[XviD-devel] new thread about on the fly dequant

Edouard Gomez ed.gomez at free.fr
Wed Jun 30 00:24:36 CEST 2004


Hey hey,

I did proper testing of the on the fly dequant patch i sent a
while.

Test procedure:
 - encode the same sequence at different bitrates
   738kbps, 2029kbps, 4015kbps, 6423kbps (saturated codec)
 - see how decoding time relates to zero/nonzero ratio
 - ensure the output is the same.

Brute data:
# bitrate zeros nonzeros time

# on the fly dequant
738 2074487 3418343 32.564
2029 26341644 17603943 37.288
4015 63499650 41808387 41.054
6423 89326200 64513839 43.391


# usual dequant
738 2074487 3418343 32.987
2029 26341644 17603943 38.067
4015 63499650 41808387 44.177
6423 89326200 64513839 49.968

Analysis:
At low bitrates, where the ratio is nearly 1:1, both methods
are equivalent, but as soon as the ratio is in favour of zero
coeffs, then the 'on the fly dequant' code scales better.

Do you guys see something my test misses and that you think it
is important to test before CVS commit (or first my branch
commit) ?

-- 
Edouard Gomez


More information about the XviD-devel mailing list