[pstl] Add additional namespace qualifications

[rodgert]

Summary

Guarding against unintended ADL issues

pstl/trunk/include/pstl/internal/algorithm_impl.h

__brick_any_of(const _ForwardIterator __first, const _ForwardIterator __last, _Pred __pred,
     return __unseq_backend::__simd_or(__first, __last - __first, __pred);
 };
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Pred, class _IsVector>
 bool
 __pattern_any_of(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Pred __pred,
                  _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
 {
-    return __brick_any_of(__first, __last, __pred, __is_vector);
+    return __internal::__brick_any_of(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Pred, class _IsVector>
 bool
 __pattern_any_of(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Pred __pred,
                  _IsVector __is_vector, /*parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
-        return __parallel_or(std::forward<_ExecutionPolicy>(__exec), __first, __last,
+    return __internal::__except_handler([&]() {
+        return __internal::__parallel_or(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                          [__pred, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                                 return __brick_any_of(__i, __j, __pred, __is_vector);
+                                             return __internal::__brick_any_of(__i, __j, __pred, __is_vector);
                                          });
     });
 }
 #endif
 
 // [alg.foreach]
 // for_each_n with no policy
 
 template <class _ForwardIterator, class _Size, class _Function>

 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Function, class _IsVector>
 void
 __pattern_walk1(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Function __f,
                 _IsVector __is_vector,
                 /*parallel=*/std::false_type) noexcept
 {
-    __brick_walk1(__first, __last, __f, __is_vector);
+    __internal::__brick_walk1(__first, __last, __f, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Function, class _IsVector>
 void
 __pattern_walk1(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Function __f,
                 _IsVector __is_vector,
                 /*parallel=*/std::true_type)
 {
-    __except_handler([&]() {
+    __internal::__except_handler([&]() {
         __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                       [__f, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                                          __brick_walk1(__i, __j, __f, __is_vector);
+                                          __internal::__brick_walk1(__i, __j, __f, __is_vector);
                                       });
     });
 }
 #endif
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Brick>
 void
 __pattern_walk_brick(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Brick __brick,
                      /*parallel=*/std::false_type) noexcept
 {
     __brick(__first, __last);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Brick>
 void
 __pattern_walk_brick(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Brick __brick,
                      /*parallel=*/std::true_type)
 {
-    __except_handler([&]() {
+    __internal::__except_handler([&]() {
         __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                       [__brick](_ForwardIterator __i, _ForwardIterator __j) { __brick(__i, __j); });
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // walk1_n
 //------------------------------------------------------------------------
 template <class _ForwardIterator, class _Size, class _Function>
 _ForwardIterator
 __brick_walk1_n(_ForwardIterator __first, _Size __n, _Function __f, /*_IsVectorTag=*/std::false_type)
 {
-    return __for_each_n_it_serial(__first, __n,
+    return __internal::__for_each_n_it_serial(__first, __n,
                                               [&__f](_ForwardIterator __it) { __f(*__it); }); // calling serial version
 }
 
 template <class _RandomAccessIterator, class _DifferenceType, class _Function>
 _RandomAccessIterator
 __brick_walk1_n(_RandomAccessIterator __first, _DifferenceType __n, _Function __f,
                 /*vectorTag=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_walk_1(__first, __n, __f);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Function, class _IsVector>
 _ForwardIterator
 __pattern_walk1_n(_ExecutionPolicy&&, _ForwardIterator __first, _Size __n, _Function __f, _IsVector __is_vector,
                   /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_walk1_n(__first, __n, __f, __is_vector);
+    return __internal::__brick_walk1_n(__first, __n, __f, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Function, class _IsVector>
 _RandomAccessIterator
 __pattern_walk1_n(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _Size __n, _Function __f,
                   _IsVector __is_vector,
                   /*is_parallel=*/std::true_type)
 {
-    __pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __first + __n, __f, __is_vector, std::true_type());
+    __internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __first + __n, __f, __is_vector,
+                                std::true_type());
     return __first + __n;
 }
 #endif
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Brick>
 _ForwardIterator
 __pattern_walk_brick_n(_ExecutionPolicy&&, _ForwardIterator __first, _Size __n, _Brick __brick,
                        /*is_parallel=*/std::false_type) noexcept
 {
     return __brick(__first, __n);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Brick>
 _RandomAccessIterator
 __pattern_walk_brick_n(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _Size __n, _Brick __brick,
                        /*is_parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::__parallel_for(
             std::forward<_ExecutionPolicy>(__exec), __first, __first + __n,
             [__brick](_RandomAccessIterator __i, _RandomAccessIterator __j) { __brick(__i, __j - __i); });
         return __first + __n;
     });
 }
 #endif
 
__brick_walk2_n(_ForwardIterator1 __first1, _Size __n, _ForwardIterator2 __first2, _Function __f,
     return __unseq_backend::__simd_walk_2(__first1, __n, __first2, __f);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Function, class _IsVector>
 _ForwardIterator2
 __pattern_walk2(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                 _Function __f, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
 {
-    return __brick_walk2(__first1, __last1, __first2, __f, __is_vector);
+    return __internal::__brick_walk2(__first1, __last1, __first2, __f, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Function, class _IsVector>
 _ForwardIterator2
 __pattern_walk2(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                 _ForwardIterator2 __first2, _Function __f, _IsVector __is_vector, /*parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::__parallel_for(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
             [__f, __first1, __first2, __is_vector](_ForwardIterator1 __i, _ForwardIterator1 __j) {
-                __brick_walk2(__i, __j, __first2 + (__i - __first1), __f, __is_vector);
+                __internal::__brick_walk2(__i, __j, __first2 + (__i - __first1), __f, __is_vector);
             });
         return __first2 + (__last1 - __first1);
     });
 }
 #endif
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Function,
           class _IsVector>
 _ForwardIterator2
 __pattern_walk2_n(_ExecutionPolicy&&, _ForwardIterator1 __first1, _Size n, _ForwardIterator2 __first2, _Function f,
                   _IsVector is_vector, /*parallel=*/std::false_type) noexcept
 {
-    return __brick_walk2_n(__first1, n, __first2, f, is_vector);
+    return __internal::__brick_walk2_n(__first1, n, __first2, f, is_vector);
 }
 
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2,
           class _Function, class _IsVector>
 _RandomAccessIterator2
 __pattern_walk2_n(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _Size n, _RandomAccessIterator2 __first2,
                   _Function f, _IsVector is_vector, /*parallel=*/std::true_type)
 {
-    return __pattern_walk2(std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + n, __first2, f, is_vector,
-                           std::true_type());
+    return __internal::__pattern_walk2(std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + n, __first2, f,
+                                       is_vector, std::true_type());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Brick>
 _ForwardIterator2
 __pattern_walk2_brick(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                       _ForwardIterator2 __first2, _Brick __brick, /*parallel=*/std::false_type) noexcept
 {
     return __brick(__first1, __last1, __first2);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Brick>
 _RandomAccessIterator2
 __pattern_walk2_brick(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
                       _RandomAccessIterator2 __first2, _Brick __brick, /*parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::__parallel_for(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
             [__first1, __first2, __brick](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
                 __brick(__i, __j, __first2 + (__i - __first1));
             });
         return __first2 + (__last1 - __first1);
     });
 }
 #endif
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2, class _Brick>
 _RandomAccessIterator2
 __pattern_walk2_brick_n(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _Size __n,
                         _RandomAccessIterator2 __first2, _Brick __brick, /*parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::__parallel_for(
             std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
             [__first1, __first2, __brick](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
                 __brick(__i, __j - __i, __first2 + (__i - __first1));
             });
         return __first2 + __n;
     });
 }

 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator3,
           class _Function, class _IsVector>
 _ForwardIterator3
 __pattern_walk3(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                 _ForwardIterator3 __first3, _Function __f, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
 {
-    return __brick_walk3(__first1, __last1, __first2, __first3, __f, __is_vector);
+    return __internal::__brick_walk3(__first1, __last1, __first2, __first3, __f, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
           class _RandomAccessIterator3, class _Function, class _IsVector>
 _RandomAccessIterator3
 __pattern_walk3(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
                 _RandomAccessIterator2 __first2, _RandomAccessIterator3 __first3, _Function __f, _IsVector __is_vector,
                 /*parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::__parallel_for(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
             [__f, __first1, __first2, __first3, __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
-                __brick_walk3(__i, __j, __first2 + (__i - __first1), __first3 + (__i - __first1), __f, __is_vector);
+                __internal::__brick_walk3(__i, __j, __first2 + (__i - __first1), __first3 + (__i - __first1), __f,
+                                          __is_vector);
             });
         return __first3 + (__last1 - __first1);
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // equal

 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
           class _IsVector>
 bool
 __pattern_equal(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                 _BinaryPredicate __p, _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
 {
-    return __brick_equal(__first1, __last1, __first2, __p, __is_vector);
+    return __internal::__brick_equal(__first1, __last1, __first2, __p, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate,
           class _IsVector>
 bool
 __pattern_equal(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
                 _RandomAccessIterator2 __first2, _BinaryPredicate __p, _IsVector __is_vector,
                 /*is_parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
-        return !__parallel_or(
+    return __internal::__except_handler([&]() {
+        return !__internal::__parallel_or(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
             [__first1, __first2, __p, __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
-                return !__brick_equal(__i, __j, __first2 + (__i - __first1), __p, __is_vector);
+                return !__internal::__brick_equal(__i, __j, __first2 + (__i - __first1), __p, __is_vector);
             });
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // find_if
 //------------------------------------------------------------------------

 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
 _ForwardIterator
 __pattern_find_if(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred,
                   _IsVector __is_vector,
                   /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_find_if(__first, __last, __pred, __is_vector);
+    return __internal::__brick_find_if(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
 _ForwardIterator
 __pattern_find_if(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred,
                   _IsVector __is_vector,
                   /*is_parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
-        return __parallel_find(
+    return __internal::__except_handler([&]() {
+        return __internal::__parallel_find(
             std::forward<_ExecutionPolicy>(__exec), __first, __last,
             [__pred, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                return __brick_find_if(__i, __j, __pred, __is_vector);
+                return __internal::__brick_find_if(__i, __j, __pred, __is_vector);
             },
             std::less<typename std::iterator_traits<_ForwardIterator>::difference_type>(),
             /*is_first=*/true);
     });
 }
 #endif
 
 //------------------------------------------------------------------------
__find_subrange(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator1 __global_last,
     {
         return __last;
     }
 
     auto __cur = __last;
     while (__first != __last && (__global_last - __first >= __n2))
     {
         // find position of *s_first in [first, last) (it can be start of subsequence)
-        __first = __brick_find_if(__first, __last,
-                                  __equal_value_by_pred<_ValueType, _BinaryPredicate>(*__s_first, __pred), __is_vector);
+        __first = __internal::__brick_find_if(
+            __first, __last, __equal_value_by_pred<_ValueType, _BinaryPredicate>(*__s_first, __pred), __is_vector);
 
         // if position that was found previously is the start of subsequence
         // then we can exit the loop (b_first == true) or keep the position
         // (b_first == false)
         if (__first != __last && (__global_last - __first >= __n2) &&
-            __brick_equal(__s_first + 1, __s_last, __first + 1, __pred, __is_vector))
+            __internal::__brick_equal(__s_first + 1, __s_last, __first + 1, __pred, __is_vector))
         {
             if (__b_first)
             {
                 return __first;
             }
             else
             {
                 __cur = __first;
__find_subrange(_RandomAccessIterator __first, _RandomAccessIterator __last, _RandomAccessIterator __global_last,
     {
         return __last; // According to the standard last shall be returned when count < 1
     }
 
     auto __n = __global_last - __first;
     auto __unary_pred = __equal_value_by_pred<_Tp, _BinaryPredicate>(__value, __pred);
     while (__first != __last && (__global_last - __first >= __count))
     {
-        __first = __brick_find_if(__first, __last, __unary_pred, __is_vector);
+        __first = __internal::__brick_find_if(__first, __last, __unary_pred, __is_vector);
 
         // check that all of elements in [first+1, first+count) equal to value
         if (__first != __last && (__global_last - __first >= __count) &&
-            !__brick_any_of(__first + 1, __first + __count, __not_pred<decltype(__unary_pred)>(__unary_pred),
-                            __is_vector))
+            !__internal::__brick_any_of(__first + 1, __first + __count,
+                                        __not_pred<decltype(__unary_pred)>(__unary_pred), __is_vector))
         {
             return __first;
         }
         else if (__first == __last)
         {
             break;
         }
         else

 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
           class _IsVector>
 _ForwardIterator1
 __pattern_find_end(_ExecutionPolicy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
                    _ForwardIterator2 __s_last, _BinaryPredicate __pred, _IsVector __is_vector,
                    /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_find_end(__first, __last, __s_first, __s_last, __pred, __is_vector);
+    return __internal::__brick_find_end(__first, __last, __s_first, __s_last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
           class _IsVector>
 _ForwardIterator1
 __pattern_find_end(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last,
                    _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred,
                    _IsVector __is_vector, /*is_parallel=*/std::true_type) noexcept
 {
     if (__last - __first == __s_last - __s_first)
     {
-        const bool __res = __pattern_equal(std::forward<_ExecutionPolicy>(__exec), __first, __last, __s_first, __pred,
-                                           __is_vector, std::true_type());
+        const bool __res = __internal::__pattern_equal(std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                       __s_first, __pred, __is_vector, std::true_type());
         return __res ? __first : __last;
     }
     else
     {
-        return __except_handler([&]() {
-            return __parallel_find(
+        return __internal::__except_handler([&]() {
+            return __internal::__parallel_find(
                 std::forward<_ExecutionPolicy>(__exec), __first, __last,
                 [__last, __s_first, __s_last, __pred, __is_vector](_ForwardIterator1 __i, _ForwardIterator1 __j) {
-                    return __find_subrange(__i, __j, __last, __s_first, __s_last, __pred, false, __is_vector);
+                    return __internal::__find_subrange(__i, __j, __last, __s_first, __s_last, __pred, false,
+                                                       __is_vector);
                 },
                 std::greater<typename std::iterator_traits<_ForwardIterator1>::difference_type>(), /*is_first=*/false);
         });
     }
 }
 #endif
 
 //------------------------------------------------------------------------

 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
           class _IsVector>
 _ForwardIterator1
 __pattern_find_first_of(_ExecutionPolicy&&, _ForwardIterator1 __first, _ForwardIterator1 __last,
                         _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred,
                         _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_find_first_of(__first, __last, __s_first, __s_last, __pred, __is_vector);
+    return __internal::__brick_find_first_of(__first, __last, __s_first, __s_last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
           class _IsVector>
 _ForwardIterator1
 __pattern_find_first_of(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last,
                         _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred,
                         _IsVector __is_vector, /*is_parallel=*/std::true_type) noexcept
 {
-    return __except_handler([&]() {
-        return __parallel_find(
+    return __internal::__except_handler([&]() {
+        return __internal::__parallel_find(
             std::forward<_ExecutionPolicy>(__exec), __first, __last,
             [__s_first, __s_last, __pred, __is_vector](_ForwardIterator1 __i, _ForwardIterator1 __j) {
-                return __brick_find_first_of(__i, __j, __s_first, __s_last, __pred, __is_vector);
+                return __internal::__brick_find_first_of(__i, __j, __s_first, __s_last, __pred, __is_vector);
             },
             std::less<typename std::iterator_traits<_ForwardIterator1>::difference_type>(), /*is_first=*/true);
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // search
 //------------------------------------------------------------------------
 template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _ForwardIterator1
 __brick_search(_ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
                _ForwardIterator2 __s_last, _BinaryPredicate __pred, /*vector=*/std::false_type) noexcept
 {
     return std::search(__first, __last, __s_first, __s_last, __pred);
 }
 
 template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _ForwardIterator1
 __brick_search(_ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
                _ForwardIterator2 __s_last, _BinaryPredicate __pred, /*vector=*/std::true_type) noexcept
 {
-    return __find_subrange(__first, __last, __last, __s_first, __s_last, __pred, true, std::true_type());
+    return __internal::__find_subrange(__first, __last, __last, __s_first, __s_last, __pred, true, std::true_type());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
           class _IsVector>
 _ForwardIterator1
 __pattern_search(_ExecutionPolicy&&, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
                  _ForwardIterator2 __s_last, _BinaryPredicate __pred, _IsVector __is_vector,
                  /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_search(__first, __last, __s_first, __s_last, __pred, __is_vector);
+    return __internal::__brick_search(__first, __last, __s_first, __s_last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
           class _IsVector>
 _ForwardIterator1
 __pattern_search(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last,
                  _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred,
                  _IsVector __is_vector,
                  /*is_parallel=*/std::true_type) noexcept
 {
     if (__last - __first == __s_last - __s_first)
     {
-        const bool __res = __pattern_equal(std::forward<_ExecutionPolicy>(__exec), __first, __last, __s_first, __pred,
-                                           __is_vector, std::true_type());
+        const bool __res = __internal::__pattern_equal(std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                       __s_first, __pred, __is_vector, std::true_type());
         return __res ? __first : __last;
     }
     else
     {
-        return __except_handler([&]() {
-            return __parallel_find(
+        return __internal::__except_handler([&]() {
+            return __internal::__parallel_find(
                 std::forward<_ExecutionPolicy>(__exec), __first, __last,
                 [__last, __s_first, __s_last, __pred, __is_vector](_ForwardIterator1 __i, _ForwardIterator1 __j) {
-                    return __find_subrange(__i, __j, __last, __s_first, __s_last, __pred, true, __is_vector);
+                    return __internal::__find_subrange(__i, __j, __last, __s_first, __s_last, __pred, true,
+                                                       __is_vector);
                 },
                 std::less<typename std::iterator_traits<_ForwardIterator1>::difference_type>(), /*is_first=*/true);
         });
     }
 }
 #endif
 
