[patch] Fix std::notify_all_at_thread_exit test for older glibc

[Jonathan Wakely]

I'm seeing this test timeout on glibc 2.13, which I think is because
it doesn't provide __cxa_atexit_thread_impl and so the thread_local
destructor and the notify_all() happen in an unspecified order.

Tested x86_64-linux, committed to trunk.

commit 299704c621dd7afaee7c5fb2a354b40ef41c2eba
Author: Jonathan Wakely <jwak...@redhat.com>
Date:   Thu Dec 11 17:12:17 2014 +0000

    	* testsuite/30_threads/condition_variable/members/3.cc: Only use
    	a thread_local when __cxa_thread_atexit_impl is available.

diff --git a/libstdc++-v3/testsuite/30_threads/condition_variable/members/3.cc b/libstdc++-v3/testsuite/30_threads/condition_variable/members/3.cc
index 0da545d..1788bcf 100644
--- a/libstdc++-v3/testsuite/30_threads/condition_variable/members/3.cc
+++ b/libstdc++-v3/testsuite/30_threads/condition_variable/members/3.cc
@@ -41,7 +41,12 @@ void func()
 {
   std::unique_lock<std::mutex> lock{mx};
   std::notify_all_at_thread_exit(cv, std::move(lock));
+#if _GLIBCXX_HAVE___CXA_THREAD_ATEXIT_IMPL
+  // Correct order of thread_local destruction needs __cxa_thread_atexit_impl
   static thread_local Inc inc;
+#else
+  Inc inc;
+#endif
 }
 
 int main()