Re: [PATCH 0/2] libm/riscv: Fixing machine-specific fma/sqrt

[Joseph Myers <joseph@codesourcery.com>]

On Thu, 3 Sep 2020, Keith Packard via Newlib wrote:

> I've got a bunch of errno/exception tests that I've written for picolibc
> that catch missing errno values, which is what the sqrt difference would
> have caused. As for fma, you'd have to construct a test that checked for
> the difference between a true fma and (a*b)+c that the general code in
> newlib currently uses.

glibc has a great many tests of expected libm function return values, 
errno and exceptions, both manually maintained (libm-test-*.inc) and with 
expected results generated by MPFR (auto-libm-test-in generating 
auto-libm-test-out-* via gen-auto-libm-tests.c).  It would be interesting 
to see what those tests show up as bugs when run on different libm 
implementations (and vice versa, of course, running other libm 
implementations' tests on glibc).  However, it's probably nontrivial to 
get the tests running on other implementations, since they freely make use 
of glibc-specific features, and the particular glibc choices they test for 
regarding errno, exceptions and expected accuracy of results may not match 
the choices made by other libm implementations.

-- 
Joseph S. Myers
joseph@codesourcery.com