[libcxx-commits] [libcxx] [libc++][pstl] Introduce a "statpart" backend (PR #230540)
Michael G. Kazakov via libcxx-commits
libcxx-commits at lists.llvm.org
Fri Oct 16 05:15:08 PDT 2026
https://github.com/mikekazakov updated https://github.com/llvm/llvm-project/pull/230540
>From 4f222cde8246d1c139493786b3421921b33fd645 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Fri, 9 Oct 2026 15:53:38 +0100
Subject: [PATCH 01/24] Introduce a "statpart" backend
---
libcxx/CMakeLists.txt | 2 +
libcxx/include/CMakeLists.txt | 1 +
libcxx/include/__config_site.in | 1 +
libcxx/include/__pstl/backend.h | 6 +
libcxx/include/__pstl/backend_fwd.h | 3 +
libcxx/include/__pstl/backends/statpart.h | 1207 +++++++++++++++++
libcxx/include/module.modulemap.in | 3 +
libcxx/src/CMakeLists.txt | 6 +
libcxx/src/pstl/statpart.cpp | 26 +
.../test/libcxx/transitive_includes/cxx17.csv | 4 +
.../test/libcxx/transitive_includes/cxx20.csv | 4 +
.../test/libcxx/transitive_includes/cxx23.csv | 4 +
.../test/libcxx/transitive_includes/cxx26.csv | 4 +
13 files changed, 1271 insertions(+)
create mode 100644 libcxx/include/__pstl/backends/statpart.h
create mode 100644 libcxx/src/pstl/statpart.cpp
diff --git a/libcxx/CMakeLists.txt b/libcxx/CMakeLists.txt
index d163f0f579c395..5478c38afa781d 100644
--- a/libcxx/CMakeLists.txt
+++ b/libcxx/CMakeLists.txt
@@ -803,6 +803,8 @@ elseif(LIBCXX_PSTL_BACKEND STREQUAL "std_thread")
config_define(1 _LIBCPP_PSTL_BACKEND_STD_THREAD)
elseif(LIBCXX_PSTL_BACKEND STREQUAL "libdispatch")
config_define(1 _LIBCPP_PSTL_BACKEND_LIBDISPATCH)
+elseif(LIBCXX_PSTL_BACKEND STREQUAL "statpart")
+ config_define(1 _LIBCPP_PSTL_BACKEND_STATPART)
else()
message(FATAL_ERROR "LIBCXX_PSTL_BACKEND is set to ${LIBCXX_PSTL_BACKEND}, which is not a valid backend.
Valid backends are: serial, std_thread and libdispatch")
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index 1ee2081561fd97..da2546a54bc04d 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -695,6 +695,7 @@ set(files
__pstl/backends/default.h
__pstl/backends/libdispatch.h
__pstl/backends/serial.h
+ __pstl/backends/statpart.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__config_site.in b/libcxx/include/__config_site.in
index e6e33a2eeb2a9e..b0a2230c42c21d 100644
--- a/libcxx/include/__config_site.in
+++ b/libcxx/include/__config_site.in
@@ -36,6 +36,7 @@
#cmakedefine _LIBCPP_PSTL_BACKEND_SERIAL
#cmakedefine _LIBCPP_PSTL_BACKEND_STD_THREAD
#cmakedefine _LIBCPP_PSTL_BACKEND_LIBDISPATCH
+#cmakedefine _LIBCPP_PSTL_BACKEND_STATPART
// Hardening.
#cmakedefine _LIBCPP_HARDENING_MODE_DEFAULT @_LIBCPP_HARDENING_MODE_DEFAULT@
diff --git a/libcxx/include/__pstl/backend.h b/libcxx/include/__pstl/backend.h
index 5980b0708cd340..75ccb5063c2919 100644
--- a/libcxx/include/__pstl/backend.h
+++ b/libcxx/include/__pstl/backend.h
@@ -30,8 +30,14 @@ _LIBCPP_PUSH_MACROS
# elif defined(_LIBCPP_PSTL_BACKEND_LIBDISPATCH)
# include <__pstl/backends/default.h>
# include <__pstl/backends/libdispatch.h>
+# elif defined(_LIBCPP_PSTL_BACKEND_STATPART)
+# include <__pstl/backends/default.h>
+# include <__pstl/backends/statpart.h>
# endif
+// statpart pulls <cerrno> in transitively via __thread/thread.h, make sure this include is not is conditional/flaky
+# include <cerrno>
+
#endif // _LIBCPP_STD_VER >= 17
_LIBCPP_POP_MACROS
diff --git a/libcxx/include/__pstl/backend_fwd.h b/libcxx/include/__pstl/backend_fwd.h
index 2665209887944f..31ae6581e1a52d 100644
--- a/libcxx/include/__pstl/backend_fwd.h
+++ b/libcxx/include/__pstl/backend_fwd.h
@@ -51,6 +51,7 @@ struct __default_backend_tag;
struct __libdispatch_backend_tag;
struct __serial_backend_tag;
struct __std_thread_backend_tag;
+struct __statpart_backend_tag;
# if defined(_LIBCPP_PSTL_BACKEND_SERIAL)
using __current_configuration _LIBCPP_NODEBUG = __backend_configuration<__serial_backend_tag, __default_backend_tag>;
@@ -60,6 +61,8 @@ using __current_configuration _LIBCPP_NODEBUG =
# elif defined(_LIBCPP_PSTL_BACKEND_LIBDISPATCH)
using __current_configuration _LIBCPP_NODEBUG =
__backend_configuration<__libdispatch_backend_tag, __default_backend_tag>;
+# elif defined(_LIBCPP_PSTL_BACKEND_STATPART)
+using __current_configuration _LIBCPP_NODEBUG = __backend_configuration<__statpart_backend_tag, __default_backend_tag>;
# else
// ...New vendors can add parallel backends here...
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
new file mode 100644
index 00000000000000..41e473d663f3ed
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -0,0 +1,1207 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_H
+#define _LIBCPP___PSTL_STATPART_H
+
+#include <__algorithm/find_end.h>
+#include <__algorithm/find_if.h>
+#include <__algorithm/for_each.h>
+#include <__algorithm/is_heap_until.h>
+#include <__algorithm/lower_bound.h>
+#include <__algorithm/merge.h>
+#include <__algorithm/min.h>
+#include <__algorithm/min_element.h>
+#include <__algorithm/minmax_element.h>
+#include <__algorithm/mismatch.h>
+#include <__algorithm/move.h>
+#include <__algorithm/reverse.h>
+#include <__algorithm/search.h>
+#include <__algorithm/search_n.h>
+#include <__algorithm/stable_sort.h>
+#include <__algorithm/swap_ranges.h>
+#include <__algorithm/transform.h>
+#include <__algorithm/upper_bound.h>
+#include <__atomic/atomic.h>
+#include <__config>
+#include <__functional/identity.h>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__iterator/move_iterator.h>
+#include <__iterator/next.h>
+#include <__iterator/reverse_iterator.h>
+#include <__memory/allocator.h>
+#include <__memory/construct_at.h>
+#include <__memory/destroy.h>
+#include <__memory/uninitialized_algorithms.h>
+#include <__memory/unique_ptr.h>
+#include <__numeric/reduce.h>
+#include <__numeric/transform_reduce.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/cpu_algos/cpu_traits.h>
+#include <__pstl/cpu_algos/find_if.h>
+#include <__thread/thread.h>
+#include <__type_traits/is_assignable.h>
+#include <__type_traits/is_constructible.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__type_traits/is_trivially_copyable.h>
+#include <__utility/convert_to_integral.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+#include <__utility/pair.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+namespace __statpart {
+
+// The backbone for the backend: applies the function to each of the indices [0, __chunk_count] in parallel.
+// The actual number of parallel workers and the order of application is unspecified.
+_LIBCPP_EXPORTED_FROM_ABI void
+__apply(size_t __chunk_count, void* __context, void (*__func)(void* __context, size_t __chunk)) noexcept;
+
+// The maximum number of partitions to be processed per each CPU
+constexpr inline size_t __max_partitions_per_cpu = 8;
+
+// The minimum number of elements to sort serially in a single partition
+constexpr inline size_t __min_sort_partition_size = 256;
+
+enum class __partitioning_result {
+ // returned when the partitioning result is not nullopt
+ __ok,
+ // returned when there is not enough element to produce a single partition or the final number of partitions is less
+ // than 2, i.e. there's no reason to process in parallel
+ __not_enough_elements,
+ // later: returned memory couldn't be allocated for the stashed forward iterators
+ __bad_alloc
+};
+
+template <class _Iterator>
+struct _LIBCPP_HIDE_FROM_ABI __iterator_range {
+ _Iterator __first;
+ _Iterator __last;
+};
+
+// Later to support forward iterators via a second implicit template argument.
+template <class _Iterator>
+struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
+ _LIBCPP_HIDE_FROM_ABI size_t __partitions_count() const { return __partitions_count_; }
+
+ _LIBCPP_HIDE_FROM_ABI __iterator_range<_Iterator> __partition(size_t __n) const {
+ _Iterator __first = __base_ + __n * __partition_size_ + (__n < __remainder_ ? __n : __remainder_);
+ _Iterator __last = __first + __partition_size_ + (__n < __remainder_ ? 1 : 0);
+ return __iterator_range<_Iterator>{__first, __last};
+ }
+
+ _LIBCPP_HIDE_FROM_ABI _Iterator __end() const { return __base_ + __count_; }
+
+private:
+ _Iterator __base_;
+ size_t __count_;
+ size_t __partition_size_;
+ size_t __remainder_;
+ size_t __partitions_count_;
+
+ template <class _Iter>
+ friend optional<__iterator_partitions<_Iter>>
+ __partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t) noexcept;
+};
+
+template <class _ForwardIterator>
+_LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>>
+__partition(_ForwardIterator __first,
+ _ForwardIterator __last,
+ __partitioning_result& __result,
+ size_t __min_partition_size = 1,
+ size_t __prefix_to_skip = 0,
+ size_t __suffix_to_skip = 0) noexcept {
+ // forward iteratos to be supported later
+ static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
+ // TODO: assert( __min_partition_size > 0 )?
+
+ size_t __count = static_cast<size_t>(std::distance(__first, __last));
+ if (__count < __prefix_to_skip + __suffix_to_skip) {
+ __result = __partitioning_result::__not_enough_elements;
+ return {};
+ }
+
+ size_t __effective_count = __count - __prefix_to_skip - __suffix_to_skip;
+ size_t __workers_count = thread::hardware_concurrency();
+ size_t __max_partitions = __workers_count * __max_partitions_per_cpu;
+ size_t __partition_count = std::min(__max_partitions, __effective_count / __min_partition_size);
+
+ if (__partition_count < 2) {
+ // TODO: verify that's what we want
+ __result = __partitioning_result::__not_enough_elements;
+ return {}; // not enough input data to process in parallel => no reason to build partitions
+ }
+
+ __iterator_partitions<_ForwardIterator> __p;
+ __p.__base_ = __first + __prefix_to_skip;
+ __p.__count_ = __count;
+ __p.__partition_size_ = __count / __partition_count;
+ __p.__remainder_ = __count % __partition_count;
+ __p.__partitions_count_ = __partition_count;
+ return __p;
+}
+
+// for_each with explicit partitioning
+template <class _Iterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI void __for_each_partition(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
+ struct __ctx_t {
+ const __iterator_partitions<_Iterator>& __partitions_;
+ _Functor& __func_;
+ } __ctx{__partitions, __func};
+ __statpart::__apply(__partitions.__partitions_count(), &__ctx, [](void* __context, size_t __partition) {
+ __ctx_t& __ctx = *static_cast<__ctx_t*>(__context);
+ auto __range = __ctx.__partitions_.__partition(__partition);
+ __ctx.__func_(__range.__first, __range.__last);
+ });
+}
+
+// Calls __func(index) for every index in [0, __count) in parallel
+template <class _Functor>
+_LIBCPP_HIDE_FROM_ABI void __for_each_index(size_t __count, _Functor __func) {
+ __statpart::__apply(__count, &__func, [](void* __context, size_t __index) {
+ (*static_cast<_Functor*>(__context))(__index);
+ });
+}
+
+// TODO: ...
+template <class _Iterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI void
+__for_each_partition_index(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
+ __statpart::__for_each_index(__partitions.__partitions_count(), std::move(__func));
+}
+
+// for_each with default implicit partitioning
+template <class _Iterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each(_Iterator __first, _Iterator __last, _Functor __func) {
+ if (__first == __last)
+ return __empty{};
+ __partitioning_result __partition_result;
+ auto __partitions = __statpart::__partition(__first, __last, __partition_result);
+ if (__partitions) {
+ __statpart::__for_each_partition(*__partitions, std::move(__func));
+ return __empty{};
+ } else if (__partition_result == __partitioning_result::__bad_alloc) {
+ return nullopt;
+ } else {
+ __func(__first, __last);
+ return __empty{};
+ }
+}
+
+// Applies the 3-legged function to sub-ranges of the input ranges in parallel:
+// f(sub_range1_first, sub_range1_last, sub_range2_first)
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BrickFunction>
+_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each_iter_pair(
+ _RandomAccessIterator1 __first1,
+ _RandomAccessIterator1 __last1,
+ _RandomAccessIterator2 __first2,
+ _BrickFunction __f) {
+ return __for_each(
+ __first1,
+ __last1,
+ [__first1, __first2, __f = std::move(__f)](
+ _RandomAccessIterator1 __brick_first1, _RandomAccessIterator1 __brick_last1) {
+ _RandomAccessIterator2 __brick_first2 = __first2 + (__brick_first1 - __first1);
+ __f(__brick_first1, __brick_last1, __brick_first2);
+ });
+}
+
+template <class _Index, class _Brick, class _Compare>
+_LIBCPP_HIDE_FROM_ABI optional<_Index>
+__find(_Index __first, _Index __last, _Brick __f, _Compare __comp, bool __b_first) {
+ typedef typename std::iterator_traits<_Index>::difference_type _DifferenceType;
+ const _DifferenceType __n = __last - __first;
+ _DifferenceType __initial_dist = __b_first ? __n : -1;
+ std::atomic<_DifferenceType> __extremum(__initial_dist);
+ // TODO: find out what is better here: parallel_for or parallel_reduce
+ auto __res = __statpart::__for_each(__first, __last, [__comp, __f, __first, &__extremum](_Index __i, _Index __j) {
+ // See "Reducing Contention Through Priority Updates", PPoPP '13, for discussion of
+ // why using a shared variable scales fairly well in this situation.
+ if (__comp(__i - __first, __extremum)) {
+ _Index __result = __f(__i, __j);
+ // If not '__last' returned then we found what we want so put this to extremum
+ if (__result != __j) {
+ const _DifferenceType __k = __result - __first;
+ for (_DifferenceType __old = __extremum; __comp(__k, __old); __old = __extremum) {
+ __extremum.compare_exchange_weak(__old, __k);
+ }
+ }
+ }
+ });
+ if (!__res)
+ return nullopt;
+ return __extremum.load() != __initial_dist ? __first + __extremum.load() : __last;
+}
+
+template <class _RandomAccessIterator, class _Transform, class _Value, class _Combiner, class _Reduction>
+_LIBCPP_HIDE_FROM_ABI optional<_Value> __transform_reduce(
+ _RandomAccessIterator __first,
+ _RandomAccessIterator __last,
+ _Transform __transform,
+ _Value __init,
+ _Combiner __combiner,
+ _Reduction __reduction) {
+ if (__first == __last)
+ return __init;
+
+ // Split into partitions with at least 3 elements in each
+ __partitioning_result __partitioning_result;
+ auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, 3);
+ if (!__partitions) {
+ if (__partitioning_result == __partitioning_result::__bad_alloc) {
+ // Failed to allocate memory for the partitions, propagate the error up.
+ return nullopt;
+ } else {
+ // Not enough elements to run in parallel, switch to serial implementation.
+ return __reduction(std::move(__first), std::move(__last), std::move(__init));
+ }
+ }
+
+ auto __destroy = [__count = __partitions->__partitions_count()](_Value* __ptr) {
+ std::destroy_n(__ptr, __count);
+ std::allocator<_Value>().deallocate(__ptr, __count);
+ };
+
+ // TODO: use __uninitialized_buffer
+ // TODO: allocate one element per worker instead of one element per chunk
+ unique_ptr<_Value[], decltype(__destroy)> __values(
+ std::allocator<_Value>().allocate(__partitions->__partitions_count()), __destroy);
+
+ // __dispatch_apply is noexcept
+ __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
+ auto __range = __partitions->__partition(__index);
+ std::__construct_at(
+ __values.get() + __index,
+ __reduction(__range.__first + 2,
+ __range.__last,
+ __combiner(__transform(__range.__first), __transform(__range.__first + 1))));
+ });
+
+ return std::reduce(
+ std::make_move_iterator(__values.get()),
+ std::make_move_iterator(__values.get() + __partitions->__partitions_count()),
+ std::move(__init),
+ __combiner);
+}
+
+template <class _RandomAccessIterator1,
+ class _RandomAccessIterator2,
+ class _RandomAccessIterator3,
+ class _Compare,
+ class _LeafMerge>
+_LIBCPP_HIDE_FROM_ABI optional<__empty>
+__merge(_RandomAccessIterator1 __first1,
+ _RandomAccessIterator1 __last1,
+ _RandomAccessIterator2 __first2,
+ _RandomAccessIterator2 __last2,
+ _RandomAccessIterator3 __result,
+ _Compare __comp,
+ _LeafMerge __leaf_merge) noexcept {
+ // Run the merge such that the longest range is partitioned.
+ // __merge_sub_ranges swaps the ranges implicitly in case the second range was the longest one.
+ auto __run =
+ [__result](auto __long_first,
+ auto __long_last,
+ auto __short_first,
+ auto __short_last,
+ auto __find_cutpoint,
+ auto __merge_sub_ranges) -> optional<__empty> {
+ // Partition the longest input range evenly.
+ __partitioning_result __partitioning_result;
+ auto __partitions = __statpart::__partition(__long_first, __long_last, __partitioning_result);
+ if (!__partitions) {
+ if (__partitioning_result == __partitioning_result::__bad_alloc)
+ return nullopt;
+ // Not enough elements to run in parallel, merge everything serially.
+ __merge_sub_ranges(__long_first, __long_last, __short_first, __short_last, __result);
+ return __empty{};
+ }
+
+ // Process the partitions of the longest range in parallel.
+ __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
+ // Get the sub-range inside the longer input range.
+ auto __range = __partitions->__partition(__index);
+
+ // Find the matching sub-range of the shorter input range: it will hold the elements that come after the last
+ // element of the previous long partition and up to the last element of this partition.
+ // This generally runs two binary searches in the smaller input range, edge partitions use existing bounds.
+ auto __short_begin =
+ __index == 0 ? __short_first : __find_cutpoint(__short_first, __short_last, __range.__first[-1]);
+ auto __short_end =
+ __index == __partitions->__partitions_count() - 1
+ ? __short_last
+ : __find_cutpoint(__short_first, __short_last, __range.__last[-1]);
+
+ // Derive the position in the output range and merge the sub-ranges.
+ auto __result_first = __result + (__range.__first - __long_first) + (__short_begin - __short_first);
+ __merge_sub_ranges(__range.__first, __range.__last, __short_begin, __short_end, __result_first);
+ });
+ return __empty{};
+ };
+
+ if (__last1 - __first1 > __last2 - __first2) {
+ // Preserve the order of the input ranges, partition the first range.
+ // Equal elements of the second range go after the cut => use lower_bound when cutting.
+ return __run(
+ __first1,
+ __last1,
+ __first2,
+ __last2,
+ [&](auto __first, auto __last, const auto& __value) {
+ return std::lower_bound(__first, __last, __value, __comp);
+ },
+ [&](auto __long_first, auto __long_last, auto __short_first, auto __short_last, auto __result_first) {
+ __leaf_merge(__long_first, __long_last, __short_first, __short_last, __result_first, __comp);
+ });
+ } else {
+ // Swap the order of the input ranges, partition the second range.
+ // Equal elements of the first range go before the cut => use upper_bound when cutting.
+ return __run(
+ __first2,
+ __last2,
+ __first1,
+ __last1,
+ [&](auto __first, auto __last, const auto& __value) {
+ return std::upper_bound(__first, __last, __value, __comp);
+ },
+ [&](auto __long_first, auto __long_last, auto __short_first, auto __short_last, auto __result_first) {
+ __leaf_merge(__short_first, __short_last, __long_first, __long_last, __result_first, __comp);
+ });
+ }
+}
+
+template <class _RandomAccessIterator, class _Comp, class _LeafSort>
+_LIBCPP_HIDE_FROM_ABI optional<__empty>
+__stable_sort(_RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp, _LeafSort __leaf_sort) {
+ const auto __size = __last - __first;
+
+ __partitioning_result __partitioning_result;
+ auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, __min_sort_partition_size);
+ if (!__partitions) {
+ if (__partitioning_result == __partitioning_result::__bad_alloc)
+ return nullopt;
+ // Not enough elements to run in parallel, sort everything serially.
+ __leaf_sort(__first, __last, __comp);
+ return __empty{};
+ }
+
+ using _Value = __iterator_value_type<_RandomAccessIterator>;
+
+ auto __destroy = [__size](_Value* __ptr) {
+ std::destroy_n(__ptr, __size);
+ std::allocator<_Value>().deallocate(__ptr, __size);
+ };
+
+ // TODO: use __uninitialized_buffer
+ unique_ptr<_Value[], decltype(__destroy)> __values(std::allocator<_Value>().allocate(__size), __destroy);
+
+ // Initialize all elements to a moved-from state
+ // TODO: Don't do this - this can be done in the first merge - see https://llvm.org/PR63928
+ std::__construct_at(__values.get(), std::move(*__first));
+ for (__iterator_difference_type<_RandomAccessIterator> __i = 1; __i != __size; ++__i) {
+ std::__construct_at(__values.get() + __i, std::move(__values.get()[__i - 1]));
+ }
+ *__first = std::move(__values.get()[__size - 1]);
+
+ __statpart::__for_each_partition(
+ *__partitions, [&__leaf_sort, &__comp](_RandomAccessIterator __chunk_first, _RandomAccessIterator __chunk_last) {
+ __leaf_sort(std::move(__chunk_first), std::move(__chunk_last), __comp);
+ });
+
+ // A chunk span is a run of __chunk_span_size consecutive partitions, the last chunk can be shorter in case of odd
+ // number of chunk spans. Each pass merges the chunk spans pairwise, so the chunk span size doubles and the chunk
+ // count is halved (rounded up).
+ const size_t __partition_count = __partitions->__partitions_count();
+ size_t __chunk_span_size = 1;
+ size_t __chunk_spans = __partition_count;
+
+ // Offsets of the beginning and the end of the chunk from __first
+ auto __chunk_begin = [&](size_t __chunk) {
+ return __partitions->__partition(__chunk * __chunk_span_size).__first - __first;
+ };
+ auto __chunk_end = [&](size_t __chunk) {
+ return __partitions->__partition(std::min((__chunk + 1) * __chunk_span_size, __partition_count) - 1).__last -
+ __first;
+ };
+
+ bool __objects_are_in_buffer = false;
+ do {
+ size_t __chunk_spans_pairs = (__chunk_spans + 1) / 2;
+ auto __merge_chunks = [&](auto __from_first, auto __to_first) {
+ __statpart::__for_each_index(__chunk_spans_pairs, [&](size_t __span_pair_index) {
+ size_t __left_chunk = 2 * __span_pair_index;
+ size_t __right_chunk = __left_chunk + 1;
+ auto __pair_first = __chunk_begin(__left_chunk);
+ if (__right_chunk == __chunk_spans) {
+ // The odd chunk has no counterpart to merge with => carry it over to the other buffer as-is.
+ std::move(__from_first + __pair_first, __from_first + __chunk_end(__left_chunk), __to_first + __pair_first);
+ } else {
+ // Merge the two chunk spans into the other buffer
+ auto __pair_mid = __chunk_begin(__right_chunk);
+ auto __pair_last = __chunk_end(__right_chunk);
+ std::merge(std::make_move_iterator(__from_first + __pair_first),
+ std::make_move_iterator(__from_first + __pair_mid),
+ std::make_move_iterator(__from_first + __pair_mid),
+ std::make_move_iterator(__from_first + __pair_last),
+ __to_first + __pair_first,
+ __comp);
+ }
+ });
+ };
+
+ if (__objects_are_in_buffer)
+ __merge_chunks(__values.get(), __first);
+ else
+ __merge_chunks(__first, __values.get());
+ __objects_are_in_buffer = !__objects_are_in_buffer;
+
+ __chunk_span_size *= 2;
+ __chunk_spans = __chunk_spans_pairs;
+ } while (__chunk_spans > 1);
+
+ if (__objects_are_in_buffer) {
+ std::move(__values.get(), __values.get() + __size, __first);
+ }
+
+ return __empty{};
+}
+
+} // namespace __statpart
+
+template <class _ExecutionPolicy>
+struct __find_end<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator1>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _BinaryPredicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
+ _DifferenceType __size2 = __last2 - __first2; // The length of the needle to search for.
+ if (__size2 == 0) {
+ return __last1; // If the needle length is zero, the last iterator is returned.
+ }
+ _DifferenceType __size1 = __last1 - __first1;
+ if (__size1 < __size2) {
+ return __last1; // The range is too small to contain the requested number of consecutive elements.
+ }
+ // Calculate the length of the tail where a potential match cannot start by definition.
+ _DifferenceType __crop = __size2 - 1;
+ // We're only interested in the range where a potential match can start: [first, last - crop)
+ _ForwardIterator1 __last1_cropped = __last1 - __crop;
+ // Run a parallel chunked find_if, covering the range where a potential match can start.
+ auto __res = __statpart::__find(
+ __first1,
+ __last1_cropped,
+ [__first2, __last2, __crop, &__pred](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
+ // Uncrop the range to allow std::find_end to find a full match, which can go beyond __brick_last.
+ _ForwardIterator1 __brick_last_uncropped = __brick_last + __crop;
+ // Run a serial std::find_end inside each of the chunks in parallel.
+ _ForwardIterator1 __ret = std::find_end(__brick_first, __brick_last_uncropped, __first2, __last2, __pred);
+ // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
+ // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __parallel_find().
+ return __ret == __brick_last_uncropped ? __brick_last : __ret;
+ },
+ greater<>{}, // `greater` here means the highest index among the matches
+ false // `false` here means we want the last match, not the first
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ if (*__res == __last1_cropped) {
+ return __last1; // No match was found in the range.
+ }
+ return *__res; // Return the successful match.
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::find_end(
+ std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __for_each<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Function>
+ _LIBCPP_HIDE_FROM_ABI optional<__empty>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Function __func) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __statpart::__for_each(
+ std::move(__first), std::move(__last), [&](auto __partition_first, auto __partition_last) {
+ std::for_each(__partition_first, __partition_last, __func);
+ });
+ } else {
+ std::for_each(std::move(__first), std::move(__last), std::move(__func));
+ return __empty{};
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __find_if<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Predicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __statpart::__find(
+ std::move(__first),
+ std::move(__last),
+ [&__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ return std::find_if(__brick_first, __brick_last, __pred);
+ },
+ less<>{},
+ true);
+ } else {
+ return std::find_if(std::move(__first), std::move(__last), std::move(__pred));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __is_heap_until<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _RandomAccessIterator, class _Comp>
+ _LIBCPP_HIDE_FROM_ABI optional<_RandomAccessIterator>
+ operator()(_Policy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy>) {
+ if (__last - __first < 2)
+ return __last; // Any sequence with less than 2 elements is a heap
+ // Run a parallel find in chunks over [first+1, last) to validate every element except the root.
+ return __statpart::__find(
+ __first + 1,
+ __last,
+ [__first, &__comp](_RandomAccessIterator __child_first, _RandomAccessIterator __child_last) {
+ // This algorithm assumes that __find() will never pass an empty brick,
+ // i.e. that __child_first != __child_last.
+ using _DifferenceType = typename std::iterator_traits<_RandomAccessIterator>::difference_type;
+
+ // Derive the indices of the children and the iterators of their parents
+ _DifferenceType __child_first_idx = __child_first - __first;
+ _DifferenceType __child_last_idx = __child_last - __first;
+ _RandomAccessIterator __parent = __first + (__child_first_idx - 1) / 2;
+ _RandomAccessIterator __parent_last = __first + (__child_last_idx - 1) / 2;
+
+ // If we're starting from a right child => process this element separately in a prologue
+ if (__child_first_idx % 2 == 0) {
+ if (__comp(*__parent, *__child_first)) // Check the right child
+ return __child_first;
+ ++__parent;
+ ++__child_first;
+ }
+
+ // Iterate over the parents and check their left and right children
+ for (; __parent != __parent_last; ++__parent) {
+ if (__comp(*__parent, *__child_first)) // Check the left child
+ return __child_first;
+ ++__child_first;
+
+ if (__comp(*__parent, *__child_first)) // Check the right child
+ return __child_first;
+ ++__child_first;
+ }
+
+ // If we're ending with a right child => __parent_last also includes the left child.
+ // Process this accessible element separately in an epilogue.
+ if (__child_last_idx % 2 == 0) {
+ if (__comp(*__parent, *__child_first)) // Check the left child
+ return __child_first;
+ }
+
+ return __child_last; // No violations found in this brick
+ },
+ less<>{}, // `less` here means the lowest index among the matches
+ true // `true` here means we want the first match, not the last
+ );
+ } else {
+ return std::is_heap_until(std::move(__first), std::move(__last), std::move(__comp));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __merge<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardOutIterator, class _Comp>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator> operator()(
+ _Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _ForwardOutIterator __result,
+ _Comp __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
+ auto __res = __statpart::__merge(
+ __first1,
+ __last1,
+ __first2,
+ __last2,
+ __result,
+ __comp,
+ [](_ForwardIterator1 __brick_first1,
+ _ForwardIterator1 __brick_last1,
+ _ForwardIterator2 __brick_first2,
+ _ForwardIterator2 __brick_last2,
+ _ForwardOutIterator __brick_result,
+ _Comp __g_comp) {
+ std::merge(std::move(__brick_first1),
+ std::move(__brick_last1),
+ std::move(__brick_first2),
+ std::move(__brick_last2),
+ std::move(__brick_result),
+ std::move(__g_comp));
+ });
+ if (!__res)
+ return nullopt;
+ return __result + (__last1 - __first1) + (__last2 - __first2);
+ } else {
+ return std::merge(__first1, __last1, __first2, __last2, __result, __comp);
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __min_element<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Compare>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ if (__first == __last) {
+ return __last; // nothing to do
+ }
+
+ _ForwardIterator __init = __first;
+ ++__first;
+ if (__first == __last) {
+ return __init; // the only element is the minimum
+ }
+
+ // A reduction that returns an iterator pointing to the lowest element, left bias in case of a tie
+ auto __iter_reduce = [&__comp](_ForwardIterator __lhs, _ForwardIterator __rhs) {
+ return __comp(*__rhs, *__lhs) ? __rhs : __lhs;
+ };
+
+ // Perform a parallel reduction of iterators [first+1, last) with 'first' as init.
