Reapply "[libc++] Optimize std::find_if" (#175903) (#175921)

#175913 removed that `__builtin_assume_dereferenceable(ptr, 0)` implies
`ptr != nullptr`, which should allow us to use the builtin with LLVM 23.

This reverts commit 776c09c212.

libcxx/include/CMakeLists.txt

@@ -607,6 +607,7 @@ set(files
   __memory/unique_temporary_buffer.h
   __memory/uses_allocator.h
   __memory/uses_allocator_construction.h
+  __memory/valid_range.h
   __memory_resource/memory_resource.h
   __memory_resource/monotonic_buffer_resource.h
   __memory_resource/polymorphic_allocator.h
@@ -931,7 +932,6 @@ set(files
   __utility/in_place.h
   __utility/integer_sequence.h
   __utility/is_pointer_in_range.h
-  __utility/is_valid_range.h
   __utility/lazy_synth_three_way_comparator.h
   __utility/move.h
   __utility/no_destroy.h

libcxx/include/__algorithm/find_if.h

@@ -11,6 +11,7 @@
 #define _LIBCPP___ALGORITHM_FIND_IF_H
 
 #include <__config>
+#include <__memory/valid_range.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header
@@ -21,6 +22,8 @@ _LIBCPP_BEGIN_NAMESPACE_STD
 template <class _InputIterator, class _Predicate>
 [[__nodiscard__]] inline _LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX20 _InputIterator
 find_if(_InputIterator __first, _InputIterator __last, _Predicate __pred) {
+  std::__assume_valid_range(__first, __last);
+
   for (; __first != __last; ++__first)
     if (__pred(*__first))
       break;

libcxx/include/__memory/valid_range.h

@@ -0,0 +1,74 @@
+//===----------------------------------------------------------------------===//
+//
+// 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___MEMORY_VALID_RANGE_H
+#define _LIBCPP___MEMORY_VALID_RANGE_H
+
+#include <__algorithm/comp.h>
+#include <__assert>
+#include <__config>
+#include <__iterator/iterator_traits.h>
+#include <__memory/assume_aligned.h>
+#include <__memory/pointer_traits.h>
+#include <__type_traits/is_constant_evaluated.h>
+#include <__type_traits/is_same.h>
+#include <__type_traits/remove_cvref.h>
+
+#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
+#  pragma GCC system_header
+#endif
+
+_LIBCPP_BEGIN_NAMESPACE_STD
+
+// A valid range as defined by the C++ Standard has the following constraints:
+// - [__first, __last) is dereferenceable
+// - __last is reachable from __first
+// - if __first and __last are contiguous iterators, the pointers they "decay to" are correctly aligned according to the
+// language rules for pointers
+
+// This function attempts to detect invalid ranges as defined above. Specifically, it checks bullet (2). This also means
+// that it doesn't return whether a range is actually valid, but only whether a range is definitely not valid.
+// The checks may be extended in the future.
+template <class _Tp>
+_LIBCPP_CONSTEXPR_SINCE_CXX14 _LIBCPP_HIDE_FROM_ABI _LIBCPP_NO_SANITIZE("address") bool
+__is_valid_range(const _Tp* __first, const _Tp* __last) {
+  if (__libcpp_is_constant_evaluated()) {
+    // If this is not a constant during constant evaluation, that is because __first and __last are not
+    // part of the same allocation. If they are part of the same allocation, we must still make sure they
+    // are ordered properly.
+    return __builtin_constant_p(__first <= __last) && __first <= __last;
+  }
+
+  return !__less<>()(__last, __first);
+}
+
+// This function allows the compiler to assume that [__first, __last) is a valid range as defined above.
+//
+// In practice, we only add explicit assumptions for bullets (1) and (3). These assumptions allow (currently only
+// clang-based compilers) to auto-vectorize algorithms that contain early returns.
+template <class _Iter, class _Sent>
+_LIBCPP_HIDE_FROM_ABI _LIBCPP_CONSTEXPR_SINCE_CXX14 void
+__assume_valid_range([[__maybe_unused__]] _Iter&& __first, [[__maybe_unused__]] _Sent&& __last) {
+#if defined(_LIBCPP_CLANG_VER) && _LIBCPP_CLANG_VER >= 2300 && !defined(_LIBCPP_CXX03_LANG)
+  if constexpr (__libcpp_is_contiguous_iterator<__remove_cvref_t<_Iter>>::value &&
+                is_same<__remove_cvref_t<_Iter>, __remove_cvref_t<_Sent>>::value) {
+    _LIBCPP_ASSERT_INTERNAL(std::__is_valid_range(std::__to_address(__first), std::__to_address(__last)),
+                            "Valid range assumption does not hold");
+    if (!__libcpp_is_constant_evaluated()) {
+      using __value_type = typename iterator_traits<__remove_cvref_t<_Iter>>::value_type;
+      __builtin_assume_dereferenceable(std::__to_address(__first), (__last - __first) * sizeof(__value_type));
+      (void)std::__assume_aligned<_LIBCPP_ALIGNOF(__value_type)>(std::__to_address(__first));
+      (void)std::__assume_aligned<_LIBCPP_ALIGNOF(__value_type)>(std::__to_address(__last));
+    }
+  }
+#endif
+}
+
+_LIBCPP_END_NAMESPACE_STD
+
+#endif // _LIBCPP___MEMORY_VALID_RANGE_H

libcxx/include/__utility/is_pointer_in_range.h

@@ -12,12 +12,12 @@
 #include <__algorithm/comp.h>
 #include <__assert>
 #include <__config>
+#include <__memory/valid_range.h>
 #include <__type_traits/enable_if.h>
 #include <__type_traits/integral_constant.h>
 #include <__type_traits/is_constant_evaluated.h>
 #include <__type_traits/void_t.h>
 #include <__utility/declval.h>
-#include <__utility/is_valid_range.h>
 
