public inbox for libstdc++-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r14-2109] libstdc++: Synchronize PSTL with upstream
@ 2023-06-26 18:35 Thomas Rodgers
  0 siblings, 0 replies; only message in thread
From: Thomas Rodgers @ 2023-06-26 18:35 UTC (permalink / raw)
  To: gcc-cvs, libstdc++-cvs

https://gcc.gnu.org/g:3162ca09dbdc2ea3ff1d4b752edf76955b6d94f9

commit r14-2109-g3162ca09dbdc2ea3ff1d4b752edf76955b6d94f9
Author: Thomas Rodgers <trodgers@redhat.com>
Date:   Mon Jun 26 11:30:21 2023 -0700

    libstdc++: Synchronize PSTL with upstream
    
    This patch rebases the C++17 parallel algorithms implementation (pstl)
    against the current upstream version, commit 843c12d6a.
    
    This version does not currently include the recently added OpenMP
    backend, that will be considered for a future version.
    
    libstdc++-v3/ChangeLog:
            * include/pstl/algorithm_fwd.h: Synchronize with upstream.
            * include/pstl/algorithm_impl.h: Likewise.
            * include/pstl/execution_defs.h: Likewise.
            * include/pstl/execution_impl.h: Likewise.
            * include/pstl/glue_algorithm_impl.h: Likewise.
            * include/pstl/glue_execution_defs.h: Likewise.
            * include/pstl/glue_memory_impl.h: Likewise.
            * include/pstl/glue_numeric_impl.h: Likewise.
            * include/pstl/memory_impl.h: Likewise.
            * include/pstl/numeric_fwd.h: Likewise.
            * include/pstl/numeric_impl.h: Likewise.
            * include/pstl/parallel_backend.h: Likewise.
            * include/pstl/parallel_backend_serial.h: Likewise.
            * include/pstl/parallel_backend_tbb.h: Likewise.
            * include/pstl/parallel_impl.h: Likewise.
            * include/pstl/pstl_config.h: Likewise.
            * include/pstl/unseq_backend_simd.h: Likewise.
            * include/pstl/utils.h: Likewise.
            * testsuite/20_util/specialized_algorithms/pstl/uninitialized_construct.cc:
            Likewise.
            * testsuite/20_util/specialized_algorithms/pstl/uninitialized_copy_move.cc:
            Likewise.
            * testsuite/20_util/specialized_algorithms/pstl/uninitialized_fill_destroy.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_merge/inplace_merge.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_merge/merge.cc: Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/copy_if.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/copy_move.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/fill.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/generate.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/is_partitioned.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/partition.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/partition_copy.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/remove.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/remove_copy.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/replace.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/replace_copy.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/rotate.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/rotate_copy.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/swap_ranges.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/transform_binary.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/transform_unary.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/unique.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_modifying_operations/unique_copy_equal.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/adjacent_find.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/all_of.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/any_of.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/count.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/equal.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/find.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/find_end.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/find_first_of.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/find_if.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/for_each.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/mismatch.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/none_of.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/nth_element.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/reverse.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/reverse_copy.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_nonmodifying/search_n.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/includes.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/is_heap.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/is_sorted.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/lexicographical_compare.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/minmax_element.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/partial_sort.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/partial_sort_copy.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/set.cc:
            Likewise.
            * testsuite/25_algorithms/pstl/alg_sorting/sort.cc:
            Likewise.
            * testsuite/26_numerics/pstl/numeric_ops/adjacent_difference.cc:
            Likewise.
            * testsuite/26_numerics/pstl/numeric_ops/reduce.cc:
            Likewise.
            * testsuite/26_numerics/pstl/numeric_ops/scan.cc:
            Likewise.
            * testsuite/26_numerics/pstl/numeric_ops/transform_reduce.cc:
            Likewise.
            * testsuite/26_numerics/pstl/numeric_ops/transform_scan.cc:
            Likewise.
            * testsuite/util/pstl/test_utils.h:
            Likewise.

Diff:
---
 libstdc++-v3/include/pstl/algorithm_fwd.h          |  951 +++----
 libstdc++-v3/include/pstl/algorithm_impl.h         | 2830 ++++++++++----------
 libstdc++-v3/include/pstl/execution_defs.h         |   76 +-
 libstdc++-v3/include/pstl/execution_impl.h         |  155 +-
 libstdc++-v3/include/pstl/glue_algorithm_impl.h    |  545 ++--
 libstdc++-v3/include/pstl/glue_execution_defs.h    |    8 +-
 libstdc++-v3/include/pstl/glue_memory_impl.h       |  293 +-
 libstdc++-v3/include/pstl/glue_numeric_impl.h      |   83 +-
 libstdc++-v3/include/pstl/memory_impl.h            |   20 +-
 libstdc++-v3/include/pstl/numeric_fwd.h            |   94 +-
 libstdc++-v3/include/pstl/numeric_impl.h           |  309 ++-
 libstdc++-v3/include/pstl/parallel_backend.h       |    6 +
 .../include/pstl/parallel_backend_serial.h         |   29 +-
 libstdc++-v3/include/pstl/parallel_backend_tbb.h   |   30 +-
 libstdc++-v3/include/pstl/parallel_impl.h          |   19 +-
 libstdc++-v3/include/pstl/pstl_config.h            |   79 +-
 libstdc++-v3/include/pstl/unseq_backend_simd.h     |   22 +-
 libstdc++-v3/include/pstl/utils.h                  |   65 +-
 .../pstl/uninitialized_construct.cc                |    4 +-
 .../pstl/uninitialized_copy_move.cc                |   14 +-
 .../pstl/uninitialized_fill_destroy.cc             |    2 +-
 .../25_algorithms/pstl/alg_merge/inplace_merge.cc  |   22 +-
 .../25_algorithms/pstl/alg_merge/merge.cc          |    8 +-
 .../pstl/alg_modifying_operations/copy_if.cc       |   10 +-
 .../pstl/alg_modifying_operations/copy_move.cc     |   37 +-
 .../pstl/alg_modifying_operations/fill.cc          |    4 +-
 .../pstl/alg_modifying_operations/generate.cc      |   15 +-
 .../alg_modifying_operations/is_partitioned.cc     |   12 +-
 .../pstl/alg_modifying_operations/partition.cc     |   19 +-
 .../alg_modifying_operations/partition_copy.cc     |   19 +-
 .../pstl/alg_modifying_operations/remove.cc        |   31 +-
 .../pstl/alg_modifying_operations/remove_copy.cc   |   11 +-
 .../pstl/alg_modifying_operations/replace.cc       |   15 +-
 .../pstl/alg_modifying_operations/replace_copy.cc  |    8 +-
 .../pstl/alg_modifying_operations/rotate.cc        |   14 +-
 .../pstl/alg_modifying_operations/rotate_copy.cc   |   10 +-
 .../pstl/alg_modifying_operations/swap_ranges.cc   |    4 +-
 .../alg_modifying_operations/transform_binary.cc   |   11 +-
 .../alg_modifying_operations/transform_unary.cc    |    2 +-
 .../pstl/alg_modifying_operations/unique.cc        |   22 +-
 .../alg_modifying_operations/unique_copy_equal.cc  |   16 +-
 .../pstl/alg_nonmodifying/adjacent_find.cc         |   14 +-
 .../25_algorithms/pstl/alg_nonmodifying/all_of.cc  |    6 +-
 .../25_algorithms/pstl/alg_nonmodifying/any_of.cc  |    6 +-
 .../25_algorithms/pstl/alg_nonmodifying/count.cc   |    6 +-
 .../25_algorithms/pstl/alg_nonmodifying/equal.cc   |    6 +-
 .../25_algorithms/pstl/alg_nonmodifying/find.cc    |   14 +-
 .../pstl/alg_nonmodifying/find_end.cc              |  110 +-
 .../pstl/alg_nonmodifying/find_first_of.cc         |   14 +-
 .../25_algorithms/pstl/alg_nonmodifying/find_if.cc |   12 +-
 .../pstl/alg_nonmodifying/for_each.cc              |    4 +-
 .../pstl/alg_nonmodifying/mismatch.cc              |    4 +-
 .../25_algorithms/pstl/alg_nonmodifying/none_of.cc |    6 +-
 .../pstl/alg_nonmodifying/nth_element.cc           |   13 +-
 .../25_algorithms/pstl/alg_nonmodifying/reverse.cc |   14 +-
 .../pstl/alg_nonmodifying/reverse_copy.cc          |   13 +-
 .../pstl/alg_nonmodifying/search_n.cc              |   14 +-
 .../25_algorithms/pstl/alg_sorting/includes.cc     |    7 +-
 .../25_algorithms/pstl/alg_sorting/is_heap.cc      |   15 +-
 .../25_algorithms/pstl/alg_sorting/is_sorted.cc    |    4 +-
 .../pstl/alg_sorting/lexicographical_compare.cc    |    4 +-
 .../pstl/alg_sorting/minmax_element.cc             |    2 +-
 .../25_algorithms/pstl/alg_sorting/partial_sort.cc |   18 +-
 .../pstl/alg_sorting/partial_sort_copy.cc          |   19 +-
 .../25_algorithms/pstl/alg_sorting/set.cc          |  155 +-
 .../25_algorithms/pstl/alg_sorting/sort.cc         |   10 +-
 .../pstl/numeric_ops/adjacent_difference.cc        |   38 +-
 .../26_numerics/pstl/numeric_ops/reduce.cc         |    4 +-
 .../testsuite/26_numerics/pstl/numeric_ops/scan.cc |   13 +-
 .../pstl/numeric_ops/transform_reduce.cc           |   33 +-
 .../26_numerics/pstl/numeric_ops/transform_scan.cc |   24 +-
 libstdc++-v3/testsuite/util/pstl/test_utils.h      |   94 +-
 72 files changed, 3334 insertions(+), 3286 deletions(-)

diff --git a/libstdc++-v3/include/pstl/algorithm_fwd.h b/libstdc++-v3/include/pstl/algorithm_fwd.h
index e85125c4dc7..1ba9555e2a0 100644
--- a/libstdc++-v3/include/pstl/algorithm_fwd.h
+++ b/libstdc++-v3/include/pstl/algorithm_fwd.h
@@ -10,6 +10,7 @@
 #ifndef _PSTL_ALGORITHM_FWD_H
 #define _PSTL_ALGORITHM_FWD_H
 
+#include <iterator>
 #include <type_traits>
 #include <utility>
 
@@ -27,20 +28,18 @@ bool
 __brick_any_of(const _ForwardIterator, const _ForwardIterator, _Pred,
                /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _Pred>
+template <class _RandomAccessIterator, class _Pred>
 bool
-__brick_any_of(const _ForwardIterator, const _ForwardIterator, _Pred,
+__brick_any_of(const _RandomAccessIterator, const _RandomAccessIterator, _Pred,
                /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Pred, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Pred>
 bool
-__pattern_any_of(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Pred, _IsVector,
-                 /*parallel=*/std::false_type) noexcept;
+__pattern_any_of(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Pred) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Pred, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Pred>
 bool
-__pattern_any_of(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Pred, _IsVector,
-                 /*parallel=*/std::true_type);
+__pattern_any_of(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Pred);
 
 //------------------------------------------------------------------------
 // walk1 (pseudo)
@@ -56,25 +55,22 @@ template <class _RandomAccessIterator, class _Function>
 void __brick_walk1(_RandomAccessIterator, _RandomAccessIterator, _Function,
                    /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Function>
 void
-__pattern_walk1(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Function, _IsVector,
-                /*parallel=*/std::false_type) noexcept;
+__pattern_walk1(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Function) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Function>
 void
-__pattern_walk1(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Function, _IsVector,
-                /*parallel=*/std::true_type);
+__pattern_walk1(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Function);
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Brick>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Brick>
 void
-__pattern_walk_brick(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Brick,
-                     /*parallel=*/std::false_type) noexcept;
+__pattern_walk_brick(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Brick) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Brick>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Brick>
 void
-__pattern_walk_brick(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Brick,
-                     /*parallel=*/std::true_type);
+__pattern_walk_brick(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                     _Brick);
 
 //------------------------------------------------------------------------
 // walk1_n
@@ -88,25 +84,21 @@ template <class _RandomAccessIterator, class _DifferenceType, class _Function>
 _RandomAccessIterator __brick_walk1_n(_RandomAccessIterator, _DifferenceType, _Function,
                                       /*vectorTag=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Function>
 _ForwardIterator
-__pattern_walk1_n(_ExecutionPolicy&&, _ForwardIterator, _Size, _Function, _IsVector,
-                  /*is_parallel=*/std::false_type) noexcept;
+__pattern_walk1_n(_Tag, _ExecutionPolicy&&, _ForwardIterator, _Size, _Function) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Function>
 _RandomAccessIterator
-__pattern_walk1_n(_ExecutionPolicy&&, _RandomAccessIterator, _Size, _Function, _IsVector,
-                  /*is_parallel=*/std::true_type);
+__pattern_walk1_n(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _Size, _Function);
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Brick>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Brick>
 _ForwardIterator
-__pattern_walk_brick_n(_ExecutionPolicy&&, _ForwardIterator, _Size, _Brick,
-                       /*is_parallel=*/std::false_type) noexcept;
+__pattern_walk_brick_n(_Tag, _ExecutionPolicy&&, _ForwardIterator, _Size, _Brick) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Brick>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Brick>
 _RandomAccessIterator
-__pattern_walk_brick_n(_ExecutionPolicy&&, _RandomAccessIterator, _Size, _Brick,
-                       /*is_parallel=*/std::true_type);
+__pattern_walk_brick_n(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _Size, _Brick);
 
 //------------------------------------------------------------------------
 // walk2 (pseudo)
@@ -118,60 +110,60 @@ template <class _ForwardIterator1, class _ForwardIterator2, class _Function>
 _ForwardIterator2 __brick_walk2(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Function,
                                 /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Function>
-_ForwardIterator2 __brick_walk2(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Function,
-                                /*vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Function>
+_RandomAccessIterator2 __brick_walk2(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2, _Function,
+                                     /*vector=*/std::true_type) noexcept;
 
 template <class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Function>
 _ForwardIterator2 __brick_walk2_n(_ForwardIterator1, _Size, _ForwardIterator2, _Function,
                                   /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Function>
-_ForwardIterator2 __brick_walk2_n(_ForwardIterator1, _Size, _ForwardIterator2, _Function,
-                                  /*vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2, class _Function>
+_RandomAccessIterator2 __brick_walk2_n(_RandomAccessIterator1, _Size, _RandomAccessIterator2, _Function,
+                                       /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Function>
 _ForwardIterator2
-__pattern_walk2(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Function, _IsVector,
-                /*parallel=*/std::false_type) noexcept;
+__pattern_walk2(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Function) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Function, class _IsVector>
-_ForwardIterator2
-__pattern_walk2(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Function, _IsVector,
-                /*parallel=*/std::true_type);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Function>
+_RandomAccessIterator2
+__pattern_walk2(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                _RandomAccessIterator2, _Function);
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Function,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2,
+          class _Function>
 _ForwardIterator2
-__pattern_walk2_n(_ExecutionPolicy&&, _ForwardIterator1, _Size, _ForwardIterator2, _Function, _IsVector,
-                  /*parallel=*/std::false_type) noexcept;
+__pattern_walk2_n(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _Size, _ForwardIterator2, _Function) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2,
-          class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _Size,
+          class _RandomAccessIterator2, class _Function>
 _RandomAccessIterator2
-__pattern_walk2_n(_ExecutionPolicy&&, _RandomAccessIterator1, _Size, _RandomAccessIterator2, _Function, _IsVector,
-                  /*parallel=*/std::true_type);
+__pattern_walk2_n(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _Size, _RandomAccessIterator2,
+                  _Function);
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Brick>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Brick>
 _ForwardIterator2
-__pattern_walk2_brick(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Brick,
-                      /*parallel=*/std::false_type) noexcept;
+__pattern_walk2_brick(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                      _Brick) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Brick>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Brick>
 _RandomAccessIterator2
-__pattern_walk2_brick(_ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
-                      _Brick,
-                      /*parallel=*/std::true_type);
+__pattern_walk2_brick(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                      _RandomAccessIterator2, _Brick);
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Brick>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2,
+          class _Brick>
 _ForwardIterator2
-__pattern_walk2_brick_n(_ExecutionPolicy&&, _ForwardIterator1, _Size, _ForwardIterator2, _Brick,
-                        /*parallel=*/std::false_type) noexcept;
+__pattern_walk2_brick_n(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _Size, _ForwardIterator2, _Brick) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2, class _Brick>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _Size,
+          class _RandomAccessIterator2, class _Brick>
 _RandomAccessIterator2
-__pattern_walk2_brick_n(_ExecutionPolicy&&, _RandomAccessIterator1, _Size, _RandomAccessIterator2, _Brick,
-                        /*parallel=*/std::true_type);
+__pattern_walk2_brick_n(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _Size,
+                        _RandomAccessIterator2, _Brick);
 
