What is the most precise Xvid Decoder Implementation (quality side)

[3ngel]

Well, pheraps it can appear a stupid question (or violating the rule "no best"), but because of i can see a difference when using the Mpeg2 Dgindex setting it to "IEEE Float reference", and because of i see there many development of the Xvid decoder (with FFDshow giving the possibility to select the idct algorithm) i was wondering what is the most "precise" decoder actually developed.
Mine is a curiosity both practical (in order to squeeze of even the last bit of quality out of a stream that in xvid case in most cases are low bitrate so that's even more important) and academic (for culture).
They output the same quality no matter what? In case there is a difference why not develop an "IEEE float" algorithm?

Moreover i don't see this question in the FAQ so it could be useful to add.

Thanks

[clsid]

To get the highest possible decoding quality for MPEG-4 you should select the iDCT for decoding that matches the DCT algorithm that was used for encoding.

[3ngel]

Mm, that's interesting, but it seems to me that for example in the classic xvid codec i can't select the dtc algorithm but only the "quantitazion type" (h263 or Mpeg).
So at this point, what's dct xvid is using to do compress? Or it does depends on the "custom matrices" (or it does depend on the profile too)?

[imcold]

It uses Skal's fdct implementation, and it doesn't depend on the matrix. Matrices are related to quantization, they don't have any effect on chosen fdct algorithm.

[3ngel]

I see, so i'm supposed to use the xvid decoder (wich i presume uses the same idct) or setting ffdshow to use Skal, to obtain the best quality?

Thanks :)

[clsid]

In recent builds of ffdshow that would be the 'Xvid' iDCT (how surpising :)), a further optimized version of Skal's iDCT.

But at 'auto', I think that is also what it uses. So you don't need to change anything.

[3ngel]

I see, much thank for the reply :)