 //------------------------------------------------------------------------
 // search_n
 //------------------------------------------------------------------------
 template <class _ForwardIterator, class _Size, class _Tp, class _BinaryPredicate>
 _ForwardIterator
 __brick_search_n(_ForwardIterator __first, _ForwardIterator __last, _Size __count, const _Tp& __value,
                  _BinaryPredicate __pred, /*vector=*/std::false_type) noexcept
 {
     return std::search_n(__first, __last, __count, __value, __pred);
 }
 
 template <class _ForwardIterator, class _Size, class _Tp, class _BinaryPredicate>
 _ForwardIterator
 __brick_search_n(_ForwardIterator __first, _ForwardIterator __last, _Size __count, const _Tp& __value,
                  _BinaryPredicate __pred, /*vector=*/std::true_type) noexcept
 {
-    return __find_subrange(__first, __last, __last, __count, __value, __pred, std::true_type());
+    return __internal::__find_subrange(__first, __last, __last, __count, __value, __pred, std::true_type());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Tp, class _BinaryPredicate,
           class _IsVector>
 _ForwardIterator
 __pattern_search_n(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Size __count,
                    const _Tp& __value, _BinaryPredicate __pred, _IsVector __is_vector,
                    /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_search_n(__first, __last, __count, __value, __pred, __is_vector);
+    return __internal::__brick_search_n(__first, __last, __count, __value, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Tp, class _BinaryPredicate,
           class _IsVector>
 _RandomAccessIterator
 __pattern_search_n(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                    _Size __count, const _Tp& __value, _BinaryPredicate __pred, _IsVector __is_vector,
                    /*is_parallel=*/std::true_type) noexcept
 {
     if (__last - __first == __count)
     {
-        const bool __result = !__pattern_any_of(
+        const bool __result = !__internal::__pattern_any_of(
             std::forward<_ExecutionPolicy>(__exec), __first, __last,
             [&__value, &__pred](const _Tp& __val) { return !__pred(__val, __value); }, __is_vector,
             /*is_parallel*/ std::true_type());
         return __result ? __first : __last;
     }
     else
     {
-        return __except_handler([&__exec, __first, __last, __count, &__value, __pred, __is_vector]() {
-            return __parallel_find(
+        return __internal::__except_handler([&__exec, __first, __last, __count, &__value, __pred, __is_vector]() {
+            return __internal::__parallel_find(
                 std::forward<_ExecutionPolicy>(__exec), __first, __last,
                 [__last, __count, &__value, __pred, __is_vector](_RandomAccessIterator __i, _RandomAccessIterator __j) {
-                    return __find_subrange(__i, __j, __last, __count, __value, __pred, __is_vector);
+                    return __internal::__find_subrange(__i, __j, __last, __count, __value, __pred, __is_vector);
                 },
                 std::less<typename std::iterator_traits<_RandomAccessIterator>::difference_type>(), /*is_first=*/true);
         });
     }
 }
 #endif
 
 //------------------------------------------------------------------------

 template <class _ForwardIterator, class _OutputIterator, class _Assigner>
 void
 __brick_copy_by_mask(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
                      bool* __restrict __mask, _Assigner __assigner, /*vector=*/std::true_type) noexcept
 {
 #if (__PSTL_MONOTONIC_PRESENT)
     __unseq_backend::__simd_copy_by_mask(__first, __last - __first, __result, __mask, __assigner);
 #else
-    __brick_copy_by_mask(__first, __last, __result, __mask, __assigner, std::false_type());
+    __internal::__brick_copy_by_mask(__first, __last, __result, __mask, __assigner, std::false_type());
 #endif
 }
 
 template <class _ForwardIterator, class _OutputIterator1, class _OutputIterator2>
 void
 __brick_partition_by_mask(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator1 __out_true,
                           _OutputIterator2 __out_false, bool* __mask, /*vector=*/std::false_type) noexcept
 {

 template <class _RandomAccessIterator, class _OutputIterator1, class _OutputIterator2>
 void
 __brick_partition_by_mask(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator1 __out_true,
                           _OutputIterator2 __out_false, bool* __mask, /*vector=*/std::true_type) noexcept
 {
 #if (__PSTL_MONOTONIC_PRESENT)
     __unseq_backend::__simd_partition_by_mask(__first, __last - __first, __out_true, __out_false, __mask);
 #else
-    __brick_partition_by_mask(__first, __last, __out_true, __out_false, __mask, std::false_type());
+    __internal::__brick_partition_by_mask(__first, __last, __out_true, __out_false, __mask, std::false_type());
 #endif
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryPredicate, class _IsVector>
 _OutputIterator
 __pattern_copy_if(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
                   _UnaryPredicate __pred, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
 {
-    return __brick_copy_if(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_copy_if(__first, __last, __result, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryPredicate,
           class _IsVector>
 _OutputIterator
 __pattern_copy_if(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                   _OutputIterator __result, _UnaryPredicate __pred, _IsVector __is_vector, /*parallel=*/std::true_type)
 {
     typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
     const _DifferenceType __n = __last - __first;
     if (_DifferenceType(1) < __n)
     {
         __par_backend::__buffer<bool> __mask_buf(__n);
-        return __except_handler([&__exec, __n, __first, __result, __is_vector, __pred, &__mask_buf]() {
+        return __internal::__except_handler([&__exec, __n, __first, __result, __is_vector, __pred, &__mask_buf]() {
             bool* __mask = __mask_buf.get();
             _DifferenceType __m{};
             __par_backend::parallel_strict_scan(
                 std::forward<_ExecutionPolicy>(__exec), __n, _DifferenceType(0),
                 [=](_DifferenceType __i, _DifferenceType __len) { // Reduce
-                    return __brick_calc_mask_1<_DifferenceType>(__first + __i, __first + (__i + __len), __mask + __i,
-                                                                __pred, __is_vector)
+                    return __internal::__brick_calc_mask_1<_DifferenceType>(__first + __i, __first + (__i + __len),
+                                                                            __mask + __i, __pred, __is_vector)
                         .first;
                 },
                 std::plus<_DifferenceType>(),                                                // Combine
                 [=](_DifferenceType __i, _DifferenceType __len, _DifferenceType __initial) { // Scan
-                    __brick_copy_by_mask(
+                    __internal::__brick_copy_by_mask(
                         __first + __i, __first + (__i + __len), __result + __initial, __mask + __i,
                         [](_RandomAccessIterator __x, _OutputIterator __z) { *__z = *__x; }, __is_vector);
                 },
                 [&__m](_DifferenceType __total) { __m = __total; });
             return __result + __m;
         });
     }
     // trivial sequence - use serial algorithm
-    return __brick_copy_if(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_copy_if(__first, __last, __result, __pred, __is_vector);
 }
 #endif
 
 //------------------------------------------------------------------------
 // count
 //------------------------------------------------------------------------
 template <class _ForwardIterator, class _Predicate>
 typename std::iterator_traits<_ForwardIterator>::difference_type
__brick_count(_ForwardIterator __first, _ForwardIterator __last, _Predicate __pred,
     return std::count_if(__first, __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
 typename std::iterator_traits<_ForwardIterator>::difference_type
 __pattern_count(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred,
                 /* is_parallel */ std::false_type, _IsVector __is_vector) noexcept
 {
-    return __brick_count(__first, __last, __pred, __is_vector);
+    return __internal::__brick_count(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
 typename std::iterator_traits<_ForwardIterator>::difference_type
 __pattern_count(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred,
                 /* is_parallel */ std::true_type, _IsVector __is_vector)
 {
     typedef typename std::iterator_traits<_ForwardIterator>::difference_type _SizeType;
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         return __par_backend::__parallel_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first, __last, _SizeType(0),
             [__pred, __is_vector](_ForwardIterator __begin, _ForwardIterator __end, _SizeType __value) -> _SizeType {
-                return __value + __brick_count(__begin, __end, __pred, __is_vector);
+                return __value + __internal::__brick_count(__begin, __end, __pred, __is_vector);
             },
             std::plus<_SizeType>());
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // unique
__brick_unique(_ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
     return std::unique(__first, __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
 _ForwardIterator
 __pattern_unique(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
                  _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_unique(__first, __last, __pred, __is_vector);
+    return __internal::__brick_unique(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 // That function is shared between two algorithms - remove_if (__pattern_remove_if) and unique (pattern unique). But a mask calculation is different.
 // So, a caller passes _CalcMask brick into remove_elements.
 template <class _ExecutionPolicy, class _ForwardIterator, class _CalcMask, class _IsVector>
 _ForwardIterator
-remove_elements(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _CalcMask __calc_mask,
+__remove_elements(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _CalcMask __calc_mask,
                   _IsVector __is_vector)
 {
     typedef typename std::iterator_traits<_ForwardIterator>::difference_type _DifferenceType;
     typedef typename std::iterator_traits<_ForwardIterator>::value_type _Tp;
     _DifferenceType __n = __last - __first;
     __par_backend::__buffer<bool> __mask_buf(__n);
     // 1. find a first iterator that should be removed
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         bool* __mask = __mask_buf.get();
         _DifferenceType __min = __par_backend::__parallel_reduce(
             std::forward<_ExecutionPolicy>(__exec), _DifferenceType(0), __n, __n,
             [__first, __mask, &__calc_mask, __is_vector](_DifferenceType __i, _DifferenceType __j,
                                                          _DifferenceType __local_min) -> _DifferenceType {
                 // Create mask
                 __calc_mask(__mask + __i, __mask + __j, __first + __i);
 
                 // if minimum was found in a previous range we shouldn't do anymore
                 if (__local_min < __i)
                 {
                     return __local_min;
                 }
                 // find first iterator that should be removed
-                bool* __result = __brick_find_if(
-                    __mask + __i, __mask + __j, [](bool __val) { return !__val; }, __is_vector);
+                bool* __result = __internal::__brick_find_if(__mask + __i, __mask + __j,
+                                                             [](bool __val) { return !__val; }, __is_vector);
                 if (__result - __mask == __j)
                 {
                     return __local_min;
                 }
                 return std::min(__local_min, _DifferenceType(__result - __mask));
             },
             [](_DifferenceType __local_min1, _DifferenceType __local_min2) -> _DifferenceType {
                 return std::min(__local_min1, __local_min2);
return __internal::__except_handler([&]() {
         __par_backend::__buffer<_Tp> __buf(__n);
         _Tp* __result = __buf.get();
         __mask += __min;
         _DifferenceType __m{};
         // 2. Elements that doesn't satisfy pred are moved to result
         __par_backend::parallel_strict_scan(
             std::forward<_ExecutionPolicy>(__exec), __n, _DifferenceType(0),
             [__mask, __is_vector](_DifferenceType __i, _DifferenceType __len) {
-                return __brick_count(
-                    __mask + __i, __mask + __i + __len, [](bool __val) { return __val; }, __is_vector);
+                return __internal::__brick_count(__mask + __i, __mask + __i + __len, [](bool __val) { return __val; },
+                                                 __is_vector);
             },
             std::plus<_DifferenceType>(),
             [=](_DifferenceType __i, _DifferenceType __len, _DifferenceType __initial) {
-                __brick_copy_by_mask(
+                __internal::__brick_copy_by_mask(
                     __first + __i, __first + __i + __len, __result + __initial, __mask + __i,
                     [](_ForwardIterator __x, _Tp* __z) {
-                        __invoke_if_else(
-                            std::is_trivial<_Tp>(), [&]() { *__z = std::move(*__x); },
+                        __internal::__invoke_if_else(std::is_trivial<_Tp>(), [&]() { *__z = std::move(*__x); },
                                                      [&]() { ::new (std::addressof(*__z)) _Tp(std::move(*__x)); });
                     },
                     __is_vector);
             },
             [&__m](_DifferenceType __total) { __m = __total; });
 
         // 3. Elements from result are moved to [first, last)
         __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __result, __result + __m,
                                       [__result, __first, __is_vector](_Tp* __i, _Tp* __j) {
-                                          __brick_move(__i, __j, __first + (__i - __result), __is_vector);
+                                          __internal::__brick_move(__i, __j, __first + (__i - __result), __is_vector);
                                       });
         return __first + __m;
     });
 }
 #endif
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
 _ForwardIterator
 __pattern_unique(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
                  _IsVector __is_vector, /*is_parallel=*/std::true_type) noexcept
 {
     typedef typename std::iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
     if (__first == __last)
     {
         return __last;
     }
     if (__first + 1 == __last || __first + 2 == __last)
     {
         // Trivial sequence - use serial algorithm
-        return __brick_unique(__first, __last, __pred, __is_vector);
+        return __internal::__brick_unique(__first, __last, __pred, __is_vector);
     }
-    return remove_elements(
+    return __internal::__remove_elements(
         std::forward<_ExecutionPolicy>(__exec), ++__first, __last,
         [&__pred, __is_vector](bool* __b, bool* __e, _ForwardIterator __it) {
-            __brick_walk3(
+            __internal::__brick_walk3(
                 __b, __e, __it - 1, __it,
                 [&__pred](bool& __x, _ReferenceType __y, _ReferenceType __z) { __x = !__pred(__y, __z); }, __is_vector);
         },
         __is_vector);
 }
 #endif
 
 //------------------------------------------------------------------------

 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryPredicate,
           class _IsVector>
 _OutputIterator
 __pattern_unique_copy(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
                       _BinaryPredicate __pred, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
 {
-    return __brick_unique_copy(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_unique_copy(__first, __last, __result, __pred, __is_vector);
 }
 
 template <class _DifferenceType, class _RandomAccessIterator, class _BinaryPredicate>
 _DifferenceType
 __brick_calc_mask_2(_RandomAccessIterator __first, _RandomAccessIterator __last, bool* __restrict __mask,
                     _BinaryPredicate __pred, /*vector=*/std::false_type) noexcept
 {
     _DifferenceType __count = 0;

 {
     typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
     const _DifferenceType __n = __last - __first;
     if (_DifferenceType(2) < __n)
     {
         __par_backend::__buffer<bool> __mask_buf(__n);
         if (_DifferenceType(2) < __n)
         {
-            return __except_handler([&__exec, __n, __first, __result, __pred, __is_vector, &__mask_buf]() {
+            return __internal::__except_handler([&__exec, __n, __first, __result, __pred, __is_vector, &__mask_buf]() {
                 bool* __mask = __mask_buf.get();
                 _DifferenceType __m{};
                 __par_backend::parallel_strict_scan(
                     std::forward<_ExecutionPolicy>(__exec), __n, _DifferenceType(0),
                     [=](_DifferenceType __i, _DifferenceType __len) -> _DifferenceType { // Reduce
                         _DifferenceType __extra = 0;
                         if (__i == 0)
                         {
                             // Special boundary case
                             __mask[__i] = true;
                             if (--__len == 0)
                                 return 1;
                             ++__i;
                             ++__extra;
                         }
-                        return __brick_calc_mask_2<_DifferenceType>(__first + __i, __first + (__i + __len),
+                        return __internal::__brick_calc_mask_2<_DifferenceType>(__first + __i, __first + (__i + __len),
                                                                                 __mask + __i, __pred, __is_vector) +
                                __extra;
                     },
                     std::plus<_DifferenceType>(),                                                // Combine
                     [=](_DifferenceType __i, _DifferenceType __len, _DifferenceType __initial) { // Scan
                         // Phase 2 is same as for __pattern_copy_if
-                        __brick_copy_by_mask(
+                        __internal::__brick_copy_by_mask(
                             __first + __i, __first + (__i + __len), __result + __initial, __mask + __i,
                             [](_RandomAccessIterator __x, _OutputIterator __z) { *__z = *__x; }, __is_vector);
                     },
                     [&__m](_DifferenceType __total) { __m = __total; });
                 return __result + __m;
             });
         }
     }
     // trivial sequence - use serial algorithm
-    return __brick_unique_copy(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_unique_copy(__first, __last, __result, __pred, __is_vector);
 }
 #endif
 
