does anyone know if GPU mipmap generation takes colour space into account when blending the source image pixels? (or I guess texels? I'm not a graphics guy)
I ask because blending linear RGB values produces different results to blending nonlinear RGB values (with the transfer function pre-applied, e.g. sRGB OETF, HLG, or PQ) and that can be quite important for colour accuracy.
I remember tackling this years ago but I can't remember what my conclusion was.