Message-ID: <3DA58143.1020702@riosoftware.com>
Date: Thu, 10 Oct 2002 09:31:47 -0400
From: Colin Sarsfield <colin@riosoftware.com>
Organization: Rio Software, LLC
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.0.1) Gecko/20020826
X-Accept-Language: en-us, en
MIME-Version: 1.0
Newsgroups: borland.public.delphi.language.basm
Subject: Re: Looking for a really fast hash algorithm
References: <3d8f7618@newsgroups.borland.com> <3d8fce54$1@newsgroups.borland.com> <3d902010@newsgroups.borland.com> <3d91976b$2@newsgroups.borland.com> <3D92DF1D.794F9EA9@optimalcode.com> <3d95390a$1@newsgroups.borland.com> <3D95B23A.DCF87058@optimalcode.com> <3d95b826$1@newsgroups.borland.com>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit
NNTP-Posting-Host: 204.111.85.64
X-Trace: newsgroups.borland.com 1034256623 204.111.85.64 (10 Oct 2002 06:30:23 -0700)
Lines: 40
Path: newsgroups.borland.com!not-for-mail
Xref: newsgroups.borland.com borland.public.delphi.language.basm:532

Will DeWitt wrote:
> Almost.  =)  Re-ran the same tests as before and the gap was narrowed down
> to less than 1 percent (in one test case it eeked out a lead tho). I made
> one or two minor modifications to the code and now it's faster in all
> situations (short or long).  The first change was right at the top-- I
> replaced the lea ecx, [eax-4] / mov edx, [ecx] combo with a simple mov edx,
> [eax-4].  I think changed the mov cl,[eax+edx] after @Loop1 to movzx ecx,
> [eax+edx].  In this form it beats my original HashASM.  =)
> 
> function HashRL3New(const S: string): integer;

Here's a function that's about 28% faster :)  (on my Athlon XP 1400+)

function HashN(const S: String): Integer;
var
   I: Integer;
   PC1, PC2: PChar;
   P1, P2: PInteger;
begin
   Result := 0;
   I := Length(S);
   P1 := PInteger(@S[1]);
   P2 := PInteger(@S[Succ(I and $fffffffc)]);
   while P1 <> P2 do begin
     Result := Result * 988053892081
         + (P1^ and 255) * 991026973
         + ((P1^ shr 8) and 255) * 994009
         + ((P1^ shr 16) and 255) * 997
         + (P1^ shr 24);
     Inc(P1);
   end;
   PC1 := PChar(P1);
   PC2 := @S[I];
   Inc(PC2);
   while PC1 <> PC2 do begin
     Result := Result * 997 + Ord(PC1^);
     Inc(PC1);
   end;
end;