Mercurial > hg > sv-dependency-builds

annotate src/fftw-3.3.5/TODO @ 83:ae30d91d2ffe

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Replace these with versions built using an older toolset (so as to avoid ABI compatibilities when linking on Ubuntu 14.04 for packaging purposes)
author Chris Cannam
date Fri, 07 Feb 2020 11:51:13 +0000
parents 2cd0e3b3e1fd
children
rev   line source
Chris@42 1 TODO before FFTW-$2\pi$:
Chris@42 2
Chris@42 3 * figure out how to autodetect NEON at runtime
Chris@42 4
Chris@42 5 * figure out the arm cycle counter business
Chris@42 6
Chris@42 7 * Wisdom: make it clear that it is specific to the exact fftw version
Chris@42 8 and configuration. Report error codes when reading wisdom. Maybe
Chris@42 9 have multiple system wisdom files, one per version?
Chris@42 10
Chris@42 11 * DCT/DST codelets? which kinds?
Chris@42 12
Chris@42 13 * investigate the addition-chain trig computation
Chris@42 14
Chris@42 15 * I can't believe that there isn't a closed form for the omega
Chris@42 16 array in Rader.
Chris@42 17
Chris@42 18 * convolution problem type(s)
Chris@42 19
Chris@42 20 * Explore the idea of having n < 0 in tensors, possibly to mean
Chris@42 21 inverse DFT.
Chris@42 22
Chris@42 23 * better estimator: possibly, let "other" cost be coef * n, where
Chris@42 24 coef is a per-solver constant determined via some big numerical
Chris@42 25 optimization/fit.
Chris@42 26
Chris@42 27 * vector radix, multidimensional codelets
Chris@42 28
Chris@42 29 * it may be a good idea to unify all those little loops that do
Chris@42 30 copying, (X[i], X[n-i]) <- (X[i] + X[n-i], X[i] - X[n-i]),
Chris@42 31 and multiplication of vectors by twiddle factors.
Chris@42 32
Chris@42 33 * Pruned FFTs (basically, a vecloop that skips zeros).
Chris@42 34
Chris@42 35 * Try FFTPACK-style back-and-forth (Stockham) FFT. (We tried this a
Chris@42 36 few years ago and it was slower, but perhaps matters have changed.)
Chris@42 37
Chris@42 38 * Generate assembly directly for more processors, or maybe fork gcc. =)
Chris@42 39
Chris@42 40 * ensure that threaded solvers generate (block_size % 4 == 0)
Chris@42 41 to allow SIMD to be used.
Chris@42 42
Chris@42 43 * memoize triggen.