Fix missing bytesize check for safe find vectorization with LLVM

stl/inc/algorithm

@@ -5774,7 +5774,7 @@ namespace ranges {
             constexpr bool _Allow_vectorization =
                 conjunction_v<_Is_trivially_ranges_swappable<_Elem>, negation<is_volatile<_Elem>>>;
 
-            if constexpr (_Allow_vectorization && _Nx <= 8 && (_Nx & (_Nx - 1)) == 0) {
+            if constexpr (_Allow_vectorization && _Is_vector_element_size<_Nx>) {
                 if (!_STD is_constant_evaluated()) {
                     _STD _Reverse_vectorized<_Nx>(_STD to_address(_First), _STD to_address(_Last));
                     return;
@@ -5834,7 +5834,7 @@ _CONSTEXPR20 _OutIt reverse_copy(_BidIt _First, _BidIt _Last, _OutIt _Dest) {
         _Is_trivially_copy_assignable_returning_same_reference<_Elem>, negation<is_volatile<_Elem>>>;
     constexpr size_t _Nx                = sizeof(_Elem);
 
-    if constexpr (_Allow_vectorization && _Nx <= 8 && (_Nx & (_Nx - 1)) == 0) {
+    if constexpr (_Allow_vectorization && _Is_vector_element_size<_Nx>) {
 #if _HAS_CXX20
         if (!_STD is_constant_evaluated())
 #endif // _HAS_CXX20
@@ -5924,7 +5924,7 @@ namespace ranges {
                     _Is_trivially_copy_assignable_returning_same_reference<_Elem>, negation<is_volatile<_Elem>>>;
                 constexpr size_t _Nx                = sizeof(_Elem);
 
-                if constexpr (_Allow_vectorization && _Nx <= 8 && (_Nx & (_Nx - 1)) == 0) {
+                if constexpr (_Allow_vectorization && _Is_vector_element_size<_Nx>) {
                     if (!_STD is_constant_evaluated()) {
                         const auto _Dest_dist = static_cast<iter_difference_t<_Out>>(_Last - _First);
                         _STD _Contiguous_iter_verify(_Output, _Dest_dist);

stl/inc/xutility

@@ -5789,11 +5789,11 @@ template <class _Iter1, class _Iter2, class _Pr>
 constexpr bool _Equal_memcmp_is_safe =
     _Equal_memcmp_is_safe_helper<remove_const_t<_Iter1>, remove_const_t<_Iter2>, remove_const_t<_Pr>>;
 
-#if _VECTORIZED_SEARCH || _VECTORIZED_ADJACENT_FIND || _VECTORIZED_FIND_END || _VECTORIZED_FIND_FIRST_OF \
-    || _VECTORIZED_UNIQUE || _VECTORIZED_UNIQUE_COPY
 template <size_t _Size>
 constexpr bool _Is_vector_element_size = _Size == 1 || _Size == 2 || _Size == 4 || _Size == 8;
 
