[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

work around inaccurate 'long double' math functions on NetBSD

From: Bruno Haible
Subject: work around inaccurate 'long double' math functions on NetBSD
Date: Mon, 21 Jan 2019 00:39:51 +0100
User-agent: KMail/5.1.3 (Linux/4.4.0-141-generic; KDE/5.18.0; x86_64; ; )

On NetBSD 8.0 (x86_64, x86, and sparc64) I'm seeing test failures:

FAIL: test-exp2l
FAIL: test-expl
FAIL: test-expm1l
FAIL: test-log10l
FAIL: test-log1pl
FAIL: test-log2l
FAIL: test-logl

The results have apparently only 54 bits of precision, even though
'long double' has 64 or 113 bits of mantissa, respectively.

Also this failure, due to wrong results for negative arguments.

FAIL: test-rintl

This series of patches adds workarounds, so that all these tests pass.


Attachment: 0001-exp2l-Work-around-inaccurate-implementation-on-NetBS.patch
Description: Text Data

Attachment: 0002-expl-Work-around-inaccurate-implementation-on-NetBSD.patch
Description: Text Data

Attachment: 0003-expm1l-Work-around-inaccurate-implementation-on-NetB.patch
Description: Text Data

Attachment: 0004-logl-Work-around-inaccurate-implementation-on-NetBSD.patch
Description: Text Data

Attachment: 0005-log10l-Work-around-inaccurate-implementation-on-NetB.patch
Description: Text Data

Attachment: 0006-rintl-Override-broken-implementation-on-NetBSD.patch
Description: Text Data

reply via email to

[Prev in Thread] Current Thread [Next in Thread]