Speaking of the hash table (branch)

[diep]

I've already extensively tested this with a lockless hashtable. With a CRC you can easily measure this way faster than completely messing up the bus locking nonstop.

In Diep there are more collissions than write errors with a 64 bits hashkey!

[...]

Note that your normal crafty is using a lockless hashtable! Don't bring in other idiotic factors you're not using in practice! Changing the discussion here just to 'save face' is what they do in Asia Bob, over here you're busted.

Long before you wrote/posted about Tim Mann inventing the XOR manner of doing things lockless, others were already using a 8 bits CRC...

Vincent, it seems you accept that the problem discussed here is solved with lockless hashing but persists without. Right?

Sven

Moment i'm parsing your english here.

Can you say it in simpler english?

Realize Bob has assembler inline codes which just work for 1 architecture. Sometimes some dudes change it for him, but then it has a problem at another.

That's the horror of bitboards and Bob's approach you know. Everything is inline assembler.

That's why they kicked him out of specint i guess.

Don't let Bob talk his way out here. He's just posting some nonsense then trying to expalin the nonsense, whereas i measured for months. Shared my research there publicly already for years. If you scroll back i posted all this already years ago. Yet no one CARED back then.

Right now it's hot to start such discussions just because of me.

If you would want to bugfix everything in crafty, you have a fulltime job of course so i'm ignoring bob's output.

I refer to his publication about SMP around ICGA june 1997 which was complete fraud there as well. This guy just makes up numbers at home.

He's not qualified to test *anything*, as he'll just shout something and then later on will try to 'explain' his initial idiocies.

Over here in Netherlands he would've been fired as a professor long long time ago for making up numbers and officially publishing those in a refereed journal.

But this is typically what happens in computerchess and shows the 'average' researchers quailty here!

[MODERATION]
There is a chance that bringing old issues like this may lead to an escalation and become a an off-topic fight. Please, to the contributors of this thread, refrain from insults and do not drive this thread to become personal. We will avoid this. The leash is very short in the technical sub-forum. Please, keep it civil, technical, and on-topic.

Adam, Julien, Miguel.

[Sven]

I've already extensively tested this with a lockless hashtable. With a CRC you can easily measure this way faster than completely messing up the bus locking nonstop.

In Diep there are more collissions than write errors with a 64 bits hashkey!

[...]

Note that your normal crafty is using a lockless hashtable! Don't bring in other idiotic factors you're not using in practice! Changing the discussion here just to 'save face' is what they do in Asia Bob, over here you're busted.

Long before you wrote/posted about Tim Mann inventing the XOR manner of doing things lockless, others were already using a 8 bits CRC...

Vincent, it seems you accept that the problem discussed here is solved with lockless hashing but persists without. Right?

Sven

Moment i'm parsing your english here.

Can you say it in simpler english?

Realize Bob has assembler inline codes which just work for 1 architecture. Sometimes some dudes change it for him, but then it has a problem at another.

That's the horror of bitboards and Bob's approach you know. Everything is inline assembler.

That's why they kicked him out of specint i guess.

Don't let Bob talk his way out here. He's just posting some nonsense then trying to expalin the nonsense, whereas i measured for months. Shared my research there publicly already for years. If you scroll back i posted all this already years ago. Yet no one CARED back then.

Right now it's hot to start such discussions just because of me.

If you would want to bugfix everything in crafty, you have a fulltime job of course so i'm ignoring bob's output.

I refer to his publication about SMP around ICGA june 1997 which was complete fraud there as well. This guy just makes up numbers at home.

He's not qualified to test *anything*, as he'll just shout something and then later on will try to 'explain' his initial idiocies.

Over here in Netherlands he would've been fired as a professor long long time ago for making up numbers and officially publishing those in a refereed journal.

But this is typically what happens in computerchess and shows the 'average' researchers quailty here!

Please address your complaints about Bob to the right person, not to me.

I certainly disagree with Bob in some areas as you might know (e.g. the Rybka/Fruit case) and some discussions with Bob were certainly "difficult" somehow but I have no reason to raise any doubts on his qualification in computer science, on his programming and testing abilities, on the quality of his publications, or on anything else of that kind. Certainly 20 or 30 years old software like Crafty is not "optimal" from my viewpoint of a software engineer etc. but at least it seems to work very well compared to many other chess programs in the world, and especially its SMP implementation is definitely of high quality.

As to my english: no, I can't say it in a simpler english, sorry.

Sven