Re:
>Anyway, it crunches about 1.2mb/sec... you know, I'm working on an 
>arithmetic coder driven by probabilistic models, and I was so happy that the whole 
>thing (except coding, which is what's still missing) crunched more than 
>2kb/sec, mainly because the first version was about 25 or more times slower... and I 
>also have a delta coder for PCM data which crunches about 35kb/sec and I was 
>so happy about it... and being only that slow I could use them without 
>waiting until detecting a significant proton decay in the universe...
>
>And now you come up and say "1.2mb/sec with primitives".
>
>Then I remember testing the probabilistic models with Jitter, to see the 
>performance increase at least twofold and only getting 25% more speed, because 
>when I use MessageTally to test the performance of the model I get a lot of 
>time spent in primitives. And the thing is already better that what the book 
>says in terms of performance... snif...
>
>Andres.

If you can find a nice compact inner loop to turn into a primitive,
you might see a speedup of 10-30. Check out the implementation of
compressToByteArray for an example.

	-- John