 //------------------------------------------------------------------------
 // walk3 (pseudo)
@@ -188,18 +180,17 @@ _RandomAccessIterator3 __brick_walk3(_RandomAccessIterator1, _RandomAccessIterat
                                      _RandomAccessIterator3, _Function,
                                      /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator3,
-          class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator3,
+          class _Function>
 _ForwardIterator3
-__pattern_walk3(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator3,
-                _Function, _IsVector,
-                /*parallel=*/std::false_type) noexcept;
+__pattern_walk3(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator3,
+                _Function) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
-          class _RandomAccessIterator3, class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _RandomAccessIterator3, class _Function>
 _RandomAccessIterator3
-__pattern_walk3(_ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
-                _RandomAccessIterator3, _Function, _IsVector, /*parallel=*/std::true_type);
+__pattern_walk3(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                _RandomAccessIterator2, _RandomAccessIterator3, _Function);
 
 //------------------------------------------------------------------------
 // equal
@@ -213,17 +204,16 @@ template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Bin
 bool __brick_equal(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2, _BinaryPredicate,
                    /* is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _BinaryPredicate,
-                _IsVector, /* is_parallel = */ std::false_type) noexcept;
+__pattern_equal(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                _BinaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
-                _BinaryPredicate, _IsVector, /* is_parallel = */ std::true_type);
+__pattern_equal(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                _RandomAccessIterator2, _BinaryPredicate);
 
 template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 bool __brick_equal(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2, _BinaryPredicate,
@@ -233,17 +223,16 @@ template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Bin
 bool __brick_equal(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator2,
                    _BinaryPredicate, /* is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                _BinaryPredicate, _IsVector, /* is_parallel = */ std::false_type) noexcept;
+__pattern_equal(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
+                _BinaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
-                _RandomAccessIterator2, _BinaryPredicate, _IsVector, /* is_parallel = */ std::true_type);
+__pattern_equal(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                _RandomAccessIterator2, _RandomAccessIterator2, _BinaryPredicate);
 
 //------------------------------------------------------------------------
 // find_if
@@ -257,15 +246,14 @@ template <class _RandomAccessIterator, class _Predicate>
 _RandomAccessIterator __brick_find_if(_RandomAccessIterator, _RandomAccessIterator, _Predicate,
                                       /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
 _ForwardIterator
-__pattern_find_if(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Predicate, _IsVector,
-                  /*is_parallel=*/std::false_type) noexcept;
+__pattern_find_if(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Predicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
-_ForwardIterator
-__pattern_find_if(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Predicate, _IsVector,
-                  /*is_parallel=*/std::true_type);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Predicate>
+_RandomAccessIterator
+__pattern_find_if(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                  _Predicate);
 
 //------------------------------------------------------------------------
 // find_end
@@ -276,24 +264,21 @@ _ForwardIterator1 __brick_find_end(_ForwardIterator1, _ForwardIterator1, _Forwar
                                    _BinaryPredicate,
                                    /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
-_ForwardIterator1 __brick_find_end(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                                   _BinaryPredicate,
-                                   /*__is_vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator1 __brick_find_end(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                        _RandomAccessIterator2, _BinaryPredicate,
+                                        /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _ForwardIterator1
-__pattern_find_end(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                   _BinaryPredicate, _IsVector,
-                   /*is_parallel=*/std::false_type) noexcept;
+__pattern_find_end(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
+                   _BinaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
-_ForwardIterator1
-__pattern_find_end(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                   _BinaryPredicate, _IsVector,
-                   /*is_parallel=*/std::true_type) noexcept;
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
+_RandomAccessIterator1
+__pattern_find_end(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                   _RandomAccessIterator2, _RandomAccessIterator2, _BinaryPredicate) noexcept;
 
 //------------------------------------------------------------------------
 // find_first_of
@@ -304,22 +289,21 @@ _ForwardIterator1 __brick_find_first_of(_ForwardIterator1, _ForwardIterator1, _F
                                         _BinaryPredicate,
                                         /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
-_ForwardIterator1 __brick_find_first_of(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                                        _BinaryPredicate,
-                                        /*__is_vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator1 __brick_find_first_of(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                             _RandomAccessIterator2, _BinaryPredicate,
+                                             /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _ForwardIterator1
-__pattern_find_first_of(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                        _BinaryPredicate, _IsVector, /*is_parallel=*/std::false_type) noexcept;
+__pattern_find_first_of(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                        _ForwardIterator2, _BinaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
-_ForwardIterator1
-__pattern_find_first_of(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                        _BinaryPredicate, _IsVector, /*is_parallel=*/std::true_type) noexcept;
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
+_RandomAccessIterator1
+__pattern_find_first_of(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                        _RandomAccessIterator2, _RandomAccessIterator2, _BinaryPredicate) noexcept;
 
 //------------------------------------------------------------------------
 // search
@@ -330,24 +314,21 @@ _ForwardIterator1 __brick_search(_ForwardIterator1, _ForwardIterator1, _ForwardI
                                  _BinaryPredicate,
                                  /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
-_ForwardIterator1 __brick_search(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                                 _BinaryPredicate,
-                                 /*vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator1 __brick_search(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                      _RandomAccessIterator2, _BinaryPredicate,
+                                      /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _ForwardIterator1
-__pattern_search(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                 _BinaryPredicate, _IsVector,
-                 /*is_parallel=*/std::false_type) noexcept;
+__pattern_search(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
+                 _BinaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
-_ForwardIterator1
-__pattern_search(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                 _BinaryPredicate, _IsVector,
-                 /*is_parallel=*/std::true_type) noexcept;
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
+_RandomAccessIterator1
+__pattern_search(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                 _RandomAccessIterator2, _RandomAccessIterator2, _BinaryPredicate) noexcept;
 
 //------------------------------------------------------------------------
 // search_n
@@ -358,24 +339,21 @@ _ForwardIterator
 __brick_search_n(_ForwardIterator, _ForwardIterator, _Size, const _Tp&, _BinaryPredicate,
                  /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _Size, class _Tp, class _BinaryPredicate>
-_ForwardIterator
-__brick_search_n(_ForwardIterator, _ForwardIterator, _Size, const _Tp&, _BinaryPredicate,
+template <class _RandomAccessIterator, class _Size, class _Tp, class _BinaryPredicate>
+_RandomAccessIterator
+__brick_search_n(_RandomAccessIterator, _RandomAccessIterator, _Size, const _Tp&, _BinaryPredicate,
                  /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Tp, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Tp, class _BinaryPredicate>
 _ForwardIterator
-__pattern_search_n(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Size, const _Tp&, _BinaryPredicate,
-		   _IsVector,
-                   /*is_parallel=*/std::false_type) noexcept;
+__pattern_search_n(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Size, const _Tp&,
+                   _BinaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Tp, class _BinaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Tp,
+          class _BinaryPredicate>
 _RandomAccessIterator
-__pattern_search_n(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Size, const _Tp&,
-                   _BinaryPredicate, _IsVector,
-                   /*is_parallel=*/std::true_type) noexcept;
+__pattern_search_n(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Size,
+                   const _Tp&, _BinaryPredicate) noexcept;
 
 //------------------------------------------------------------------------
 // copy_n
@@ -385,8 +363,8 @@ template <class _ForwardIterator, class _Size, class _OutputIterator>
 _OutputIterator __brick_copy_n(_ForwardIterator, _Size, _OutputIterator,
                                /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _Size, class _OutputIterator>
-_OutputIterator __brick_copy_n(_ForwardIterator, _Size, _OutputIterator,
+template <class _RandomAccessIterator, class _Size, class _OutputIterator>
+_OutputIterator __brick_copy_n(_RandomAccessIterator, _Size, _OutputIterator,
                                /*vector=*/std::true_type) noexcept;
 
 //------------------------------------------------------------------------
@@ -417,14 +395,12 @@ _OutputIterator __brick_move(_RandomAccessIterator, _RandomAccessIterator, _Outp
 // swap_ranges
 //------------------------------------------------------------------------
 template <class _ForwardIterator, class _OutputIterator>
-_OutputIterator
-__brick_swap_ranges(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
-                    /*vector=*/std::false_type) noexcept;
+_OutputIterator __brick_swap_ranges(_ForwardIterator, _ForwardIterator, _OutputIterator,
+                                    /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _OutputIterator>
-_OutputIterator
-__brick_swap_ranges(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
-                    /*vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator, class _OutputIterator>
+_OutputIterator __brick_swap_ranges(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator,
+                                    /*vector=*/std::true_type) noexcept;
 
 //------------------------------------------------------------------------
 // copy_if
@@ -434,8 +410,8 @@ template <class _ForwardIterator, class _OutputIterator, class _UnaryPredicate>
 _OutputIterator __brick_copy_if(_ForwardIterator, _ForwardIterator, _OutputIterator, _UnaryPredicate,
                                 /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _OutputIterator, class _UnaryPredicate>
-_OutputIterator __brick_copy_if(_ForwardIterator, _ForwardIterator, _OutputIterator, _UnaryPredicate,
+template <class _RandomAccessIterator, class _OutputIterator, class _UnaryPredicate>
+_OutputIterator __brick_copy_if(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator, _UnaryPredicate,
                                 /*vector=*/std::true_type) noexcept;
 
 template <class _DifferenceType, class _ForwardIterator, class _UnaryPredicate>
@@ -452,9 +428,9 @@ void
 __brick_copy_by_mask(_ForwardIterator, _ForwardIterator, _OutputIterator, bool*,
                      /*vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _OutputIterator>
+template <class _RandomAccessIterator, class _OutputIterator>
 void
-__brick_copy_by_mask(_ForwardIterator, _ForwardIterator, _OutputIterator, bool* __restrict,
+__brick_copy_by_mask(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator, bool* __restrict,
                      /*vector=*/std::true_type) noexcept;
 
 template <class _ForwardIterator, class _OutputIterator1, class _OutputIterator2>
@@ -467,24 +443,24 @@ void
 __brick_partition_by_mask(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator1, _OutputIterator2, bool*,
                           /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryPredicate>
 _OutputIterator
-__pattern_copy_if(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator, _UnaryPredicate, _IsVector,
-                  /*parallel=*/std::false_type) noexcept;
+__pattern_copy_if(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator,
+                  _UnaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator,
+          class _UnaryPredicate>
 _OutputIterator
-__pattern_copy_if(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _OutputIterator, _UnaryPredicate,
-                  _IsVector, /*parallel=*/std::true_type);
+__pattern_copy_if(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                  _OutputIterator, _UnaryPredicate);
 
 //------------------------------------------------------------------------
 // count
 //------------------------------------------------------------------------
 
-template <class _ForwardIterator, class _Predicate>
-typename std::iterator_traits<_ForwardIterator>::difference_type
-    __brick_count(_ForwardIterator, _ForwardIterator, _Predicate,
+template <class _RandomAccessIterator, class _Predicate>
+typename std::iterator_traits<_RandomAccessIterator>::difference_type
+    __brick_count(_RandomAccessIterator, _RandomAccessIterator, _Predicate,
                   /* is_vector = */ std::true_type) noexcept;
 
 template <class _ForwardIterator, class _Predicate>
@@ -492,15 +468,14 @@ typename std::iterator_traits<_ForwardIterator>::difference_type
     __brick_count(_ForwardIterator, _ForwardIterator, _Predicate,
                   /* is_vector = */ std::false_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
 typename std::iterator_traits<_ForwardIterator>::difference_type
-__pattern_count(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Predicate,
-                /* is_parallel */ std::false_type, _IsVector) noexcept;
+__pattern_count(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Predicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
-typename std::iterator_traits<_ForwardIterator>::difference_type
-__pattern_count(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Predicate,
-                /* is_parallel */ std::true_type, _IsVector);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Predicate>
+typename std::iterator_traits<_RandomAccessIterator>::difference_type
+__pattern_count(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                _Predicate);
 
 //------------------------------------------------------------------------
 // unique
@@ -510,37 +485,35 @@ template <class _ForwardIterator, class _BinaryPredicate>
 _ForwardIterator __brick_unique(_ForwardIterator, _ForwardIterator, _BinaryPredicate,
                                 /*is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _BinaryPredicate>
-_ForwardIterator __brick_unique(_ForwardIterator, _ForwardIterator, _BinaryPredicate,
-                                /*is_vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator, class _BinaryPredicate>
+_RandomAccessIterator __brick_unique(_RandomAccessIterator, _RandomAccessIterator, _BinaryPredicate,
+                                     /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate>
 _ForwardIterator
-__pattern_unique(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _BinaryPredicate, _IsVector,
-                 /*is_parallel=*/std::false_type) noexcept;
+__pattern_unique(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _BinaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
-_ForwardIterator
-__pattern_unique(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _BinaryPredicate, _IsVector,
-                 /*is_parallel=*/std::true_type) noexcept;
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _BinaryPredicate>
+_RandomAccessIterator
+__pattern_unique(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                 _BinaryPredicate) noexcept;
 
 //------------------------------------------------------------------------
 // unique_copy
 //------------------------------------------------------------------------
 
 template <class _ForwardIterator, class _OutputIterator, class _BinaryPredicate>
-  _OutputIterator __brick_unique_copy(_ForwardIterator, _ForwardIterator, _OutputIterator, _BinaryPredicate,
+_OutputIterator __brick_unique_copy(_ForwardIterator, _ForwardIterator, _OutputIterator, _BinaryPredicate,
                                    /*vector=*/std::false_type) noexcept;
 
 template <class _RandomAccessIterator, class _OutputIterator, class _BinaryPredicate>
 _OutputIterator __brick_unique_copy(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator, _BinaryPredicate,
                                     /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryPredicate>
 _OutputIterator
-__pattern_unique_copy(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator, _BinaryPredicate,
-                      _IsVector, /*parallel=*/std::false_type) noexcept;
+__pattern_unique_copy(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator,
+                      _BinaryPredicate) noexcept;
 
 template <class _ExecutionPolicy, class _DifferenceType, class _RandomAccessIterator, class _BinaryPredicate>
 _DifferenceType
@@ -552,11 +525,11 @@ _DifferenceType
 __brick_calc_mask_2(_RandomAccessIterator, _RandomAccessIterator, bool* __restrict, _BinaryPredicate,
                     /*vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _BinaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator,
+          class _BinaryPredicate>
 _OutputIterator
-__pattern_unique_copy(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _OutputIterator,
-                      _BinaryPredicate, _IsVector, /*parallel=*/std::true_type);
+__pattern_unique_copy(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                      _OutputIterator, _BinaryPredicate);
 
 //------------------------------------------------------------------------
 // reverse
@@ -566,27 +539,25 @@ template <class _BidirectionalIterator>
 void __brick_reverse(_BidirectionalIterator, _BidirectionalIterator,
                      /*__is_vector=*/std::false_type) noexcept;
 
-template <class _BidirectionalIterator>
-void __brick_reverse(_BidirectionalIterator, _BidirectionalIterator,
+template <class _RandomAccessIterator>
+void __brick_reverse(_RandomAccessIterator, _RandomAccessIterator,
                      /*__is_vector=*/std::true_type) noexcept;
 
 template <class _BidirectionalIterator>
 void __brick_reverse(_BidirectionalIterator, _BidirectionalIterator, _BidirectionalIterator,
                      /*is_vector=*/std::false_type) noexcept;
 
-template <class _BidirectionalIterator>
-void __brick_reverse(_BidirectionalIterator, _BidirectionalIterator, _BidirectionalIterator,
+template <class _RandomAccessIterator>
+void __brick_reverse(_RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator,
                      /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator>
 void
-__pattern_reverse(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _IsVector,
-                  /*is_parallel=*/std::false_type) noexcept;
+__pattern_reverse(_Tag, _ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator>
 void
-__pattern_reverse(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _IsVector,
-                  /*is_parallel=*/std::true_type);
+__pattern_reverse(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator);
 
 //------------------------------------------------------------------------
 // reverse_copy
@@ -596,19 +567,19 @@ template <class _BidirectionalIterator, class _OutputIterator>
 _OutputIterator __brick_reverse_copy(_BidirectionalIterator, _BidirectionalIterator, _OutputIterator,
                                      /*is_vector=*/std::false_type) noexcept;
 
