[libc++] Introduce __product_iterator_traits and optimise flat_map::insert (#139454)

Fixes #108624

This allows `flat_map::insert(Iter, Iter)` to directly forward to
underlying containers' `insert(Iter, Iter)`, instead of inserting one
element at a time, when input models "product iterator". atm,
`flat_map::iterator` and `zip_view::iterator` are "product iterator"s.

This gives about almost 10x speed up in my benchmark with -03 (for both
before and after)

```cpp
Benchmark                                                          Time             CPU      Time Old      Time New       CPU Old       CPU New
-----------------------------------------------------------------------------------------------------------------------------------------------
flat_map::insert_product_iterator_flat_map/32                   -0.5028         -0.5320           149            74           149            70
flat_map::insert_product_iterator_flat_map/1024                 -0.8617         -0.8618          3113           430          3112           430
flat_map::insert_product_iterator_flat_map/8192                 -0.8877         -0.8877         26682          2995         26679          2995
flat_map::insert_product_iterator_flat_map/65536                -0.8769         -0.8769        226235         27844        226221         27841
flat_map::insert_product_iterator_zip/32                        -0.5844         -0.5844           162            67           162            67
flat_map::insert_product_iterator_zip/1024                      -0.8754         -0.8754          3427           427          3427           427
flat_map::insert_product_iterator_zip/8192                      -0.8934         -0.8934         28134          3000         28132          3000
flat_map::insert_product_iterator_zip/65536                     -0.8783         -0.8783        229783         27960        229767         27958
OVERALL_GEOMEAN                                                 -0.8319         -0.8332             0             0             0             0
```

---------

Co-authored-by: Louis Dionne <[email protected]>

Author: huixie90

libcxx/docs/ReleaseNotes/21.rst

@@ -81,6 +81,9 @@ Improvements and New Features
 - The ``bitset::to_string`` function has been optimized, resulting in a performance improvement of up to 8.3x for bitsets
   with uniformly distributed zeros and ones, and up to 13.5x and 16.1x for sparse and dense bitsets, respectively.
 
+- The ``flat_map::insert`` and ``flat_set::insert_range`` have been optimized, resulting in a performance improvement of up
+  to 10x for inserting elements into a ``flat_map`` when the input range is a ``flat_map`` or a ``zip_view``.
+
 Deprecations and Removals
 -------------------------
 

libcxx/include/CMakeLists.txt

@@ -498,6 +498,7 @@ set(files
   __iterator/ostreambuf_iterator.h
   __iterator/permutable.h
   __iterator/prev.h
+  __iterator/product_iterator.h
   __iterator/projected.h
   __iterator/ranges_iterator_traits.h
   __iterator/readable_traits.h

libcxx/include/__flat_map/key_value_iterator.h

@@ -13,9 +13,12 @@
 #include <__compare/three_way_comparable.h>
 #include <__concepts/convertible_to.h>
 #include <__config>
+#include <__cstddef/size_t.h>
 #include <__iterator/iterator_traits.h>
+#include <__iterator/product_iterator.h>
 #include <__memory/addressof.h>
 #include <__type_traits/conditional.h>
+#include <__utility/forward.h>
 #include <__utility/move.h>
 #include <__utility/pair.h>
 
@@ -57,6 +60,8 @@ struct __key_value_iterator {
   template <class, class, class, bool>
   friend struct __key_value_iterator;
 
+  friend struct __product_iterator_traits<__key_value_iterator>;
+
 public:
   using iterator_concept = random_access_iterator_tag;
   // `__key_value_iterator` only satisfy "Cpp17InputIterator" named requirements, because
@@ -181,6 +186,29 @@ struct __key_value_iterator {
   }
 };
 
+template <class _Owner, class _KeyContainer, class _MappedContainer, bool _Const>
+struct __product_iterator_traits<__key_value_iterator<_Owner, _KeyContainer, _MappedContainer, _Const>> {
+  static constexpr size_t __size = 2;
+
+  template <size_t _Nth, class _Iter>
+  _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX26 static decltype(auto) __get_iterator_element(_Iter&& __it)
+    requires(_Nth <= 1)
+  {
+    if constexpr (_Nth == 0) {
+      return std::forward<_Iter>(__it).__key_iter_;
+    } else {
+      return std::forward<_Iter>(__it).__mapped_iter_;
+    }
+  }
+
+  template <class _KeyIter, class _MappedIter>
+  _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX26 static auto
+  __make_product_iterator(_KeyIter&& __key_iter, _MappedIter&& __mapped_iter) {
+    return __key_value_iterator<_Owner, _KeyContainer, _MappedContainer, _Const>(
+        std::forward<_KeyIter>(__key_iter), std::forward<_MappedIter>(__mapped_iter));
+  }
+};
+
 _LIBCPP_END_NAMESPACE_STD
 
