Syzygy / egbb discussion

[Daniel Shawul]

Mike I am sorry your thread is hijacked and I definately have nothing against you. Therse guys are spreading lies,lies,lise everywhere. I think now that yet another false rumor about 'multi-threading ' problems is shown to be yet another butchering by Houdart, you should let it finish which won't take long.

For those keeping score

a) Houdini totally butchering egbb implemntation and blaming me for this problmes. Proven lie.
http://talkchess.com/forum/viewtopic.php?t=49684

b) Houdini badly implementing SMP for EGTB probing and blamin Scorpio,Nalimov,and Gaviota TBs here and everywhere. Yet again proven atleast for scorpio. I will not be surprized if Nalimov,Gaviota implementation had codes that needs modification for >8 processors, better check your code Miguel!

I should also note that i never butted in these discussion if scorpio bitbases are not directly or indirectly referenced , as was the case here. "I have never seen such elo with other TBs" warants a response from authors of other TBs IMO. Please understand that these guys are spreading lies about scorpio egbbs all over the place (learned about the bashing in Rybka forum years ago just recently). I wish I could say to you that I will ignore it but i had enough already.

[ernest]

Come on Daniel, everybody here considers that your bitbases are great and this thread can really bring a lot of progress in understanding how to use them best.
And I am sure that Robert is completely open to making them work in an efficient way. If difficulties are found, exploring them in depth is also progress!

[Houdini]

Come on Daniel, everybody here considers that your bitbases are great and this thread can really bring a lot of progress in understanding how to use them best.
And I am sure that Robert is completely open to making them work in an efficient way. If difficulties are found, exploring them in depth is also progress!

The egbb are actually working very well in Houdini. Just use them in analysis - not for engine matches - and you'll do fine!

Note that I've never said otherwise - I've already challenged Daniel to find a single quote in which I say negative things about egbb, so far no reply.

[Daniel Shawul]

Come on Ernest Bonnem, you think you are smart implying that scorpio bitbases are hard to use?? Please let me educate you a bit, an engine author needs like 100 lines of code to use them! Compare that to others TBs where they are statically linked with code all over the place. I am 100% sure they are the easiest to implement along with knightdreamer's bitbase compared to Nalimov, Gaviota, Syzygy etc.. The dll way such a relief for programmers but I don't blame you if you don't understand that with negligible cost.

Houdart is such a butcher of code and I am not sure what he will pull next to amuse us. I never owned Houdini before so I did not know how he was using them. That is twice already he got owned, so I have him in my pocket besides my credit/debit card

You should know that my job is done once I provide the WDL data but I go a long way than that to help authors.

[Daniel Shawul]

Egbbs are never compiled for more than 8 processors . Post the egbbdll.cpp code you used so that I could show you.

You said this

To understand what I'm saying, I invite you to run some 16-core analysis on an endgame position, and you'll see that Syzygy bases reduce the overhead by an order of magnitude compared to the three other table base types.

They don't work with 8 cores well unless I changed the probing code. Scorpio played without EGBBs in TCEC because Martin used 16 cores. So how come YOU of all people fix my broken code for me and play with them with 32 cores??? It bogles the mind ....

[Houdini]

Daniel, I don't have time for your nonsense.
As I told you before, Houdini 4 will use Syzygy bases which is a far superior solution in many respects.
egbb are good, but Syzygy are excellent. Deal with it...

[Daniel Shawul]

You are a liar, period. Scorpio egbbs never work with more than 8 processor , except in Houdini the magician's land.

[Houdini]

You are a liar, period. Scorpio egbbs never work with more than 8 processor , except in Houdini the magician's land.

For compassion's sake, let me indulge you this final time.
Here's the start of my egbb.cpp:

Code: Select all

#include "common.h"
#include "egbbdll.h"

static const int WIN_SCORE = 9900;

enum {
	DECOMP_IN_RAM,DECOMP_IN_DISK,COMP_IN_RAM,COMP_IN_DISK
};
static const int VALUE[4] = {
	DRAW, WIN, LOSS, PREDICTED
};

#define is_in_disk(x)    ((x) & 1)
#define is_comp(x)       ((x) & 2)

static SEARCHER searchers[MAX_THREADS];

Note the last line.
You see, I've really managed to replace your static "8" by my variable MAX_THREADS (which is 6 in Houdini Standard and 32 in Houdini Pro).
Yes we can!

[Daniel Shawul]

How about this one? You see there is a hard-coded 8 ...

Code: Select all

PSEARCHER psearcher;
	l_lock(searcher_lock);
	for(int i = 0;i < 8;i++) {
		if(!searchers[i].used) {
			psearcher = &searchers[i];
			psearcher->used = 1;
			break;
		}
	}
	l_unlock(searcher_lock);

[Houdini]

You want to see all the places where the constant "8" is replaced by the constant "MAX_THREADS"?

I don't think any further discussion will be useful here.
If you want to see Houdini's egbb code, send an e-mail to houdini@cruxis.be and I'll send you the files. You may find them interesting...

Now back on topic about the Syzygy bases!