-template <class _BidirectionalIterator, class _OutputIterator>
-_OutputIterator __brick_reverse_copy(_BidirectionalIterator, _BidirectionalIterator, _OutputIterator,
+template <class _RandomAccessIterator, class _OutputIterator>
+_OutputIterator __brick_reverse_copy(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator,
                                      /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _OutputIterator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator, class _OutputIterator>
 _OutputIterator
-__pattern_reverse_copy(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _OutputIterator, _IsVector,
-                       /*is_parallel=*/std::false_type) noexcept;
+__pattern_reverse_copy(_Tag, _ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator,
+                       _OutputIterator) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _OutputIterator, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator>
 _OutputIterator
-__pattern_reverse_copy(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _OutputIterator, _IsVector,
-                       /*is_parallel=*/std::true_type);
+__pattern_reverse_copy(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                       _OutputIterator);
 
 //------------------------------------------------------------------------
 // rotate
@@ -618,19 +589,18 @@ template <class _ForwardIterator>
 _ForwardIterator __brick_rotate(_ForwardIterator, _ForwardIterator, _ForwardIterator,
                                 /*is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator>
-_ForwardIterator __brick_rotate(_ForwardIterator, _ForwardIterator, _ForwardIterator,
-                                /*is_vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator>
+_RandomAccessIterator __brick_rotate(_RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator,
+                                     /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator>
 _ForwardIterator
-__pattern_rotate(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _ForwardIterator, _IsVector,
-                 /*is_parallel=*/std::false_type) noexcept;
+__pattern_rotate(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _ForwardIterator) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _IsVector>
-_ForwardIterator
-__pattern_rotate(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _ForwardIterator, _IsVector,
-                 /*is_parallel=*/std::true_type);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator>
+_RandomAccessIterator
+__pattern_rotate(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                 _RandomAccessIterator);
 
 //------------------------------------------------------------------------
 // rotate_copy
@@ -640,21 +610,20 @@ template <class _ForwardIterator, class _OutputIterator>
 _OutputIterator __brick_rotate_copy(_ForwardIterator, _ForwardIterator, _ForwardIterator, _OutputIterator,
                                     /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _OutputIterator>
-_OutputIterator __brick_rotate_copy(_ForwardIterator, _ForwardIterator, _ForwardIterator, _OutputIterator,
+template <class _RandomAccessIterator, class _OutputIterator>
+_OutputIterator __brick_rotate_copy(_RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator,
+                                    _OutputIterator,
                                     /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator>
 _OutputIterator
-__pattern_rotate_copy(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _ForwardIterator, _OutputIterator,
-                      _IsVector,
-                      /*is_parallel=*/std::false_type) noexcept;
+__pattern_rotate_copy(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _ForwardIterator,
+                      _OutputIterator) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator>
 _OutputIterator
-__pattern_rotate_copy(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _ForwardIterator, _OutputIterator,
-                      _IsVector,
-                      /*is_parallel=*/std::true_type);
+__pattern_rotate_copy(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                      _RandomAccessIterator, _OutputIterator);
 
 //------------------------------------------------------------------------
 // is_partitioned
@@ -664,19 +633,18 @@ template <class _ForwardIterator, class _UnaryPredicate>
 bool __brick_is_partitioned(_ForwardIterator, _ForwardIterator, _UnaryPredicate,
                             /*is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _UnaryPredicate>
-bool __brick_is_partitioned(_ForwardIterator, _ForwardIterator, _UnaryPredicate,
+template <class _RandomAccessIterator, class _UnaryPredicate>
+bool __brick_is_partitioned(_RandomAccessIterator, _RandomAccessIterator, _UnaryPredicate,
                             /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 bool
-__pattern_is_partitioned(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate, _IsVector,
-                         /*is_parallel=*/std::false_type) noexcept;
+__pattern_is_partitioned(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
 bool
-__pattern_is_partitioned(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate, _IsVector,
-                         /*is_parallel=*/std::true_type);
+__pattern_is_partitioned(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                         _UnaryPredicate);
 
 //------------------------------------------------------------------------
 // partition
@@ -686,19 +654,18 @@ template <class _ForwardIterator, class _UnaryPredicate>
 _ForwardIterator __brick_partition(_ForwardIterator, _ForwardIterator, _UnaryPredicate,
                                    /*is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _UnaryPredicate>
-_ForwardIterator __brick_partition(_ForwardIterator, _ForwardIterator, _UnaryPredicate,
-                                   /*is_vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator __brick_partition(_RandomAccessIterator, _RandomAccessIterator, _UnaryPredicate,
+                                        /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 _ForwardIterator
-__pattern_partition(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate, _IsVector,
-                    /*is_parallel=*/std::false_type) noexcept;
+__pattern_partition(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
-_ForwardIterator
-__pattern_partition(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate, _IsVector,
-                    /*is_parallel=*/std::true_type);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__pattern_partition(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                    _UnaryPredicate);
 
 //------------------------------------------------------------------------
 // stable_partition
@@ -708,21 +675,19 @@ template <class _BidirectionalIterator, class _UnaryPredicate>
 _BidirectionalIterator __brick_stable_partition(_BidirectionalIterator, _BidirectionalIterator, _UnaryPredicate,
                                                 /*__is_vector=*/std::false_type) noexcept;
 
-template <class _BidirectionalIterator, class _UnaryPredicate>
-_BidirectionalIterator __brick_stable_partition(_BidirectionalIterator, _BidirectionalIterator, _UnaryPredicate,
-                                                /*__is_vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator __brick_stable_partition(_RandomAccessIterator, _RandomAccessIterator, _UnaryPredicate,
+                                               /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate>
 _BidirectionalIterator
-__pattern_stable_partition(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _UnaryPredicate,
-                           _IsVector,
-                           /*is_parallelization=*/std::false_type) noexcept;
+__pattern_stable_partition(_Tag, _ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator,
+                           _UnaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate, class _IsVector>
-_BidirectionalIterator
-__pattern_stable_partition(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _UnaryPredicate,
-                           _IsVector,
-                           /*is_parallelization=*/std::true_type) noexcept;
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__pattern_stable_partition(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                           _UnaryPredicate) noexcept;
 
 //------------------------------------------------------------------------
 // partition_copy
@@ -733,134 +698,123 @@ std::pair<_OutputIterator1, _OutputIterator2>
     __brick_partition_copy(_ForwardIterator, _ForwardIterator, _OutputIterator1, _OutputIterator2, _UnaryPredicate,
                            /*is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator, class _OutputIterator1, class _OutputIterator2, class _UnaryPredicate>
-std::pair<_OutputIterator1, _OutputIterator2>
-    __brick_partition_copy(_ForwardIterator, _ForwardIterator, _OutputIterator1, _OutputIterator2, _UnaryPredicate,
-                           /*is_vector=*/std::true_type) noexcept;
+template <class _RandomAccessIterator, class _OutputIterator1, class _OutputIterator2, class _UnaryPredicate>
+std::pair<_OutputIterator1, _OutputIterator2> __brick_partition_copy(_RandomAccessIterator, _RandomAccessIterator,
+                                                                     _OutputIterator1, _OutputIterator2,
+                                                                     _UnaryPredicate,
+                                                                     /*is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator1, class _OutputIterator2,
-          class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator1, class _OutputIterator2,
+          class _UnaryPredicate>
 std::pair<_OutputIterator1, _OutputIterator2>
-__pattern_partition_copy(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator1, _OutputIterator2,
-                         _UnaryPredicate, _IsVector,
-                         /*is_parallelization=*/std::false_type) noexcept;
+__pattern_partition_copy(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator1,
+                         _OutputIterator2, _UnaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator1, class _OutputIterator2,
-          class _UnaryPredicate, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator1,
+          class _OutputIterator2, class _UnaryPredicate>
 std::pair<_OutputIterator1, _OutputIterator2>
-__pattern_partition_copy(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _OutputIterator1,
-                         _OutputIterator2, _UnaryPredicate, _IsVector,
-                         /*is_parallelization=*/std::true_type);
+__pattern_partition_copy(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                         _OutputIterator1, _OutputIterator2, _UnaryPredicate);
 
 //------------------------------------------------------------------------
 // sort
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector,
-          class _IsMoveConstructible>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsMoveConstructible>
 void
-__pattern_sort(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare, _IsVector /*is_vector*/,
-               /*is_parallel=*/std::false_type, _IsMoveConstructible) noexcept;
+__pattern_sort(_Tag, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare,
+               _IsMoveConstructible) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_sort(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare, _IsVector /*is_vector*/,
-               /*is_parallel=*/std::true_type,
+__pattern_sort(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare,
                /*is_move_constructible=*/std::true_type);
 
 //------------------------------------------------------------------------
 // stable_sort
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_stable_sort(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare,
-                      _IsVector /*is_vector*/,
-                      /*is_parallel=*/std::false_type) noexcept;
+__pattern_stable_sort(_Tag, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_stable_sort(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare,
-                      _IsVector /*is_vector*/,
-                      /*is_parallel=*/std::true_type);
+__pattern_stable_sort(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                      _Compare);
 
 //------------------------------------------------------------------------
 // partial_sort
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_partial_sort(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator,
-                       _Compare, _IsVector,
-                       /*is_parallel=*/std::false_type) noexcept;
+__pattern_partial_sort(_Tag, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator,
+                       _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_partial_sort(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator,
-                       _Compare, _IsVector,
-                       /*is_parallel=*/std::true_type);
+__pattern_partial_sort(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                       _RandomAccessIterator, _Compare);
 
 //------------------------------------------------------------------------
 // partial_sort_copy
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _RandomAccessIterator, class _Compare, class _IsVector>
-_RandomAccessIterator
-__pattern_partial_sort_copy(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _RandomAccessIterator,
-                            _RandomAccessIterator, _Compare, _IsVector,
-                            /*is_parallel=*/std::false_type) noexcept;
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Compare>
+_RandomAccessIterator2
+__pattern_partial_sort_copy(_Tag, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                            _RandomAccessIterator2, _RandomAccessIterator2, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _RandomAccessIterator, class _Compare, class _IsVector>
-_RandomAccessIterator
-__pattern_partial_sort_copy(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _RandomAccessIterator,
-                            _RandomAccessIterator, _Compare, _IsVector,
-                            /*is_parallel=*/std::true_type);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Compare>
+_RandomAccessIterator2
+__pattern_partial_sort_copy(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1,
+                            _RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator2, _Compare);
 
 //------------------------------------------------------------------------
 // adjacent_find
 //------------------------------------------------------------------------
 
-template <class _ForwardIterator, class _BinaryPredicate>
-_ForwardIterator
-__brick_adjacent_find(_ForwardIterator, _ForwardIterator, _BinaryPredicate,
-                      /* _IsVector = */ std::true_type, bool) noexcept;
+template <class _RandomAccessIterator, class _BinaryPredicate>
+_RandomAccessIterator
+__brick_adjacent_find(_RandomAccessIterator, _RandomAccessIterator, _BinaryPredicate,
+                      /* IsVector = */ std::true_type, bool) noexcept;
 
 template <class _ForwardIterator, class _BinaryPredicate>
 _ForwardIterator
 __brick_adjacent_find(_ForwardIterator, _ForwardIterator, _BinaryPredicate,
-                      /* _IsVector = */ std::false_type, bool) noexcept;
+                      /* IsVector = */ std::false_type, bool) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate>
 _ForwardIterator
-__pattern_adjacent_find(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _BinaryPredicate,
-                        /* is_parallel */ std::false_type, _IsVector, bool) noexcept;
+__pattern_adjacent_find(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _BinaryPredicate, bool) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _BinaryPredicate, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _BinaryPredicate>
 _RandomAccessIterator
-__pattern_adjacent_find(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _BinaryPredicate,
-                        /* is_parallel */ std::true_type, _IsVector, bool);
+__pattern_adjacent_find(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                        _BinaryPredicate, bool);
 
 //------------------------------------------------------------------------
 // nth_element
 //------------------------------------------------------------------------
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_nth_element(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator, _Compare,
-                      _IsVector,
-                      /*is_parallel=*/std::false_type) noexcept;
+__pattern_nth_element(_Tag, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator,
+                      _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_nth_element(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator, _Compare,
-                      _IsVector,
-                      /*is_parallel=*/std::true_type) noexcept;
+__pattern_nth_element(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                      _RandomAccessIterator, _Compare) noexcept;
 
 //------------------------------------------------------------------------
 // fill, fill_n
 //------------------------------------------------------------------------
-template <class _ForwardIterator, class _Tp>
+template <class _RandomAccessIterator, class _Tp>
 void
-__brick_fill(_ForwardIterator, _ForwardIterator, const _Tp&,
+__brick_fill(_RandomAccessIterator, _RandomAccessIterator, const _Tp&,
              /* __is_vector = */ std::true_type) noexcept;
 
 template <class _ForwardIterator, class _Tp>
@@ -868,19 +822,17 @@ void
 __brick_fill(_ForwardIterator, _ForwardIterator, const _Tp&,
              /* __is_vector = */ std::false_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Tp>
 void
-__pattern_fill(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, const _Tp&,
-               /*is_parallel=*/std::false_type, _IsVector) noexcept;
+__pattern_fill(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, const _Tp&) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _IsVector>
-_ForwardIterator
-__pattern_fill(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, const _Tp&,
-               /*is_parallel=*/std::true_type, _IsVector);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Tp>
+_RandomAccessIterator
+__pattern_fill(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, const _Tp&);
 
-template <class _OutputIterator, class _Size, class _Tp>
-_OutputIterator
-__brick_fill_n(_OutputIterator, _Size, const _Tp&,
+template <class _RandomAccessIterator, class _Size, class _Tp>
+_RandomAccessIterator
+__brick_fill_n(_RandomAccessIterator, _Size, const _Tp&,
                /* __is_vector = */ std::true_type) noexcept;
 
 template <class _OutputIterator, class _Size, class _Tp>
@@ -888,15 +840,13 @@ _OutputIterator
 __brick_fill_n(_OutputIterator, _Size, const _Tp&,
                /* __is_vector = */ std::false_type) noexcept;
 
-template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Tp, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _OutputIterator, class _Size, class _Tp>
 _OutputIterator
-__pattern_fill_n(_ExecutionPolicy&&, _OutputIterator, _Size, const _Tp&,
-                 /*is_parallel=*/std::false_type, _IsVector) noexcept;
+__pattern_fill_n(_Tag, _ExecutionPolicy&&, _OutputIterator, _Size, const _Tp&) noexcept;
 
-template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Tp, class _IsVector>
-_OutputIterator
-__pattern_fill_n(_ExecutionPolicy&&, _OutputIterator, _Size, const _Tp&,
-                 /*is_parallel=*/std::true_type, _IsVector);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Tp>
+_RandomAccessIterator
+__pattern_fill_n(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _Size, const _Tp&);
 
 //------------------------------------------------------------------------
 // generate, generate_n
@@ -910,33 +860,30 @@ template <class _ForwardIterator, class _Generator>
 void __brick_generate(_ForwardIterator, _ForwardIterator, _Generator,
                       /* is_vector = */ std::false_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Generator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Generator>
 void
-__pattern_generate(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Generator,
-                   /*is_parallel=*/std::false_type, _IsVector) noexcept;
+__pattern_generate(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Generator) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Generator, class _IsVector>
-_ForwardIterator
-__pattern_generate(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Generator,
-                   /*is_parallel=*/std::true_type, _IsVector);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Generator>
+_RandomAccessIterator
+__pattern_generate(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                   _Generator);
 
-template <class _OutputIterator, class _Size, class _Generator>
-  _OutputIterator __brick_generate_n(_OutputIterator, _Size, _Generator,
-                                  /* is_vector = */ std::true_type) noexcept;
+template <class _RandomAccessIterator, class _Size, class _Generator>
+_RandomAccessIterator __brick_generate_n(_RandomAccessIterator, _Size, _Generator,
+                                         /* is_vector = */ std::true_type) noexcept;
 
 template <class _OutputIterator, class _Size, class _Generator>
-  _OutputIterator __brick_generate_n(_OutputIterator, _Size, _Generator,
+_OutputIterator __brick_generate_n(_OutputIterator, _Size, _Generator,
                                   /* is_vector = */ std::false_type) noexcept;
 
-template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Generator, class _IsVector>
-  _OutputIterator
-__pattern_generate_n(_ExecutionPolicy&&, _OutputIterator, _Size, _Generator,
-                     /*is_parallel=*/std::false_type, _IsVector) noexcept;
+template <class _Tag, class _ExecutionPolicy, class _OutputIterator, class _Size, class _Generator>
+_OutputIterator
+__pattern_generate_n(_Tag, _ExecutionPolicy&&, _OutputIterator, _Size, _Generator) noexcept;
 
-template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Generator, class _IsVector>
-  _OutputIterator
-__pattern_generate_n(_ExecutionPolicy&&, _OutputIterator, _Size, _Generator,
-                     /*is_parallel=*/std::true_type, _IsVector);
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Generator>
+_RandomAccessIterator
+__pattern_generate_n(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _Size, _Generator);
 
