|
Chris@10
|
1 TODO before FFTW-$2\pi$:
|
|
Chris@10
|
2
|
|
Chris@10
|
3 * Wisdom: make it clear that it is specific to the exact fftw version
|
|
Chris@10
|
4 and configuration. Report error codes when reading wisdom. Maybe
|
|
Chris@10
|
5 have multiple system wisdom files, one per version?
|
|
Chris@10
|
6
|
|
Chris@10
|
7 * DCT/DST codelets? which kinds?
|
|
Chris@10
|
8
|
|
Chris@10
|
9 * investigate the addition-chain trig computation
|
|
Chris@10
|
10
|
|
Chris@10
|
11 * I can't believe that there isn't a closed form for the omega
|
|
Chris@10
|
12 array in Rader.
|
|
Chris@10
|
13
|
|
Chris@10
|
14 * convolution problem type(s)
|
|
Chris@10
|
15
|
|
Chris@10
|
16 * Explore the idea of having n < 0 in tensors, possibly to mean
|
|
Chris@10
|
17 inverse DFT.
|
|
Chris@10
|
18
|
|
Chris@10
|
19 * better estimator: possibly, let "other" cost be coef * n, where
|
|
Chris@10
|
20 coef is a per-solver constant determined via some big numerical
|
|
Chris@10
|
21 optimization/fit.
|
|
Chris@10
|
22
|
|
Chris@10
|
23 * vector radix, multidimensional codelets
|
|
Chris@10
|
24
|
|
Chris@10
|
25 * it may be a good idea to unify all those little loops that do
|
|
Chris@10
|
26 copying, (X[i], X[n-i]) <- (X[i] + X[n-i], X[i] - X[n-i]),
|
|
Chris@10
|
27 and multiplication of vectors by twiddle factors.
|
|
Chris@10
|
28
|
|
Chris@10
|
29 * Pruned FFTs (basically, a vecloop that skips zeros).
|
|
Chris@10
|
30
|
|
Chris@10
|
31 * Try FFTPACK-style back-and-forth (Stockham) FFT. (We tried this a
|
|
Chris@10
|
32 few years ago and it was slower, but perhaps matters have changed.)
|
|
Chris@10
|
33
|
|
Chris@10
|
34 * Generate assembly directly for more processors, or maybe fork gcc. =)
|
|
Chris@10
|
35
|
|
Chris@10
|
36 * ensure that threaded solvers generate (block_size % 4 == 0)
|
|
Chris@10
|
37 to allow SIMD to be used.
|
|
Chris@10
|
38
|
|
Chris@10
|
39 * memoize triggen.
|
