@zigc/lib 0.17.0-dev.27 → 0.17.0-dev.296

package/libcxx/include/__algorithm/fill_n.h

@@ -9,11 +9,14 @@
 #ifndef _LIBCPP___ALGORITHM_FILL_N_H
 #define _LIBCPP___ALGORITHM_FILL_N_H
 
-#include <__algorithm/min.h>
+#include <__algorithm/for_each_n_segment.h>
+#include <__algorithm/specialized_algorithms.h>
 #include <__config>
-#include <__fwd/bit_reference.h>
-#include <__memory/pointer_traits.h>
+#include <__iterator/iterator_traits.h>
+#include <__iterator/segmented_iterator.h>
+#include <__type_traits/enable_if.h>
 #include <__utility/convert_to_integral.h>
+#include <__utility/move.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
@@ -26,56 +29,39 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 
 // fill_n isn't specialized for std::memset, because the compiler already optimizes the loop to a call to std::memset.
 
-template <class _OutputIterator, class _Size, class _Tp>
-inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _OutputIterator
-__fill_n(_OutputIterator __first, _Size __n, const _Tp& __value);
-
-template <bool _FillVal, class _Cp>
-_LIBCPP_CONSTEXPR_SINCE_CXX20 _LIBCPP_HIDE_FROM_ABI void
-__fill_n_bool(__bit_iterator<_Cp, false> __first, typename __size_difference_type_traits<_Cp>::size_type __n) {
-  using _It            = __bit_iterator<_Cp, false>;
-  using __storage_type = typename _It::__storage_type;
-
-  const int __bits_per_word = _It::__bits_per_word;
-  // do first partial word
-  if (__first.__ctz_ != 0) {
-    __storage_type __clz_f = static_cast<__storage_type>(__bits_per_word - __first.__ctz_);
-    __storage_type __dn    = std::min(__clz_f, __n);
-    std::__fill_masked_range(std::__to_address(__first.__seg_), __clz_f - __dn, __first.__ctz_, _FillVal);
-    __n -= __dn;
-    ++__first.__seg_;
-  }
-  // do middle whole words
-  __storage_type __nw = __n / __bits_per_word;
-  std::__fill_n(std::__to_address(__first.__seg_), __nw, _FillVal ? static_cast<__storage_type>(-1) : 0);
-  __n -= __nw * __bits_per_word;
-  // do last partial word
-  if (__n > 0) {
-    __first.__seg_ += __nw;
-    std::__fill_masked_range(std::__to_address(__first.__seg_), __bits_per_word - __n, 0u, _FillVal);
-  }
-}
-
-template <class _Cp, class _Size>
-inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 __bit_iterator<_Cp, false>
-__fill_n(__bit_iterator<_Cp, false> __first, _Size __n, const bool& __value) {
-  if (__n > 0) {
-    if (__value)
-      std::__fill_n_bool<true>(__first, __n);
-    else
-      std::__fill_n_bool<false>(__first, __n);
-  }
-  return __first + __n;
-}
-
-template <class _OutputIterator, class _Size, class _Tp>
+template <
+    class _OutputIterator,
+    class _Size,
+    class _Tp,
+    __enable_if_t<!__specialized_algorithm<_Algorithm::__fill_n, __single_iterator<_OutputIterator> >::__has_algorithm,
+                  int> = 0>
 inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _OutputIterator
 __fill_n(_OutputIterator __first, _Size __n, const _Tp& __value) {
+#ifndef _LIBCPP_CXX03_LANG
+  if constexpr (__is_segmented_iterator_v<_OutputIterator>) {
+    using __local_iterator = typename __segmented_iterator_traits<_OutputIterator>::__local_iterator;
+    if constexpr (__has_random_access_iterator_category<__local_iterator>::value) {
+      return std::__for_each_n_segment(__first, __n, [&](__local_iterator __lfirst, __local_iterator __llast) {
+        std::__fill_n(__lfirst, __llast - __lfirst, __value);
+      });
+    }
+  }
+#endif
   for (; __n > 0; ++__first, (void)--__n)
     *__first = __value;
   return __first;
 }
 
