Re: Re: Re: MMX version of 4x4 Forward Transform

It has no effect when the qp>11 
for qp<12 the result might differ (the math has to do something with the righ-shifts in the first pass eq. 8-270 / 8-271), 
but the error should be minor or zero.
This was defined so that all implementations will comply exactly.

<[email protected]>
Re: Re: [Hdot264-devel] MMX version of 4x4 Forward Transform

    Why that emphasis on _first_ vertical, and _then_ horizontal?  I'm not a matrix expert, but i
have checked and the order seems not to affect the result.  The dynamic range of the intermediate
values should be the same, so it is not an accuracy issue.

    /AITOR

----- Original Message ----- 
From: [email protected] 
To: [email protected] ; [email protected] 
Sent: Monday, March 31, 2003 12:24 PM
Subject: Re: Re: [Hdot264-devel] MMX version of 4x4 Forward Transform




<[email protected]>
Re: [Hdot264-devel] MMX version of 4x4 Forward Transform

For those who are working on MMX transform, please pay attention to a (new) clarification in the latest draft.
Section 8.5.8 (where the transform is defined), first the vertical transform should be applied, then the horizontal.
enjoy




Walla! Mail, Get Your Private, Free E-mail from Walla! at:
http://mail.walla.co.il

-----------------------------------------------------------------------
Walla! Mail, Get Your Private, Free E-mail from Walla! at:
http://mail.walla.co.il