 //------------------------------------------------------------------------
 // reverse
 //------------------------------------------------------------------------
 template <class _BidirectionalIterator>
 void

 }
 
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _IsVector>
 void
 __pattern_reverse(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
                   _IsVector _is_vector,
                   /*is_parallel=*/std::false_type) noexcept
 {
-    __brick_reverse(__first, __last, _is_vector);
+    __internal::__brick_reverse(__first, __last, _is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _IsVector>
 void
 __pattern_reverse(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __last,
                   _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     __par_backend::__parallel_for(
         std::forward<_ExecutionPolicy>(__exec), __first, __first + (__last - __first) / 2,
         [__is_vector, __first, __last](_BidirectionalIterator __inner_first, _BidirectionalIterator __inner_last) {
-            __brick_reverse(__inner_first, __inner_last, __last - (__inner_first - __first), __is_vector);
+            __internal::__brick_reverse(__inner_first, __inner_last, __last - (__inner_first - __first), __is_vector);
         });
 }
 #endif
 
 //------------------------------------------------------------------------
 // reverse_copy
 //------------------------------------------------------------------------
 
return __unseq_backend::__simd_walk_2(std::reverse_iterator<_BidirectionalIterator>(__last), __last - __first,
                                           __d_first, [](_ReferenceType1 __x, _ReferenceType2 __y) { __y = __x; });
 }
 
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _OutputIterator, class _IsVector>
 _OutputIterator
 __pattern_reverse_copy(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
                        _OutputIterator __d_first, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_reverse_copy(__first, __last, __d_first, __is_vector);
+    return __internal::__brick_reverse_copy(__first, __last, __d_first, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _OutputIterator, class _IsVector>
 _OutputIterator
 __pattern_reverse_copy(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __last,
                        _OutputIterator __d_first, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     auto __len = __last - __first;
     __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                   [__is_vector, __first, __len, __d_first](_BidirectionalIterator __inner_first,
                                                                            _BidirectionalIterator __inner_last) {
-                                      __brick_reverse_copy(__inner_first, __inner_last,
-                                                           __d_first + (__len - (__inner_last - __first)), __is_vector);
+                                      __internal::__brick_reverse_copy(__inner_first, __inner_last,
+                                                                       __d_first + (__len - (__inner_last - __first)),
+                                                                       __is_vector);
                                   });
     return __d_first + __len;
 }
 #endif
 
 //------------------------------------------------------------------------
 // rotate
 //------------------------------------------------------------------------
__brick_rotate(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
     return __ret;
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _IsVector>
 _ForwardIterator
 __pattern_rotate(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
                  _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_rotate(__first, __middle, __last, __is_vector);
+    return __internal::__brick_rotate(__first, __middle, __last, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _IsVector>
 _ForwardIterator
 __pattern_rotate(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __middle,
                  _ForwardIterator __last, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     typedef typename std::iterator_traits<_ForwardIterator>::value_type _Tp;
     auto __n = __last - __first;
     auto __m = __middle - __first;
     if (__m <= __n / 2)
     {
         __par_backend::__buffer<_Tp> __buf(__n - __m);
-        return __except_handler([&__exec, __n, __m, __first, __middle, __last, __is_vector, &__buf]() {
+        return __internal::__except_handler([&__exec, __n, __m, __first, __middle, __last, __is_vector, &__buf]() {
             _Tp* __result = __buf.get();
             __par_backend::__parallel_for(
                 std::forward<_ExecutionPolicy>(__exec), __middle, __last,
                 [__middle, __result, __is_vector](_ForwardIterator __b, _ForwardIterator __e) {
-                    __brick_uninitialized_move(__b, __e, __result + (__b - __middle), __is_vector);
+                    __internal::__brick_uninitialized_move(__b, __e, __result + (__b - __middle), __is_vector);
                 });
 
             __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __middle,
                                           [__last, __middle, __is_vector](_ForwardIterator __b, _ForwardIterator __e) {
-                                              __brick_move(__b, __e, __b + (__last - __middle), __is_vector);
+                                              __internal::__brick_move(__b, __e, __b + (__last - __middle),
+                                                                       __is_vector);
                                           });
 
             __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __result, __result + (__n - __m),
                                           [__first, __result, __is_vector](_Tp* __b, _Tp* __e) {
-                                              __brick_move(__b, __e, __first + (__b - __result), __is_vector);
+                                              __internal::__brick_move(__b, __e, __first + (__b - __result),
+                                                                       __is_vector);
                                           });
 
             return __first + (__last - __middle);
         });
     }
     else
     {
         __par_backend::__buffer<_Tp> __buf(__m);
-        return __except_handler([&__exec, __n, __m, __first, __middle, __last, __is_vector, &__buf]() {
+        return __internal::__except_handler([&__exec, __n, __m, __first, __middle, __last, __is_vector, &__buf]() {
             _Tp* __result = __buf.get();
             __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __middle,
                                           [__first, __result, __is_vector](_ForwardIterator __b, _ForwardIterator __e) {
-                                              __brick_uninitialized_move(__b, __e, __result + (__b - __first),
-                                                                         __is_vector);
+                                              __internal::__brick_uninitialized_move(
+                                                  __b, __e, __result + (__b - __first), __is_vector);
                                           });
 
             __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __middle, __last,
                                           [__first, __middle, __is_vector](_ForwardIterator __b, _ForwardIterator __e) {
-                                              __brick_move(__b, __e, __first + (__b - __middle), __is_vector);
+                                              __internal::__brick_move(__b, __e, __first + (__b - __middle),
+                                                                       __is_vector);
                                           });
 
             __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __result, __result + __m,
                                           [__n, __m, __first, __result, __is_vector](_Tp* __b, _Tp* __e) {
-                                              __brick_move(__b, __e, __first + ((__n - __m) + (__b - __result)),
-                                                           __is_vector);
+                                              __internal::__brick_move(
+                                                  __b, __e, __first + ((__n - __m) + (__b - __result)), __is_vector);
                                           });
 
             return __first + (__last - __middle);
         });
     }
 }
 #endif
 
__brick_rotate_copy(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
     return std::rotate_copy(__first, __middle, __last, __result);
 }
 
 template <class _ForwardIterator, class _OutputIterator>
 _OutputIterator
 __brick_rotate_copy(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
                     _OutputIterator __result, /*__is_vector=*/std::true_type) noexcept
 {
-    _OutputIterator __res = __brick_copy(__middle, __last, __result, std::true_type());
-    return __brick_copy(__first, __middle, __res, std::true_type());
+    _OutputIterator __res = __internal::__brick_copy(__middle, __last, __result, std::true_type());
+    return __internal::__brick_copy(__first, __middle, __res, std::true_type());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _IsVector>
 _OutputIterator
 __pattern_rotate_copy(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
                       _OutputIterator __result, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_rotate_copy(__first, __middle, __last, __result, __is_vector);
+    return __internal::__brick_rotate_copy(__first, __middle, __last, __result, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _IsVector>
 _OutputIterator
 __pattern_rotate_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __middle,
                       _ForwardIterator __last, _OutputIterator __result, _IsVector __is_vector,
                       /*is_parallel=*/std::true_type)
 {
     __par_backend::__parallel_for(
         std::forward<_ExecutionPolicy>(__exec), __first, __last,
         [__first, __last, __middle, __result, __is_vector](_ForwardIterator __b, _ForwardIterator __e) {
             if (__b > __middle)
             {
-                __brick_copy(__b, __e, __result + (__b - __middle), __is_vector);
+                __internal::__brick_copy(__b, __e, __result + (__b - __middle), __is_vector);
             }
             else
             {
                 _OutputIterator __new_result = __result + ((__last - __middle) + (__b - __first));
                 if (__e < __middle)
                 {
-                    __brick_copy(__b, __e, __new_result, __is_vector);
+                    __internal::__brick_copy(__b, __e, __new_result, __is_vector);
                 }
                 else
                 {
-                    __brick_copy(__b, __middle, __new_result, __is_vector);
-                    __brick_copy(__middle, __e, __result, __is_vector);
+                    __internal::__brick_copy(__b, __middle, __new_result, __is_vector);
+                    __internal::__brick_copy(__middle, __e, __result, __is_vector);
                 }
             }
         });
     return __result + (__last - __first);
 }
 #endif
 
 //------------------------------------------------------------------------
__brick_is_partitioned(_ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
     }
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
 bool
 __pattern_is_partitioned(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
                          _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_is_partitioned(__first, __last, __pred, __is_vector);
+    return __internal::__brick_is_partitioned(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
 bool
 __pattern_is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
                          _UnaryPredicate __pred, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     if (__first == __last)
     {
         return true;
     }
     else
     {
-        return __except_handler([&]() {
+        return __internal::__except_handler([&]() {
             // State of current range:
             // broken     - current range is not partitioned by pred
             // all_true   - all elements in current range satisfy pred
             // all_false  - all elements in current range don't satisfy pred
             // true_false - elements satisfy pred are placed before elements that don't satisfy pred
             enum _ReduceType
             {
                 __not_init = -1,
else
                         return __broken;
                     }
                     _ReduceType __res = __not_init;
                     // if first element satisfy pred
                     if (__pred(*__i))
                     {
                         // find first element that don't satisfy pred
                         _ForwardIterator __x =
-                            __brick_find_if(__i + 1, __j, __not_pred<_UnaryPredicate>(__pred), __is_vector);
+                            __internal::__brick_find_if(__i + 1, __j, __not_pred<_UnaryPredicate>(__pred), __is_vector);
                         if (__x != __j)
                         {
                             // find first element after "x" that satisfy pred
-                            _ForwardIterator __y = __brick_find_if(__x + 1, __j, __pred, __is_vector);
+                            _ForwardIterator __y = __internal::__brick_find_if(__x + 1, __j, __pred, __is_vector);
                             // if it was found then range isn't partitioned by pred
                             if (__y != __j)
                             {
                                 return __broken;
                             }
                             else
                             {
                                 __res = __true_false;
                             }
                         }
                         else
                         {
                             __res = __all_true;
                         }
                     }
                     else
                     { // if first element doesn't satisfy pred
                         // then we should find the first element that satisfy pred.
                         // If we found it then range isn't partitioned by pred
-                        if (__brick_find_if(__i + 1, __j, __pred, __is_vector) != __j)
+                        if (__internal::__brick_find_if(__i + 1, __j, __pred, __is_vector) != __j)
                         {
                             return __broken;
                         }
                         else
                         {
                             __res = __all_false;
                         }
                     }
__brick_partition(_ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
     return std::partition(__first, __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
 _ForwardIterator
 __pattern_partition(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
                     _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_partition(__first, __last, __pred, __is_vector);
+    return __internal::__brick_partition(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
 _ForwardIterator
 __pattern_partition(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
                     _UnaryPredicate __pred, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
 
     // partitioned range: elements before pivot satisfy pred (true part),
     //                    elements after pivot don't satisfy pred (false part)
     struct _PartitionRange
     {
         _ForwardIterator __begin;
         _ForwardIterator __pivot;
         _ForwardIterator __end;
     };
 
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         _PartitionRange __init{__last, __last, __last};
 
         // lambda for merging two partitioned ranges to one partitioned range
         auto __reductor = [&__exec, __is_vector](_PartitionRange __val1, _PartitionRange __val2) -> _PartitionRange {
             auto __size1 = __val1.__end - __val1.__pivot;
             auto __size2 = __val2.__pivot - __val2.__begin;
             auto __new_begin = __val2.__begin - (__val1.__end - __val1.__begin);
 
             // if all elements in left range satisfy pred then we can move new pivot to pivot of right range
             if (__val1.__end == __val1.__pivot)
             {
                 return {__new_begin, __val2.__pivot, __val2.__end};
             }
             // if true part of right range greater than false part of left range
             // then we should swap the false part of left range and last part of true part of right range
             else if (__size2 > __size1)
             {
                 __par_backend::__parallel_for(
                     std::forward<_ExecutionPolicy>(__exec), __val1.__pivot, __val1.__pivot + __size1,
                     [__val1, __val2, __size1, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                        __brick_swap_ranges(__i, __j, (__val2.__pivot - __size1) + (__i - __val1.__pivot), __is_vector);
+                        __internal::__brick_swap_ranges(__i, __j, (__val2.__pivot - __size1) + (__i - __val1.__pivot),
+                                                        __is_vector);
                     });
                 return {__new_begin, __val2.__pivot - __size1, __val2.__end};
             }
             // else we should swap the first part of false part of left range and true part of right range
             else
             {
                 __par_backend::__parallel_for(
                     std::forward<_ExecutionPolicy>(__exec), __val1.__pivot, __val1.__pivot + __size2,
                     [__val1, __val2, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                        __brick_swap_ranges(__i, __j, __val2.__begin + (__i - __val1.__pivot), __is_vector);
+                        __internal::__brick_swap_ranges(__i, __j, __val2.__begin + (__i - __val1.__pivot), __is_vector);
                     });
                 return {__new_begin, __val1.__pivot + __size2, __val2.__end};
             }
         };
 
         _PartitionRange __result = __par_backend::__parallel_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first, __last, __init,
             [__pred, __is_vector, __reductor](_ForwardIterator __i, _ForwardIterator __j,
                                               _PartitionRange __value) -> _PartitionRange {
                 //1. serial partition
-                _ForwardIterator __pivot = __brick_partition(__i, __j, __pred, __is_vector);
+                _ForwardIterator __pivot = __internal::__brick_partition(__i, __j, __pred, __is_vector);
 
                 // 2. merging of two ranges (left and right respectively)
                 return __reductor(__value, {__i, __pivot, __j});
             },
             __reductor);
         return __result.__pivot;
     });
 }

 }
 
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate, class _IsVector>
 _BidirectionalIterator
 __pattern_stable_partition(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
                            _UnaryPredicate __pred, _IsVector __is_vector,
                            /*is_parallelization=*/std::false_type) noexcept
 {
-    return __brick_stable_partition(__first, __last, __pred, __is_vector);
+    return __internal::__brick_stable_partition(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate, class _IsVector>
 _BidirectionalIterator
 __pattern_stable_partition(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __last,
                            _UnaryPredicate __pred, _IsVector __is_vector,
                            /*is_parallelization=*/std::true_type) noexcept
 {
     // partitioned range: elements before pivot satisfy pred (true part),
     //                    elements after pivot don't satisfy pred (false part)
     struct _PartitionRange
     {
         _BidirectionalIterator __begin;
         _BidirectionalIterator __pivot;
         _BidirectionalIterator __end;
     };
 
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         _PartitionRange __init{__last, __last, __last};
 
         // lambda for merging two partitioned ranges to one partitioned range
         auto __reductor = [__is_vector](_PartitionRange __val1, _PartitionRange __val2) -> _PartitionRange {
             auto __size1 = __val1.__end - __val1.__pivot;
             auto __new_begin = __val2.__begin - (__val1.__end - __val1.__begin);
 
             // if all elements in left range satisfy pred then we can move new pivot to pivot of right range
             if (__val1.__end == __val1.__pivot)
             {
                 return {__new_begin, __val2.__pivot, __val2.__end};
             }
             // if true part of right range greater than false part of left range
             // then we should swap the false part of left range and last part of true part of right range
             else
             {
-                __brick_rotate(__val1.__pivot, __val2.__begin, __val2.__pivot, __is_vector);
+                __internal::__brick_rotate(__val1.__pivot, __val2.__begin, __val2.__pivot, __is_vector);
                 return {__new_begin, __val2.__pivot - __size1, __val2.__end};
             }
         };
 
         _PartitionRange __result = __par_backend::__parallel_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first, __last, __init,
             [&__pred, __is_vector, __reductor](_BidirectionalIterator __i, _BidirectionalIterator __j,
                                                _PartitionRange __value) -> _PartitionRange {
                 //1. serial stable_partition
-                _BidirectionalIterator __pivot = __brick_stable_partition(__i, __j, __pred, __is_vector);
+                _BidirectionalIterator __pivot = __internal::__brick_stable_partition(__i, __j, __pred, __is_vector);
 
