[libc++] Forward to memcmp in lexicographical_compare

libcxx/include/__algorithm/comp.h

@@ -11,6 +11,7 @@
 
 #include <__config>
 #include <__type_traits/desugars_to.h>
+#include <__type_traits/is_integral.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
@@ -42,7 +43,7 @@ struct __less<void, void> {
 };
 
 template <class _Tp>
-inline const bool __desugars_to_v<__less_tag, __less<>, _Tp, _Tp> = true;
+inline const bool __desugars_to_v<__totally_ordered_less_tag, __less<>, _Tp, _Tp> = is_integral<_Tp>::value;
 
 _LIBCPP_END_NAMESPACE_STD
 

libcxx/include/__algorithm/lexicographical_compare.h

@@ -11,39 +11,96 @@
 
 #include <__algorithm/comp.h>
 #include <__algorithm/comp_ref_type.h>
+#include <__algorithm/max.h>
+#include <__algorithm/mismatch.h>
+#include <__algorithm/simd_utils.h>
 #include <__config>
+#include <__functional/identity.h>
 #include <__iterator/iterator_traits.h>
+#include <__string/constexpr_c_functions.h>
+#include <__type_traits/desugars_to.h>
+#include <__type_traits/invoke.h>
+#include <__type_traits/is_equality_comparable.h>
+#include <__type_traits/is_integral.h>
+#include <__type_traits/is_volatile.h>
+#include <cstring>
+#include <cwchar>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
 #endif
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
-template <class _Compare, class _InputIterator1, class _InputIterator2>
+template <class _Iter1, class _Sent1, class _Iter2, class _Sent2, class _Proj1, class _Proj2, class _Comp>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 bool __lexicographical_compare(
-    _InputIterator1 __first1,
-    _InputIterator1 __last1,
-    _InputIterator2 __first2,
-    _InputIterator2 __last2,
-    _Compare __comp) {
-  for (; __first2 != __last2; ++__first1, (void)++__first2) {
-    if (__first1 == __last1 || __comp(*__first1, *__first2))
+    _Iter1 __first1, _Sent1 __last1, _Iter2 __first2, _Sent2 __last2, _Comp& __comp, _Proj1& __proj1, _Proj2& __proj2) {
+  while (__first2 != __last2) {
+    if (__first1 == __last1 ||
+        std::__invoke(__comp, std::__invoke(__proj1, *__first1), std::__invoke(__proj2, *__first2)))
       return true;
-    if (__comp(*__first2, *__first1))
+    if (std::__invoke(__comp, std::__invoke(__proj2, *__first2), std::__invoke(__proj1, *__first1)))
       return false;
+    ++__first1;
+    ++__first2;
   }
   return false;
 }
 
+#if _LIBCPP_VECTORIZE_ALGORITHMS
+
+template <class _Tp,
+          class _Proj1,
+          class _Proj2,
+          class _Comp,
+          __enable_if_t<__desugars_to_v<__totally_ordered_less_tag, _Comp, _Tp, _Tp> && !is_volatile<_Tp>::value &&
+                            __libcpp_is_trivially_equality_comparable<_Tp, _Tp>::value &&
+                            __is_identity<_Proj1>::value && __is_identity<_Proj2>::value,
+                        int> = 0>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 bool __lexicographical_compare(
+    _Tp* __first1, _Tp* __last1, _Tp* __first2, _Tp* __last2, _Comp& __comp, _Proj1& __proj1, _Proj2& __proj2) {
+  if constexpr (__is_trivially_lexicographically_comparable_v<_Tp, _Tp>) {
+    auto __res =
+        std::__constexpr_memcmp(__first1, __first2, __element_count(std::min(__last1 - __first1, __last2 - __first2)));
+    if (__res == 0)
+      return __last1 - __first1 < __last2 - __first2;
+    return __res < 0;
+  }
+#  ifndef _LIBCPP_HAS_NO_WIDE_CHARACTERS
+  else if constexpr (is_same<__remove_cv_t<_Tp>, wchar_t>::value) {
+    auto __res = std::__constexpr_wmemcmp(__first1, __first2, std::min(__last1 - __first1, __last2 - __first2));
+    if (__res == 0)
+      return __last1 - __first1 < __last2 - __first2;
+    return __res < 0;
+  }
+#  endif // _LIBCPP_HAS_NO_WIDE_CHARACTERS
+  else {
+    auto __res = std::mismatch(__first1, __last1, __first2, __last2);
+    if (__res.second == __last2)
+      return false;
+    if (__res.first == __last1)
+      return true;
+    return *__res.first < *__res.second;
+  }
+}
+#endif // _LIBCPP_VECTORIZE_ALGORITHMS
+
 template <class _InputIterator1, class _InputIterator2, class _Compare>
 _LIBCPP_NODISCARD inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 bool lexicographical_compare(
     _InputIterator1 __first1,
     _InputIterator1 __last1,
     _InputIterator2 __first2,
     _InputIterator2 __last2,
     _Compare __comp) {
-  return std::__lexicographical_compare<__comp_ref_type<_Compare> >(__first1, __last1, __first2, __last2, __comp);
+  __identity __proj;
+  return std::__lexicographical_compare(
+      std::__unwrap_iter(__first1),
+      std::__unwrap_iter(__last1),
+      std::__unwrap_iter(__first2),
+      std::__unwrap_iter(__last2),
+      __comp,
+      __proj,
+      __proj);
 }
 
