#include "tensorstore/util/bfloat16.h"
BFloat16& tensorstore::BFloat16::operator=(float v);

Possibly lossy conversion from float.