This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Enable autoinlining for size at O2
On Thu, Sep 06, 2007 at 01:33:59PM +0200, Uros Bizjak wrote:
> On 9/5/07, Uros Bizjak <ubizjak@gmail.com> wrote:
>
> > > The problem does not reproduce for me (my notebook is SSE3 but for some
> > > reason the testcase works just fine in both 32bit and 64bit. One place
> > > that strikes wrong is the aligned access. Can you please check if the
> > > attached patch fix the problem for you and if not how the testcase is
> > > failing?
> > >
> >
> > The problem is in infamous x87 excess precision on ilp32 target. For
> > example, in sse3-addsubpd.c, we have a function that directly compares
> > two doubles for inequality:
>
> This patch adds -mfpmath=sse to all sse3-* tests that could be
> affected by excess x87 precision.
>
> Patch was tested on i686-pc-linux-gnu and is committed to mainline SVN.
>
> 2007-09-06 Uros Bizjak <ubizjak@gmail.com>
>
> * gcc.target/i386/sse3-addsubpd.c: Add -mfpmath=sse to dg-options.
> * gcc.target/i386/sse3-addsubps.c: Ditto.
> * gcc.target/i386/sse3-haddpd.c: Ditto.
> * gcc.target/i386/sse3-haddps.c: Ditto.
> * gcc.target/i386/sse3-hsubpd.c: Ditto.
> * gcc.target/i386/sse3-hsubps.c: Ditto.
> * gcc.target/i386/sse3-lddqu.c: Ditto.
> * gcc.target/i386/sse3-movddup.c: Ditto.
> * gcc.target/i386/sse3-movshdup.c: Ditto.
> * gcc.target/i386/sse3-movsldup.c: Ditto.
>
That looks good to me.
Thanks.
H.J.