+ return __statpart::__transform_reduce(
+ std::move(__first),
+ std::move(__last),
+ __identity{}, // No transformations
+ std::move(__init), // Use the first iterator as the init element
+ __iter_reduce, // Reduction of 2 elements
+ [&__iter_reduce, &__comp](auto __brick_first, auto __brick_last, auto __brick_init) {
+ // Reduction of an iterator range + init element: use the serial version to find the minimum among
+ // the iterators and then reduce with the init element.
+ return __iter_reduce(__brick_init, std::min_element(__brick_first, __brick_last, __comp));
+ });
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::min_element(std::move(__first), std::move(__last), std::move(__comp));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __minmax_element<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Compare>
+ _LIBCPP_HIDE_FROM_ABI optional<pair<_ForwardIterator, _ForwardIterator>>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ using _IterPair = pair<_ForwardIterator, _ForwardIterator>;
+
+ if (__first == __last) {
+ return _IterPair{__last, __last}; // Nothing to do
+ }
+
+ _IterPair __init = {__first, __first};
+ ++__first;
+ if (__first == __last) {
+ return __init; // The only element is both the minimum and the maximum
+ }
+
+ // A reduction that returns a pair of iterators pointing to the minimum and maximum elements.
+ // In a case of a tie the minimum iterators are biased left and the maximum iterators are biased right.
+ auto __iter_reduce = [&__comp](_IterPair __lhs, _IterPair __rhs) {
+ return _IterPair{__comp(*__rhs.first, *__lhs.first) ? __rhs.first : __lhs.first,
+ __comp(*__rhs.second, *__lhs.second) ? __lhs.second : __rhs.second};
+ };
+
+ // Perform a parallel reduction of iterators [first+1, last) with {first, first} as init.
+ return __statpart::__transform_reduce(
+ std::move(__first),
+ std::move(__last),
+ [](auto __it) { return _IterPair{__it, __it}; }, // Transform an iterator into an iterator pair
+ std::move(__init), // Use the pair of first iterators as the init element
+ __iter_reduce, // Reduction of 2 minmax pairs
+ [&__iter_reduce, &__comp](auto __brick_first, auto __brick_last, auto __brick_init) {
+ // Reduction of an iterator range + init element: use the serial version to find the minmax among
+ // the iterators and then reduce it with the init element.
+ return __iter_reduce(__brick_init, std::minmax_element(__brick_first, __brick_last, __comp));
+ });
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::minmax_element(std::move(__first), std::move(__last), std::move(__comp));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __mismatch<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
+ _LIBCPP_HIDE_FROM_ABI optional<pair<_ForwardIterator1, _ForwardIterator2>>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _Predicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ // Look for a mismatch only in the prefix of the two ranges.
+ auto __n = std::min(__last1 - __first1, __last2 - __first2);
+ // Find a position in the first range where the predicate is false against the corresponding position in the
+ // second range.
+ auto __res = __statpart::__find(
+ __first1,
+ __first1 + __n,
+ [&__pred, __first1, __first2](_ForwardIterator1 __brick_first1, _ForwardIterator1 __brick_last1) {
+ // Run the sequential mismatch algorithm on these ranges:
+ // [__brick_first1, __brick_last1) and
+ // [__first2 + (__brick_first1 - __first1), __first2 + (__brick_last1 - __first1))
+ auto __brick_first2 = __first2 + (__brick_first1 - __first1);
+ return std::mismatch(std::move(__brick_first1), std::move(__brick_last1), std::move(__brick_first2), __pred)
+ .first;
+ },
+ less<>{}, // `less` here means the lowest index among the mismatches
+ true // `true` here means we want the first mismatch, not the last
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ auto __idx = *__res - __first1;
+ return pair<_ForwardIterator1, _ForwardIterator2>{std::move(*__res), __first2 + __idx};
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ // Unsequenced execution is also implicitly covered by the sequential implementation.
+ return std::mismatch(
+ std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __reverse<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _BidirectionalIterator>
+ _LIBCPP_HIDE_FROM_ABI optional<__empty>
+ operator()(_Policy&&, _BidirectionalIterator __first, _BidirectionalIterator __last) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_BidirectionalIterator>::value) {
+ // Perform a chunked for_each on the first half of the range.
+ // Odd-sized ranges will leave the middle element untouched, which is correct for this algorithm:
+ // its reversed position is the same.
+ auto __n = (__last - __first) / 2;
+ return __statpart::__for_each(
+ __first, __first + __n, [__first, __last](_BidirectionalIterator __i, _BidirectionalIterator __j) {
+ // Derive the last position of the mirrored range by counting from the end.
+ _BidirectionalIterator __mirror_last = __last - (__i - __first);
+ // Swap the elements in the range of the first half with their mirrored counterparts in the second half.
+ std::swap_ranges(std::move(__i),
+ std::move(__j),
+ std::reverse_iterator<_BidirectionalIterator>(std::move(__mirror_last)));
+ });
+ } else {
+ // Non-random access iterators currently cannot be processed in parallel, use the sequential implementation.
+ std::reverse(std::move(__first), std::move(__last));
+ return __empty{};
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __search<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator1>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _BinaryPredicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
+ _DifferenceType __size2 = __last2 - __first2; // The length of the needle to search for.
+ if (__size2 == 0) {
+ return __first1; // If the needle length is zero, the first iterator is returned.
+ }
+ _DifferenceType __size1 = __last1 - __first1;
+ if (__size1 < __size2) {
+ return __last1; // The range is too small to contain the requested number of consecutive elements.
+ }
+ // Calculate the length of the tail where a potential match cannot start by definition.
+ _DifferenceType __crop = __size2 - 1;
+ // We're only interested in the range where a potential match can start: [first, last - crop)
+ _ForwardIterator1 __last1_cropped = __last1 - __crop;
+ // Run a parallel chunked find_if, covering the range where a potential match can start.
+ auto __res = __statpart::__find(
+ __first1,
+ __last1_cropped,
+ [__first2, __last2, __crop, &__pred](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
+ // Uncrop the range to allow std::search to find a full match, which can go beyond __brick_last.
+ _ForwardIterator1 __brick_last_uncropped = __brick_last + __crop;
+ // Run a serial std::search inside each of the chunks in parallel.
+ _ForwardIterator1 __ret = std::search(__brick_first, __brick_last_uncropped, __first2, __last2, __pred);
+ // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
+ // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __find().
+ return __ret == __brick_last_uncropped ? __brick_last : __ret;
+ },
+ less<>{}, // `less` here means the lowest index among the matches
+ true // `true` here means we want the first match, not the last
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ if (*__res == __last1_cropped) {
+ return __last1; // No match was found in the range.
+ }
+ return *__res; // Return the successful match.
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::search(
+ std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __search_n<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Size, class _Tp, class _Predicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
+ operator()(_Policy&&,
+ _ForwardIterator __first,
+ _ForwardIterator __last,
+ _Size __count,
+ const _Tp& __value,
+ _Predicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ typedef typename std::iterator_traits<_ForwardIterator>::difference_type _DifferenceType;
+ _DifferenceType __integral_count = std::__convert_to_integral(__count);
+ if (__integral_count <= 0) {
+ return __first; // If the count is non-positive, the first iterator is returned.
+ }
+ _DifferenceType __size = __last - __first;
+ if (__size < __integral_count) {
+ return __last; // The range is too small to contain the requested number of consecutive elements.
+ }
+ // Calculate the length of the tail where a potential match cannot start by definition.
+ _DifferenceType __crop = __integral_count - 1;
+ // We're only interested in the range where a potential match can start: [first, last - crop)
+ _ForwardIterator __last2 = __last - __crop;
+ // Run a parallel chunked find_if, covering the range where a potential match can start.
+ auto __res = __statpart::__find(
+ __first,
+ __last2,
+ [__integral_count, __crop, &__value, &__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ // Uncrop the range to allow std::search_n to find a full match, which can go beyond __brick_last.
+ _ForwardIterator __brick_last_uncropped = __brick_last + __crop;
+ // Run a serial std::search_n inside each of the chunks in parallel.
+ _ForwardIterator __ret =
+ std::search_n(__brick_first, __brick_last_uncropped, __integral_count, __value, __pred);
+ // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
+ // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __find().
+ return __ret == __brick_last_uncropped ? __brick_last : __ret;
+ },
+ less<>{}, // `less` here means the lowest index among the matches
+ true // `true` here means we want the first match, not the last
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ if (*__res == __last2) {
+ return __last; // No match was found in the range.
+ }
+ return *__res; // Return the successful match.
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::search_n(std::move(__first), std::move(__last), __count, __value, std::move(__pred));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __stable_sort<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _RandomAccessIterator, class _Comp>
+ _LIBCPP_HIDE_FROM_ABI optional<__empty>
+ operator()(_Policy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy>) {
+ return __statpart::__stable_sort(
+ __first, __last, __comp, [](_RandomAccessIterator __g_first, _RandomAccessIterator __g_last, _Comp __g_comp) {
+ std::stable_sort(__g_first, __g_last, __g_comp);
+ });
+ } else {
+ std::stable_sort(__first, __last, __comp);
+ return __empty{};
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __swap_ranges<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
+ _Policy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ auto __res = __statpart::__for_each_iter_pair(
+ __first1,
+ __last1,
+ __first2,
+ [](_ForwardIterator1 __brick_first1, _ForwardIterator1 __brick_last1, _ForwardIterator2 __brick_first2) {
+ std::swap_ranges(std::move(__brick_first1), std::move(__brick_last1), std::move(__brick_first2));
+ });
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ return __first2 + (__last1 - __first1);
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::swap_ranges(std::move(__first1), std::move(__last1), std::move(__first2));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __transform<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _ForwardOutIterator, class _UnaryOperation>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator> operator()(
+ _Policy&&, _ForwardIterator __first, _ForwardIterator __last, _ForwardOutIterator __result, _UnaryOperation __op)
+ const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
+ __statpart::__for_each(
+ __first, __last, [__op, __first, __result](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ return std::transform(__brick_first, __brick_last, __result + (__brick_first - __first), __op);
+ });
+ return __result + (__last - __first);
+ } else {
+ return std::transform(__first, __last, __result, __op);
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __transform_binary<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy,
+ class _ForwardIterator1,
+ class _ForwardIterator2,
+ class _ForwardOutIterator,
+ class _BinaryOperation>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardOutIterator __result,
+ _BinaryOperation __op) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
+ auto __res = __statpart::__for_each(
+ __first1,
+ __last1,
+ [__op, __first1, __first2, __result](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
+ return std::transform(
+ __brick_first,
+ __brick_last,
+ __first2 + (__brick_first - __first1),
+ __result + (__brick_first - __first1),
+ __op);
+ });
+ if (!__res)
+ return nullopt;
+ return __result + (__last1 - __first1);
+ } else {
+ return std::transform(__first1, __last1, __first2, __result, __op);
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __transform_reduce<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation>
+ _LIBCPP_HIDE_FROM_ABI optional<_Tp>
+ operator()(_Policy&&,
+ _ForwardIterator __first,
+ _ForwardIterator __last,
+ _Tp __init,
+ _BinaryOperation __reduce,
+ _UnaryOperation __transform) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __statpart::__transform_reduce(
+ std::move(__first),
+ std::move(__last),
+ [__transform](_ForwardIterator __iter) { return __transform(*__iter); },
+ std::move(__init),
+ __reduce,
+ [__transform, __reduce](auto __brick_first, auto __brick_last, _Tp __brick_init) {
+ return std::transform_reduce(
+ std::move(__brick_first),
+ std::move(__brick_last),
+ std::move(__brick_init),
+ std::move(__reduce),
+ std::move(__transform));
+ });
+ } else {
+ return std::transform_reduce(
+ std::move(__first), std::move(__last), std::move(__init), std::move(__reduce), std::move(__transform));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __transform_reduce_binary<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy,
+ class _ForwardIterator1,
+ class _ForwardIterator2,
+ class _Tp,
+ class _BinaryOperation1,
+ class _BinaryOperation2>
+ _LIBCPP_HIDE_FROM_ABI optional<_Tp> operator()(
+ _Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _Tp __init,
+ _BinaryOperation1 __reduce,
+ _BinaryOperation2 __transform) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ return __statpart::__transform_reduce(
+ __first1,
+ std::move(__last1),
+ [__first1, __first2, __transform](_ForwardIterator1 __iter) {
+ return __transform(*__iter, *(__first2 + (__iter - __first1)));
+ },
+ std::move(__init),
+ std::move(__reduce),
+ [__first1, __first2, __reduce, __transform](
+ _ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _Tp __brick_init) {
+ return std::transform_reduce(
+ __brick_first,
+ std::move(__brick_last),
+ __first2 + (__brick_first - __first1),
+ std::move(__brick_init),
+ std::move(__reduce),
+ std::move(__transform));
+ });
+ } else {
+ return std::transform_reduce(
+ std::move(__first1),
+ std::move(__last1),
+ std::move(__first2),
+ std::move(__init),
+ std::move(__reduce),
+ std::move(__transform));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __uninitialized_copy<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
+ _Policy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ auto __res = __statpart::__for_each_iter_pair(
+ __first,
+ __last,
+ __result,
+ [](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _ForwardIterator2 __brick_result) {
+ std::uninitialized_copy(std::move(__brick_first), std::move(__brick_last), std::move(__brick_result));
+ });
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ return __result + (__last - __first);
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::uninitialized_copy(std::move(__first), std::move(__last), std::move(__result));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __uninitialized_move<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
+ _Policy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ auto __res = __statpart::__for_each_iter_pair(
+ __first,
+ __last,
+ __result,
+ [](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _ForwardIterator2 __brick_result) {
+ std::uninitialized_move(std::move(__brick_first), std::move(__brick_last), std::move(__brick_result));
+ });
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ return __result + (__last - __first);
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::uninitialized_move(std::move(__first), std::move(__last), std::move(__result));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 28ae9ee8de46a3..87a1ef5ed8292f 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2416,6 +2416,9 @@ module std {
header "__pstl/backends/serial.h"
export std_core.utility_core.empty
}
+ module statpart {
+ header "__pstl/backends/statpart.h"
+ }
module std_thread {
header "__pstl/backends/std_thread.h"
export std.pstl.cpu_algos
diff --git a/libcxx/src/CMakeLists.txt b/libcxx/src/CMakeLists.txt
index e029f9f29d5ea2..d49dd69e66f95b 100644
--- a/libcxx/src/CMakeLists.txt
+++ b/libcxx/src/CMakeLists.txt
@@ -310,6 +310,12 @@ if (LIBCXX_PSTL_BACKEND STREQUAL "libdispatch")
)
endif()
+if (LIBCXX_PSTL_BACKEND STREQUAL "statpart")
+ list(APPEND LIBCXX_EXPERIMENTAL_SOURCES
+ pstl/statpart.cpp
+ )
+endif()
+
if (LIBCXX_ENABLE_LOCALIZATION AND LIBCXX_ENABLE_FILESYSTEM AND LIBCXX_ENABLE_TIME_ZONE_DATABASE)
list(APPEND LIBCXX_EXPERIMENTAL_SOURCES
experimental/include/tzdb/time_zone_private.h
diff --git a/libcxx/src/pstl/statpart.cpp b/libcxx/src/pstl/statpart.cpp
new file mode 100644
index 00000000000000..903ebbf000128e
--- /dev/null
+++ b/libcxx/src/pstl/statpart.cpp
@@ -0,0 +1,26 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include <__config>
+#include <__pstl/backends/statpart.h>
+#include <dispatch/dispatch.h>
+
+// Hardcoded to use libdispatch for now.
+// CMake switches to be introduced later.
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl::__statpart {
+
+void __apply(size_t chunk_count, void* context, void (*func)(void* context, size_t chunk)) noexcept {
+ ::dispatch_apply_f(chunk_count, DISPATCH_APPLY_AUTO, context, func);
+}
+
+} // namespace __pstl::__statpart
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
diff --git a/libcxx/test/libcxx/transitive_includes/cxx17.csv b/libcxx/test/libcxx/transitive_includes/cxx17.csv
index e0a92259202c48..5db6a12c1dd278 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx17.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx17.csv
@@ -1,4 +1,5 @@
algorithm cctype
+algorithm cerrno
algorithm climits
algorithm compare
algorithm cstdint
@@ -528,6 +529,7 @@ map tuple
map version
mdspan version
memory cctype
+memory cerrno
memory climits
memory compare
memory cstdint
@@ -564,6 +566,7 @@ mutex version
new version
numbers version
numeric cctype
+numeric cerrno
numeric climits
numeric compare
numeric cstdint
@@ -640,6 +643,7 @@ queue tuple
queue vector
queue version
random cctype
+random cerrno
random climits
random cmath
random compare
diff --git a/libcxx/test/libcxx/transitive_includes/cxx20.csv b/libcxx/test/libcxx/transitive_includes/cxx20.csv
index 3c5e9439f100cc..1e40f6abcadd6a 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx20.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx20.csv
@@ -1,4 +1,5 @@
algorithm cctype
+algorithm cerrno
algorithm climits
algorithm compare
algorithm cstdint
@@ -594,6 +595,7 @@ map tuple
map version
mdspan version
memory cctype
+memory cerrno
memory climits
memory compare
memory cstdint
@@ -630,6 +632,7 @@ mutex version
new version
numbers version
numeric cctype
+numeric cerrno
numeric climits
numeric compare
numeric cstdint
@@ -716,6 +719,7 @@ queue tuple
queue vector
queue version
random cctype
+random cerrno
random climits
random cmath
random compare
diff --git a/libcxx/test/libcxx/transitive_includes/cxx23.csv b/libcxx/test/libcxx/transitive_includes/cxx23.csv
index 8c03f1c3c2251e..f3c6c4c9796553 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx23.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx23.csv
@@ -1,4 +1,5 @@
algorithm cctype
+algorithm cerrno
algorithm climits
algorithm compare
algorithm cstdint
@@ -628,6 +629,7 @@ mdspan span
mdspan stdexcept
mdspan version
memory cctype
+memory cerrno
memory climits
memory compare
memory cstdint
@@ -664,6 +666,7 @@ mutex version
new version
numbers version
numeric cctype
+numeric cerrno
numeric climits
numeric compare
numeric cstdint
@@ -758,6 +761,7 @@ queue tuple
queue vector
queue version
random cctype
+random cerrno
random climits
random cmath
random compare
diff --git a/libcxx/test/libcxx/transitive_includes/cxx26.csv b/libcxx/test/libcxx/transitive_includes/cxx26.csv
index 800d65aed4704c..269a32ca1befe0 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx26.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx26.csv
@@ -1,4 +1,5 @@
algorithm cctype
+algorithm cerrno
algorithm climits
algorithm compare
algorithm cstdint
@@ -620,6 +621,7 @@ mdspan span
mdspan stdexcept
mdspan version
memory cctype
+memory cerrno
memory climits
memory compare
memory cstdint
@@ -656,6 +658,7 @@ mutex version
new version
numbers version
numeric cctype
+numeric cerrno
numeric climits
numeric compare
numeric cstdint
@@ -743,6 +746,7 @@ queue tuple
queue vector
queue version
random cctype
+random cerrno
random climits
random compare
random cstdint
>From eb04b45b236442822aae0a80373023ea13274fd5 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Fri, 9 Oct 2026 16:32:03 +0100
Subject: [PATCH 02/24] Update includes for C++29 as well
---
libcxx/test/libcxx/transitive_includes/cxx29.csv | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/libcxx/test/libcxx/transitive_includes/cxx29.csv b/libcxx/test/libcxx/transitive_includes/cxx29.csv
index 800d65aed4704c..269a32ca1befe0 100644
--- a/libcxx/test/libcxx/transitive_includes/cxx29.csv
+++ b/libcxx/test/libcxx/transitive_includes/cxx29.csv
@@ -1,4 +1,5 @@
algorithm cctype
+algorithm cerrno
algorithm climits
algorithm compare
algorithm cstdint
@@ -620,6 +621,7 @@ mdspan span
mdspan stdexcept
mdspan version
memory cctype
+memory cerrno
memory climits
memory compare
memory cstdint
@@ -656,6 +658,7 @@ mutex version
new version
numbers version
numeric cctype
+numeric cerrno
numeric climits
numeric compare
numeric cstdint
@@ -743,6 +746,7 @@ queue tuple
queue vector
queue version
random cctype
+random cerrno
random climits
random compare
random cstdint
>From 738efff6945043e3b64ecef455f549887cbe8d64 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Fri, 9 Oct 2026 17:47:27 +0100
Subject: [PATCH 03/24] Use the statpart backend on Apple platforms to run it
against CI
---
libcxx/CMakeLists.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/libcxx/CMakeLists.txt b/libcxx/CMakeLists.txt
index 5478c38afa781d..55b7a5fc709b37 100644
--- a/libcxx/CMakeLists.txt
+++ b/libcxx/CMakeLists.txt
@@ -321,7 +321,7 @@ option(LIBCXX_HAS_EXTERNAL_THREAD_API
option(LIBCXX_HAS_C11_THREAD_API "Ignore auto-detection and force use of C11 thread API" OFF)
if (APPLE)
- set(LIBCXX_PSTL_BACKEND_DEFAULT_VALUE "libdispatch")
+ set(LIBCXX_PSTL_BACKEND_DEFAULT_VALUE "statpart")
elseif(LIBCXX_ENABLE_THREADS)
set(LIBCXX_PSTL_BACKEND_DEFAULT_VALUE "std_thread")
else()
>From 3a57228a1c21d6c2985fc5a25103a9b09a80ee09 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Mon, 12 Oct 2026 13:27:31 +0100
Subject: [PATCH 04/24] Polished __partition() a bit, added some basic unit
tests
---
libcxx/include/__pstl/backends/statpart.h | 25 +--
.../pstl.statpart.partition.pass.cpp | 181 ++++++++++++++++++
.../libcxx/test/features/libcxx_macros.py | 1 +
3 files changed, 195 insertions(+), 12 deletions(-)
create mode 100644 libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 41e473d663f3ed..715027c6bd706a 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -121,17 +121,18 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
template <class _Iter>
friend optional<__iterator_partitions<_Iter>>
- __partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t) noexcept;
+ __partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
};
template <class _ForwardIterator>
-_LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>>
-__partition(_ForwardIterator __first,
- _ForwardIterator __last,
- __partitioning_result& __result,
- size_t __min_partition_size = 1,
- size_t __prefix_to_skip = 0,
- size_t __suffix_to_skip = 0) noexcept {
+_LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partition(
+ _ForwardIterator __first,
+ _ForwardIterator __last,
+ __partitioning_result& __result,
+ size_t __min_partition_size = 1,
+ size_t __prefix_to_skip = 0,
+ size_t __suffix_to_skip = 0,
+ size_t __workers_count = thread::hardware_concurrency()) noexcept {
// forward iteratos to be supported later
static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
// TODO: assert( __min_partition_size > 0 )?
@@ -143,7 +144,6 @@ __partition(_ForwardIterator __first,
}
size_t __effective_count = __count - __prefix_to_skip - __suffix_to_skip;
- size_t __workers_count = thread::hardware_concurrency();
size_t __max_partitions = __workers_count * __max_partitions_per_cpu;
size_t __partition_count = std::min(__max_partitions, __effective_count / __min_partition_size);
@@ -155,10 +155,11 @@ __partition(_ForwardIterator __first,
__iterator_partitions<_ForwardIterator> __p;
__p.__base_ = __first + __prefix_to_skip;
- __p.__count_ = __count;
- __p.__partition_size_ = __count / __partition_count;
- __p.__remainder_ = __count % __partition_count;
+ __p.__count_ = __effective_count;
+ __p.__partition_size_ = __effective_count / __partition_count;
+ __p.__remainder_ = __effective_count % __partition_count;
__p.__partitions_count_ = __partition_count;
+ __result = __partitioning_result::__ok;
return __p;
}
diff --git a/libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp b/libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp
new file mode 100644
index 00000000000000..e9c973b94e7a23
--- /dev/null
+++ b/libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp
@@ -0,0 +1,181 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+// <algorithm>
+
+// REQUIRES: std-at-least-c++17
+// REQUIRES: libcpp-pstl-backend-statpart
+// UNSUPPORTED: libcpp-has-no-incomplete-pstl
+
+// template <class Iterator>
+// optional<__iterator_partitions<Iterator>>
+// __partition(Iterator __first,
+// Iterator __last,
+// __partitioning_result& __result,
+// size_t __min_partition_size,
+// size_t __prefix_to_skip,
+// size_t __suffix_to_skip,
+// size_t __workers_count)
+
+#include <__pstl/backends/statpart.h>
+#include <algorithm>
+#include <iterator>
+#include <cassert>
+#include <cstddef>
+#include <vector>
+
+template <class Container>
+void test_partition() {
+ auto part_eq = [](const auto& partition, const auto& container, size_t first_pos, size_t last_pos) {
+ return partition.__first == std::next(container.begin(), first_pos) &&
+ partition.__last == std::next(container.begin(), last_pos);
+ };
+
+ { // 1 element, 8 workers: no partitions
+ Container v(1);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 0, 0, 8);
+ assert(partitions == std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__not_enough_elements);
+ }
+ { // 2 elements, 8 workers: 2 partitions
+ // [0..1), [1..2)
+ Container v(2);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 0, 0, 8);
+ assert(partitions != std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
+ assert(partitions->__partitions_count() == 2);
+ assert(partitions->__end() == v.end());
+ assert(part_eq(partitions->__partition(0), v, 0, 1));
+ assert(part_eq(partitions->__partition(1), v, 1, 2));
+ }
+ { // 2 elements, min_size=2, 8 workers: no partitions
+ // [0..1), [1..2)
+ Container v(2);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 2, 0, 8);
+ assert(partitions == std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__not_enough_elements);
+ }
+ { // 20 elements, 2 workers: 16 partitions
+ // [0..2), [2..4), [4..6), [6..8), [8..9), [9..10), ..., [19, 20)
+ Container v(20);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 0, 0, 2);
+ assert(partitions != std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
+ assert(partitions->__partitions_count() == 16);
+ assert(partitions->__end() == v.end());
+ assert(part_eq(partitions->__partition(0), v, 0, 2));
+ assert(part_eq(partitions->__partition(1), v, 2, 4));
+ assert(part_eq(partitions->__partition(2), v, 4, 6));
+ assert(part_eq(partitions->__partition(3), v, 6, 8));
+ assert(part_eq(partitions->__partition(4), v, 8, 9));
+ assert(part_eq(partitions->__partition(5), v, 9, 10));
+ assert(part_eq(partitions->__partition(6), v, 10, 11));
+ assert(part_eq(partitions->__partition(7), v, 11, 12));
+ assert(part_eq(partitions->__partition(8), v, 12, 13));
+ assert(part_eq(partitions->__partition(9), v, 13, 14));
+ assert(part_eq(partitions->__partition(10), v, 14, 15));
+ assert(part_eq(partitions->__partition(11), v, 15, 16));
+ assert(part_eq(partitions->__partition(12), v, 16, 17));
+ assert(part_eq(partitions->__partition(13), v, 17, 18));
+ assert(part_eq(partitions->__partition(14), v, 18, 19));
+ assert(part_eq(partitions->__partition(15), v, 19, 20));
+ }
+ { // 20 elements, min_size=2, 2 workers: 10 partitions
+ // [0..2), [2..4), [4..6), ..., [18, 20)
+ Container v(20);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 2, 0, 0, 2);
+ assert(partitions != std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
+ assert(partitions->__partitions_count() == 10);
+ assert(partitions->__end() == v.end());
+ assert(part_eq(partitions->__partition(0), v, 0, 2));
+ assert(part_eq(partitions->__partition(1), v, 2, 4));
+ assert(part_eq(partitions->__partition(2), v, 4, 6));
+ assert(part_eq(partitions->__partition(3), v, 6, 8));
+ assert(part_eq(partitions->__partition(4), v, 8, 10));
+ assert(part_eq(partitions->__partition(5), v, 10, 12));
+ assert(part_eq(partitions->__partition(6), v, 12, 14));
+ assert(part_eq(partitions->__partition(7), v, 14, 16));
+ assert(part_eq(partitions->__partition(8), v, 16, 18));
+ assert(part_eq(partitions->__partition(9), v, 18, 20));
+ }
+ { // 20 elements, min_size=3, 2 workers: 6 partitions
+ // [0..4), [4..8), [8..11), [11..14), [14..17), [17..20)
+ Container v(20);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 0, 0, 2);
+ assert(partitions != std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
+ assert(partitions->__partitions_count() == 6);
+ assert(partitions->__end() == v.end());
+ assert(part_eq(partitions->__partition(0), v, 0, 4));
+ assert(part_eq(partitions->__partition(1), v, 4, 8));
+ assert(part_eq(partitions->__partition(2), v, 8, 11));
+ assert(part_eq(partitions->__partition(3), v, 11, 14));
+ assert(part_eq(partitions->__partition(4), v, 14, 17));
+ assert(part_eq(partitions->__partition(5), v, 17, 20));
+ }
+ { // 20 elements, min_size=3, prefix_skip=1, 2 workers: 6 partitions
+ // [1..5), [5..8), [8..11), [11..14), [14..17), [17..20)
+ Container v(20);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 1, 0, 2);
+ assert(partitions != std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
+ assert(partitions->__partitions_count() == 6);
+ assert(partitions->__end() == v.end());
+ assert(part_eq(partitions->__partition(0), v, 1, 5));
+ assert(part_eq(partitions->__partition(1), v, 5, 8));
+ assert(part_eq(partitions->__partition(2), v, 8, 11));
+ assert(part_eq(partitions->__partition(3), v, 11, 14));
+ assert(part_eq(partitions->__partition(4), v, 14, 17));
+ assert(part_eq(partitions->__partition(5), v, 17, 20));
+ }
+ { // 20 elements, min_size=3, suffix_skip=1, 2 workers: 6 partitions
+ // [0..4), [4..8), [8..11), [11..14), [14..17), [17..20)
+ Container v(20);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 0, 1, 2);
+ assert(partitions != std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
+ assert(partitions->__partitions_count() == 6);
+ assert(partitions->__end() == std::prev(v.end(), 1));
+ assert(part_eq(partitions->__partition(0), v, 0, 4));
+ assert(part_eq(partitions->__partition(1), v, 4, 7));
+ assert(part_eq(partitions->__partition(2), v, 7, 10));
+ assert(part_eq(partitions->__partition(3), v, 10, 13));
+ assert(part_eq(partitions->__partition(4), v, 13, 16));
+ assert(part_eq(partitions->__partition(5), v, 16, 19));
+ }
+ { // 20 elements, min_size=3, prefix_skip=1, suffix_skip=1, 2 workers: 6 partitions
+ // [0..4), [4..8), [8..11), [11..14), [14..17), [17..20)
+ Container v(20);
+ std::__pstl::__statpart::__partitioning_result result;
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 1, 1, 2);
+ assert(partitions != std::nullopt);
+ assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
+ assert(partitions->__partitions_count() == 6);
+ assert(partitions->__end() == std::prev(v.end(), 1));
+ assert(part_eq(partitions->__partition(0), v, 1, 4));
+ assert(part_eq(partitions->__partition(1), v, 4, 7));
+ assert(part_eq(partitions->__partition(2), v, 7, 10));
+ assert(part_eq(partitions->__partition(3), v, 10, 13));
+ assert(part_eq(partitions->__partition(4), v, 13, 16));
+ assert(part_eq(partitions->__partition(5), v, 16, 19));
+ }
+}
+
+int main(int, char**) {
+ test_partition<std::vector<int>>();
+ // later: test_partition<std::list<int>>();
+}
diff --git a/libcxx/utils/libcxx/test/features/libcxx_macros.py b/libcxx/utils/libcxx/test/features/libcxx_macros.py
index 33d569e75e6bcb..2feeeb3eb6a879 100644
--- a/libcxx/utils/libcxx/test/features/libcxx_macros.py
+++ b/libcxx/utils/libcxx/test/features/libcxx_macros.py
@@ -34,6 +34,7 @@
"_LIBCPP_DEPRECATED_ABI_DISABLE_PAIR_TRIVIAL_COPY_CTOR": "libcpp-deprecated-abi-disable-pair-trivial-copy-ctor",
"_LIBCPP_ABI_NO_COMPRESSED_PAIR_PADDING": "libcpp-abi-no-compressed-pair-padding",
"_LIBCPP_PSTL_BACKEND_LIBDISPATCH": "libcpp-pstl-backend-libdispatch",
+ "_LIBCPP_PSTL_BACKEND_STATPART": "libcpp-pstl-backend-statpart",
"_LIBCPP_ABI_USE_SMALL_DEQUE_BLOCK_SIZE": "libcpp-abi-use-small-deque-block-size",
}
for macro, feature in macros.items():
>From f3e9747fe8327102293afb45c07625f563bd9de8 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Mon, 12 Oct 2026 13:43:07 +0100
Subject: [PATCH 05/24] More comments
---
libcxx/include/__pstl/backends/statpart.h | 23 ++++++++++++++++++-----
1 file changed, 18 insertions(+), 5 deletions(-)
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 715027c6bd706a..11a1264fe6b117 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -72,7 +72,7 @@ _LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
namespace __statpart {
-// The backbone for the backend: applies the function to each of the indices [0, __chunk_count] in parallel.