+template <class _OutIter,
+          class _Size,
+          class _Tp,
+          __enable_if_t<__specialized_algorithm<_Algorithm::__fill_n, __single_iterator<_OutIter> >::__has_algorithm,
+                        int> = 0>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _OutIter __fill_n(_OutIter __first, _Size __n, const _Tp& __value) {
+  return __specialized_algorithm<_Algorithm::__fill_n, __single_iterator<_OutIter> >()(
+      std::move(__first), __n, __value);
+}
+
 template <class _OutputIterator, class _Size, class _Tp>
 inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _OutputIterator
 fill_n(_OutputIterator __first, _Size __n, const _Tp& __value) {

package/libcxx/include/__algorithm/find.h

@@ -12,16 +12,19 @@
 
 #include <__algorithm/find_segment_if.h>
 #include <__algorithm/min.h>
+#include <__algorithm/simd_utils.h>
 #include <__algorithm/unwrap_iter.h>
 #include <__bit/countr.h>
 #include <__bit/invert_if.h>
 #include <__config>
+#include <__cstddef/size_t.h>
 #include <__functional/identity.h>
 #include <__fwd/bit_reference.h>
 #include <__iterator/segmented_iterator.h>
 #include <__string/constexpr_c_functions.h>
 #include <__type_traits/enable_if.h>
 #include <__type_traits/invoke.h>
+#include <__type_traits/is_constant_evaluated.h>
 #include <__type_traits/is_equality_comparable.h>
 #include <__type_traits/is_integral.h>
 #include <__type_traits/is_signed.h>
@@ -44,46 +47,108 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 // generic implementation
 template <class _Iter, class _Sent, class _Tp, class _Proj>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Iter
-__find(_Iter __first, _Sent __last, const _Tp& __value, _Proj& __proj) {
+__find_loop(_Iter __first, _Sent __last, const _Tp& __value, _Proj& __proj) {
   for (; __first != __last; ++__first)
     if (std::__invoke(__proj, *__first) == __value)
       break;
   return __first;
 }
 
-// trivially equality comparable implementations
-template <class _Tp,
-          class _Up,
-          class _Proj,
-          __enable_if_t<__is_identity<_Proj>::value && __libcpp_is_trivially_equality_comparable<_Tp, _Up>::value &&
-                            sizeof(_Tp) == 1,
-                        int> = 0>
-_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp* __find(_Tp* __first, _Tp* __last, const _Up& __value, _Proj&) {
-  if (auto __ret = std::__constexpr_memchr(__first, __value, __last - __first))
-    return __ret;
-  return __last;
+template <class _Iter, class _Sent, class _Tp, class _Proj>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Iter
+__find(_Iter __first, _Sent __last, const _Tp& __value, _Proj& __proj) {
+  return std::__find_loop(std::move(__first), std::move(__last), __value, __proj);
 }
 
