Unclear on the algorithm's use of DCT

[seydar]

(The issue title is labeled as best as I can — I recognize that the fault for this issue lies with me)

I'm trying to implement blurhash in Haskell, and I'm trying to stick to the Algorithm.md page as much as possible. In addition, I'm trying to use the DCT2 function that's provided by FFTW3 so that I can best understand how the DCT functions in relation to the algorithm (as opposed to just translating the code directly).

Lines 132-134 encapsulate the logic of the algorithm (https://paste.tomsmeding.com/kgoLDWh8) as I understand it: convert all the sRGB values to linear, perform a DCT on the array, convert the linear values back to sRGB, and go on from there.

Unfortunately, I am unable to get an answer that makes sense: the DCT results are all pegged high at 255.

Can someone help me see the error of my ways here?