[gcc r17-3419] libstdc++: optimize std::for_each for segmented iterators
Yuao Ma
yuao@gcc.gnu.org
Wed Aug 19 12:58:03 GMT 2026
https://gcc.gnu.org/g:2d11dfb0e2edd9e3ea451987b844c76b12bb09af
commit r17-3419-g2d11dfb0e2edd9e3ea451987b844c76b12bb09af
Author: Yuao Ma <c8ef@outlook.com>
Date: Mon Aug 3 23:05:04 2026 +0800
libstdc++: optimize std::for_each for segmented iterators
Similar to r17-2859-g15505d012872dd, we can optimize std::for_each for
segemented iterators.
libstdc++-v3/ChangeLog:
* include/bits/stl_algo.h (__for_each): Add segemented
iterators logic. Split out naive for-loop from ...
(for_each): ... here.
* testsuite/performance/25_algorithms/for_each.cc: New test.
Diff:
---
libstdc++-v3/include/bits/stl_algo.h | 33 ++++++++++++++++++++--
.../performance/25_algorithms/for_each.cc | 32 +++++++++++++++++++++
2 files changed, 63 insertions(+), 2 deletions(-)
diff --git a/libstdc++-v3/include/bits/stl_algo.h b/libstdc++-v3/include/bits/stl_algo.h
index 800c176cd5bd..6a9e133ce955 100644
--- a/libstdc++-v3/include/bits/stl_algo.h
+++ b/libstdc++-v3/include/bits/stl_algo.h
@@ -132,6 +132,36 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
return __first;
}
+ // Apply __f to each element in [__first, __last)
+ // Dispatches to __for_each_segment for segmented iterators
+ // (e.g. deque::iterator).
+ // Returns an iterator equal to __last.
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wc++17-extensions" // if constexpr
+#pragma GCC diagnostic ignored "-Wc++20-extensions" // template lambda
+ template<typename _InputIterator, typename _Function>
+ _GLIBCXX20_CONSTEXPR
+ _InputIterator
+ __for_each(_InputIterator __first, _InputIterator __last, _Function& __f)
+ {
+#if __cplusplus >= 201103L
+ if constexpr (__enable_for_each_segment<_InputIterator>)
+ {
+ std::__for_each_segment(__first, __last,
+ [&]<typename _Iter>(_Iter __lfirst, _Iter __llast)
+ { return std::__for_each(__lfirst, __llast, __f); });
+ return __last;
+ }
+ else
+#endif // C++11
+ {
+ for (; __first != __last; ++__first)
+ __f(*__first);
+ return __first;
+ }
+ }
+#pragma GCC diagnostic pop
+
// set_difference
// set_intersection
// set_symmetric_difference
@@ -3813,8 +3843,7 @@ _GLIBCXX_BEGIN_NAMESPACE_ALGO
// concept requirements
__glibcxx_function_requires(_InputIteratorConcept<_InputIterator>)
__glibcxx_requires_valid_range(__first, __last);
- for (; __first != __last; ++__first)
- __f(*__first);
+ std::__for_each(__first, __last, __f);
return __f; // N.B. [alg.foreach] says std::move(f) but it's redundant.
}
diff --git a/libstdc++-v3/testsuite/performance/25_algorithms/for_each.cc b/libstdc++-v3/testsuite/performance/25_algorithms/for_each.cc
new file mode 100644
index 000000000000..37af5ab59816
--- /dev/null
+++ b/libstdc++-v3/testsuite/performance/25_algorithms/for_each.cc
@@ -0,0 +1,32 @@
+#include <testsuite_performance.h>
+
+#include <algorithm>
+#include <deque>
+#include <list>
+#include <vector>
+
+const std::size_t size = 8192;
+
+template <typename Container>
+void bench_seq(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::for_each(c.begin(), c.end(),
+ [](T& x) { x = std::min<T>(100, std::max<T>(x, 10)); });
+ stop_counters(time, resource);
+ report_performance(__FILE__, label, time, resource);
+ clear_counters(time, resource);
+}
+
+int main() {
+ using namespace __gnu_test;
+ time_counter time;
+ resource_counter resource;
+
+ 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);
+}
More information about the Libstdc++-cvs
mailing list