- The biggest missing feature by far: optimized, strided (and non-strided) memory copies. You can see where I was going with this. 1D transforms are fairly straight-forward, but 2D and beyond will require a bit of work.
- Further testing of rdft-ffmpeg.c and reodft-ffmpeg.c . I did do some very basic forward transformation tests to ensure that I had the correct output for forward transforms, but not for reverse.
- Ideally, a few extremely fast transforms in neon / asm for small, non-power of two cases. These would compliment the extremely fast power-of-two cases and also make composite transforms incredibly faster (much like on x86).
I'll definitely be tackling those issues after my thesis is handed in. Only 16 more days and then I'm done in academia for the rest of my life!