                 // 2. merging of two ranges (left and right respectively)
                 return __reductor(__value, {__i, __pivot, __j});
             },
             __reductor);
         return __result.__pivot;
     });
 }

 
 template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator1, class _OutputIterator2,
           class _UnaryPredicate, class _IsVector>
 std::pair<_OutputIterator1, _OutputIterator2>
 __pattern_partition_copy(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
                          _OutputIterator1 __out_true, _OutputIterator2 __out_false, _UnaryPredicate __pred,
                          _IsVector __is_vector, /*is_parallelization=*/std::false_type) noexcept
 {
-    return __brick_partition_copy(__first, __last, __out_true, __out_false, __pred, __is_vector);
+    return __internal::__brick_partition_copy(__first, __last, __out_true, __out_false, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator1, class _OutputIterator2,
           class _UnaryPredicate, class _IsVector>
 std::pair<_OutputIterator1, _OutputIterator2>
 __pattern_partition_copy(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                          _OutputIterator1 __out_true, _OutputIterator2 __out_false, _UnaryPredicate __pred,
                          _IsVector __is_vector, /*is_parallelization=*/std::true_type)
 {
     typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
     typedef std::pair<_DifferenceType, _DifferenceType> _ReturnType;
     const _DifferenceType __n = __last - __first;
     if (_DifferenceType(1) < __n)
     {
         __par_backend::__buffer<bool> __mask_buf(__n);
-        return __except_handler([&__exec, __n, __first, __out_true, __out_false, __is_vector, __pred, &__mask_buf]() {
+        return __internal::__except_handler([&__exec, __n, __first, __out_true, __out_false, __is_vector, __pred,
+                                             &__mask_buf]() {
             bool* __mask = __mask_buf.get();
             _ReturnType __m{};
             __par_backend::parallel_strict_scan(
                 std::forward<_ExecutionPolicy>(__exec), __n, std::make_pair(_DifferenceType(0), _DifferenceType(0)),
                 [=](_DifferenceType __i, _DifferenceType __len) { // Reduce
-                    return __brick_calc_mask_1<_DifferenceType>(__first + __i, __first + (__i + __len), __mask + __i,
-                                                                __pred, __is_vector);
+                    return __internal::__brick_calc_mask_1<_DifferenceType>(__first + __i, __first + (__i + __len),
+                                                                            __mask + __i, __pred, __is_vector);
                 },
                 [](const _ReturnType& __x, const _ReturnType& __y) -> _ReturnType {
                     return std::make_pair(__x.first + __y.first, __x.second + __y.second);
                 },                                                                       // Combine
                 [=](_DifferenceType __i, _DifferenceType __len, _ReturnType __initial) { // Scan
-                    __brick_partition_by_mask(__first + __i, __first + (__i + __len), __out_true + __initial.first,
-                                              __out_false + __initial.second, __mask + __i, __is_vector);
+                    __internal::__brick_partition_by_mask(__first + __i, __first + (__i + __len),
+                                                          __out_true + __initial.first, __out_false + __initial.second,
+                                                          __mask + __i, __is_vector);
                 },
                 [&__m](_ReturnType __total) { __m = __total; });
             return std::make_pair(__out_true + __m.first, __out_false + __m.second);
         });
     }
     // trivial sequence - use serial algorithm
-    return __brick_partition_copy(__first, __last, __out_true, __out_false, __pred, __is_vector);
+    return __internal::__brick_partition_copy(__first, __last, __out_true, __out_false, __pred, __is_vector);
 }
 #endif
 
 //------------------------------------------------------------------------
 // sort
 //------------------------------------------------------------------------
 
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector,
           class _IsMoveConstructible>
 void
 __pattern_sort(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
                _IsVector /*is_vector*/, /*is_parallel=*/std::false_type, _IsMoveConstructible) noexcept
 {
     std::sort(__first, __last, __comp);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
 void
 __pattern_sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
                _IsVector /*is_vector*/, /*is_parallel=*/std::true_type, /*is_move_constructible=*/std::true_type)
 {
-    __except_handler([&]() {
-        __par_backend::__parallel_stable_sort(
-            std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
-            [](_RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp) {
-                std::sort(__first, __last, __comp);
-            },
+    __internal::__except_handler([&]() {
+        __par_backend::__parallel_stable_sort(std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
+                                              [](_RandomAccessIterator __first, _RandomAccessIterator __last,
+                                                 _Compare __comp) { std::sort(__first, __last, __comp); },
                                               __last - __first);
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // stable_sort
 //------------------------------------------------------------------------
 
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
 void
 __pattern_stable_sort(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
                       _IsVector /*is_vector*/, /*is_parallel=*/std::false_type) noexcept
 {
     std::stable_sort(__first, __last, __comp);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
 void
 __pattern_stable_sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                       _Compare __comp, _IsVector /*is_vector*/, /*is_parallel=*/std::true_type)
 {
-    __except_handler([&]() {
+    __internal::__except_handler([&]() {
         __par_backend::__parallel_stable_sort(std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
                                               [](_RandomAccessIterator __first, _RandomAccessIterator __last,
                                                  _Compare __comp) { std::stable_sort(__first, __last, __comp); });
     });
 }
 #endif
 
 //------------------------------------------------------------------------

 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
 void
 __pattern_partial_sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __middle,
                        _RandomAccessIterator __last, _Compare __comp, _IsVector, /*is_parallel=*/std::true_type)
 {
     const auto __n = __middle - __first;
-    __except_handler([&]() {
+    __internal::__except_handler([&]() {
         __par_backend::__parallel_stable_sort(
             std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
             [__n](_RandomAccessIterator __begin, _RandomAccessIterator __end, _Compare __comp) {
                 if (__n < __end - __begin)
                     std::partial_sort(__begin, __begin + __n, __end, __comp);
                 else
                     std::sort(__begin, __end, __comp);
             },
__pattern_partial_sort_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
                             _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     if (__last == __first || __d_last == __d_first)
     {
         return __d_first;
     }
     auto __n1 = __last - __first;
     auto __n2 = __d_last - __d_first;
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         if (__n2 >= __n1)
         {
             __par_backend::__parallel_stable_sort(
                 std::forward<_ExecutionPolicy>(__exec), __d_first, __d_first + __n1, __comp,
                 [__first, __d_first, __is_vector](_RandomAccessIterator __i, _RandomAccessIterator __j,
                                                   _Compare __comp) {
                     _ForwardIterator __i1 = __first + (__i - __d_first);
                     _ForwardIterator __j1 = __first + (__j - __d_first);
 
                 // 1. Copy elements from input to output
 #if !__PSTL_ICC_18_OMP_SIMD_BROKEN
-                    __brick_copy(__i1, __j1, __i, __is_vector);
+                    __internal::__brick_copy(__i1, __j1, __i, __is_vector);
 #else
                     std::copy(__i1, __j1, __i);
 #endif
                     // 2. Sort elements in output sequence
                     std::sort(__i, __j, __comp);
                 },
                 __n1);
             return __d_first + __n1;
         }
         else
         {
             typedef typename std::iterator_traits<_ForwardIterator>::value_type _T1;
             typedef typename std::iterator_traits<_RandomAccessIterator>::value_type _T2;
             __par_backend::__buffer<_T1> __buf(__n1);
             _T1* __r = __buf.get();
 
-            __par_backend::__parallel_stable_sort(
-                std::forward<_ExecutionPolicy>(__exec), __r, __r + __n1, __comp,
+            __par_backend::__parallel_stable_sort(std::forward<_ExecutionPolicy>(__exec), __r, __r + __n1, __comp,
                                                   [__n2, __first, __r](_T1* __i, _T1* __j, _Compare __comp) {
                                                       _ForwardIterator __it = __first + (__i - __r);
 
                                                       // 1. Copy elements from input to raw memory
                                                       for (_T1* __k = __i; __k != __j; ++__k, ++__it)
                                                       {
                                                           ::new (__k) _T2(*__it);
                                                       }
 
                                                       // 2. Sort elements in temporary __buffer
                                                       if (__n2 < __j - __i)
                                                           std::partial_sort(__i, __i + __n2, __j, __comp);
                                                       else
                                                           std::sort(__i, __j, __comp);
                                                   },
                                                   __n2);
 
             // 3. Move elements from temporary __buffer to output
             __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __r, __r + __n2,
                                           [__r, __d_first, __is_vector](_T1* __i, _T1* __j) {
-                                              __brick_move(__i, __j, __d_first + (__i - __r), __is_vector);
+                                              __internal::__brick_move(__i, __j, __d_first + (__i - __r), __is_vector);
                                           });
             return __d_first + __n2;
         }
     });
 }
 #endif
 
 //------------------------------------------------------------------------
__brick_adjacent_find(_ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
     return std::adjacent_find(__first, __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
 _ForwardIterator
 __pattern_adjacent_find(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
                         /* is_parallel */ std::false_type, _IsVector __is_vector, bool __or_semantic) noexcept
 {
-    return __brick_adjacent_find(__first, __last, __pred, __is_vector, __or_semantic);
+    return __internal::__brick_adjacent_find(__first, __last, __pred, __is_vector, __or_semantic);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _BinaryPredicate, class _IsVector>
 _RandomAccessIterator
 __pattern_adjacent_find(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                         _BinaryPredicate __pred, /* is_parallel */ std::true_type, _IsVector __is_vector,
                         bool __or_semantic)
 {
     if (__last - __first < 2)
         return __last;
 
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         return __par_backend::__parallel_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first, __last, __last,
             [__last, __pred, __is_vector, __or_semantic](_RandomAccessIterator __begin, _RandomAccessIterator __end,
                                                          _RandomAccessIterator __value) -> _RandomAccessIterator {
                 // TODO: investigate performance benefits from the use of shared variable for the result,
                 // checking (compare_and_swap idiom) its __value at __first.
                 if (__or_semantic && __value < __last)
                 { //found
                     __par_backend::__cancel_execution();
                     return __value;
                 }
 
                 if (__value > __begin)
                 {
                     // modify __end to check the predicate on the boundary __values;
                     // TODO: to use a custom range with boundaries overlapping
                     // TODO: investigate what if we remove "if" below and run algorithm on range [__first, __last-1)
                     // then check the pair [__last-1, __last)
                     if (__end != __last)
                         ++__end;
 
                     //correct the global result iterator if the "brick" returns a local "__last"
                     const _RandomAccessIterator __res =
-                        __brick_adjacent_find(__begin, __end, __pred, __is_vector, __or_semantic);
+                        __internal::__brick_adjacent_find(__begin, __end, __pred, __is_vector, __or_semantic);
                     if (__res < __end)
                         __value = __res;
                 }
                 return __value;
             },
             [](_RandomAccessIterator __x, _RandomAccessIterator __y) -> _RandomAccessIterator {
                 return __x < __y ? __x : __y;
             } //reduce a __value
if (__first == __last || __nth == __last)
         return;
     }
 
     using std::iter_swap;
     typedef typename std::iterator_traits<_RandomAccessIterator>::value_type _Tp;
     _RandomAccessIterator __x;
     do
     {
-        __x = __pattern_partition(
-            std::forward<_ExecutionPolicy>(__exec), __first + 1, __last,
-            [&__comp, __first](const _Tp& __x) { return __comp(__x, *__first); }, __is_vector,
+        __x = __internal::__pattern_partition(std::forward<_ExecutionPolicy>(__exec), __first + 1, __last,
+                                              [&__comp, __first](const _Tp& __x) { return __comp(__x, *__first); },
+                                              __is_vector,
                                               /*is_parallel=*/std::true_type());
         --__x;
         if (__x != __first)
         {
             iter_swap(__first, __x);
         }
         // if x > nth then our new range for partition is [first, x)
         if (__x - __nth > 0)
         {
__brick_fill(_ForwardIterator __first, _ForwardIterator __last, const _Tp& __value,
     std::fill(__first, __last, __value);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _IsVector>
 void
 __pattern_fill(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value,
                /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
 {
-    __brick_fill(__first, __last, __value, __is_vector);
+    __internal::__brick_fill(__first, __last, __value, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _IsVector>
 _ForwardIterator
 __pattern_fill(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value,
                /*is_parallel=*/std::true_type, _IsVector __is_vector)
 {
-    return __except_handler([&__exec, __first, __last, &__value, __is_vector]() {
+    return __internal::__except_handler([&__exec, __first, __last, &__value, __is_vector]() {
         __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                       [&__value, __is_vector](_ForwardIterator __begin, _ForwardIterator __end) {
-                                          __brick_fill(__begin, __end, __value, __is_vector);
+                                          __internal::__brick_fill(__begin, __end, __value, __is_vector);
                                       });
         return __last;
     });
 }
 #endif
 
 template <class _OutputIterator, class _Size, class _Tp>
 _OutputIterator
__brick_fill_n(_OutputIterator __first, _Size __count, const _Tp& __value, /* __is_vector = */ std::false_type) noexcept
     return std::fill_n(__first, __count, __value);
 }
 
 template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Tp, class _IsVector>
 _OutputIterator
 __pattern_fill_n(_ExecutionPolicy&&, _OutputIterator __first, _Size __count, const _Tp& __value,
                  /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
 {
-    return __brick_fill_n(__first, __count, __value, __is_vector);
+    return __internal::__brick_fill_n(__first, __count, __value, __is_vector);
 }
 
 template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Tp, class _IsVector>
 _OutputIterator
 __pattern_fill_n(_ExecutionPolicy&& __exec, _OutputIterator __first, _Size __count, const _Tp& __value,
                  /*is_parallel=*/std::true_type, _IsVector __is_vector)
 {
-    return __pattern_fill(std::forward<_ExecutionPolicy>(__exec), __first, __first + __count, __value, std::true_type(),
-                          __is_vector);
+    return __internal::__pattern_fill(std::forward<_ExecutionPolicy>(__exec), __first, __first + __count, __value,
+                                      std::true_type(), __is_vector);
 }
 
 //------------------------------------------------------------------------
 // generate, generate_n
 //------------------------------------------------------------------------
 template <class _RandomAccessIterator, class _Generator>
 void
 __brick_generate(_RandomAccessIterator __first, _RandomAccessIterator __last, _Generator __g,
__brick_generate(_ForwardIterator __first, _ForwardIterator __last, _Generator __g,
     std::generate(__first, __last, __g);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Generator, class _IsVector>
 void
 __pattern_generate(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Generator __g,
                    /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
 {
-    __brick_generate(__first, __last, __g, __is_vector);
+    __internal::__brick_generate(__first, __last, __g, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Generator, class _IsVector>
 _ForwardIterator
 __pattern_generate(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Generator __g,
                    /*is_parallel=*/std::true_type, _IsVector __is_vector)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                       [__g, __is_vector](_ForwardIterator __begin, _ForwardIterator __end) {
-                                          __brick_generate(__begin, __end, __g, __is_vector);
+                                          __internal::__brick_generate(__begin, __end, __g, __is_vector);
                                       });
         return __last;
     });
 }
 #endif
 
 template <class OutputIterator, class Size, class _Generator>
 OutputIterator
__brick_generate_n(OutputIterator __first, Size __count, _Generator __g, /* is_vector = */ std::false_type) noexcept
     return std::generate_n(__first, __count, __g);
 }
 
 template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Generator, class _IsVector>
 _OutputIterator
 __pattern_generate_n(_ExecutionPolicy&&, _OutputIterator __first, _Size __count, _Generator __g,
                      /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
 {
-    return __brick_generate_n(__first, __count, __g, __is_vector);
+    return __internal::__brick_generate_n(__first, __count, __g, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Generator, class _IsVector>
 _OutputIterator
 __pattern_generate_n(_ExecutionPolicy&& __exec, _OutputIterator __first, _Size __count, _Generator __g,
                      /*is_parallel=*/std::true_type, _IsVector __is_vector)
 {
     static_assert(__is_random_access_iterator<_OutputIterator>::value,
                   "Pattern-brick error. Should be a random access iterator.");
-    return __pattern_generate(std::forward<_ExecutionPolicy>(__exec), __first, __first + __count, __g, std::true_type(),
-                              __is_vector);
+    return __internal::__pattern_generate(std::forward<_ExecutionPolicy>(__exec), __first, __first + __count, __g,
+                                          std::true_type(), __is_vector);
 }
 #endif
 
 //------------------------------------------------------------------------
 // remove
 //------------------------------------------------------------------------
 
 template <class _ForwardIterator, class _UnaryPredicate>

 #endif
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
 _ForwardIterator
 __pattern_remove_if(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
                     _IsVector __is_vector, /*is_parallel*/ std::false_type) noexcept
 {
-    return __brick_remove_if(__first, __last, __pred, __is_vector);
+    return __internal::__brick_remove_if(__first, __last, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
 _ForwardIterator
 __pattern_remove_if(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
                     _UnaryPredicate __pred, _IsVector __is_vector, /*is_parallel*/ std::true_type) noexcept
 {
     typedef typename std::iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
     if (__first == __last || __first + 1 == __last)
     {
         // Trivial sequence - use serial algorithm
-        return __brick_remove_if(__first, __last, __pred, __is_vector);
+        return __internal::__brick_remove_if(__first, __last, __pred, __is_vector);
     }
 