 //------------------------------------------------------------------------
 // remove
@@ -949,15 +896,14 @@ template <class _RandomAccessIterator, class _UnaryPredicate>
 _RandomAccessIterator __brick_remove_if(_RandomAccessIterator, _RandomAccessIterator, _UnaryPredicate,
                                         /* __is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 _ForwardIterator
-__pattern_remove_if(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate, _IsVector,
-                    /*is_parallel*/ std::false_type) noexcept;
+__pattern_remove_if(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
-_ForwardIterator
-__pattern_remove_if(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _UnaryPredicate, _IsVector,
-                    /*is_parallel*/ std::true_type) noexcept;
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__pattern_remove_if(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                    _UnaryPredicate) noexcept;
 
 //------------------------------------------------------------------------
 // merge
@@ -968,23 +914,22 @@ _OutputIterator __brick_merge(_ForwardIterator1, _ForwardIterator1, _ForwardIter
                               _OutputIterator, _Compare,
                               /* __is_vector = */ std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator __brick_merge(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                              _OutputIterator, _Compare,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator, class _Compare>
+_OutputIterator __brick_merge(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                              _RandomAccessIterator2, _OutputIterator, _Compare,
                               /* __is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_merge(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                _OutputIterator, _Compare, _IsVector, /* is_parallel = */ std::false_type) noexcept;
+__pattern_merge(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
+                _OutputIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _OutputIterator, class _Compare>
 _OutputIterator
-__pattern_merge(_ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
-                _RandomAccessIterator2, _OutputIterator, _Compare, _IsVector,
-                /* is_parallel = */ std::true_type);
+__pattern_merge(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                _RandomAccessIterator2, _RandomAccessIterator2, _OutputIterator, _Compare);
 
 //------------------------------------------------------------------------
 // inplace_merge
@@ -994,37 +939,34 @@ template <class _BidirectionalIterator, class _Compare>
 void __brick_inplace_merge(_BidirectionalIterator, _BidirectionalIterator, _BidirectionalIterator, _Compare,
                            /* __is_vector = */ std::false_type) noexcept;
 
-template <class _BidirectionalIterator, class _Compare>
-void __brick_inplace_merge(_BidirectionalIterator, _BidirectionalIterator, _BidirectionalIterator, _Compare,
+template <class _RandomAccessIterator, class _Compare>
+void __brick_inplace_merge(_RandomAccessIterator, _RandomAccessIterator, _RandomAccessIterator, _Compare,
                            /* __is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator, class _Compare>
 void
-__pattern_inplace_merge(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _BidirectionalIterator,
-                        _Compare, _IsVector,
-                        /* is_parallel = */ std::false_type) noexcept;
+__pattern_inplace_merge(_Tag, _ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator,
+                        _BidirectionalIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_inplace_merge(_ExecutionPolicy&&, _BidirectionalIterator, _BidirectionalIterator, _BidirectionalIterator,
-                        _Compare, _IsVector,
-                        /*is_parallel=*/std::true_type);
+__pattern_inplace_merge(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                        _RandomAccessIterator, _Compare);
 
 //------------------------------------------------------------------------
 // includes
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare>
 bool
-__pattern_includes(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                   _Compare, _IsVector,
-                   /*is_parallel=*/std::false_type) noexcept;
+__pattern_includes(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
+                   _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Compare>
 bool
-__pattern_includes(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                   _Compare, _IsVector,
-                   /*is_parallel=*/std::true_type);
+__pattern_includes(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                   _RandomAccessIterator2, _RandomAccessIterator2, _Compare);
 
 //------------------------------------------------------------------------
 // set_union
@@ -1035,22 +977,22 @@ _OutputIterator __brick_set_union(_ForwardIterator1, _ForwardIterator1, _Forward
                                   _OutputIterator, _Compare,
                                   /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator __brick_set_union(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                                  _OutputIterator, _Compare,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator, class _Compare>
+_OutputIterator __brick_set_union(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                  _RandomAccessIterator2, _OutputIterator, _Compare,
                                   /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_set_union(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                    _OutputIterator, _Compare, _IsVector, /*is_parallel=*/std::false_type) noexcept;
+__pattern_set_union(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                    _ForwardIterator2, _OutputIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _OutputIterator, class _Compare>
 _OutputIterator
-__pattern_set_union(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                    _OutputIterator, _Compare, _IsVector, /*is_parallel=*/std::true_type);
+__pattern_set_union(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                    _RandomAccessIterator2, _RandomAccessIterator2, _OutputIterator, _Compare);
 
 //------------------------------------------------------------------------
 // set_intersection
@@ -1061,23 +1003,23 @@ _OutputIterator __brick_set_intersection(_ForwardIterator1, _ForwardIterator1, _
                                          _OutputIterator, _Compare,
                                          /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator __brick_set_intersection(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                                         _OutputIterator, _Compare,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator, class _Compare>
+_OutputIterator __brick_set_intersection(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                         _RandomAccessIterator2, _OutputIterator, _Compare,
                                          /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_set_intersection(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
-                           _ForwardIterator2, _OutputIterator, _Compare, _IsVector,
-                           /*is_parallel=*/std::false_type) noexcept;
+__pattern_set_intersection(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                           _ForwardIterator2, _OutputIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _OutputIterator, class _Compare>
 _OutputIterator
-__pattern_set_intersection(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
-                           _ForwardIterator2, _OutputIterator, _Compare, _IsVector, /*is_parallel=*/std::true_type);
+__pattern_set_intersection(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1,
+                           _RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator2, _OutputIterator,
+                           _Compare);
 
 //------------------------------------------------------------------------
 // set_difference
@@ -1088,22 +1030,22 @@ _OutputIterator __brick_set_difference(_ForwardIterator1, _ForwardIterator1, _Fo
                                        _OutputIterator, _Compare,
                                        /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator __brick_set_difference(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                                       _OutputIterator, _Compare,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator, class _Compare>
+_OutputIterator __brick_set_difference(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                       _RandomAccessIterator2, _OutputIterator, _Compare,
                                        /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_set_difference(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                         _OutputIterator, _Compare, _IsVector, /*is_parallel=*/std::false_type) noexcept;
+__pattern_set_difference(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                         _ForwardIterator2, _OutputIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _OutputIterator, class _Compare>
 _OutputIterator
-__pattern_set_difference(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                         _OutputIterator, _Compare, _IsVector, /*is_parallel=*/std::true_type);
+__pattern_set_difference(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                         _RandomAccessIterator2, _RandomAccessIterator2, _OutputIterator, _Compare);
 
 //------------------------------------------------------------------------
 // set_symmetric_difference
@@ -1114,24 +1056,23 @@ _OutputIterator __brick_set_symmetric_difference(_ForwardIterator1, _ForwardIter
                                                  _ForwardIterator2, _OutputIterator, _Compare,
                                                  /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator __brick_set_symmetric_difference(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
-                                                 _ForwardIterator2, _OutputIterator, _Compare,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator, class _Compare>
+_OutputIterator __brick_set_symmetric_difference(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                                 _RandomAccessIterator2, _OutputIterator, _Compare,
                                                  /*__is_vector=*/std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_set_symmetric_difference(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
-                                   _ForwardIterator2, _OutputIterator, _Compare, _IsVector,
-                                   /*is_parallel=*/std::false_type) noexcept;
+__pattern_set_symmetric_difference(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                                   _ForwardIterator2, _OutputIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _OutputIterator, class _Compare>
 _OutputIterator
-__pattern_set_symmetric_difference(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
-                                   _ForwardIterator2, _OutputIterator, _Compare, _IsVector,
-                                   /*is_parallel=*/std::true_type);
+__pattern_set_symmetric_difference(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1,
+                                   _RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator2,
+                                   _OutputIterator, _Compare);
 
 //------------------------------------------------------------------------
 // is_heap_until
@@ -1145,15 +1086,14 @@ template <class _RandomAccessIterator, class _Compare>
 _RandomAccessIterator __brick_is_heap_until(_RandomAccessIterator, _RandomAccessIterator, _Compare,
                                             /* __is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 _RandomAccessIterator
-__pattern_is_heap_until(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare, _IsVector,
-                        /* is_parallel = */ std::false_type) noexcept;
+__pattern_is_heap_until(_Tag, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 _RandomAccessIterator
-__pattern_is_heap_until(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare, _IsVector,
-                        /* is_parallel = */ std::true_type) noexcept;
+__pattern_is_heap_until(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                        _Compare) noexcept;
 
 //------------------------------------------------------------------------
 // min_element
@@ -1163,19 +1103,18 @@ template <typename _ForwardIterator, typename _Compare>
 _ForwardIterator __brick_min_element(_ForwardIterator, _ForwardIterator, _Compare,
                                      /* __is_vector = */ std::false_type) noexcept;
 
-template <typename _ForwardIterator, typename _Compare>
-_ForwardIterator __brick_min_element(_ForwardIterator, _ForwardIterator, _Compare,
-                                     /* __is_vector = */ std::true_type) noexcept;
+template <typename _RandomAccessIterator, typename _Compare>
+_RandomAccessIterator __brick_min_element(_RandomAccessIterator, _RandomAccessIterator, _Compare,
+                                          /* __is_vector = */ std::true_type) noexcept;
 
-template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
+template <typename _Tag, typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare>
 _ForwardIterator
-__pattern_min_element(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Compare, _IsVector,
-                      /* is_parallel = */ std::false_type) noexcept;
+__pattern_min_element(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Compare) noexcept;
 
-template <typename _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare, typename _IsVector>
+template <typename _IsVector, typename _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare>
 _RandomAccessIterator
-__pattern_min_element(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _Compare, _IsVector,
-                      /* is_parallel = */ std::true_type);
+__pattern_min_element(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                      _Compare);
 
 //------------------------------------------------------------------------
 // minmax_element
@@ -1185,19 +1124,19 @@ template <typename _ForwardIterator, typename _Compare>
 std::pair<_ForwardIterator, _ForwardIterator> __brick_minmax_element(_ForwardIterator, _ForwardIterator, _Compare,
                                                                      /* __is_vector = */ std::false_type) noexcept;
 
-template <typename _ForwardIterator, typename _Compare>
-std::pair<_ForwardIterator, _ForwardIterator> __brick_minmax_element(_ForwardIterator, _ForwardIterator, _Compare,
-                                                                     /* __is_vector = */ std::true_type) noexcept;
+template <typename _RandomAccessIterator, typename _Compare>
+std::pair<_RandomAccessIterator, _RandomAccessIterator>
+    __brick_minmax_element(_RandomAccessIterator, _RandomAccessIterator, _Compare,
+                           /* __is_vector = */ std::true_type) noexcept;
 
-template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
+template <typename _Tag, typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare>
 std::pair<_ForwardIterator, _ForwardIterator>
-__pattern_minmax_element(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Compare, _IsVector,
-                         /* is_parallel = */ std::false_type) noexcept;
+__pattern_minmax_element(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Compare) noexcept;
 
-template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
-std::pair<_ForwardIterator, _ForwardIterator>
-__pattern_minmax_element(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Compare, _IsVector,
-                         /* is_parallel = */ std::true_type);
+template <typename _IsVector, typename _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare>
+std::pair<_RandomAccessIterator, _RandomAccessIterator>
+__pattern_minmax_element(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                         _Compare);
 
 //------------------------------------------------------------------------
 // mismatch
@@ -1208,22 +1147,22 @@ std::pair<_ForwardIterator1, _ForwardIterator2> __brick_mismatch(_ForwardIterato
                                                                  _ForwardIterator2, _ForwardIterator2, _Predicate,
                                                                  /* __is_vector = */ std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
-std::pair<_ForwardIterator1, _ForwardIterator2> __brick_mismatch(_ForwardIterator1, _ForwardIterator1,
-                                                                 _ForwardIterator2, _ForwardIterator2, _Predicate,
-                                                                 /* __is_vector = */ std::true_type) noexcept;
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Predicate>
+std::pair<_RandomAccessIterator1, _RandomAccessIterator2>
+    __brick_mismatch(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator2,
+                     _Predicate,
+                     /* __is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
 std::pair<_ForwardIterator1, _ForwardIterator2>
-__pattern_mismatch(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                   _Predicate, _IsVector,
-                   /* is_parallel = */ std::false_type) noexcept;
+__pattern_mismatch(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
+                   _Predicate) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Predicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Predicate>
 std::pair<_RandomAccessIterator1, _RandomAccessIterator2>
-__pattern_mismatch(_ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
-                   _RandomAccessIterator2, _Predicate, _IsVector, /* is_parallel = */ std::true_type) noexcept;
+__pattern_mismatch(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1,
+                   _RandomAccessIterator2, _RandomAccessIterator2, _Predicate) noexcept;
 
 //------------------------------------------------------------------------
 // lexicographical_compare
@@ -1234,20 +1173,22 @@ bool __brick_lexicographical_compare(_ForwardIterator1, _ForwardIterator1, _Forw
                                      _Compare,
                                      /* __is_vector = */ std::false_type) noexcept;
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Compare>
-bool __brick_lexicographical_compare(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _ForwardIterator2,
-                                     _Compare,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Compare>
+bool __brick_lexicographical_compare(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
+                                     _RandomAccessIterator2, _Compare,
                                      /* __is_vector = */ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare>
 bool
-__pattern_lexicographical_compare(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
-                                  _ForwardIterator2, _Compare, _IsVector, /* is_parallel = */ std::false_type) noexcept;
+__pattern_lexicographical_compare(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
+                                  _ForwardIterator2, _Compare) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Compare>
 bool
-__pattern_lexicographical_compare(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2,
-                                  _ForwardIterator2, _Compare, _IsVector, /* is_parallel = */ std::true_type) noexcept;
+__pattern_lexicographical_compare(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1,
+                                  _RandomAccessIterator1, _RandomAccessIterator2, _RandomAccessIterator2,
+                                  _Compare) noexcept;
 
 } // namespace __internal
 } // namespace __pstl
diff --git a/libstdc++-v3/include/pstl/algorithm_impl.h b/libstdc++-v3/include/pstl/algorithm_impl.h
index 32c19990ed0..aec8c7484e7 100644
--- a/libstdc++-v3/include/pstl/algorithm_impl.h
+++ b/libstdc++-v3/include/pstl/algorithm_impl.h
@@ -41,33 +41,35 @@ __brick_any_of(const _ForwardIterator __first, const _ForwardIterator __last, _P
     return std::any_of(__first, __last, __pred);
 };
 
-template <class _ForwardIterator, class _Pred>
+template <class _RandomAccessIterator, class _Pred>
 bool
-__brick_any_of(const _ForwardIterator __first, const _ForwardIterator __last, _Pred __pred,
+__brick_any_of(const _RandomAccessIterator __first, const _RandomAccessIterator __last, _Pred __pred,
                /*__is_vector=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_or(__first, __last - __first, __pred);
 };
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Pred, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Pred>
 bool
-__pattern_any_of(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Pred __pred,
-                 _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
+__pattern_any_of(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Pred __pred) noexcept
 {
-    return __internal::__brick_any_of(__first, __last, __pred, __is_vector);
+    return __internal::__brick_any_of(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Pred, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Pred>
 bool
-__pattern_any_of(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Pred __pred,
-                 _IsVector __is_vector, /*parallel=*/std::true_type)
+__pattern_any_of(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                 _RandomAccessIterator __last, _Pred __pred)
 {
-    return __internal::__except_handler([&]() {
-        return __internal::__parallel_or(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                         [__pred, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                                             return __internal::__brick_any_of(__i, __j, __pred, __is_vector);
-                                         });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __internal::__parallel_or(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                             [__pred](_RandomAccessIterator __i, _RandomAccessIterator __j)
+                                             { return __internal::__brick_any_of(__i, __j, __pred, _IsVector{}); });
+        });
 }
 
