Embed root node into rtree_t.

This avoids one atomic operation per tree access.

include/jemalloc/internal/rtree_structs.h

@@ -39,13 +39,17 @@ struct rtree_ctx_s {
 #ifndef _MSC_VER
 	JEMALLOC_ALIGNED(CACHELINE)
 #endif
-	rtree_ctx_cache_elm_t cache[RTREE_CTX_NCACHE];
+	rtree_ctx_cache_elm_t	cache[RTREE_CTX_NCACHE];
 };
 
 struct rtree_s {
-	/* An rtree_{internal,leaf}_elm_t *. */
-	atomic_p_t	root;
-	malloc_mutex_t	init_lock;
+	malloc_mutex_t		init_lock;
+	/* Number of elements based on rtree_levels[0].bits. */
+#if RTREE_HEIGHT > 1
+	rtree_node_elm_t	root[1U << (RTREE_NSB/RTREE_HEIGHT)];
+#else
+	rtree_leaf_elm_t	root[1U << (RTREE_NSB/RTREE_HEIGHT)];
+#endif
 };
 
 #endif /* JEMALLOC_INTERNAL_RTREE_STRUCTS_H */