+// The backbone of the backend: applies the function to each of the indices [0, __chunk_count] in parallel.
// The actual number of parallel workers and the order of application is unspecified.
_LIBCPP_EXPORTED_FROM_ABI void
__apply(size_t __chunk_count, void* __context, void (*__func)(void* __context, size_t __chunk)) noexcept;
@@ -99,7 +99,9 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_range {
_Iterator __last;
};
-// Later to support forward iterators via a second implicit template argument.
+// __iterator_partitions<Iter> holds a finite per-determined set of partitions and allows to query each iterator range
+// in O(1) time.
+// TODO: Later to support forward iterators via a second implicit template argument.
template <class _Iterator>
struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
_LIBCPP_HIDE_FROM_ABI size_t __partitions_count() const { return __partitions_count_; }
@@ -124,6 +126,14 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
__partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
};
+// Splits the input iterator range into a finite set of partitions, such as that every iterator sub-range can queried in
+// a O(1) time later.
+// In case partitioning failed returns nullopt and reports the reason in __result.
+// __min_partition_size specifies the minimal number of elements in every partition.
+// __prefix_to_skip and __suffix_to_skip allow to optionally trim the input range by specified numbers of elements
+// before partitioning.
+// __workers_count is normally implicitly queried from hardware_concurrency() and should be explicitly specified only in
+// unit tests.
template <class _ForwardIterator>
_LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partition(
_ForwardIterator __first,
@@ -136,6 +146,7 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
// forward iteratos to be supported later
static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
// TODO: assert( __min_partition_size > 0 )?
+ // TODO: assert( __workers_count > 0 )?
size_t __count = static_cast<size_t>(std::distance(__first, __last));
if (__count < __prefix_to_skip + __suffix_to_skip) {
@@ -163,7 +174,8 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
return __p;
}
-// for_each with explicit partitioning
+// for_each with explicit partitioning:
+// calls __func in parallel with every iterator ranges provided in __partitions
template <class _Iterator, class _Functor>
_LIBCPP_HIDE_FROM_ABI void __for_each_partition(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
struct __ctx_t {
@@ -185,14 +197,15 @@ _LIBCPP_HIDE_FROM_ABI void __for_each_index(size_t __count, _Functor __func) {
});
}
-// TODO: ...
+// for_each that calls __func in parallel against every partition index available in __partitions
template <class _Iterator, class _Functor>
_LIBCPP_HIDE_FROM_ABI void
__for_each_partition_index(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
__statpart::__for_each_index(__partitions.__partitions_count(), std::move(__func));
}
-// for_each with default implicit partitioning
+// for_each with default implicit partitioning:
+// calls __func in parallel with every element in [__first, __last), provided as iterator ranges.
template <class _Iterator, class _Functor>
_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each(_Iterator __first, _Iterator __last, _Functor __func) {
if (__first == __last)
>From ed528d4c619311a46d19cefd4cff063a4d9eee27 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Mon, 12 Oct 2026 18:13:20 +0100
Subject: [PATCH 06/24] Restructuring the statpart backend for easier
maintenance
---
libcxx/include/CMakeLists.txt | 3 +-
libcxx/include/__pstl/backends/statpart.h | 434 +--------------
.../include/__pstl/backends/statpart/basis.h | 494 ++++++++++++++++++
libcxx/include/module.modulemap.in | 5 +
4 files changed, 503 insertions(+), 433 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/basis.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index da2546a54bc04d..5b99cc3ba58565 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -695,7 +695,8 @@ set(files
__pstl/backends/default.h
__pstl/backends/libdispatch.h
__pstl/backends/serial.h
- __pstl/backends/statpart.h
+ __pstl/backends/statpart.h
+ __pstl/backends/statpart/basis.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 11a1264fe6b117..788bb713c4fae6 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -9,6 +9,8 @@
#ifndef _LIBCPP___PSTL_STATPART_H
#define _LIBCPP___PSTL_STATPART_H
+#include <__pstl/backends/statpart/basis.h>
+
#include <__algorithm/find_end.h>
#include <__algorithm/find_if.h>
#include <__algorithm/for_each.h>
@@ -70,438 +72,6 @@ _LIBCPP_PUSH_MACROS
_LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
-namespace __statpart {
-
-// The backbone of the backend: applies the function to each of the indices [0, __chunk_count] in parallel.
-// The actual number of parallel workers and the order of application is unspecified.
-_LIBCPP_EXPORTED_FROM_ABI void
-__apply(size_t __chunk_count, void* __context, void (*__func)(void* __context, size_t __chunk)) noexcept;
-
-// The maximum number of partitions to be processed per each CPU
-constexpr inline size_t __max_partitions_per_cpu = 8;
-
-// The minimum number of elements to sort serially in a single partition
-constexpr inline size_t __min_sort_partition_size = 256;
-
-enum class __partitioning_result {
- // returned when the partitioning result is not nullopt
- __ok,
- // returned when there is not enough element to produce a single partition or the final number of partitions is less
- // than 2, i.e. there's no reason to process in parallel
- __not_enough_elements,
- // later: returned memory couldn't be allocated for the stashed forward iterators
- __bad_alloc
-};
-
-template <class _Iterator>
-struct _LIBCPP_HIDE_FROM_ABI __iterator_range {
- _Iterator __first;
- _Iterator __last;
-};
-
-// __iterator_partitions<Iter> holds a finite per-determined set of partitions and allows to query each iterator range
-// in O(1) time.
-// TODO: Later to support forward iterators via a second implicit template argument.
-template <class _Iterator>
-struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
- _LIBCPP_HIDE_FROM_ABI size_t __partitions_count() const { return __partitions_count_; }
-
- _LIBCPP_HIDE_FROM_ABI __iterator_range<_Iterator> __partition(size_t __n) const {
- _Iterator __first = __base_ + __n * __partition_size_ + (__n < __remainder_ ? __n : __remainder_);
- _Iterator __last = __first + __partition_size_ + (__n < __remainder_ ? 1 : 0);
- return __iterator_range<_Iterator>{__first, __last};
- }
-
- _LIBCPP_HIDE_FROM_ABI _Iterator __end() const { return __base_ + __count_; }
-
-private:
- _Iterator __base_;
- size_t __count_;
- size_t __partition_size_;
- size_t __remainder_;
- size_t __partitions_count_;
-
- template <class _Iter>
- friend optional<__iterator_partitions<_Iter>>
- __partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
-};
-
-// Splits the input iterator range into a finite set of partitions, such as that every iterator sub-range can queried in
-// a O(1) time later.
-// In case partitioning failed returns nullopt and reports the reason in __result.
-// __min_partition_size specifies the minimal number of elements in every partition.
-// __prefix_to_skip and __suffix_to_skip allow to optionally trim the input range by specified numbers of elements
-// before partitioning.
-// __workers_count is normally implicitly queried from hardware_concurrency() and should be explicitly specified only in
-// unit tests.
-template <class _ForwardIterator>
-_LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partition(
- _ForwardIterator __first,
- _ForwardIterator __last,
- __partitioning_result& __result,
- size_t __min_partition_size = 1,
- size_t __prefix_to_skip = 0,
- size_t __suffix_to_skip = 0,
- size_t __workers_count = thread::hardware_concurrency()) noexcept {
- // forward iteratos to be supported later
- static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
- // TODO: assert( __min_partition_size > 0 )?
- // TODO: assert( __workers_count > 0 )?
-
- size_t __count = static_cast<size_t>(std::distance(__first, __last));
- if (__count < __prefix_to_skip + __suffix_to_skip) {
- __result = __partitioning_result::__not_enough_elements;
- return {};
- }
-
- size_t __effective_count = __count - __prefix_to_skip - __suffix_to_skip;
- size_t __max_partitions = __workers_count * __max_partitions_per_cpu;
- size_t __partition_count = std::min(__max_partitions, __effective_count / __min_partition_size);
-
- if (__partition_count < 2) {
- // TODO: verify that's what we want
- __result = __partitioning_result::__not_enough_elements;
- return {}; // not enough input data to process in parallel => no reason to build partitions
- }
-
- __iterator_partitions<_ForwardIterator> __p;
- __p.__base_ = __first + __prefix_to_skip;
- __p.__count_ = __effective_count;
- __p.__partition_size_ = __effective_count / __partition_count;
- __p.__remainder_ = __effective_count % __partition_count;
- __p.__partitions_count_ = __partition_count;
- __result = __partitioning_result::__ok;
- return __p;
-}
-
-// for_each with explicit partitioning:
-// calls __func in parallel with every iterator ranges provided in __partitions
-template <class _Iterator, class _Functor>
-_LIBCPP_HIDE_FROM_ABI void __for_each_partition(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
- struct __ctx_t {
- const __iterator_partitions<_Iterator>& __partitions_;
- _Functor& __func_;
- } __ctx{__partitions, __func};
- __statpart::__apply(__partitions.__partitions_count(), &__ctx, [](void* __context, size_t __partition) {
- __ctx_t& __ctx = *static_cast<__ctx_t*>(__context);
- auto __range = __ctx.__partitions_.__partition(__partition);
- __ctx.__func_(__range.__first, __range.__last);
- });
-}
-
-// Calls __func(index) for every index in [0, __count) in parallel
-template <class _Functor>
-_LIBCPP_HIDE_FROM_ABI void __for_each_index(size_t __count, _Functor __func) {
- __statpart::__apply(__count, &__func, [](void* __context, size_t __index) {
- (*static_cast<_Functor*>(__context))(__index);
- });
-}
-
-// for_each that calls __func in parallel against every partition index available in __partitions
-template <class _Iterator, class _Functor>
-_LIBCPP_HIDE_FROM_ABI void
-__for_each_partition_index(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
- __statpart::__for_each_index(__partitions.__partitions_count(), std::move(__func));
-}
-
-// for_each with default implicit partitioning:
-// calls __func in parallel with every element in [__first, __last), provided as iterator ranges.
-template <class _Iterator, class _Functor>
-_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each(_Iterator __first, _Iterator __last, _Functor __func) {
- if (__first == __last)
- return __empty{};
- __partitioning_result __partition_result;
- auto __partitions = __statpart::__partition(__first, __last, __partition_result);
- if (__partitions) {
- __statpart::__for_each_partition(*__partitions, std::move(__func));
- return __empty{};
- } else if (__partition_result == __partitioning_result::__bad_alloc) {
- return nullopt;
- } else {
- __func(__first, __last);
- return __empty{};
- }
-}
-
-// Applies the 3-legged function to sub-ranges of the input ranges in parallel:
-// f(sub_range1_first, sub_range1_last, sub_range2_first)
-template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BrickFunction>
-_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each_iter_pair(
- _RandomAccessIterator1 __first1,
- _RandomAccessIterator1 __last1,
- _RandomAccessIterator2 __first2,
- _BrickFunction __f) {
- return __for_each(
- __first1,
- __last1,
- [__first1, __first2, __f = std::move(__f)](
- _RandomAccessIterator1 __brick_first1, _RandomAccessIterator1 __brick_last1) {
- _RandomAccessIterator2 __brick_first2 = __first2 + (__brick_first1 - __first1);
- __f(__brick_first1, __brick_last1, __brick_first2);
- });
-}
-
-template <class _Index, class _Brick, class _Compare>
-_LIBCPP_HIDE_FROM_ABI optional<_Index>
-__find(_Index __first, _Index __last, _Brick __f, _Compare __comp, bool __b_first) {
- typedef typename std::iterator_traits<_Index>::difference_type _DifferenceType;
- const _DifferenceType __n = __last - __first;
- _DifferenceType __initial_dist = __b_first ? __n : -1;
- std::atomic<_DifferenceType> __extremum(__initial_dist);
- // TODO: find out what is better here: parallel_for or parallel_reduce
- auto __res = __statpart::__for_each(__first, __last, [__comp, __f, __first, &__extremum](_Index __i, _Index __j) {
- // See "Reducing Contention Through Priority Updates", PPoPP '13, for discussion of
- // why using a shared variable scales fairly well in this situation.
- if (__comp(__i - __first, __extremum)) {
- _Index __result = __f(__i, __j);
- // If not '__last' returned then we found what we want so put this to extremum
- if (__result != __j) {
- const _DifferenceType __k = __result - __first;
- for (_DifferenceType __old = __extremum; __comp(__k, __old); __old = __extremum) {
- __extremum.compare_exchange_weak(__old, __k);
- }
- }
- }
- });
- if (!__res)
- return nullopt;
- return __extremum.load() != __initial_dist ? __first + __extremum.load() : __last;
-}
-
-template <class _RandomAccessIterator, class _Transform, class _Value, class _Combiner, class _Reduction>
-_LIBCPP_HIDE_FROM_ABI optional<_Value> __transform_reduce(
- _RandomAccessIterator __first,
- _RandomAccessIterator __last,
- _Transform __transform,
- _Value __init,
- _Combiner __combiner,
- _Reduction __reduction) {
- if (__first == __last)
- return __init;
-
- // Split into partitions with at least 3 elements in each
- __partitioning_result __partitioning_result;
- auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, 3);
- if (!__partitions) {
- if (__partitioning_result == __partitioning_result::__bad_alloc) {
- // Failed to allocate memory for the partitions, propagate the error up.
- return nullopt;
- } else {
- // Not enough elements to run in parallel, switch to serial implementation.
- return __reduction(std::move(__first), std::move(__last), std::move(__init));
- }
- }
-
- auto __destroy = [__count = __partitions->__partitions_count()](_Value* __ptr) {
- std::destroy_n(__ptr, __count);
- std::allocator<_Value>().deallocate(__ptr, __count);
- };
-
- // TODO: use __uninitialized_buffer
- // TODO: allocate one element per worker instead of one element per chunk
- unique_ptr<_Value[], decltype(__destroy)> __values(
- std::allocator<_Value>().allocate(__partitions->__partitions_count()), __destroy);
-
- // __dispatch_apply is noexcept
- __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
- auto __range = __partitions->__partition(__index);
- std::__construct_at(
- __values.get() + __index,
- __reduction(__range.__first + 2,
- __range.__last,
- __combiner(__transform(__range.__first), __transform(__range.__first + 1))));
- });
-
- return std::reduce(
- std::make_move_iterator(__values.get()),
- std::make_move_iterator(__values.get() + __partitions->__partitions_count()),
- std::move(__init),
- __combiner);
-}
-
-template <class _RandomAccessIterator1,
- class _RandomAccessIterator2,
- class _RandomAccessIterator3,
- class _Compare,
- class _LeafMerge>
-_LIBCPP_HIDE_FROM_ABI optional<__empty>
-__merge(_RandomAccessIterator1 __first1,
- _RandomAccessIterator1 __last1,
- _RandomAccessIterator2 __first2,
- _RandomAccessIterator2 __last2,
- _RandomAccessIterator3 __result,
- _Compare __comp,
- _LeafMerge __leaf_merge) noexcept {
- // Run the merge such that the longest range is partitioned.
- // __merge_sub_ranges swaps the ranges implicitly in case the second range was the longest one.
- auto __run =
- [__result](auto __long_first,
- auto __long_last,
- auto __short_first,
- auto __short_last,
- auto __find_cutpoint,
- auto __merge_sub_ranges) -> optional<__empty> {
- // Partition the longest input range evenly.
- __partitioning_result __partitioning_result;
- auto __partitions = __statpart::__partition(__long_first, __long_last, __partitioning_result);
- if (!__partitions) {
- if (__partitioning_result == __partitioning_result::__bad_alloc)
- return nullopt;
- // Not enough elements to run in parallel, merge everything serially.
- __merge_sub_ranges(__long_first, __long_last, __short_first, __short_last, __result);
- return __empty{};
- }
-
- // Process the partitions of the longest range in parallel.
- __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
- // Get the sub-range inside the longer input range.
- auto __range = __partitions->__partition(__index);
-
- // Find the matching sub-range of the shorter input range: it will hold the elements that come after the last
- // element of the previous long partition and up to the last element of this partition.
- // This generally runs two binary searches in the smaller input range, edge partitions use existing bounds.
- auto __short_begin =
- __index == 0 ? __short_first : __find_cutpoint(__short_first, __short_last, __range.__first[-1]);
- auto __short_end =
- __index == __partitions->__partitions_count() - 1
- ? __short_last
- : __find_cutpoint(__short_first, __short_last, __range.__last[-1]);
-
- // Derive the position in the output range and merge the sub-ranges.
- auto __result_first = __result + (__range.__first - __long_first) + (__short_begin - __short_first);
- __merge_sub_ranges(__range.__first, __range.__last, __short_begin, __short_end, __result_first);
- });
- return __empty{};
- };
-
- if (__last1 - __first1 > __last2 - __first2) {
- // Preserve the order of the input ranges, partition the first range.
- // Equal elements of the second range go after the cut => use lower_bound when cutting.
- return __run(
- __first1,
- __last1,
- __first2,
- __last2,
- [&](auto __first, auto __last, const auto& __value) {
- return std::lower_bound(__first, __last, __value, __comp);
- },
- [&](auto __long_first, auto __long_last, auto __short_first, auto __short_last, auto __result_first) {
- __leaf_merge(__long_first, __long_last, __short_first, __short_last, __result_first, __comp);
- });
- } else {
- // Swap the order of the input ranges, partition the second range.
- // Equal elements of the first range go before the cut => use upper_bound when cutting.
- return __run(
- __first2,
- __last2,
- __first1,
- __last1,
- [&](auto __first, auto __last, const auto& __value) {
- return std::upper_bound(__first, __last, __value, __comp);
- },
- [&](auto __long_first, auto __long_last, auto __short_first, auto __short_last, auto __result_first) {
- __leaf_merge(__short_first, __short_last, __long_first, __long_last, __result_first, __comp);
- });
- }
-}
-
-template <class _RandomAccessIterator, class _Comp, class _LeafSort>
-_LIBCPP_HIDE_FROM_ABI optional<__empty>
-__stable_sort(_RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp, _LeafSort __leaf_sort) {
- const auto __size = __last - __first;
-
- __partitioning_result __partitioning_result;
- auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, __min_sort_partition_size);
- if (!__partitions) {
- if (__partitioning_result == __partitioning_result::__bad_alloc)
- return nullopt;
- // Not enough elements to run in parallel, sort everything serially.
- __leaf_sort(__first, __last, __comp);
- return __empty{};
- }
-
- using _Value = __iterator_value_type<_RandomAccessIterator>;
-
- auto __destroy = [__size](_Value* __ptr) {
- std::destroy_n(__ptr, __size);
- std::allocator<_Value>().deallocate(__ptr, __size);
- };
-
- // TODO: use __uninitialized_buffer
- unique_ptr<_Value[], decltype(__destroy)> __values(std::allocator<_Value>().allocate(__size), __destroy);
-
- // Initialize all elements to a moved-from state
- // TODO: Don't do this - this can be done in the first merge - see https://llvm.org/PR63928
- std::__construct_at(__values.get(), std::move(*__first));
- for (__iterator_difference_type<_RandomAccessIterator> __i = 1; __i != __size; ++__i) {
- std::__construct_at(__values.get() + __i, std::move(__values.get()[__i - 1]));
- }
- *__first = std::move(__values.get()[__size - 1]);
-
- __statpart::__for_each_partition(
- *__partitions, [&__leaf_sort, &__comp](_RandomAccessIterator __chunk_first, _RandomAccessIterator __chunk_last) {
- __leaf_sort(std::move(__chunk_first), std::move(__chunk_last), __comp);
- });
-
- // A chunk span is a run of __chunk_span_size consecutive partitions, the last chunk can be shorter in case of odd
- // number of chunk spans. Each pass merges the chunk spans pairwise, so the chunk span size doubles and the chunk
- // count is halved (rounded up).
- const size_t __partition_count = __partitions->__partitions_count();
- size_t __chunk_span_size = 1;
- size_t __chunk_spans = __partition_count;
-
- // Offsets of the beginning and the end of the chunk from __first
- auto __chunk_begin = [&](size_t __chunk) {
- return __partitions->__partition(__chunk * __chunk_span_size).__first - __first;
- };
- auto __chunk_end = [&](size_t __chunk) {
- return __partitions->__partition(std::min((__chunk + 1) * __chunk_span_size, __partition_count) - 1).__last -
- __first;
- };
-
- bool __objects_are_in_buffer = false;
- do {
- size_t __chunk_spans_pairs = (__chunk_spans + 1) / 2;
- auto __merge_chunks = [&](auto __from_first, auto __to_first) {
- __statpart::__for_each_index(__chunk_spans_pairs, [&](size_t __span_pair_index) {
- size_t __left_chunk = 2 * __span_pair_index;
- size_t __right_chunk = __left_chunk + 1;
- auto __pair_first = __chunk_begin(__left_chunk);
- if (__right_chunk == __chunk_spans) {
- // The odd chunk has no counterpart to merge with => carry it over to the other buffer as-is.
- std::move(__from_first + __pair_first, __from_first + __chunk_end(__left_chunk), __to_first + __pair_first);
- } else {
- // Merge the two chunk spans into the other buffer
- auto __pair_mid = __chunk_begin(__right_chunk);
- auto __pair_last = __chunk_end(__right_chunk);
- std::merge(std::make_move_iterator(__from_first + __pair_first),
- std::make_move_iterator(__from_first + __pair_mid),
- std::make_move_iterator(__from_first + __pair_mid),
- std::make_move_iterator(__from_first + __pair_last),
- __to_first + __pair_first,
- __comp);
- }
- });
- };
-
- if (__objects_are_in_buffer)
- __merge_chunks(__values.get(), __first);
- else
- __merge_chunks(__first, __values.get());
- __objects_are_in_buffer = !__objects_are_in_buffer;
-
- __chunk_span_size *= 2;
- __chunk_spans = __chunk_spans_pairs;
- } while (__chunk_spans > 1);
-
- if (__objects_are_in_buffer) {
- std::move(__values.get(), __values.get() + __size, __first);
- }
-
- return __empty{};
-}
-
-} // namespace __statpart
template <class _ExecutionPolicy>
struct __find_end<__statpart_backend_tag, _ExecutionPolicy> {
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
new file mode 100644
index 00000000000000..709eee0504bc4e
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -0,0 +1,494 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART__BASIS_H
+#define _LIBCPP___PSTL_STATPART__BASIS_H
+
+#include <__algorithm/lower_bound.h>
+#include <__algorithm/merge.h>
+#include <__algorithm/min.h>
+#include <__algorithm/move.h>
+#include <__algorithm/upper_bound.h>
+#include <__atomic/atomic.h>
+#include <__config>
+#include <__functional/identity.h>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/distance.h>
+#include <__iterator/iterator_traits.h>
+#include <__iterator/move_iterator.h>
+#include <__memory/allocator.h>
+#include <__memory/construct_at.h>
+#include <__memory/destroy.h>
+#include <__memory/unique_ptr.h>
+#include <__numeric/reduce.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__thread/thread.h>
+#include <__type_traits/is_assignable.h>
+#include <__type_traits/is_constructible.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__type_traits/is_trivially_copyable.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+namespace __statpart {
+
+// The backbone of the backend: applies the function to each of the indices [0, __chunk_count] in parallel.
+// The actual number of parallel workers and the order of application is unspecified.
+_LIBCPP_EXPORTED_FROM_ABI void
+__apply(size_t __chunk_count, void* __context, void (*__func)(void* __context, size_t __chunk)) noexcept;
+
+// The maximum number of partitions to be processed per each CPU
+constexpr inline size_t __max_partitions_per_cpu = 8;
+
+// The minimum number of elements to sort serially in a single partition
+constexpr inline size_t __min_sort_partition_size = 256;
+
+enum class __partitioning_result {
+ // returned when the partitioning result is not nullopt
+ __ok,
+ // returned when there is not enough element to produce a single partition or the final number of partitions is less
+ // than 2, i.e. there's no reason to process in parallel
+ __not_enough_elements,
+ // later: returned memory couldn't be allocated for the stashed forward iterators
+ __bad_alloc
+};
+
+template <class _Iterator>
+struct _LIBCPP_HIDE_FROM_ABI __iterator_range {
+ _Iterator __first;
+ _Iterator __last;
+};
+
+// __iterator_partitions<Iter> holds a finite per-determined set of partitions and allows to query each iterator range
+// in O(1) time.
+// TODO: Later to support forward iterators via a second implicit template argument.
+template <class _Iterator>
+struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
+ _LIBCPP_HIDE_FROM_ABI size_t __partitions_count() const { return __partitions_count_; }
+
+ _LIBCPP_HIDE_FROM_ABI __iterator_range<_Iterator> __partition(size_t __n) const {
+ _Iterator __first = __base_ + __n * __partition_size_ + (__n < __remainder_ ? __n : __remainder_);
+ _Iterator __last = __first + __partition_size_ + (__n < __remainder_ ? 1 : 0);
+ return __iterator_range<_Iterator>{__first, __last};
+ }
+
+ _LIBCPP_HIDE_FROM_ABI _Iterator __end() const { return __base_ + __count_; }
+
+private:
+ _Iterator __base_;
+ size_t __count_;
+ size_t __partition_size_;
+ size_t __remainder_;
+ size_t __partitions_count_;
+
+ template <class _Iter>
+ friend optional<__iterator_partitions<_Iter>>
+ __partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
+};
+
+// Splits the input iterator range into a finite set of partitions, such as that every iterator sub-range can queried in
+// a O(1) time later.
+// In case partitioning failed returns nullopt and reports the reason in __result.
+// __min_partition_size specifies the minimal number of elements in every partition.
+// __prefix_to_skip and __suffix_to_skip allow to optionally trim the input range by specified numbers of elements
+// before partitioning.
+// __workers_count is normally implicitly queried from hardware_concurrency() and should be explicitly specified only in
+// unit tests.
+template <class _ForwardIterator>
+_LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partition(
+ _ForwardIterator __first,
+ _ForwardIterator __last,
+ __partitioning_result& __result,
+ size_t __min_partition_size = 1,
+ size_t __prefix_to_skip = 0,
+ size_t __suffix_to_skip = 0,
+ size_t __workers_count = thread::hardware_concurrency()) noexcept {
+ // forward iteratos to be supported later
+ static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
+ // TODO: assert( __min_partition_size > 0 )?
+ // TODO: assert( __workers_count > 0 )?
