle of a leap-frog pass would thus not built on the lost positions just generated by its first cycle. Only i...

I am trying to get going writing a baseline generator for DTZ. (It need not be efficient in any way, as long as it gives the correct result in reasonable time for small tablebases, i.e. 4-man in 24h is enough.) Correct result is the priority. I have also looked at order of computation. When computin...

For writing I copy from /dev/zero to a dedicated partition. (/dev/null is always empty, no matter how much you write to it.) It is safe to test reading from the windows partition, writing is of course not safe. Read and write speed ought to be about the same speed when not involving a file system mu...

I use GNU/Linux. To test device /dev/hda at beginning I type: dd if=/dev/hda of=/dev/null bs=80M count=1K To test a 1TB disk near end I type: dd if=/dev/hda of=/dev/null bs=80M count=1K skip=11k (skips 11k*80MB=880GB so test is at end of disk) With a sata drive the device would likely be /dev/sda or...

I bought a new computer last month with Athlon 64 Phenom x4 9950 and nominally 8400 MB/s memory. I measured the memory speed with read+simple computation+write and got about 8000 MB/s (counting read+write as 2 operations). Before I decided on it I looked at the memory benchmarks at Tom's Hardware an...

Another possibility is to examine all moves and positions (except perhaps conversion moves which only need to be examined in the beginning). Then you only need to do a one ply forward search/position to increase depth 1 ply and you never run into the problem of positions that are not updated. Luckil...

What is DTZ? DTC = Distance To Conversion I know that. ... Thanks so far, I thought I was almost ready with my program but it seems I have to work on the speed :P DTZ is distance to zeroing the game counter with respect to the 50 move rule. Similar to DTC, only all pawn moves reset the counter too....

Both h.g.muller and I have discussed how to generate DTZ tablebases efficiently. For pawnless 7-man endings, my partitioning scheme works well as long as there are at least 2-3 black pieces and his works well as long as there are at least 3-4 white pieces. Things our schemes have in common are that ...

Hmmm... I can't think of anything else than a sort of RunLength pre-compression for that idea of yours. Is that the way you should handle it, Koistinen? No, not run-length, it is removal of illegal positions: int encode(inbuf, outbuf, firstindex, count) { index = firstindex outcount = 0 for i=1..co...

One way to avoid storing illegal positions is by storing positions in blocks and then filtering out illegal positions before compression.
When decompressing a block, use the same filtering function to put them back.

Would you publish your code for others to use? (under some free software licence)

notnale wrote:so basically I have to learn to program and make my own?

Looks like doing KRK DTZ50 with free software would advance the state of the art.

Me too, in particular working generators that are free software.

If it gets too complicated, try simplifying, say by using simple indexing and no mirroring and suchlike. Let every piece have 64 possible positions, kings too. That is what I'll do next. ps. "totally wrong" is not a good explanation of what is wrong with something. Try to give some example...

In what way is the index wrong? Does it not work as a minimal perfect hash? White king: KW Black king: KB ix: 0..2016 (inside of envelope calculation as the outside was already overwritten :) To compute an index from all pieces you might consider KB+64*(R+64*(ix+2017*KW)). This is not minimally perf...

New pieces, new rules, (possibly flawed) chessic logic, intuition ... I'm afraid this appears to me to be so baroque that I will not understand what the output means or what it's integrity is. My understanding is that it is the cpu time and disk use that is the important output of the benchmark. At...

I noticed that by not storing illegal positions and exploiting symmetry you can save quite a bit of i/o as well as some disk space. $ gcc -O2 -Wall countpos.c $ ./a.out uint64_t poscount[8]= { 462, 28056, 1707888, 102455640, 6044812200, 350598895920, 19984136644080 }; I think it is best to implement...

Thank you for the web-page titles. We have different goals but in part they overlap. My interest is mainly in how to efficiently compute, with computing all 7-man endgames for standard chess a part of that. Also I like affordable computing, so as long as the hardware is not too expensive it might in...

I wrote a small program to compute how much the symmetry of multiple pieces of same type can reduce size of database for the 7-man endgames and got the following results: $ gcc -O2 -Wall count.c $ ./a.out |grep -v [Pp]|sort|awk '{print $1}'|uniq -c 4 120, 44 12, 56 1, 44 24, 276 2, 156 4, 156 6, $ ....

Now I think I understand your K-streaming better. Say the white pieces are KQR, then you might assign moves within a-b, c-d, e-f, g-h to one leap and the other king moves to the other. Then load KabQxR, (x being the fixed position of the queen and the actual position the one after mirrorings). Do th...

Nice explanation of K-slice Streaming. Now I at least understand it in part. The way I'd do it might be different: Have one leap include pawn moves that cause a diagonal flip and the other moves that cause a horisontal flip. Vertical flips would only ever be caused by moves to/from d4 so I'd simply ...

With 16GB of ram it would be fairly easy to make your leapfrog algorithm work well for all 7-man endgames, right? I would guess that within a year, getting a PC with 16GB of ram will be fairly common and not all that expensive. Today 8GB of ram is not too expensive and there are plenty of motherboar...

How about this format: Decide on an evaluation function for entries and a competition period. Let people send in entries when they like and publish it and an evaluation as soon as possible. When competition period is ended, divide prize in proportion to how far each entry pushed the state of the art...

Perform 1 ply and probe the table again. By definition, the position after 1 ply is d-1 plies from a zeroing move. The value returned by the table for that position is either k or k-1. This isn't true if the ply was a zeroing move True, but if you can convert right away to a winning position (or if...