The floating point functions are here rather than compiler-rt because the libc softfloat code allows us to set the rounding mode.
Andrew Turner