 // [alg.foreach]
@@ -102,46 +104,51 @@ __brick_walk1(_RandomAccessIterator __first, _RandomAccessIterator __last, _Func
     __unseq_backend::__simd_walk_1(__first, __last - __first, __f);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Function>
 void
-__pattern_walk1(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Function __f,
-                _IsVector __is_vector,
-                /*parallel=*/std::false_type) noexcept
+__pattern_walk1(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Function __f) noexcept
 {
-    __internal::__brick_walk1(__first, __last, __f, __is_vector);
+    __internal::__brick_walk1(__first, __last, __f, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Function>
 void
-__pattern_walk1(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Function __f,
-                _IsVector __is_vector,
-                /*parallel=*/std::true_type)
-{
-    __internal::__except_handler([&]() {
-        __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                      [__f, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                                          __internal::__brick_walk1(__i, __j, __f, __is_vector);
-                                      });
-    });
+__pattern_walk1(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                _RandomAccessIterator __last, _Function __f)
+{
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                          [__f](_RandomAccessIterator __i, _RandomAccessIterator __j)
+                                          { __internal::__brick_walk1(__i, __j, __f, _IsVector{}); });
+        });
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Brick>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Brick>
 void
-__pattern_walk_brick(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Brick __brick,
-                     /*parallel=*/std::false_type) noexcept
+__pattern_walk_brick(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                     _Brick __brick) noexcept
 {
     __brick(__first, __last);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Brick>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Brick>
 void
-__pattern_walk_brick(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Brick __brick,
-                     /*parallel=*/std::true_type)
+__pattern_walk_brick(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                     _RandomAccessIterator __last, _Brick __brick)
 {
-    __internal::__except_handler([&]() {
-        __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                      [__brick](_ForwardIterator __i, _ForwardIterator __j) { __brick(__i, __j); });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                          [__brick](_RandomAccessIterator __i, _RandomAccessIterator __j)
+                                          { __brick(__i, __j); });
+        });
 }
 
 //------------------------------------------------------------------------
@@ -163,44 +170,45 @@ __brick_walk1_n(_RandomAccessIterator __first, _DifferenceType __n, _Function __
     return __unseq_backend::__simd_walk_1(__first, __n, __f);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Function>
 _ForwardIterator
-__pattern_walk1_n(_ExecutionPolicy&&, _ForwardIterator __first, _Size __n, _Function __f, _IsVector __is_vector,
-                  /*is_parallel=*/std::false_type) noexcept
+__pattern_walk1_n(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _Size __n, _Function __f) noexcept
 {
-    return __internal::__brick_walk1_n(__first, __n, __f, __is_vector);
+    return __internal::__brick_walk1_n(__first, __n, __f, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Function>
 _RandomAccessIterator
-__pattern_walk1_n(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _Size __n, _Function __f,
-                  _IsVector __is_vector,
-                  /*is_parallel=*/std::true_type)
+__pattern_walk1_n(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first, _Size __n,
+                  _Function __f)
 {
-    __internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __first + __n, __f, __is_vector,
-                                std::true_type());
+    __internal::__pattern_walk1(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __first + __n, __f);
+
     return __first + __n;
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Brick>
+template <class _Tag, 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
+__pattern_walk_brick_n(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _Size __n, _Brick __brick) noexcept
 {
     return __brick(__first, __n);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Brick>
+template <class _IsVector, 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)
+__pattern_walk_brick_n(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                       _Size __n, _Brick __brick)
 {
-    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;
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __first + __n,
+                [__brick](_RandomAccessIterator __i, _RandomAccessIterator __j) { __brick(__i, __j - __i); });
+            return __first + __n;
+        });
 }
 
 //------------------------------------------------------------------------
@@ -218,9 +226,10 @@ __brick_walk2(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIte
     return __first2;
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Function>
-_ForwardIterator2
-__brick_walk2(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2, _Function __f,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Function>
+_RandomAccessIterator2
+__brick_walk2(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+              _Function __f,
               /*vector=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_walk_2(__first1, __last1 - __first1, __first2, __f);
@@ -236,102 +245,115 @@ __brick_walk2_n(_ForwardIterator1 __first1, _Size __n, _ForwardIterator2 __first
     return __first2;
 }
 
-template <class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Function>
-_ForwardIterator2
-__brick_walk2_n(_ForwardIterator1 __first1, _Size __n, _ForwardIterator2 __first2, _Function __f,
+template <class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2, class _Function>
+_RandomAccessIterator2
+__brick_walk2_n(_RandomAccessIterator1 __first1, _Size __n, _RandomAccessIterator2 __first2, _Function __f,
                 /*vector=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_walk_2(__first1, __n, __first2, __f);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Function>
 _ForwardIterator2
-__pattern_walk2(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                _Function __f, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
+__pattern_walk2(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                _ForwardIterator2 __first2, _Function __f) noexcept
 {
-    return __internal::__brick_walk2(__first1, __last1, __first2, __f, __is_vector);
+    return __internal::__brick_walk2(__first1, __last1, __first2, __f, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Function>
+_RandomAccessIterator2
+__pattern_walk2(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _Function __f)
 {
-    return __internal::__except_handler([&]() {
-        __par_backend::__parallel_for(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
-            [__f, __first1, __first2, __is_vector](_ForwardIterator1 __i, _ForwardIterator1 __j) {
-                __internal::__brick_walk2(__i, __j, __first2 + (__i - __first1), __f, __is_vector);
-            });
-        return __first2 + (__last1 - __first1);
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                [__f, __first1, __first2](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+                { __internal::__brick_walk2(__i, __j, __first2 + (__i - __first1), __f, _IsVector{}); });
+            return __first2 + (__last1 - __first1);
+        });
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Function,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2,
+          class _Function>
 _ForwardIterator2
-__pattern_walk2_n(_ExecutionPolicy&&, _ForwardIterator1 __first1, _Size __n, _ForwardIterator2 __first2, _Function __f,
-                  _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
+__pattern_walk2_n(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _Size __n, _ForwardIterator2 __first2,
+                  _Function __f) noexcept
 {
-    return __internal::__brick_walk2_n(__first1, __n, __first2, __f, __is_vector);
+    return __internal::__brick_walk2_n(__first1, __n, __first2, __f, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2,
-          class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _Size,
+          class _RandomAccessIterator2, class _Function>
 _RandomAccessIterator2
-__pattern_walk2_n(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _Size __n, _RandomAccessIterator2 __first2,
-                  _Function __f, _IsVector __is_vector, /*parallel=*/std::true_type)
+__pattern_walk2_n(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                  _Size __n, _RandomAccessIterator2 __first2, _Function __f)
 {
-    return __internal::__pattern_walk2(std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n, __first2, __f,
-                                       __is_vector, std::true_type());
+    return __internal::__pattern_walk2(__tag, std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
+                                       __first2, __f);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Brick>
+template <class _Tag, 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
+__pattern_walk2_brick(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                      _ForwardIterator2 __first2, _Brick __brick) noexcept
 {
     return __brick(__first1, __last1, __first2);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Brick>
+template <class _IsVector, 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)
+__pattern_walk2_brick(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                      _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _Brick __brick)
 {
-    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);
-    });
-}
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
 
-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 __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;
-    });
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                [__first1, __first2, __brick](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+                { __brick(__i, __j, __first2 + (__i - __first1)); });
+            return __first2 + (__last1 - __first1);
+        });
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2, class _Brick>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2,
+          class _Brick>
 _ForwardIterator2
-__pattern_walk2_brick_n(_ExecutionPolicy&&, _ForwardIterator1 __first1, _Size __n, _ForwardIterator2 __first2,
-                        _Brick __brick, /*parallel=*/std::false_type) noexcept
+__pattern_walk2_brick_n(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _Size __n, _ForwardIterator2 __first2,
+                        _Brick __brick) noexcept
 {
     return __brick(__first1, __n, __first2);
 }
 
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _Size,
+          class _RandomAccessIterator2, class _Brick>
+_RandomAccessIterator2
+__pattern_walk2_brick_n(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                        _Size __n, _RandomAccessIterator2 __first2, _Brick __brick)
+{
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
+                [__first1, __first2, __brick](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+                { __brick(__i, __j - __i, __first2 + (__i - __first1)); });
+            return __first2 + __n;
+        });
+}
+
 //------------------------------------------------------------------------
 // walk3 (pseudo)
 //
@@ -355,31 +377,35 @@ __brick_walk3(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _
     return __unseq_backend::__simd_walk_3(__first1, __last1 - __first1, __first2, __first3, __f);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator3,
-          class _Function, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator3,
+          class _Function>
 _ForwardIterator3
-__pattern_walk3(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                _ForwardIterator3 __first3, _Function __f, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
+__pattern_walk3(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                _ForwardIterator2 __first2, _ForwardIterator3 __first3, _Function __f) noexcept
 {
-    return __internal::__brick_walk3(__first1, __last1, __first2, __first3, __f, __is_vector);
+    return __internal::__brick_walk3(__first1, __last1, __first2, __first3, __f, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
-          class _RandomAccessIterator3, class _Function, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _RandomAccessIterator3, class _Function>
 _RandomAccessIterator3
-__pattern_walk3(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
-                _RandomAccessIterator2 __first2, _RandomAccessIterator3 __first3, _Function __f, _IsVector __is_vector,
-                /*parallel=*/std::true_type)
+__pattern_walk3(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _RandomAccessIterator3 __first3,
+                _Function __f)
 {
-    return __internal::__except_handler([&]() {
-        __par_backend::__parallel_for(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
-            [__f, __first1, __first2, __first3, __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
-                __internal::__brick_walk3(__i, __j, __first2 + (__i - __first1), __first3 + (__i - __first1), __f,
-                                          __is_vector);
-            });
-        return __first3 + (__last1 - __first1);
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                [__f, __first1, __first2, __first3](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
+                    __internal::__brick_walk3(__i, __j, __first2 + (__i - __first1), __first3 + (__i - __first1), __f,
+                                              _IsVector{});
+                });
+            return __first3 + (__last1 - __first1);
+        });
 }
 
 //------------------------------------------------------------------------
@@ -405,34 +431,37 @@ __brick_equal(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _
     return __unseq_backend::__simd_first(__first1, __last1 - __first1, __first2, std::not_fn(__p)).first == __last1;
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                _ForwardIterator2 __last2, _BinaryPredicate __p, _IsVector __is_vector, /* is_parallel = */
-                std::false_type) noexcept
+__pattern_equal(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                _ForwardIterator2 __first2, _ForwardIterator2 __last2, _BinaryPredicate __p) noexcept
 {
-    return __internal::__brick_equal(__first1, __last1, __first2, __last2, __p, __is_vector);
+    return __internal::__brick_equal(__first1, __last1, __first2, __last2, __p, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
-                _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2, _BinaryPredicate __p,
-                _IsVector __is_vector, /*is_parallel=*/std::true_type)
+__pattern_equal(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                _BinaryPredicate __p)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__last1 - __first1 != __last2 - __first2)
         return false;
 
-    return __internal::__except_handler([&]() {
-        return !__internal::__parallel_or(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
-            [__first1, __first2, __p, __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
-                return !__internal::__brick_equal(__i, __j, __first2 + (__i - __first1), __first2 + (__j - __first1),
-                                                  __p, __is_vector);
-            });
-    });
+    return __internal::__except_handler(
+        [&]()
+        {
+            return !__internal::__parallel_or(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                [__first1, __first2, __p](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+                {
+                    return !__internal::__brick_equal(__i, __j, __first2 + (__i - __first1),
+                                                      __first2 + (__j - __first1), __p, _IsVector{});
+                });
+        });
 }
 
 //------------------------------------------------------------------------
@@ -455,29 +484,30 @@ __brick_equal(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _
     return __unseq_backend::__simd_first(__first1, __last1 - __first1, __first2, std::not_fn(__p)).first == __last1;
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                _BinaryPredicate __p, _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
+__pattern_equal(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                _ForwardIterator2 __first2, _BinaryPredicate __p) noexcept
 {
-    return __internal::__brick_equal(__first1, __last1, __first2, __p, __is_vector);
+    return __internal::__brick_equal(__first1, __last1, __first2, __p, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
 bool
-__pattern_equal(_ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
-                _RandomAccessIterator2 __first2, _BinaryPredicate __p, _IsVector __is_vector,
-                /*is_parallel=*/std::true_type)
+__pattern_equal(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _BinaryPredicate __p)
 {
-    return __internal::__except_handler([&]() {
-        return !__internal::__parallel_or(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
-            [__first1, __first2, __p, __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
-                return !__internal::__brick_equal(__i, __j, __first2 + (__i - __first1), __p, __is_vector);
-            });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            return !__internal::__parallel_or(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                [__first1, __first2, __p](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+                { return !__internal::__brick_equal(__i, __j, __first2 + (__i - __first1), __p, _IsVector{}); });
+        });
 }
 
 //------------------------------------------------------------------------
@@ -502,30 +532,31 @@ __brick_find_if(_RandomAccessIterator __first, _RandomAccessIterator __last, _Pr
         [&__pred](_RandomAccessIterator __it, _SizeType __i) { return __pred(__it[__i]); });
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
 _ForwardIterator
-__pattern_find_if(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred,
-                  _IsVector __is_vector,
-                  /*is_parallel=*/std::false_type) noexcept
+__pattern_find_if(_Tag __tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                  _Predicate __pred) noexcept
 {
-    return __internal::__brick_find_if(__first, __last, __pred, __is_vector);
+    return __internal::__brick_find_if(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Predicate>
+_RandomAccessIterator
+__pattern_find_if(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                  _RandomAccessIterator __last, _Predicate __pred)
 {
-    return __internal::__except_handler([&]() {
-        return __internal::__parallel_find(
-            std::forward<_ExecutionPolicy>(__exec), __first, __last,
-            [__pred, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                return __internal::__brick_find_if(__i, __j, __pred, __is_vector);
-            },
-            std::less<typename std::iterator_traits<_ForwardIterator>::difference_type>(),
-            /*is_first=*/true);
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __internal::__parallel_find(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                [__pred](_RandomAccessIterator __i, _RandomAccessIterator __j)
+                { return __internal::__brick_find_if(__i, __j, __pred, _IsVector{}); },
+                std::less<typename std::iterator_traits<_RandomAccessIterator>::difference_type>(),
+                /*is_first=*/true);
+        });
 }
 
 //------------------------------------------------------------------------
@@ -632,48 +663,51 @@ __brick_find_end(_ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIt
     return std::find_end(__first, __last, __s_first, __s_last, __pred);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
-_ForwardIterator1
-__brick_find_end(_ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
-                 _ForwardIterator2 __s_last, _BinaryPredicate __pred, /*__is_vector=*/std::true_type) noexcept
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator1
+__brick_find_end(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __s_first,
+                 _RandomAccessIterator2 __s_last, _BinaryPredicate __pred, /*__is_vector=*/std::true_type) noexcept
 {
     return __find_subrange(__first, __last, __last, __s_first, __s_last, __pred, false, std::true_type());
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _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
+__pattern_find_end(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first, _ForwardIterator1 __last,
+                   _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred) noexcept
 {
-    return __internal::__brick_find_end(__first, __last, __s_first, __s_last, __pred, __is_vector);
+    return __internal::__brick_find_end(__first, __last, __s_first, __s_last, __pred, typename _Tag::__is_vector{});
 }
 
-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
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
+_RandomAccessIterator1
+__pattern_find_end(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                   _RandomAccessIterator1 __last, _RandomAccessIterator2 __s_first, _RandomAccessIterator2 __s_last,
+                   _BinaryPredicate __pred) noexcept
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__last - __first == __s_last - __s_first)
     {
-        const bool __res = __internal::__pattern_equal(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                                       __s_first, __pred, __is_vector, std::true_type());
+        const bool __res = __internal::__pattern_equal(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                       __s_first, __pred);
         return __res ? __first : __last;
     }
     else
     {
-        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 __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);
-        });
+        return __internal::__except_handler(
+            [&]()
+            {
+                return __internal::__parallel_find(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                    [__last, __s_first, __s_last, __pred](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
+                        return __internal::__find_subrange(__i, __j, __last, __s_first, __s_last, __pred, false,
+                                                           _IsVector{});
+                    },
+                    std::greater<typename std::iterator_traits<_RandomAccessIterator1>::difference_type>(),
+                    /*is_first=*/false);
+            });
     }
 }
 
@@ -688,95 +722,99 @@ __brick_find_first_of(_ForwardIterator1 __first, _ForwardIterator1 __last, _Forw
     return std::find_first_of(__first, __last, __s_first, __s_last, __pred);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
-_ForwardIterator1
-__brick_find_first_of(_ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
-                      _ForwardIterator2 __s_last, _BinaryPredicate __pred, /*__is_vector=*/std::true_type) noexcept
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator1
+__brick_find_first_of(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __s_first,
+                      _RandomAccessIterator2 __s_last, _BinaryPredicate __pred, /*__is_vector=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_find_first_of(__first, __last, __s_first, __s_last, __pred);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _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
+__pattern_find_first_of(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first, _ForwardIterator1 __last,
+                        _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred) noexcept
 {
-    return __internal::__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,
+                                             typename _Tag::__is_vector{});
 }
 
-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
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
+_RandomAccessIterator1
+__pattern_find_first_of(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                        _RandomAccessIterator1 __last, _RandomAccessIterator2 __s_first,
+                        _RandomAccessIterator2 __s_last, _BinaryPredicate __pred) noexcept
 {
-    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 __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);
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __internal::__parallel_find(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                [__s_first, __s_last, __pred](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+                { return __internal::__brick_find_first_of(__i, __j, __s_first, __s_last, __pred, _IsVector{}); },
+                std::less<typename std::iterator_traits<_RandomAccessIterator1>::difference_type>(), /*is_first=*/true);
+        });
 }
 