 template <class _InputIterator1, class _InputIterator2>

libcxx/include/__algorithm/ranges_lexicographical_compare.h

@@ -9,6 +9,8 @@
 #ifndef _LIBCPP___ALGORITHM_RANGES_LEXICOGRAPHICAL_COMPARE_H
 #define _LIBCPP___ALGORITHM_RANGES_LEXICOGRAPHICAL_COMPARE_H
 
+#include <__algorithm/lexicographical_compare.h>
+#include <__algorithm/unwrap_range.h>
 #include <__config>
 #include <__functional/identity.h>
 #include <__functional/invoke.h>
@@ -34,23 +36,24 @@ namespace ranges {
 namespace __lexicographical_compare {
 struct __fn {
   template <class _Iter1, class _Sent1, class _Iter2, class _Sent2, class _Proj1, class _Proj2, class _Comp>
-  _LIBCPP_HIDE_FROM_ABI constexpr static bool __lexicographical_compare_impl(
+  static _LIBCPP_HIDE_FROM_ABI constexpr bool __lexicographical_compare_unwrap(
       _Iter1 __first1,
       _Sent1 __last1,
       _Iter2 __first2,
       _Sent2 __last2,
       _Comp& __comp,
       _Proj1& __proj1,
       _Proj2& __proj2) {
-    while (__first2 != __last2) {
-      if (__first1 == __last1 || std::invoke(__comp, std::invoke(__proj1, *__first1), std::invoke(__proj2, *__first2)))
-        return true;
-      if (std::invoke(__comp, std::invoke(__proj2, *__first2), std::invoke(__proj1, *__first1)))
-        return false;
-      ++__first1;
-      ++__first2;
-    }
-    return false;
+    auto [__first1_un, __last1_un] = std::__unwrap_range(std::move(__first1), std::move(__last1));
+    auto [__first2_un, __last2_un] = std::__unwrap_range(std::move(__first2), std::move(__last2));
+    return std::__lexicographical_compare(
+        std::move(__first1_un),
+        std::move(__last1_un),
+        std::move(__first2_un),
+        std::move(__last2_un),
+        __comp,
+        __proj1,
+        __proj2);
   }
 
   template <input_iterator _Iter1,
@@ -68,7 +71,7 @@ struct __fn {
       _Comp __comp   = {},
       _Proj1 __proj1 = {},
       _Proj2 __proj2 = {}) const {
-    return __lexicographical_compare_impl(
+    return __lexicographical_compare_unwrap(
         std::move(__first1), std::move(__last1), std::move(__first2), std::move(__last2), __comp, __proj1, __proj2);
   }
 
@@ -80,7 +83,7 @@ struct __fn {
                 _Comp = ranges::less>
   [[nodiscard]] _LIBCPP_HIDE_FROM_ABI constexpr bool operator()(
       _Range1&& __range1, _Range2&& __range2, _Comp __comp = {}, _Proj1 __proj1 = {}, _Proj2 __proj2 = {}) const {
-    return __lexicographical_compare_impl(
+    return __lexicographical_compare_unwrap(
         ranges::begin(__range1),
         ranges::end(__range1),
         ranges::begin(__range2),

libcxx/include/__algorithm/sort.h

@@ -129,25 +129,23 @@ __sort5(_ForwardIterator __x1,
 
 // The comparator being simple is a prerequisite for using the branchless optimization.
 template <class _Tp>
-struct __is_simple_comparator : false_type {};
-template <>
-struct __is_simple_comparator<__less<>&> : true_type {};
+struct __is_simple_comparator : __is_less_than_predicate<_Tp> {};
+template <class _Tp>
+struct __is_simple_comparator<__less<_Tp>&> : true_type {};
 template <class _Tp>
 struct __is_simple_comparator<less<_Tp>&> : true_type {};
 template <class _Tp>
 struct __is_simple_comparator<greater<_Tp>&> : true_type {};
 #if _LIBCPP_STD_VER >= 20
 template <>
-struct __is_simple_comparator<ranges::less&> : true_type {};
-template <>
-struct __is_simple_comparator<ranges::greater&> : true_type {};
+struct __is_simple_comparator<ranges::greater> : true_type {};
 #endif
 
 template <class _Compare, class _Iter, class _Tp = typename iterator_traits<_Iter>::value_type>
 using __use_branchless_sort =
     integral_constant<bool,
                       __libcpp_is_contiguous_iterator<_Iter>::value && sizeof(_Tp) <= sizeof(void*) &&
-                          is_arithmetic<_Tp>::value && __is_simple_comparator<_Compare>::value>;
+                          is_arithmetic<_Tp>::value && __is_simple_comparator<__remove_cvref_t<_Compare>>::value>;
 
 namespace __detail {
 

libcxx/include/__functional/ranges_operations.h

@@ -100,7 +100,7 @@ template <class _Tp, class _Up>
 inline const bool __desugars_to_v<__equal_tag, ranges::equal_to, _Tp, _Up> = true;
 
 template <class _Tp, class _Up>
-inline const bool __desugars_to_v<__less_tag, ranges::less, _Tp, _Up> = true;
+inline const bool __desugars_to_v<__totally_ordered_less_tag, ranges::less, _Tp, _Up> = true;
 
