Hi everyone,
I'm still waiting on a few graphs to be generated for my next blog
post (should be done within the next 1/2 hour), but I thought I would
push the text-only version here.
And has it finished yet? 
Hi Koen,
Right before I was about to submit my post, I'd re-generated a bunch
of graphs and ran into a really mysterious problem where ffmpeg wasn't
getting past N=32. Kind of frustrating. I think I have it fixed though
and will be back tonight to finish it up.
I also just realized that I'd completely neglected optimized memory
copy routines (these are what FFTW refers to as rank-0 transforms)
which deal with buffering: i.e. copying for out-of-place transforms,
but also turning strided data into non-strided data (which is what
ffmpeg expects). You can probably imagine that optimized copy routines
would greatly speed up just about everything. I've identified exactly
which files need to be considered, and I'll have a look at it tonight
after I get back from my lab.
C