|
|
Message-Id: <mwk0sona13.fsf@tomate.loria.fr>
Date: Thu, 07 Jan 2021 14:36:56 +0100
From: Paul Zimmermann <Paul.Zimmermann@...ia.fr>
To: musl@...ts.openwall.com
Subject: large errors with long double functions
Hi,
I notice large errors for some long double functions (on x86_64) with
musl-1.2.1:
expm1l 0x2.c516edaf5e96a33cp+12l 9700.65
lgammal -0x2.74ff92c01f0d82acp+0l 9.08e+19
sinhl 0x2.c516edaf5e96a33cp+12l 9700.65
tgammal -0x2.8d19fd20f3aa62cp+4l 3.69e+19
powl 0xd.64229ca81fffep+16368l,-0x1.b9b6d40000000002p-48l 532.341
The lgammal/tgammal errors are not surprising, but those for expm1l and sinhl
are, and the fact that the input/error are the same for expm1l and sinhl
suggest that one function is using the other one, and maybe that latter one
only provides double precision accuracy.
For powl also it is surprising, since other libraries typically yield
errors < 1 ulp.
Best regards,
Paul
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.