+
+ size_t __count = static_cast<size_t>(std::distance(__first, __last));
+ if (__count < __prefix_to_skip + __suffix_to_skip) {
+ __result = __partitioning_result::__not_enough_elements;
+ return {};
+ }
+
+ size_t __effective_count = __count - __prefix_to_skip - __suffix_to_skip;
+ size_t __max_partitions = __workers_count * __max_partitions_per_cpu;
+ size_t __partition_count = std::min(__max_partitions, __effective_count / __min_partition_size);
+
+ if (__partition_count < 2) {
+ // TODO: verify that's what we want
+ __result = __partitioning_result::__not_enough_elements;
+ return {}; // not enough input data to process in parallel => no reason to build partitions
+ }
+
+ __iterator_partitions<_ForwardIterator> __p;
+ __p.__base_ = __first + __prefix_to_skip;
+ __p.__count_ = __effective_count;
+ __p.__partition_size_ = __effective_count / __partition_count;
+ __p.__remainder_ = __effective_count % __partition_count;
+ __p.__partitions_count_ = __partition_count;
+ __result = __partitioning_result::__ok;
+ return __p;
+}
+
+// for_each with explicit partitioning:
+// calls __func in parallel with every iterator ranges provided in __partitions
+template <class _Iterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI void __for_each_partition(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
+ struct __ctx_t {
+ const __iterator_partitions<_Iterator>& __partitions_;
+ _Functor& __func_;
+ } __ctx{__partitions, __func};
+ __statpart::__apply(__partitions.__partitions_count(), &__ctx, [](void* __context, size_t __partition) {
+ __ctx_t& __ctx = *static_cast<__ctx_t*>(__context);
+ auto __range = __ctx.__partitions_.__partition(__partition);
+ __ctx.__func_(__range.__first, __range.__last);
+ });
+}
+
+// Calls __func(index) for every index in [0, __count) in parallel
+template <class _Functor>
+_LIBCPP_HIDE_FROM_ABI void __for_each_index(size_t __count, _Functor __func) {
+ __statpart::__apply(__count, &__func, [](void* __context, size_t __index) {
+ (*static_cast<_Functor*>(__context))(__index);
+ });
+}
+
+// for_each that calls __func in parallel against every partition index available in __partitions
+template <class _Iterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI void
+__for_each_partition_index(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
+ __statpart::__for_each_index(__partitions.__partitions_count(), std::move(__func));
+}
+
+// for_each with default implicit partitioning:
+// calls __func in parallel with every element in [__first, __last), provided as iterator ranges.
+template <class _Iterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each(_Iterator __first, _Iterator __last, _Functor __func) {
+ if (__first == __last)
+ return __empty{};
+ __partitioning_result __partition_result;
+ auto __partitions = __statpart::__partition(__first, __last, __partition_result);
+ if (__partitions) {
+ __statpart::__for_each_partition(*__partitions, std::move(__func));
+ return __empty{};
+ } else if (__partition_result == __partitioning_result::__bad_alloc) {
+ return nullopt;
+ } else {
+ __func(__first, __last);
+ return __empty{};
+ }
+}
+
+// Applies the 3-legged function to sub-ranges of the input ranges in parallel:
+// f(sub_range1_first, sub_range1_last, sub_range2_first)
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BrickFunction>
+_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each_iter_pair(
+ _RandomAccessIterator1 __first1,
+ _RandomAccessIterator1 __last1,
+ _RandomAccessIterator2 __first2,
+ _BrickFunction __f) {
+ return __for_each(
+ __first1,
+ __last1,
+ [__first1, __first2, __f = std::move(__f)](
+ _RandomAccessIterator1 __brick_first1, _RandomAccessIterator1 __brick_last1) {
+ _RandomAccessIterator2 __brick_first2 = __first2 + (__brick_first1 - __first1);
+ __f(__brick_first1, __brick_last1, __brick_first2);
+ });
+}
+
+template <class _Index, class _Brick, class _Compare>
+_LIBCPP_HIDE_FROM_ABI optional<_Index>
+__find(_Index __first, _Index __last, _Brick __f, _Compare __comp, bool __b_first) {
+ typedef typename std::iterator_traits<_Index>::difference_type _DifferenceType;
+ const _DifferenceType __n = __last - __first;
+ _DifferenceType __initial_dist = __b_first ? __n : -1;
+ std::atomic<_DifferenceType> __extremum(__initial_dist);
+ // TODO: find out what is better here: parallel_for or parallel_reduce
+ auto __res = __statpart::__for_each(__first, __last, [__comp, __f, __first, &__extremum](_Index __i, _Index __j) {
+ // See "Reducing Contention Through Priority Updates", PPoPP '13, for discussion of
+ // why using a shared variable scales fairly well in this situation.
+ if (__comp(__i - __first, __extremum)) {
+ _Index __result = __f(__i, __j);
+ // If not '__last' returned then we found what we want so put this to extremum
+ if (__result != __j) {
+ const _DifferenceType __k = __result - __first;
+ for (_DifferenceType __old = __extremum; __comp(__k, __old); __old = __extremum) {
+ __extremum.compare_exchange_weak(__old, __k);
+ }
+ }
+ }
+ });
+ if (!__res)
+ return nullopt;
+ return __extremum.load() != __initial_dist ? __first + __extremum.load() : __last;
+}
+
+template <class _RandomAccessIterator, class _Transform, class _Value, class _Combiner, class _Reduction>
+_LIBCPP_HIDE_FROM_ABI optional<_Value> __transform_reduce(
+ _RandomAccessIterator __first,
+ _RandomAccessIterator __last,
+ _Transform __transform,
+ _Value __init,
+ _Combiner __combiner,
+ _Reduction __reduction) {
+ if (__first == __last)
+ return __init;
+
+ // Split into partitions with at least 3 elements in each
+ __partitioning_result __partitioning_result;
+ auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, 3);
+ if (!__partitions) {
+ if (__partitioning_result == __partitioning_result::__bad_alloc) {
+ // Failed to allocate memory for the partitions, propagate the error up.
+ return nullopt;
+ } else {
+ // Not enough elements to run in parallel, switch to serial implementation.
+ return __reduction(std::move(__first), std::move(__last), std::move(__init));
+ }
+ }
+
+ auto __destroy = [__count = __partitions->__partitions_count()](_Value* __ptr) {
+ std::destroy_n(__ptr, __count);
+ std::allocator<_Value>().deallocate(__ptr, __count);
+ };
+
+ // TODO: use __uninitialized_buffer
+ // TODO: allocate one element per worker instead of one element per chunk
+ unique_ptr<_Value[], decltype(__destroy)> __values(
+ std::allocator<_Value>().allocate(__partitions->__partitions_count()), __destroy);
+
+ // __dispatch_apply is noexcept
+ __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
+ auto __range = __partitions->__partition(__index);
+ std::__construct_at(
+ __values.get() + __index,
+ __reduction(__range.__first + 2,
+ __range.__last,
+ __combiner(__transform(__range.__first), __transform(__range.__first + 1))));
+ });
+
+ return std::reduce(
+ std::make_move_iterator(__values.get()),
+ std::make_move_iterator(__values.get() + __partitions->__partitions_count()),
+ std::move(__init),
+ __combiner);
+}
+
+template <class _RandomAccessIterator1,
+ class _RandomAccessIterator2,
+ class _RandomAccessIterator3,
+ class _Compare,
+ class _LeafMerge>
+_LIBCPP_HIDE_FROM_ABI optional<__empty>
+__merge(_RandomAccessIterator1 __first1,
+ _RandomAccessIterator1 __last1,
+ _RandomAccessIterator2 __first2,
+ _RandomAccessIterator2 __last2,
+ _RandomAccessIterator3 __result,
+ _Compare __comp,
+ _LeafMerge __leaf_merge) noexcept {
+ // Run the merge such that the longest range is partitioned.
+ // __merge_sub_ranges swaps the ranges implicitly in case the second range was the longest one.
+ auto __run =
+ [__result](auto __long_first,
+ auto __long_last,
+ auto __short_first,
+ auto __short_last,
+ auto __find_cutpoint,
+ auto __merge_sub_ranges) -> optional<__empty> {
+ // Partition the longest input range evenly.
+ __partitioning_result __partitioning_result;
+ auto __partitions = __statpart::__partition(__long_first, __long_last, __partitioning_result);
+ if (!__partitions) {
+ if (__partitioning_result == __partitioning_result::__bad_alloc)
+ return nullopt;
+ // Not enough elements to run in parallel, merge everything serially.
+ __merge_sub_ranges(__long_first, __long_last, __short_first, __short_last, __result);
+ return __empty{};
+ }
+
+ // Process the partitions of the longest range in parallel.
+ __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
+ // Get the sub-range inside the longer input range.
+ auto __range = __partitions->__partition(__index);
+
+ // Find the matching sub-range of the shorter input range: it will hold the elements that come after the last
+ // element of the previous long partition and up to the last element of this partition.
+ // This generally runs two binary searches in the smaller input range, edge partitions use existing bounds.
+ auto __short_begin =
+ __index == 0 ? __short_first : __find_cutpoint(__short_first, __short_last, __range.__first[-1]);
+ auto __short_end =
+ __index == __partitions->__partitions_count() - 1
+ ? __short_last
+ : __find_cutpoint(__short_first, __short_last, __range.__last[-1]);
+
+ // Derive the position in the output range and merge the sub-ranges.
+ auto __result_first = __result + (__range.__first - __long_first) + (__short_begin - __short_first);
+ __merge_sub_ranges(__range.__first, __range.__last, __short_begin, __short_end, __result_first);
+ });
+ return __empty{};
+ };
+
+ if (__last1 - __first1 > __last2 - __first2) {
+ // Preserve the order of the input ranges, partition the first range.
+ // Equal elements of the second range go after the cut => use lower_bound when cutting.
+ return __run(
+ __first1,
+ __last1,
+ __first2,
+ __last2,
+ [&](auto __first, auto __last, const auto& __value) {
+ return std::lower_bound(__first, __last, __value, __comp);
+ },
+ [&](auto __long_first, auto __long_last, auto __short_first, auto __short_last, auto __result_first) {
+ __leaf_merge(__long_first, __long_last, __short_first, __short_last, __result_first, __comp);
+ });
+ } else {
+ // Swap the order of the input ranges, partition the second range.
+ // Equal elements of the first range go before the cut => use upper_bound when cutting.
+ return __run(
+ __first2,
+ __last2,
+ __first1,
+ __last1,
+ [&](auto __first, auto __last, const auto& __value) {
+ return std::upper_bound(__first, __last, __value, __comp);
+ },
+ [&](auto __long_first, auto __long_last, auto __short_first, auto __short_last, auto __result_first) {
+ __leaf_merge(__short_first, __short_last, __long_first, __long_last, __result_first, __comp);
+ });
+ }
+}
+
+template <class _RandomAccessIterator, class _Comp, class _LeafSort>
+_LIBCPP_HIDE_FROM_ABI optional<__empty>
+__stable_sort(_RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp, _LeafSort __leaf_sort) {
+ const auto __size = __last - __first;
+
+ __partitioning_result __partitioning_result;
+ auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, __min_sort_partition_size);
+ if (!__partitions) {
+ if (__partitioning_result == __partitioning_result::__bad_alloc)
+ return nullopt;
+ // Not enough elements to run in parallel, sort everything serially.
+ __leaf_sort(__first, __last, __comp);
+ return __empty{};
+ }
+
+ using _Value = __iterator_value_type<_RandomAccessIterator>;
+
+ auto __destroy = [__size](_Value* __ptr) {
+ std::destroy_n(__ptr, __size);
+ std::allocator<_Value>().deallocate(__ptr, __size);
+ };
+
+ // TODO: use __uninitialized_buffer
+ unique_ptr<_Value[], decltype(__destroy)> __values(std::allocator<_Value>().allocate(__size), __destroy);
+
+ // Initialize all elements to a moved-from state
+ // TODO: Don't do this - this can be done in the first merge - see https://llvm.org/PR63928
+ std::__construct_at(__values.get(), std::move(*__first));
+ for (__iterator_difference_type<_RandomAccessIterator> __i = 1; __i != __size; ++__i) {
+ std::__construct_at(__values.get() + __i, std::move(__values.get()[__i - 1]));
+ }
+ *__first = std::move(__values.get()[__size - 1]);
+
+ __statpart::__for_each_partition(
+ *__partitions, [&__leaf_sort, &__comp](_RandomAccessIterator __chunk_first, _RandomAccessIterator __chunk_last) {
+ __leaf_sort(std::move(__chunk_first), std::move(__chunk_last), __comp);
+ });
+
+ // A chunk span is a run of __chunk_span_size consecutive partitions, the last chunk can be shorter in case of odd
+ // number of chunk spans. Each pass merges the chunk spans pairwise, so the chunk span size doubles and the chunk
+ // count is halved (rounded up).
+ const size_t __partition_count = __partitions->__partitions_count();
+ size_t __chunk_span_size = 1;
+ size_t __chunk_spans = __partition_count;
+
+ // Offsets of the beginning and the end of the chunk from __first
+ auto __chunk_begin = [&](size_t __chunk) {
+ return __partitions->__partition(__chunk * __chunk_span_size).__first - __first;
+ };
+ auto __chunk_end = [&](size_t __chunk) {
+ return __partitions->__partition(std::min((__chunk + 1) * __chunk_span_size, __partition_count) - 1).__last -
+ __first;
+ };
+
+ bool __objects_are_in_buffer = false;
+ do {
+ size_t __chunk_spans_pairs = (__chunk_spans + 1) / 2;
+ auto __merge_chunks = [&](auto __from_first, auto __to_first) {
+ __statpart::__for_each_index(__chunk_spans_pairs, [&](size_t __span_pair_index) {
+ size_t __left_chunk = 2 * __span_pair_index;
+ size_t __right_chunk = __left_chunk + 1;
+ auto __pair_first = __chunk_begin(__left_chunk);
+ if (__right_chunk == __chunk_spans) {
+ // The odd chunk has no counterpart to merge with => carry it over to the other buffer as-is.
+ std::move(__from_first + __pair_first, __from_first + __chunk_end(__left_chunk), __to_first + __pair_first);
+ } else {
+ // Merge the two chunk spans into the other buffer
+ auto __pair_mid = __chunk_begin(__right_chunk);
+ auto __pair_last = __chunk_end(__right_chunk);
+ std::merge(std::make_move_iterator(__from_first + __pair_first),
+ std::make_move_iterator(__from_first + __pair_mid),
+ std::make_move_iterator(__from_first + __pair_mid),
+ std::make_move_iterator(__from_first + __pair_last),
+ __to_first + __pair_first,
+ __comp);
+ }
+ });
+ };
+
+ if (__objects_are_in_buffer)
+ __merge_chunks(__values.get(), __first);
+ else
+ __merge_chunks(__first, __values.get());
+ __objects_are_in_buffer = !__objects_are_in_buffer;
+
+ __chunk_span_size *= 2;
+ __chunk_spans = __chunk_spans_pairs;
+ } while (__chunk_spans > 1);
+
+ if (__objects_are_in_buffer) {
+ std::move(__values.get(), __values.get() + __size, __first);
+ }
+
+ return __empty{};
+}
+
+} // namespace __statpart
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART__BASIS_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 87a1ef5ed8292f..153d67fc05d54d 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2417,6 +2417,11 @@ module std {
export std_core.utility_core.empty
}
module statpart {
+ module basis {
+ header "__pstl/backends/statpart/basis.h"
+ }
+ }
+ module statpart_old {
header "__pstl/backends/statpart.h"
}
module std_thread {
>From 1b3a65f26b0161a438e52ba90a4e005ee457e953 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Tue, 13 Oct 2026 10:10:18 +0100
Subject: [PATCH 07/24] Moved find_end to a separate file
---
libcxx/include/CMakeLists.txt | 1 +
libcxx/include/__pstl/backends/statpart.h | 58 +---------
.../include/__pstl/backends/statpart/basis.h | 6 +-
.../__pstl/backends/statpart/find_end.h | 102 ++++++++++++++++++
libcxx/include/module.modulemap.in | 3 +
5 files changed, 110 insertions(+), 60 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/find_end.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index 5b99cc3ba58565..c7830a12eaf009 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -697,6 +697,7 @@ set(files
__pstl/backends/serial.h
__pstl/backends/statpart.h
__pstl/backends/statpart/basis.h
+ __pstl/backends/statpart/find_end.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 788bb713c4fae6..1ac299caeae200 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -10,6 +10,7 @@
#define _LIBCPP___PSTL_STATPART_H
#include <__pstl/backends/statpart/basis.h>
+#include <__pstl/backends/statpart/find_end.h>
#include <__algorithm/find_end.h>
#include <__algorithm/find_if.h>
@@ -73,63 +74,6 @@ _LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
-template <class _ExecutionPolicy>
-struct __find_end<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator1>
- operator()(_Policy&&,
- _ForwardIterator1 __first1,
- _ForwardIterator1 __last1,
- _ForwardIterator2 __first2,
- _ForwardIterator2 __last2,
- _BinaryPredicate __pred) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
- typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
- _DifferenceType __size2 = __last2 - __first2; // The length of the needle to search for.
- if (__size2 == 0) {
- return __last1; // If the needle length is zero, the last iterator is returned.
- }
- _DifferenceType __size1 = __last1 - __first1;
- if (__size1 < __size2) {
- return __last1; // The range is too small to contain the requested number of consecutive elements.
- }
- // Calculate the length of the tail where a potential match cannot start by definition.
- _DifferenceType __crop = __size2 - 1;
- // We're only interested in the range where a potential match can start: [first, last - crop)
- _ForwardIterator1 __last1_cropped = __last1 - __crop;
- // Run a parallel chunked find_if, covering the range where a potential match can start.
- auto __res = __statpart::__find(
- __first1,
- __last1_cropped,
- [__first2, __last2, __crop, &__pred](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
- // Uncrop the range to allow std::find_end to find a full match, which can go beyond __brick_last.
- _ForwardIterator1 __brick_last_uncropped = __brick_last + __crop;
- // Run a serial std::find_end inside each of the chunks in parallel.
- _ForwardIterator1 __ret = std::find_end(__brick_first, __brick_last_uncropped, __first2, __last2, __pred);
- // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
- // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __parallel_find().
- return __ret == __brick_last_uncropped ? __brick_last : __ret;
- },
- greater<>{}, // `greater` here means the highest index among the matches
- false // `false` here means we want the last match, not the first
- );
- if (!__res) {
- return std::nullopt; // Failed to run the algorithm, propagate the error.
- }
- if (*__res == __last1_cropped) {
- return __last1; // No match was found in the range.
- }
- return *__res; // Return the successful match.
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::find_end(
- std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
- }
- }
-};
-
template <class _ExecutionPolicy>
struct __for_each<__statpart_backend_tag, _ExecutionPolicy> {
template <class _Policy, class _ForwardIterator, class _Function>
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index 709eee0504bc4e..75dc2a5e159dcf 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART__BASIS_H
-#define _LIBCPP___PSTL_STATPART__BASIS_H
+#ifndef _LIBCPP___PSTL_STATPART_BASIS_H
+#define _LIBCPP___PSTL_STATPART_BASIS_H
#include <__algorithm/lower_bound.h>
#include <__algorithm/merge.h>
@@ -491,4 +491,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART__BASIS_H
+#endif // _LIBCPP___PSTL_STATPART_BASIS_H
diff --git a/libcxx/include/__pstl/backends/statpart/find_end.h b/libcxx/include/__pstl/backends/statpart/find_end.h
new file mode 100644
index 00000000000000..2f022024077258
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/find_end.h
@@ -0,0 +1,102 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_FIND_END_H
+#define _LIBCPP___PSTL_STATPART_FIND_END_H
+
+#include <__algorithm/find_end.h>
+#include <__config>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __find_end<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator1>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _BinaryPredicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
+ _DifferenceType __size2 = __last2 - __first2; // The length of the needle to search for.
+ if (__size2 == 0) {
+ return __last1; // If the needle length is zero, the last iterator is returned.
+ }
+ _DifferenceType __size1 = __last1 - __first1;
+ if (__size1 < __size2) {
+ return __last1; // The range is too small to contain the requested number of consecutive elements.
+ }
+ // Calculate the length of the tail where a potential match cannot start by definition.
+ _DifferenceType __crop = __size2 - 1;
+ // We're only interested in the range where a potential match can start: [first, last - crop)
+ _ForwardIterator1 __last1_cropped = __last1 - __crop;
+ // Run a parallel chunked find_if, covering the range where a potential match can start.
+ auto __res = __statpart::__find(
+ __first1,
+ __last1_cropped,
+ [__first2, __last2, __crop, &__pred](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
+ // Uncrop the range to allow std::find_end to find a full match, which can go beyond __brick_last.
+ _ForwardIterator1 __brick_last_uncropped = __brick_last + __crop;
+ // Run a serial std::find_end inside each of the chunks in parallel.
+ _ForwardIterator1 __ret = std::find_end(__brick_first, __brick_last_uncropped, __first2, __last2, __pred);
+ // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
+ // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __parallel_find().
+ return __ret == __brick_last_uncropped ? __brick_last : __ret;
+ },
+ greater<>{}, // `greater` here means the highest index among the matches
+ false // `false` here means we want the last match, not the first
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ if (*__res == __last1_cropped) {
+ return __last1; // No match was found in the range.
+ }
+ return *__res; // Return the successful match.
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::find_end(
+ std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_FIND_END_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 153d67fc05d54d..dfe7eea64cb542 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2420,6 +2420,9 @@ module std {
module basis {
header "__pstl/backends/statpart/basis.h"
}
+ module find_end {
+ header "__pstl/backends/statpart/find_end.h"
+ }
}
module statpart_old {
header "__pstl/backends/statpart.h"
>From f7d7832ca578ff66e1bbb2edaa638813b2db878e Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Tue, 13 Oct 2026 11:09:16 +0100
Subject: [PATCH 08/24] Reorg
---
libcxx/include/CMakeLists.txt | 4 +
libcxx/include/__pstl/backends/statpart.h | 147 +-----------------
.../__pstl/backends/statpart/find_if.h | 67 ++++++++
.../__pstl/backends/statpart/for_each.h | 63 ++++++++
.../__pstl/backends/statpart/is_heap_until.h | 105 +++++++++++++
.../include/__pstl/backends/statpart/merge.h | 88 +++++++++++
libcxx/include/module.modulemap.in | 14 +-
7 files changed, 343 insertions(+), 145 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/find_if.h
create mode 100644 libcxx/include/__pstl/backends/statpart/for_each.h
create mode 100644 libcxx/include/__pstl/backends/statpart/is_heap_until.h
create mode 100644 libcxx/include/__pstl/backends/statpart/merge.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index c7830a12eaf009..2edea9fad38c61 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -698,6 +698,10 @@ set(files
__pstl/backends/statpart.h
__pstl/backends/statpart/basis.h
__pstl/backends/statpart/find_end.h
+ __pstl/backends/statpart/find_if.h
+ __pstl/backends/statpart/for_each.h
+ __pstl/backends/statpart/is_heap_until.h
+ __pstl/backends/statpart/merge.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 1ac299caeae200..5396380a9c51da 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -11,6 +11,10 @@
#include <__pstl/backends/statpart/basis.h>
#include <__pstl/backends/statpart/find_end.h>
+#include <__pstl/backends/statpart/find_if.h>
+#include <__pstl/backends/statpart/for_each.h>
+#include <__pstl/backends/statpart/is_heap_until.h>
+#include <__pstl/backends/statpart/merge.h>
#include <__algorithm/find_end.h>
#include <__algorithm/find_if.h>
@@ -74,149 +78,6 @@ _LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
-template <class _ExecutionPolicy>
-struct __for_each<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator, class _Function>
- _LIBCPP_HIDE_FROM_ABI optional<__empty>
- operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Function __func) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
- return __statpart::__for_each(
- std::move(__first), std::move(__last), [&](auto __partition_first, auto __partition_last) {
- std::for_each(__partition_first, __partition_last, __func);
- });
- } else {
- std::for_each(std::move(__first), std::move(__last), std::move(__func));
- return __empty{};
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __find_if<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator, class _Predicate>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
- operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
- return __statpart::__find(
- std::move(__first),
- std::move(__last),
- [&__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
- return std::find_if(__brick_first, __brick_last, __pred);
- },
- less<>{},
- true);
- } else {
- return std::find_if(std::move(__first), std::move(__last), std::move(__pred));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __is_heap_until<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _RandomAccessIterator, class _Comp>
- _LIBCPP_HIDE_FROM_ABI optional<_RandomAccessIterator>
- operator()(_Policy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy>) {
- if (__last - __first < 2)
- return __last; // Any sequence with less than 2 elements is a heap
- // Run a parallel find in chunks over [first+1, last) to validate every element except the root.
- return __statpart::__find(
- __first + 1,
- __last,
- [__first, &__comp](_RandomAccessIterator __child_first, _RandomAccessIterator __child_last) {
- // This algorithm assumes that __find() will never pass an empty brick,
- // i.e. that __child_first != __child_last.
- using _DifferenceType = typename std::iterator_traits<_RandomAccessIterator>::difference_type;
-
- // Derive the indices of the children and the iterators of their parents
- _DifferenceType __child_first_idx = __child_first - __first;
- _DifferenceType __child_last_idx = __child_last - __first;
- _RandomAccessIterator __parent = __first + (__child_first_idx - 1) / 2;
- _RandomAccessIterator __parent_last = __first + (__child_last_idx - 1) / 2;
-
- // If we're starting from a right child => process this element separately in a prologue
- if (__child_first_idx % 2 == 0) {
- if (__comp(*__parent, *__child_first)) // Check the right child
- return __child_first;
- ++__parent;
- ++__child_first;
- }
-
- // Iterate over the parents and check their left and right children
- for (; __parent != __parent_last; ++__parent) {
- if (__comp(*__parent, *__child_first)) // Check the left child
- return __child_first;
- ++__child_first;
-
- if (__comp(*__parent, *__child_first)) // Check the right child
- return __child_first;
- ++__child_first;
- }
-
- // If we're ending with a right child => __parent_last also includes the left child.
- // Process this accessible element separately in an epilogue.
- if (__child_last_idx % 2 == 0) {
- if (__comp(*__parent, *__child_first)) // Check the left child
- return __child_first;
- }
-
- return __child_last; // No violations found in this brick
- },
- less<>{}, // `less` here means the lowest index among the matches
- true // `true` here means we want the first match, not the last
- );
- } else {
- return std::is_heap_until(std::move(__first), std::move(__last), std::move(__comp));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __merge<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardOutIterator, class _Comp>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator> operator()(
- _Policy&&,
- _ForwardIterator1 __first1,
- _ForwardIterator1 __last1,
- _ForwardIterator2 __first2,
- _ForwardIterator2 __last2,
- _ForwardOutIterator __result,
- _Comp __comp) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
- auto __res = __statpart::__merge(
- __first1,
- __last1,
- __first2,
- __last2,
- __result,
- __comp,
- [](_ForwardIterator1 __brick_first1,
- _ForwardIterator1 __brick_last1,
- _ForwardIterator2 __brick_first2,
- _ForwardIterator2 __brick_last2,
- _ForwardOutIterator __brick_result,
- _Comp __g_comp) {
- std::merge(std::move(__brick_first1),
- std::move(__brick_last1),
- std::move(__brick_first2),
- std::move(__brick_last2),
- std::move(__brick_result),
- std::move(__g_comp));
- });
- if (!__res)
- return nullopt;
- return __result + (__last1 - __first1) + (__last2 - __first2);
- } else {
- return std::merge(__first1, __last1, __first2, __last2, __result, __comp);
- }
- }
-};
-
template <class _ExecutionPolicy>
struct __min_element<__statpart_backend_tag, _ExecutionPolicy> {
template <class _Policy, class _ForwardIterator, class _Compare>
diff --git a/libcxx/include/__pstl/backends/statpart/find_if.h b/libcxx/include/__pstl/backends/statpart/find_if.h
new file mode 100644
index 00000000000000..c7273a75e54bba
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/find_if.h
@@ -0,0 +1,67 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_FIND_IF_H
+#define _LIBCPP___PSTL_STATPART_FIND_IF_H
+
+#include <__algorithm/find_if.h>
+#include <__config>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __find_if<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Predicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __statpart::__find(
+ std::move(__first),
+ std::move(__last),
+ [&__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ return std::find_if(__brick_first, __brick_last, __pred);
+ },
+ less<>{},
+ true);
+ } else {
+ return std::find_if(std::move(__first), std::move(__last), std::move(__pred));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_FIND_IF_H
diff --git a/libcxx/include/__pstl/backends/statpart/for_each.h b/libcxx/include/__pstl/backends/statpart/for_each.h
new file mode 100644
index 00000000000000..b2199a828cb185
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/for_each.h
@@ -0,0 +1,63 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_FOR_EACH_H
+#define _LIBCPP___PSTL_STATPART_FOR_EACH_H
+
+#include <__algorithm/for_each.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __for_each<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Function>
+ _LIBCPP_HIDE_FROM_ABI optional<__empty>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Function __func) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __statpart::__for_each(
+ std::move(__first), std::move(__last), [&](auto __partition_first, auto __partition_last) {
+ std::for_each(__partition_first, __partition_last, __func);
+ });
+ } else {
+ std::for_each(std::move(__first), std::move(__last), std::move(__func));
+ return __empty{};
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_FOR_EACH_H
diff --git a/libcxx/include/__pstl/backends/statpart/is_heap_until.h b/libcxx/include/__pstl/backends/statpart/is_heap_until.h
new file mode 100644
index 00000000000000..88e0c11de11a2e
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/is_heap_until.h
@@ -0,0 +1,105 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_IS_HEAP_UNTIL_H
+#define _LIBCPP___PSTL_STATPART_IS_HEAP_UNTIL_H
+
+#include <__algorithm/is_heap_until.h>
+#include <__config>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __is_heap_until<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _RandomAccessIterator, class _Comp>
+ _LIBCPP_HIDE_FROM_ABI optional<_RandomAccessIterator>
+ operator()(_Policy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy>) {
+ if (__last - __first < 2)
+ return __last; // Any sequence with less than 2 elements is a heap
+ // Run a parallel find in chunks over [first+1, last) to validate every element except the root.
+ return __statpart::__find(
+ __first + 1,
+ __last,
+ [__first, &__comp](_RandomAccessIterator __child_first, _RandomAccessIterator __child_last) {
+ // This algorithm assumes that __find() will never pass an empty brick,
+ // i.e. that __child_first != __child_last.
