rontolisp:bits-float16

(rontolisp:bits-float16 integer)

Returns the float an IEEE 754 binary16 (f16) bit pattern encodes; only the low sixteen bits of the argument are read. Every one of the 65536 patterns decodes to a float -- infinities and NaNs included -- but the round trip through rontolisp:float16-bits is not always the identity on a NaN's payload: the JDK's own Float.float16ToFloat/Float.floatToFloat16 pair, which this primitive is built on, can quiet a signalling NaN on the way through. rontolisp:bits-bfloat16/rontolisp:bfloat16-bits has no such gap.

The second value is the point: 0.1 is not an f16 value, and the round trip shows exactly which float the width can hold.