* `roll16 $mDst0, $mSrc0, $mSrc1`
* `roll16 $aDst0, $aSrc0, $aSrc1`

Perform a SIMD *roll* permutation on the 4 x 16-bit values across 2
source registers. Equivalent to a SIMD *roll* operation on 8 x 8-bit
values.