This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Fix math functions triggering invalid exceptions


On Wed, Feb 15, 2012 at 03:35:28PM -0600, Ryan S. Arnold wrote:
> On Wed, Feb 15, 2012 at 3:29 PM, Aurelien Jarno <aurelien@aurel32.net> wrote:
> > On Wed, Feb 15, 2012 at 08:55:52PM +0000, Joseph S. Myers wrote:
> ...
> >> So I think you should repost with those fixes, and then commit if noone
> >> else spots any problems and if it's been tested on both an affected
> >> platform and one that didn't previously have the issue you report.
> >>
> >
> > Thanks for the review. Here is the new patch.
> >
> >
> > 2012-02-15 ?Aurelien Jarno ?<aurelien@aurel32.net>
> >
> > ? ? ? ?* math/w_acos.c: Use non-signaling floating-point comparisons.
> > ? ? ? ?* math/w_acosf.c: Likewise.
> > ? ? ? ?* math/w_acosh.c: Likewise.
> > ? ? ? ?* math/w_acoshf.c: Likewise.
> > ? ? ? ?* math/w_acoshl.c: Likewise.
> > ? ? ? ?* math/w_acosl.c: Likewise.
> > ? ? ? ?* math/w_asin.c: Likewise.
> > ? ? ? ?* math/w_asinf.c: Likewise.
> > ? ? ? ?* math/w_asinl.c: Likewise.
> > ? ? ? ?* math/w_atanh.c: Likewise.
> > ? ? ? ?* math/w_atanhf.c: Likewise.
> > ? ? ? ?* math/w_atanhl.c: Likewise.
> > ? ? ? ?* math/w_exp2.c: Likewise.
> > ? ? ? ?* math/w_exp2f.c: Likewise.
> > ? ? ? ?* math/w_exp2l.c: Likewise.
> > ? ? ? ?* math/w_j0.c: Likewise.
> > ? ? ? ?* math/w_j0f.c: Likewise.
> > ? ? ? ?* math/w_j0l.c: Likewise.
> > ? ? ? ?* math/w_j1.c: Likewise.
> > ? ? ? ?* math/w_j1f.c: Likewise.
> > ? ? ? ?* math/w_j1l.c: Likewise.
> > ? ? ? ?* math/w_jn.c: Likewise.
> > ? ? ? ?* math/w_jnf.c: Likewise.
> > ? ? ? ?* math/w_log.c: Likewise.
> > ? ? ? ?* math/w_log10.c: Likewise.
> > ? ? ? ?* math/w_log10f.c: Likewise.
> > ? ? ? ?* math/w_log10l.c: Likewise.
> > ? ? ? ?* math/w_log2.c: Likewise.
> > ? ? ? ?* math/w_log2f.c: Likewise.
> > ? ? ? ?* math/w_log2l.c: Likewise.
> > ? ? ? ?* math/w_logf.c: Likewise.
> > ? ? ? ?* math/w_logl.c: Likewise.
> > ? ? ? ?* math/w_sqrt.c: Likewise.
> > ? ? ? ?* math/w_sqrtf.c: Likewise.
> > ? ? ? ?* math/w_sqrtl.c: Likewise.
> > ? ? ? ?* sysdeps/ieee754/dbl-64/e_atanh.c: Likewise.
> > ? ? ? ?* sysdeps/ieee754/dbl-64/w_exp.c: Likewise.
> > ? ? ? ?* sysdeps/ieee754/flt-32/e_atanhf.c: Likewise.
> > ? ? ? ?* sysdeps/ieee754/flt-32/w_expf.c: Likewise.
> > ? ? ? ?* sysdeps/ieee754/ldbl-96/w_expl.c: Likewise.
> 
> Please hold up on checking this in.  We'd like some time to
> investigate the performance implications of these additional
> comparisons which very well may be unnecessary on Power.
> 

I haven't look at the generated code, but at the assembly level it
should simply replace the fcmpo instruction by fmcpu. I doubt it
will make any change on the performance side.

-- 
Aurelien Jarno	                        GPG: 1024D/F1BCDB73
aurelien@aurel32.net                 http://www.aurel32.net


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]