Enable double to float shrinking optimizations for binary functions like 'fmin/fmax...