also moves some tests to the appropriate code block
there is an argument to be made that the caller "chose" Float32 so we should respect the numerical conclusions that arise from that, but it seems a bit unreasonable to expect callers to be able to predict the implications in this case
also moves some tests to the appropriate code block
there is an argument to be made that the caller "chose" Float32 so we should respect the numerical conclusions that arise from that, but it seems a bit unreasonable to expect callers to be able to predict the implications in this case