For something like retinex, the added precision may not be necessary and introduce needless slowdowns. We should figure out a way to bring the speed back up if possible, be it by introducing a new function for the approximation, adding a parameter, or figuring out a faster way to do an accurate blur.
This commit https://github.com/Jaded-Encoding-Thaumaturgy/vs-rgtools/commit/c6b27072fbdb573ed5714892f9f123988f216384 reportedly caused a severe performance loss when called from retinex. Seems tied to the downscaled gaussian approximation or what-have-you.
For something like retinex, the added precision may not be necessary and introduce needless slowdowns. We should figure out a way to bring the speed back up if possible, be it by introducing a new function for the approximation, adding a parameter, or figuring out a faster way to do an accurate blur.