This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Update sparc ULPs.
On Mon, 24 Sep 2012, David Miller wrote:
> I'll do more work today to spot the location of the underflow.
My guess is it's the computation x * x, as I said.
If x is small enough (absolute value less than LDBL_EPSILON / 2.0L, I
think) then the correct rounded-to-nearest result of log1pl (x) is just x,
so code can test for that case and avoid the more complicated calculations
with potential underflow (while taking care to generate underflow if
correct). Properly, that's x with inexact exception (if not 0), and
underflow exception if x is subnormal (or, on tinyness-before-rounding
machines, if x is +LDBL_MIN, but that detail isn't something I'd expect
glibc to implement).
--
Joseph S. Myers
joseph@codesourcery.com