-    return remove_elements(
+    return __internal::__remove_elements(
         std::forward<_ExecutionPolicy>(__exec), __first, __last,
         [&__pred, __is_vector](bool* __b, bool* __e, _ForwardIterator __it) {
-            __brick_walk2(
-                __b, __e, __it, [&__pred](bool& __x, _ReferenceType __y) { __x = !__pred(__y); }, __is_vector);
+            __internal::__brick_walk2(__b, __e, __it, [&__pred](bool& __x, _ReferenceType __y) { __x = !__pred(__y); },
+                                      __is_vector);
         },
         __is_vector);
 }
 #endif
 
 //------------------------------------------------------------------------
 // merge
 //------------------------------------------------------------------------

 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_merge(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                 _ForwardIterator2 __last2, _OutputIterator __d_first, _Compare __comp, _IsVector __is_vector,
                 /* is_parallel = */ std::false_type) noexcept
 {
-    return __brick_merge(__first1, __last1, __first2, __last2, __d_first, __comp, __is_vector);
+    return __internal::__brick_merge(__first1, __last1, __first2, __last2, __d_first, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_merge(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
                 _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2, _OutputIterator __d_first,
                 _Compare __comp, _IsVector __is_vector, /* is_parallel = */ std::true_type)
 {
     __par_backend::__parallel_merge(
         std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __d_first, __comp,
         [__is_vector](_RandomAccessIterator1 __f1, _RandomAccessIterator1 __l1, _RandomAccessIterator2 __f2,
-                      _RandomAccessIterator2 __l2, _OutputIterator __f3,
-                      _Compare __comp) { return __brick_merge(__f1, __l1, __f2, __l2, __f3, __comp, __is_vector); });
+                      _RandomAccessIterator2 __l2, _OutputIterator __f3, _Compare __comp) {
+            return __internal::__brick_merge(__f1, __l1, __f2, __l2, __f3, __comp, __is_vector);
+        });
     return __d_first + (__last1 - __first1) + (__last2 - __first2);
 }
 #endif
 
 //------------------------------------------------------------------------
 // inplace_merge
 //------------------------------------------------------------------------
 template <class _BidirectionalIterator, class _Compare>

 }
 
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _Compare, class _IsVector>
 void
 __pattern_inplace_merge(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __middle,
                         _BidirectionalIterator __last, _Compare __comp, _IsVector __is_vector,
                         /* is_parallel = */ std::false_type) noexcept
 {
-    __brick_inplace_merge(__first, __middle, __last, __comp, __is_vector);
+    __internal::__brick_inplace_merge(__first, __middle, __last, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _Compare, class _IsVector>
 void
 __pattern_inplace_merge(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __middle,
                         _BidirectionalIterator __last, _Compare __comp, _IsVector __is_vector,
                         /*is_parallel=*/std::true_type)
 {
     if (__first == __last || __first == __middle || __middle == __last)
     {
         return;
     }
     typedef typename std::iterator_traits<_BidirectionalIterator>::value_type _Tp;
     auto __n = __last - __first;
     __par_backend::__buffer<_Tp> __buf(__n);
     _Tp* __r = __buf.get();
-    __except_handler([&]() {
+    __internal::__except_handler([&]() {
         auto __move_values = [](_BidirectionalIterator __x, _Tp* __z) {
-            __invoke_if_else(
-                std::is_trivial<_Tp>(), [&]() { *__z = std::move(*__x); },
+            __internal::__invoke_if_else(std::is_trivial<_Tp>(), [&]() { *__z = std::move(*__x); },
                                          [&]() { ::new (std::addressof(*__z)) _Tp(std::move(*__x)); });
         };
 
         auto __move_sequences = [](_BidirectionalIterator __first1, _BidirectionalIterator __last1, _Tp* __first2) {
-            return __brick_uninitialized_move(__first1, __last1, __first2, _IsVector());
+            return __internal::__brick_uninitialized_move(__first1, __last1, __first2, _IsVector());
         };
 
         __par_backend::__parallel_merge(
             std::forward<_ExecutionPolicy>(__exec), __first, __middle, __middle, __last, __r, __comp,
             [__n, __move_values, __move_sequences](_BidirectionalIterator __f1, _BidirectionalIterator __l1,
                                                    _BidirectionalIterator __f2, _BidirectionalIterator __l2, _Tp* __f3,
                                                    _Compare __comp) {
                 auto __func = __par_backend::__serial_move_merge<decltype(__move_values), decltype(__move_sequences)>(
                     __n, __move_values, __move_sequences);
                 __func(__f1, __l1, __f2, __l2, __f3, __comp);
                 return __f3 + (__l1 - __f1) + (__l2 - __f2);
             });
 
         __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __r, __r + __n,
                                       [__r, __first, __is_vector](_Tp* __i, _Tp* __j) {
-                                          __brick_move(__i, __j, __first + (__i - __r), __is_vector);
+                                          __internal::__brick_move(__i, __j, __first + (__i - __r), __is_vector);
                                       });
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // includes
 //------------------------------------------------------------------------
__pattern_includes(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
 
     __first1 = std::lower_bound(__first1, __last1, *__first2, __comp);
     if (__first1 == __last1)
         return false;
 
     if (__last2 - __first2 == 1)
         return !__comp(*__first1, *__first2) && !__comp(*__first2, *__first1);
 
-    return __except_handler([&]() {
-        return !__parallel_or(
+    return __internal::__except_handler([&]() {
+        return !__internal::__parallel_or(
             std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
             [__first1, __last1, __first2, __last2, &__comp](_ForwardIterator2 __i, _ForwardIterator2 __j) {
                 assert(__j > __i);
                 //assert(__j - __i > 1);
 
                 //1. moving boundaries to "consume" subsequence of equal elements
                 auto __is_equal = [&__comp](_ForwardIterator2 __a, _ForwardIterator2 __b) -> bool {
                     return !__comp(*__a, *__b) && !__comp(*__b, *__a);
struct _SetRange
         }
     };
 
     const _DifferenceType __n1 = __last1 - __first1;
     const _DifferenceType __n2 = __last2 - __first2;
 
     __par_backend::__buffer<_T> __buf(__size_func(__n1, __n2));
 
-    return __except_handler([&__exec, __n1, __first1, __last1, __first2, __last2, __result, __is_vector, __comp,
-                             __size_func, __set_op, &__buf]() {
+    return __internal::__except_handler([&__exec, __n1, __first1, __last1, __first2, __last2, __result, __is_vector,
+                                         __comp, __size_func, __set_op, &__buf]() {
         auto __buffer = __buf.get();
         _DifferenceType __m{};
         auto __scan = [=](_DifferenceType, _DifferenceType, const _SetRange& __s) { // Scan
             if (!__s.empty())
-                __brick_move(__buffer + __s.__buf_pos, __buffer + (__s.__buf_pos + __s.__len), __result + __s.__pos,
-                             __is_vector);
+                __internal::__brick_move(__buffer + __s.__buf_pos, __buffer + (__s.__buf_pos + __s.__len),
+                                         __result + __s.__pos, __is_vector);
         };
         __par_backend::parallel_strict_scan(
             std::forward<_ExecutionPolicy>(__exec), __n1, _SetRange{0, 0, 0}, //-1, 0},
             [=](_DifferenceType __i, _DifferenceType __len) {                 // Reduce
                 //[__b; __e) - a subrange of the first sequence, to reduce
                 _ForwardIterator1 __b = __first1 + __i, __e = __first1 + (__i + __len);
 
                 //try searching for the first element which not equal to *__b
__parallel_set_union_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                         _Compare __comp, _SetUnionOp __set_union_op, _IsVector __is_vector)
 {
     typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
 
     const auto __n1 = __last1 - __first1;
     const auto __n2 = __last2 - __first2;
 
     auto copy_range1 = [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res) {
-        return __brick_copy(__begin, __end, __res, __is_vector);
+        return __internal::__brick_copy(__begin, __end, __res, __is_vector);
     };
     auto copy_range2 = [__is_vector](_ForwardIterator2 __begin, _ForwardIterator2 __end, _OutputIterator __res) {
-        return __brick_copy(__begin, __end, __res, __is_vector);
+        return __internal::__brick_copy(__begin, __end, __res, __is_vector);
     };
 
     // {1} {}: parallel copying just first sequence
     if (__n2 == 0)
-        return __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result, copy_range1,
-                                     std::true_type());
+        return __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
+                                                 copy_range1, std::true_type());
 
     // {} {2}: parallel copying justmake  second sequence
     if (__n1 == 0)
-        return __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2, __result, copy_range2,
-                                     std::true_type());
+        return __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2, __result,
+                                                 copy_range2, std::true_type());
 
     // testing  whether the sequences are intersected
-    _ForwardIterator1 __left_bound_seq_1 = lower_bound(__first1, __last1, *__first2, __comp);
+    _ForwardIterator1 __left_bound_seq_1 = std::lower_bound(__first1, __last1, *__first2, __comp);
 
     if (__left_bound_seq_1 == __last1)
     {
         //{1} < {2}: seq2 is wholly greater than seq1, so, do parallel copying seq1 and seq2
         __par_backend::__parallel_invoke(
             std::forward<_ExecutionPolicy>(__exec),
             [=] {
-                __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result, copy_range1,
-                                      std::true_type());
+                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
+                                                  copy_range1, std::true_type());
             },
             [=] {
-                __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2, __result + __n1,
-                                      copy_range2, std::true_type());
+                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
+                                                  __result + __n1, copy_range2, std::true_type());
             });
         return __result + __n1 + __n2;
     }
 
     // testing  whether the sequences are intersected
-    _ForwardIterator2 __left_bound_seq_2 = lower_bound(__first2, __last2, *__first1, __comp);
+    _ForwardIterator2 __left_bound_seq_2 = std::lower_bound(__first2, __last2, *__first1, __comp);
 
     if (__left_bound_seq_2 == __last2)
     {
         //{2} < {1}: seq2 is wholly greater than seq1, so, do parallel copying seq1 and seq2
         __par_backend::__parallel_invoke(
             std::forward<_ExecutionPolicy>(__exec),
             [=] {
-                __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2, __result, copy_range2,
-                                      std::true_type());
+                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2, __result,
+                                                  copy_range2, std::true_type());
             },
             [=] {
-                __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result + __n2,
-                                      copy_range1, std::true_type());
+                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                                                  __result + __n2, copy_range1, std::true_type());
             });
         return __result + __n1 + __n2;
     }
 
     const auto __m1 = __left_bound_seq_1 - __first1;
     if (__m1 > __set_algo_cut_off)
     {
         auto __res_or = __result;
         __result += __m1; //we know proper offset due to [first1; left_bound_seq_1) < [first2; last2)
         __par_backend::__parallel_invoke(
             std::forward<_ExecutionPolicy>(__exec),
             //do parallel copying of [first1; left_bound_seq_1)
             [=] {
-                __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __left_bound_seq_1, __res_or,
-                                      copy_range1, std::true_type());
+                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __left_bound_seq_1,
+                                                  __res_or, copy_range1, std::true_type());
             },
             [=, &__result] {
-                __result = __parallel_set_op(
+                __result = __internal::__parallel_set_op(
                     std::forward<_ExecutionPolicy>(__exec), __left_bound_seq_1, __last1, __first2, __last2, __result,
                     __comp, [](_DifferenceType __n, _DifferenceType __m) { return __n + __m; }, __set_union_op,
                     __is_vector);
             });
         return __result;
     }
 
     const auto __m2 = __left_bound_seq_2 - __first2;
     assert(__m1 == 0 || __m2 == 0);
     if (__m2 > __set_algo_cut_off)
     {
         auto __res_or = __result;
         __result += __m2; //we know proper offset due to [first2; left_bound_seq_2) < [first1; last1)
         __par_backend::__parallel_invoke(
             std::forward<_ExecutionPolicy>(__exec),
             //do parallel copying of [first2; left_bound_seq_2)
             [=] {
-                __pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __left_bound_seq_2, __res_or,
-                                      copy_range2, std::true_type());
+                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __left_bound_seq_2,
+                                                  __res_or, copy_range2, std::true_type());
             },
             [=, &__result] {
-                __result = __parallel_set_op(
+                __result = __internal::__parallel_set_op(
                     std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __left_bound_seq_2, __last2, __result,
                     __comp, [](_DifferenceType __n, _DifferenceType __m) { return __n + __m; }, __set_union_op,
                     __is_vector);
             });
         return __result;
     }
 
-    return __parallel_set_op(
+    return __internal::__parallel_set_op(
         std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
         [](_DifferenceType __n, _DifferenceType __m) { return __n + __m; }, __set_union_op, __is_vector);
 }
 #endif
 
 //------------------------------------------------------------------------
 // set_union
 //------------------------------------------------------------------------

 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_union(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                     _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result, _Compare __comp,
                     _IsVector __is_vector,
                     /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_set_union(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
+    return __internal::__brick_set_union(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_union(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                     _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result, _Compare __comp,
                     _IsVector __is_vector, /*__is_parallel=*/std::true_type)
 {
 
     const auto __n1 = __last1 - __first1;
     const auto __n2 = __last2 - __first2;
 
     // use serial algorithm
     if (__n1 + __n2 <= __set_algo_cut_off)
         return std::set_union(__first1, __last1, __first2, __last2, __result, __comp);
 
     typedef typename std::iterator_traits<_OutputIterator>::value_type _T;
-    return __parallel_set_union_op(
+    return __internal::__parallel_set_union_op(
         std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
         [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2, _ForwardIterator2 __last2,
            _T* __result,
            _Compare __comp) { return std::set_union(__first1, __last1, __first2, __last2, __result, __comp); },
         __is_vector);
 }
 #endif
 

 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_intersection(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                            _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
                            _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_set_intersection(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
+    return __internal::__brick_set_intersection(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_intersection(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                            _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
                            _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     typedef typename std::iterator_traits<_OutputIterator>::value_type _T;
     typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
 
     const auto __n1 = __last1 - __first1;
     const auto __n2 = __last2 - __first2;
 
     // intersection is empty
     if (__n1 == 0 || __n2 == 0)
         return __result;
 
     // testing  whether the sequences are intersected
-    _ForwardIterator1 __left_bound_seq_1 = lower_bound(__first1, __last1, *__first2, __comp);
+    _ForwardIterator1 __left_bound_seq_1 = std::lower_bound(__first1, __last1, *__first2, __comp);
     //{1} < {2}: seq 2 is wholly greater than seq 1, so, the intersection is empty
     if (__left_bound_seq_1 == __last1)
         return __result;
 
     // testing  whether the sequences are intersected
-    _ForwardIterator2 __left_bound_seq_2 = lower_bound(__first2, __last2, *__first1, __comp);
+    _ForwardIterator2 __left_bound_seq_2 = std::lower_bound(__first2, __last2, *__first1, __comp);
     //{2} < {1}: seq 1 is wholly greater than seq 2, so, the intersection is empty
     if (__left_bound_seq_2 == __last2)
         return __result;
 
     const auto __m1 = __last1 - __left_bound_seq_1 + __n2;
     if (__m1 > __set_algo_cut_off)
     {
         //we know proper offset due to [first1; left_bound_seq_1) < [first2; last2)
-        return __parallel_set_op(
+        return __internal::__parallel_set_op(
             std::forward<_ExecutionPolicy>(__exec), __left_bound_seq_1, __last1, __first2, __last2, __result, __comp,
             [](_DifferenceType __n, _DifferenceType __m) { return std::min(__n, __m); },
             [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                _ForwardIterator2 __last2, _T* __result, _Compare __comp) {
                 return std::set_intersection(__first1, __last1, __first2, __last2, __result, __comp);
             },
             __is_vector);
     }
 
     const auto __m2 = __last2 - __left_bound_seq_2 + __n1;
     if (__m2 > __set_algo_cut_off)
     {
         //we know proper offset due to [first2; left_bound_seq_2) < [first1; last1)
-        __result = __parallel_set_op(
+        __result = __internal::__parallel_set_op(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __left_bound_seq_2, __last2, __result, __comp,
             [](_DifferenceType __n, _DifferenceType __m) { return std::min(__n, __m); },
             [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                _ForwardIterator2 __last2, _T* __result, _Compare __comp) {
                 return std::set_intersection(__first2, __last2, __first1, __last1, __result, __comp);
             },
             __is_vector);
         return __result;

 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_difference(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                          _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
                          _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_set_difference(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
+    return __internal::__brick_set_difference(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_difference(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                          _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
                          _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     typedef typename std::iterator_traits<_OutputIterator>::value_type _T;
     typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
 
     const auto __n1 = __last1 - __first1;
     const auto __n2 = __last2 - __first2;
 
