From: Robert Lee <rhlee@nwu.edu>
Subject: Re: Assember (in-line) help
Date: 21 Nov 1997 00:00:00 GMT
Message-ID: <347586A1.F9F6F4F9@nwu.edu>
References: <34746066.8462150@news.cjnetworks.com> <652f66$bci@cocoa.brown.edu> <3474df1b.2228469@news.cjnetworks.com> <653rup$i00@cocoa.brown.edu>
Organization: Northwestern University, Evanston, IL, US
Newsgroups: comp.lang.pascal.delphi.misc




John_Mertus wrote:

> In article <3474df1b.2228469@news.cjnetworks.com>,
> nospam.aburton@mizehouser.com says...
> >
> >Thank you to everyone.  I had to temporarily rely on some math routine
> >to swap things around.  As you can guess that is a lot slower.
> >
>   Damn, I realized my second routine had the same problem as xchg,a
> faster, better  Pentium Pro, II routine is.
>
>  Procedure  Swap32c (Var LongVar : LongInt) ; Assembler ;
>    Asm
>       mov esi,longvar
>
>       xor   eax,eax
>       mov   ax,[esi]
>       xor   edx,edx
>       mov   dx,[esi+2]
>
>       mov   [esi],dh
>       mov   [esi+1],dl
>       mov   [esi+2],ah
>       mov   [esi+3],al
>
>    End {Swap32} ;

  Actually there are clock penalties for addressing the same dWord in the
same clock cycle and writing to the same register in same cycle so the
above takes 8 clocks.  The best anwser is still bswap. Its one cycle on a
pentium.  By the way LongVar is loaded into EAX and so there's no reason to
move it to ESI.

Procedure  Swap32d (Var LongVar : LongInt) ; Assembler ;
   Asm
        bswap LongVar // LongVar=eax
   End {Swap32} ;


Bob Lee