From: "Steve Patton" <leaphe AT pemail DOT net>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: does anyone have anyideas on how to optimize this code
Date: 28 Nov 1997 18:12:26 GMT
Organization: AT&T WorldNet Services
Lines: 85
Message-ID: <65n1ia$q6h@bgtnsc03.worldnet.att.net>
References: <346C8822 DOT D3725C8E AT earthlink DOT net>
NNTP-Posting-Host: 12.67.2.170
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk



STEVEN S. FALLS <ELN/broadview AT earthlink DOT net> wrote in article
<346C8822 DOT D3725C8E AT earthlink DOT net>...
> Hi, my name is Ardy and I have a pixelplotting ruteen that needs to be
> optmized. This is what I got, please help!
>             thanks,
>                     -Ardy
> 
> This was writen in NASM:
> 
> BITS 32
> GLOBAL _SetPix__Fiic
> GLOBAL _SetPixRam__Fiic
> GLOBAL _GetPix__Fii
> GLOBAL _GetPixRam__Fii
> GLOBAL _ClearPg__Fv
> GLOBAL _ClearPgRam__Fv
> GLOBAL _CopyPg__Fv
> GLOBAL _Line__Fiiiic
> EXTERN __go32_info_block
> EXTERN _YTbl
> EXTERN _RamPg
> 
> SECTION .text
> _SetPix__Fiic:
>         push ebp
>         mov edx,[_YTbl]
>         mov ebx,[esp+12]
>         mov ebx,[edx+ebx*4]
>         add ebx,[esp+8]
>         mov al,[esp+16]
>         mov fs,[__go32_info_block+26]
>         add ebx,0A0000h
>         mov [fs:ebx],al
>         pop ebp
>         ret
> _SetPixRam__Fiic:
> 

Ok, that was original, here's one with suggestions.

BITS 32
GLOBAL _SetPix__Fiic
GLOBAL _SetPixRam__Fiic
GLOBAL _GetPix__Fii
GLOBAL _GetPixRam__Fii
GLOBAL _ClearPg__Fv
GLOBAL _ClearPgRam__Fv
GLOBAL _CopyPg__Fv
GLOBAL _Line__Fiiiic
EXTERN __go32_info_block
EXTERN _YTbl
EXTERN _RamPg

SECTION .text
_SetPix__Fiic:
        push ebp
        mov edx,[_YTbl]
        mov ebx,[esp+12]
        mov ebx,[edx+ebx*4]  // Look below
        add ebx,[esp+8]
        mov al,[esp+16]
        mov fs,[__go32_info_block+26]
        add ebx,0A0000h
        mov [fs:ebx],al
        pop ebp
        ret
_SetPixRam__Fiic:


I don't know if GCC will take care of this, but it appears that you're
multiplying ebx by 4.  You can simply do this

shl $2, %ebx        // what should be AT&T format

I don't know if that's the correct syntax (I'm used to Intel format), so if
you get an error, just try

shl %ebx, $2      // format used by Intel

That will do it in 1 clock cycel on a 486, as opposed to 10-50(about)
clocks on a 486.  You might try that and see if that helps you at all.

-Steve