Re: Bit-exact floating-point computations

Christian Bau <christian.bau@cbau.freeserve.co.uk> writes:
>> "Write Once, Run Anywhere"?
>> Without bit-exact correctness, it's but a shameful lie.
>
> But this is a totally different thing: To get bit identical results
> (for sine, cosine and so on) you don't need the ridiculous precision
> that Java prescribes. What you want is that sin (1.0e300) produces the
> same result on your machine as on mine. That is easily achieved by
> defining the actual code that the sin implementation uses, and by
> defining semantics of + - * / as bit exact as Java does anyway.


Java does the right thing here: because the only future-proof way of
defining sin(1.0e300) is the bit-exact correct result. Standardizing
a kluge is the best way to build heaps of rubbish in the future so as
to be backward compatible with ancient Java applications when all modern
processors of the day (say in 2010) compute sin(1.0e300) without problem.


An acceptable compromise would be do define a large interval inside which
bit-exact correctness is guaranteed, and outside of which warranties are
less stringent.


Allowing spurious discrepancy for the sake of speed is just Wrong(TM).
Standardizing on wrong answers is Evil(TM).


[ François-René ÐVB Rideau | Reflection&Cybernethics | http://fare.tunes.org ]

Return to the comp.compilers page.
Search the comp.compilers archives again.