[Enzyme] Mark assert applicable as non-differentiable

[wsmoses]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.

Project coverage is 70.73%. Comparing base (cefd827) to head (fb6ecef).

Files with missing lines	Patch %	Lines
ext/FFTWEnzymeExt.jl	0.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #307      +/-   ##
==========================================
- Coverage   70.99%   70.73%   -0.27%     
==========================================
  Files           5        6       +1     
  Lines         531      533       +2     
==========================================
  Hits          377      377              
- Misses        154      156       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[wsmoses]

@stevengj mind giving this a review?

[ChrisRackauckas]

Add a downstream test of Enzyme differentiation on FFTW?

[wsmoses]

It doesn’t yet differentiate it fully, but this is a necessary pre requisite

[giordano]

I suspect this needs to add Enzyme to the extras to make Pkg happy. But also, might be easier to move the extension to Enzyme? Pkgextensions are symmetrical, it doesn't really matter which of the two packages hosts them.

[wsmoses]

Yeah that’s always an option. My inclination here is to put in an FFTW extension since the functions marked inactive are really internals to fftw and I’d prefer not to touch internals of a package from another package. FFTW.jl doesn’t change that much though so it might be fine

[giordano]

This is relevant: https://discourse.julialang.org/t/autodifferentiation-with-fft-with-enzyme/128674/2. CC @danielwe

[danielwe]

Rules should probably live in AbstractFFTs.jl. There's an abandoned draft PR: JuliaMath/AbstractFFTs.jl#103. If you intercept at that level, you probably don't need to worry about marking any FFTW.jl internals inactive---at least I haven't needed that for the rule I shared on Discourse to work.