-#if _LIBCPP_HAS_WIDE_CHARACTERS
+#if _LIBCPP_VECTORIZE_ALGORITHMS
+template <class _Tp, class _Up>
+[[__nodiscard__]] _LIBCPP_HIDE_FROM_ABI
+_LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp* __find_vectorized(_Tp* __first, _Tp* __last, _Up __value) {
+  if (!__libcpp_is_constant_evaluated()) {
+    constexpr size_t __unroll_count = 4;
+    constexpr size_t __vec_size     = __native_vector_size<_Tp>;
+    using __vec                     = __simd_vector<_Tp, __vec_size>;
+
+    auto __orig_first = __first;
+
+    auto __values = static_cast<__simd_vector<_Tp, __vec_size>>(__value); // broadcast the value
+    while (static_cast<size_t>(__last - __first) >= __unroll_count * __vec_size) [[__unlikely__]] {
+      __vec __lhs[__unroll_count];
+
+      for (size_t __i = 0; __i != __unroll_count; ++__i)
+        __lhs[__i] = std::__load_vector<__vec>(__first + __i * __vec_size);
+
+      for (size_t __i = 0; __i != __unroll_count; ++__i) {
+        if (auto __cmp_res = __lhs[__i] == __values; std::__any_of(__cmp_res)) {
+          auto __offset = __i * __vec_size + std::__find_first_set(__cmp_res);
+          return __first + __offset;
+        }
+      }
+
+      __first += __unroll_count * __vec_size;
+    }
+
+    // check the remaining 0-3 vectors
+    while (static_cast<size_t>(__last - __first) >= __vec_size) {
+      if (auto __cmp_res = std::__load_vector<__vec>(__first) == __values; std::__any_of(__cmp_res)) {
+        return __first + std::__find_first_set(__cmp_res);
+      }
+      __first += __vec_size;
+    }
+
+    if (__last - __first == 0)
+      return __first;
+
+    // Check if we can load elements in front of the current pointer. If that's the case load a vector at
+    // (last - vector_size) to check the remaining elements
+    if (static_cast<size_t>(__first - __orig_first) >= __vec_size) {
+      __first = __last - __vec_size;
+      return __first + std::__find_first_set(std::__load_vector<__vec>(__first) == __values);
+    }
+  }
+
+  __identity __proj;
+  return std::__find_loop(__first, __last, __value, __proj);
+}
+#endif
+
+#ifndef _LIBCPP_CXX03_LANG
+// trivially equality comparable implementations
 template <class _Tp,
           class _Up,
           class _Proj,
-          __enable_if_t<__is_identity<_Proj>::value && __libcpp_is_trivially_equality_comparable<_Tp, _Up>::value &&
-                            sizeof(_Tp) == sizeof(wchar_t) && _LIBCPP_ALIGNOF(_Tp) >= _LIBCPP_ALIGNOF(wchar_t),
-                        int> = 0>
+          __enable_if_t<__is_identity<_Proj>::value && __is_trivially_equality_comparable_v<_Tp, _Up>, int> = 0>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp* __find(_Tp* __first, _Tp* __last, const _Up& __value, _Proj&) {
-  if (auto __ret = std::__constexpr_wmemchr(__first, __value, __last - __first))
-    return __ret;
-  return __last;
+  if constexpr (sizeof(_Tp) == 1) {
+    if (auto __ret = std::__constexpr_memchr(__first, __value, __last - __first))
+      return __ret;
+    return __last;
+  }
+#  if _LIBCPP_HAS_WIDE_CHARACTERS
+  else if constexpr (sizeof(_Tp) == sizeof(wchar_t) && _LIBCPP_ALIGNOF(_Tp) >= _LIBCPP_ALIGNOF(wchar_t)) {
+    if (auto __ret = std::__constexpr_wmemchr(__first, __value, __last - __first))
+      return __ret;
+    return __last;
+  }
+#  endif
+#  if _LIBCPP_VECTORIZE_ALGORITHMS
+  else if constexpr (is_integral<_Tp>::value) {
+    return std::__find_vectorized(__first, __last, __value);
+  }
+#  endif
+  else {
+    __identity __proj;
+    return std::__find_loop(__first, __last, __value, __proj);
+  }
 }
-#endif // _LIBCPP_HAS_WIDE_CHARACTERS
+#endif
 
 // TODO: This should also be possible to get right with different signedness
 // cast integral types to allow vectorization
 template <class _Tp,
           class _Up,
           class _Proj,
-          __enable_if_t<__is_identity<_Proj>::value && !__libcpp_is_trivially_equality_comparable<_Tp, _Up>::value &&
+          __enable_if_t<__is_identity<_Proj>::value && !__is_trivially_equality_comparable_v<_Tp, _Up> &&
                             is_integral<_Tp>::value && is_integral<_Up>::value &&
                             is_signed<_Tp>::value == is_signed<_Up>::value,
                         int> = 0>
@@ -143,31 +208,23 @@ __find(__bit_iterator<_Cp, _IsConst> __first, __bit_iterator<_Cp, _IsConst> __la
 
 // segmented iterator implementation
 
-template <class>
-struct __find_segment;
-
 template <class _SegmentedIterator,
           class _Tp,
           class _Proj,
-          __enable_if_t<__is_segmented_iterator<_SegmentedIterator>::value, int> = 0>
+          __enable_if_t<__is_segmented_iterator_v<_SegmentedIterator>, int> = 0>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _SegmentedIterator
 __find(_SegmentedIterator __first, _SegmentedIterator __last, const _Tp& __value, _Proj& __proj) {
-  return std::__find_segment_if(std::move(__first), std::move(__last), __find_segment<_Tp>(__value), __proj);
+  using __local_iterator = typename __segmented_iterator_traits<_SegmentedIterator>::__local_iterator;
+  return std::__find_segment_if(
+      std::move(__first),
+      std::move(__last),
+      [&__value](__local_iterator __lfirst, __local_iterator __llast, _Proj& __lproj) {
+        return std::__rewrap_iter(
+            __lfirst, std::__find(std::__unwrap_iter(__lfirst), std::__unwrap_iter(__llast), __value, __lproj));
+      },
+      __proj);
 }
 
