Re: [Openexr-devel] Conversion between half-float and float

[Paul Schneider]

Hi, Kennant,

I haven't looked at the Derek's code that closely, and I don't have 
much experience with Altivec programming (and I have zero experience 
with SSE).  I know of a few people who are working on half-to-float 
conversions using SSE and they report significant speedups, but I don't 
think that any of them are in a position to discuss their techniques at 
this time (I wouldn't want to speak for others, though).

Derek, if you're reading this and feel like sharing any thoughts on 
implementing half-to-float conversions using SIMD processors, I'm sure 
your insight would be appreciated :)

- Paul


On Jun 25, 2004, at 10:17 PM, Kennant Tom wrote:

> Hello Paul,
>
> I have read through that piece of code before. I have thought about 
> implementing the conversion using the SIMD2 intrinsics. It seems that 
> the conversion involves more than a few logical shift operations and 
> the while-loop for denormalizing normalized values is difficult to 
> implement neatly with SIMD2 intrinsics. Do you know the actual logic 
> behind the implementation using the AltiVec functions.
>
> Regards,
>
> Kennant
>
>
> > From: Paul Schneider <address@hidden>
> > To: "Kennant Tom" <address@hidden>
> > CC: address@hidden
> > Subject: Re: [Openexr-devel] Conversion between half-float and float
> > Date: Fri, 25 Jun 2004 14:23:34 -0700
> >
> >
> > Hi, Kennant,
> >
> > If you don't have a nice video card, but you do have a Mac, you might 
> > be able to use these Altivec-enabled conversion routines.  If you're 
> > on an x86 machine and you have some experience with SSE, you might be 
> > able to adapt these algorithms to do SIMD conversions using SSE.
> >
> > From: "Derek Ledbetter" <address@hidden>
> > Date: August 28, 2003 4:54:05 PM PDT
> > To: <address@hidden>
> > Subject: [Openexr-devel] AltiVec functions that operate on "half"
> >
> > I wrote two functions that use AltiVec (the vector processing unit on
> >  the PowerPC G4 and G5) to convert between "half"s and "float"s.?I 
> > found
> >  that converting a large array of "float"s to "half"s is about three
> >  times faster using AltiVec, and it's about 1.7 times as fast going 
> > the
> >  other way.?I figure that they should be useful for other functions 
> > too.
> >
> >  <http://www.serve.com/derekl/openexr/AltiVecHalfConversion.html>
> >  --
> >  Derek Ledbetter
> >  address@hidden
> >
> >
> >
> >  _______________________________________________
> >  Openexr-devel mailing list
> >  address@hidden
> > http://mail.nongnu.org/mailman/listinfo/openexr-devel
> >
> >
> >
> >
> > On Jun 24, 2004, at 8:36 PM, Kennant Tom wrote:
> >
> > > Hello all,
> > >
> > > Does anybody out there know a way that can convert between 
> > > half-float and float faster than the implementation provided the 
> > > OpenExr group?
> > >
> > > Regards,
> > >
> > > Kennant
> > >
> > > _________________________________________________________________
> > > Surf the net and talk on the phone with Xtra JetStream @  
> > > http://xtra.co.nz/jetstream
> > >
> > >
> > >
> > > _______________________________________________
> > > Openexr-devel mailing list
> > > address@hidden
> > > http://lists.nongnu.org/mailman/listinfo/openexr-devel
>
> _________________________________________________________________
> There’s never been a better time to get Xtra JetStream @  
> http://xtra.co.nz/jetstream