More cleanup of `general_` functions

This extracts as much of the common logic from the nd-transforms as possible into generic functions. With the transform-specific logic handled by the exec functor. I've also tried to abstract all of the data movement and vector-type reordering into overloads of the various copy_ functions.

This does currently keep r2c and c2r as their own special cases since they are subtly different in a few ways.

changed the description

added 1 commit

• ac891a30 - Fix hartley transforms by disallowing inplace FFT output

Compare with previous version

I like the changes a lot, but something still seems to break for multi-D transforms - perhaps normalization related?

Does test.py run through cleanly for you?

I'm mostly away from computers this week, but will try to have a closer look later on!

added 1 commit

• 15278280 - fix

Compare with previous version

Found it :) I took the liberty of pushing directly onto your branch.

Thanks for the fix. I have trouble running test.py because all of the longdouble transforms fail on my system due to Windows WSL bugs. It's hard to sort through thousands of test failures.

I wish I knew of a way to detect WSL and mark the tests as xfail.

I have trouble running test.py because all of the longdouble transforms fail on my system due to Windows WSL bugs

I have an easy workaround for this on some other branch. Will see if I can commit this.

added 1 commit

• 8aa2d5cd - xfail long double tests on WSL and platforms without extended precision

Compare with previous version

I actually found a method searching around: https://www.reddit.com/r/bashonubuntuonwindows/comments/85jghk/how_to_allow_python_to_know_im_on_windows/dvxwy9t/

I'm now seeing test failures for the dct but it's only an error of ~1e-14. Do these fail for you or is it just WSL being buggy?

I think WSL just has an inaccurate libmath implementation. The DCTs currently call cos instead of the special sincos code in pocketfft - that may have been a bad idea.

That sounds likely. I can take a look at the cos calls to see if it fixes the issues if you like.

If you like to play with this, feel free! But it's probably going to be a bit tricky; alternatively you could just relax the tolerances a bit (maybe also only on Windows).

merged

mentioned in commit bef3070e