Re: [Patch, libquadmath, committed] fix signbit call

[Tobias Burnus <burnus at net-b dot de>]

Tilo Schwarz wrote:
> On Sun, 17 Feb 2013 14:18:02 +0100, Tobias Burnus <burnus@net-b.de> 
> wrote:
> > I have now committed the attached patch (as Rev. 196109).
>
> Thank you!

Well, I have also to thank you for finding and reporting this bug.

> Now I get with both configurations
>
> ../configure --disable-multilib --enable-languages=fortran
> --disable-bootstrap
>
> and
>
> ../configure --disable-multilib --enable-languages=fortran
> --disable-bootstrap CFLAGS="-g -O0" CXXFLAGS="-g -O0" 
> CFLAGS_FOR_BUILD="-g
> -O0" CFLAGS_FOR_TARGET="-g -O0" FCFLAGS="-g -O0"
>
> almost the same results for
>
> make check-fortran
>
> But one test, gfortran.dg/quad_2.f90 fails for the 2nd config:
>
> It seems to be this line:
> 34    if (abs (fp2 - fp4)/fp2 > epsilon(fp2)) call abort()

There is something wrong with the string-to-real(16) conversion for -O0. 
As I couldn't quickly spot the problem, I have filled a bug to track it: 
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56379

Tobias