 #endif // _LIBCPP_STD_VER >= 20
 

libcxx/include/__fwd/bit_reference.h

@@ -20,6 +20,9 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 template <class _Cp, bool _IsConst, typename _Cp::__storage_type = 0>
 class __bit_iterator;
 
+template <class _Tp>
+struct __is_less_than_predicate : false_type {};
+
 _LIBCPP_END_NAMESPACE_STD
 
 #endif // _LIBCPP___FWD_BIT_REFERENCE_H

libcxx/include/__string/constexpr_c_functions.h

@@ -64,13 +64,13 @@ inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 size_t __constexpr_st
   return __builtin_strlen(reinterpret_cast<const char*>(__str));
 }
 
-// Because of __libcpp_is_trivially_lexicographically_comparable we know that comparing the object representations is
+// Because of __is_trivially_lexicographically_comparable_v we know that comparing the object representations is
 // equivalent to a std::memcmp. Since we have multiple objects contiguously in memory, we can call memcmp once instead
 // of invoking it on every object individually.
 template <class _Tp, class _Up>
 _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 int
 __constexpr_memcmp(const _Tp* __lhs, const _Up* __rhs, __element_count __n) {
-  static_assert(__libcpp_is_trivially_lexicographically_comparable<_Tp, _Up>::value,
+  static_assert(__is_trivially_lexicographically_comparable_v<_Tp, _Up>,
                 "_Tp and _Up have to be trivially lexicographically comparable");
 
   auto __count = static_cast<size_t>(__n);

libcxx/include/__type_traits/desugars_to.h

@@ -20,7 +20,7 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 // Tags to represent the canonical operations
 struct __equal_tag {};
 struct __plus_tag {};
-struct __less_tag {};
+struct __totally_ordered_less_tag {};
 
 // This class template is used to determine whether an operation "desugars"
 // (or boils down) to a given canonical operation.

libcxx/include/__type_traits/is_equality_comparable.h

@@ -15,10 +15,12 @@
 #include <__type_traits/is_integral.h>
 #include <__type_traits/is_same.h>
 #include <__type_traits/is_signed.h>
+#include <__type_traits/is_unsigned.h>
 #include <__type_traits/is_void.h>
 #include <__type_traits/remove_cv.h>
 #include <__type_traits/void_t.h>
 #include <__utility/declval.h>
+#include <cstddef>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header

libcxx/include/__type_traits/is_trivially_lexicographically_comparable.h

@@ -16,6 +16,7 @@
 #include <__type_traits/remove_cv.h>
 #include <__type_traits/void_t.h>
 #include <__utility/declval.h>
+#include <cstddef>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
@@ -40,13 +41,22 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 // unsigned integer types with sizeof(T) > 1: depending on the endianness, the LSB might be the first byte to be
 //                                            compared. This means that when comparing unsigned(129) and unsigned(2)
 //                                            using memcmp(), the result would be that 2 > 129.
-//                                            TODO: Do we want to enable this on big-endian systems?
+
+template <class _Tp>
+inline const bool __is_std_byte_v = false;
+
+#if _LIBCPP_STD_VER >= 17
+template <>
+inline const bool __is_std_byte_v<byte> = true;
+#endif
 
 template <class _Tp, class _Up>
-struct __libcpp_is_trivially_lexicographically_comparable
-    : integral_constant<bool,
-                        is_same<__remove_cv_t<_Tp>, __remove_cv_t<_Up> >::value && sizeof(_Tp) == 1 &&
-                            is_unsigned<_Tp>::value> {};
+inline const bool __is_trivially_lexicographically_comparable_v =
+    is_same<__remove_cv_t<_Tp>, __remove_cv_t<_Up> >::value &&
+#ifdef _LIBCPP_LITTLE_ENDIAN
+    sizeof(_Tp) == 1 &&
+#endif
+    (is_unsigned<_Tp>::value || __is_std_byte_v<_Tp>);
 
 _LIBCPP_END_NAMESPACE_STD
 

libcxx/include/__utility/pair.h

@@ -34,6 +34,7 @@
 #include <__type_traits/is_nothrow_constructible.h>
 #include <__type_traits/is_same.h>
 #include <__type_traits/is_swappable.h>
+#include <__type_traits/is_trivially_lexicographically_comparable.h>
 #include <__type_traits/is_trivially_relocatable.h>
 #include <__type_traits/nat.h>
 #include <__type_traits/remove_cvref.h>