Author: Christophe Theron
Date: 20:54:10 03/16/01
Go up one level in this thread
On March 16, 2001 at 19:52:30, Tony Werten wrote:
>On March 15, 2001 at 19:30:13, Olaf Jenkner wrote:
>
>>>>A look at the generated code shows some horrible things:
>>>>
>>>>Example: if(a[i]=x)or(a[i]=y)or(a[i]=z)then ...
>>>> Delphi does not know that it can load the value
>>>> of a[i] in a register only on once.
>>>>
>>>>I think many things could be done better without a great optimization.
>>>
>>>Or change the code to
>>>
>>>tmp:=a[i];
>>>if (tmp=x) or (tmp=y) or (tmp=z) then ...
>>>
>>>BTW the behavour of the array a[] is different depending if it's a global,local
>>>or dynamic array. So is its speed. Global dynamic is fastest if the size of the
>>>elements is a multiple of 32 bit.
>>>
>>>That's what I mean with you have to know the compiler.
>>>
>>>cheers,
>>>Tony
>>>
>>>>
>>>>OJe
>>
>>Why do I have to program around compiler weaknesses?
>>Perhaps the variable tmp is not hold in a register.
>>What I mean: In simple cases, as shown above, I wish
>>a better behaviour of compilers. I know that there are
>>compilers which are extremely optimizing to minimize
>>RAM accesses or to shrink the code or to minimize the
>>number of clock ticks. All of them I can't see when I
>>look at the generated code of Delphi. It's a pity because
>>I don't know much about C++ and have no experience in it.
>
>I'll have to agree with you that it would be nice if the compiler would see this
>kind of stuff. But remember that it's made to work in general. Sometimes it's
>about speed sometimes about size. ( I would like that switch in Delphi )
>
>Because you are the one who knows what you want, you have to send the compiler
>in the right direction. Learn some assembly and you'll find you can improve the
>compiler even more because you know what it should do.
>
>But these things are not the most important in chessprogramming, the algoritmes
>you use are. You can try a lot of things in C or Pascal, change them, and when
>you finally decide to stick with it, that's the time you can start to optimize.
>
>Oh, and BTW, if you think your program is too slow, just say you go for
>knowledge rather than speed.
>
>A countryman programmer of me mulptiplied his nps by 7 before showing it on
>screen just to stop people from saying it was very slow. ( he also used the
>"never accept material from a better program because you're probably not winning
>anything, you just think you are" algoritm )
Did it work? :)
And what about positional gifts? :)
Christophe
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.