[libc++] Forward to std::{,w}memchr in std::find

Reviewed By: #libc, ldionne

Spies: Mordante, libcxx-commits, ldionne, mikhail.ramalho

Differential Revision: https://reviews.llvm.org/D144394

libcxx/benchmarks/CMakeLists.txt

@@ -159,6 +159,7 @@ endfunction()
 set(BENCHMARK_TESTS
     algorithms.partition_point.bench.cpp
     algorithms/equal.bench.cpp
+    algorithms/find.bench.cpp
     algorithms/lower_bound.bench.cpp
     algorithms/make_heap.bench.cpp
     algorithms/make_heap_then_sort_heap.bench.cpp

libcxx/benchmarks/algorithms/find.bench.cpp

@@ -0,0 +1,49 @@
+//===----------------------------------------------------------------------===//
+//
+// 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 <algorithm>
+#include <benchmark/benchmark.h>
+#include <cstring>
+#include <random>
+#include <vector>
+
+template <class T>
+static void bm_find(benchmark::State& state) {
+  std::vector<T> vec1(state.range(), '1');
+  std::mt19937_64 rng(std::random_device{}());
+
+  for (auto _ : state) {
+    auto idx  = rng() % vec1.size();
+    vec1[idx] = '2';
+    benchmark::DoNotOptimize(vec1);
+    benchmark::DoNotOptimize(std::find(vec1.begin(), vec1.end(), T('2')));
+    vec1[idx] = '1';
+  }
+}
+BENCHMARK(bm_find<char>)->DenseRange(1, 8)->Range(16, 1 << 20);
+BENCHMARK(bm_find<short>)->DenseRange(1, 8)->Range(16, 1 << 20);
+BENCHMARK(bm_find<int>)->DenseRange(1, 8)->Range(16, 1 << 20);
+
+template <class T>
+static void bm_ranges_find(benchmark::State& state) {
+  std::vector<T> vec1(state.range(), '1');
+  std::mt19937_64 rng(std::random_device{}());
+
+  for (auto _ : state) {
+    auto idx  = rng() % vec1.size();
+    vec1[idx] = '2';
+    benchmark::DoNotOptimize(vec1);
+    benchmark::DoNotOptimize(std::ranges::find(vec1, T('2')));
+    vec1[idx] = '1';
+  }
+}
+BENCHMARK(bm_ranges_find<char>)->DenseRange(1, 8)->Range(16, 1 << 20);
+BENCHMARK(bm_ranges_find<short>)->DenseRange(1, 8)->Range(16, 1 << 20);
+BENCHMARK(bm_ranges_find<int>)->DenseRange(1, 8)->Range(16, 1 << 20);
+
+BENCHMARK_MAIN();

libcxx/include/__algorithm/find.h

@@ -10,23 +10,68 @@
 #ifndef _LIBCPP___ALGORITHM_FIND_H
 #define _LIBCPP___ALGORITHM_FIND_H
 
+#include <__algorithm/unwrap_iter.h>
 #include <__config>
+#include <__functional/identity.h>
+#include <__functional/invoke.h>
+#include <__string/constexpr_c_functions.h>
+#include <__type_traits/is_same.h>
+
+#ifndef _LIBCPP_HAS_NO_WIDE_CHARACTERS
+#  include <cwchar>
+#endif
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
 #endif
 
 _LIBCPP_BEGIN_NAMESPACE_STD
 
-template <class _InputIterator, class _Tp>
-_LIBCPP_NODISCARD_EXT inline _LIBCPP_INLINE_VISIBILITY _LIBCPP_CONSTEXPR_SINCE_CXX20 _InputIterator
-find(_InputIterator __first, _InputIterator __last, const _Tp& __value) {
+template <class _Iter, class _Sent, class _Tp, class _Proj>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Iter
+__find_impl(_Iter __first, _Sent __last, const _Tp& __value, _Proj& __proj) {
   for (; __first != __last; ++__first)
-    if (*__first == __value)
+    if (std::__invoke(__proj, *__first) == __value)
       break;
   return __first;
 }
 