+#if _VECTORIZED_SEARCH || _VECTORIZED_ADJACENT_FIND || _VECTORIZED_FIND_END || _VECTORIZED_FIND_FIRST_OF \
+    || _VECTORIZED_UNIQUE || _VECTORIZED_UNIQUE_COPY
 // Can we activate the vector algorithms for search, adjacent_find, find_end, find_first_of, and default_searcher?
 // (Also used indirectly for unique and unique_copy.)
 template <class _It1, class _It2, class _Pr>
@@ -6344,23 +6344,25 @@ struct _Vector_alg_in_find_is_safe_object_pointers<_Ty1*, _Ty2*>
 // Can we activate the vector algorithms to find a value in a range of elements?
 template <class _Ty, class _Elem>
 constexpr bool _Vector_alg_in_find_is_safe_elem =
-    _Is_same_and_builtin_trivially_equality_comparable<_Ty, _Elem>
-    || disjunction_v<
+    _Is_vector_element_size<sizeof(_Elem)>
+    && (_Is_same_and_builtin_trivially_equality_comparable<_Ty, _Elem>
+        || disjunction_v<
 #ifdef __cpp_lib_byte
-        // We're finding a std::byte in a range of std::byte.
-        conjunction<is_same<_Ty, byte>, is_same<_Elem, byte>>,
+            // We're finding a std::byte in a range of std::byte.
+            conjunction<is_same<_Ty, byte>, is_same<_Elem, byte>>,
 #endif // defined(__cpp_lib_byte)
-       // We're finding an integer in a range of integers.
-       // This case is the one that requires careful runtime handling in _Could_compare_equal_to_value_type.
-        conjunction<is_integral<_Ty>, is_integral<_Elem>>,
-        // We're finding an (object or function) pointer in a range of pointers of the same type.
-        conjunction<is_pointer<_Ty>, is_same<_Ty, _Elem>>,
-        // We're finding a nullptr in a range of (object or function) pointers.
-        conjunction<is_null_pointer<_Ty>, is_pointer<_Elem>>,
-        // We're finding an object pointer in a range of object pointers, and:
-        // - One of the pointer types is a cv void*.
-        // - One of the pointer types is a cv1 U* and the other is a cv2 U*.
-        _Vector_alg_in_find_is_safe_object_pointers<_Ty, _Elem>>;
+
+            // We're finding an integer in a range of integers.
+            // This case is the one that requires careful runtime handling in _Could_compare_equal_to_value_type.
+            conjunction<is_integral<_Ty>, is_integral<_Elem>>,
+            // We're finding an (object or function) pointer in a range of pointers of the same type.
+            conjunction<is_pointer<_Ty>, is_same<_Ty, _Elem>>,
+            // We're finding a nullptr in a range of (object or function) pointers.
+            conjunction<is_null_pointer<_Ty>, is_pointer<_Elem>>,
+            // We're finding an object pointer in a range of object pointers, and:
+            // - One of the pointer types is a cv void*.
+            // - One of the pointer types is a cv1 U* and the other is a cv2 U*.
+            _Vector_alg_in_find_is_safe_object_pointers<_Ty, _Elem>>);
 
 // Can we activate the vector algorithms for find, count, remove, remove_copy, and ranges::find_last?
 // (Also used indirectly for replace and search_n.)
@@ -6840,7 +6842,7 @@ _CONSTEXPR20 void reverse(const _BidIt _First, const _BidIt _Last) { // reverse
         _Is_trivially_swappable<_Elem>, negation<is_volatile<_Elem>>>;
     constexpr size_t _Nx                = sizeof(_Elem);
 
-    if constexpr (_Allow_vectorization && _Nx <= 8 && (_Nx & (_Nx - 1)) == 0) {
+    if constexpr (_Allow_vectorization && _Is_vector_element_size<sizeof(_Elem)>) {
 #if _HAS_CXX20
         if (!_STD is_constant_evaluated())
 #endif // _HAS_CXX20

tests/std/test.lst

@@ -281,6 +281,7 @@ tests\GH_005800_stable_sort_large_alignment
 tests\GH_005816_numeric_limits_traps
 tests\GH_005968_headers_provide_begin_end
 tests\GH_005974_asan_annotate_optional
+tests\GH_006294_vectorized_find_is_safe
 tests\LWG2381_num_get_floating_point
 tests\LWG2510_tag_classes
 tests\LWG2597_complex_branch_cut

tests/std/tests/GH_000431_equal_memcmp_is_safe/test.compile.pass.cpp

@@ -157,6 +157,13 @@ struct DefaultComparisonOddSize {
 
     bool operator==(const DefaultComparisonOddSize&) const noexcept = default;
 };
+
+struct DefaultComparisonLargeSize {
+    long long i;
+    long long j;
+
+    bool operator==(const DefaultComparisonLargeSize&) const noexcept = default;
+};
 #endif // _HAS_CXX20
 