-template <class _Tp>
-struct __find_segment {
-  const _Tp& __value_;
-
-  _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR __find_segment(const _Tp& __value) : __value_(__value) {}
-
-  template <class _InputIterator, class _Proj>
-  _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR _InputIterator
-  operator()(_InputIterator __first, _InputIterator __last, _Proj& __proj) const {
-    return std::__find(__first, __last, __value_, __proj);
-  }
-};
-
 // public API
 template <class _InputIterator, class _Tp>
 [[__nodiscard__]] inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _InputIterator

package/libcxx/include/__algorithm/find_end.h

@@ -76,6 +76,111 @@ _LIBCPP_HIDE_FROM_ABI inline _LIBCPP_CONSTEXPR_SINCE_CXX14 pair<_Iter1, _Iter1>
   }
 }
 
+template <class _AlgPolicy,
+          class _Pred,
+          class _Iter1,
+          class _Sent1,
+          class _Iter2,
+          class _Sent2,
+          class _Proj1,
+          class _Proj2>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 pair<_Iter1, _Iter1> __find_end_impl(
+    _Iter1 __first1,
+    _Sent1 __sent1,
+    _Iter2 __first2,
+    _Sent2 __sent2,
+    _Pred& __pred,
+    _Proj1& __proj1,
+    _Proj2& __proj2,
+    bidirectional_iterator_tag,
+    bidirectional_iterator_tag) {
+  auto __last1 = _IterOps<_AlgPolicy>::next(__first1, __sent1);
+  auto __last2 = _IterOps<_AlgPolicy>::next(__first2, __sent2);
+  // modeled after search algorithm (in reverse)
+  if (__first2 == __last2)
+    return std::make_pair(__last1, __last1); // Everything matches an empty sequence
+  _Iter1 __l1 = __last1;
+  _Iter2 __l2 = __last2;
+  --__l2;
+  while (true) {
+    // Find last element in sequence 1 that matches *(__last2-1), with a mininum of loop checks
+    while (true) {
+      if (__first1 == __l1) // return __last1 if no element matches *__first2
+        return std::make_pair(__last1, __last1);
+      if (std::__invoke(__pred, std::__invoke(__proj1, *--__l1), std::__invoke(__proj2, *__l2)))
+        break;
+    }
+    // *__l1 matches *__l2, now match elements before here
+    _Iter1 __match_last = __l1;
+    _Iter1 __m1         = __l1;
+    _Iter2 __m2         = __l2;
+    while (true) {
+      if (__m2 == __first2) // If pattern exhausted, __m1 is the answer (works for 1 element pattern)
+        return std::make_pair(__m1, ++__match_last);
+      if (__m1 == __first1) // Otherwise if source exhaused, pattern not found
+        return std::make_pair(__last1, __last1);
+
+      // if there is a mismatch, restart with a new __l1
+      if (!std::__invoke(__pred, std::__invoke(__proj1, *--__m1), std::__invoke(__proj2, *--__m2))) {
+        break;
+      } // else there is a match, check next elements
+    }
+  }
+}
+
+template <class _AlgPolicy,
+          class _Pred,
+          class _Iter1,
+          class _Sent1,
+          class _Iter2,
+          class _Sent2,
+          class _Proj1,
+          class _Proj2>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 pair<_Iter1, _Iter1> __find_end_impl(
+    _Iter1 __first1,
+    _Sent1 __sent1,
+    _Iter2 __first2,
+    _Sent2 __sent2,
+    _Pred& __pred,
+    _Proj1& __proj1,
+    _Proj2& __proj2,
+    random_access_iterator_tag,
+    random_access_iterator_tag) {
+  typedef typename iterator_traits<_Iter1>::difference_type _D1;
+  auto __last1 = _IterOps<_AlgPolicy>::next(__first1, __sent1);
+  auto __last2 = _IterOps<_AlgPolicy>::next(__first2, __sent2);
+  // Take advantage of knowing source and pattern lengths.  Stop short when source is smaller than pattern
+  auto __len2 = __last2 - __first2;
+  if (__len2 == 0)
+    return std::make_pair(__last1, __last1);
+  auto __len1 = __last1 - __first1;
+  if (__len1 < __len2)
+    return std::make_pair(__last1, __last1);
+  const _Iter1 __s = __first1 + _D1(__len2 - 1); // End of pattern match can't go before here
+  _Iter1 __l1      = __last1;
+  _Iter2 __l2      = __last2;
+  --__l2;
+  while (true) {
+    while (true) {
+      if (__s == __l1)
+        return std::make_pair(__last1, __last1);
+      if (std::__invoke(__pred, std::__invoke(__proj1, *--__l1), std::__invoke(__proj2, *__l2)))
+        break;
+    }
+    _Iter1 __last_match = __l1;
+    _Iter1 __m1         = __l1;
+    _Iter2 __m2         = __l2;
+    while (true) {
+      if (__m2 == __first2)
+        return std::make_pair(__m1, ++__last_match);
+      // no need to check range on __m1 because __s guarantees we have enough source
+      if (!std::__invoke(__pred, std::__invoke(__proj1, *--__m1), std::__invoke(__proj2, *--__m2))) {
+        break;
+      }
+    }
+  }
+}
+
 template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 [[__nodiscard__]] inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _ForwardIterator1 __find_end_classic(
     _ForwardIterator1 __first1,

