glHint - precision for fragment programs

Considering Nvidia’s problem of having ARB_fragment_program running in 32 bit float mode by default (and hence getting creamed by ATI’s 24 bit floats) I was wondering if it would be possible to have “hint” flags like there is with fog. (ie. GL_FRAGMENT_PROGRAM_NICEST, GL_FRAGMENT_PROGRAM_FASTEST ) As games could probably live with 16 bit floats for most things and people who needed it could use the full precion when necessary.

It would be best if this hint was applied on a per-program basis (ie you could have full precision and half precision fragment programs in the one program)

Nix. the above, just read in the spec that this already exists

One problem is that different types of calculations require different precision. Most color blending operations used in games will do fine in 16-bit precision, but computing texture addresses (e.g. for dependent texturing) 16 bits is definitely not enough.