     // {} \ {2}: the difference is empty
     if (__n1 == 0)
         return __result;
 
     // {1} \ {}: parallel copying just first sequence
     if (__n2 == 0)
-        return __pattern_walk2_brick(
+        return __internal::__pattern_walk2_brick(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
             [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res) {
-                return __brick_copy(__begin, __end, __res, __is_vector);
+                return __internal::__brick_copy(__begin, __end, __res, __is_vector);
             },
             std::true_type());
 
     // testing  whether the sequences are intersected
-    _ForwardIterator1 __left_bound_seq_1 = lower_bound(__first1, __last1, *__first2, __comp);
+    _ForwardIterator1 __left_bound_seq_1 = std::lower_bound(__first1, __last1, *__first2, __comp);
     //{1} < {2}: seq 2 is wholly greater than seq 1, so, parallel copying just first sequence
     if (__left_bound_seq_1 == __last1)
-        return __pattern_walk2_brick(
+        return __internal::__pattern_walk2_brick(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
             [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res) {
-                return __brick_copy(__begin, __end, __res, __is_vector);
+                return __internal::__brick_copy(__begin, __end, __res, __is_vector);
             },
             std::true_type());
 
     // testing  whether the sequences are intersected
-    _ForwardIterator2 __left_bound_seq_2 = lower_bound(__first2, __last2, *__first1, __comp);
+    _ForwardIterator2 __left_bound_seq_2 = std::lower_bound(__first2, __last2, *__first1, __comp);
     //{2} < {1}: seq 1 is wholly greater than seq 2, so, parallel copying just first sequence
     if (__left_bound_seq_2 == __last2)
-        return __pattern_walk2_brick(
+        return __internal::__pattern_walk2_brick(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
             [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res) {
-                return __brick_copy(__begin, __end, __res, __is_vector);
+                return __internal::__brick_copy(__begin, __end, __res, __is_vector);
             },
             std::true_type());
 
     if (__n1 + __n2 > __set_algo_cut_off)
-        return __parallel_set_op(
+        return __internal::__parallel_set_op(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
             [](_DifferenceType __n, _DifferenceType __m) { return __n; },
             [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                _ForwardIterator2 __last2, _T* __result,
                _Compare __comp) { return std::set_difference(__first1, __last1, __first2, __last2, __result, __comp); },
             __is_vector);
 
     // use serial algorithm

 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_symmetric_difference(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                                    _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
                                    _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_set_symmetric_difference(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
+    return __internal::__brick_set_symmetric_difference(__first1, __last1, __first2, __last2, __result, __comp,
+                                                        __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
           class _Compare, class _IsVector>
 _OutputIterator
 __pattern_set_symmetric_difference(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                                    _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
                                    _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
 
     const auto __n1 = __last1 - __first1;
     const auto __n2 = __last2 - __first2;
 
     // use serial algorithm
     if (__n1 + __n2 <= __set_algo_cut_off)
         return std::set_symmetric_difference(__first1, __last1, __first2, __last2, __result, __comp);
 
     typedef typename std::iterator_traits<_OutputIterator>::value_type _T;
-    return __parallel_set_union_op(
+    return __internal::__parallel_set_union_op(
         std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
         [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2, _ForwardIterator2 __last2,
            _T* __result, _Compare __comp) {
             return std::set_symmetric_difference(__first1, __last1, __first2, __last2, __result, __comp);
         },
         __is_vector);
 }
 #endif
return __unseq_backend::__simd_first(
         [&__comp](_RandomAccessIterator __it, _SizeType __i) { return __comp(__it[(__i - 1) / 2], __it[__i]); });
 }
 
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
 _RandomAccessIterator
 __pattern_is_heap_until(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last,
                         _Compare __comp, _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
 {
-    return __brick_is_heap_until(__first, __last, __comp, __is_vector);
+    return __internal::__brick_is_heap_until(__first, __last, __comp, __is_vector);
 }
 
 template <class _RandomAccessIterator, class _DifferenceType, class _Compare>
 _RandomAccessIterator
 __is_heap_until_local(_RandomAccessIterator __first, _DifferenceType __begin, _DifferenceType __end, _Compare __comp,
                       /* __is_vector = */ std::false_type) noexcept
 {
     _DifferenceType __i = __begin;

 template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
 _RandomAccessIterator
 __pattern_is_heap_until(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                         _Compare __comp, _IsVector __is_vector, /* is_parallel = */ std::true_type) noexcept
 {
     if (__last - __first < 2)
         return __last;
 
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         return __parallel_find(
             std::forward<_ExecutionPolicy>(__exec), __first, __last,
             [__first, __comp, __is_vector](_RandomAccessIterator __i, _RandomAccessIterator __j) {
-                return __is_heap_until_local(__first, __i - __first, __j - __first, __comp, __is_vector);
+                return __internal::__is_heap_until_local(__first, __i - __first, __j - __first, __comp, __is_vector);
             },
             std::less<typename std::iterator_traits<_RandomAccessIterator>::difference_type>(), /*is_first=*/true);
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // min_element

 #endif
 }
 
 template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
 _ForwardIterator
 __pattern_min_element(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp,
                       _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
 {
-    return __brick_min_element(__first, __last, __comp, __is_vector);
+    return __internal::__brick_min_element(__first, __last, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <typename _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare, typename _IsVector>
 _RandomAccessIterator
 __pattern_min_element(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                       _Compare __comp, _IsVector __is_vector, /* is_parallel = */ std::true_type)
 {
     if (__first == __last)
         return __last;
 
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         return __par_backend::__parallel_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first + 1, __last, __first,
             [=](_RandomAccessIterator __begin, _RandomAccessIterator __end,
                 _RandomAccessIterator __init) -> _RandomAccessIterator {
-                const _RandomAccessIterator subresult = __brick_min_element(__begin, __end, __comp, __is_vector);
-                return __cmp_iterators_by_values(__init, subresult, __comp);
+                const _RandomAccessIterator subresult =
+                    __internal::__brick_min_element(__begin, __end, __comp, __is_vector);
+                return __internal::__cmp_iterators_by_values(__init, subresult, __comp);
             },
             [=](_RandomAccessIterator __it1, _RandomAccessIterator __it2) -> _RandomAccessIterator {
-                return __cmp_iterators_by_values(__it1, __it2, __comp);
+                return __internal::__cmp_iterators_by_values(__it1, __it2, __comp);
             });
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // minmax_element
 //------------------------------------------------------------------------

 #endif
 }
 
 template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
 std::pair<_ForwardIterator, _ForwardIterator>
 __pattern_minmax_element(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp,
                          _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
 {
-    return __brick_minmax_element(__first, __last, __comp, __is_vector);
+    return __internal::__brick_minmax_element(__first, __last, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
 std::pair<_ForwardIterator, _ForwardIterator>
 __pattern_minmax_element(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp,
                          _IsVector __is_vector, /* is_parallel = */ std::true_type)
 {
     if (__first == __last)
         return std::make_pair(__first, __first);
 
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         typedef std::pair<_ForwardIterator, _ForwardIterator> _Result;
 
         return __par_backend::__parallel_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first + 1, __last, std::make_pair(__first, __first),
             [=](_ForwardIterator __begin, _ForwardIterator __end, _Result __init) -> _Result {
-                const _Result __subresult = __brick_minmax_element(__begin, __end, __comp, __is_vector);
-                return std::make_pair(
-                    __cmp_iterators_by_values(__subresult.first, __init.first, __comp),
-                    __cmp_iterators_by_values(__init.second, __subresult.second, __not_pred<_Compare>(__comp)));
+                const _Result __subresult = __internal::__brick_minmax_element(__begin, __end, __comp, __is_vector);
+                return std::make_pair(__internal::__cmp_iterators_by_values(__subresult.first, __init.first, __comp),
+                                      __internal::__cmp_iterators_by_values(__init.second, __subresult.second,
+                                                                            __not_pred<_Compare>(__comp)));
             },
             [=](_Result __p1, _Result __p2) -> _Result {
                 return std::make_pair(
-                    __cmp_iterators_by_values(__p1.first, __p2.first, __comp),
-                    __cmp_iterators_by_values(__p2.second, __p1.second, __not_pred<_Compare>(__comp)));
+                    __internal::__cmp_iterators_by_values(__p1.first, __p2.first, __comp),
+                    __internal::__cmp_iterators_by_values(__p2.second, __p1.second, __not_pred<_Compare>(__comp)));
             });
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // mismatch
 //------------------------------------------------------------------------

 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate, class _IsVector>
 std::pair<_ForwardIterator1, _ForwardIterator2>
 __pattern_mismatch(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                    _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Predicate __pred, _IsVector __is_vector,
                    /* is_parallel = */ std::false_type) noexcept
 {
-    return __brick_mismatch(__first1, __last1, __first2, __last2, __pred, __is_vector);
+    return __internal::__brick_mismatch(__first1, __last1, __first2, __last2, __pred, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Predicate,
           class _IsVector>
 std::pair<_RandomAccessIterator1, _RandomAccessIterator2>
 __pattern_mismatch(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
                    _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2, _Predicate __pred,
                    _IsVector __is_vector, /* is_parallel = */ std::true_type) noexcept
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         auto __n = std::min(__last1 - __first1, __last2 - __first2);
-        auto __result = __parallel_find(
+        auto __result = __internal::__parallel_find(
             std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
             [__first1, __first2, __pred, __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
-                return __brick_mismatch(__i, __j, __first2 + (__i - __first1), __first2 + (__j - __first1), __pred,
-                                        __is_vector)
+                return __internal::__brick_mismatch(__i, __j, __first2 + (__i - __first1), __first2 + (__j - __first1),
+                                                    __pred, __is_vector)
                     .first;
             },
             std::less<typename std::iterator_traits<_RandomAccessIterator1>::difference_type>(), /*is_first=*/true);
         return std::make_pair(__result, __first2 + (__result - __first1));
     });
 }
 #endif
 

 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
 bool
 __pattern_lexicographical_compare(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                                   _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp,
                                   _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
 {
-    return __brick_lexicographical_compare(__first1, __last1, __first2, __last2, __comp, __is_vector);
+    return __internal::__brick_lexicographical_compare(__first1, __last1, __first2, __last2, __comp, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
 bool
 __pattern_lexicographical_compare(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                                   _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp,
                                   _IsVector __is_vector, /* is_parallel = */ std::true_type) noexcept
 {
     if (__first2 == __last2)
     { // if second sequence is empty
         return false;
     }
     else if (__first1 == __last1)
     { // if first sequence is empty
         return true;
     }
     else
     {
         typedef typename std::iterator_traits<_ForwardIterator1>::reference _RefType1;
         typedef typename std::iterator_traits<_ForwardIterator2>::reference _RefType2;
         --__last1;
         --__last2;
         auto __n = std::min(__last1 - __first1, __last2 - __first2);
-        auto __result = __parallel_find(
+        auto __result = __internal::__parallel_find(
             std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
             [__first1, __first2, &__comp, __is_vector](_ForwardIterator1 __i, _ForwardIterator1 __j) {
-                return __brick_mismatch(
-                           __i, __j, __first2 + (__i - __first1), __first2 + (__j - __first1),
+                return __internal::__brick_mismatch(__i, __j, __first2 + (__i - __first1), __first2 + (__j - __first1),
                                                     [&__comp](const _RefType1 __x, const _RefType2 __y) {
                                                         return !__comp(__x, __y) && !__comp(__y, __x);
                                                     },
                                                     __is_vector)
                     .first;
             },
             std::less<typename std::iterator_traits<_ForwardIterator1>::difference_type>(), /*is_first=*/true);
 
         if (__result == __last1 && __first2 + (__result - __first1) != __last2)
         { // if first sequence shorter than second
             return !__comp(*(__first2 + (__result - __first1)), *__result);
         }

pstl/trunk/include/pstl/internal/execution_defs.h

 
 // 2.3, Execution policy type trait
 template <class T>
 struct is_execution_policy : std::false_type
 {
 };
 
 template <>
-struct is_execution_policy<sequenced_policy> : std::true_type
+struct is_execution_policy<__pstl::execution::sequenced_policy> : std::true_type
 {
 };
 #if __PSTL_USE_PAR_POLICIES
 template <>
-struct is_execution_policy<parallel_policy> : std::true_type
+struct is_execution_policy<__pstl::execution::parallel_policy> : std::true_type
 {
 };
 template <>
-struct is_execution_policy<parallel_unsequenced_policy> : std::true_type
+struct is_execution_policy<__pstl::execution::parallel_unsequenced_policy> : std::true_type
 {
 };
 #endif
 template <>
-struct is_execution_policy<unsequenced_policy> : std::true_type
+struct is_execution_policy<__pstl::execution::unsequenced_policy> : std::true_type
 {
 };
 
 #if __PSTL_CPP14_VARIABLE_TEMPLATES_PRESENT
 template <class T>
-constexpr bool is_execution_policy_v = is_execution_policy<T>::value;
+constexpr bool is_execution_policy_v = __pstl::execution::is_execution_policy<T>::value;
 #endif
 
 } // namespace v1
 } // namespace execution
 
 namespace __internal
 {
 template <class ExecPolicy, class T>
 using __enable_if_execution_policy =
     typename std::enable_if<__pstl::execution::is_execution_policy<typename std::decay<ExecPolicy>::type>::value,
                             T>::type;
 } // namespace __internal
 
 } // namespace __pstl
 
 #endif /* __PSTL_execution_policy_defs_H */

pstl/trunk/include/pstl/internal/execution_impl.h

 {
     return __a;
 }
 
 /* iterator */
 template <typename _IteratorType, typename... _OtherIteratorTypes>
 struct __is_random_access_iterator
 {
-    static constexpr bool value =
-        __is_random_access_iterator<_IteratorType>::value && __is_random_access_iterator<_OtherIteratorTypes...>::value;
+    static constexpr bool value = __internal::__is_random_access_iterator<_IteratorType>::value &&
+                                  __internal::__is_random_access_iterator<_OtherIteratorTypes...>::value;
     typedef std::integral_constant<bool, value> type;
 };
 
 template <typename _IteratorType>
 struct __is_random_access_iterator<_IteratorType>
     : std::is_same<typename std::iterator_traits<_IteratorType>::iterator_category, std::random_access_iterator_tag>
 {
 };

 {
     typedef std::true_type allow_parallel;
     typedef std::true_type allow_unsequenced;
     typedef std::true_type allow_vector;
 };
 #endif
 
 template <typename _ExecutionPolicy>
-using __collector_t = typename __policy_traits<typename std::decay<_ExecutionPolicy>::type>::__collector_type;
+using __collector_t =
+    typename __internal::__policy_traits<typename std::decay<_ExecutionPolicy>::type>::__collector_type;
 
 template <typename _ExecutionPolicy>
-using __allow_vector = typename __policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_vector;
+using __allow_vector =
+    typename __internal::__policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_vector;
 
 template <typename _ExecutionPolicy>
-using __allow_unsequenced = typename __policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_unsequenced;
+using __allow_unsequenced =
+    typename __internal::__policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_unsequenced;
 
 template <typename _ExecutionPolicy>
-using __allow_parallel = typename __policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_parallel;
+using __allow_parallel =
+    typename __internal::__policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_parallel;
 
 template <typename _ExecutionPolicy, typename... _IteratorTypes>
 auto
 __is_vectorization_preferred(_ExecutionPolicy&& __exec)
-    -> decltype(__lazy_and(__exec.__allow_vector(), typename __is_random_access_iterator<_IteratorTypes...>::type()))
+    -> decltype(__internal::__lazy_and(__exec.__allow_vector(),
+                                       typename __internal::__is_random_access_iterator<_IteratorTypes...>::type()))
 {
-    return __lazy_and(__exec.__allow_vector(), typename __is_random_access_iterator<_IteratorTypes...>::type());
+    return __internal::__lazy_and(__exec.__allow_vector(),
+                                  typename __internal::__is_random_access_iterator<_IteratorTypes...>::type());
 }
 
 template <typename _ExecutionPolicy, typename... _IteratorTypes>
 auto
 __is_parallelization_preferred(_ExecutionPolicy&& __exec)
-    -> decltype(__lazy_and(__exec.__allow_parallel(), typename __is_random_access_iterator<_IteratorTypes...>::type()))
+    -> decltype(__internal::__lazy_and(__exec.__allow_parallel(),
+                                       typename __internal::__is_random_access_iterator<_IteratorTypes...>::type()))
 {
-    return __lazy_and(__exec.__allow_parallel(), typename __is_random_access_iterator<_IteratorTypes...>::type());
+    return __internal::__lazy_and(__exec.__allow_parallel(),
+                                  typename __internal::__is_random_access_iterator<_IteratorTypes...>::type());
 }
 
 template <typename policy, typename... _IteratorTypes>
 struct __prefer_unsequenced_tag
 {
-    static constexpr bool value =
-        __allow_unsequenced<policy>::value && __is_random_access_iterator<_IteratorTypes...>::value;
+    static constexpr bool value = __internal::__allow_unsequenced<policy>::value &&
+                                  __internal::__is_random_access_iterator<_IteratorTypes...>::value;
     typedef std::integral_constant<bool, value> type;
 };
 
 template <typename policy, typename... _IteratorTypes>
 struct __prefer_parallel_tag
 {
-    static constexpr bool value =
-        __allow_parallel<policy>::value && __is_random_access_iterator<_IteratorTypes...>::value;
+    static constexpr bool value = __internal::__allow_parallel<policy>::value &&
+                                  __internal::__is_random_access_iterator<_IteratorTypes...>::value;
     typedef std::integral_constant<bool, value> type;
 };
 
