Fix ldbl-128ibm logl inaccuracy (bug 16386).
authorJoseph Myers <joseph@codesourcery.com>
Thu, 2 Jan 2014 16:35:46 +0000 (16:35 +0000)
committerJoseph Myers <joseph@codesourcery.com>
Thu, 2 Jan 2014 16:35:46 +0000 (16:35 +0000)
commit819e5d50dd4d10dc359037eba74c70e74cb42739
treefe51e811c4c6bb9e91bf2ab3730d34aa8c03452c
parent90b6a1e55fdbb3e6a3132742c26eb19666ad9e33
Fix ldbl-128ibm logl inaccuracy (bug 16386).

This patch fixes bug 16386, ldbl-128ibm logl inaccuracy (with
consequent inaccuracy for lgammal) for arguments where the high double
is subnormal, which showed up while attempting to regenerate ulps for
powerpc-nofpu for 2.19.  The problem here is logic failing to allow
for subnormals when calculating the exponent of the argument.  Tested
for powerpc-nofpu.

* sysdeps/ieee754/ldbl-128ibm/e_logl.c (__ieee754_logl): Adjust
numbers with subnormal high part when calculating exponent.
ChangeLog
NEWS
sysdeps/ieee754/ldbl-128ibm/e_logl.c