f56f1d7735
to floating-point, the result is a quiet NaN. The current implementation may return a signaling NaN, and the vendor has no plans for changing this, for reasons explained in the comment I added.