This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: Build all of glibc with -frounding-math
- From: "Joseph S. Myers" <joseph at codesourcery dot com>
- To: Andreas Jaeger <aj at suse dot de>
- Cc: libc-alpha at sources dot redhat dot com
- Date: Fri, 4 May 2012 10:51:45 +0000 (UTC)
- Subject: Re: Build all of glibc with -frounding-math
- References: <87zk9owdp0.fsf@linux-6dv9.site>
On Fri, 4 May 2012, Andreas Jaeger wrote:
> On x86-64 I got these new ULPs with test-float (and ifloat):
I think ulps updates are fine here, probably disabling an optimization
that was invalid for -frounding-math has just happened to produce slightly
different results here. (Since the decimal expected values in
libm-test.inc are always interpreted by the compiler in round-to-nearest
mode, 1ulp errors in directed-rounding tests like this aren't really
meaningful anyway.)
> 2012-05-04 Andreas Jaeger <aj@suse.de>
>
> * Makeconfig (+math-flags): New, set to -frounding-math.
> (+cflags): Add +math-flags so that all of glibc gets compiled with
> it.
I think this should go in - but since it does apply to the build of every
file in glibc, it would be a good idea to wait a few days to see if either
anyone objects, or other people also agree it should go in.
Once it's in I'll see if bug 10846 is fixed - and if so I may look at some
point at supporting tests for underflow exceptions in libm-test.inc so we
can add a test for that bug and see what the general state of giving
underflow exceptions from libm is like.
--
Joseph S. Myers
joseph@codesourcery.com