[mdlug] For The Mathematically Inclined

[LAP]

Raymond Ingles wrote:

>
> I wonder if anyone will take this test suite and see if they can improve
> GLIBC? That's certainly been done with testing tools that find bugs in the
> Linux kernel
>

This work was just a comparison of all the math libraries.

GLIBC is well aware of its own accuracy:

https://www.gnu.org/software/libc/manual/html_node/Errors-in-Math-Functions.html

>From the above link:

"the GNU C Library does not aim for correctly rounded results for
functions in the math library..."

Then they go on to explain their accuracy goals.

But from the comparison paper GLIBC does a good job being second only
to the Intel MKL.

The real question, for me, is how many scientific/engineering people
actually use GLIBC regularly.  Most, I would suspect, use commercial
software like Matlab or Mathematica.  These commercial packages don't
use GLIBC but a lot (most?) of them do use GMP, or the GNU Mulit-Precision
software.

Mathematics and engineering with GNU/Linux is a big area and much
could be said.

GNU/Linux is big on servers, embedded systems, and supercomputers
but not so big on desktop workstations, and that probably includes
scientific/enginering workstations.