 } // namespace __internal
 } // namespace __pstl
 
 #endif /* __PSTL_execution_impl_H */

pstl/trunk/include/pstl/internal/glue_algorithm_impl.h

 
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _Tp>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 replace_if(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
            const _Tp& __new_value)
 {
     using namespace __pstl;
     typedef typename iterator_traits<_ForwardIterator>::reference _ElementType;
-    __internal::__pattern_walk1(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last,
+    __internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                 [&__pred, &__new_value](_ElementType __elem) {
                                     if (__pred(__elem))
                                     {
                                         __elem = __new_value;
                                     }
                                 },
                                 __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
                                 __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 replace(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __old_value,
         const _Tp& __new_value)
 {
     std::replace_if(std::forward<_ExecutionPolicy>(__exec), __first, __last,

pstl/trunk/include/pstl/internal/glue_memory_impl.h

return __internal::__invoke_if_else(
             return __internal::__pattern_walk2_brick(
                 std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
                 [__is_vector](_InputIterator __begin, _InputIterator __end, _ForwardIterator __res) {
                     return __internal::__brick_copy(__begin, __end, __res, __is_vector);
                 },
                 __is_parallel);
         },
         [&]() {
-            return __internal::__pattern_walk2(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+            return __internal::__pattern_walk2(std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
                                                [](_ReferenceType1 __val1, _ReferenceType2 __val2) {
                                                    ::new (std::addressof(__val2)) _ValueType2(__val1);
                                                },
                                                __is_vector, __is_parallel);
         });
 }
 
 template <class _ExecutionPolicy, class _InputIterator, class _Size, class _ForwardIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 uninitialized_copy_n(_ExecutionPolicy&& __exec, _InputIterator __first, _Size __n, _ForwardIterator __result)
 {
     typedef typename iterator_traits<_InputIterator>::value_type _ValueType1;
return __internal::__invoke_if_else(
             return __internal::__pattern_walk2_brick_n(
                 std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
                 [__is_vector](_InputIterator __begin, _Size __sz, _ForwardIterator __res) {
                     return __internal::__brick_copy_n(__begin, __sz, __res, __is_vector);
                 },
                 __is_parallel);
         },
         [&]() {
-            return __internal::__pattern_walk2_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
+            return __internal::__pattern_walk2_n(std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
                                                  [](_ReferenceType1 __val1, _ReferenceType2 __val2) {
                                                      ::new (std::addressof(__val2)) _ValueType2(__val1);
                                                  },
                                                  __is_vector, __is_parallel);
         });
 }
 
 // [uninitialized.move]
 
 template <class _ExecutionPolicy, class _InputIterator, class _ForwardIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 uninitialized_move(_ExecutionPolicy&& __exec, _InputIterator __first, _InputIterator __last, _ForwardIterator __result)
return __internal::__invoke_if_else(
             return __internal::__pattern_walk2_brick(
                 std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
                 [__is_vector](_InputIterator __begin, _InputIterator __end, _ForwardIterator __res) {
                     return __internal::__brick_copy(__begin, __end, __res, __is_vector);
                 },
                 __is_parallel);
         },
         [&]() {
-            return __internal::__pattern_walk2(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+            return __internal::__pattern_walk2(std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
                                                [](_ReferenceType1 __val1, _ReferenceType2 __val2) {
                                                    ::new (std::addressof(__val2)) _ValueType2(std::move(__val1));
                                                },
                                                __is_vector, __is_parallel);
         });
 }
 
 template <class _ExecutionPolicy, class _InputIterator, class _Size, class _ForwardIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 uninitialized_move_n(_ExecutionPolicy&& __exec, _InputIterator __first, _Size __n, _ForwardIterator __result)
 {
     typedef typename iterator_traits<_InputIterator>::value_type _ValueType1;
return __internal::__invoke_if_else(
             return __internal::__pattern_walk2_brick_n(
                 std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
                 [__is_vector](_InputIterator __begin, _Size __sz, _ForwardIterator __res) {
                     return __internal::__brick_copy_n(__begin, __sz, __res, __is_vector);
                 },
                 __is_parallel);
         },
         [&]() {
-            return __internal::__pattern_walk2_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
+            return __internal::__pattern_walk2_n(std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
                                                  [](_ReferenceType1 __val1, _ReferenceType2 __val2) {
                                                      ::new (std::addressof(__val2)) _ValueType2(std::move(__val1));
                                                  },
                                                  __is_vector, __is_parallel);
         });
 }
 
 // [uninitialized.fill]
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 uninitialized_fill(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value)
 {
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
     using namespace __pstl;
 
     const auto __is_parallel = __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
     const auto __is_vector = __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
 
-    __internal::__invoke_if_else(
-        std::is_arithmetic<_ValueType>(),
+    __internal::__invoke_if_else(std::is_arithmetic<_ValueType>(),
                                  [&]() {
                                      __internal::__pattern_walk_brick(
                                          std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                          [&__value, &__is_vector](_ForwardIterator __begin, _ForwardIterator __end) {
                                              __internal::__brick_fill(__begin, __end, _ValueType(__value), __is_vector);
                                          },
                                          __is_parallel);
                                  },
                                  [&]() {
-            __internal::__pattern_walk1(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                [&__value](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(__value); }, __is_vector,
-                __is_parallel);
+                                     __internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first,
+                                                                 __last,
+                                                                 [&__value](_ReferenceType __val) {
+                                                                     ::new (std::addressof(__val)) _ValueType(__value);
+                                                                 },
+                                                                 __is_vector, __is_parallel);
                                  });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Tp>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 uninitialized_fill_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __n, const _Tp& __value)
 {
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
destroy(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last)
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
     using namespace __pstl;
 
     const auto __is_parallel = __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
     const auto __is_vector = __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
 
     __internal::__invoke_if_not(std::is_trivially_destructible<_ValueType>(), [&]() {
-        __internal::__pattern_walk1(
-            std::forward<_ExecutionPolicy>(__exec), __first, __last, [](_ReferenceType __val) { __val.~_ValueType(); },
-            __is_vector, __is_parallel);
+        __internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                    [](_ReferenceType __val) { __val.~_ValueType(); }, __is_vector, __is_parallel);
     });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 destroy_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __n)
 {
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
     using namespace __pstl;
 
     const auto __is_parallel = __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
     const auto __is_vector = __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
 
     return __internal::__invoke_if_else(
         std::is_trivially_destructible<_ValueType>(), [&]() { return std::next(__first, __n); },
         [&]() {
-            return __internal::__pattern_walk1_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n, [](_ReferenceType __val) { __val.~_ValueType(); },
-                __is_vector, __is_parallel);
+            return __internal::__pattern_walk1_n(std::forward<_ExecutionPolicy>(__exec), __first, __n,
+                                                 [](_ReferenceType __val) { __val.~_ValueType(); }, __is_vector,
+                                                 __is_parallel);
         });
 }
 
 // [uninitialized.construct.default]
 
 template <class _ExecutionPolicy, class _ForwardIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 uninitialized_default_construct(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last)
 {
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
     using namespace __pstl;
 
     const auto __is_parallel = __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
     const auto __is_vector = __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
 
     __internal::__invoke_if_not(std::is_trivial<_ValueType>(), [&]() {
-        __internal::__pattern_walk1(
-            std::forward<_ExecutionPolicy>(__exec), __first, __last,
-            [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; }, __is_vector, __is_parallel);
+        __internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                    [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; }, __is_vector,
+                                    __is_parallel);
     });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 uninitialized_default_construct_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __n)
 {
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
     using namespace __pstl;
 
     const auto __is_parallel = __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
     const auto __is_vector = __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
 
-    return __internal::__invoke_if_else(
-        std::is_trivial<_ValueType>(), [&]() { return std::next(__first, __n); },
+    return __internal::__invoke_if_else(std::is_trivial<_ValueType>(), [&]() { return std::next(__first, __n); },
                                         [&]() {
                                             return __internal::__pattern_walk1_n(
                                                 std::forward<_ExecutionPolicy>(__exec), __first, __n,
-                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; }, __is_vector, __is_parallel);
+                                                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; },
+                                                __is_vector, __is_parallel);
                                         });
 }
 
 // [uninitialized.construct.value]
 
 template <class _ExecutionPolicy, class _ForwardIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 uninitialized_value_construct(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last)
 {
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
     using namespace __pstl;
 
     const auto __is_parallel = __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
     const auto __is_vector = __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
 
     __internal::__invoke_if_else(
         std::is_trivial<_ValueType>(),
         [&]() {
-            __internal::__pattern_walk_brick(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last,
+            __internal::__pattern_walk_brick(std::forward<_ExecutionPolicy>(__exec), __first, __last,
                                              [__is_vector](_ForwardIterator __begin, _ForwardIterator __end) {
                                                  __internal::__brick_fill(__begin, __end, _ValueType(), __is_vector);
                                              },
                                              __is_parallel);
         },
         [&]() {
-            __internal::__pattern_walk1(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(); }, __is_vector, __is_parallel);
+            __internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                        [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(); },
+                                        __is_vector, __is_parallel);
         });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 uninitialized_value_construct_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __n)
 {
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
     using namespace __pstl;
 
     const auto __is_parallel = __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
     const auto __is_vector = __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
 
     return __internal::__invoke_if_else(
         std::is_trivial<_ValueType>(),
         [&]() {
-            return __internal::__pattern_walk_brick_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n,
+            return __internal::__pattern_walk_brick_n(std::forward<_ExecutionPolicy>(__exec), __first, __n,
                                                       [__is_vector](_ForwardIterator __begin, _Size __count) {
-                    return __internal::__brick_fill_n(__begin, __count, _ValueType(), __is_vector);
+                                                          return __internal::__brick_fill_n(__begin, __count,
+                                                                                            _ValueType(), __is_vector);
                                                       },
                                                       __is_parallel);
         },
         [&]() {
             return __internal::__pattern_walk1_n(
                 std::forward<_ExecutionPolicy>(__exec), __first, __n,
                 [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(); }, __is_vector, __is_parallel);
         });
 }
 
 } // namespace std
 
 #endif /* __PSTL_glue_memory_imple_H */

pstl/trunk/include/pstl/internal/numeric_impl.h

 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Tp,
           class _BinaryOperation1, class _BinaryOperation2, class _IsVector>
 _Tp
 __pattern_transform_reduce(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
                            _RandomAccessIterator2 __first2, _Tp __init, _BinaryOperation1 __binary_op1,
                            _BinaryOperation2 __binary_op2, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         return __par_backend::__parallel_transform_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
             [__first1, __first2, __binary_op2](_RandomAccessIterator1 __i) mutable {
                 return __binary_op2(*__i, *(__first2 + (__i - __first1)));
             },
             __init,
             __binary_op1, // Combine
             [__first1, __first2, __binary_op1, __binary_op2,
              __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j, _Tp __init) -> _Tp {
-                return __brick_transform_reduce(__i, __j, __first2 + (__i - __first1), __init, __binary_op1,
+                return __internal::__brick_transform_reduce(__i, __j, __first2 + (__i - __first1), __init, __binary_op1,
                                                             __binary_op2, __is_vector);
             });
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // transform_reduce (version with unary and binary functions)
 //------------------------------------------------------------------------

 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation,
           class _IsVector>
 _Tp
 __pattern_transform_reduce(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Tp __init,
                            _BinaryOperation __binary_op, _UnaryOperation __unary_op, _IsVector __is_vector,
                            /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_transform_reduce(__first, __last, __init, __binary_op, __unary_op, __is_vector);
+    return __internal::__brick_transform_reduce(__first, __last, __init, __binary_op, __unary_op, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation,
           class _IsVector>
 _Tp
 __pattern_transform_reduce(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Tp __init,
                            _BinaryOperation __binary_op, _UnaryOperation __unary_op, _IsVector __is_vector,
                            /*is_parallel=*/std::true_type)
 {
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         return __par_backend::__parallel_transform_reduce(
             std::forward<_ExecutionPolicy>(__exec), __first, __last,
             [__unary_op](_ForwardIterator __i) mutable { return __unary_op(*__i); }, __init, __binary_op,
             [__unary_op, __binary_op, __is_vector](_ForwardIterator __i, _ForwardIterator __j, _Tp __init) {
-                return __brick_transform_reduce(__i, __j, __init, __binary_op, __unary_op, __is_vector);
+                return __internal::__brick_transform_reduce(__i, __j, __init, __binary_op, __unary_op, __is_vector);
             });
     });
 }
 #endif
 
 //------------------------------------------------------------------------
 // transform_exclusive_scan
 //
__brick_transform_scan(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
                        _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op, _Inclusive,
                        /*is_vector=*/std::true_type) noexcept
 {
 #if (__PSTL_UDS_PRESENT)
     return __unseq_backend::__simd_scan(__first, __last - __first, __result, __unary_op, __init, __binary_op,
                                         _Inclusive());
 #else
     // We need to call serial brick here to call function for inclusive and exclusive scan that depends on _Inclusive() value
-    return __brick_transform_scan(__first, __last, __result, __unary_op, __init, __binary_op, _Inclusive(),
+    return __internal::__brick_transform_scan(__first, __last, __result, __unary_op, __init, __binary_op, _Inclusive(),
                                               /*is_vector=*/std::false_type());
 #endif
 }
 
 template <class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp, class _BinaryOperation,
           class _Inclusive>
 typename std::enable_if<is_arithmetic_udop<_Tp, _BinaryOperation>::value, std::pair<_OutputIterator, _Tp>>::type
 __brick_transform_scan(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
                        _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op, _Inclusive,
                        /*is_vector=*/std::true_type) noexcept
 {
-    return __brick_transform_scan(__first, __last, __result, __unary_op, __init, __binary_op, _Inclusive(),
+    return __internal::__brick_transform_scan(__first, __last, __result, __unary_op, __init, __binary_op, _Inclusive(),
                                               /*is_vector=*/std::false_type());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
           class _BinaryOperation, class _Inclusive, class _IsVector>
 _OutputIterator
 __pattern_transform_scan(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
                          _OutputIterator __result, _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op,
                          _Inclusive, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
 {
-    return __brick_transform_scan(__first, __last, __result, __unary_op, __init, __binary_op, _Inclusive(), __is_vector)
+    return __internal::__brick_transform_scan(__first, __last, __result, __unary_op, __init, __binary_op, _Inclusive(),
+                                              __is_vector)
         .first;
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
           class _BinaryOperation, class _Inclusive, class _IsVector>
 typename std::enable_if<!std::is_floating_point<_Tp>::value, _OutputIterator>::type
 __pattern_transform_scan(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                          _OutputIterator __result, _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op,
                          _Inclusive, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
 