 #endif // _LIBCPP_STD_VER >= 23

libcxx/include/__flat_map/utils.h

@@ -11,6 +11,7 @@
 #define _LIBCPP___FLAT_MAP_UTILS_H
 
 #include <__config>
+#include <__iterator/product_iterator.h>
 #include <__type_traits/container_traits.h>
 #include <__utility/exception_guard.h>
 #include <__utility/forward.h>
@@ -79,8 +80,6 @@ struct __flat_map_utils {
     return typename _Map::iterator(std::move(__key_it), std::move(__mapped_it));
   }
 
-  // TODO: We could optimize this, see
-  // https://github.com/llvm/llvm-project/issues/108624
   template <class _Map, class _InputIterator, class _Sentinel>
   _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX26 static typename _Map::size_type
   __append(_Map& __map, _InputIterator __first, _Sentinel __last) {
@@ -93,6 +92,25 @@ struct __flat_map_utils {
     }
     return __num_appended;
   }
+
+  template <class _Map, class _InputIterator>
+  _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX26 static typename _Map::size_type
+  __append(_Map& __map, _InputIterator __first, _InputIterator __last)
+    requires __is_product_iterator_of_size<_InputIterator, 2>::value
+  {
+    auto __s1 = __map.__containers_.keys.size();
+    __map.__containers_.keys.insert(
+        __map.__containers_.keys.end(),
+        __product_iterator_traits<_InputIterator>::template __get_iterator_element<0>(__first),
+        __product_iterator_traits<_InputIterator>::template __get_iterator_element<0>(__last));
+
+    __map.__containers_.values.insert(
+        __map.__containers_.values.end(),
+        __product_iterator_traits<_InputIterator>::template __get_iterator_element<1>(__first),
+        __product_iterator_traits<_InputIterator>::template __get_iterator_element<1>(__last));
+
+    return __map.__containers_.keys.size() - __s1;
+  }
 };
 _LIBCPP_END_NAMESPACE_STD
 

libcxx/include/__iterator/product_iterator.h

@@ -0,0 +1,76 @@
+//===----------------------------------------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef _LIBCPP___ITERATOR_PRODUCT_ITERATOR_H
+#define _LIBCPP___ITERATOR_PRODUCT_ITERATOR_H
+
+// Product iterators are iterators that contain two or more underlying iterators.
+//
+// For example, std::flat_map stores its data into two separate containers, and its iterator
+// is a proxy over two separate underlying iterators. The concept of product iterators
+// allows algorithms to operate over these underlying iterators separately, opening the
+// door to various optimizations.
+//
+// If __product_iterator_traits can be instantiated, the following functions and associated types must be provided:
+// - static constexpr size_t Traits::__size
+//   The number of underlying iterators inside the product iterator.
+//
+// - template <size_t _N>
+//   static decltype(auto) Traits::__get_iterator_element(It&& __it)
+//   Returns the _Nth iterator element of the given product iterator.
+//
+// - template <class... _Iters>
+//   static _Iterator __make_product_iterator(_Iters&&...);
+//   Creates a product iterator from the given underlying iterators.
+
+#include <__config>
+#include <__cstddef/size_t.h>
+#include <__type_traits/enable_if.h>
+#include <__type_traits/integral_constant.h>
+#include <__utility/declval.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+template <class _Iterator>
+struct __product_iterator_traits;
+/* exposition-only:
+{
+  static constexpr size_t __size = ...;
+
+  template <size_t _N, class _Iter>
+  static decltype(auto) __get_iterator_element(_Iter&&);
+
+  template <class... _Iters>
+  static _Iterator __make_product_iterator(_Iters&&...);
+};
+*/
+
+template <class _Tp, size_t = 0>
+struct __is_product_iterator : false_type {};
+
+template <class _Tp>
+struct __is_product_iterator<_Tp, sizeof(__product_iterator_traits<_Tp>) * 0> : true_type {};
+
+template <class _Tp, size_t _Size, class = void>
+struct __is_product_iterator_of_size : false_type {};
+
+template <class _Tp, size_t _Size>
+struct __is_product_iterator_of_size<_Tp, _Size, __enable_if_t<__product_iterator_traits<_Tp>::__size == _Size> >
+    : true_type {};
+
+template <class _Iterator, size_t _Nth>
+using __product_iterator_element_t _LIBCPP_NODEBUG =
+    decltype(__product_iterator_traits<_Iterator>::template __get_iterator_element<_Nth>(std::declval<_Iterator>()));
+
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP___ITERATOR_PRODUCT_ITERATOR_H

libcxx/include/__ranges/zip_view.h

@@ -23,6 +23,7 @@
 #include <__iterator/iter_move.h>
 #include <__iterator/iter_swap.h>
 #include <__iterator/iterator_traits.h>
+#include <__iterator/product_iterator.h>
 #include <__ranges/access.h>
 #include <__ranges/all.h>
 #include <__ranges/concepts.h>
@@ -251,6 +252,10 @@ class zip_view<_Views...>::__iterator : public __zip_view_iterator_category_base
 
   friend class zip_view<_Views...>;
 