+ using _DifferenceType = typename std::iterator_traits<_RandomAccessIterator>::difference_type;
+
+ // Derive the indices of the children and the iterators of their parents
+ _DifferenceType __child_first_idx = __child_first - __first;
+ _DifferenceType __child_last_idx = __child_last - __first;
+ _RandomAccessIterator __parent = __first + (__child_first_idx - 1) / 2;
+ _RandomAccessIterator __parent_last = __first + (__child_last_idx - 1) / 2;
+
+ // If we're starting from a right child => process this element separately in a prologue
+ if (__child_first_idx % 2 == 0) {
+ if (__comp(*__parent, *__child_first)) // Check the right child
+ return __child_first;
+ ++__parent;
+ ++__child_first;
+ }
+
+ // Iterate over the parents and check their left and right children
+ for (; __parent != __parent_last; ++__parent) {
+ if (__comp(*__parent, *__child_first)) // Check the left child
+ return __child_first;
+ ++__child_first;
+
+ if (__comp(*__parent, *__child_first)) // Check the right child
+ return __child_first;
+ ++__child_first;
+ }
+
+ // If we're ending with a right child => __parent_last also includes the left child.
+ // Process this accessible element separately in an epilogue.
+ if (__child_last_idx % 2 == 0) {
+ if (__comp(*__parent, *__child_first)) // Check the left child
+ return __child_first;
+ }
+
+ return __child_last; // No violations found in this brick
+ },
+ less<>{}, // `less` here means the lowest index among the matches
+ true // `true` here means we want the first match, not the last
+ );
+ } else {
+ return std::is_heap_until(std::move(__first), std::move(__last), std::move(__comp));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_IS_HEAP_UNTIL_H
diff --git a/libcxx/include/__pstl/backends/statpart/merge.h b/libcxx/include/__pstl/backends/statpart/merge.h
new file mode 100644
index 00000000000000..bd9ef68d413a26
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/merge.h
@@ -0,0 +1,88 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_MERGE_H
+#define _LIBCPP___PSTL_STATPART_MERGE_H
+
+#include <__algorithm/merge.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __merge<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardOutIterator, class _Comp>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator> operator()(
+ _Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _ForwardOutIterator __result,
+ _Comp __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
+ auto __res = __statpart::__merge(
+ __first1,
+ __last1,
+ __first2,
+ __last2,
+ __result,
+ __comp,
+ [](_ForwardIterator1 __brick_first1,
+ _ForwardIterator1 __brick_last1,
+ _ForwardIterator2 __brick_first2,
+ _ForwardIterator2 __brick_last2,
+ _ForwardOutIterator __brick_result,
+ _Comp __g_comp) {
+ std::merge(std::move(__brick_first1),
+ std::move(__brick_last1),
+ std::move(__brick_first2),
+ std::move(__brick_last2),
+ std::move(__brick_result),
+ std::move(__g_comp));
+ });
+ if (!__res)
+ return nullopt;
+ return __result + (__last1 - __first1) + (__last2 - __first2);
+ } else {
+ return std::merge(__first1, __last1, __first2, __last2, __result, __comp);
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_MERGE_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index dfe7eea64cb542..30450755d9fad6 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2423,8 +2423,18 @@ module std {
module find_end {
header "__pstl/backends/statpart/find_end.h"
}
- }
- module statpart_old {
+ module find_if {
+ header "__pstl/backends/statpart/find_if.h"
+ }
+ module for_each {
+ header "__pstl/backends/statpart/for_each.h"
+ }
+ module is_heap_until {
+ header "__pstl/backends/statpart/is_heap_until.h"
+ }
+ module merge {
+ header "__pstl/backends/statpart/merge.h"
+ }
header "__pstl/backends/statpart.h"
}
module std_thread {
>From 7d328ddb9b6ccaee45d714b7419064bd41372c2d Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Tue, 13 Oct 2026 11:39:13 +0100
Subject: [PATCH 09/24] Reorg
---
libcxx/include/CMakeLists.txt | 3 +
libcxx/include/__pstl/backends/statpart.h | 134 +-----------------
.../__pstl/backends/statpart/min_element.h | 85 +++++++++++
.../__pstl/backends/statpart/minmax_element.h | 90 ++++++++++++
.../__pstl/backends/statpart/mismatch.h | 91 ++++++++++++
libcxx/include/module.modulemap.in | 9 ++
6 files changed, 281 insertions(+), 131 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/min_element.h
create mode 100644 libcxx/include/__pstl/backends/statpart/minmax_element.h
create mode 100644 libcxx/include/__pstl/backends/statpart/mismatch.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index 2edea9fad38c61..629488639a88f2 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -702,6 +702,9 @@ set(files
__pstl/backends/statpart/for_each.h
__pstl/backends/statpart/is_heap_until.h
__pstl/backends/statpart/merge.h
+ __pstl/backends/statpart/min_element.h
+ __pstl/backends/statpart/minmax_element.h
+ __pstl/backends/statpart/mismatch.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 5396380a9c51da..101f6e536d5346 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -15,6 +15,9 @@
#include <__pstl/backends/statpart/for_each.h>
#include <__pstl/backends/statpart/is_heap_until.h>
#include <__pstl/backends/statpart/merge.h>
+#include <__pstl/backends/statpart/min_element.h>
+#include <__pstl/backends/statpart/minmax_element.h>
+#include <__pstl/backends/statpart/mismatch.h>
#include <__algorithm/find_end.h>
#include <__algorithm/find_if.h>
@@ -78,137 +81,6 @@ _LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
-template <class _ExecutionPolicy>
-struct __min_element<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator, class _Compare>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
- operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
- if (__first == __last) {
- return __last; // nothing to do
- }
-
- _ForwardIterator __init = __first;
- ++__first;
- if (__first == __last) {
- return __init; // the only element is the minimum
- }
-
- // A reduction that returns an iterator pointing to the lowest element, left bias in case of a tie
- auto __iter_reduce = [&__comp](_ForwardIterator __lhs, _ForwardIterator __rhs) {
- return __comp(*__rhs, *__lhs) ? __rhs : __lhs;
- };
-
- // Perform a parallel reduction of iterators [first+1, last) with 'first' as init.
- return __statpart::__transform_reduce(
- std::move(__first),
- std::move(__last),
- __identity{}, // No transformations
- std::move(__init), // Use the first iterator as the init element
- __iter_reduce, // Reduction of 2 elements
- [&__iter_reduce, &__comp](auto __brick_first, auto __brick_last, auto __brick_init) {
- // Reduction of an iterator range + init element: use the serial version to find the minimum among
- // the iterators and then reduce with the init element.
- return __iter_reduce(__brick_init, std::min_element(__brick_first, __brick_last, __comp));
- });
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::min_element(std::move(__first), std::move(__last), std::move(__comp));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __minmax_element<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator, class _Compare>
- _LIBCPP_HIDE_FROM_ABI optional<pair<_ForwardIterator, _ForwardIterator>>
- operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
- using _IterPair = pair<_ForwardIterator, _ForwardIterator>;
-
- if (__first == __last) {
- return _IterPair{__last, __last}; // Nothing to do
- }
-
- _IterPair __init = {__first, __first};
- ++__first;
- if (__first == __last) {
- return __init; // The only element is both the minimum and the maximum
- }
-
- // A reduction that returns a pair of iterators pointing to the minimum and maximum elements.
- // In a case of a tie the minimum iterators are biased left and the maximum iterators are biased right.
- auto __iter_reduce = [&__comp](_IterPair __lhs, _IterPair __rhs) {
- return _IterPair{__comp(*__rhs.first, *__lhs.first) ? __rhs.first : __lhs.first,
- __comp(*__rhs.second, *__lhs.second) ? __lhs.second : __rhs.second};
- };
-
- // Perform a parallel reduction of iterators [first+1, last) with {first, first} as init.
- return __statpart::__transform_reduce(
- std::move(__first),
- std::move(__last),
- [](auto __it) { return _IterPair{__it, __it}; }, // Transform an iterator into an iterator pair
- std::move(__init), // Use the pair of first iterators as the init element
- __iter_reduce, // Reduction of 2 minmax pairs
- [&__iter_reduce, &__comp](auto __brick_first, auto __brick_last, auto __brick_init) {
- // Reduction of an iterator range + init element: use the serial version to find the minmax among
- // the iterators and then reduce it with the init element.
- return __iter_reduce(__brick_init, std::minmax_element(__brick_first, __brick_last, __comp));
- });
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::minmax_element(std::move(__first), std::move(__last), std::move(__comp));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __mismatch<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
- _LIBCPP_HIDE_FROM_ABI optional<pair<_ForwardIterator1, _ForwardIterator2>>
- operator()(_Policy&&,
- _ForwardIterator1 __first1,
- _ForwardIterator1 __last1,
- _ForwardIterator2 __first2,
- _ForwardIterator2 __last2,
- _Predicate __pred) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
- // Look for a mismatch only in the prefix of the two ranges.
- auto __n = std::min(__last1 - __first1, __last2 - __first2);
- // Find a position in the first range where the predicate is false against the corresponding position in the
- // second range.
- auto __res = __statpart::__find(
- __first1,
- __first1 + __n,
- [&__pred, __first1, __first2](_ForwardIterator1 __brick_first1, _ForwardIterator1 __brick_last1) {
- // Run the sequential mismatch algorithm on these ranges:
- // [__brick_first1, __brick_last1) and
- // [__first2 + (__brick_first1 - __first1), __first2 + (__brick_last1 - __first1))
- auto __brick_first2 = __first2 + (__brick_first1 - __first1);
- return std::mismatch(std::move(__brick_first1), std::move(__brick_last1), std::move(__brick_first2), __pred)
- .first;
- },
- less<>{}, // `less` here means the lowest index among the mismatches
- true // `true` here means we want the first mismatch, not the last
- );
- if (!__res) {
- return std::nullopt; // Failed to run the algorithm, propagate the error.
- }
- auto __idx = *__res - __first1;
- return pair<_ForwardIterator1, _ForwardIterator2>{std::move(*__res), __first2 + __idx};
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- // Unsequenced execution is also implicitly covered by the sequential implementation.
- return std::mismatch(
- std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
- }
- }
-};
-
template <class _ExecutionPolicy>
struct __reverse<__statpart_backend_tag, _ExecutionPolicy> {
template <class _Policy, class _BidirectionalIterator>
diff --git a/libcxx/include/__pstl/backends/statpart/min_element.h b/libcxx/include/__pstl/backends/statpart/min_element.h
new file mode 100644
index 00000000000000..09e3117c81d475
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/min_element.h
@@ -0,0 +1,85 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_MIN_ELEMENT_H
+#define _LIBCPP___PSTL_STATPART_MIN_ELEMENT_H
+
+#include <__algorithm/min_element.h>
+#include <__config>
+#include <__functional/identity.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __min_element<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Compare>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ if (__first == __last) {
+ return __last; // nothing to do
+ }
+
+ _ForwardIterator __init = __first;
+ ++__first;
+ if (__first == __last) {
+ return __init; // the only element is the minimum
+ }
+
+ // A reduction that returns an iterator pointing to the lowest element, left bias in case of a tie
+ auto __iter_reduce = [&__comp](_ForwardIterator __lhs, _ForwardIterator __rhs) {
+ return __comp(*__rhs, *__lhs) ? __rhs : __lhs;
+ };
+
+ // Perform a parallel reduction of iterators [first+1, last) with 'first' as init.
+ return __statpart::__transform_reduce(
+ std::move(__first),
+ std::move(__last),
+ __identity{}, // No transformations
+ std::move(__init), // Use the first iterator as the init element
+ __iter_reduce, // Reduction of 2 elements
+ [&__iter_reduce, &__comp](auto __brick_first, auto __brick_last, auto __brick_init) {
+ // Reduction of an iterator range + init element: use the serial version to find the minimum among
+ // the iterators and then reduce with the init element.
+ return __iter_reduce(__brick_init, std::min_element(__brick_first, __brick_last, __comp));
+ });
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::min_element(std::move(__first), std::move(__last), std::move(__comp));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_MIN_ELEMENT_H
diff --git a/libcxx/include/__pstl/backends/statpart/minmax_element.h b/libcxx/include/__pstl/backends/statpart/minmax_element.h
new file mode 100644
index 00000000000000..bde4bb515999b4
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/minmax_element.h
@@ -0,0 +1,90 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_MINMAX_ELEMENT_H
+#define _LIBCPP___PSTL_STATPART_MINMAX_ELEMENT_H
+
+#include <__algorithm/minmax_element.h>
+#include <__config>
+#include <__functional/identity.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+#include <__utility/pair.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __minmax_element<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Compare>
+ _LIBCPP_HIDE_FROM_ABI optional<pair<_ForwardIterator, _ForwardIterator>>
+ operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ using _IterPair = pair<_ForwardIterator, _ForwardIterator>;
+
+ if (__first == __last) {
+ return _IterPair{__last, __last}; // Nothing to do
+ }
+
+ _IterPair __init = {__first, __first};
+ ++__first;
+ if (__first == __last) {
+ return __init; // The only element is both the minimum and the maximum
+ }
+
+ // A reduction that returns a pair of iterators pointing to the minimum and maximum elements.
+ // In a case of a tie the minimum iterators are biased left and the maximum iterators are biased right.
+ auto __iter_reduce = [&__comp](_IterPair __lhs, _IterPair __rhs) {
+ return _IterPair{__comp(*__rhs.first, *__lhs.first) ? __rhs.first : __lhs.first,
+ __comp(*__rhs.second, *__lhs.second) ? __lhs.second : __rhs.second};
+ };
+
+ // Perform a parallel reduction of iterators [first+1, last) with {first, first} as init.
+ return __statpart::__transform_reduce(
+ std::move(__first),
+ std::move(__last),
+ [](auto __it) { return _IterPair{__it, __it}; }, // Transform an iterator into an iterator pair
+ std::move(__init), // Use the pair of first iterators as the init element
+ __iter_reduce, // Reduction of 2 minmax pairs
+ [&__iter_reduce, &__comp](auto __brick_first, auto __brick_last, auto __brick_init) {
+ // Reduction of an iterator range + init element: use the serial version to find the minmax among
+ // the iterators and then reduce it with the init element.
+ return __iter_reduce(__brick_init, std::minmax_element(__brick_first, __brick_last, __comp));
+ });
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::minmax_element(std::move(__first), std::move(__last), std::move(__comp));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_MINMAX_ELEMENT_H
diff --git a/libcxx/include/__pstl/backends/statpart/mismatch.h b/libcxx/include/__pstl/backends/statpart/mismatch.h
new file mode 100644
index 00000000000000..b049b690e12bf8
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/mismatch.h
@@ -0,0 +1,91 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_MISMATCH_H
+#define _LIBCPP___PSTL_STATPART_MISMATCH_H
+
+#include <__algorithm/min.h>
+#include <__algorithm/mismatch.h>
+#include <__config>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+#include <__utility/pair.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __mismatch<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
+ _LIBCPP_HIDE_FROM_ABI optional<pair<_ForwardIterator1, _ForwardIterator2>>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _Predicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ // Look for a mismatch only in the prefix of the two ranges.
+ auto __n = std::min(__last1 - __first1, __last2 - __first2);
+ // Find a position in the first range where the predicate is false against the corresponding position in the
+ // second range.
+ auto __res = __statpart::__find(
+ __first1,
+ __first1 + __n,
+ [&__pred, __first1, __first2](_ForwardIterator1 __brick_first1, _ForwardIterator1 __brick_last1) {
+ // Run the sequential mismatch algorithm on these ranges:
+ // [__brick_first1, __brick_last1) and
+ // [__first2 + (__brick_first1 - __first1), __first2 + (__brick_last1 - __first1))
+ auto __brick_first2 = __first2 + (__brick_first1 - __first1);
+ return std::mismatch(std::move(__brick_first1), std::move(__brick_last1), std::move(__brick_first2), __pred)
+ .first;
+ },
+ less<>{}, // `less` here means the lowest index among the mismatches
+ true // `true` here means we want the first mismatch, not the last
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ auto __idx = *__res - __first1;
+ return pair<_ForwardIterator1, _ForwardIterator2>{std::move(*__res), __first2 + __idx};
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ // Unsequenced execution is also implicitly covered by the sequential implementation.
+ return std::mismatch(
+ std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_MISMATCH_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 30450755d9fad6..a33e9441270119 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2435,6 +2435,15 @@ module std {
module merge {
header "__pstl/backends/statpart/merge.h"
}
+ module min_element {
+ header "__pstl/backends/statpart/min_element.h"
+ }
+ module minmax_element {
+ header "__pstl/backends/statpart/minmax_element.h"
+ }
+ module mismatch {
+ header "__pstl/backends/statpart/mismatch.h"
+ }
header "__pstl/backends/statpart.h"
}
module std_thread {
>From 45fec3699dd64df560d4457a9faa5952833b3797 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Tue, 13 Oct 2026 11:54:45 +0100
Subject: [PATCH 10/24] Reorg
---
libcxx/include/CMakeLists.txt | 2 +
libcxx/include/__pstl/backends/statpart.h | 87 +--------------
.../__pstl/backends/statpart/reverse.h | 75 +++++++++++++
.../include/__pstl/backends/statpart/search.h | 102 ++++++++++++++++++
libcxx/include/module.modulemap.in | 6 ++
5 files changed, 187 insertions(+), 85 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/reverse.h
create mode 100644 libcxx/include/__pstl/backends/statpart/search.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index 629488639a88f2..1da365a5bd0909 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -705,6 +705,8 @@ set(files
__pstl/backends/statpart/min_element.h
__pstl/backends/statpart/minmax_element.h
__pstl/backends/statpart/mismatch.h
+ __pstl/backends/statpart/reverse.h
+ __pstl/backends/statpart/search.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 101f6e536d5346..ff9abb66477917 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -18,6 +18,8 @@
#include <__pstl/backends/statpart/min_element.h>
#include <__pstl/backends/statpart/minmax_element.h>
#include <__pstl/backends/statpart/mismatch.h>
+#include <__pstl/backends/statpart/reverse.h>
+#include <__pstl/backends/statpart/search.h>
#include <__algorithm/find_end.h>
#include <__algorithm/find_if.h>
@@ -81,91 +83,6 @@ _LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
-template <class _ExecutionPolicy>
-struct __reverse<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _BidirectionalIterator>
- _LIBCPP_HIDE_FROM_ABI optional<__empty>
- operator()(_Policy&&, _BidirectionalIterator __first, _BidirectionalIterator __last) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_BidirectionalIterator>::value) {
- // Perform a chunked for_each on the first half of the range.
- // Odd-sized ranges will leave the middle element untouched, which is correct for this algorithm:
- // its reversed position is the same.
- auto __n = (__last - __first) / 2;
- return __statpart::__for_each(
- __first, __first + __n, [__first, __last](_BidirectionalIterator __i, _BidirectionalIterator __j) {
- // Derive the last position of the mirrored range by counting from the end.
- _BidirectionalIterator __mirror_last = __last - (__i - __first);
- // Swap the elements in the range of the first half with their mirrored counterparts in the second half.
- std::swap_ranges(std::move(__i),
- std::move(__j),
- std::reverse_iterator<_BidirectionalIterator>(std::move(__mirror_last)));
- });
- } else {
- // Non-random access iterators currently cannot be processed in parallel, use the sequential implementation.
- std::reverse(std::move(__first), std::move(__last));
- return __empty{};
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __search<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator1>
- operator()(_Policy&&,
- _ForwardIterator1 __first1,
- _ForwardIterator1 __last1,
- _ForwardIterator2 __first2,
- _ForwardIterator2 __last2,
- _BinaryPredicate __pred) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
- typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
- _DifferenceType __size2 = __last2 - __first2; // The length of the needle to search for.
- if (__size2 == 0) {
- return __first1; // If the needle length is zero, the first iterator is returned.
- }
- _DifferenceType __size1 = __last1 - __first1;
- if (__size1 < __size2) {
- return __last1; // The range is too small to contain the requested number of consecutive elements.
- }
- // Calculate the length of the tail where a potential match cannot start by definition.
- _DifferenceType __crop = __size2 - 1;
- // We're only interested in the range where a potential match can start: [first, last - crop)
- _ForwardIterator1 __last1_cropped = __last1 - __crop;
- // Run a parallel chunked find_if, covering the range where a potential match can start.
- auto __res = __statpart::__find(
- __first1,
- __last1_cropped,
- [__first2, __last2, __crop, &__pred](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
- // Uncrop the range to allow std::search to find a full match, which can go beyond __brick_last.
- _ForwardIterator1 __brick_last_uncropped = __brick_last + __crop;
- // Run a serial std::search inside each of the chunks in parallel.
- _ForwardIterator1 __ret = std::search(__brick_first, __brick_last_uncropped, __first2, __last2, __pred);
- // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
- // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __find().
- return __ret == __brick_last_uncropped ? __brick_last : __ret;
- },
- less<>{}, // `less` here means the lowest index among the matches
- true // `true` here means we want the first match, not the last
- );
- if (!__res) {
- return std::nullopt; // Failed to run the algorithm, propagate the error.
- }
- if (*__res == __last1_cropped) {
- return __last1; // No match was found in the range.
- }
- return *__res; // Return the successful match.
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::search(
- std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
- }
- }
-};
-
template <class _ExecutionPolicy>
struct __search_n<__statpart_backend_tag, _ExecutionPolicy> {
template <class _Policy, class _ForwardIterator, class _Size, class _Tp, class _Predicate>
diff --git a/libcxx/include/__pstl/backends/statpart/reverse.h b/libcxx/include/__pstl/backends/statpart/reverse.h
new file mode 100644
index 00000000000000..4e3d147475568b
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/reverse.h
@@ -0,0 +1,75 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_REVERSE_H
+#define _LIBCPP___PSTL_STATPART_REVERSE_H
+
+#include <__algorithm/reverse.h>
+#include <__algorithm/swap_ranges.h>
+#include <__config>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__iterator/reverse_iterator.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __reverse<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _BidirectionalIterator>
+ _LIBCPP_HIDE_FROM_ABI optional<__empty>
+ operator()(_Policy&&, _BidirectionalIterator __first, _BidirectionalIterator __last) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_BidirectionalIterator>::value) {
+ // Perform a chunked for_each on the first half of the range.
+ // Odd-sized ranges will leave the middle element untouched, which is correct for this algorithm:
+ // its reversed position is the same.
+ auto __n = (__last - __first) / 2;
+ return __statpart::__for_each(
+ __first, __first + __n, [__first, __last](_BidirectionalIterator __i, _BidirectionalIterator __j) {
+ // Derive the last position of the mirrored range by counting from the end.
+ _BidirectionalIterator __mirror_last = __last - (__i - __first);
+ // Swap the elements in the range of the first half with their mirrored counterparts in the second half.
+ std::swap_ranges(std::move(__i),
+ std::move(__j),
+ std::reverse_iterator<_BidirectionalIterator>(std::move(__mirror_last)));
+ });
+ } else {
+ // Non-random access iterators currently cannot be processed in parallel, use the sequential implementation.
+ std::reverse(std::move(__first), std::move(__last));
+ return __empty{};
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_REVERSE_H
diff --git a/libcxx/include/__pstl/backends/statpart/search.h b/libcxx/include/__pstl/backends/statpart/search.h
new file mode 100644
index 00000000000000..a831fb8ef63d0e
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/search.h
@@ -0,0 +1,102 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_SEARCH_H
+#define _LIBCPP___PSTL_STATPART_SEARCH_H
+
+#include <__algorithm/search.h>
+#include <__config>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __search<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator1>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardIterator2 __last2,
+ _BinaryPredicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
+ _DifferenceType __size2 = __last2 - __first2; // The length of the needle to search for.
+ if (__size2 == 0) {
+ return __first1; // If the needle length is zero, the first iterator is returned.
+ }
+ _DifferenceType __size1 = __last1 - __first1;
+ if (__size1 < __size2) {
+ return __last1; // The range is too small to contain the requested number of consecutive elements.
+ }
+ // Calculate the length of the tail where a potential match cannot start by definition.
+ _DifferenceType __crop = __size2 - 1;
+ // We're only interested in the range where a potential match can start: [first, last - crop)
+ _ForwardIterator1 __last1_cropped = __last1 - __crop;
+ // Run a parallel chunked find_if, covering the range where a potential match can start.
+ auto __res = __statpart::__find(
+ __first1,
+ __last1_cropped,
+ [__first2, __last2, __crop, &__pred](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
+ // Uncrop the range to allow std::search to find a full match, which can go beyond __brick_last.
+ _ForwardIterator1 __brick_last_uncropped = __brick_last + __crop;
+ // Run a serial std::search inside each of the chunks in parallel.
+ _ForwardIterator1 __ret = std::search(__brick_first, __brick_last_uncropped, __first2, __last2, __pred);
+ // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
+ // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __find().
+ return __ret == __brick_last_uncropped ? __brick_last : __ret;
+ },
+ less<>{}, // `less` here means the lowest index among the matches
+ true // `true` here means we want the first match, not the last
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ if (*__res == __last1_cropped) {
+ return __last1; // No match was found in the range.
+ }
+ return *__res; // Return the successful match.
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::search(
+ std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), std::move(__pred));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_SEARCH_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index a33e9441270119..42b219a06ec8d3 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2444,6 +2444,12 @@ module std {
module mismatch {
header "__pstl/backends/statpart/mismatch.h"
}
+ module reverse {
+ header "__pstl/backends/statpart/reverse.h"
+ }
+ module search {
+ header "__pstl/backends/statpart/search.h"
+ }
header "__pstl/backends/statpart.h"
}
module std_thread {
>From eb220f8e97f99dd81c34718dab66d3ab13c50043 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Tue, 13 Oct 2026 15:19:02 +0100
Subject: [PATCH 11/24] Reorg
---
libcxx/include/CMakeLists.txt | 4 +
libcxx/include/__pstl/backends/statpart.h | 161 +-----------------
.../__pstl/backends/statpart/search_n.h | 102 +++++++++++
.../__pstl/backends/statpart/stable_sort.h | 61 +++++++
.../__pstl/backends/statpart/swap_ranges.h | 70 ++++++++
.../__pstl/backends/statpart/transform.h | 102 +++++++++++
libcxx/include/module.modulemap.in | 12 ++
7 files changed, 355 insertions(+), 157 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/search_n.h
create mode 100644 libcxx/include/__pstl/backends/statpart/stable_sort.h
create mode 100644 libcxx/include/__pstl/backends/statpart/swap_ranges.h
create mode 100644 libcxx/include/__pstl/backends/statpart/transform.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index 1da365a5bd0909..f2657b9810faa0 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -707,6 +707,10 @@ set(files
__pstl/backends/statpart/mismatch.h
__pstl/backends/statpart/reverse.h
__pstl/backends/statpart/search.h
+ __pstl/backends/statpart/search_n.h
+ __pstl/backends/statpart/stable_sort.h
+ __pstl/backends/statpart/swap_ranges.h
+ __pstl/backends/statpart/transform.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index ff9abb66477917..61c484f91318f2 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -20,6 +20,10 @@
#include <__pstl/backends/statpart/mismatch.h>
#include <__pstl/backends/statpart/reverse.h>
#include <__pstl/backends/statpart/search.h>
+#include <__pstl/backends/statpart/search_n.h>
+#include <__pstl/backends/statpart/stable_sort.h>
+#include <__pstl/backends/statpart/swap_ranges.h>
+#include <__pstl/backends/statpart/transform.h>
#include <__algorithm/find_end.h>
#include <__algorithm/find_if.h>
@@ -83,163 +87,6 @@ _LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
-template <class _ExecutionPolicy>
-struct __search_n<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator, class _Size, class _Tp, class _Predicate>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
- operator()(_Policy&&,
- _ForwardIterator __first,
- _ForwardIterator __last,
- _Size __count,
- const _Tp& __value,
- _Predicate __pred) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
- typedef typename std::iterator_traits<_ForwardIterator>::difference_type _DifferenceType;
- _DifferenceType __integral_count = std::__convert_to_integral(__count);
- if (__integral_count <= 0) {
- return __first; // If the count is non-positive, the first iterator is returned.
- }
- _DifferenceType __size = __last - __first;
- if (__size < __integral_count) {
- return __last; // The range is too small to contain the requested number of consecutive elements.
- }
- // Calculate the length of the tail where a potential match cannot start by definition.
- _DifferenceType __crop = __integral_count - 1;
- // We're only interested in the range where a potential match can start: [first, last - crop)
- _ForwardIterator __last2 = __last - __crop;
- // Run a parallel chunked find_if, covering the range where a potential match can start.
- auto __res = __statpart::__find(
- __first,
- __last2,
- [__integral_count, __crop, &__value, &__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
- // Uncrop the range to allow std::search_n to find a full match, which can go beyond __brick_last.
- _ForwardIterator __brick_last_uncropped = __brick_last + __crop;
- // Run a serial std::search_n inside each of the chunks in parallel.
- _ForwardIterator __ret =
- std::search_n(__brick_first, __brick_last_uncropped, __integral_count, __value, __pred);
- // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
- // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __find().
- return __ret == __brick_last_uncropped ? __brick_last : __ret;
- },
- less<>{}, // `less` here means the lowest index among the matches
- true // `true` here means we want the first match, not the last
- );
- if (!__res) {
- return std::nullopt; // Failed to run the algorithm, propagate the error.
- }
- if (*__res == __last2) {
- return __last; // No match was found in the range.
- }
- return *__res; // Return the successful match.
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::search_n(std::move(__first), std::move(__last), __count, __value, std::move(__pred));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __stable_sort<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _RandomAccessIterator, class _Comp>
- _LIBCPP_HIDE_FROM_ABI optional<__empty>
- operator()(_Policy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy>) {
- return __statpart::__stable_sort(
- __first, __last, __comp, [](_RandomAccessIterator __g_first, _RandomAccessIterator __g_last, _Comp __g_comp) {
- std::stable_sort(__g_first, __g_last, __g_comp);
- });
- } else {
- std::stable_sort(__first, __last, __comp);
- return __empty{};
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __swap_ranges<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
- _Policy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
- auto __res = __statpart::__for_each_iter_pair(
- __first1,
- __last1,
- __first2,
- [](_ForwardIterator1 __brick_first1, _ForwardIterator1 __brick_last1, _ForwardIterator2 __brick_first2) {
- std::swap_ranges(std::move(__brick_first1), std::move(__brick_last1), std::move(__brick_first2));
- });
- if (!__res) {
- return std::nullopt; // Failed to run the algorithm, propagate the error.
