This code has too few legal range checks, so you can tell it so do things the codec can't support. First you can't use odd numbers for the windowing as the image is CFA RAW, color pattern is an even number repeat -- technically I could switch the bayer phase to handle odd numbers -- currently values these are rounded. Next the codec likes horizontal encodes of a 16 multiple and vertical of 2. So this should work -(8,0,1928,960) -- except I introduced a bug it doesn't. Fix coming.
I did have a look at the BMDFilm output curve, and it is a little odd. It is likely has BMD sensor calibration stuff in it, the curve we use for Protune and others, mathematically models the mapping of linear to log and back again, the BMD curve don't fit our models. PanaLog, S-Log, C-Log, Protune etc. are all nice mathematical equations, but not BMD. If we were to use that curve natively, either a lot of code changes to the codec, or an approximation of the curve would produce (mild) color errors in the matrix, white balance and saturation (all the linear light operations.)