Function float_format 

pub fn float_format(kind: FloatKind, target: &TargetInfo) -> Format

The binary format a real floating type has on target.

Not derivable from float_width, which is why it is a separate question: the width of a long double on SysV x86-64 is a hundred and twenty eight bits and its format is the eighty bit x87 one, and a compiler that picked the format by the size would fold every long double constant on that target with seventeen decimal digits too many.