package/libcxx/include/__algorithm/for_each.h

@@ -11,45 +11,41 @@
 #define _LIBCPP___ALGORITHM_FOR_EACH_H
 
 #include <__algorithm/for_each_segment.h>
+#include <__algorithm/specialized_algorithms.h>
 #include <__config>
 #include <__functional/identity.h>
 #include <__iterator/segmented_iterator.h>
-#include <__type_traits/enable_if.h>
 #include <__type_traits/invoke.h>
-#include <__utility/move.h>
+#include <__type_traits/is_same.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
 #endif
 
-_LIBCPP_PUSH_MACROS
-#include <__undef_macros>
-
 _LIBCPP_BEGIN_NAMESPACE_STD
 
 template <class _InputIterator, class _Sent, class _Func, class _Proj>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _InputIterator
-__for_each(_InputIterator __first, _Sent __last, _Func& __f, _Proj& __proj) {
+__for_each(_InputIterator __first, _Sent __last, _Func& __func, _Proj& __proj) {
+#ifndef _LIBCPP_CXX03_LANG
+  if constexpr (using _SpecialAlg =
+                    __specialized_algorithm<_Algorithm::__for_each, __iterator_pair<_InputIterator, _Sent>>;
+                _SpecialAlg::__has_algorithm) {
+    _SpecialAlg()(__first, __last, __func, __proj);
+    return __last;
+  } else if constexpr (is_same<_InputIterator, _Sent>::value && __is_segmented_iterator_v<_InputIterator>) {
+    using __local_iterator_t = typename __segmented_iterator_traits<_InputIterator>::__local_iterator;
+    std::__for_each_segment(__first, __last, [&](__local_iterator_t __lfirst, __local_iterator_t __llast) {
+      std::__for_each(__lfirst, __llast, __func, __proj);
+    });
+    return __last;
+  }
+#endif
   for (; __first != __last; ++__first)
-    std::__invoke(__f, std::__invoke(__proj, *__first));
+    std::__invoke(__func, std::__invoke(__proj, *__first));
   return __first;
 }
 
-#ifndef _LIBCPP_CXX03_LANG
-template <class _SegmentedIterator,
-          class _Func,
-          class _Proj,
-          __enable_if_t<__is_segmented_iterator<_SegmentedIterator>::value, int> = 0>
-_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _SegmentedIterator
-__for_each(_SegmentedIterator __first, _SegmentedIterator __last, _Func& __func, _Proj& __proj) {
-  using __local_iterator_t = typename __segmented_iterator_traits<_SegmentedIterator>::__local_iterator;
-  std::__for_each_segment(__first, __last, [&](__local_iterator_t __lfirst, __local_iterator_t __llast) {
-    std::__for_each(__lfirst, __llast, __func, __proj);
-  });
-  return __last;
-}
-#endif // !_LIBCPP_CXX03_LANG
-
 template <class _InputIterator, class _Func>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _Func
 for_each(_InputIterator __first, _InputIterator __last, _Func __f) {
@@ -60,6 +56,4 @@ for_each(_InputIterator __first, _InputIterator __last, _Func __f) {
 
