As far as CPU utilization, I suspect it's due to cache misses. A pxa270 with identical size cache and association size 32 runs the encoding at about 4% of the cpu; however the 2420, with its 4 associations and similar speed needs around 50%. There is a section of code that moves memory around on every pass so we'll target that first.

There was interest from some people at the university that partners with INdT to put the codec on the dsp, but it was far from official.