[XviD-devel] overflow being reported incorrectly

Jim Hauxwell xvid-devel@xvid.org
Sat, 4 Jan 2003 12:42:00 -0000


yes you're correct in that it does something.  I scan read it, and the
implication is that on athlon its zero latency and zero switch cost (which
is not the same as doing nothing).

If they are both equal then I would switch back to emms

Jim

> -----Original Message-----
> From: xvid-devel-admin@xvid.org [mailto:xvid-devel-admin@xvid.org]On
> Behalf Of Christoph Lampert
> Sent: 04 January 2003 12:18
> To: xvid-devel@xvid.org
> Subject: RE: [XviD-devel] overflow being reported incorrectly
>
>
> On Sat, 4 Jan 2003, Jim Hauxwell wrote:
>
> > Hi,
> >
> > I came across the femms problem this morning also.  seems like intel and
> > microsoft compilers cant build for it.  Even the intrinsics version
> > _m_femms() dosn't work so I've just commented it out.
> >
> > As a matter of interest, why has someone added this?  According
> to AMD doc I
> > read this morning, the instructions femms and emms have zero latency and
> > dont do anything.
>
> On Athlon it has zero latency, on K6-2 it doesn't.
> And it surely does do something: It switches registers back from MMX/3Dnow
> to floating point mode. If you don't switch before using floats, you end
> up with garbled float results.
>
> I think FEMMS is AMDs "faster" version of EMMS, maybe we should switch
> back to EMMS, and if it's slower on Pete's AMD K6, we can make a
> special macro for him to avoid problems.
>
> :-)
>
> gruel
>
>
>
> >
> >
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/2262
> 1.pdf
>
> Maybe someone can shed some light on it.
>
> Jim
>
> > -----Original Message-----
> > From: xvid-devel-admin@xvid.org [mailto:xvid-devel-admin@xvid.org]On
> > Behalf Of Dirk Knop
> > Sent: 04 January 2003 11:21
> > To: xvid-devel@xvid.org
> > Subject: Re: [XviD-devel] overflow being reported incorrectly
> >
> >
> > Hi,
> >
> > suxen_drol wrote:
> >
> > >fix commited (stable & dev).
> > >
> > >a potential problem: our dequant functions are only good for coeff
> > >levels in the [-256,255] range.
> > >
> >
> > Uhm, we should definatly fix that ;)
> >
> > Another problem with the latest commit:
> > D:\data\PROGTEMP\XviD-dev-api-3\xvidcore\src\utils\emms.c(79): error:
> > __asm unknown opcode 'FEMMS'
> >
> > .... is there a way to explicit call nasm for instructions in c-files?
> > masm doesn't seem to understand that instruction (processor pack is
> > installed). (Or does intel install an assembler as well? I wonder why it
> > doesn't understand that then).
> >
> > Best regards
> > Dirk
> >
> >
> > _______________________________________________
> > XviD-devel mailing list
> > XviD-devel@xvid.org
> > http://list.xvid.org/mailman/listinfo/xvid-devel
>
> _______________________________________________
> XviD-devel mailing list
> XviD-devel@xvid.org
> http://list.xvid.org/mailman/listinfo/xvid-devel
>

_______________________________________________
XviD-devel mailing list
XviD-devel@xvid.org
http://list.xvid.org/mailman/listinfo/xvid-devel