[Bug libstdc++/58800] New: [4.8 regression, I think] std::nth_element segfaults on valid input

["luto at mit dot edu" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58800

            Bug ID: 58800
           Summary: [4.8 regression, I think] std::nth_element segfaults
                    on valid input
           Product: gcc
           Version: 4.8.1
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: libstdc++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: luto at mit dot edu

This program segfaults on gcc 4.8.1 from Ubuntu 13.10.  I'm building a copy of
the 4.8 branch to see if I can reproduce it there.

This is on x86_64.  I used a gross hack to specify the input so I don't have to
think about precision issues.  The numbers are all well-behaved values near
0.9.

I'll attach preprocessed source.  The preprocessed source crashes even if built
with gcc 4.7.something.

#include <algorithm>
#include <stdint.h>
//#include <iostream>

double to_double(uint64_t x)
{
        union {double d; uint64_t x;} u;
        u.x = x;
        return u.d;
}

int main()
{
        std::vector<double> v = {
                to_double(4606672070890243784),
                to_double(4606672025854247510),
                to_double(4606671800674266141),
                to_double(4606671575494284772),
                to_double(4606672115926240057),
                to_double(4606672160962236330),
                to_double(4606672070890243784),
        };

//        for (auto i : v)
//                std::cout << i << std::endl;

        std::nth_element(v.begin(), v.begin() + 3, v.end());
        return 0;
}