On 21 April 2013 20:31, Paul Khuong <pvk@pvk.ca> wrote:

> IIUC, rtoym didn't implement quad floats as much as double doubles: they're not "just" floats with 128 bits of

Oh, that's interesting. I didn't follow his work in any sort of detail beyond being aware of it. Doesn't really change my stance on arbitrary-precision floats being better an entirely new subclass of NUMBER rather than LONG-FLOAT...

Read an interesting discussion pro/contra double-double vs quad floats on the glibc mailing list. To quote: "No one will use software IEEE binary128 and we don't have hardware binary128." ...so there are clearly points in favor of using double-double as long-float today, but there's also a 2012 paper on a lower-cost FMA for binary128, so maybe we'll actually see hardware for it surfacing before too long? One can hope...

On similar topic, a 16 bit short float would also neat as well ...but again, neither here or now. :)


 -- Nikodemus