This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] [RFC] Change SET_RESTORE_ROUND to just set the roundingmode
On Thu, 14 Mar 2013, Adhemerval Zanella wrote:
> Hi Joseph,
>
> I have opened bug BZ#15268. I couldn't find a easy way to add a testcase in the
> existing math mainly because although the exp call underflow, the result is
> not inexact and the underflow mask is set on test-* cases to not trap. I open
> to suggestion on how to add a testcase or if is really need to add one.
The final exp result certainly isn't exact. There may be two separate
bugs here: missing underflow exceptions when the internal computation that
underflows is exact although the final result isn't, and missing underflow
exceptions in both cases on powerpc.
Maybe you can find an input that *does* raise the underflow flag on other
architectures, because the internal calculation is inexact, but does not
do so on powerpc, and use that in the testcase in libm-test.inc? Then
file a separate bug for the case where the internal calculation is exact
so the underflow flag doesn't get raised (in the absence of trapping on
underflow) on any architecture?
--
Joseph S. Myers
joseph@codesourcery.com