 #ifdef __cpp_lib_is_pointer_interconvertible
@@ -529,10 +536,12 @@ STATIC_ASSERT(test_equal_memcmp_is_safe_for_types<magic, DefaultComparisonOddSiz
 // The only difference between _Equal_memcmp_is_safe and _Vector_alg_in_search_is_safe is how the magic works
 STATIC_ASSERT(_Equal_memcmp_is_safe<DefaultComparison*, DefaultComparison*, equal_to<>> == magic);
 STATIC_ASSERT(_Equal_memcmp_is_safe<DefaultComparisonOddSize*, DefaultComparisonOddSize*, equal_to<>> == magic);
+STATIC_ASSERT(_Equal_memcmp_is_safe<DefaultComparisonLargeSize*, DefaultComparisonLargeSize*, equal_to<>> == magic);
 #if _VECTORIZED_SEARCH || _VECTORIZED_ADJACENT_FIND || _VECTORIZED_FIND_END || _VECTORIZED_FIND_FIRST_OF \
     || _VECTORIZED_UNIQUE || _VECTORIZED_UNIQUE_COPY
 STATIC_ASSERT(_Vector_alg_in_search_is_safe<DefaultComparison*, DefaultComparison*, equal_to<>> == magic);
 STATIC_ASSERT(!_Vector_alg_in_search_is_safe<DefaultComparisonOddSize*, DefaultComparisonOddSize*, equal_to<>>);
+STATIC_ASSERT(!_Vector_alg_in_search_is_safe<DefaultComparisonLargeSize*, DefaultComparisonLargeSize*, equal_to<>>);
 #endif // ^^^ _Vector_alg_in_search_is_safe is available ^^^
 #endif // _HAS_CXX20
 

tests/std/tests/GH_006294_vectorized_find_is_safe/env.lst

@@ -0,0 +1,4 @@
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+RUNALL_INCLUDE ..\usual_matrix.lst

tests/std/tests/GH_006294_vectorized_find_is_safe/test.compile.pass.cpp

@@ -0,0 +1,63 @@
+// Copyright (c) Microsoft Corporation.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#include <algorithm>
+#include <cstdint>
+
+using namespace std;
+
+#define STATIC_ASSERT(...) static_assert(__VA_ARGS__, #__VA_ARGS__)
+
+#if _HAS_CXX20
+
+#ifdef __clang__
+constexpr bool vectorized_trivial_comparison = true;
+#else
+constexpr bool vectorized_trivial_comparison = false;
+#endif
+
+struct DefaultComparison {
+    int32_t i;
+
+    bool operator==(const DefaultComparison&) const noexcept = default;
+};
+
+struct DefaultComparisonOddSize {
+    int32_t i;
+    int16_t j;
+
+    bool operator==(const DefaultComparisonOddSize&) const noexcept = default;
+};
+
+struct DefaultComparisonLargeSize {
+    int32_t i;
+    int32_t j;
+
+    bool operator==(const DefaultComparisonLargeSize&) const noexcept = default;
+};
+
+struct DefaultComparisonTooLargeSize {
+    int64_t i;
+    int64_t j;
+
+    bool operator==(const DefaultComparisonTooLargeSize&) const noexcept = default;
+};
+
+#endif // _HAS_CXX20
+
+STATIC_ASSERT(
+    _Vector_alg_in_find_is_safe<int*, int>); // ints are trivially equality comparable and 4 bytes, so should be allowed
+STATIC_ASSERT(_Vector_alg_in_find_is_safe<long*,
+    long>); // long is an integral type with a supported element size (4 or 8 bytes), so should be allowed
+
+STATIC_ASSERT(_Vector_alg_in_find_is_safe<int const*, int>); // "Top level const should not change the answer"
+STATIC_ASSERT(!_Vector_alg_in_find_is_safe<float*, float>); // "Don't allow floating point"
+STATIC_ASSERT(!_Vector_alg_in_find_is_safe<int volatile*, int>); // "Don't allow volatile"
+
+#if _HAS_CXX20
+STATIC_ASSERT(_Vector_alg_in_find_is_safe<DefaultComparison*, DefaultComparison> == vectorized_trivial_comparison);
+STATIC_ASSERT(_Vector_alg_in_find_is_safe<DefaultComparisonLargeSize*, DefaultComparisonLargeSize>
+              == vectorized_trivial_comparison);
+STATIC_ASSERT(!_Vector_alg_in_find_is_safe<DefaultComparisonOddSize*, DefaultComparisonOddSize>);
+STATIC_ASSERT(!_Vector_alg_in_find_is_safe<DefaultComparisonTooLargeSize*, DefaultComparisonTooLargeSize>);
+#endif // _HAS_CXX20