rontolisp:bits-bfloat16

(rontolisp:bits-bfloat16 integer)

Returns the float a bfloat16 bit pattern encodes; only the low sixteen bits of the argument are read. Widening is exact and total, so this never rounds -- every pattern names a float, and rontolisp:bfloat16-bits takes it back unchanged.

The second value is the point: 0.1 is not a bfloat16, and the round trip shows exactly which float the width can hold.