[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANv4PNkKXLgmBFFrgZNLkJqoP49x4FGA_+8cK+kxLT0SHG1Y_w@mail.gmail.com>
Date: Sat, 2 Feb 2019 21:05:13 -0500
From: Morten Welinder <mwelinder@...il.com>
To: musl@...ts.openwall.com
Subject: Re: Possible Mistype in exp.c
Thanks.
> there are no universal tests for math functions [...]
I'm painfully aware of that, :-/ However, there are a few collections
of test values out there. Of varying size and quality.
crlibm has fairly large tables of test values. See, for example,
https://github.com/taschini/crlibm/blob/master/tests/sin.testdata
John Burkardt has a pile at
http://people.sc.fsu.edu/~jburkardt/c_src/test_values/test_values.c
These tests are not ieee-854 based, but base 10. For arctanh, for
example, the test value 0.999999 really is supposed to be fraction
999999/1000000 and not whatever that rounds to as a double. That
reference value reflects that.
Testing is important. Cases in point: crlibm had a crazy return value
for expm1( 2 * -2.079401950488110); glibc 2.23 was released with a
broken cos function.
Morten
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
