Re: floating point

Bruce Dawson <comments@cygnus-software.com> wrote:
> 2) Those who want predictable double precision results aren't happy
> because the results they need are impossible to get all the time.
> Although, with rounding set to double precision they probably do get
> them 99.999% of the time - any other guesses?


If you store every intermediate result to catch out of range exponents
then setting the default precision to double does work okay - except
for denormalised numbers where you still get double rounding. And you
don't have to set the precision in single precision before doing the
operation and storing to get it to work okay - even when dividing or
getting square roots. Storing and reloading does of course crucify
performance but the alternative is using software emulation.


> In short, if you want to design an FPU that has an ultra-fast or
> ultra-precise mode you have to make sure it can be turned off
> completely, for those who want predictability. And, you have to make
> sure that turning it off is trivial - forcing compiler writers to do
> anything more than set a bit is unacceptable - they won't do it.


Completely agree. Intel should fix their FPU, I can't see why they
haven't done so long ago as the changes would be pretty trivial
in hardware terms even if they kept a heritage mode.
--
David McQuillan

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