diff --git a/opal/include/opal/sys/amd64/atomic.h b/opal/include/opal/sys/amd64/atomic.h
index cf8d099bad..39cc3f0c4f 100644
--- a/opal/include/opal/sys/amd64/atomic.h
+++ b/opal/include/opal/sys/amd64/atomic.h
@@ -44,6 +44,10 @@
 
 #define OPAL_HAVE_ATOMIC_CMPSET_64 1
 
+#define OPAL_HAVE_ATOMIC_SWAP_32 1
+
+#define OPAL_HAVE_ATOMIC_SWAP_64 1
+
 /**********************************************************************
  *
  * Memory Barriers
@@ -119,6 +123,40 @@ static inline int opal_atomic_cmpset_64( volatile int64_t *addr,
 #define opal_atomic_cmpset_acq_64 opal_atomic_cmpset_64
 #define opal_atomic_cmpset_rel_64 opal_atomic_cmpset_64
 
+#if OMPI_GCC_INLINE_ASSEMBLY
+
+static inline int32_t opal_atomic_swap_32( volatile int32_t *addr,
+					   int32_t newval)
+{
+    int32_t oldval;
+
+    __asm__ __volatile__("xchg %0, %1" :
+			 "=r" (oldval), "=m" (*addr) :
+			 "0" (newval), "m" (*addr) :
+			 "memory");
+    return oldval;
+}
+
+#endif /* OMPI_GCC_INLINE_ASSEMBLY */
+
+#if OMPI_GCC_INLINE_ASSEMBLY
+
+static inline int64_t opal_atomic_swap_64( volatile int64_t *addr,
+                                           int64_t newval)
+{
+    int64_t oldval;
+
+    __asm__ __volatile__("xchgq %1, %0" :
+			 "=r" (oldval) :
+			 "m" (*addr), "0" (newval) :
+			 "memory");
+    return oldval;
+}
+
+#endif /* OMPI_GCC_INLINE_ASSEMBLY */
+
+
+
 #if OMPI_GCC_INLINE_ASSEMBLY
 
 #define OPAL_HAVE_ATOMIC_MATH_32 1
diff --git a/opal/include/opal/sys/atomic.h b/opal/include/opal/sys/atomic.h
index 4911f0c259..01e83668d3 100644
--- a/opal/include/opal/sys/atomic.h
+++ b/opal/include/opal/sys/atomic.h
@@ -121,6 +121,8 @@ typedef struct opal_atomic_lock_t opal_atomic_lock_t;
 #define OPAL_HAVE_INLINE_ATOMIC_SUB_32 0
 #define OPAL_HAVE_INLINE_ATOMIC_ADD_64 0
 #define OPAL_HAVE_INLINE_ATOMIC_SUB_64 0
+#define OPAL_HAVE_INLINE_ATOMIC_SWAP_32 0
+#define OPAL_HAVE_INLINE_ATOMIC_SWAP_64 0
 #else
 #define OPAL_HAVE_INLINE_ATOMIC_MEM_BARRIER 1
 #define OPAL_HAVE_INLINE_ATOMIC_CMPSET_32 1
@@ -129,6 +131,8 @@ typedef struct opal_atomic_lock_t opal_atomic_lock_t;
 #define OPAL_HAVE_INLINE_ATOMIC_SUB_32 1
 #define OPAL_HAVE_INLINE_ATOMIC_ADD_64 1
 #define OPAL_HAVE_INLINE_ATOMIC_SUB_64 1
+#define OPAL_HAVE_INLINE_ATOMIC_SWAP_32 1
+#define OPAL_HAVE_INLINE_ATOMIC_SWAP_64 1
 #endif
 
 /**********************************************************************
@@ -561,6 +565,16 @@ static inline int opal_atomic_cmpset_rel_ptr(volatile void* addr,
 
 #endif /* (OPAL_HAVE_ATOMIC_CMPSET_32 || OPAL_HAVE_ATOMIC_CMPSET_64) */
 
+#if defined(DOXYGEN) || (OPAL_HAVE_ATOMIC_SWAP_32 || OPAL_HAVE_ATOMIC_SWAP_64)
+
+#if SIZEOF_VOID_P == 4 && OPAL_HAVE_ATOMIC_SWAP_32
+#define opal_atomic_swap_ptr(addr, value) opal_atomic_swap_32((int32_t *) addr, value)
+#elif SIZEOF_VOID_P == 8 && OPAL_HAVE_ATOMIC_SWAP_64
+#define opal_atomic_swap_ptr(addr, value) opal_atomic_swap_64((int64_t *) addr, value)
+#endif
+
+#endif /* (OPAL_HAVE_ATOMIC_SWAP_32 || OPAL_HAVE_ATOMIC_SWAP_64) */
+
 #if defined(DOXYGEN) || (OPAL_HAVE_ATOMIC_MATH_32 || OPAL_HAVE_ATOMIC_MATH_64)
 
 static inline void opal_atomic_add_xx(volatile void* addr,