Re: [ros-dev] [ros-diffs] [greatlrd] 23826: Wrote RtlUshotByteSwap RtlUlongByteSwap and RtlUlonglongByteSwap to asm code. but we need a C api for header to linking it right. Put the asm version to i386

[mrnobo1024]

--- Alex Ionescu <ionucu(a)videotron.ca> wrote:

>   mrnobo1024 wrote:
> >   The new implementations for RtlFillMemory,
> > RtlZeroMemory, and 
>  RtlMoveMemory
> >   have a bug where they don't write the last
> > 1-3 bytes if the length isn't 
>  a
> >   multiple of the word size. They have an "or
> > ecx, ecx" which needs to be 
>  "or
> >   ecx, edx". 
>  
>  Noted, thanks! Surprised it still booted! 
Looking at them again I noticed a couple other problems. At the end of
RtlCompareMemory it subtracts the length (esp+20) instead of the buffer start
(esp+12), and RtlCompareMemoryUlong subtracts from esi when it's using edi.

I also noticed some of the functions clear the direction flag, and some
don't.  This would be either a bug or just unnecessary code. I don't know if
the Win32 calling conventions require the flag to be clear before calling a
function or not.

> >   Also the new
> > RtlUlonglongByteSwap is sub-optimal, it swaps eax and
> >  edx when they could have just been read in reverse order like the other
> >  version did. 
>  
>  Not true. BSWAP will pair on the pipe and be executed much faster. 
I meant that instead of mov edx,[esp+8] mov eax,[esp+4] it could be mov
edx,[esp+4] mov eax,[esp+8]. I don't think that would affect instruction pairing.

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com