+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_impl(_Tp* __first, _Tp* __last, const _Up& __value, _Proj&) {
+  if (auto __ret = std::__constexpr_memchr(__first, __value, __last - __first))
+    return __ret;
+  return __last;
+}
+
+#ifndef _LIBCPP_HAS_NO_WIDE_CHARACTERS
+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>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp*
+__find_impl(_Tp* __first, _Tp* __last, const _Up& __value, _Proj&) {
+  if (auto __ret = std::__constexpr_wmemchr(__first, __value, __last - __first))
+    return __ret;
+  return __last;
+}
+#endif // _LIBCPP_HAS_NO_WIDE_CHARACTERS
+
+template <class _InputIterator, class _Tp>
+_LIBCPP_NODISCARD_EXT inline _LIBCPP_INLINE_VISIBILITY _LIBCPP_CONSTEXPR_SINCE_CXX20 _InputIterator
+find(_InputIterator __first, _InputIterator __last, const _Tp& __value) {
+  __identity __proj;
+  return std::__rewrap_iter(
+      __first, std::__find_impl(std::__unwrap_iter(__first), std::__unwrap_iter(__last), __value, __proj));
+}
+
 _LIBCPP_END_NAMESPACE_STD
 
 #endif // _LIBCPP___ALGORITHM_FIND_H

libcxx/include/__algorithm/ranges_find.h

@@ -9,7 +9,9 @@
 #ifndef _LIBCPP___ALGORITHM_RANGES_FIND_H
 #define _LIBCPP___ALGORITHM_RANGES_FIND_H
 
+#include <__algorithm/find.h>
 #include <__algorithm/ranges_find_if.h>
+#include <__algorithm/unwrap_range.h>
 #include <__config>
 #include <__functional/identity.h>
 #include <__functional/invoke.h>
@@ -33,20 +35,30 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 namespace ranges {
 namespace __find {
 struct __fn {
+  template <class _Iter, class _Sent, class _Tp, class _Proj>
+  _LIBCPP_HIDE_FROM_ABI static constexpr _Iter
+  __find_unwrap(_Iter __first, _Sent __last, const _Tp& __value, _Proj& __proj) {
+    if constexpr (forward_iterator<_Iter>) {
+      auto [__first_un, __last_un] = std::__unwrap_range(__first, std::move(__last));
+      return std::__rewrap_range<_Sent>(
+          std::move(__first), std::__find_impl(std::move(__first_un), std::move(__last_un), __value, __proj));
+    } else {
+      return std::__find_impl(std::move(__first), std::move(__last), __value, __proj);
+    }
+  }
+
   template <input_iterator _Ip, sentinel_for<_Ip> _Sp, class _Tp, class _Proj = identity>
     requires indirect_binary_predicate<ranges::equal_to, projected<_Ip, _Proj>, const _Tp*>
   _LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr
   _Ip operator()(_Ip __first, _Sp __last, const _Tp& __value, _Proj __proj = {}) const {
-    auto __pred = [&](auto&& __e) { return std::forward<decltype(__e)>(__e) == __value; };
-    return ranges::__find_if_impl(std::move(__first), std::move(__last), __pred, __proj);
+    return __find_unwrap(std::move(__first), std::move(__last), __value, __proj);
   }
 
   template <input_range _Rp, class _Tp, class _Proj = identity>
     requires indirect_binary_predicate<ranges::equal_to, projected<iterator_t<_Rp>, _Proj>, const _Tp*>
   _LIBCPP_NODISCARD_EXT _LIBCPP_HIDE_FROM_ABI constexpr
   borrowed_iterator_t<_Rp> operator()(_Rp&& __r, const _Tp& __value, _Proj __proj = {}) const {
-    auto __pred = [&](auto&& __e) { return std::forward<decltype(__e)>(__e) == __value; };
-    return ranges::__find_if_impl(ranges::begin(__r), ranges::end(__r), __pred, __proj);
+    return __find_unwrap(ranges::begin(__r), ranges::end(__r), __value, __proj);
   }
 };
 } // namespace __find

libcxx/include/__string/char_traits.h

@@ -244,7 +244,7 @@ struct _LIBCPP_TEMPLATE_VIS char_traits<char>
     const char_type* find(const char_type* __s, size_t __n, const char_type& __a) _NOEXCEPT {
       if (__n == 0)
           return nullptr;
-      return std::__constexpr_char_memchr(__s, static_cast<int>(__a), __n);
+      return std::__constexpr_memchr(__s, __a, __n);
     }
 
