Author: Landon Rabern
Date: 23:39:12 01/24/01
Go up one level in this thread
On January 25, 2001 at 01:33:23, Larry Griffiths wrote:
>On January 24, 2001 at 23:08:47, Landon Rabern wrote:
>
>>On January 24, 2001 at 22:49:08, Landon Rabern wrote:
>>
>>>On January 24, 2001 at 03:43:03, Severi Salminen wrote:
>>>
>>>>>How much of a speed increase did you get from using MMX? I got only about a 35%
>>>>>increase.
>>>>
>>>>"Only"??? Are you mad :) 35% speedup from using different set of instructions is
>>>>_a lot_!! Are you now fast or were you slow back then?
>>>>
>>>>Severi
>>>
>>>I said only because I was under the impression that Larry was getting a lot
>>>faster, but I guess not.
>>>
>>>I was always pretty fast and now faster. Around 400,000 nps on a PIII-500 with
>>>my whole search and eval and everything on. My eval is not that complicated
>>>right now and could use some work and my move ordering can be improved some. I
>>>am at school now and do not have much time except to work on a Neural Network
>>>for my eval which I convinced them to give me credit for.
>>>
>>>The version of my program that is released 5.1 I think does not have the MMX
>>>stuff because I did not want to deal with making it compatible for above and
>>>below PII.
>>>
>>>Regards,
>>>
>>>Landon
>>
>>If anyone cares, here is the MMX code I wrote for it. I would not doubt it if
>>tehre were some good ways to optimize it.
>>
>>Regards,
>>
>>Landon
>>
>>#define gBlackBishopQueenCaps(fs) \
>>{ \
>> __asm mov eax,board\
>> __asm movq mm0,[eax].A1H8board\
>> __asm mov ebx,fs\
>> __asm shl ebx,2\
>> __asm movd mm1,[A1H8shift+ebx]\
>> __asm psrlq mm0,mm1\
>> __asm pand mm0,[A1H8rotMask+ebx]\
>> __asm movd mm2,[A8H1shift+ebx]\
>> __asm movq mm3,[eax].A8H1board\
>> __asm psrlq mm3,mm2\
>> __asm pand mm3,[A8H1rotMask+ebx]\
>> __asm shl ebx,9\
>> __asm movd ecx,mm0\
>> __asm shl ecx,3\
>> __asm movq mm4,[attacksA1H8+ebx+ecx]\
>> __asm movd ecx,mm3\
>> __asm shl ecx,3\
>> __asm movq mm5,[attacksA8H1+ebx+ecx]\
>> __asm por mm4,mm5\
>> __asm pand mm4,whitePieces\
>> __asm movq toMap,mm4\
>>}
>
>Looks simular to my code. I have GenCaps, GenMoves and GenCapsMoves.
>The GenCapsMoves produces both a capture and a move bitboard. I save the move
>bitboard in my derived piece object so that I can just grab the move bitboard
>when GenMoves is called after a GenCapture.
>
>Here is my Queen GenCapsMoves...
>
>#define defbbGenSplitQueenCapturesMoves(fsq,bbrankocpieces)\
> {\
> _ESI = (unsigned long)fsq;\
> asm lea EDI,bbArray1;\
> asm movzx EAX,bbRankContentsOffset[ESI];\
> asm movzx EBX,bbFileContentsOffset[ESI];\
> asm movzx ECX,bbLDiagContentsOffset[ESI];\
> asm movzx EDX,bbRDiagContentsOffset[ESI];\
> asm movzx EAX,byte ptr [EDI+EAX+bbRankOccupied*8];\
> asm movzx EBX,byte ptr [EDI+EBX+bbFileOccupied*8];\
> asm movzx ECX,byte ptr [EDI+ECX+bbLDiagOccupied*8];\
> asm movzx EDX,byte ptr [EDI+EDX+bbRDiagOccupied*8];\
> asm shl ESI,0x0b;\
> asm movq mm4,[EDI+bbRankOccupied*8];\
> asm movq mm0,[bbRankAdjacentEmptySquaresPieces+ESI+EAX*8];\
> asm por mm0,[bbFileAdjacentEmptySquaresPieces+ESI+EBX*8];\
> asm por mm0,[bbLDiagAdjacentEmptySquaresPieces+ESI+ECX*8];\
> asm por mm0,[bbRDiagAdjacentEmptySquaresPieces+ESI+EDX*8];\
> asm pandn mm4,mm0;\
> asm pand mm0,[EDI+bbrankocpieces*8];\
> asm movq bbMoves,mm4;\
> asm movq bbCaptures,mm0;\
> }
What if you get a cutoff and do not need the moves. I only gen caps then only
gen moves after I have tried all the caps, so that if I get a cutoff, the work
of generating moves is saved.
I also use the MMX registers for check detection which seems to speed that up a
little.
Have you tried coding the whole gereation routine in assembler. Like the loop
to take moves out of the bitboard, etc. if so, any improvement?
Regards,
Landon
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.