- }
- return __first2 + (__last1 - __first1);
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::swap_ranges(std::move(__first1), std::move(__last1), std::move(__first2));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __transform<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator, class _ForwardOutIterator, class _UnaryOperation>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator> operator()(
- _Policy&&, _ForwardIterator __first, _ForwardIterator __last, _ForwardOutIterator __result, _UnaryOperation __op)
- const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
- __statpart::__for_each(
- __first, __last, [__op, __first, __result](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
- return std::transform(__brick_first, __brick_last, __result + (__brick_first - __first), __op);
- });
- return __result + (__last - __first);
- } else {
- return std::transform(__first, __last, __result, __op);
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __transform_binary<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy,
- class _ForwardIterator1,
- class _ForwardIterator2,
- class _ForwardOutIterator,
- class _BinaryOperation>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator>
- operator()(_Policy&&,
- _ForwardIterator1 __first1,
- _ForwardIterator1 __last1,
- _ForwardIterator2 __first2,
- _ForwardOutIterator __result,
- _BinaryOperation __op) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
- auto __res = __statpart::__for_each(
- __first1,
- __last1,
- [__op, __first1, __first2, __result](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
- return std::transform(
- __brick_first,
- __brick_last,
- __first2 + (__brick_first - __first1),
- __result + (__brick_first - __first1),
- __op);
- });
- if (!__res)
- return nullopt;
- return __result + (__last1 - __first1);
- } else {
- return std::transform(__first1, __last1, __first2, __result, __op);
- }
- }
-};
-
template <class _ExecutionPolicy>
struct __transform_reduce<__statpart_backend_tag, _ExecutionPolicy> {
template <class _Policy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation>
diff --git a/libcxx/include/__pstl/backends/statpart/search_n.h b/libcxx/include/__pstl/backends/statpart/search_n.h
new file mode 100644
index 00000000000000..1ab5b72f2c5b67
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/search_n.h
@@ -0,0 +1,102 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_SEARCH_N_H
+#define _LIBCPP___PSTL_STATPART_SEARCH_N_H
+
+#include <__algorithm/search_n.h>
+#include <__config>
+#include <__functional/operations.h>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/convert_to_integral.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __search_n<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Size, class _Tp, class _Predicate>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
+ operator()(_Policy&&,
+ _ForwardIterator __first,
+ _ForwardIterator __last,
+ _Size __count,
+ const _Tp& __value,
+ _Predicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ typedef typename std::iterator_traits<_ForwardIterator>::difference_type _DifferenceType;
+ _DifferenceType __integral_count = std::__convert_to_integral(__count);
+ if (__integral_count <= 0) {
+ return __first; // If the count is non-positive, the first iterator is returned.
+ }
+ _DifferenceType __size = __last - __first;
+ if (__size < __integral_count) {
+ return __last; // The range is too small to contain the requested number of consecutive elements.
+ }
+ // Calculate the length of the tail where a potential match cannot start by definition.
+ _DifferenceType __crop = __integral_count - 1;
+ // We're only interested in the range where a potential match can start: [first, last - crop)
+ _ForwardIterator __last2 = __last - __crop;
+ // Run a parallel chunked find_if, covering the range where a potential match can start.
+ auto __res = __statpart::__find(
+ __first,
+ __last2,
+ [__integral_count, __crop, &__value, &__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ // Uncrop the range to allow std::search_n to find a full match, which can go beyond __brick_last.
+ _ForwardIterator __brick_last_uncropped = __brick_last + __crop;
+ // Run a serial std::search_n inside each of the chunks in parallel.
+ _ForwardIterator __ret =
+ std::search_n(__brick_first, __brick_last_uncropped, __integral_count, __value, __pred);
+ // The returned iterator is either a match inside [__brick_first, __brick_last) or a miss encoded as
+ // __brick_last_uncropped. Return the miss as __brick_last to conform to expectations of __find().
+ return __ret == __brick_last_uncropped ? __brick_last : __ret;
+ },
+ less<>{}, // `less` here means the lowest index among the matches
+ true // `true` here means we want the first match, not the last
+ );
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ if (*__res == __last2) {
+ return __last; // No match was found in the range.
+ }
+ return *__res; // Return the successful match.
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::search_n(std::move(__first), std::move(__last), __count, __value, std::move(__pred));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_SEARCH_N_H
diff --git a/libcxx/include/__pstl/backends/statpart/stable_sort.h b/libcxx/include/__pstl/backends/statpart/stable_sort.h
new file mode 100644
index 00000000000000..22708ba4a8fcd1
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/stable_sort.h
@@ -0,0 +1,61 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_STABLE_SORT_H
+#define _LIBCPP___PSTL_STATPART_STABLE_SORT_H
+
+#include <__algorithm/stable_sort.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __stable_sort<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _RandomAccessIterator, class _Comp>
+ _LIBCPP_HIDE_FROM_ABI optional<__empty>
+ operator()(_Policy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy>) {
+ return __statpart::__stable_sort(
+ __first, __last, __comp, [](_RandomAccessIterator __g_first, _RandomAccessIterator __g_last, _Comp __g_comp) {
+ std::stable_sort(__g_first, __g_last, __g_comp);
+ });
+ } else {
+ std::stable_sort(__first, __last, __comp);
+ return __empty{};
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_STABLE_SORT_H
diff --git a/libcxx/include/__pstl/backends/statpart/swap_ranges.h b/libcxx/include/__pstl/backends/statpart/swap_ranges.h
new file mode 100644
index 00000000000000..71a4c40cebe516
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/swap_ranges.h
@@ -0,0 +1,70 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_SWAP_RANGES_H
+#define _LIBCPP___PSTL_STATPART_SWAP_RANGES_H
+
+#include <__algorithm/swap_ranges.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __swap_ranges<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
+ _Policy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ auto __res = __statpart::__for_each_iter_pair(
+ __first1,
+ __last1,
+ __first2,
+ [](_ForwardIterator1 __brick_first1, _ForwardIterator1 __brick_last1, _ForwardIterator2 __brick_first2) {
+ std::swap_ranges(std::move(__brick_first1), std::move(__brick_last1), std::move(__brick_first2));
+ });
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ return __first2 + (__last1 - __first1);
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::swap_ranges(std::move(__first1), std::move(__last1), std::move(__first2));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_SWAP_RANGES_H
diff --git a/libcxx/include/__pstl/backends/statpart/transform.h b/libcxx/include/__pstl/backends/statpart/transform.h
new file mode 100644
index 00000000000000..69914bc2c5055b
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/transform.h
@@ -0,0 +1,102 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_TRANSFORM_H
+#define _LIBCPP___PSTL_STATPART_TRANSFORM_H
+
+#include <__algorithm/transform.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __transform<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _ForwardOutIterator, class _UnaryOperation>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator> operator()(
+ _Policy&&, _ForwardIterator __first, _ForwardIterator __last, _ForwardOutIterator __result, _UnaryOperation __op)
+ const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
+ __statpart::__for_each(
+ __first, __last, [__op, __first, __result](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ return std::transform(__brick_first, __brick_last, __result + (__brick_first - __first), __op);
+ });
+ return __result + (__last - __first);
+ } else {
+ return std::transform(__first, __last, __result, __op);
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __transform_binary<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy,
+ class _ForwardIterator1,
+ class _ForwardIterator2,
+ class _ForwardOutIterator,
+ class _BinaryOperation>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardOutIterator>
+ operator()(_Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _ForwardOutIterator __result,
+ _BinaryOperation __op) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardOutIterator>::value) {
+ auto __res = __statpart::__for_each(
+ __first1,
+ __last1,
+ [__op, __first1, __first2, __result](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last) {
+ return std::transform(
+ __brick_first,
+ __brick_last,
+ __first2 + (__brick_first - __first1),
+ __result + (__brick_first - __first1),
+ __op);
+ });
+ if (!__res)
+ return nullopt;
+ return __result + (__last1 - __first1);
+ } else {
+ return std::transform(__first1, __last1, __first2, __result, __op);
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_TRANSFORM_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 42b219a06ec8d3..92c2118c8760e8 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2450,6 +2450,18 @@ module std {
module search {
header "__pstl/backends/statpart/search.h"
}
+ module search_n {
+ header "__pstl/backends/statpart/search_n.h"
+ }
+ module stable_sort {
+ header "__pstl/backends/statpart/stable_sort.h"
+ }
+ module swap_ranges {
+ header "__pstl/backends/statpart/swap_ranges.h"
+ }
+ module transform {
+ header "__pstl/backends/statpart/transform.h"
+ }
header "__pstl/backends/statpart.h"
}
module std_thread {
>From 1211da0a6b1341e93648297795d0f9c08710f696 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Tue, 13 Oct 2026 15:40:36 +0100
Subject: [PATCH 12/24] Reorg
---
libcxx/include/CMakeLists.txt | 3 +
libcxx/include/__pstl/backends/statpart.h | 207 +-----------------
.../backends/statpart/transform_reduce.h | 126 +++++++++++
.../backends/statpart/uninitialized_copy.h | 69 ++++++
.../backends/statpart/uninitialized_move.h | 69 ++++++
libcxx/include/module.modulemap.in | 9 +
6 files changed, 279 insertions(+), 204 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/transform_reduce.h
create mode 100644 libcxx/include/__pstl/backends/statpart/uninitialized_copy.h
create mode 100644 libcxx/include/__pstl/backends/statpart/uninitialized_move.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index f2657b9810faa0..e3670e150f16cb 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -711,6 +711,9 @@ set(files
__pstl/backends/statpart/stable_sort.h
__pstl/backends/statpart/swap_ranges.h
__pstl/backends/statpart/transform.h
+ __pstl/backends/statpart/transform_reduce.h
+ __pstl/backends/statpart/uninitialized_copy.h
+ __pstl/backends/statpart/uninitialized_move.h
__pstl/backends/std_thread.h
__pstl/cpu_algos/any_of.h
__pstl/cpu_algos/cpu_traits.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 61c484f91318f2..665d3d36d8a437 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -24,209 +24,8 @@
#include <__pstl/backends/statpart/stable_sort.h>
#include <__pstl/backends/statpart/swap_ranges.h>
#include <__pstl/backends/statpart/transform.h>
-
-#include <__algorithm/find_end.h>
-#include <__algorithm/find_if.h>
-#include <__algorithm/for_each.h>
-#include <__algorithm/is_heap_until.h>
-#include <__algorithm/lower_bound.h>
-#include <__algorithm/merge.h>
-#include <__algorithm/min.h>
-#include <__algorithm/min_element.h>
-#include <__algorithm/minmax_element.h>
-#include <__algorithm/mismatch.h>
-#include <__algorithm/move.h>
-#include <__algorithm/reverse.h>
-#include <__algorithm/search.h>
-#include <__algorithm/search_n.h>
-#include <__algorithm/stable_sort.h>
-#include <__algorithm/swap_ranges.h>
-#include <__algorithm/transform.h>
-#include <__algorithm/upper_bound.h>
-#include <__atomic/atomic.h>
-#include <__config>
-#include <__functional/identity.h>
-#include <__functional/operations.h>
-#include <__iterator/concepts.h>
-#include <__iterator/iterator_traits.h>
-#include <__iterator/move_iterator.h>
-#include <__iterator/next.h>
-#include <__iterator/reverse_iterator.h>
-#include <__memory/allocator.h>
-#include <__memory/construct_at.h>
-#include <__memory/destroy.h>
-#include <__memory/uninitialized_algorithms.h>
-#include <__memory/unique_ptr.h>
-#include <__numeric/reduce.h>
-#include <__numeric/transform_reduce.h>
-#include <__optional/nullopt_t.h>
-#include <__optional/optional.h>
-#include <__pstl/backend_fwd.h>
-#include <__pstl/cpu_algos/cpu_traits.h>
-#include <__pstl/cpu_algos/find_if.h>
-#include <__thread/thread.h>
-#include <__type_traits/is_assignable.h>
-#include <__type_traits/is_constructible.h>
-#include <__type_traits/is_execution_policy.h>
-#include <__type_traits/is_trivially_copyable.h>
-#include <__utility/convert_to_integral.h>
-#include <__utility/empty.h>
-#include <__utility/move.h>
-#include <__utility/pair.h>
-
-#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
-# pragma GCC system_header
-#endif
-
-_LIBCPP_PUSH_MACROS
-#include <__undef_macros>
-
-#if _LIBCPP_STD_VER >= 17
-
-_LIBCPP_BEGIN_NAMESPACE_STD
-_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
-namespace __pstl {
-
-template <class _ExecutionPolicy>
-struct __transform_reduce<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation>
- _LIBCPP_HIDE_FROM_ABI optional<_Tp>
- operator()(_Policy&&,
- _ForwardIterator __first,
- _ForwardIterator __last,
- _Tp __init,
- _BinaryOperation __reduce,
- _UnaryOperation __transform) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
- return __statpart::__transform_reduce(
- std::move(__first),
- std::move(__last),
- [__transform](_ForwardIterator __iter) { return __transform(*__iter); },
- std::move(__init),
- __reduce,
- [__transform, __reduce](auto __brick_first, auto __brick_last, _Tp __brick_init) {
- return std::transform_reduce(
- std::move(__brick_first),
- std::move(__brick_last),
- std::move(__brick_init),
- std::move(__reduce),
- std::move(__transform));
- });
- } else {
- return std::transform_reduce(
- std::move(__first), std::move(__last), std::move(__init), std::move(__reduce), std::move(__transform));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __transform_reduce_binary<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy,
- class _ForwardIterator1,
- class _ForwardIterator2,
- class _Tp,
- class _BinaryOperation1,
- class _BinaryOperation2>
- _LIBCPP_HIDE_FROM_ABI optional<_Tp> operator()(
- _Policy&&,
- _ForwardIterator1 __first1,
- _ForwardIterator1 __last1,
- _ForwardIterator2 __first2,
- _Tp __init,
- _BinaryOperation1 __reduce,
- _BinaryOperation2 __transform) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
- return __statpart::__transform_reduce(
- __first1,
- std::move(__last1),
- [__first1, __first2, __transform](_ForwardIterator1 __iter) {
- return __transform(*__iter, *(__first2 + (__iter - __first1)));
- },
- std::move(__init),
- std::move(__reduce),
- [__first1, __first2, __reduce, __transform](
- _ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _Tp __brick_init) {
- return std::transform_reduce(
- __brick_first,
- std::move(__brick_last),
- __first2 + (__brick_first - __first1),
- std::move(__brick_init),
- std::move(__reduce),
- std::move(__transform));
- });
- } else {
- return std::transform_reduce(
- std::move(__first1),
- std::move(__last1),
- std::move(__first2),
- std::move(__init),
- std::move(__reduce),
- std::move(__transform));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __uninitialized_copy<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
- _Policy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
- auto __res = __statpart::__for_each_iter_pair(
- __first,
- __last,
- __result,
- [](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _ForwardIterator2 __brick_result) {
- std::uninitialized_copy(std::move(__brick_first), std::move(__brick_last), std::move(__brick_result));
- });
- if (!__res) {
- return std::nullopt; // Failed to run the algorithm, propagate the error.
- }
- return __result + (__last - __first);
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::uninitialized_copy(std::move(__first), std::move(__last), std::move(__result));
- }
- }
-};
-
-template <class _ExecutionPolicy>
-struct __uninitialized_move<__statpart_backend_tag, _ExecutionPolicy> {
- template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
- _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
- _Policy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result) const noexcept {
- if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
- __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
- auto __res = __statpart::__for_each_iter_pair(
- __first,
- __last,
- __result,
- [](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _ForwardIterator2 __brick_result) {
- std::uninitialized_move(std::move(__brick_first), std::move(__brick_last), std::move(__brick_result));
- });
- if (!__res) {
- return std::nullopt; // Failed to run the algorithm, propagate the error.
- }
- return __result + (__last - __first);
- } else {
- // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
- return std::uninitialized_move(std::move(__first), std::move(__last), std::move(__result));
- }
- }
-};
-
-} // namespace __pstl
-_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
-_LIBCPP_END_NAMESPACE_STD
-
-#endif // _LIBCPP_STD_VER >= 17
-
-_LIBCPP_POP_MACROS
+#include <__pstl/backends/statpart/transform_reduce.h>
+#include <__pstl/backends/statpart/uninitialized_copy.h>
+#include <__pstl/backends/statpart/uninitialized_move.h>
#endif // _LIBCPP___PSTL_STATPART_H
diff --git a/libcxx/include/__pstl/backends/statpart/transform_reduce.h b/libcxx/include/__pstl/backends/statpart/transform_reduce.h
new file mode 100644
index 00000000000000..c57fc7f0c269c8
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/transform_reduce.h
@@ -0,0 +1,126 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_TRANSFORM_REDUCE_H
+#define _LIBCPP___PSTL_STATPART_TRANSFORM_REDUCE_H
+
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__numeric/transform_reduce.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __transform_reduce<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation>
+ _LIBCPP_HIDE_FROM_ABI optional<_Tp>
+ operator()(_Policy&&,
+ _ForwardIterator __first,
+ _ForwardIterator __last,
+ _Tp __init,
+ _BinaryOperation __reduce,
+ _UnaryOperation __transform) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __statpart::__transform_reduce(
+ std::move(__first),
+ std::move(__last),
+ [__transform](_ForwardIterator __iter) { return __transform(*__iter); },
+ std::move(__init),
+ __reduce,
+ [__transform, __reduce](auto __brick_first, auto __brick_last, _Tp __brick_init) {
+ return std::transform_reduce(
+ std::move(__brick_first),
+ std::move(__brick_last),
+ std::move(__brick_init),
+ std::move(__reduce),
+ std::move(__transform));
+ });
+ } else {
+ return std::transform_reduce(
+ std::move(__first), std::move(__last), std::move(__init), std::move(__reduce), std::move(__transform));
+ }
+ }
+};
+
+template <class _ExecutionPolicy>
+struct __transform_reduce_binary<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy,
+ class _ForwardIterator1,
+ class _ForwardIterator2,
+ class _Tp,
+ class _BinaryOperation1,
+ class _BinaryOperation2>
+ _LIBCPP_HIDE_FROM_ABI optional<_Tp> operator()(
+ _Policy&&,
+ _ForwardIterator1 __first1,
+ _ForwardIterator1 __last1,
+ _ForwardIterator2 __first2,
+ _Tp __init,
+ _BinaryOperation1 __reduce,
+ _BinaryOperation2 __transform) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ return __statpart::__transform_reduce(
+ __first1,
+ std::move(__last1),
+ [__first1, __first2, __transform](_ForwardIterator1 __iter) {
+ return __transform(*__iter, *(__first2 + (__iter - __first1)));
+ },
+ std::move(__init),
+ std::move(__reduce),
+ [__first1, __first2, __reduce, __transform](
+ _ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _Tp __brick_init) {
+ return std::transform_reduce(
+ __brick_first,
+ std::move(__brick_last),
+ __first2 + (__brick_first - __first1),
+ std::move(__brick_init),
+ std::move(__reduce),
+ std::move(__transform));
+ });
+ } else {
+ return std::transform_reduce(
+ std::move(__first1),
+ std::move(__last1),
+ std::move(__first2),
+ std::move(__init),
+ std::move(__reduce),
+ std::move(__transform));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_TRANSFORM_REDUCE_H
diff --git a/libcxx/include/__pstl/backends/statpart/uninitialized_copy.h b/libcxx/include/__pstl/backends/statpart/uninitialized_copy.h
new file mode 100644
index 00000000000000..1c2fba6295ff26
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/uninitialized_copy.h
@@ -0,0 +1,69 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_UNITIALIZED_COPY_H
+#define _LIBCPP___PSTL_STATPART_UNITIALIZED_COPY_H
+
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__memory/uninitialized_algorithms.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __uninitialized_copy<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
+ _Policy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ auto __res = __statpart::__for_each_iter_pair(
+ __first,
+ __last,
+ __result,
+ [](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _ForwardIterator2 __brick_result) {
+ std::uninitialized_copy(std::move(__brick_first), std::move(__brick_last), std::move(__brick_result));
+ });
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ return __result + (__last - __first);
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::uninitialized_copy(std::move(__first), std::move(__last), std::move(__result));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_UNITIALIZED_COPY_H
diff --git a/libcxx/include/__pstl/backends/statpart/uninitialized_move.h b/libcxx/include/__pstl/backends/statpart/uninitialized_move.h
new file mode 100644
index 00000000000000..5596b16b087b4b
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/uninitialized_move.h
@@ -0,0 +1,69 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_STATPART_UNITIALIZED_MOVE_H
+#define _LIBCPP___PSTL_STATPART_UNITIALIZED_MOVE_H
+
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__iterator/iterator_traits.h>
+#include <__memory/uninitialized_algorithms.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _ExecutionPolicy>
+struct __uninitialized_move<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator1, class _ForwardIterator2>
+ _LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator2> operator()(
+ _Policy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator1>::value &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator2>::value) {
+ auto __res = __statpart::__for_each_iter_pair(
+ __first,
+ __last,
+ __result,
+ [](_ForwardIterator1 __brick_first, _ForwardIterator1 __brick_last, _ForwardIterator2 __brick_result) {
+ std::uninitialized_move(std::move(__brick_first), std::move(__brick_last), std::move(__brick_result));
+ });
+ if (!__res) {
+ return std::nullopt; // Failed to run the algorithm, propagate the error.
+ }
+ return __result + (__last - __first);
+ } else {
+ // Non-random access iterators cannot be processed in parallel, fall back to the sequential implementation.
+ return std::uninitialized_move(std::move(__first), std::move(__last), std::move(__result));
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_STATPART_UNITIALIZED_MOVE_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 92c2118c8760e8..4540b7ccf77c9d 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2462,6 +2462,15 @@ module std {
module transform {
header "__pstl/backends/statpart/transform.h"
}
+ module transform_reduce {
+ header "__pstl/backends/statpart/transform_reduce.h"
+ }
+ module uninitialized_copy {
+ header "__pstl/backends/statpart/uninitialized_copy.h"
+ }
+ module uninitialized_move {
+ header "__pstl/backends/statpart/uninitialized_move.h"
+ }
header "__pstl/backends/statpart.h"
}
module std_thread {
>From dfa5a4b9fdcbb40d449f195e37e530755e02b023 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Tue, 13 Oct 2026 17:29:52 +0100
Subject: [PATCH 13/24] Polished the comments, added an assertion on the min
partition size, added earlier bailing out when there's not enough elements.
---
.../include/__pstl/backends/statpart/basis.h | 47 +++++++++++--------
1 file changed, 28 insertions(+), 19 deletions(-)
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index 75dc2a5e159dcf..a5d66d808b8dd5 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -14,6 +14,7 @@
#include <__algorithm/min.h>
#include <__algorithm/move.h>
#include <__algorithm/upper_bound.h>
+#include <__assert>
#include <__atomic/atomic.h>
#include <__config>
#include <__functional/identity.h>
@@ -52,24 +53,27 @@ namespace __pstl {
namespace __statpart {
-// The backbone of the backend: applies the function to each of the indices [0, __chunk_count] in parallel.
+// The backbone of the backend: applies the function to each of the indices [0, __chunk_count) in parallel.
// The actual number of parallel workers and the order of application is unspecified.
_LIBCPP_EXPORTED_FROM_ABI void
__apply(size_t __chunk_count, void* __context, void (*__func)(void* __context, size_t __chunk)) noexcept;
-// The maximum number of partitions to be processed per each CPU
+// The maximum number of partitions to be processed per CPU
constexpr inline size_t __max_partitions_per_cpu = 8;
// The minimum number of elements to sort serially in a single partition
constexpr inline size_t __min_sort_partition_size = 256;
enum class __partitioning_result {
- // returned when the partitioning result is not nullopt
+ // Returned when the partitioning result is not a nullopt.
__ok,
- // returned when there is not enough element to produce a single partition or the final number of partitions is less
- // than 2, i.e. there's no reason to process in parallel
+ // Returned when there are not enough element to produce a single partition or the final number of partitions is fewer
+ // than 2, i.e. there's no reason to process the input in parallel.
__not_enough_elements,
- // later: returned memory couldn't be allocated for the stashed forward iterators
+ // There's fewer than 2 processors available in the system, or their number is unknown, which is treated as having
+ // a single processor. In either case, fall back to serial processing.
+ __no_parallelism,
+ // Reserved for later: returned when memory could not be allocated for the stashed forward iterators.
__bad_alloc
};
@@ -79,9 +83,9 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_range {
_Iterator __last;
};
-// __iterator_partitions<Iter> holds a finite per-determined set of partitions and allows to query each iterator range
+// __iterator_partitions<Iter> holds a finite pre-determined set of partitions and allows querying each iterator range
// in O(1) time.
-// TODO: Later to support forward iterators via a second implicit template argument.
+// TODO: Support forward iterators later via a second implicit template argument and its specializations.
template <class _Iterator>
struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
_LIBCPP_HIDE_FROM_ABI size_t __partitions_count() const { return __partitions_count_; }
@@ -106,11 +110,11 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
__partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
};
-// Splits the input iterator range into a finite set of partitions, such as that every iterator sub-range can queried in
+// Splits the input iterator range into a finite set of partitions, such that every iterator sub-range can queried in
// a O(1) time later.
-// In case partitioning failed returns nullopt and reports the reason in __result.
+// If partitioning fails, returns nullopt and reports the reason in __result.
// __min_partition_size specifies the minimal number of elements in every partition.
-// __prefix_to_skip and __suffix_to_skip allow to optionally trim the input range by specified numbers of elements
+// __prefix_to_skip and __suffix_to_skip allows optionally trimming the input range by the specified number of elements
// before partitioning.
// __workers_count is normally implicitly queried from hardware_concurrency() and should be explicitly specified only in
// unit tests.
@@ -123,13 +127,19 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
size_t __prefix_to_skip = 0,
size_t __suffix_to_skip = 0,
size_t __workers_count = thread::hardware_concurrency()) noexcept {
- // forward iteratos to be supported later
+ // Forward iterators will be supported later.
static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
- // TODO: assert( __min_partition_size > 0 )?
- // TODO: assert( __workers_count > 0 )?
+
+ _LIBCPP_ASSERT_INTERNAL(__min_partition_size >= 1, "minimum partition size must be at least 1 element");
+
+ if (__workers_count < 2) {
+ // There's no available parallelism in the system, no reason to perform partitioning.
+ __result = __partitioning_result::__no_parallelism;
+ return {};
+ }
size_t __count = static_cast<size_t>(std::distance(__first, __last));
- if (__count < __prefix_to_skip + __suffix_to_skip) {
+ if (__count <= __prefix_to_skip + __suffix_to_skip) {
__result = __partitioning_result::__not_enough_elements;
return {};
}
@@ -139,7 +149,6 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
size_t __partition_count = std::min(__max_partitions, __effective_count / __min_partition_size);
if (__partition_count < 2) {
- // TODO: verify that's what we want
__result = __partitioning_result::__not_enough_elements;
return {}; // not enough input data to process in parallel => no reason to build partitions
}
@@ -155,7 +164,7 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
}
// for_each with explicit partitioning:
-// calls __func in parallel with every iterator ranges provided in __partitions
+// calls __func in parallel with every iterator range provided in __partitions.
template <class _Iterator, class _Functor>
_LIBCPP_HIDE_FROM_ABI void __for_each_partition(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
struct __ctx_t {
@@ -169,7 +178,7 @@ _LIBCPP_HIDE_FROM_ABI void __for_each_partition(const __iterator_partitions<_Ite
});
}
-// Calls __func(index) for every index in [0, __count) in parallel
+// Calls __func(index) for every index in [0, __count) in parallel.
template <class _Functor>
_LIBCPP_HIDE_FROM_ABI void __for_each_index(size_t __count, _Functor __func) {
__statpart::__apply(__count, &__func, [](void* __context, size_t __index) {
@@ -177,7 +186,7 @@ _LIBCPP_HIDE_FROM_ABI void __for_each_index(size_t __count, _Functor __func) {
});
}
-// for_each that calls __func in parallel against every partition index available in __partitions
+// for_each that calls __func in parallel for every partition index available in __partitions.
template <class _Iterator, class _Functor>
_LIBCPP_HIDE_FROM_ABI void
__for_each_partition_index(const __iterator_partitions<_Iterator>& __partitions, _Functor __func) {
>From 42cb3b18e1c5a72869a90e751e0d6f99ba895090 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Wed, 14 Oct 2026 09:35:40 +0100
Subject: [PATCH 14/24] Reverted the default on Apple to libdispatch, added a
verbose check in statpart.cpp
---
libcxx/CMakeLists.txt | 2 +-
libcxx/src/pstl/statpart.cpp | 7 ++++++-
2 files changed, 7 insertions(+), 2 deletions(-)
diff --git a/libcxx/CMakeLists.txt b/libcxx/CMakeLists.txt
index 55b7a5fc709b37..5478c38afa781d 100644
--- a/libcxx/CMakeLists.txt
+++ b/libcxx/CMakeLists.txt
@@ -321,7 +321,7 @@ option(LIBCXX_HAS_EXTERNAL_THREAD_API
option(LIBCXX_HAS_C11_THREAD_API "Ignore auto-detection and force use of C11 thread API" OFF)
if (APPLE)
- set(LIBCXX_PSTL_BACKEND_DEFAULT_VALUE "statpart")
+ set(LIBCXX_PSTL_BACKEND_DEFAULT_VALUE "libdispatch")
elseif(LIBCXX_ENABLE_THREADS)
set(LIBCXX_PSTL_BACKEND_DEFAULT_VALUE "std_thread")
else()
diff --git a/libcxx/src/pstl/statpart.cpp b/libcxx/src/pstl/statpart.cpp
index 903ebbf000128e..6822288fa71a1e 100644
--- a/libcxx/src/pstl/statpart.cpp
+++ b/libcxx/src/pstl/statpart.cpp
@@ -8,11 +8,16 @@
#include <__config>
#include <__pstl/backends/statpart.h>
-#include <dispatch/dispatch.h>
// Hardcoded to use libdispatch for now.
// CMake switches to be introduced later.
+#if __has_include(<dispatch/dispatch.h>)
+# include <dispatch/dispatch.h>
+#else
+# error "Currently statpart only supports parallelism provided by libdispatch."
