[XviD-devel] [In progress] devapi4 -- mpeg matrices

[Edouard Gomez]

Edouard Gomez (ed.gomez at free.fr) wrote:
> read: they sux.

The PSNR meter confirms that:

ASM version + forced quant_mpeg_xxx_mmx
===============================================================================
Test name: underworld-trailer
Using cached first pass stats (hash - a1c5ab3ce4420b302dd41ee9886379ea)
 
Results:
xvid:     Min PSNR y : 23,19 dB, u : 34,66 dB, v : 35,37 dB
xvid: Average PSNR y : 36,49 dB, u : 42,70 dB, v : 45,36 dB
xvid:     Max PSNR y : 99,99 dB, u : 99,99 dB, v : 99,99 dB
===============================================================================

C only lib:
===============================================================================
Test name: underworld-trailer
Using cached first pass stats (hash - a1c5ab3ce4420b302dd41ee9886379ea)
 
Results:
xvid:     Min PSNR y : 32,29 dB, u : 39,62 dB, v : 41,77 dB
xvid: Average PSNR y : 39,11 dB, u : 43,10 dB, v : 45,38 dB
xvid:     Max PSNR y : 99,99 dB, u : 99,99 dB, v : 99,99 dB
===============================================================================

Asm versions with XMM quantization functions give similar results to
the C only version.

So i  think we should  just write the  MMX versions again.  XMM versions
should use  less registers to  allow matrix passing, a  complete rewrite
seems not to be needed. Now up to you, assemblers guys, to do your job.

-- 
Edouard Gomez