S. Newhouse wrote:
> R Fateman wrote:
>
>> I think that your programs are way too elaborate.
>> if a+b is a correctly rounded answer, then a lower bound on a+b is to
>> adjust by one unit in the last place. (or even half a unit if you can
>> do that.)
>>
>> IEEE floats and bfloats are correctly rounded.
>> or should be.
>>
>>
>>
>>
> I see that the IEEE 754 standard gives precise rounding rules for up to
> 80 bits (extended precision), but I don't see anything about higher
> precision than 80 bits.
> Where is the source file for bfloat?
>
>
src/float.lisp.
Richard Fateman wrote most (all?) of that, so if he says they're
correctly rounded, they are correctly rounded, barring bugs, of course.
I think bfloat uses the same rules for rounding as IEEE 754.
Ray