https://gcc.gnu.org/g:63f3f898bbfda3678b28ebf923bda5282b4ae654
commit r17-4229-g63f3f898bbfda3678b28ebf923bda5282b4ae654 Author: Yuao Ma <[email protected]> Date: Fri Aug 21 22:31:19 2026 +0800 libstdc++: optimize ranges::for_each for segmented iterators Similar to r17-3419-g2d11dfb0e2edd9, we can optimize std::for_each for segemented iterators. libstdc++-v3/ChangeLog: * include/bits/ranges_algo.h (ranges::__for_each): Add segemented iterators logic. Split out naive for-loop from ... (__for_each_fn::operator()): ... here. * testsuite/performance/25_algorithms/for_each.cc: Add benchmark for ranges::for_each. Diff: --- libstdc++-v3/include/bits/ranges_algo.h | 33 ++++++++++++++++++++-- .../performance/25_algorithms/for_each.cc | 22 ++++++++++++++- 2 files changed, 51 insertions(+), 4 deletions(-) diff --git a/libstdc++-v3/include/bits/ranges_algo.h b/libstdc++-v3/include/bits/ranges_algo.h index 4330d3e70b84..f4c70dc70291 100644 --- a/libstdc++-v3/include/bits/ranges_algo.h +++ b/libstdc++-v3/include/bits/ranges_algo.h @@ -211,6 +211,33 @@ namespace ranges template<typename _Iter, typename _Fp> using for_each_result = in_fun_result<_Iter, _Fp>; + // Apply __f to the result of applying __proj to each element in + // [__first, __last). + // Dispatches to __for_each_segment for segmented iterators + // (e.g. deque::iterator). + // Returns an iterator equal to __last. + template<typename _InputIterator, typename _Sentinel, typename _Function, + typename _Proj> + constexpr _InputIterator + __for_each(_InputIterator __first, _Sentinel __last, _Function&& __f, + _Proj& __proj) + { + if constexpr (__segmented_iterator<_InputIterator> + && same_as<_InputIterator, _Sentinel>) + { + std::__for_each_segment(__first, __last, + [&](auto __lfirst, auto __llast) + { return ranges::__for_each(__lfirst, __llast, __f, __proj); }); + return __last; + } + else + { + for (; __first != __last; ++__first) + std::__invoke(__f, std::__invoke(__proj, *__first)); + return __first; + } + } + struct __for_each_fn { template<input_iterator _Iter, sentinel_for<_Iter> _Sent, @@ -219,9 +246,9 @@ namespace ranges constexpr for_each_result<_Iter, _Fun> operator()(_Iter __first, _Sent __last, _Fun __f, _Proj __proj = {}) const { - for (; __first != __last; ++__first) - std::__invoke(__f, std::__invoke(__proj, *__first)); - return { std::move(__first), std::move(__f) }; + auto __end = ranges::__for_each(std::move(__first), std::move(__last), __f, + __proj); + return { std::move(__end), std::move(__f) }; } template<input_range _Range, typename _Proj = identity, diff --git a/libstdc++-v3/testsuite/performance/25_algorithms/for_each.cc b/libstdc++-v3/testsuite/performance/25_algorithms/for_each.cc index 37af5ab59816..65b13be6c21a 100644 --- a/libstdc++-v3/testsuite/performance/25_algorithms/for_each.cc +++ b/libstdc++-v3/testsuite/performance/25_algorithms/for_each.cc @@ -15,7 +15,20 @@ void bench_seq(const char* label, __gnu_test::time_counter& time, start_counters(time, resource); for (int i = 0; i < 20000; ++i) std::for_each(c.begin(), c.end(), - [](T& x) { x = std::min<T>(100, std::max<T>(x, 10)); }); + [](T& x) { x = std::clamp<T>(x, 10, 100); }); + stop_counters(time, resource); + report_performance(__FILE__, label, time, resource); + clear_counters(time, resource); +} + +template <typename Container> +void bench_seq_ranges(const char* label, __gnu_test::time_counter& time, + __gnu_test::resource_counter& resource) { + using T = typename Container::value_type; + Container c(size, 1); + start_counters(time, resource); + for (int i = 0; i < 20000; ++i) + std::ranges::for_each(c, [](T& x) { x = std::clamp<T>(x, 10, 100); }); stop_counters(time, resource); report_performance(__FILE__, label, time, resource); clear_counters(time, resource); @@ -29,4 +42,11 @@ int main() { bench_seq<std::vector<int>>("std::for_each vector<int>", time, resource); bench_seq<std::deque<int>>("std::for_each deque<int>", time, resource); bench_seq<std::list<int>>("std::for_each list<int>", time, resource); + + bench_seq_ranges<std::vector<int>>("std::ranges::for_each vector<int>", time, + resource); + bench_seq_ranges<std::deque<int>>("std::ranges::for_each deque<int>", time, + resource); + bench_seq_ranges<std::list<int>>("std::ranges::for_each list<int>", time, + resource); }
