Hard float for mips16 means essentially to compile as soft float but to
use a runtime library for soft float that is written with native mips32
floating point instructions (those runtime routines run in mips32 hard
float mode).
The patch reviewed by Reed Kotler.
llvm-svn: 195123