 #if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
 #  pragma GCC system_header

libcxx/include/__utility/is_valid_range.h

@@ -1,37 +0,0 @@
-//===----------------------------------------------------------------------===//
-//
-// 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___UTILITY_IS_VALID_RANGE_H
-#define _LIBCPP___UTILITY_IS_VALID_RANGE_H
-
-#include <__algorithm/comp.h>
-#include <__config>
-#include <__type_traits/is_constant_evaluated.h>
-
-#if !defined(_LIBCPP_HAS_NO_PRAGMA_SYSTEM_HEADER)
-#  pragma GCC system_header
-#endif
-
-_LIBCPP_BEGIN_NAMESPACE_STD
-
-template <class _Tp>
-_LIBCPP_CONSTEXPR_SINCE_CXX14 _LIBCPP_HIDE_FROM_ABI _LIBCPP_NO_SANITIZE("address") bool
-__is_valid_range(const _Tp* __first, const _Tp* __last) {
-  if (__libcpp_is_constant_evaluated()) {
-    // If this is not a constant during constant evaluation, that is because __first and __last are not
-    // part of the same allocation. If they are part of the same allocation, we must still make sure they
-    // are ordered properly.
-    return __builtin_constant_p(__first <= __last) && __first <= __last;
-  }
-
-  return !__less<>()(__last, __first);
-}
-
-_LIBCPP_END_NAMESPACE_STD
-
-#endif // _LIBCPP___UTILITY_IS_VALID_RANGE_H

libcxx/include/module.modulemap.in

@@ -1690,6 +1690,7 @@ module std [system] {
     }
     module uses_allocator                     { header "__memory/uses_allocator.h" }
     module uses_allocator_construction        { header "__memory/uses_allocator_construction.h" }
+    module valid_range                        { header "__memory/valid_range.h" }
 
     header "memory"
     export *
@@ -2172,7 +2173,6 @@ module std [system] {
     }
     module integer_sequence                { header "__utility/integer_sequence.h" }
     module is_pointer_in_range             { header "__utility/is_pointer_in_range.h" }
-    module is_valid_range                  { header "__utility/is_valid_range.h" }
     module lazy_synth_three_way_comparator { header "__utility/lazy_synth_three_way_comparator.h" }
     module move                            { header "__utility/move.h" }
     module no_destroy                      { header "__utility/no_destroy.h" }

libcxx/include/streambuf

@@ -117,8 +117,8 @@ protected:
 #    include <__assert>
 #    include <__fwd/streambuf.h>
 #    include <__locale>
+#    include <__memory/valid_range.h>
 #    include <__type_traits/is_same.h>
-#    include <__utility/is_valid_range.h>
 #    include <__utility/scope_guard.h>
 #    include <climits>
 #    include <ios>

libcxx/test/benchmarks/algorithms/nonmodifying/find.bench.cpp

@@ -22,30 +22,18 @@
 int main(int argc, char** argv) {
   auto std_find    = [](auto first, auto last, auto const& value) { return std::find(first, last, value); };
   auto std_find_if = [](auto first, auto last, auto const& value) {
-    return std::find_if(first, last, [&](auto element) {
-      benchmark::DoNotOptimize(element);
-      return element == value;
-    });
+    return std::find_if(first, last, [&](auto element) { return element == value; });
   };
   auto std_find_if_not = [](auto first, auto last, auto const& value) {
-    return std::find_if_not(first, last, [&](auto element) {
-      benchmark::DoNotOptimize(element);
-      return element != value;
-    });
+    return std::find_if_not(first, last, [&](auto element) { return element != value; });
   };
 
   auto ranges_find    = [](auto first, auto last, auto const& value) { return std::ranges::find(first, last, value); };
   auto ranges_find_if = [](auto first, auto last, auto const& value) {
-    return std::ranges::find_if(first, last, [&](auto element) {
-      benchmark::DoNotOptimize(element);
-      return element == value;
-    });
+    return std::ranges::find_if(first, last, [&](auto element) { return element == value; });
   };
   auto ranges_find_if_not = [](auto first, auto last, auto const& value) {
-    return std::ranges::find_if_not(first, last, [&](auto element) {
-      benchmark::DoNotOptimize(element);
-      return element != value;
-    });
+    return std::ranges::find_if_not(first, last, [&](auto element) { return element != value; });
   };
 
   auto register_benchmarks = [&](auto bm, std::string comment) {

libcxx/test/libcxx/utilities/is_valid_range.pass.cpp

@@ -6,7 +6,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include <__utility/is_valid_range.h>
+#include <__memory/valid_range.h>
 #include <cassert>
 
 #include "test_macros.h"

libcxx/test/std/algorithms/alg.nonmodifying/alg.find/find_if.pass.cpp

@@ -56,6 +56,12 @@ TEST_CONSTEXPR_CXX17 bool test() {
   test_iter<forward_iterator<int*> >();
   test_iter<bidirectional_iterator<int*> >();
   test_iter<random_access_iterator<int*> >();
+
+  { // Check that [nullptr, nullptr) is an accepted range
+    auto result = std::find_if(static_cast<int*>(nullptr), static_cast<int*>(nullptr), [](int i) { return i == 0; });
+    assert(result == nullptr);
+  }
+
   return true;
 }