Accuracy of Mathematical Functions in Single, Double, and Quadruple Precision

[Joseph Myers]

On Fri, 4 Dec 2020, Paul Zimmermann wrote:

>        Hi,
> 
> I have published a new version of my note, including results for double and
> quadruple precision, and also for bivariate functions (atan2, hypot, pow):
> 
> https://homepages.loria.fr/PZimmermann/papers/#accuracy

Thanks.  The errors above 9ulps that you found for tgamma and lgamma for 
binary128 would be good to mention in the bugs you opened for those 
functions for double, as they're evidence improvements are needed for 
those functions for binary128 as well.

> In short, the GNU libc is the second best library in terms of accuracy,
> behind the Intel Math Library.

I think the FreeBSD libm would be worth testing as well and may well be 
quite good - but may be harder to include as you'd need to run the tests 
on a FreeBSD system.

-- 
Joseph S. Myers
joseph@codesourcery.com