+#endif
+
_LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl::__statpart {
>From eaf172271662f3a24999193ed83396900a356d58 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Wed, 14 Oct 2026 09:43:17 +0100
Subject: [PATCH 15/24] Match the signature of __partition more closely when
declaring a friend
---
libcxx/include/__pstl/backends/statpart/basis.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index a5d66d808b8dd5..01b5211fd67c18 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -106,7 +106,7 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
size_t __partitions_count_;
template <class _Iter>
- friend optional<__iterator_partitions<_Iter>>
+ friend optional<__iterator_partitions<_Iter>> _LIBCPP_HIDE_FROM_ABI
__partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
};
>From 2ae9c49fa3916fb4d882902503b51767b8ca079a Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Wed, 14 Oct 2026 11:31:09 +0100
Subject: [PATCH 16/24] More typo fixes
---
libcxx/include/__pstl/backend.h | 2 +-
libcxx/include/__pstl/backends/statpart/basis.h | 2 +-
2 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/libcxx/include/__pstl/backend.h b/libcxx/include/__pstl/backend.h
index 75ccb5063c2919..1626a929f9707d 100644
--- a/libcxx/include/__pstl/backend.h
+++ b/libcxx/include/__pstl/backend.h
@@ -35,7 +35,7 @@ _LIBCPP_PUSH_MACROS
# include <__pstl/backends/statpart.h>
# endif
-// statpart pulls <cerrno> in transitively via __thread/thread.h, make sure this include is not is conditional/flaky
+// statpart pulls <cerrno> in transitively via __thread/thread.h, make sure this include is not conditional/flaky.
# include <cerrno>
#endif // _LIBCPP_STD_VER >= 17
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index 01b5211fd67c18..470785bd8b3f47 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -110,7 +110,7 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
__partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
};
-// Splits the input iterator range into a finite set of partitions, such that every iterator sub-range can queried in
+// Splits the input iterator range into a finite set of partitions, such that every iterator sub-range can be queried in
// a O(1) time later.
// If partitioning fails, returns nullopt and reports the reason in __result.
// __min_partition_size specifies the minimal number of elements in every partition.
>From 0200644cf5f98efbea976a3d30e64c7e7055eb11 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Wed, 14 Oct 2026 17:54:19 +0100
Subject: [PATCH 17/24] Fixed the include guards
---
libcxx/include/__pstl/backends/statpart.h | 6 +++---
libcxx/include/__pstl/backends/statpart/basis.h | 6 +++---
libcxx/include/__pstl/backends/statpart/find_end.h | 6 +++---
libcxx/include/__pstl/backends/statpart/find_if.h | 6 +++---
libcxx/include/__pstl/backends/statpart/for_each.h | 6 +++---
libcxx/include/__pstl/backends/statpart/is_heap_until.h | 6 +++---
libcxx/include/__pstl/backends/statpart/merge.h | 6 +++---
libcxx/include/__pstl/backends/statpart/min_element.h | 6 +++---
libcxx/include/__pstl/backends/statpart/minmax_element.h | 6 +++---
libcxx/include/__pstl/backends/statpart/mismatch.h | 6 +++---
libcxx/include/__pstl/backends/statpart/reverse.h | 6 +++---
libcxx/include/__pstl/backends/statpart/search.h | 6 +++---
libcxx/include/__pstl/backends/statpart/search_n.h | 6 +++---
libcxx/include/__pstl/backends/statpart/stable_sort.h | 6 +++---
libcxx/include/__pstl/backends/statpart/swap_ranges.h | 6 +++---
libcxx/include/__pstl/backends/statpart/transform.h | 6 +++---
libcxx/include/__pstl/backends/statpart/transform_reduce.h | 6 +++---
.../include/__pstl/backends/statpart/uninitialized_copy.h | 6 +++---
.../include/__pstl/backends/statpart/uninitialized_move.h | 6 +++---
19 files changed, 57 insertions(+), 57 deletions(-)
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index 665d3d36d8a437..a77671588874ff 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_H
-#define _LIBCPP___PSTL_STATPART_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_H
#include <__pstl/backends/statpart/basis.h>
#include <__pstl/backends/statpart/find_end.h>
@@ -28,4 +28,4 @@
#include <__pstl/backends/statpart/uninitialized_copy.h>
#include <__pstl/backends/statpart/uninitialized_move.h>
-#endif // _LIBCPP___PSTL_STATPART_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_H
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index 470785bd8b3f47..754946dc977207 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_BASIS_H
-#define _LIBCPP___PSTL_STATPART_BASIS_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_BASIS_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_BASIS_H
#include <__algorithm/lower_bound.h>
#include <__algorithm/merge.h>
@@ -500,4 +500,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_BASIS_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_BASIS_H
diff --git a/libcxx/include/__pstl/backends/statpart/find_end.h b/libcxx/include/__pstl/backends/statpart/find_end.h
index 2f022024077258..5324d840219120 100644
--- a/libcxx/include/__pstl/backends/statpart/find_end.h
+++ b/libcxx/include/__pstl/backends/statpart/find_end.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_FIND_END_H
-#define _LIBCPP___PSTL_STATPART_FIND_END_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_FIND_END_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_FIND_END_H
#include <__algorithm/find_end.h>
#include <__config>
@@ -99,4 +99,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_FIND_END_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_FIND_END_H
diff --git a/libcxx/include/__pstl/backends/statpart/find_if.h b/libcxx/include/__pstl/backends/statpart/find_if.h
index c7273a75e54bba..0913e91a18e3f2 100644
--- a/libcxx/include/__pstl/backends/statpart/find_if.h
+++ b/libcxx/include/__pstl/backends/statpart/find_if.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_FIND_IF_H
-#define _LIBCPP___PSTL_STATPART_FIND_IF_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_FIND_IF_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_FIND_IF_H
#include <__algorithm/find_if.h>
#include <__config>
@@ -64,4 +64,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_FIND_IF_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_FIND_IF_H
diff --git a/libcxx/include/__pstl/backends/statpart/for_each.h b/libcxx/include/__pstl/backends/statpart/for_each.h
index b2199a828cb185..53aeef91fad574 100644
--- a/libcxx/include/__pstl/backends/statpart/for_each.h
+++ b/libcxx/include/__pstl/backends/statpart/for_each.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_FOR_EACH_H
-#define _LIBCPP___PSTL_STATPART_FOR_EACH_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_FOR_EACH_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_FOR_EACH_H
#include <__algorithm/for_each.h>
#include <__config>
@@ -60,4 +60,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_FOR_EACH_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_FOR_EACH_H
diff --git a/libcxx/include/__pstl/backends/statpart/is_heap_until.h b/libcxx/include/__pstl/backends/statpart/is_heap_until.h
index 88e0c11de11a2e..e983a4835272a3 100644
--- a/libcxx/include/__pstl/backends/statpart/is_heap_until.h
+++ b/libcxx/include/__pstl/backends/statpart/is_heap_until.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_IS_HEAP_UNTIL_H
-#define _LIBCPP___PSTL_STATPART_IS_HEAP_UNTIL_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_IS_HEAP_UNTIL_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_IS_HEAP_UNTIL_H
#include <__algorithm/is_heap_until.h>
#include <__config>
@@ -102,4 +102,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_IS_HEAP_UNTIL_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_IS_HEAP_UNTIL_H
diff --git a/libcxx/include/__pstl/backends/statpart/merge.h b/libcxx/include/__pstl/backends/statpart/merge.h
index bd9ef68d413a26..5c3ccc7d596576 100644
--- a/libcxx/include/__pstl/backends/statpart/merge.h
+++ b/libcxx/include/__pstl/backends/statpart/merge.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_MERGE_H
-#define _LIBCPP___PSTL_STATPART_MERGE_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_MERGE_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_MERGE_H
#include <__algorithm/merge.h>
#include <__config>
@@ -85,4 +85,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_MERGE_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_MERGE_H
diff --git a/libcxx/include/__pstl/backends/statpart/min_element.h b/libcxx/include/__pstl/backends/statpart/min_element.h
index 09e3117c81d475..d267c54732ab91 100644
--- a/libcxx/include/__pstl/backends/statpart/min_element.h
+++ b/libcxx/include/__pstl/backends/statpart/min_element.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_MIN_ELEMENT_H
-#define _LIBCPP___PSTL_STATPART_MIN_ELEMENT_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_MIN_ELEMENT_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_MIN_ELEMENT_H
#include <__algorithm/min_element.h>
#include <__config>
@@ -82,4 +82,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_MIN_ELEMENT_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_MIN_ELEMENT_H
diff --git a/libcxx/include/__pstl/backends/statpart/minmax_element.h b/libcxx/include/__pstl/backends/statpart/minmax_element.h
index bde4bb515999b4..40138be1b6e735 100644
--- a/libcxx/include/__pstl/backends/statpart/minmax_element.h
+++ b/libcxx/include/__pstl/backends/statpart/minmax_element.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_MINMAX_ELEMENT_H
-#define _LIBCPP___PSTL_STATPART_MINMAX_ELEMENT_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_MINMAX_ELEMENT_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_MINMAX_ELEMENT_H
#include <__algorithm/minmax_element.h>
#include <__config>
@@ -87,4 +87,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_MINMAX_ELEMENT_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_MINMAX_ELEMENT_H
diff --git a/libcxx/include/__pstl/backends/statpart/mismatch.h b/libcxx/include/__pstl/backends/statpart/mismatch.h
index b049b690e12bf8..e2a4ee64da1b28 100644
--- a/libcxx/include/__pstl/backends/statpart/mismatch.h
+++ b/libcxx/include/__pstl/backends/statpart/mismatch.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_MISMATCH_H
-#define _LIBCPP___PSTL_STATPART_MISMATCH_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_MISMATCH_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_MISMATCH_H
#include <__algorithm/min.h>
#include <__algorithm/mismatch.h>
@@ -88,4 +88,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_MISMATCH_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_MISMATCH_H
diff --git a/libcxx/include/__pstl/backends/statpart/reverse.h b/libcxx/include/__pstl/backends/statpart/reverse.h
index 4e3d147475568b..9b9aee1e6ddbde 100644
--- a/libcxx/include/__pstl/backends/statpart/reverse.h
+++ b/libcxx/include/__pstl/backends/statpart/reverse.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_REVERSE_H
-#define _LIBCPP___PSTL_STATPART_REVERSE_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_REVERSE_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_REVERSE_H
#include <__algorithm/reverse.h>
#include <__algorithm/swap_ranges.h>
@@ -72,4 +72,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_REVERSE_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_REVERSE_H
diff --git a/libcxx/include/__pstl/backends/statpart/search.h b/libcxx/include/__pstl/backends/statpart/search.h
index a831fb8ef63d0e..8c238b6fe055f1 100644
--- a/libcxx/include/__pstl/backends/statpart/search.h
+++ b/libcxx/include/__pstl/backends/statpart/search.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_SEARCH_H
-#define _LIBCPP___PSTL_STATPART_SEARCH_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_SEARCH_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_SEARCH_H
#include <__algorithm/search.h>
#include <__config>
@@ -99,4 +99,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_SEARCH_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_SEARCH_H
diff --git a/libcxx/include/__pstl/backends/statpart/search_n.h b/libcxx/include/__pstl/backends/statpart/search_n.h
index 1ab5b72f2c5b67..394f8823328665 100644
--- a/libcxx/include/__pstl/backends/statpart/search_n.h
+++ b/libcxx/include/__pstl/backends/statpart/search_n.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_SEARCH_N_H
-#define _LIBCPP___PSTL_STATPART_SEARCH_N_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_SEARCH_N_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_SEARCH_N_H
#include <__algorithm/search_n.h>
#include <__config>
@@ -99,4 +99,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_SEARCH_N_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_SEARCH_N_H
diff --git a/libcxx/include/__pstl/backends/statpart/stable_sort.h b/libcxx/include/__pstl/backends/statpart/stable_sort.h
index 22708ba4a8fcd1..aa42da2a453c49 100644
--- a/libcxx/include/__pstl/backends/statpart/stable_sort.h
+++ b/libcxx/include/__pstl/backends/statpart/stable_sort.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_STABLE_SORT_H
-#define _LIBCPP___PSTL_STATPART_STABLE_SORT_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_STABLE_SORT_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_STABLE_SORT_H
#include <__algorithm/stable_sort.h>
#include <__config>
@@ -58,4 +58,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_STABLE_SORT_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_STABLE_SORT_H
diff --git a/libcxx/include/__pstl/backends/statpart/swap_ranges.h b/libcxx/include/__pstl/backends/statpart/swap_ranges.h
index 71a4c40cebe516..0e6c50aedef02b 100644
--- a/libcxx/include/__pstl/backends/statpart/swap_ranges.h
+++ b/libcxx/include/__pstl/backends/statpart/swap_ranges.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_SWAP_RANGES_H
-#define _LIBCPP___PSTL_STATPART_SWAP_RANGES_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_SWAP_RANGES_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_SWAP_RANGES_H
#include <__algorithm/swap_ranges.h>
#include <__config>
@@ -67,4 +67,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_SWAP_RANGES_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_SWAP_RANGES_H
diff --git a/libcxx/include/__pstl/backends/statpart/transform.h b/libcxx/include/__pstl/backends/statpart/transform.h
index 69914bc2c5055b..0e303ff4334dfc 100644
--- a/libcxx/include/__pstl/backends/statpart/transform.h
+++ b/libcxx/include/__pstl/backends/statpart/transform.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_TRANSFORM_H
-#define _LIBCPP___PSTL_STATPART_TRANSFORM_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_TRANSFORM_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_TRANSFORM_H
#include <__algorithm/transform.h>
#include <__config>
@@ -99,4 +99,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_TRANSFORM_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_TRANSFORM_H
diff --git a/libcxx/include/__pstl/backends/statpart/transform_reduce.h b/libcxx/include/__pstl/backends/statpart/transform_reduce.h
index c57fc7f0c269c8..80110503e696dd 100644
--- a/libcxx/include/__pstl/backends/statpart/transform_reduce.h
+++ b/libcxx/include/__pstl/backends/statpart/transform_reduce.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_TRANSFORM_REDUCE_H
-#define _LIBCPP___PSTL_STATPART_TRANSFORM_REDUCE_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_TRANSFORM_REDUCE_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_TRANSFORM_REDUCE_H
#include <__config>
#include <__iterator/concepts.h>
@@ -123,4 +123,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_TRANSFORM_REDUCE_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_TRANSFORM_REDUCE_H
diff --git a/libcxx/include/__pstl/backends/statpart/uninitialized_copy.h b/libcxx/include/__pstl/backends/statpart/uninitialized_copy.h
index 1c2fba6295ff26..e9dd9f026ceba0 100644
--- a/libcxx/include/__pstl/backends/statpart/uninitialized_copy.h
+++ b/libcxx/include/__pstl/backends/statpart/uninitialized_copy.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_UNITIALIZED_COPY_H
-#define _LIBCPP___PSTL_STATPART_UNITIALIZED_COPY_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_UNITIALIZED_COPY_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_UNITIALIZED_COPY_H
#include <__config>
#include <__iterator/concepts.h>
@@ -66,4 +66,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_UNITIALIZED_COPY_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_UNITIALIZED_COPY_H
diff --git a/libcxx/include/__pstl/backends/statpart/uninitialized_move.h b/libcxx/include/__pstl/backends/statpart/uninitialized_move.h
index 5596b16b087b4b..5b763c28b4d86a 100644
--- a/libcxx/include/__pstl/backends/statpart/uninitialized_move.h
+++ b/libcxx/include/__pstl/backends/statpart/uninitialized_move.h
@@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//
-#ifndef _LIBCPP___PSTL_STATPART_UNITIALIZED_MOVE_H
-#define _LIBCPP___PSTL_STATPART_UNITIALIZED_MOVE_H
+#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_UNITIALIZED_MOVE_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_UNITIALIZED_MOVE_H
#include <__config>
#include <__iterator/concepts.h>
@@ -66,4 +66,4 @@ _LIBCPP_END_NAMESPACE_STD
_LIBCPP_POP_MACROS
-#endif // _LIBCPP___PSTL_STATPART_UNITIALIZED_MOVE_H
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_UNITIALIZED_MOVE_H
>From b7469c4c65db2bde6e339bc94a33b68f5da84d7c Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Thu, 15 Oct 2026 12:38:12 +0100
Subject: [PATCH 18/24] Removed the misleading comment
---
libcxx/src/pstl/statpart.cpp | 1 -
1 file changed, 1 deletion(-)
diff --git a/libcxx/src/pstl/statpart.cpp b/libcxx/src/pstl/statpart.cpp
index 6822288fa71a1e..09fb8ddd5b3699 100644
--- a/libcxx/src/pstl/statpart.cpp
+++ b/libcxx/src/pstl/statpart.cpp
@@ -10,7 +10,6 @@
#include <__pstl/backends/statpart.h>
// Hardcoded to use libdispatch for now.
-// CMake switches to be introduced later.
#if __has_include(<dispatch/dispatch.h>)
# include <dispatch/dispatch.h>
>From e0e302dc6c7526c9e8e8a3808c32b81dfed3281f Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Thu, 15 Oct 2026 12:39:20 +0100
Subject: [PATCH 19/24] Qualify a function call
---
libcxx/include/__pstl/backends/statpart/basis.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index 754946dc977207..89f6a8c3e8b43d 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -220,7 +220,7 @@ _LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each_iter_pair(
_RandomAccessIterator1 __last1,
_RandomAccessIterator2 __first2,
_BrickFunction __f) {
- return __for_each(
+ return __statpart::__for_each(
__first1,
__last1,
[__first1, __first2, __f = std::move(__f)](
>From 746a7cc30bd38c8d2628fff4a3de36e2e9bff0bf Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Thu, 15 Oct 2026 16:18:58 +0100
Subject: [PATCH 20/24] __partition() no longer uses return arguments
---
.../include/__pstl/backends/statpart/basis.h | 101 ++++-----
.../pstl.statpart.partition.pass.cpp | 198 +++++++++---------
2 files changed, 147 insertions(+), 152 deletions(-)
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index 89f6a8c3e8b43d..2efe82f7c07a36 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -64,7 +64,7 @@ constexpr inline size_t __max_partitions_per_cpu = 8;
// The minimum number of elements to sort serially in a single partition
constexpr inline size_t __min_sort_partition_size = 256;
-enum class __partitioning_result {
+enum class __partitioning_status {
// Returned when the partitioning result is not a nullopt.
__ok,
// Returned when there are not enough element to produce a single partition or the final number of partitions is fewer
@@ -77,6 +77,9 @@ enum class __partitioning_result {
__bad_alloc
};
+template <class _Iterator>
+struct _LIBCPP_HIDE_FROM_ABI __partitioning_result;
+
template <class _Iterator>
struct _LIBCPP_HIDE_FROM_ABI __iterator_range {
_Iterator __first;
@@ -106,8 +109,16 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
size_t __partitions_count_;
template <class _Iter>
- friend optional<__iterator_partitions<_Iter>> _LIBCPP_HIDE_FROM_ABI
- __partition(_Iter, _Iter, __partitioning_result&, size_t, size_t, size_t, size_t) noexcept;
+ friend __partitioning_result<_Iter> _LIBCPP_HIDE_FROM_ABI
+ __partition(_Iter, _Iter, size_t, size_t, size_t, size_t) noexcept;
+};
+
+// This is basically a barebones std::expected<__iterator_partitions<_Iterator>, __partitioning_status>.
+// __partition() uses this to avoid resorting to return arguments.
+template <class _Iterator>
+struct _LIBCPP_HIDE_FROM_ABI __partitioning_result {
+ optional<__iterator_partitions<_Iterator>> __value_;
+ __partitioning_status __status_ = __partitioning_status::__ok;
};
// Splits the input iterator range into a finite set of partitions, such that every iterator sub-range can be queried in
@@ -119,14 +130,13 @@ struct _LIBCPP_HIDE_FROM_ABI __iterator_partitions {
// __workers_count is normally implicitly queried from hardware_concurrency() and should be explicitly specified only in
// unit tests.
template <class _ForwardIterator>
-_LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partition(
- _ForwardIterator __first,
- _ForwardIterator __last,
- __partitioning_result& __result,
- size_t __min_partition_size = 1,
- size_t __prefix_to_skip = 0,
- size_t __suffix_to_skip = 0,
- size_t __workers_count = thread::hardware_concurrency()) noexcept {
+_LIBCPP_HIDE_FROM_ABI __partitioning_result<_ForwardIterator>
+__partition(_ForwardIterator __first,
+ _ForwardIterator __last,
+ size_t __min_partition_size = 1,
+ size_t __prefix_to_skip = 0,
+ size_t __suffix_to_skip = 0,
+ size_t __workers_count = thread::hardware_concurrency()) noexcept {
// Forward iterators will be supported later.
static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
@@ -134,14 +144,12 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
if (__workers_count < 2) {
// There's no available parallelism in the system, no reason to perform partitioning.
- __result = __partitioning_result::__no_parallelism;
- return {};
+ return {nullopt, __partitioning_status::__no_parallelism};
}
size_t __count = static_cast<size_t>(std::distance(__first, __last));
if (__count <= __prefix_to_skip + __suffix_to_skip) {
- __result = __partitioning_result::__not_enough_elements;
- return {};
+ return {nullopt, __partitioning_status::__not_enough_elements};
}
size_t __effective_count = __count - __prefix_to_skip - __suffix_to_skip;
@@ -149,8 +157,8 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
size_t __partition_count = std::min(__max_partitions, __effective_count / __min_partition_size);
if (__partition_count < 2) {
- __result = __partitioning_result::__not_enough_elements;
- return {}; // not enough input data to process in parallel => no reason to build partitions
+ // not enough input data to process in parallel => no reason to build partitions
+ return {nullopt, __partitioning_status::__not_enough_elements};
}
__iterator_partitions<_ForwardIterator> __p;
@@ -159,8 +167,7 @@ _LIBCPP_HIDE_FROM_ABI optional<__iterator_partitions<_ForwardIterator>> __partit
__p.__partition_size_ = __effective_count / __partition_count;
__p.__remainder_ = __effective_count % __partition_count;
__p.__partitions_count_ = __partition_count;
- __result = __partitioning_result::__ok;
- return __p;
+ return {std::move(__p), __partitioning_status::__ok};
}
// for_each with explicit partitioning:
@@ -199,12 +206,11 @@ template <class _Iterator, class _Functor>
_LIBCPP_HIDE_FROM_ABI optional<__empty> __for_each(_Iterator __first, _Iterator __last, _Functor __func) {
if (__first == __last)
return __empty{};
- __partitioning_result __partition_result;
- auto __partitions = __statpart::__partition(__first, __last, __partition_result);
- if (__partitions) {
- __statpart::__for_each_partition(*__partitions, std::move(__func));
+ auto __partitions = __statpart::__partition(__first, __last);
+ if (__partitions.__value_) {
+ __statpart::__for_each_partition(*__partitions.__value_, std::move(__func));
return __empty{};
- } else if (__partition_result == __partitioning_result::__bad_alloc) {
+ } else if (__partitions.__status_ == __partitioning_status::__bad_alloc) {
return nullopt;
} else {
__func(__first, __last);
@@ -269,10 +275,9 @@ _LIBCPP_HIDE_FROM_ABI optional<_Value> __transform_reduce(
return __init;
// Split into partitions with at least 3 elements in each
- __partitioning_result __partitioning_result;
- auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, 3);
- if (!__partitions) {
- if (__partitioning_result == __partitioning_result::__bad_alloc) {
+ auto __partitions = __statpart::__partition(__first, __last, 3);
+ if (!__partitions.__value_) {
+ if (__partitions.__status_ == __partitioning_status::__bad_alloc) {
// Failed to allocate memory for the partitions, propagate the error up.
return nullopt;
} else {
@@ -281,7 +286,7 @@ _LIBCPP_HIDE_FROM_ABI optional<_Value> __transform_reduce(
}
}
- auto __destroy = [__count = __partitions->__partitions_count()](_Value* __ptr) {
+ auto __destroy = [__count = __partitions.__value_->__partitions_count()](_Value* __ptr) {
std::destroy_n(__ptr, __count);
std::allocator<_Value>().deallocate(__ptr, __count);
};
@@ -289,11 +294,11 @@ _LIBCPP_HIDE_FROM_ABI optional<_Value> __transform_reduce(
// TODO: use __uninitialized_buffer
// TODO: allocate one element per worker instead of one element per chunk
unique_ptr<_Value[], decltype(__destroy)> __values(
- std::allocator<_Value>().allocate(__partitions->__partitions_count()), __destroy);
+ std::allocator<_Value>().allocate(__partitions.__value_->__partitions_count()), __destroy);
// __dispatch_apply is noexcept
- __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
- auto __range = __partitions->__partition(__index);
+ __statpart::__for_each_partition_index(*__partitions.__value_, [&](size_t __index) {
+ auto __range = __partitions.__value_->__partition(__index);
std::__construct_at(
__values.get() + __index,
__reduction(__range.__first + 2,
@@ -303,7 +308,7 @@ _LIBCPP_HIDE_FROM_ABI optional<_Value> __transform_reduce(
return std::reduce(
std::make_move_iterator(__values.get()),
- std::make_move_iterator(__values.get() + __partitions->__partitions_count()),
+ std::make_move_iterator(__values.get() + __partitions.__value_->__partitions_count()),
std::move(__init),
__combiner);
}
@@ -331,10 +336,9 @@ __merge(_RandomAccessIterator1 __first1,
auto __find_cutpoint,
auto __merge_sub_ranges) -> optional<__empty> {
// Partition the longest input range evenly.
- __partitioning_result __partitioning_result;
- auto __partitions = __statpart::__partition(__long_first, __long_last, __partitioning_result);
- if (!__partitions) {
- if (__partitioning_result == __partitioning_result::__bad_alloc)
+ auto __partitions = __statpart::__partition(__long_first, __long_last);
+ if (!__partitions.__value_) {
+ if (__partitions.__status_ == __partitioning_status::__bad_alloc)
return nullopt;
// Not enough elements to run in parallel, merge everything serially.
__merge_sub_ranges(__long_first, __long_last, __short_first, __short_last, __result);
@@ -342,9 +346,9 @@ __merge(_RandomAccessIterator1 __first1,
}
// Process the partitions of the longest range in parallel.
- __statpart::__for_each_partition_index(*__partitions, [&](size_t __index) {
+ __statpart::__for_each_partition_index(*__partitions.__value_, [&](size_t __index) {
// Get the sub-range inside the longer input range.
- auto __range = __partitions->__partition(__index);
+ auto __range = __partitions.__value_->__partition(__index);
// Find the matching sub-range of the shorter input range: it will hold the elements that come after the last
// element of the previous long partition and up to the last element of this partition.
@@ -352,7 +356,7 @@ __merge(_RandomAccessIterator1 __first1,
auto __short_begin =
__index == 0 ? __short_first : __find_cutpoint(__short_first, __short_last, __range.__first[-1]);
auto __short_end =
- __index == __partitions->__partitions_count() - 1
+ __index == __partitions.__value_->__partitions_count() - 1
? __short_last
: __find_cutpoint(__short_first, __short_last, __range.__last[-1]);
@@ -399,10 +403,9 @@ _LIBCPP_HIDE_FROM_ABI optional<__empty>
__stable_sort(_RandomAccessIterator __first, _RandomAccessIterator __last, _Comp __comp, _LeafSort __leaf_sort) {
const auto __size = __last - __first;
- __partitioning_result __partitioning_result;
- auto __partitions = __statpart::__partition(__first, __last, __partitioning_result, __min_sort_partition_size);
- if (!__partitions) {
- if (__partitioning_result == __partitioning_result::__bad_alloc)
+ auto __partitions = __statpart::__partition(__first, __last, __min_sort_partition_size);
+ if (!__partitions.__value_) {
+ if (__partitions.__status_ == __partitioning_status::__bad_alloc)
return nullopt;
// Not enough elements to run in parallel, sort everything serially.
__leaf_sort(__first, __last, __comp);
@@ -428,23 +431,25 @@ __stable_sort(_RandomAccessIterator __first, _RandomAccessIterator __last, _Comp
*__first = std::move(__values.get()[__size - 1]);
__statpart::__for_each_partition(
- *__partitions, [&__leaf_sort, &__comp](_RandomAccessIterator __chunk_first, _RandomAccessIterator __chunk_last) {
+ *__partitions.__value_,
+ [&__leaf_sort, &__comp](_RandomAccessIterator __chunk_first, _RandomAccessIterator __chunk_last) {
__leaf_sort(std::move(__chunk_first), std::move(__chunk_last), __comp);
});
// A chunk span is a run of __chunk_span_size consecutive partitions, the last chunk can be shorter in case of odd
// number of chunk spans. Each pass merges the chunk spans pairwise, so the chunk span size doubles and the chunk
// count is halved (rounded up).
