[libc++] Optimize multi{map,set}::insert

Author: philnik777

libcxx/include/__tree

@@ -970,6 +970,35 @@ public:
     __emplace_hint_multi(__p, std::move(__value));
   }
 
+  template <class _InIter, class _Sent>
+  _LIBCPP_HIDE_FROM_ABI void __insert_range_multi(_InIter __first, _Sent __last) {
+    if (__first == __last)
+      return;
+
+    if (__root() == nullptr) { // Make sure we always have a root node
+      __node_holder __node = __construct_node(*__first);
+      __insert_node_at(__end_node(), __end_node()->__left_, static_cast<__node_base_pointer>(__node.release()));
+      ++__first;
+    }
+
+    auto __max_node = static_cast<__node_pointer>(std::__tree_max(static_cast<__node_base_pointer>(__root())));
+
+    for (; __first != __last; ++__first) {
+      __node_holder __node = __construct_node(*__first);
+      // Always check the max node first. This optimizes for sorted ranges inserted at the end.
+      if (!value_comp()(__node->__get_value(), __max_node->__get_value())) { // __node >= __max_val
+        __insert_node_at(static_cast<__end_node_pointer>(__max_node),
+                         __max_node->__right_,
+                         static_cast<__node_base_pointer>(__node.get()));
+        __max_node = __node.release();
+      } else {
+        __end_node_pointer __parent;
+        __node_base_pointer& __child = __find_leaf_high(__parent, __node->__value_);
+        __insert_node_at(__parent, __child, static_cast<__node_base_pointer>(__node.release()));
+      }
+    }
+  }
+
   _LIBCPP_HIDE_FROM_ABI pair<iterator, bool> __node_assign_unique(const value_type& __v, __node_pointer __dest);
 
   _LIBCPP_HIDE_FROM_ABI iterator __node_insert_multi(__node_pointer __nd);

libcxx/include/map

@@ -593,6 +593,7 @@ erase_if(multimap<Key, T, Compare, Allocator>& c, Predicate pred);  // C++20
 #  include <__memory/unique_ptr.h>
 #  include <__memory_resource/polymorphic_allocator.h>
 #  include <__node_handle>
+#  include <__ranges/access.h>
 #  include <__ranges/concepts.h>
 #  include <__ranges/container_compatible_range.h>
 #  include <__ranges/from_range.h>
@@ -1749,17 +1750,13 @@ public:
 
   template <class _InputIterator>
   _LIBCPP_HIDE_FROM_ABI void insert(_InputIterator __f, _InputIterator __l) {
-    for (const_iterator __e = cend(); __f != __l; ++__f)
-      __tree_.__emplace_hint_multi(__e.__i_, *__f);
+    __tree_.__insert_range_multi(__f, __l);
   }
 
 #  if _LIBCPP_STD_VER >= 23
   template <_ContainerCompatibleRange<value_type> _Range>
   _LIBCPP_HIDE_FROM_ABI void insert_range(_Range&& __range) {
-    const_iterator __end = cend();
-    for (auto&& __element : __range) {
-      __tree_.__emplace_hint_multi(__end.__i_, std::forward<decltype(__element)>(__element));
-    }
+    __tree_.__insert_range_multi(ranges::begin(__range), ranges::end(__range));
   }
 #  endif
 

libcxx/include/set

@@ -530,6 +530,7 @@ erase_if(multiset<Key, Compare, Allocator>& c, Predicate pred);  // C++20
 #  include <__memory/allocator_traits.h>
 #  include <__memory_resource/polymorphic_allocator.h>
 #  include <__node_handle>
+#  include <__ranges/access.h>
 #  include <__ranges/concepts.h>
 #  include <__ranges/container_compatible_range.h>
 #  include <__ranges/from_range.h>
@@ -1205,18 +1206,14 @@ public:
   }
 
   template <class _InputIterator>
-  _LIBCPP_HIDE_FROM_ABI void insert(_InputIterator __f, _InputIterator __l) {
-    for (const_iterator __e = cend(); __f != __l; ++__f)
-      __tree_.__emplace_hint_multi(__e, *__f);
+  _LIBCPP_HIDE_FROM_ABI void insert(_InputIterator __first, _InputIterator __last) {
+    __tree_.__insert_range_multi(__first, __last);
   }
 
 #  if _LIBCPP_STD_VER >= 23
   template <_ContainerCompatibleRange<value_type> _Range>
   _LIBCPP_HIDE_FROM_ABI void insert_range(_Range&& __range) {
-    const_iterator __end = cend();
-    for (auto&& __element : __range) {
-      __tree_.__emplace_hint_multi(__end, std::forward<decltype(__element)>(__element));
-    }
+    __tree_.__insert_range_multi(ranges::begin(__range), ranges::end(__range));
   }
 #  endif