[Bug fortran/14396] New: [gfortran] Intrinsic SQRT and CSQRT give wrong answer with COMPLEX argument.

["bdavis9659 at comcast dot net" <gcc-bugzilla at gcc dot gnu dot org>]

Distilled from test FM833 of the NIST F77 compiler validation suite.

        COMPLEX BVC,AVC
        BVC = (3.0, 4.0)
        AVC = CSQRT(BVC) - SQRT(BVC)
        print*,'CSQRT(BVC)=',CSQRT(BVC),' SQRT(BVC)=',SQRT(BVC)
        print*,'CSQRT(BVC) - SQRT(BVC)',avc
        end
 
 
$ /usr/local/bin/gfortran -static z.f
$ ./a.out
CSQRT(BVC)= ( NaN,   0.000000    )  SQRT(BVC)= ( NaN,   0.000000    )
CSQRT(BVC) - SQRT(BVC) ( NaN,   0.000000    )


Same code with g77:

$ g77 z.f
$ ./a.out
 CSQRT(BVC)= (2.,1.) SQRT(BVC)= (2.,1.)
 CSQRT(BVC) - SQRT(BVC) (0.,0.)

-- 
           Summary: [gfortran] Intrinsic SQRT and CSQRT give wrong answer
                    with COMPLEX argument.
           Product: gcc
           Version: tree-ssa
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: bdavis9659 at comcast dot net
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14396