D124 - The division sequence is refused, and says which numbers are missing

Status: assumed

v_div_scale_f32, v_div_fmas_f32 and v_div_fixup_f32 are in model::BLOCKED with individual reasons rather than translated.

They implement float division together: pre-scale an operand by a power of two so the reciprocal that follows cannot overflow, refine, then scale back and substitute results for the special cases - zero over zero, infinity over infinity, signed zeros, NaNs. The thresholds that decide when to scale, and the table of substitutions, are stated in the published instruction set and are not available here.

A first guess would be exact for ordinary values and wrong at the extremes, which is the worst shape a numerical bug takes: it survives every test somebody thinks to write and appears years later as a rendering artefact nobody can reproduce. Refusing costs one fixture; guessing costs the ability to trust any division.

Worth distinguishing from the other entries in that list: exp is blocked on a subsystem that does not exist, and these are blocked on documentation that has not been consulted. The second is much cheaper to unblock, and the reason strings say so.