 //------------------------------------------------------------------------
 // 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
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator1
+__brick_search(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __s_first,
+               _RandomAccessIterator2 __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
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator1
+__brick_search(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __s_first,
+               _RandomAccessIterator2 __s_last, _BinaryPredicate __pred, /*vector=*/std::true_type) noexcept
 {
     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>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
 _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
+__pattern_search(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first, _ForwardIterator1 __last,
+                 _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred) noexcept
 {
-    return __internal::__brick_search(__first, __last, __s_first, __s_last, __pred, __is_vector);
+    return __internal::__brick_search(__first, __last, __s_first, __s_last, __pred, typename _Tag::__is_vector{});
 }
 
-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
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
+_RandomAccessIterator1
+__pattern_search(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                 _RandomAccessIterator1 __last, _RandomAccessIterator2 __s_first, _RandomAccessIterator2 __s_last,
+                 _BinaryPredicate __pred) noexcept
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__last - __first == __s_last - __s_first)
     {
-        const bool __res = __internal::__pattern_equal(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                                       __s_first, __pred, __is_vector, std::true_type());
+        const bool __res = __internal::__pattern_equal(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                       __s_first, __pred);
         return __res ? __first : __last;
     }
     else
     {
-        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 __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);
-        });
+        return __internal::__except_handler(
+            [&]()
+            {
+                return __internal::__parallel_find(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                    [__last, __s_first, __s_last, __pred](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
+                        return __internal::__find_subrange(__i, __j, __last, __s_first, __s_last, __pred, true,
+                                                           _IsVector{});
+                    },
+                    std::less<typename std::iterator_traits<_RandomAccessIterator1>::difference_type>(),
+                    /*is_first=*/true);
+            });
     }
 }
 
@@ -791,49 +829,49 @@ __brick_search_n(_ForwardIterator __first, _ForwardIterator __last, _Size __coun
     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,
+template <class _RandomAccessIterator, class _Size, class _Tp, class _BinaryPredicate>
+_RandomAccessIterator
+__brick_search_n(_RandomAccessIterator __first, _RandomAccessIterator __last, _Size __count, const _Tp& __value,
                  _BinaryPredicate __pred, /*vector=*/std::true_type) noexcept
 {
     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>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Tp, class _BinaryPredicate>
 _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
+__pattern_search_n(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Size __count,
+                   const _Tp& __value, _BinaryPredicate __pred) noexcept
 {
-    return __internal::__brick_search_n(__first, __last, __count, __value, __pred, __is_vector);
+    return __internal::__brick_search_n(__first, __last, __count, __value, __pred, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Tp, class _BinaryPredicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Tp,
+          class _BinaryPredicate>
 _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
+__pattern_search_n(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                   _RandomAccessIterator __last, _Size __count, const _Tp& __value, _BinaryPredicate __pred) noexcept
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (static_cast<_Size>(__last - __first) == __count)
     {
-        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());
+        const bool __result =
+            !__internal::__pattern_any_of(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                          [&__value, &__pred](const _Tp& __val) { return !__pred(__val, __value); });
         return __result ? __first : __last;
     }
     else
     {
-        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 __internal::__find_subrange(__i, __j, __last, __count, __value, __pred, __is_vector);
-                },
-                std::less<typename std::iterator_traits<_RandomAccessIterator>::difference_type>(), /*is_first=*/true);
-        });
+        return __internal::__except_handler(
+            [&__exec, __first, __last, __count, &__value, __pred]()
+            {
+                return __internal::__parallel_find(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                    [__last, __count, &__value, __pred](_RandomAccessIterator __i, _RandomAccessIterator __j)
+                    { return __internal::__find_subrange(__i, __j, __last, __count, __value, __pred, _IsVector{}); },
+                    std::less<typename std::iterator_traits<_RandomAccessIterator>::difference_type>(),
+                    /*is_first=*/true);
+            });
     }
 }
 
@@ -848,12 +886,14 @@ __brick_copy_n(_ForwardIterator __first, _Size __n, _OutputIterator __result, /*
     return std::copy_n(__first, __n, __result);
 }
 
-template <class _ForwardIterator, class _Size, class _OutputIterator>
-_OutputIterator
-__brick_copy_n(_ForwardIterator __first, _Size __n, _OutputIterator __result, /*vector=*/std::true_type) noexcept
+template <class _RandomAccessIterator1, class _Size, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__brick_copy_n(_RandomAccessIterator1 __first, _Size __n, _RandomAccessIterator2 __result,
+               /*vector=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_assign(
-        __first, __n, __result, [](_ForwardIterator __first, _OutputIterator __result) { *__result = *__first; });
+        __first, __n, __result,
+        [](_RandomAccessIterator1 __first, _RandomAccessIterator2 __result) { *__result = *__first; });
 }
 
 //------------------------------------------------------------------------
@@ -867,14 +907,14 @@ __brick_copy(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator
     return std::copy(__first, __last, __result);
 }
 
-template <class _RandomAccessIterator, class _OutputIterator>
-_OutputIterator
-__brick_copy(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__brick_copy(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
              /*vector=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_assign(
         __first, __last - __first, __result,
-        [](_RandomAccessIterator __first, _OutputIterator __result) { *__result = *__first; });
+        [](_RandomAccessIterator1 __first, _RandomAccessIterator2 __result) { *__result = *__first; });
 }
 
 //------------------------------------------------------------------------
@@ -888,36 +928,38 @@ __brick_move(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator
     return std::move(__first, __last, __result);
 }
 
-template <class _RandomAccessIterator, class _OutputIterator>
-_OutputIterator
-__brick_move(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__brick_move(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
              /*vector=*/std::true_type) noexcept
 {
     return __unseq_backend::__simd_assign(
         __first, __last - __first, __result,
-        [](_RandomAccessIterator __first, _OutputIterator __result) { *__result = std::move(*__first); });
+        [](_RandomAccessIterator1 __first, _RandomAccessIterator2 __result) { *__result = std::move(*__first); });
 }
 
 struct __brick_move_destroy
 {
-    template <typename _Iterator, typename _OutputIterator>
-    _OutputIterator
-    operator()(_Iterator __first, _Iterator __last, _OutputIterator __result, /*vec*/ std::true_type) const
+    template <typename _RandomAccessIterator1, typename _RandomAccessIterator2>
+    _RandomAccessIterator2
+    operator()(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
+               /*vec*/ std::true_type) const
     {
-        using _IteratorValueType = typename std::iterator_traits<_Iterator>::value_type;
+        using _IteratorValueType = typename std::iterator_traits<_RandomAccessIterator1>::value_type;
 
         return __unseq_backend::__simd_assign(__first, __last - __first, __result,
-                                              [](_Iterator __first, _OutputIterator __result) {
+                                              [](_RandomAccessIterator1 __first, _RandomAccessIterator2 __result) {
                                                   *__result = std::move(*__first);
                                                   (*__first).~_IteratorValueType();
                                               });
     }
 
-    template <typename _Iterator, typename _OutputIterator>
-    _OutputIterator
-    operator()(_Iterator __first, _Iterator __last, _OutputIterator __result, /*vec*/ std::false_type) const
+    template <typename _RandomAccessIterator1, typename _RandomAccessIterator2>
+    _RandomAccessIterator2
+    operator()(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
+               /*vec*/ std::false_type) const
     {
-        using _IteratorValueType = typename std::iterator_traits<_Iterator>::value_type;
+        using _IteratorValueType = typename std::iterator_traits<_RandomAccessIterator1>::value_type;
 
         for (; __first != __last; ++__first, ++__result)
         {
@@ -939,14 +981,14 @@ __brick_swap_ranges(_ForwardIterator __first, _ForwardIterator __last, _OutputIt
     return std::swap_ranges(__first, __last, __result);
 }
 
-template <class _ForwardIterator, class _OutputIterator>
-_OutputIterator
-__brick_swap_ranges(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__brick_swap_ranges(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
                     /*vector=*/std::true_type) noexcept
 {
     using std::iter_swap;
     return __unseq_backend::__simd_assign(__first, __last - __first, __result,
-                                          iter_swap<_ForwardIterator, _OutputIterator>);
+                                          iter_swap<_RandomAccessIterator1, _RandomAccessIterator2>);
 }
 
 //------------------------------------------------------------------------
@@ -960,12 +1002,13 @@ __brick_copy_if(_ForwardIterator __first, _ForwardIterator __last, _OutputIterat
     return std::copy_if(__first, __last, __result, __pred);
 }
 
-template <class _ForwardIterator, class _OutputIterator, class _UnaryPredicate>
-_OutputIterator
-__brick_copy_if(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result, _UnaryPredicate __pred,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _UnaryPredicate>
+_RandomAccessIterator2
+__brick_copy_if(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
+                _UnaryPredicate __pred,
                 /*vector=*/std::true_type) noexcept
 {
-#if (_PSTL_MONOTONIC_PRESENT)
+#if defined(_PSTL_MONOTONIC_PRESENT)
     return __unseq_backend::__simd_copy_if(__first, __last - __first, __result, __pred);
 #else
     return std::copy_if(__first, __last, __result, __pred);
@@ -981,7 +1024,7 @@ __brick_calc_mask_1(_ForwardIterator __first, _ForwardIterator __last, bool* __r
     auto __count_true = _DifferenceType(0);
     auto __size = __last - __first;
 
-    static_assert(__is_random_access_iterator<_ForwardIterator>::value,
+    static_assert(__are_random_access_iterators<_ForwardIterator>::value,
                   "Pattern-brick error. Should be a random access iterator.");
 
     for (; __first != __last; ++__first, ++__mask)
@@ -1019,12 +1062,12 @@ __brick_copy_by_mask(_ForwardIterator __first, _ForwardIterator __last, _OutputI
     }
 }
 
-template <class _ForwardIterator, class _OutputIterator, class _Assigner>
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Assigner>
 void
-__brick_copy_by_mask(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
+__brick_copy_by_mask(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
                      bool* __restrict __mask, _Assigner __assigner, /*vector=*/std::true_type) noexcept
 {
-#if (_PSTL_MONOTONIC_PRESENT)
+#if defined(_PSTL_MONOTONIC_PRESENT)
     __unseq_backend::__simd_copy_by_mask(__first, __last - __first, __result, __mask, __assigner);
 #else
     __internal::__brick_copy_by_mask(__first, __last, __result, __mask, __assigner, std::false_type());
@@ -1051,67 +1094,72 @@ __brick_partition_by_mask(_ForwardIterator __first, _ForwardIterator __last, _Ou
     }
 }
 
-template <class _RandomAccessIterator, class _OutputIterator1, class _OutputIterator2>
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _RandomAccessIterator3>
 void
-__brick_partition_by_mask(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator1 __out_true,
-                          _OutputIterator2 __out_false, bool* __mask, /*vector=*/std::true_type) noexcept
+__brick_partition_by_mask(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last,
+                          _RandomAccessIterator2 __out_true, _RandomAccessIterator3 __out_false, bool* __mask,
+                          /*vector=*/std::true_type) noexcept
 {
-#if (_PSTL_MONOTONIC_PRESENT)
+#if defined(_PSTL_MONOTONIC_PRESENT)
     __unseq_backend::__simd_partition_by_mask(__first, __last - __first, __out_true, __out_false, __mask);
 #else
     __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>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryPredicate>
 _OutputIterator
-__pattern_copy_if(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
-                  _UnaryPredicate __pred, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
+__pattern_copy_if(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
+                  _UnaryPredicate __pred) noexcept
 {
-    return __internal::__brick_copy_if(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_copy_if(__first, __last, __result, __pred, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _UnaryPredicate>
+_RandomAccessIterator2
+__pattern_copy_if(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                  _RandomAccessIterator1 __last, _RandomAccessIterator2 __result, _UnaryPredicate __pred)
 {
-    typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::difference_type _DifferenceType;
     const _DifferenceType __n = __last - __first;
     if (_DifferenceType(1) < __n)
     {
         __par_backend::__buffer<bool> __mask_buf(__n);
-        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 __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
-                    __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;
-        });
+        return __internal::__except_handler(
+            [&__exec, __n, __first, __result, __pred, &__mask_buf]()
+            {
+                bool* __mask = __mask_buf.get();
+                _DifferenceType __m{};
+                __par_backend::__parallel_strict_scan(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __n, _DifferenceType(0),
+                    [=](_DifferenceType __i, _DifferenceType __len) { // Reduce
+                        return __internal::__brick_calc_mask_1<_DifferenceType>(__first + __i, __first + (__i + __len),
+                                                                                __mask + __i, __pred, _IsVector{})
+                            .first;
+                    },
+                    std::plus<_DifferenceType>(),                                                // Combine
+                    [=](_DifferenceType __i, _DifferenceType __len, _DifferenceType __initial) { // Scan
+                        __internal::__brick_copy_by_mask(
+                            __first + __i, __first + (__i + __len), __result + __initial, __mask + __i,
+                            [](_RandomAccessIterator1 __x, _RandomAccessIterator2 __z) { *__z = *__x; }, _IsVector{});
+                    },
+                    [&__m](_DifferenceType __total) { __m = __total; });
+                return __result + __m;
+            });
     }
     // trivial sequence - use serial algorithm
-    return __internal::__brick_copy_if(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_copy_if(__first, __last, __result, __pred, _IsVector{});
 }
 
 //------------------------------------------------------------------------
 // count
 //------------------------------------------------------------------------
-template <class _ForwardIterator, class _Predicate>
-typename std::iterator_traits<_ForwardIterator>::difference_type
-__brick_count(_ForwardIterator __first, _ForwardIterator __last, _Predicate __pred,
+template <class _RandomAccessIterator, class _Predicate>
+typename std::iterator_traits<_RandomAccessIterator>::difference_type
+__brick_count(_RandomAccessIterator __first, _RandomAccessIterator __last, _Predicate __pred,
               /* is_vector = */ std::true_type) noexcept
 {
     return __unseq_backend::__simd_count(__first, __last - __first, __pred);
@@ -1125,66 +1173,70 @@ __brick_count(_ForwardIterator __first, _ForwardIterator __last, _Predicate __pr
     return std::count_if(__first, __last, __pred);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
 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
+__pattern_count(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred) noexcept
 {
-    return __internal::__brick_count(__first, __last, __pred, __is_vector);
+    return __internal::__brick_count(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Predicate>
+typename std::iterator_traits<_RandomAccessIterator>::difference_type
+__pattern_count(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                _RandomAccessIterator __last, _Predicate __pred)
 {
-    typedef typename std::iterator_traits<_ForwardIterator>::difference_type _SizeType;
-    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 + __internal::__brick_count(__begin, __end, __pred, __is_vector);
-            },
-            std::plus<_SizeType>());
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _SizeType;
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __par_backend::__parallel_reduce(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last, _SizeType(0),
+                [__pred](_RandomAccessIterator __begin, _RandomAccessIterator __end, _SizeType __value) -> _SizeType
+                { return __value + __internal::__brick_count(__begin, __end, __pred, _IsVector{}); },
+                std::plus<_SizeType>());
+        });
 }
 
 //------------------------------------------------------------------------
 // unique
 //------------------------------------------------------------------------
 
-template <class _ForwardIterator, class _BinaryPredicate>
-_ForwardIterator
-__brick_unique(_ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
+template <class _RandomAccessIterator, class _BinaryPredicate>
+_RandomAccessIterator
+__brick_unique(_RandomAccessIterator __first, _RandomAccessIterator __last, _BinaryPredicate __pred,
                /*is_vector=*/std::false_type) noexcept
 {
     return std::unique(__first, __last, __pred);
 }
 
-template <class _ForwardIterator, class _BinaryPredicate>
-_ForwardIterator
-__brick_unique(_ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
+template <class _RandomAccessIterator, class _BinaryPredicate>
+_RandomAccessIterator
+__brick_unique(_RandomAccessIterator __first, _RandomAccessIterator __last, _BinaryPredicate __pred,
                /*is_vector=*/std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::unique(__first, __last, __pred);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate>
 _ForwardIterator
-__pattern_unique(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
-                 _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+__pattern_unique(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                 _BinaryPredicate __pred) noexcept
 {
-    return __internal::__brick_unique(__first, __last, __pred, __is_vector);
+    return __internal::__brick_unique(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
 // 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>
+template <class _IsVector, class _ExecutionPolicy, class _ForwardIterator, class _CalcMask>
 _ForwardIterator
-__remove_elements(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _CalcMask __calc_mask,
-                  _IsVector __is_vector)
+__remove_elements(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _ForwardIterator __first,
+                  _ForwardIterator __last, _CalcMask __calc_mask)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     typedef typename std::iterator_traits<_ForwardIterator>::difference_type _DifferenceType;
     typedef typename std::iterator_traits<_ForwardIterator>::value_type _Tp;
     _DifferenceType __n = __last - __first;
@@ -1193,9 +1245,10 @@ __remove_elements(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardI
     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 {
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), _DifferenceType(0), __n, __n,
+            [__first, __mask, &__calc_mask](_DifferenceType __i, _DifferenceType __j,
+                                            _DifferenceType __local_min) -> _DifferenceType
+            {
                 // Create mask
                 __calc_mask(__mask + __i, __mask + __j, __first + __i);
 