+  static constexpr bool __is_zip_view_iterator = true;
+
+  friend struct __product_iterator_traits<__iterator>;
+
 public:
   using iterator_concept = decltype(ranges::__get_zip_view_iterator_tag<_Const, _Views...>());
   using value_type       = tuple<range_value_t<__maybe_const<_Const, _Views>>...>;
@@ -468,6 +473,23 @@ inline constexpr auto zip = __zip::__fn{};
 } // namespace views
 } // namespace ranges
 
+template <class _Iterator>
+  requires _Iterator::__is_zip_view_iterator
+struct __product_iterator_traits<_Iterator> {
+  static constexpr size_t __size = tuple_size<decltype(std::declval<_Iterator>().__current_)>::value;
+
+  template <size_t _Nth, class _Iter>
+    requires(_Nth < __size)
+  _LIBCPP_HIDE_FROM_ABI static constexpr decltype(auto) __get_iterator_element(_Iter&& __it) {
+    return std::get<_Nth>(std::forward<_Iter>(__it).__current_);
+  }
+
+  template <class... _Iters>
+  _LIBCPP_HIDE_FROM_ABI static constexpr _Iterator __make_product_iterator(_Iters&&... __iters) {
+    return _Iterator(std::tuple(std::forward<_Iters>(__iters)...));
+  }
+};
+
 #endif // _LIBCPP_STD_VER >= 23
 
 _LIBCPP_END_NAMESPACE_STD

libcxx/include/module.modulemap.in

@@ -1522,6 +1522,7 @@ module std [system] {
     }
     module permutable                 { header "__iterator/permutable.h" }
     module prev                       { header "__iterator/prev.h" }
+    module product_iterator           { header "__iterator/product_iterator.h" }
     module projected                  { header "__iterator/projected.h" }
     module ranges_iterator_traits     { header "__iterator/ranges_iterator_traits.h" }
     module readable_traits            { header "__iterator/readable_traits.h" }

libcxx/test/benchmarks/containers/associative/associative_container_benchmarks.h

@@ -13,12 +13,14 @@
 #include <iterator>
 #include <random>
 #include <string>
+#include <ranges>
 #include <type_traits>
 #include <utility>
 #include <vector>
 
 #include "benchmark/benchmark.h"
 #include "../../GenerateInput.h"
+#include "test_macros.h"
 
 namespace support {
 
@@ -66,6 +68,8 @@ void associative_container_benchmarks(std::string container) {
 
   static constexpr bool is_ordered_container = requires(Container c, Key k) { c.lower_bound(k); };
 
+  static constexpr bool is_map_like = requires { typename Container::mapped_type; };
+
   // These benchmarks are structured to perform the operation being benchmarked
   // a small number of times at each iteration, in order to offset the cost of
   // PauseTiming() and ResumeTiming().
@@ -321,6 +325,48 @@ void associative_container_benchmarks(std::string container) {
     }
   });
 
+  if constexpr (is_map_like) {
+    bench("insert(iterator, iterator) (product_iterator from same type)", [=](auto& st) {
+      const std::size_t size = st.range(0);
+      std::vector<Value> in  = make_value_types(generate_unique_keys(size + (size / 10)));
+      Container source(in.begin(), in.end());
+
+      Container c;
+
+      for ([[maybe_unused]] auto _ : st) {
+        c.insert(source.begin(), source.end());
+        benchmark::DoNotOptimize(c);
+        benchmark::ClobberMemory();
+
+        st.PauseTiming();
+        c = Container();
+        st.ResumeTiming();
+      }
+    });
+
+#if TEST_STD_VER >= 23
+    bench("insert(iterator, iterator) (product_iterator from zip_view)", [=](auto& st) {
+      const std::size_t size = st.range(0);
+      std::vector<Key> keys  = generate_unique_keys(size + (size / 10));
+      std::sort(keys.begin(), keys.end());
+      std::vector<typename Container::mapped_type> mapped(keys.size());
+
+      auto source = std::views::zip(keys, mapped);
+
+      Container c;
+
+      for ([[maybe_unused]] auto _ : st) {
+        c.insert(source.begin(), source.end());
+        benchmark::DoNotOptimize(c);
+        benchmark::ClobberMemory();
+
+        st.PauseTiming();
+        c = Container();
+        st.ResumeTiming();
+      }
+    });
+#endif
+  }
   /////////////////////////
   // Erasure
   /////////////////////////

libcxx/test/benchmarks/containers/associative/flat_map.bench.cpp

@@ -6,7 +6,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-// REQUIRES: std-at-least-c++26
+// REQUIRES: std-at-least-c++23
 
 #include <flat_map>
 #include <utility>

libcxx/test/benchmarks/containers/associative/flat_multimap.bench.cpp

@@ -6,7 +6,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-// REQUIRES: std-at-least-c++26
+// REQUIRES: std-at-least-c++23
 
 #include <flat_map>