BZ #15536: Fix ulp for 128-bit IBM long double.
authorCarlos O'Donell <carlos@redhat.com>
Mon, 3 Jun 2013 18:49:48 +0000 (14:49 -0400)
committerCarlos O'Donell <carlos@redhat.com>
Mon, 3 Jun 2013 18:49:48 +0000 (14:49 -0400)
commit8b0ccb2d7fd1ec646a622a16bd64e356739ffca3
treeb7500d0f98c33e9fa91366fc95ef456eff692d15
parent3b3c4d40c18e0e340c68a02487014d0001211382
BZ #15536: Fix ulp for 128-bit IBM long double.

In 128-bit IBM long double the precision of the type
decreases as you approach subnormal numbers, equaling
that of a double for subnormal numbers. Therefore
adjust the computation in ulp to use 2^(MIN_EXP - MANT_DIG)
which is correct for FP_SUBNORMAL for all types.
ChangeLog
NEWS
math/libm-test.inc