[x264-devel] Re: alignment of arrays in common/quant.c

[Loren Merritt]

On Wed, 26 Jul 2006, Guillaume POIRIER wrote:

> I'd like to implement some, if not all routines in common/quant.c in
> Altivec. I saw that there are only MMX versions and no SSE version of
> these routines (which don't care about load alignments), so the
> respective alignment of the input arrays isn't quite obvious to me.
> As I'm a beginner in Altivec programming, I'd prefer not to have to
> deal with unaligned loads if possible.

Currently, the input arrays are declared on the stack with no specified 
alignment. But that's easy to change.

> Other than that, is there a particular reason why there's no SSE
> version of quant.c routines (other than because no one wrote them?)

Because I optimize for AMD only, where MMX is faster than SSE2 for most 
tasks (unless there's a specific need for wider registers).

--Loren Merritt

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html