     static inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20

libcxx/include/__string/constexpr_c_functions.h

@@ -14,6 +14,7 @@
 #include <__type_traits/is_equality_comparable.h>
 #include <__type_traits/is_same.h>
 #include <__type_traits/is_trivially_lexicographically_comparable.h>
+#include <__type_traits/remove_cv.h>
 #include <cstddef>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
@@ -95,20 +96,29 @@ __constexpr_memcmp_equal(const _Tp* __lhs, const _Up* __rhs, size_t __count) {
   }
 }
 
-inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 const char*
-__constexpr_char_memchr(const char* __str, int __char, size_t __count) {
-#if __has_builtin(__builtin_char_memchr)
-  return __builtin_char_memchr(__str, __char, __count);
-#else
-  if (!__libcpp_is_constant_evaluated())
-    return static_cast<const char*>(__builtin_memchr(__str, __char, __count));
-  for (; __count; --__count) {
-    if (*__str == __char)
-      return __str;
-    ++__str;
-  }
-  return nullptr;
+template <class _Tp, class _Up>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp* __constexpr_memchr(_Tp* __str, _Up __value, size_t __count) {
+  static_assert(sizeof(_Tp) == 1 && __libcpp_is_trivially_equality_comparable<_Tp, _Up>::value,
+                "Calling memchr on non-trivially equality comparable types is unsafe.");
+
+  if (__libcpp_is_constant_evaluated()) {
+// use __builtin_char_memchr to optimize constexpr evaluation if we can
+#if _LIBCPP_STD_VER >= 17 && __has_builtin(__builtin_char_memchr)
+    if constexpr (is_same_v<remove_cv_t<_Tp>, char> && is_same_v<remove_cv_t<_Up>, char>)
+      return __builtin_char_memchr(__str, __value, __count);
 #endif
+
+    for (; __count; --__count) {
+      if (*__str == __value)
+        return __str;
+      ++__str;
+    }
+    return nullptr;
+  } else {
+    char __value_buffer = 0;
+    __builtin_memcpy(&__value_buffer, &__value, sizeof(char));
+    return static_cast<_Tp*>(__builtin_memchr(__str, __value_buffer, __count));
+  }
 }
 
 _LIBCPP_END_NAMESPACE_STD

libcxx/include/cwchar

@@ -104,7 +104,11 @@ size_t wcsrtombs(char* restrict dst, const wchar_t** restrict src, size_t len,
 
 #include <__assert> // all public C++ headers provide the assertion handler
 #include <__config>
+#include <__type_traits/apply_cv.h>
 #include <__type_traits/is_constant_evaluated.h>
+#include <__type_traits/is_equality_comparable.h>
+#include <__type_traits/is_same.h>
+#include <__type_traits/remove_cv.h>
 #include <cwctype>
 
 #include <wchar.h>
@@ -222,21 +226,31 @@ __constexpr_wmemcmp(const wchar_t* __lhs, const wchar_t* __rhs, size_t __count)
 #endif
 }
 
-inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 const wchar_t*
-__constexpr_wmemchr(const wchar_t* __str, wchar_t __char, size_t __count) {
-#if __has_feature(cxx_constexpr_string_builtins)
-  return __builtin_wmemchr(__str, __char, __count);
-#else
-  if (!__libcpp_is_constant_evaluated())
-    return std::wmemchr(__str, __char, __count);
+template <class _Tp, class _Up>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 _Tp* __constexpr_wmemchr(_Tp* __str, _Up __value, size_t __count) {
+  static_assert(sizeof(_Tp) == sizeof(wchar_t)&& _LIBCPP_ALIGNOF(_Tp) >= _LIBCPP_ALIGNOF(wchar_t) &&
+                    __libcpp_is_trivially_equality_comparable<_Tp, _Tp>::value,
+                "Calling wmemchr on non-trivially equality comparable types is unsafe.");
+
+#if __has_builtin(__builtin_wmemchr)
+  if (!__libcpp_is_constant_evaluated()) {
+    wchar_t __value_buffer = 0;
+    __builtin_memcpy(&__value_buffer, &__value, sizeof(wchar_t));
+    return reinterpret_cast<_Tp*>(
+        __builtin_wmemchr(reinterpret_cast<__apply_cv_t<_Tp, wchar_t>*>(__str), __value_buffer, __count));
+  }
+#  if _LIBCPP_STD_VER >= 17
+  else if constexpr (is_same_v<remove_cv_t<_Tp>, wchar_t>)
+    return __builtin_wmemchr(__str, __value, __count);
+#  endif
+#endif // __has_builtin(__builtin_wmemchr)
 