 _LIBCPP_END_NAMESPACE_STD
 
-_LIBCPP_POP_MACROS
-
 #endif // _LIBCPP___ALGORITHM_FOR_EACH_H

package/libcxx/include/__algorithm/for_each_n.h

@@ -16,10 +16,7 @@
 #include <__functional/identity.h>
 #include <__iterator/iterator_traits.h>
 #include <__iterator/segmented_iterator.h>
-#include <__type_traits/disjunction.h>
-#include <__type_traits/enable_if.h>
 #include <__type_traits/invoke.h>
-#include <__type_traits/negation.h>
 #include <__utility/convert_to_integral.h>
 #include <__utility/move.h>
 
@@ -32,57 +29,33 @@ _LIBCPP_PUSH_MACROS
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
-template <class _InputIterator,
-          class _Size,
-          class _Func,
-          class _Proj,
-          __enable_if_t<!__has_random_access_iterator_category<_InputIterator>::value &&
-                            _Or< _Not<__is_segmented_iterator<_InputIterator> >,
-                                 _Not<__has_random_access_local_iterator<_InputIterator> > >::value,
-                        int> = 0>
+template <class _InputIterator, class _Size, class _Func, class _Proj>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _InputIterator
 __for_each_n(_InputIterator __first, _Size __orig_n, _Func& __f, _Proj& __proj) {
   typedef decltype(std::__convert_to_integral(__orig_n)) _IntegralSize;
   _IntegralSize __n = __orig_n;
-  while (__n > 0) {
-    std::__invoke(__f, std::__invoke(__proj, *__first));
-    ++__first;
-    --__n;
-  }
-  return std::move(__first);
-}
-
-template <class _RandIter,
-          class _Size,
-          class _Func,
-          class _Proj,
-          __enable_if_t<__has_random_access_iterator_category<_RandIter>::value, int> = 0>
-_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _RandIter
-__for_each_n(_RandIter __first, _Size __orig_n, _Func& __f, _Proj& __proj) {
-  typename std::iterator_traits<_RandIter>::difference_type __n = __orig_n;
-  auto __last                                                   = __first + __n;
-  std::__for_each(__first, __last, __f, __proj);
-  return __last;
-}
 
 #ifndef _LIBCPP_CXX03_LANG
-template <class _SegmentedIterator,
-          class _Size,
-          class _Func,
-          class _Proj,
-          __enable_if_t<!__has_random_access_iterator_category<_SegmentedIterator>::value &&
-                            __is_segmented_iterator<_SegmentedIterator>::value &&
-                            __has_random_access_iterator_category<
-                                typename __segmented_iterator_traits<_SegmentedIterator>::__local_iterator>::value,
-                        int> = 0>
-_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _SegmentedIterator
-__for_each_n(_SegmentedIterator __first, _Size __orig_n, _Func& __f, _Proj& __proj) {
-  using __local_iterator_t = typename __segmented_iterator_traits<_SegmentedIterator>::__local_iterator;
-  return std::__for_each_n_segment(__first, __orig_n, [&](__local_iterator_t __lfirst, __local_iterator_t __llast) {
-    std::__for_each(__lfirst, __llast, __f, __proj);
-  });
+  if constexpr (__is_segmented_iterator_v<_InputIterator>) {
+    using __local_iterator = typename __segmented_iterator_traits<_InputIterator>::__local_iterator;
+    if constexpr (__has_random_access_iterator_category<__local_iterator>::value) {
+      return std::__for_each_n_segment(__first, __orig_n, [&](__local_iterator __lfirst, __local_iterator __llast) {
+        std::__for_each(__lfirst, __llast, __f, __proj);
+      });
+    } else {
+      return std::__for_each(__first, __first + __n, __f, __proj);
+    }
+  } else
+#endif
+  {
+    while (__n > 0) {
+      std::__invoke(__f, std::__invoke(__proj, *__first));
+      ++__first;
+      --__n;
+    }
+    return std::move(__first);
+  }
 }
-#endif // !_LIBCPP_CXX03_LANG
 
 #if _LIBCPP_STD_VER >= 17
 

package/libcxx/include/__algorithm/for_each_n_segment.h

@@ -27,7 +27,7 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 template <class _SegmentedIterator, class _Size, class _Functor>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _SegmentedIterator
 __for_each_n_segment(_SegmentedIterator __first, _Size __orig_n, _Functor __func) {
-  static_assert(__is_segmented_iterator<_SegmentedIterator>::value &&
+  static_assert(__is_segmented_iterator_v<_SegmentedIterator> &&
                     __has_random_access_iterator_category<
                         typename __segmented_iterator_traits<_SegmentedIterator>::__local_iterator>::value,
                 "__for_each_n_segment only works with segmented iterators with random-access local iterators");

package/libcxx/include/__algorithm/for_each_segment.h

@@ -48,6 +48,32 @@ __for_each_segment(_SegmentedIterator __first, _SegmentedIterator __last, _Funct
   __func(_Traits::__begin(__sfirst), _Traits::__local(__last));
 }
 
