* `f32tof16 $aDst0, $aSrc0`

Convert a *single-precision* value to *f16*, using the rounding mode as
specified by `$FP_CTL.RND`/`$FP_CTL.ESR`. Supports stochastic rounding.
See fp_format_conversions.

The 16-bit result of the conversion is broadcast to (duplicated into) a
single ARF register, producing a 2-element vector of identical *f16*
values.