I got the shine code from the link in the rockbox encoder page.
To get reasonable performance on iriver I had to rewrite the code completely.
For performance reasons I merged the mdct algorithm from a later lame encoder.
Finally I optimized the code to get the best quality from integer calculation
So basically the internal technique is shine, but optimized for iriver.
The encoder quality is very close to the shine quality, which equals lame quality
This encoder does not use any of the lame advanced mp3 encoding features.
I once tried to implement joint stereo, but gave up. There is also no psymodel
For my purpose (radio recording on iriver) this encoder is quite satisfying
because of the relatively poor radio quality.
Because the current source is based on shine, which is based on lame, there are
still a lot of unused options from the original code.
And yes, the delay should still be 576 samples.