[XviD-devel] inlined SAD

[Michael Militzer]

Hi,

oops, a quick glance at the code I just sent you showed that I had modified
it (unrolled, added prefetch commands etc.). The optimizations weren't
really successful therefore I didn't commit them, but the code also didn't
become slower - it simply made no difference. However the code should still
be sufficient to test if __inline makes a difference or not...

Michael

----- Original Message -----
From: "Michael Militzer" <michael@xvid.org>
To: <xvid-devel@xvid.org>
Sent: Tuesday, September 03, 2002 4:31 PM
Subject: Re: [XviD-devel] inlined SAD


> Hi,
>
> > for a speed test sysKin and I wanted to inline as many functions as
> > possible, but this isn't possible for NASM code :-(.
> > Does any of you have inline assembler versions of sad16_mmx
> > and/or sad16_xmm in exactly the version as it's in nasm sources? gcc or
> > VC++ doesn't matter.
>
> yes, I have inlined versions of all assembler functions (VC++ inlined),
but
> not the latest versions. However it doesn't make a difference for sad
> because there weren't important changes for months...
>
> Also it's sufficient to compare inlined and non-inline - btw: I did
> something like that earlier already and didn't notice any good improvement
> from inlining, but maybe times have changed in the meantime ;-) So I'll be
> interested in hearing your results...
>
> bye
> Michael
>