@@ -1205,17 +1258,16 @@ __remove_elements(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardI
                     return __local_min;
                 }
                 // find first iterator that should be removed
-                bool* __result = __internal::__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; }, _IsVector{});
                 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);
-            });
+            [](_DifferenceType __local_min1, _DifferenceType __local_min2) -> _DifferenceType
+            { return std::min(__local_min1, __local_min2); });
 
         // No elements to remove - exit
         if (__min == __n)
@@ -1231,44 +1283,46 @@ __remove_elements(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardI
         _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 __internal::__brick_count(__mask + __i, __mask + __i + __len, [](bool __val) { return __val; },
-                                                 __is_vector);
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __n, _DifferenceType(0),
+            [__mask](_DifferenceType __i, _DifferenceType __len)
+            {
+                return __internal::__brick_count(
+                    __mask + __i, __mask + __i + __len, [](bool __val) { return __val; }, _IsVector{});
             },
             std::plus<_DifferenceType>(),
-            [=](_DifferenceType __i, _DifferenceType __len, _DifferenceType __initial) {
+            [=](_DifferenceType __i, _DifferenceType __len, _DifferenceType __initial)
+            {
                 __internal::__brick_copy_by_mask(
                     __first + __i, __first + __i + __len, __result + __initial, __mask + __i,
-                    [](_ForwardIterator __x, _Tp* __z) {
-                        __internal::__invoke_if_else(std::is_trivial<_Tp>(), [&]() { *__z = std::move(*__x); },
-                                                     [&]() { ::new (std::addressof(*__z)) _Tp(std::move(*__x)); });
+                    [](_ForwardIterator __x, _Tp* __z)
+                    {
+                        __internal::__invoke_if_else(
+                            std::is_trivial<_Tp>(), [&]() { *__z = std::move(*__x); },
+                            [&]() { ::new (std::addressof(*__z)) _Tp(std::move(*__x)); });
                     },
-                    __is_vector);
+                    _IsVector{});
             },
             [&__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) {
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __result, __result + __m,
+            [__result, __first](_Tp* __i, _Tp* __j)
+            {
                 __invoke_if_else(
-                    std::is_trivial<_Tp>(),
-                    [&]() { __brick_move(__i, __j, __first + (__i - __result), __is_vector); },
-                    [&]() {
-                        __brick_move_destroy()(__i, __j, __first + (__i - __result), __is_vector);
-                    });
+                    std::is_trivial<_Tp>(), [&]() { __brick_move(__i, __j, __first + (__i - __result), _IsVector{}); },
+                    [&]() { __brick_move_destroy()(__i, __j, __first + (__i - __result), _IsVector{}); });
             });
         return __first + __m;
     });
 }
 
-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
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _BinaryPredicate>
+_RandomAccessIterator
+__pattern_unique(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                 _RandomAccessIterator __last, _BinaryPredicate __pred) noexcept
 {
-    typedef typename std::iterator_traits<_ForwardIterator>::reference _ReferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator>::reference _ReferenceType;
 
     if (__first == __last)
     {
@@ -1277,16 +1331,16 @@ __pattern_unique(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIt
     if (__first + 1 == __last || __first + 2 == __last)
     {
         // Trivial sequence - use serial algorithm
-        return __internal::__brick_unique(__first, __last, __pred, __is_vector);
+        return __internal::__brick_unique(__first, __last, __pred, _IsVector{});
     }
     return __internal::__remove_elements(
-        std::forward<_ExecutionPolicy>(__exec), ++__first, __last,
-        [&__pred, __is_vector](bool* __b, bool* __e, _ForwardIterator __it) {
+        __tag, std::forward<_ExecutionPolicy>(__exec), ++__first, __last,
+        [&__pred](bool* __b, bool* __e, _RandomAccessIterator __it)
+        {
             __internal::__brick_walk3(
                 __b, __e, __it - 1, __it,
-                [&__pred](bool& __x, _ReferenceType __y, _ReferenceType __z) { __x = !__pred(__y, __z); }, __is_vector);
-        },
-        __is_vector);
+                [&__pred](bool& __x, _ReferenceType __y, _ReferenceType __z) { __x = !__pred(__y, __z); }, _IsVector{});
+        });
 }
 
 //------------------------------------------------------------------------
@@ -1294,32 +1348,31 @@ __pattern_unique(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIt
 //------------------------------------------------------------------------
 
 template <class _ForwardIterator, class _OutputIterator, class _BinaryPredicate>
-  _OutputIterator
+_OutputIterator
 __brick_unique_copy(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result, _BinaryPredicate __pred,
                     /*vector=*/std::false_type) noexcept
 {
     return std::unique_copy(__first, __last, __result, __pred);
 }
 
-template <class _RandomAccessIterator, class _OutputIterator, class _BinaryPredicate>
-  _OutputIterator
-__brick_unique_copy(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryPredicate>
+_RandomAccessIterator2
+__brick_unique_copy(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result,
                     _BinaryPredicate __pred, /*vector=*/std::true_type) noexcept
 {
-#if (_PSTL_MONOTONIC_PRESENT)
+#if defined(_PSTL_MONOTONIC_PRESENT)
     return __unseq_backend::__simd_unique_copy(__first, __last - __first, __result, __pred);
 #else
     return std::unique_copy(__first, __last, __result, __pred);
 #endif
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryPredicate,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryPredicate>
 _OutputIterator
-__pattern_unique_copy(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
-                      _BinaryPredicate __pred, _IsVector __is_vector, /*parallel=*/std::false_type) noexcept
+__pattern_unique_copy(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                      _OutputIterator __result, _BinaryPredicate __pred) noexcept
 {
-    return __internal::__brick_unique_copy(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_unique_copy(__first, __last, __result, __pred, typename _Tag::__is_vector{});
 }
 
 template <class _DifferenceType, class _RandomAccessIterator, class _BinaryPredicate>
@@ -1344,54 +1397,58 @@ __brick_calc_mask_2(_RandomAccessIterator __first, _RandomAccessIterator __last,
     return __unseq_backend::__simd_calc_mask_2(__first, __last - __first, __mask, __pred);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _BinaryPredicate,
-          class _IsVector>
-_OutputIterator
-__pattern_unique_copy(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
-                      _OutputIterator __result, _BinaryPredicate __pred, _IsVector __is_vector,
-                      /*parallel=*/std::true_type)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryPredicate>
+_RandomAccessIterator2
+__pattern_unique_copy(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                      _RandomAccessIterator1 __last, _RandomAccessIterator2 __result, _BinaryPredicate __pred)
 {
-    typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::difference_type _DifferenceType;
     const _DifferenceType __n = __last - __first;
     if (_DifferenceType(2) < __n)
     {
         __par_backend::__buffer<bool> __mask_buf(__n);
         if (_DifferenceType(2) < __n)
         {
-            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 __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
-                        __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;
-            });
+            return __internal::__except_handler(
+                [&__exec, __n, __first, __result, __pred, &__mask_buf]()
+                {
+                    bool* __mask = __mask_buf.get();
+                    _DifferenceType __m{};
+                    __par_backend::__parallel_strict_scan(
+                        __backend_tag{}, 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 __internal::__brick_calc_mask_2<_DifferenceType>(
+                                       __first + __i, __first + (__i + __len), __mask + __i, __pred, _IsVector{}) +
+                                   __extra;
+                        },
+                        std::plus<_DifferenceType>(),                                                // Combine
+                        [=](_DifferenceType __i, _DifferenceType __len, _DifferenceType __initial) { // Scan
+                            // Phase 2 is same as for __pattern_copy_if
+                            __internal::__brick_copy_by_mask(
+                                __first + __i, __first + (__i + __len), __result + __initial, __mask + __i,
+                                [](_RandomAccessIterator1 __x, _RandomAccessIterator2 __z) { *__z = *__x; },
+                                _IsVector{});
+                        },
+                        [&__m](_DifferenceType __total) { __m = __total; });
+                    return __result + __m;
+                });
         }
     }
     // trivial sequence - use serial algorithm
-    return __internal::__brick_unique_copy(__first, __last, __result, __pred, __is_vector);
+    return __internal::__brick_unique_copy(__first, __last, __result, __pred, _IsVector{});
 }
 
 //------------------------------------------------------------------------
@@ -1404,14 +1461,14 @@ __brick_reverse(_BidirectionalIterator __first, _BidirectionalIterator __last, /
     std::reverse(__first, __last);
 }
 
-template <class _BidirectionalIterator>
+template <class _RandomAccessIterator>
 void
-__brick_reverse(_BidirectionalIterator __first, _BidirectionalIterator __last, /*__is_vector=*/std::true_type) noexcept
+__brick_reverse(_RandomAccessIterator __first, _RandomAccessIterator __last, /*__is_vector=*/std::true_type) noexcept
 {
-    typedef typename std::iterator_traits<_BidirectionalIterator>::reference _ReferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator>::reference _ReferenceType;
 
     const auto __n = (__last - __first) / 2;
-    __unseq_backend::__simd_walk_2(__first, __n, std::reverse_iterator<_BidirectionalIterator>(__last),
+    __unseq_backend::__simd_walk_2(__first, __n, std::reverse_iterator<_RandomAccessIterator>(__last),
                                    [](_ReferenceType __x, _ReferenceType __y) {
                                        using std::swap;
                                        swap(__x, __y);
@@ -1432,39 +1489,38 @@ __brick_reverse(_BidirectionalIterator __first, _BidirectionalIterator __last, _
 }
 
 // this brick is called in parallel version, so we can use iterator arithmetic
-template <class _BidirectionalIterator>
+template <class _RandomAccessIterator>
 void
-__brick_reverse(_BidirectionalIterator __first, _BidirectionalIterator __last, _BidirectionalIterator __d_last,
+__brick_reverse(_RandomAccessIterator __first, _RandomAccessIterator __last, _RandomAccessIterator __d_last,
                 /*is_vector=*/std::true_type) noexcept
 {
-    typedef typename std::iterator_traits<_BidirectionalIterator>::reference _ReferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator>::reference _ReferenceType;
 
-    __unseq_backend::__simd_walk_2(__first, __last - __first, std::reverse_iterator<_BidirectionalIterator>(__d_last),
+    __unseq_backend::__simd_walk_2(__first, __last - __first, std::reverse_iterator<_RandomAccessIterator>(__d_last),
                                    [](_ReferenceType __x, _ReferenceType __y) {
                                        using std::swap;
                                        swap(__x, __y);
                                    });
 }
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator>
 void
-__pattern_reverse(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
-                  _IsVector __is_vector,
-                  /*is_parallel=*/std::false_type) noexcept
+__pattern_reverse(_Tag, _ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last) noexcept
 {
-    __internal::__brick_reverse(__first, __last, __is_vector);
+    __internal::__brick_reverse(__first, __last, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator>
 void
-__pattern_reverse(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __last,
-                  _IsVector __is_vector, /*is_parallel=*/std::true_type)
+__pattern_reverse(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                  _RandomAccessIterator __last)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     __par_backend::__parallel_for(
-        std::forward<_ExecutionPolicy>(__exec), __first, __first + (__last - __first) / 2,
-        [__is_vector, __first, __last](_BidirectionalIterator __inner_first, _BidirectionalIterator __inner_last) {
-            __internal::__brick_reverse(__inner_first, __inner_last, __last - (__inner_first - __first), __is_vector);
-        });
+        __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __first + (__last - __first) / 2,
+        [__first, __last](_RandomAccessIterator __inner_first, _RandomAccessIterator __inner_last)
+        { __internal::__brick_reverse(__inner_first, __inner_last, __last - (__inner_first - __first), _IsVector{}); });
 }
 
 //------------------------------------------------------------------------
@@ -1479,39 +1535,41 @@ __brick_reverse_copy(_BidirectionalIterator __first, _BidirectionalIterator __la
     return std::reverse_copy(__first, __last, __d_first);
 }
 
-template <class _BidirectionalIterator, class _OutputIterator>
-_OutputIterator
-__brick_reverse_copy(_BidirectionalIterator __first, _BidirectionalIterator __last, _OutputIterator __d_first,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__brick_reverse_copy(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __d_first,
                      /*is_vector=*/std::true_type) noexcept
 {
-    typedef typename std::iterator_traits<_BidirectionalIterator>::reference _ReferenceType1;
-    typedef typename std::iterator_traits<_OutputIterator>::reference _ReferenceType2;
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::reference _ReferenceType1;
+    typedef typename std::iterator_traits<_RandomAccessIterator2>::reference _ReferenceType2;
 
-    return __unseq_backend::__simd_walk_2(std::reverse_iterator<_BidirectionalIterator>(__last), __last - __first,
+    return __unseq_backend::__simd_walk_2(std::reverse_iterator<_RandomAccessIterator1>(__last), __last - __first,
                                           __d_first, [](_ReferenceType1 __x, _ReferenceType2 __y) { __y = __x; });
 }
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _OutputIterator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator, class _OutputIterator>
 _OutputIterator
-__pattern_reverse_copy(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
-                       _OutputIterator __d_first, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+__pattern_reverse_copy(_Tag, _ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
+                       _OutputIterator __d_first) noexcept
 {
-    return __internal::__brick_reverse_copy(__first, __last, __d_first, __is_vector);
+    return __internal::__brick_reverse_copy(__first, __last, __d_first, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__pattern_reverse_copy(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                       _RandomAccessIterator1 __last, _RandomAccessIterator2 __d_first)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     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) {
-                                      __internal::__brick_reverse_copy(__inner_first, __inner_last,
-                                                                       __d_first + (__len - (__inner_last - __first)),
-                                                                       __is_vector);
-                                  });
+    __par_backend::__parallel_for(
+        __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+        [__first, __len, __d_first](_RandomAccessIterator1 __inner_first, _RandomAccessIterator1 __inner_last)
+        {
+            __internal::__brick_reverse_copy(__inner_first, __inner_last,
+                                             __d_first + (__len - (__inner_last - __first)), _IsVector{});
+        });
     return __d_first + __len;
 }
 
@@ -1523,7 +1581,7 @@ _ForwardIterator
 __brick_rotate(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
                /*is_vector=*/std::false_type) noexcept
 {
-#if _PSTL_CPP11_STD_ROTATE_BROKEN
+#if defined(_PSTL_CPP11_STD_ROTATE_BROKEN)
     std::rotate(__first, __middle, __last);
     return std::next(__first, std::distance(__middle, __last));
 #else
@@ -1531,14 +1589,14 @@ __brick_rotate(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIter
 #endif
 }
 
-template <class _ForwardIterator>
-_ForwardIterator
-__brick_rotate(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
+template <class _RandomAccessIterator>
+_RandomAccessIterator
+__brick_rotate(_RandomAccessIterator __first, _RandomAccessIterator __middle, _RandomAccessIterator __last,
                /*is_vector=*/std::true_type) noexcept
 {
     auto __n = __last - __first;
     auto __m = __middle - __first;
-    const _ForwardIterator __ret = __first + (__last - __middle);
+    const _RandomAccessIterator __ret = __first + (__last - __middle);
 
     bool __is_left = (__m <= __n / 2);
     if (!__is_left)
@@ -1553,7 +1611,7 @@ __brick_rotate(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIter
             for (; __last - __first >= __m_2; __first += __m)
             {
                 __unseq_backend::__simd_assign(__first, __m, __first + __m,
-                                               iter_swap<_ForwardIterator, _ForwardIterator>);
+                                               iter_swap<_RandomAccessIterator, _RandomAccessIterator>);
             }
         }
         else
@@ -1561,7 +1619,7 @@ __brick_rotate(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIter
             for (; __last - __first >= __m_2; __last -= __m)
             {
                 __unseq_backend::__simd_assign(__last - __m, __m, __last - __m_2,
-                                               iter_swap<_ForwardIterator, _ForwardIterator>);
+                                               iter_swap<_RandomAccessIterator, _RandomAccessIterator>);
             }
         }
         __is_left = !__is_left;
@@ -1572,73 +1630,73 @@ __brick_rotate(_ForwardIterator __first, _ForwardIterator __middle, _ForwardIter
     return __ret;
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator>
 _ForwardIterator
-__pattern_rotate(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
-                 _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+__pattern_rotate(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __middle,
+                 _ForwardIterator __last) noexcept
 {
-    return __internal::__brick_rotate(__first, __middle, __last, __is_vector);
+    return __internal::__brick_rotate(__first, __middle, __last, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator>
+_RandomAccessIterator
+__pattern_rotate(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                 _RandomAccessIterator __middle, _RandomAccessIterator __last)
 {
-    typedef typename std::iterator_traits<_ForwardIterator>::value_type _Tp;
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    typedef typename std::iterator_traits<_RandomAccessIterator>::value_type _Tp;
     auto __n = __last - __first;
     auto __m = __middle - __first;
     if (__m <= __n / 2)
     {
         __par_backend::__buffer<_Tp> __buf(__n - __m);
-        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) {
-                    __internal::__brick_uninitialized_move(__b, __e, __result + (__b - __middle), __is_vector);
-                });
+        return __internal::__except_handler(
+            [&__exec, __n, __m, __first, __middle, __last, &__buf]()
+            {
+                _Tp* __result = __buf.get();
+                __par_backend::__parallel_for(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __middle, __last,
+                    [__middle, __result](_RandomAccessIterator __b, _RandomAccessIterator __e)
+                    { __internal::__brick_uninitialized_move(__b, __e, __result + (__b - __middle), _IsVector{}); });
 