   for (; __count; --__count) {
-    if (*__str == __char)
+    if (*__str == __value)
       return __str;
     ++__str;
   }
   return nullptr;
-#endif
 }
 
 _LIBCPP_END_NAMESPACE_STD

libcxx/test/libcxx/strings/c.strings/constexpr.cstring.compile.pass.cpp

@@ -29,6 +29,6 @@ static_assert(std::__constexpr_memcmp_equal(Banane, Banane, 6), "");
 
 constexpr bool test_constexpr_wmemchr() {
   const char str[] = "Banane";
-  return std::__constexpr_char_memchr(str, 'n', 6) == str + 2;
+  return std::__constexpr_memchr(str, 'n', 6) == str + 2;
 }
 static_assert(test_constexpr_wmemchr(), "");

libcxx/test/libcxx/transitive_includes/cxx03.csv

@@ -7,6 +7,7 @@ algorithm cstdint
 algorithm cstdlib
 algorithm cstring
 algorithm ctime
+algorithm cwchar
 algorithm execution
 algorithm initializer_list
 algorithm iosfwd
@@ -191,6 +192,7 @@ coroutine version
 cstddef version
 ctgmath ccomplex
 ctgmath cmath
+cwchar cstddef
 cwchar cwctype
 cwctype cctype
 deque algorithm
@@ -201,6 +203,7 @@ deque cstddef
 deque cstdint
 deque cstdlib
 deque cstring
+deque cwchar
 deque functional
 deque initializer_list
 deque iosfwd
@@ -950,6 +953,7 @@ vector cstddef
 vector cstdint
 vector cstdlib
 vector cstring
+vector cwchar
 vector initializer_list
 vector iosfwd
 vector limits

libcxx/test/libcxx/transitive_includes/cxx11.csv

@@ -7,6 +7,7 @@ algorithm cstdint
 algorithm cstdlib
 algorithm cstring
 algorithm ctime
+algorithm cwchar
 algorithm execution
 algorithm initializer_list
 algorithm iosfwd
@@ -191,6 +192,7 @@ coroutine version
 cstddef version
 ctgmath ccomplex
 ctgmath cmath
+cwchar cstddef
 cwchar cwctype
 cwctype cctype
 deque algorithm
@@ -201,6 +203,7 @@ deque cstddef
 deque cstdint
 deque cstdlib
 deque cstring
+deque cwchar
 deque functional
 deque initializer_list
 deque iosfwd
@@ -951,6 +954,7 @@ vector cstddef
 vector cstdint
 vector cstdlib
 vector cstring
+vector cwchar
 vector initializer_list
 vector iosfwd
 vector limits

libcxx/test/libcxx/transitive_includes/cxx14.csv

@@ -7,6 +7,7 @@ algorithm cstdint
 algorithm cstdlib
 algorithm cstring
 algorithm ctime
+algorithm cwchar
 algorithm execution
 algorithm initializer_list
 algorithm iosfwd
@@ -191,6 +192,7 @@ coroutine version
 cstddef version
 ctgmath ccomplex
 ctgmath cmath
+cwchar cstddef
 cwchar cwctype
 cwctype cctype
 deque algorithm
@@ -201,6 +203,7 @@ deque cstddef
 deque cstdint
 deque cstdlib
 deque cstring
+deque cwchar
 deque functional
 deque initializer_list
 deque iosfwd
@@ -953,6 +956,7 @@ vector cstddef
 vector cstdint
 vector cstdlib
 vector cstring
+vector cwchar
 vector initializer_list
 vector iosfwd
 vector limits