- const size_t __partition_count = __partitions->__partitions_count();
+ const size_t __partition_count = __partitions.__value_->__partitions_count();
size_t __chunk_span_size = 1;
size_t __chunk_spans = __partition_count;
// Offsets of the beginning and the end of the chunk from __first
auto __chunk_begin = [&](size_t __chunk) {
- return __partitions->__partition(__chunk * __chunk_span_size).__first - __first;
+ return __partitions.__value_->__partition(__chunk * __chunk_span_size).__first - __first;
};
auto __chunk_end = [&](size_t __chunk) {
- return __partitions->__partition(std::min((__chunk + 1) * __chunk_span_size, __partition_count) - 1).__last -
+ return __partitions.__value_->__partition(std::min((__chunk + 1) * __chunk_span_size, __partition_count) - 1)
+ .__last -
__first;
};
diff --git a/libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp b/libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp
index e9c973b94e7a23..457d97a39831de 100644
--- a/libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp
+++ b/libcxx/test/libcxx/algorithms/pstl.statpart.partition.pass.cpp
@@ -13,10 +13,9 @@
// UNSUPPORTED: libcpp-has-no-incomplete-pstl
// template <class Iterator>
-// optional<__iterator_partitions<Iterator>>
+// __partitioning_result<Iterator>
// __partition(Iterator __first,
// Iterator __last,
-// __partitioning_result& __result,
// size_t __min_partition_size,
// size_t __prefix_to_skip,
// size_t __suffix_to_skip,
@@ -38,140 +37,131 @@ void test_partition() {
{ // 1 element, 8 workers: no partitions
Container v(1);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 0, 0, 8);
- assert(partitions == std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__not_enough_elements);
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 1, 0, 0, 8);
+ assert(partitions.__value_ == std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__not_enough_elements);
}
{ // 2 elements, 8 workers: 2 partitions
// [0..1), [1..2)
Container v(2);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 0, 0, 8);
- assert(partitions != std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
- assert(partitions->__partitions_count() == 2);
- assert(partitions->__end() == v.end());
- assert(part_eq(partitions->__partition(0), v, 0, 1));
- assert(part_eq(partitions->__partition(1), v, 1, 2));
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 1, 0, 0, 8);
+ assert(partitions.__value_ != std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__ok);
+ assert(partitions.__value_->__partitions_count() == 2);
+ assert(partitions.__value_->__end() == v.end());
+ assert(part_eq(partitions.__value_->__partition(0), v, 0, 1));
+ assert(part_eq(partitions.__value_->__partition(1), v, 1, 2));
}
{ // 2 elements, min_size=2, 8 workers: no partitions
// [0..1), [1..2)
Container v(2);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 2, 0, 8);
- assert(partitions == std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__not_enough_elements);
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 1, 2, 0, 8);
+ assert(partitions.__value_ == std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__not_enough_elements);
}
{ // 20 elements, 2 workers: 16 partitions
// [0..2), [2..4), [4..6), [6..8), [8..9), [9..10), ..., [19, 20)
Container v(20);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 1, 0, 0, 2);
- assert(partitions != std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
- assert(partitions->__partitions_count() == 16);
- assert(partitions->__end() == v.end());
- assert(part_eq(partitions->__partition(0), v, 0, 2));
- assert(part_eq(partitions->__partition(1), v, 2, 4));
- assert(part_eq(partitions->__partition(2), v, 4, 6));
- assert(part_eq(partitions->__partition(3), v, 6, 8));
- assert(part_eq(partitions->__partition(4), v, 8, 9));
- assert(part_eq(partitions->__partition(5), v, 9, 10));
- assert(part_eq(partitions->__partition(6), v, 10, 11));
- assert(part_eq(partitions->__partition(7), v, 11, 12));
- assert(part_eq(partitions->__partition(8), v, 12, 13));
- assert(part_eq(partitions->__partition(9), v, 13, 14));
- assert(part_eq(partitions->__partition(10), v, 14, 15));
- assert(part_eq(partitions->__partition(11), v, 15, 16));
- assert(part_eq(partitions->__partition(12), v, 16, 17));
- assert(part_eq(partitions->__partition(13), v, 17, 18));
- assert(part_eq(partitions->__partition(14), v, 18, 19));
- assert(part_eq(partitions->__partition(15), v, 19, 20));
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 1, 0, 0, 2);
+ assert(partitions.__value_ != std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__ok);
+ assert(partitions.__value_->__partitions_count() == 16);
+ assert(partitions.__value_->__end() == v.end());
+ assert(part_eq(partitions.__value_->__partition(0), v, 0, 2));
+ assert(part_eq(partitions.__value_->__partition(1), v, 2, 4));
+ assert(part_eq(partitions.__value_->__partition(2), v, 4, 6));
+ assert(part_eq(partitions.__value_->__partition(3), v, 6, 8));
+ assert(part_eq(partitions.__value_->__partition(4), v, 8, 9));
+ assert(part_eq(partitions.__value_->__partition(5), v, 9, 10));
+ assert(part_eq(partitions.__value_->__partition(6), v, 10, 11));
+ assert(part_eq(partitions.__value_->__partition(7), v, 11, 12));
+ assert(part_eq(partitions.__value_->__partition(8), v, 12, 13));
+ assert(part_eq(partitions.__value_->__partition(9), v, 13, 14));
+ assert(part_eq(partitions.__value_->__partition(10), v, 14, 15));
+ assert(part_eq(partitions.__value_->__partition(11), v, 15, 16));
+ assert(part_eq(partitions.__value_->__partition(12), v, 16, 17));
+ assert(part_eq(partitions.__value_->__partition(13), v, 17, 18));
+ assert(part_eq(partitions.__value_->__partition(14), v, 18, 19));
+ assert(part_eq(partitions.__value_->__partition(15), v, 19, 20));
}
{ // 20 elements, min_size=2, 2 workers: 10 partitions
// [0..2), [2..4), [4..6), ..., [18, 20)
Container v(20);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 2, 0, 0, 2);
- assert(partitions != std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
- assert(partitions->__partitions_count() == 10);
- assert(partitions->__end() == v.end());
- assert(part_eq(partitions->__partition(0), v, 0, 2));
- assert(part_eq(partitions->__partition(1), v, 2, 4));
- assert(part_eq(partitions->__partition(2), v, 4, 6));
- assert(part_eq(partitions->__partition(3), v, 6, 8));
- assert(part_eq(partitions->__partition(4), v, 8, 10));
- assert(part_eq(partitions->__partition(5), v, 10, 12));
- assert(part_eq(partitions->__partition(6), v, 12, 14));
- assert(part_eq(partitions->__partition(7), v, 14, 16));
- assert(part_eq(partitions->__partition(8), v, 16, 18));
- assert(part_eq(partitions->__partition(9), v, 18, 20));
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 2, 0, 0, 2);
+ assert(partitions.__value_ != std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__ok);
+ assert(partitions.__value_->__partitions_count() == 10);
+ assert(partitions.__value_->__end() == v.end());
+ assert(part_eq(partitions.__value_->__partition(0), v, 0, 2));
+ assert(part_eq(partitions.__value_->__partition(1), v, 2, 4));
+ assert(part_eq(partitions.__value_->__partition(2), v, 4, 6));
+ assert(part_eq(partitions.__value_->__partition(3), v, 6, 8));
+ assert(part_eq(partitions.__value_->__partition(4), v, 8, 10));
+ assert(part_eq(partitions.__value_->__partition(5), v, 10, 12));
+ assert(part_eq(partitions.__value_->__partition(6), v, 12, 14));
+ assert(part_eq(partitions.__value_->__partition(7), v, 14, 16));
+ assert(part_eq(partitions.__value_->__partition(8), v, 16, 18));
+ assert(part_eq(partitions.__value_->__partition(9), v, 18, 20));
}
{ // 20 elements, min_size=3, 2 workers: 6 partitions
// [0..4), [4..8), [8..11), [11..14), [14..17), [17..20)
Container v(20);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 0, 0, 2);
- assert(partitions != std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
- assert(partitions->__partitions_count() == 6);
- assert(partitions->__end() == v.end());
- assert(part_eq(partitions->__partition(0), v, 0, 4));
- assert(part_eq(partitions->__partition(1), v, 4, 8));
- assert(part_eq(partitions->__partition(2), v, 8, 11));
- assert(part_eq(partitions->__partition(3), v, 11, 14));
- assert(part_eq(partitions->__partition(4), v, 14, 17));
- assert(part_eq(partitions->__partition(5), v, 17, 20));
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 3, 0, 0, 2);
+ assert(partitions.__value_ != std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__ok);
+ assert(partitions.__value_->__partitions_count() == 6);
+ assert(partitions.__value_->__end() == v.end());
+ assert(part_eq(partitions.__value_->__partition(0), v, 0, 4));
+ assert(part_eq(partitions.__value_->__partition(1), v, 4, 8));
+ assert(part_eq(partitions.__value_->__partition(2), v, 8, 11));
+ assert(part_eq(partitions.__value_->__partition(3), v, 11, 14));
+ assert(part_eq(partitions.__value_->__partition(4), v, 14, 17));
+ assert(part_eq(partitions.__value_->__partition(5), v, 17, 20));
}
{ // 20 elements, min_size=3, prefix_skip=1, 2 workers: 6 partitions
// [1..5), [5..8), [8..11), [11..14), [14..17), [17..20)
Container v(20);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 1, 0, 2);
- assert(partitions != std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
- assert(partitions->__partitions_count() == 6);
- assert(partitions->__end() == v.end());
- assert(part_eq(partitions->__partition(0), v, 1, 5));
- assert(part_eq(partitions->__partition(1), v, 5, 8));
- assert(part_eq(partitions->__partition(2), v, 8, 11));
- assert(part_eq(partitions->__partition(3), v, 11, 14));
- assert(part_eq(partitions->__partition(4), v, 14, 17));
- assert(part_eq(partitions->__partition(5), v, 17, 20));
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 3, 1, 0, 2);
+ assert(partitions.__value_ != std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__ok);
+ assert(partitions.__value_->__partitions_count() == 6);
+ assert(partitions.__value_->__end() == v.end());
+ assert(part_eq(partitions.__value_->__partition(0), v, 1, 5));
+ assert(part_eq(partitions.__value_->__partition(1), v, 5, 8));
+ assert(part_eq(partitions.__value_->__partition(2), v, 8, 11));
+ assert(part_eq(partitions.__value_->__partition(3), v, 11, 14));
+ assert(part_eq(partitions.__value_->__partition(4), v, 14, 17));
+ assert(part_eq(partitions.__value_->__partition(5), v, 17, 20));
}
{ // 20 elements, min_size=3, suffix_skip=1, 2 workers: 6 partitions
// [0..4), [4..8), [8..11), [11..14), [14..17), [17..20)
Container v(20);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 0, 1, 2);
- assert(partitions != std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
- assert(partitions->__partitions_count() == 6);
- assert(partitions->__end() == std::prev(v.end(), 1));
- assert(part_eq(partitions->__partition(0), v, 0, 4));
- assert(part_eq(partitions->__partition(1), v, 4, 7));
- assert(part_eq(partitions->__partition(2), v, 7, 10));
- assert(part_eq(partitions->__partition(3), v, 10, 13));
- assert(part_eq(partitions->__partition(4), v, 13, 16));
- assert(part_eq(partitions->__partition(5), v, 16, 19));
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 3, 0, 1, 2);
+ assert(partitions.__value_ != std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__ok);
+ assert(partitions.__value_->__partitions_count() == 6);
+ assert(partitions.__value_->__end() == std::prev(v.end(), 1));
+ assert(part_eq(partitions.__value_->__partition(0), v, 0, 4));
+ assert(part_eq(partitions.__value_->__partition(1), v, 4, 7));
+ assert(part_eq(partitions.__value_->__partition(2), v, 7, 10));
+ assert(part_eq(partitions.__value_->__partition(3), v, 10, 13));
+ assert(part_eq(partitions.__value_->__partition(4), v, 13, 16));
+ assert(part_eq(partitions.__value_->__partition(5), v, 16, 19));
}
{ // 20 elements, min_size=3, prefix_skip=1, suffix_skip=1, 2 workers: 6 partitions
// [0..4), [4..8), [8..11), [11..14), [14..17), [17..20)
Container v(20);
- std::__pstl::__statpart::__partitioning_result result;
- const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), result, 3, 1, 1, 2);
- assert(partitions != std::nullopt);
- assert(result == std::__pstl::__statpart::__partitioning_result::__ok);
- assert(partitions->__partitions_count() == 6);
- assert(partitions->__end() == std::prev(v.end(), 1));
- assert(part_eq(partitions->__partition(0), v, 1, 4));
- assert(part_eq(partitions->__partition(1), v, 4, 7));
- assert(part_eq(partitions->__partition(2), v, 7, 10));
- assert(part_eq(partitions->__partition(3), v, 10, 13));
- assert(part_eq(partitions->__partition(4), v, 13, 16));
- assert(part_eq(partitions->__partition(5), v, 16, 19));
+ const auto partitions = std::__pstl::__statpart::__partition(v.begin(), v.end(), 3, 1, 1, 2);
+ assert(partitions.__value_ != std::nullopt);
+ assert(partitions.__status_ == std::__pstl::__statpart::__partitioning_status::__ok);
+ assert(partitions.__value_->__partitions_count() == 6);
+ assert(partitions.__value_->__end() == std::prev(v.end(), 1));
+ assert(part_eq(partitions.__value_->__partition(0), v, 1, 4));
+ assert(part_eq(partitions.__value_->__partition(1), v, 4, 7));
+ assert(part_eq(partitions.__value_->__partition(2), v, 7, 10));
+ assert(part_eq(partitions.__value_->__partition(3), v, 10, 13));
+ assert(part_eq(partitions.__value_->__partition(4), v, 13, 16));
+ assert(part_eq(partitions.__value_->__partition(5), v, 16, 19));
}
}
>From 7ed5cb6b3dbab77dcd119404349f668033697241 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Thu, 15 Oct 2026 18:22:59 +0100
Subject: [PATCH 21/24] Brought a custom implementation of any_of() back
---
libcxx/include/CMakeLists.txt | 3 +-
libcxx/include/__pstl/backends/statpart.h | 1 +
.../include/__pstl/backends/statpart/any_of.h | 109 ++++++++++++++++++
.../include/__pstl/backends/statpart/basis.h | 12 +-
libcxx/include/module.modulemap.in | 3 +
5 files changed, 121 insertions(+), 7 deletions(-)
create mode 100644 libcxx/include/__pstl/backends/statpart/any_of.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index 91081893079c1b..3a436816468991 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -695,7 +695,8 @@ set(files
__pstl/backends/default.h
__pstl/backends/libdispatch.h
__pstl/backends/serial.h
- __pstl/backends/statpart.h
+ __pstl/backends/statpart.h
+ __pstl/backends/statpart/any_of.h
__pstl/backends/statpart/basis.h
__pstl/backends/statpart/find_end.h
__pstl/backends/statpart/find_if.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index a77671588874ff..a3a9427fb96e7e 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -9,6 +9,7 @@
#ifndef _LIBCPP___PSTL_BACKENDS_STATPART_H
#define _LIBCPP___PSTL_BACKENDS_STATPART_H
+#include <__pstl/backends/statpart/any_of.h>
#include <__pstl/backends/statpart/basis.h>
#include <__pstl/backends/statpart/find_end.h>
#include <__pstl/backends/statpart/find_if.h>
diff --git a/libcxx/include/__pstl/backends/statpart/any_of.h b/libcxx/include/__pstl/backends/statpart/any_of.h
new file mode 100644
index 00000000000000..5b966e6a651379
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/any_of.h
@@ -0,0 +1,109 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_BACKENDS_STATPART_ANY_OF_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_ANY_OF_H
+
+#include <__algorithm/any_of.h>
+#include <__assert>
+#include <__atomic/atomic.h>
+#include <__atomic/memory_order.h>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__optional/nullopt_t.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/move.h>
+#include <__utility/pair.h>
+#include <cstdint>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
+namespace __pstl {
+
+template <class _Index, class _Brick>
+_LIBCPP_HIDE_FROM_ABI optional<bool> __parallel_or(_Index __first, _Index __last, _Brick __f) {
+ std::atomic<bool> __found(false);
+ auto __ret = __statpart::__for_each(__first, __last, [__f, &__found](_Index __i, _Index __j) {
+ if (!__found.load(std::memory_order_relaxed) && __f(__i, __j)) {
+ __found.store(true, std::memory_order_relaxed);
+ }
+ });
+ if (!__ret)
+ return nullopt;
+ return static_cast<bool>(__found);
+}
+
+// TODO: check whether __simd_first() can be used here
+template <class _Index, class _DifferenceType, class _Pred>
+_LIBCPP_HIDE_FROM_ABI bool __simd_or(_Index __first, _DifferenceType __n, _Pred __pred) noexcept {
+ _DifferenceType __block_size = 4 < __n ? 4 : __n;
+ const _Index __last = __first + __n;
+ while (__last != __first) {
+ int32_t __flag = 1;
+ _PSTL_PRAGMA_SIMD_REDUCTION(& : __flag)
+ for (_DifferenceType __i = 0; __i < __block_size; ++__i)
+ if (__pred(*(__first + __i)))
+ __flag = 0;
+ if (!__flag)
+ return true;
+
+ __first += __block_size;
+ if (__last - __first >= __block_size << 1) {
+ // Double the block _Size. Any unnecessary iterations can be amortized against work done so far.
+ __block_size <<= 1;
+ } else {
+ __block_size = __last - __first;
+ }
+ }
+ return false;
+}
+
+template <class _ExecutionPolicy>
+struct __any_of<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Predicate>
+ _LIBCPP_HIDE_FROM_ABI optional<bool>
+ operator()(_Policy&& __policy, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __pstl::__parallel_or(
+ __first, __last, [&__policy, &__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ using _AnyOfUnseq = __pstl::__any_of<__statpart_backend_tag, __remove_parallel_policy_t<_ExecutionPolicy>>;
+ auto __res = _AnyOfUnseq()(std::__remove_parallel_policy(__policy), __brick_first, __brick_last, __pred);
+ _LIBCPP_ASSERT_INTERNAL(__res, "unseq/seq should never try to allocate!");
+ return *std::move(__res);
+ });
+ } else if constexpr (__is_unsequenced_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __pstl::__simd_or(__first, __last - __first, __pred);
+ } else {
+ return std::any_of(__first, __last, __pred);
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_EXPLICIT_ABI_ANNOTATIONS
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+_LIBCPP_POP_MACROS
+
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_ANY_OF_H
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index 2efe82f7c07a36..a59e9f18e4cb28 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -129,16 +129,16 @@ struct _LIBCPP_HIDE_FROM_ABI __partitioning_result {
// before partitioning.
// __workers_count is normally implicitly queried from hardware_concurrency() and should be explicitly specified only in
// unit tests.
-template <class _ForwardIterator>
-_LIBCPP_HIDE_FROM_ABI __partitioning_result<_ForwardIterator>
-__partition(_ForwardIterator __first,
- _ForwardIterator __last,
+template <class _Iterator>
+_LIBCPP_HIDE_FROM_ABI __partitioning_result<_Iterator>
+__partition(_Iterator __first,
+ _Iterator __last,
size_t __min_partition_size = 1,
size_t __prefix_to_skip = 0,
size_t __suffix_to_skip = 0,
size_t __workers_count = thread::hardware_concurrency()) noexcept {
// Forward iterators will be supported later.
- static_assert(__has_random_access_iterator_category_or_concept<_ForwardIterator>::value);
+ static_assert(__has_random_access_iterator_category_or_concept<_Iterator>::value);
_LIBCPP_ASSERT_INTERNAL(__min_partition_size >= 1, "minimum partition size must be at least 1 element");
@@ -161,7 +161,7 @@ __partition(_ForwardIterator __first,
return {nullopt, __partitioning_status::__not_enough_elements};
}
- __iterator_partitions<_ForwardIterator> __p;
+ __iterator_partitions<_Iterator> __p;
__p.__base_ = __first + __prefix_to_skip;
__p.__count_ = __effective_count;
__p.__partition_size_ = __effective_count / __partition_count;
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 4f63030a73cf32..19e2298c504e4b 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2422,6 +2422,9 @@ module std {
export std_core.utility_core.empty
}
module statpart {
+ module any_of {
+ header "__pstl/backends/statpart/any_of.h"
+ }
module basis {
header "__pstl/backends/statpart/basis.h"
}
>From 88f1fc9008fb1ec84d57ada72b99af3bf40d6276 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Fri, 16 Oct 2026 12:47:18 +0100
Subject: [PATCH 22/24] Brought back fill() as well
---
libcxx/include/CMakeLists.txt | 1 +
libcxx/include/__pstl/backends/statpart.h | 1 +
.../include/__pstl/backends/statpart/fill.h | 70 +++++++++++++++++++
libcxx/include/module.modulemap.in | 3 +
4 files changed, 75 insertions(+)
create mode 100644 libcxx/include/__pstl/backends/statpart/fill.h
diff --git a/libcxx/include/CMakeLists.txt b/libcxx/include/CMakeLists.txt
index 3a436816468991..ab985fb3579932 100644
--- a/libcxx/include/CMakeLists.txt
+++ b/libcxx/include/CMakeLists.txt
@@ -698,6 +698,7 @@ set(files
__pstl/backends/statpart.h
__pstl/backends/statpart/any_of.h
__pstl/backends/statpart/basis.h
+ __pstl/backends/statpart/fill.h
__pstl/backends/statpart/find_end.h
__pstl/backends/statpart/find_if.h
__pstl/backends/statpart/for_each.h
diff --git a/libcxx/include/__pstl/backends/statpart.h b/libcxx/include/__pstl/backends/statpart.h
index a3a9427fb96e7e..82665e66557ec1 100644
--- a/libcxx/include/__pstl/backends/statpart.h
+++ b/libcxx/include/__pstl/backends/statpart.h
@@ -11,6 +11,7 @@
#include <__pstl/backends/statpart/any_of.h>
#include <__pstl/backends/statpart/basis.h>
+#include <__pstl/backends/statpart/fill.h>
#include <__pstl/backends/statpart/find_end.h>
#include <__pstl/backends/statpart/find_if.h>
#include <__pstl/backends/statpart/for_each.h>
diff --git a/libcxx/include/__pstl/backends/statpart/fill.h b/libcxx/include/__pstl/backends/statpart/fill.h
new file mode 100644
index 00000000000000..1195505fcb00f0
--- /dev/null
+++ b/libcxx/include/__pstl/backends/statpart/fill.h
@@ -0,0 +1,70 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___PSTL_BACKENDS_STATPART_FILL_H
+#define _LIBCPP___PSTL_BACKENDS_STATPART_FILL_H
+
+#include <__algorithm/fill.h>
+#include <__assert>
+#include <__config>
+#include <__iterator/concepts.h>
+#include <__optional/optional.h>
+#include <__pstl/backend_fwd.h>
+#include <__pstl/backends/statpart/basis.h>
+#include <__type_traits/is_execution_policy.h>
+#include <__utility/empty.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+# pragma GCC system_header
+#endif
+
+#if _LIBCPP_STD_VER >= 17
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+namespace __pstl {
+
+template <class _Index, class _DifferenceType, class _Tp>
+_LIBCPP_HIDE_FROM_ABI _Index __simd_fill_n(_Index __first, _DifferenceType __n, const _Tp& __value) noexcept {
+ _PSTL_USE_NONTEMPORAL_STORES_IF_ALLOWED
+ _PSTL_PRAGMA_SIMD
+ for (_DifferenceType __i = 0; __i < __n; ++__i)
+ __first[__i] = __value;
+ return __first + __n;
+}
+
+template <class _ExecutionPolicy>
+struct __fill<__statpart_backend_tag, _ExecutionPolicy> {
+ template <class _Policy, class _ForwardIterator, class _Tp>
+ _LIBCPP_HIDE_FROM_ABI optional<__empty>
+ operator()(_Policy&& __policy, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value) const noexcept {
+ if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ return __statpart::__for_each(
+ __first, __last, [&__policy, &__value](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ using _FillUnseq = __pstl::__fill<__statpart_backend_tag, __remove_parallel_policy_t<_ExecutionPolicy>>;
+ [[maybe_unused]] auto __res =
+ _FillUnseq()(std::__remove_parallel_policy(__policy), __brick_first, __brick_last, __value);
+ _LIBCPP_ASSERT_INTERNAL(__res, "unseq/seq should never try to allocate!");
+ });
+ } else if constexpr (__is_unsequenced_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ __pstl::__simd_fill_n(__first, __last - __first, __value);
+ return __empty{};
+ } else {
+ std::fill(__first, __last, __value);
+ return __empty{};
+ }
+ }
+};
+
+} // namespace __pstl
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP_STD_VER >= 17
+
+#endif // _LIBCPP___PSTL_BACKENDS_STATPART_FILL_H
diff --git a/libcxx/include/module.modulemap.in b/libcxx/include/module.modulemap.in
index 19e2298c504e4b..0998d26d205f66 100644
--- a/libcxx/include/module.modulemap.in
+++ b/libcxx/include/module.modulemap.in
@@ -2428,6 +2428,9 @@ module std {
module basis {
header "__pstl/backends/statpart/basis.h"
}
+ module fill {
+ header "__pstl/backends/statpart/fill.h"
+ }
module find_end {
header "__pstl/backends/statpart/find_end.h"
}
>From 25e852e271fe86cd7525a955a284dc8616fc5a87 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Fri, 16 Oct 2026 13:03:03 +0100
Subject: [PATCH 23/24] Brought back the unseq codepath in find_if()
---
.../include/__pstl/backends/statpart/basis.h | 2 +
.../__pstl/backends/statpart/find_if.h | 53 +++++++++++++++++--
2 files changed, 52 insertions(+), 3 deletions(-)
diff --git a/libcxx/include/__pstl/backends/statpart/basis.h b/libcxx/include/__pstl/backends/statpart/basis.h
index a59e9f18e4cb28..349d16ae57d27c 100644
--- a/libcxx/include/__pstl/backends/statpart/basis.h
+++ b/libcxx/include/__pstl/backends/statpart/basis.h
@@ -64,6 +64,8 @@ constexpr inline size_t __max_partitions_per_cpu = 8;
// The minimum number of elements to sort serially in a single partition
constexpr inline size_t __min_sort_partition_size = 256;
+constexpr inline size_t __lane_size = 64;
+
enum class __partitioning_status {
// Returned when the partitioning result is not a nullopt.
__ok,
diff --git a/libcxx/include/__pstl/backends/statpart/find_if.h b/libcxx/include/__pstl/backends/statpart/find_if.h
index 0913e91a18e3f2..8130d99d810c69 100644
--- a/libcxx/include/__pstl/backends/statpart/find_if.h
+++ b/libcxx/include/__pstl/backends/statpart/find_if.h
@@ -35,21 +35,68 @@ _LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
+template <class _Index, class _DifferenceType, class _Compare>
+_LIBCPP_HIDE_FROM_ABI _Index
+__simd_first(_Index __first, _DifferenceType __begin, _DifferenceType __end, _Compare __comp) noexcept {
+ // Experiments show good block sizes like this
+ const _DifferenceType __block_size = 8;
+ alignas(__statpart::__lane_size) _DifferenceType __lane[__block_size] = {0};
+ while (__end - __begin >= __block_size) {
+ _DifferenceType __found = 0;
+ _PSTL_PRAGMA_SIMD_REDUCTION(| : __found) for (_DifferenceType __i = __begin; __i < __begin + __block_size; ++__i) {
+ const _DifferenceType __t = __comp(__first, __i);
+ __lane[__i - __begin] = __t;
+ __found |= __t;
+ }
+ if (__found) {
+ _DifferenceType __i;
+ // This will vectorize
+ for (__i = 0; __i < __block_size; ++__i) {
+ if (__lane[__i]) {
+ break;
+ }
+ }
+ return __first + __begin + __i;
+ }
+ __begin += __block_size;
+ }
+
+ // Keep remainder scalar
+ while (__begin != __end) {
+ if (__comp(__first, __begin)) {
+ return __first + __begin;
+ }
+ ++__begin;
+ }
+ return __first + __end;
+}
+
template <class _ExecutionPolicy>
struct __find_if<__statpart_backend_tag, _ExecutionPolicy> {
template <class _Policy, class _ForwardIterator, class _Predicate>
_LIBCPP_HIDE_FROM_ABI optional<_ForwardIterator>
- operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred) const noexcept {
+ operator()(_Policy&& __policy, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred) const noexcept {
if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
__has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
return __statpart::__find(
std::move(__first),
std::move(__last),
- [&__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
- return std::find_if(__brick_first, __brick_last, __pred);
+ [&__policy, &__pred](_ForwardIterator __brick_first, _ForwardIterator __brick_last) {
+ using _FindIfUnseq =
+ __pstl::__find_if<__statpart_backend_tag, __remove_parallel_policy_t<_ExecutionPolicy>>;
+ auto __res = _FindIfUnseq()(std::__remove_parallel_policy(__policy), __brick_first, __brick_last, __pred);
+ _LIBCPP_ASSERT_INTERNAL(__res, "unseq/seq should never try to allocate!");
+ return *std::move(__res);
},
less<>{},
true);
+ } else if constexpr (__is_unsequenced_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ using __diff_t = __iterator_difference_type<_ForwardIterator>;
+ return __pstl::__simd_first(
+ __first, __diff_t(0), __last - __first, [&__pred](_ForwardIterator __iter, __diff_t __i) {
+ return __pred(__iter[__i]);
+ });
} else {
return std::find_if(std::move(__first), std::move(__last), std::move(__pred));
}
>From 5977033c8d2ed76aa9644f966e75dc5cd7e5d457 Mon Sep 17 00:00:00 2001
From: Michael Kazakov <mike.kazakov at gmail.com>
Date: Fri, 16 Oct 2026 13:14:34 +0100
Subject: [PATCH 24/24] Brought back unseq in for_each()
---
.../__pstl/backends/statpart/for_each.h | 21 +++++++++++++++++--
1 file changed, 19 insertions(+), 2 deletions(-)
diff --git a/libcxx/include/__pstl/backends/statpart/for_each.h b/libcxx/include/__pstl/backends/statpart/for_each.h
index 53aeef91fad574..7fecc50a9cce12 100644
--- a/libcxx/include/__pstl/backends/statpart/for_each.h
+++ b/libcxx/include/__pstl/backends/statpart/for_each.h
@@ -34,17 +34,34 @@ _LIBCPP_BEGIN_NAMESPACE_STD
_LIBCPP_BEGIN_EXPLICIT_ABI_ANNOTATIONS
namespace __pstl {
+template <class _Iterator, class _DifferenceType, class _Function>
+_LIBCPP_HIDE_FROM_ABI _Iterator __simd_for_each(_Iterator __first, _DifferenceType __n, _Function __f) noexcept {
+ _PSTL_PRAGMA_SIMD
+ for (_DifferenceType __i = 0; __i < __n; ++__i)
+ __f(__first[__i]);
+
+ return __first + __n;
+}
+
template <class _ExecutionPolicy>
struct __for_each<__statpart_backend_tag, _ExecutionPolicy> {
template <class _Policy, class _ForwardIterator, class _Function>
_LIBCPP_HIDE_FROM_ABI optional<__empty>
- operator()(_Policy&&, _ForwardIterator __first, _ForwardIterator __last, _Function __func) const noexcept {
+ operator()(_Policy&& __policy, _ForwardIterator __first, _ForwardIterator __last, _Function __func) const noexcept {
if constexpr (__is_parallel_execution_policy_v<_ExecutionPolicy> &&
__has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
return __statpart::__for_each(
std::move(__first), std::move(__last), [&](auto __partition_first, auto __partition_last) {
- std::for_each(__partition_first, __partition_last, __func);
+ using _ForEachUnseq =
+ __pstl::__for_each<__statpart_backend_tag, __remove_parallel_policy_t<_ExecutionPolicy>>;
+ [[maybe_unused]] auto __res =
+ _ForEachUnseq()(std::__remove_parallel_policy(__policy), __partition_first, __partition_last, __func);
+ _LIBCPP_ASSERT_INTERNAL(__res, "unseq/seq should never try to allocate!");
});
+ } else if constexpr (__is_unsequenced_execution_policy_v<_ExecutionPolicy> &&
+ __has_random_access_iterator_category_or_concept<_ForwardIterator>::value) {
+ __pstl::__simd_for_each(__first, __last - __first, __func);
+ return __empty{};
} else {
std::for_each(std::move(__first), std::move(__last), std::move(__func));
return __empty{};
More information about the libcxx-commits
mailing list