+template <class _SegmentedIterator, class _Functor>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 void
+__for_each_segment_backward(_SegmentedIterator __first, _SegmentedIterator __last, _Functor __func) {
+  using _Traits = __segmented_iterator_traits<_SegmentedIterator>;
+
+  auto __sfirst = _Traits::__segment(__first);
+  auto __slast  = _Traits::__segment(__last);
+
+  // We are in a single segment, so we might not be at the beginning or end
+  if (__sfirst == __slast) {
+    __func(_Traits::__local(__first), _Traits::__local(__last));
+    return;
+  }
+
+  // We have more than one segment. Iterate over the last segment, since we might not start at the end
+  __func(_Traits::__begin(__slast), _Traits::__local(__last));
+  --__slast;
+  // iterate over the segments which are guaranteed to be completely in the range
+  while (__sfirst != __slast) {
+    __func(_Traits::__begin(__slast), _Traits::__end(__slast));
+    --__slast;
+  }
+  // iterate over the first segment
+  __func(_Traits::__local(__first), _Traits::__end(__slast));
+}
+
 _LIBCPP_END_NAMESPACE_STD
 
 #endif // _LIBCPP___ALGORITHM_FOR_EACH_SEGMENT_H

package/libcxx/include/__algorithm/generate.h

@@ -9,7 +9,9 @@
 #ifndef _LIBCPP___ALGORITHM_GENERATE_H
 #define _LIBCPP___ALGORITHM_GENERATE_H
 
+#include <__algorithm/for_each.h>
 #include <__config>
+#include <__utility/forward.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
@@ -20,8 +22,8 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 template <class _ForwardIterator, class _Generator>
 inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 void
 generate(_ForwardIterator __first, _ForwardIterator __last, _Generator __gen) {
-  for (; __first != __last; ++__first)
-    *__first = __gen();
+  using __iter_ref = decltype(*__first);
+  std::for_each(__first, __last, [&](__iter_ref __element) { std::forward<__iter_ref>(__element) = __gen(); });
 }
 
 _LIBCPP_END_NAMESPACE_STD

package/libcxx/include/__algorithm/generate_n.h

@@ -9,25 +9,38 @@
 #ifndef _LIBCPP___ALGORITHM_GENERATE_N_H
 #define _LIBCPP___ALGORITHM_GENERATE_N_H
 
+#include <__algorithm/for_each_n.h>
 #include <__config>
-#include <__utility/convert_to_integral.h>
+#include <__functional/identity.h>
+#include <__utility/forward.h>
+#include <__utility/move.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
 #endif
 
+_LIBCPP_PUSH_MACROS
+#include <__undef_macros>
+
 _LIBCPP_BEGIN_NAMESPACE_STD
 
+template <class _OutputIterator, class _Size, class _Generator>
+inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _OutputIterator
+__generate_n(_OutputIterator __first, _Size __orig_n, _Generator& __gen) {
+  using __iter_ref = decltype(*__first);
+  __identity __proj;
+  auto __f = [&](__iter_ref __element) { std::forward<__iter_ref>(__element) = __gen(); };
+  return std::__for_each_n(std::move(__first), __orig_n, __f, __proj);
+}
+
 template <class _OutputIterator, class _Size, class _Generator>
 inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _OutputIterator
 generate_n(_OutputIterator __first, _Size __orig_n, _Generator __gen) {
-  typedef decltype(std::__convert_to_integral(__orig_n)) _IntegralSize;
-  _IntegralSize __n = __orig_n;
-  for (; __n > 0; ++__first, (void)--__n)
-    *__first = __gen();
-  return __first;
+  return std::__generate_n(std::move(__first), __orig_n, __gen);
 }
 
 _LIBCPP_END_NAMESPACE_STD
 
+_LIBCPP_POP_MACROS
+
 #endif // _LIBCPP___ALGORITHM_GENERATE_N_H