Mark all member functions with memory models always inline v2

When a non constant memory model is passed to __atomic_*
gcc falls back to seq_cst. This drops any HLE acquire or release bits.

This can happen when <atomic> is used with -O0
as the member functions are not always inlined then and the memory
argument passed in ends up being non-constant.

v2: Use _GLIBCXX_ALWAYS_INLINE

libstdc++-v3/:

2013-05-08  Andi Kleen  <ak@linux.intel.com>

	PR target/55947
	* libstdc++-v3/include/bits/atomic_base.h
	(_GLIBCXX_ALWAYS_INLINE): Add new macro.
	(atomic_thread_fence, atomic_signal_fence, test_and_set,
	clear, store, load, exchange, compare_exchange_weak)
	compare_exchange_strong, fetch_add, fetch_sub, fetch_and,
	fetch_or, fetch_xor): Mark _GLIBCXX_ALWAYS_INLINE.

From-SVN: r198733

libstdc++-v3/ChangeLog

+2013-05-08  Andi Kleen  <ak@linux.intel.com>
+
+	PR target/55947
+	* libstdc++-v3/include/bits/atomic_base.h
+	(_GLIBCXX_ALWAYS_INLINE): Add new macro.
+	(atomic_thread_fence, atomic_signal_fence, test_and_set,
+	clear, store, load, exchange, compare_exchange_weak)
+	compare_exchange_strong, fetch_add, fetch_sub, fetch_and,
+	fetch_or, fetch_xor): Mark _GLIBCXX_ALWAYS_INLINE.
+
 2013-05-08  Jason Merrill  <jason@redhat.com>
 
 	Add std::bad_array_new_length (N2932)