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::__parallel_transform_scan(
             std::forward<_ExecutionPolicy>(__exec), __last - __first,
             [__first, __unary_op](_DifferenceType __i) mutable { return __unary_op(__first[__i]); }, __init,
             __binary_op,
             [__first, __unary_op, __binary_op](_DifferenceType __i, _DifferenceType __j, _Tp __init) {
                 // Execute serial __brick_transform_reduce, due to the explicit SIMD vectorization (reduction) requires a commutative operation for the guarantee of correct scan.
-                return __brick_transform_reduce(__first + __i, __first + __j, __init, __binary_op, __unary_op,
+                return __internal::__brick_transform_reduce(__first + __i, __first + __j, __init, __binary_op,
+                                                            __unary_op,
                                                             /*__is_vector*/ std::false_type());
             },
             [__first, __unary_op, __binary_op, __result, __is_vector](_DifferenceType __i, _DifferenceType __j,
                                                                       _Tp __init) {
-                return __brick_transform_scan(__first + __i, __first + __j, __result + __i, __unary_op, __init,
-                                              __binary_op, _Inclusive(), __is_vector)
+                return __internal::__brick_transform_scan(__first + __i, __first + __j, __result + __i, __unary_op,
+                                                          __init, __binary_op, _Inclusive(), __is_vector)
                     .second;
             });
         return __result + (__last - __first);
     });
 }
 #endif
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
           class _BinaryOperation, class _Inclusive, class _IsVector>
 typename std::enable_if<std::is_floating_point<_Tp>::value, _OutputIterator>::type
 __pattern_transform_scan(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
                          _OutputIterator __result, _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op,
                          _Inclusive, _IsVector __is_vector, /*is_parallel=*/std::true_type)
 {
     typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
     _DifferenceType __n = __last - __first;
 
     if (__n <= 0)
     {
         return __result;
     }
-    return __except_handler([&]() {
+    return __internal::__except_handler([&]() {
         __par_backend::parallel_strict_scan(
             std::forward<_ExecutionPolicy>(__exec), __n, __init,
             [__first, __unary_op, __binary_op, __result, __is_vector](_DifferenceType __i, _DifferenceType __len) {
-                return __brick_transform_scan(__first + __i, __first + (__i + __len), __result + __i, __unary_op, _Tp{},
-                                              __binary_op, _Inclusive(), __is_vector)
+                return __internal::__brick_transform_scan(__first + __i, __first + (__i + __len), __result + __i,
+                                                          __unary_op, _Tp{}, __binary_op, _Inclusive(), __is_vector)
                     .second;
             },
             __binary_op,
             [__result, &__binary_op](_DifferenceType __i, _DifferenceType __len, _Tp __initial) {
                 return *(std::transform(__result + __i, __result + __i + __len, __result + __i,
                                         [&__initial, &__binary_op](const _Tp& __x) {
                                             __PSTL_PRAGMA_FORCEINLINE
                                             return __binary_op(__initial, __x);

 
 template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryOperation,
           class _IsVector>
 _OutputIterator
 __pattern_adjacent_difference(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
                               _OutputIterator __d_first, _BinaryOperation __op, _IsVector __is_vector,
                               /*is_parallel*/ std::false_type) noexcept
 {
-    return __brick_adjacent_difference(__first, __last, __d_first, __op, __is_vector);
+    return __internal::__brick_adjacent_difference(__first, __last, __d_first, __op, __is_vector);
 }
 
 #if __PSTL_USE_PAR_POLICIES
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryOperation,
           class _IsVector>
 _ForwardIterator2
 __pattern_adjacent_difference(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last,
                               _ForwardIterator2 __d_first, _BinaryOperation __op, _IsVector __is_vector,
                               /*is_parallel=*/std::true_type)
 {
     assert(__first != __last);
     typedef typename std::iterator_traits<_ForwardIterator1>::reference _ReferenceType1;
     typedef typename std::iterator_traits<_ForwardIterator2>::reference _ReferenceType2;
 
     *__d_first = *__first;
     __par_backend::__parallel_for(
         std::forward<_ExecutionPolicy>(__exec), __first, __last - 1,
         [&__op, __is_vector, __d_first, __first](_ForwardIterator1 __b, _ForwardIterator1 __e) {
             _ForwardIterator2 __d_b = __d_first + (__b - __first);
-            __brick_walk3(
+            __internal::__brick_walk3(
                 __b, __e, __b + 1, __d_b + 1,
                 [&__op](_ReferenceType1 __x, _ReferenceType1 __y, _ReferenceType2 __z) { __z = __op(__y, __x); },
                 __is_vector);
         });
     return __d_first + (__last - __first);
 }
 #endif
 
 } // namespace __internal
 } // namespace __pstl
 
 #endif /* __PSTL_numeric_impl_H */

pstl/trunk/include/pstl/internal/parallel_backend_tbb.h

struct __par_trans_red_body
     }
 };
 
 template <class _ExecutionPolicy, class _Index, class _Up, class _Tp, class _Cp, class _Rp>
 _Tp
 __parallel_transform_reduce(_ExecutionPolicy&&, _Index __first, _Index __last, _Up __u, _Tp __init, _Cp __combine,
                             _Rp __brick_reduce)
 {
-    __par_trans_red_body<_Index, _Up, _Tp, _Cp, _Rp> __body(__u, __init, __combine, __brick_reduce);
+    __par_backend::__par_trans_red_body<_Index, _Up, _Tp, _Cp, _Rp> __body(__u, __init, __combine, __brick_reduce);
     // The grain size of 3 is used in order to provide mininum 2 elements for each body
     tbb::this_task_arena::isolate(
         [__first, __last, &__body]() { tbb::parallel_reduce(tbb::blocked_range<_Index>(__first, __last, 3), __body); });
     return __body.sum();
 }
 
 //------------------------------------------------------------------------
 // parallel_scan

 __upsweep(_Index __i, _Index __m, _Index __tilesize, _Tp* __r, _Index __lastsize, _Rp __reduce, _Cp __combine)
 {
     if (__m == 1)
         __r[0] = __reduce(__i * __tilesize, __lastsize);
     else
     {
         _Index __k = __split(__m);
         tbb::parallel_invoke(
-            [=] { __upsweep(__i, __k, __tilesize, __r, __tilesize, __reduce, __combine); },
-            [=] { __upsweep(__i + __k, __m - __k, __tilesize, __r + __k, __lastsize, __reduce, __combine); });
+            [=] { __par_backend::__upsweep(__i, __k, __tilesize, __r, __tilesize, __reduce, __combine); },
+            [=] {
+                __par_backend::__upsweep(__i + __k, __m - __k, __tilesize, __r + __k, __lastsize, __reduce, __combine);
+            });
         if (__m == 2 * __k)
             __r[__m - 1] = __combine(__r[__k - 1], __r[__m - 1]);
     }
 }
 
 template <typename _Index, typename _Tp, typename _Cp, typename _Sp>
 void
 __downsweep(_Index __i, _Index __m, _Index __tilesize, _Tp* __r, _Index __lastsize, _Tp __initial, _Cp __combine,
             _Sp __scan)
 {
     if (__m == 1)
         __scan(__i * __tilesize, __lastsize, __initial);
     else
     {
         const _Index __k = __split(__m);
-        tbb::parallel_invoke([=] { __downsweep(__i, __k, __tilesize, __r, __tilesize, __initial, __combine, __scan); },
-                             // Assumes that __combine never throws.
-                             //TODO: Consider adding a requirement for user functors to be constant.
-                             [=, &__combine] {
-                                 __downsweep(__i + __k, __m - __k, __tilesize, __r + __k, __lastsize,
-                                             __combine(__initial, __r[__k - 1]), __combine, __scan);
-                             });
+        tbb::parallel_invoke(
+            [=] { __par_backend::__downsweep(__i, __k, __tilesize, __r, __tilesize, __initial, __combine, __scan); },
+                __par_backend::__downsweep(__i + __k, __m - __k, __tilesize, __r + __k, __lastsize,
     }
 }
 
 // Adapted from Intel(R) Cilk(TM) version from cilkpub.
 // Let i:len denote a counted interval of length n starting at i.  s denotes a generalized-sum value.
 // Expected actions of the functors are:
 //     reduce(i,len) -> s  -- return reduction value of i:len.
 //     combine(s1,s2) -> s -- return merged sum
tbb::this_task_arena::isolate([=, &__combine]() {
         if (__n > 1)
         {
             _Index __p = tbb::this_task_arena::max_concurrency();
             const _Index __slack = 4;
             _Index __tilesize = (__n - 1) / (__slack * __p) + 1;
             _Index __m = (__n - 1) / __tilesize;
             __buffer<_Tp> __buf(__m + 1);
             _Tp* __r = __buf.get();
-            __upsweep(_Index(0), _Index(__m + 1), __tilesize, __r, __n - __m * __tilesize, __reduce, __combine);
+            __par_backend::__upsweep(_Index(0), _Index(__m + 1), __tilesize, __r, __n - __m * __tilesize, __reduce,
+                                     __combine);
 
             // When __apex is a no-op and __combine has no side effects, a good optimizer
             // should be able to eliminate all code between here and __apex.
             // Alternatively, provide a default value for __apex that can be
             // recognized by metaprogramming that conditionlly executes the following.
             size_t __k = __m + 1;
             _Tp __t = __r[__k - 1];
             while ((__k &= __k - 1))
                 __t = __combine(__r[__k - 1], __t);
             __apex(__combine(__initial, __t));
-            __downsweep(_Index(0), _Index(__m + 1), __tilesize, __r, __n - __m * __tilesize, __initial, __combine,
-                        __scan);
+            __par_backend::__downsweep(_Index(0), _Index(__m + 1), __tilesize, __r, __n - __m * __tilesize, __initial,
+                                       __combine, __scan);
             return;
         }
         // Fewer than 2 elements in sequence, or out of memory.  Handle has single block.
         _Tp __sum = __initial;
         if (__n)
             __sum = __combine(__sum, __reduce(_Index(0), __n));
         __apex(__sum);
         if (__n)

 {
     const _SizeType __n = _M_xe - _M_xs;
     const _SizeType __nmerge = _M_nsort > 0 ? _M_nsort : __n;
     const _SizeType __sort_cut_off = __PSTL_STABLE_SORT_CUT_OFF;
     if (__n <= __sort_cut_off)
     {
         _M_leaf_sort(_M_xs, _M_xe, _M_comp);
         if (_M_inplace != 2)
-            __init_buf(_M_xs, _M_xe, _M_zs, _M_inplace == 0);
+            __par_backend::__init_buf(_M_xs, _M_xe, _M_zs, _M_inplace == 0);
         return NULL;
     }
     else
     {
         const _RandomAccessIterator1 __xm = _M_xs + __n / 2;
         const _RandomAccessIterator2 __zm = _M_zs + (__xm - _M_xs);
         const _RandomAccessIterator2 __ze = _M_zs + __n;
         task* __m;
         auto __move_values = [](_RandomAccessIterator2 __x, _RandomAccessIterator1 __z) { *__z = std::move(*__x); };
         auto __move_sequences = [](_RandomAccessIterator2 __first1, _RandomAccessIterator2 __last1,
                                    _RandomAccessIterator1 __first2) { return std::move(__first1, __last1, __first2); };
         if (_M_inplace == 2)
-            __m = new (allocate_continuation())
+            __m = new (tbb::task::allocate_continuation())
                 __merge_task<_RandomAccessIterator2, _RandomAccessIterator2, _RandomAccessIterator1, _Compare,
                              __serial_destroy,
-                             __serial_move_merge<decltype(__move_values), decltype(__move_sequences)>>(
+                             __par_backend::__serial_move_merge<decltype(__move_values), decltype(__move_sequences)>>(
                     _M_zs, __zm, __zm, __ze, _M_xs, _M_comp, __serial_destroy(),
-                    __serial_move_merge<decltype(__move_values), decltype(__move_sequences)>(__nmerge, __move_values,
-                                                                                             __move_sequences));
+                    __par_backend::__serial_move_merge<decltype(__move_values), decltype(__move_sequences)>(
+                        __nmerge, __move_values, __move_sequences));
         else if (_M_inplace)
-            __m = new (allocate_continuation())
+            __m = new (tbb::task::allocate_continuation())
                 __merge_task<_RandomAccessIterator2, _RandomAccessIterator2, _RandomAccessIterator1, _Compare,
-                             __binary_no_op, __serial_move_merge<decltype(__move_values), decltype(__move_sequences)>>(
-                    _M_zs, __zm, __zm, __ze, _M_xs, _M_comp, __binary_no_op(),
-                    __serial_move_merge<decltype(__move_values), decltype(__move_sequences)>(__nmerge, __move_values,
-                                                                                             __move_sequences));
+                             __par_backend::__binary_no_op,
+                             __par_backend::__serial_move_merge<decltype(__move_values), decltype(__move_sequences)>>(
+                    _M_zs, __zm, __zm, __ze, _M_xs, _M_comp, __par_backend::__binary_no_op(),
+                    __par_backend::__serial_move_merge<decltype(__move_values), decltype(__move_sequences)>(
+                        __nmerge, __move_values, __move_sequences));
         else
         {
             auto __move_values = [](_RandomAccessIterator1 __x, _RandomAccessIterator2 __z) { *__z = std::move(*__x); };
             auto __move_sequences = [](_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
                                        _RandomAccessIterator2 __first2) {
                 return std::move(__first1, __last1, __first2);
             };
-            __m = new (allocate_continuation())
+            __m = new (tbb::task::allocate_continuation())
                 __merge_task<_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2, _Compare,
-                             __binary_no_op, __serial_move_merge<decltype(__move_values), decltype(__move_sequences)>>(
-                    _M_xs, __xm, __xm, _M_xe, _M_zs, _M_comp, __binary_no_op(),
-                    __serial_move_merge<decltype(__move_values), decltype(__move_sequences)>(__nmerge, __move_values,
-                                                                                             __move_sequences));
+                             __par_backend::__binary_no_op,
+                             __par_backend::__serial_move_merge<decltype(__move_values), decltype(__move_sequences)>>(
+                    _M_xs, __xm, __xm, _M_xe, _M_zs, _M_comp, __par_backend::__binary_no_op(),
+                    __par_backend::__serial_move_merge<decltype(__move_values), decltype(__move_sequences)>(
+                        __nmerge, __move_values, __move_sequences));
         }
         __m->set_ref_count(2);
         task* __right = new (__m->allocate_child())
             __stable_sort_task(__xm, _M_xe, __zm, !_M_inplace, _M_comp, _M_leaf_sort, __nmerge);
-        spawn(*__right);
-        recycle_as_child_of(*__m);
+        tbb::task::spawn(*__right);
+        tbb::task::recycle_as_child_of(*__m);
         _M_xe = __xm;
         _M_inplace = !_M_inplace;
     }
     return this;
 }
 
 template <class _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare, typename _LeafSort>
 void
__parallel_merge(_ExecutionPolicy&&, _RandomAccessIterator1 __xs, _RandomAccessIterator1 __xe,
     {
         // Fall back on serial merge
         __leaf_merge(__xs, __xe, __ys, __ye, __zs, __comp);
     }
     else
     {
         tbb::this_task_arena::isolate([=]() {
             typedef __merge_task<_RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator3, _Compare,
-                                 __binary_no_op, _LeafMerge>
+                                 __par_backend::__binary_no_op, _LeafMerge>
                 _TaskType;
             tbb::task::spawn_root_and_wait(*new (tbb::task::allocate_root()) _TaskType(
-                __xs, __xe, __ys, __ye, __zs, __comp, __binary_no_op(), __leaf_merge));
+                __xs, __xe, __ys, __ye, __zs, __comp, __par_backend::__binary_no_op(), __leaf_merge));
         });
     }
 }
 
 //------------------------------------------------------------------------
 // parallel_invoke
 //------------------------------------------------------------------------
 template <class _ExecutionPolicy, typename _F1, typename _F2>

pstl/trunk/include/pstl/internal/unseq_backend_simd.h

__simd_find_first_of(_ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
 
     // Common case
     // If first sequence larger than second then we'll run simd_first with parameters of first sequence.
     // Otherwise, vice versa.
     if (__n1 < __n2)
     {
         for (; __first != __last; ++__first)
         {
-            if (__simd_or(__s_first, __n2,
-                          __internal::__equal_value_by_pred<decltype(*__first), _BinaryPredicate>(*__first, __pred)))
+            if (__unseq_backend::__simd_or(
+                    __s_first, __n2,
             {
                 return __first;
             }
         }
     }
     else
     {
         for (; __s_first != __s_last; ++__s_first)
         {
-            const auto __result = __simd_first(__first, _DifferencType(0), __n1,
-                                               [__s_first, &__pred](_ForwardIterator1 __it, _DifferencType __i) {
-                                                   return __pred(__it[__i], *__s_first);
-                                               });
+            const auto __result = __unseq_backend::__simd_first(
+                __first, _DifferencType(0), __n1, [__s_first, &__pred](_ForwardIterator1 __it, _DifferencType __i) {
             if (__result != __last)
             {
                 return __result;
             }
         }
     }
     return __last;
 }
 
 template <class _RandomAccessIterator, class _DifferenceType, class _UnaryPredicate>
 _RandomAccessIterator
 __simd_remove_if(_RandomAccessIterator __first, _DifferenceType __n, _UnaryPredicate __pred) noexcept
 {
     // find first element we need to remove
-    auto __current =
-        __simd_first(__first, _DifferenceType(0), __n,
-                     [&__pred](_RandomAccessIterator __it, _DifferenceType __i) { return __pred(__it[__i]); });
+    auto __current = __unseq_backend::__simd_first(
+        __first, _DifferenceType(0), __n,
     __n -= __current - __first;
 
     // if we have in sequence only one element that pred(__current[1]) != false we can exit the function
     if (__n < 2)
     {
         return __current;
     }