Re: transfrom dynamic range
"Aitor Garay" <[email protected]> Wed, 2 Apr 2003 14:10:02 +0200
| Newsgroups | gmane.comp.video.h264.devel |
|---|---|
| Message-ID | <00ad01c2f910$ca90dec0$bcb31fac@HAJC0062> |
> The inverse, gains 4.
^^^^
This is correct?
Let me recapitulate:
residual [9 bits] -> transform -> [15 bits] -> quantize -> [X bits] -> inv quantize -> [15 bits] -> int transform -> [9 bits]
The standard says that input to IT is in range [-2^15...( 2^15)-1] ( as Indrajit said), and output in [-2^15...( 2^15)-33].
Then, the normalization of EQ 8-278 takes it back to 9 bits.
Another thing is that inside quantization ( using the reference software approach to change divisions by right shifts) dynamic
range go up to nearly 32 bit because of the multiplication with 14 bit constants. As i see it, this could be problematic for optimized
MMX code unless both integer multiplication/divisions are used.
/AITOR
----- Original Message -----
From: [email protected]
To: [email protected] ; [email protected]
Sent: Wednesday, April 02, 2003 12:56 PM
Subject: Re: [Hdot264-devel] transfrom dynamic range
The forward transfrom gaines 6 bits over the range of the input data ,
The inverse, gains 4.
<[email protected]>
[Hdot264-devel] transfrom dynamic range
Hi there!,
Which is the real dynamic range of the values output by the forward transform? Full 16 bit,
or even in a worst case less than that?
/AITOR
------------------------------------------------------------------------
------------------------------------------------------------------------
------------------------------------------------------------------------------
Walla! Mail, Get Your Private, Free E-mail from Walla! at:
http://mail.walla.co.il