[vorbis-dev] vorbis on playstation 2

Nicolas Pitre nico at cam.org
Thu Feb 6 18:06:23 PST 2003

On Thu, 6 Feb 2003, David Etherton wrote:

> > Interesting.  I turned on _LOW_ACCURACY_ and rebuilt the library and the
> > total cycle count for my test case went from 950M cycles to 723M cycles.
> > The two sincos lookup tables are 4K each in "full" mode which are likely
> > going to thrash the cache.
> Tried moving the LUT as-is to scratchpad memory.
> Full accuracy: 950->910M
> Low accuracy: 723->711M
> The low accuracy method is starting to look promising -- seems to me that I
> could get away with a 16x16->32 eight-wide MMX multiply to really speed
> things up, particularly since there's a muladd unit there too.

Beware that the low accuracy uses unbalanced multiplications, like 24x8->32.  
This was done to get the best audio quality after many listening tests.  If
you change that bit ratio in either direction you're likely to increase
the presence of artifacts a lot.

Since It's the lookup table values that are on the 8 bit side, I took the 
opportunity to shrink them to char arrays when low accuracy is selected.  It 
didn't happen the other way around.


--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-dev-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.

More information about the Vorbis-dev mailing list