xtensor-fftw

xtensor-fftw copied to clipboard

Rewrite README

1

Take inspiration from the other xtensor pages, e.g. [xensor-io](https://github.com/QuantStack/xtensor-io) has a nice compact one, that would be good as a header. The rest can be something like: 1. Intro 2....

egpbos

We should add items to the [numpy to xtensor cheat sheet](http://xtensor.readthedocs.io/en/latest/numpy.html) and a code snippet in the [xtensor related projects section](https://xtensor.readthedocs.io/en/latest/related.html).

egpbos

Currently, the helper functions in `helper.hpp` only work on single dimensional arrays. We should extend this to multi-dimensional arrays. `numpy.fft`'s helper functions use the `axes` keyword argument for this. This...

egpbos

In basic.hpp, there are a few exceptions to safeguard against `nullptr`s from FFTW plan creation, which for instance can happen if you try to build a plan for a multi-dimensional...

egpbos

Add FFTW accuracy tests

1

Like these: - https://dsp.stackexchange.com/questions/633/what-data-should-i-use-to-test-an-fft-implementation-and-what-accuracy-should-i - http://www.sccon.ca/sccon/fft/fft3.htm

egpbos

Find a stable place to benchmark and record benchmarks

1

Some decent machine + maybe a website to publish the recorded results. Continuous benchmarks would be even better, since then we can catch performance regressions more easily.

egpbos

Benchmark 1,2,3-dimensional functions against general n-dim ones

1

We should benchmark whether the 1-3 dimensional special functions are actually faster than the n-dimensional one. If not, we can just use one implementation function for all ranks, with simple...

egpbos

After implementing the basic numpy.fft-like interface (#6), we should benchmark it against the actual numpy.fft functions.

egpbos