-            __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __middle,
-                                          [__last, __middle, __is_vector](_ForwardIterator __b, _ForwardIterator __e) {
-                                              __internal::__brick_move(__b, __e, __b + (__last - __middle),
-                                                                       __is_vector);
-                                          });
+                __par_backend::__parallel_for(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __middle,
+                    [__last, __middle](_RandomAccessIterator __b, _RandomAccessIterator __e)
+                    { __internal::__brick_move(__b, __e, __b + (__last - __middle), _IsVector{}); });
 
-            __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __result, __result + (__n - __m),
-                                          [__first, __result, __is_vector](_Tp* __b, _Tp* __e) {
-                                              __brick_move_destroy()(
-                                                  __b, __e, __first + (__b - __result), __is_vector);
-                                          });
+                __par_backend::__parallel_for(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __result, __result + (__n - __m),
+                    [__first, __result](_Tp* __b, _Tp* __e)
+                    { __brick_move_destroy()(__b, __e, __first + (__b - __result), _IsVector{}); });
 
-            return __first + (__last - __middle);
-        });
+                return __first + (__last - __middle);
+            });
     }
     else
     {
         __par_backend::__buffer<_Tp> __buf(__m);
-        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) {
-                                              __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) {
-                                              __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_destroy()(
-                                                  __b, __e, __first + ((__n - __m) + (__b - __result)), __is_vector);
-                                          });
-
-            return __first + (__last - __middle);
-        });
+        return __internal::__except_handler(
+            [&__exec, __n, __m, __first, __middle, __last, &__buf]()
+            {
+                _Tp* __result = __buf.get();
+                __par_backend::__parallel_for(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __middle,
+                    [__first, __result](_RandomAccessIterator __b, _RandomAccessIterator __e)
+                    { __internal::__brick_uninitialized_move(__b, __e, __result + (__b - __first), _IsVector{}); });
+
+                __par_backend::__parallel_for(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __middle, __last,
+                    [__first, __middle](_RandomAccessIterator __b, _RandomAccessIterator __e)
+                    { __internal::__brick_move(__b, __e, __first + (__b - __middle), _IsVector{}); });
+
+                __par_backend::__parallel_for(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __result, __result + __m,
+                    [__n, __m, __first, __result](_Tp* __b, _Tp* __e)
+                    { __brick_move_destroy()(__b, __e, __first + ((__n - __m) + (__b - __result)), _IsVector{}); });
+
+                return __first + (__last - __middle);
+            });
     }
 }
 
@@ -1654,47 +1712,49 @@ __brick_rotate_copy(_ForwardIterator __first, _ForwardIterator __middle, _Forwar
     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
+template <class _RandomAccessIterator1, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__brick_rotate_copy(_RandomAccessIterator1 __first, _RandomAccessIterator1 __middle, _RandomAccessIterator1 __last,
+                    _RandomAccessIterator2 __result, /*__is_vector=*/std::true_type) noexcept
 {
-    _OutputIterator __res = __internal::__brick_copy(__middle, __last, __result, std::true_type());
+    _RandomAccessIterator2 __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>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator>
 _OutputIterator
-__pattern_rotate_copy(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last,
-                      _OutputIterator __result, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+__pattern_rotate_copy(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __middle,
+                      _ForwardIterator __last, _OutputIterator __result) noexcept
 {
-    return __internal::__brick_rotate_copy(__first, __middle, __last, __result, __is_vector);
+    return __internal::__brick_rotate_copy(__first, __middle, __last, __result, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2>
+_RandomAccessIterator2
+__pattern_rotate_copy(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                      _RandomAccessIterator1 __middle, _RandomAccessIterator1 __last, _RandomAccessIterator2 __result)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     __par_backend::__parallel_for(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last,
-        [__first, __last, __middle, __result, __is_vector](_ForwardIterator __b, _ForwardIterator __e) {
+        __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+        [__first, __last, __middle, __result](_RandomAccessIterator1 __b, _RandomAccessIterator1 __e)
+        {
             if (__b > __middle)
             {
-                __internal::__brick_copy(__b, __e, __result + (__b - __middle), __is_vector);
+                __internal::__brick_copy(__b, __e, __result + (__b - __middle), _IsVector{});
             }
             else
             {
-                _OutputIterator __new_result = __result + ((__last - __middle) + (__b - __first));
+                _RandomAccessIterator2 __new_result = __result + ((__last - __middle) + (__b - __first));
                 if (__e < __middle)
                 {
-                    __internal::__brick_copy(__b, __e, __new_result, __is_vector);
+                    __internal::__brick_copy(__b, __e, __new_result, _IsVector{});
                 }
                 else
                 {
-                    __internal::__brick_copy(__b, __middle, __new_result, __is_vector);
-                    __internal::__brick_copy(__middle, __e, __result, __is_vector);
+                    __internal::__brick_copy(__b, __middle, __new_result, _IsVector{});
+                    __internal::__brick_copy(__middle, __e, __result, _IsVector{});
                 }
             }
         });
@@ -1713,21 +1773,21 @@ __brick_is_partitioned(_ForwardIterator __first, _ForwardIterator __last, _Unary
     return std::is_partitioned(__first, __last, __pred);
 }
 
-template <class _ForwardIterator, class _UnaryPredicate>
+template <class _RandomAccessIterator, class _UnaryPredicate>
 bool
-__brick_is_partitioned(_ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
+__brick_is_partitioned(_RandomAccessIterator __first, _RandomAccessIterator __last, _UnaryPredicate __pred,
                        /*is_vector=*/std::true_type) noexcept
 {
-    typedef typename std::iterator_traits<_ForwardIterator>::difference_type _SizeType;
+    typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _SizeType;
     if (__first == __last)
     {
         return true;
     }
     else
     {
-        _ForwardIterator __result = __unseq_backend::__simd_first(
+        _RandomAccessIterator __result = __unseq_backend::__simd_first(
             __first, _SizeType(0), __last - __first,
-            [&__pred](_ForwardIterator __it, _SizeType __i) { return !__pred(__it[__i]); });
+            [&__pred](_RandomAccessIterator __it, _SizeType __i) { return !__pred(__it[__i]); });
         if (__result == __last)
         {
             return true;
@@ -1740,18 +1800,18 @@ __brick_is_partitioned(_ForwardIterator __first, _ForwardIterator __last, _Unary
     }
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 bool
-__pattern_is_partitioned(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
-                         _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+__pattern_is_partitioned(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                         _UnaryPredicate __pred) noexcept
 {
-    return __internal::__brick_is_partitioned(__first, __last, __pred, __is_vector);
+    return __internal::__brick_is_partitioned(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
 bool
-__pattern_is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
-                         _UnaryPredicate __pred, _IsVector __is_vector, /*is_parallel=*/std::true_type)
+__pattern_is_partitioned(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                         _RandomAccessIterator __last, _UnaryPredicate __pred)
 {
     if (__first == __last)
     {
@@ -1759,6 +1819,8 @@ __pattern_is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _F
     }
     else
     {
+        using __backend_tag = typename decltype(__tag)::__backend_tag;
+
         return __internal::__except_handler([&]() {
             // State of current range:
             // broken     - current range is not partitioned by pred
@@ -1782,9 +1844,10 @@ __pattern_is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _F
                                      __broken,     __broken,     __true_false, __broken};
 
             __init = __par_backend::__parallel_reduce(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last, __init,
-                [&__pred, &__table, __is_vector](_ForwardIterator __i, _ForwardIterator __j,
-                                                 _ReduceType __value) -> _ReduceType {
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last, __init,
+                [&__pred, &__table](_RandomAccessIterator __i, _RandomAccessIterator __j,
+                                    _ReduceType __value) -> _ReduceType
+                {
                     if (__value == __broken)
                     {
                         return __broken;
@@ -1794,12 +1857,12 @@ __pattern_is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _F
                     if (__pred(*__i))
                     {
                         // find first element that don't satisfy pred
-                        _ForwardIterator __x =
-                            __internal::__brick_find_if(__i + 1, __j, std::not_fn(__pred), __is_vector);
+                        _RandomAccessIterator __x =
+                            __internal::__brick_find_if(__i + 1, __j, std::not_fn(__pred), _IsVector{});
                         if (__x != __j)
                         {
                             // find first element after "x" that satisfy pred
-                            _ForwardIterator __y = __internal::__brick_find_if(__x + 1, __j, __pred, __is_vector);
+                            _RandomAccessIterator __y = __internal::__brick_find_if(__x + 1, __j, __pred, _IsVector{});
                             // if it was found then range isn't partitioned by pred
                             if (__y != __j)
                             {
@@ -1819,7 +1882,7 @@ __pattern_is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _F
                     { // 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 (__internal::__brick_find_if(__i + 1, __j, __pred, __is_vector) != __j)
+                        if (__internal::__brick_find_if(__i + 1, __j, __pred, _IsVector{}) != __j)
                         {
                             return __broken;
                         }
@@ -1832,7 +1895,8 @@ __pattern_is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _F
                     return (__value == -1) ? __res : __table[__value * 4 + __res];
                 },
 
-                [&__table](_ReduceType __val1, _ReduceType __val2) -> _ReduceType {
+                [&__table](_ReduceType __val1, _ReduceType __val2) -> _ReduceType
+                {
                     if (__val1 == __broken || __val2 == __broken)
                     {
                         return __broken;
@@ -1857,43 +1921,45 @@ __brick_partition(_ForwardIterator __first, _ForwardIterator __last, _UnaryPredi
     return std::partition(__first, __last, __pred);
 }
 
-template <class _ForwardIterator, class _UnaryPredicate>
-_ForwardIterator
-__brick_partition(_ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
+template <class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__brick_partition(_RandomAccessIterator __first, _RandomAccessIterator __last, _UnaryPredicate __pred,
                   /*is_vector=*/std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::partition(__first, __last, __pred);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 _ForwardIterator
-__pattern_partition(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
-                    _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+__pattern_partition(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                    _UnaryPredicate __pred) noexcept
 {
-    return __internal::__brick_partition(__first, __last, __pred, __is_vector);
+    return __internal::__brick_partition(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__pattern_partition(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                    _RandomAccessIterator __last, _UnaryPredicate __pred)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
 
     // 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;
+        _RandomAccessIterator __begin;
+        _RandomAccessIterator __pivot;
+        _RandomAccessIterator __end;
     };
 
     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 __reductor = [&__exec](_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);
@@ -1908,10 +1974,10 @@ __pattern_partition(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Forwar
             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) {
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __val1.__pivot, __val1.__pivot + __size1,
+                    [__val1, __val2, __size1](_RandomAccessIterator __i, _RandomAccessIterator __j) {
                         __internal::__brick_swap_ranges(__i, __j, (__val2.__pivot - __size1) + (__i - __val1.__pivot),
-                                                        __is_vector);
+                                                        _IsVector{});
                     });
                 return {__new_begin, __val2.__pivot - __size1, __val2.__end};
             }
@@ -1919,20 +1985,21 @@ __pattern_partition(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Forwar
             else
             {
                 __par_backend::__parallel_for(
-                    std::forward<_ExecutionPolicy>(__exec), __val1.__pivot, __val1.__pivot + __size2,
-                    [__val1, __val2, __is_vector](_ForwardIterator __i, _ForwardIterator __j) {
-                        __internal::__brick_swap_ranges(__i, __j, __val2.__begin + (__i - __val1.__pivot), __is_vector);
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __val1.__pivot, __val1.__pivot + __size2,
+                    [__val1, __val2](_RandomAccessIterator __i, _RandomAccessIterator __j) {
+                        __internal::__brick_swap_ranges(__i, __j, __val2.__begin + (__i - __val1.__pivot), _IsVector{});
                     });
                 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 {
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last, __init,
+            [__pred, __reductor](_RandomAccessIterator __i, _RandomAccessIterator __j,
+                                 _PartitionRange __value) -> _PartitionRange
+            {
                 //1. serial partition
-                _ForwardIterator __pivot = __internal::__brick_partition(__i, __j, __pred, __is_vector);
+                _RandomAccessIterator __pivot = __internal::__brick_partition(__i, __j, __pred, _IsVector{});
 
                 // 2. merging of two ranges (left and right respectively)
                 return __reductor(__value, {__i, __pivot, __j});
@@ -1954,44 +2021,45 @@ __brick_stable_partition(_BidirectionalIterator __first, _BidirectionalIterator
     return std::stable_partition(__first, __last, __pred);
 }
 
-template <class _BidirectionalIterator, class _UnaryPredicate>
-_BidirectionalIterator
-__brick_stable_partition(_BidirectionalIterator __first, _BidirectionalIterator __last, _UnaryPredicate __pred,
+template <class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__brick_stable_partition(_RandomAccessIterator __first, _RandomAccessIterator __last, _UnaryPredicate __pred,
                          /*__is_vector=*/std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::stable_partition(__first, __last, __pred);
 }
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate>
 _BidirectionalIterator
-__pattern_stable_partition(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
-                           _UnaryPredicate __pred, _IsVector __is_vector,
-                           /*is_parallelization=*/std::false_type) noexcept
+__pattern_stable_partition(_Tag, _ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __last,
+                           _UnaryPredicate __pred) noexcept
 {
-    return __internal::__brick_stable_partition(__first, __last, __pred, __is_vector);
+    return __internal::__brick_stable_partition(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
-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
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__pattern_stable_partition(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                           _RandomAccessIterator __last, _UnaryPredicate __pred) noexcept
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     // 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;
+        _RandomAccessIterator __begin;
+        _RandomAccessIterator __pivot;
+        _RandomAccessIterator __end;
     };
 
     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 __reductor = [](_PartitionRange __val1, _PartitionRange __val2) -> _PartitionRange
+        {
             auto __size1 = __val1.__end - __val1.__pivot;
             auto __new_begin = __val2.__begin - (__val1.__end - __val1.__begin);
 
@@ -2004,17 +2072,18 @@ __pattern_stable_partition(_ExecutionPolicy&& __exec, _BidirectionalIterator __f
             // then we should swap the false part of left range and last part of true part of right range
             else
             {
-                __internal::__brick_rotate(__val1.__pivot, __val2.__begin, __val2.__pivot, __is_vector);
+                __internal::__brick_rotate(__val1.__pivot, __val2.__begin, __val2.__pivot, _IsVector{});
                 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 {
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last, __init,
+            [&__pred, __reductor](_RandomAccessIterator __i, _RandomAccessIterator __j,
+                                  _PartitionRange __value) -> _PartitionRange
+            {
                 //1. serial stable_partition
-                _BidirectionalIterator __pivot = __internal::__brick_stable_partition(__i, __j, __pred, __is_vector);
+                _RandomAccessIterator __pivot = __internal::__brick_stable_partition(__i, __j, __pred, _IsVector{});
 
                 // 2. merging of two ranges (left and right respectively)
                 return __reductor(__value, {__i, __pivot, __j});
@@ -2036,170 +2105,190 @@ __brick_partition_copy(_ForwardIterator __first, _ForwardIterator __last, _Outpu
     return std::partition_copy(__first, __last, __out_true, __out_false, __pred);
 }
 
-template <class _ForwardIterator, class _OutputIterator1, class _OutputIterator2, class _UnaryPredicate>
-std::pair<_OutputIterator1, _OutputIterator2>
-__brick_partition_copy(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator1 __out_true,
-                       _OutputIterator2 __out_false, _UnaryPredicate __pred, /*is_vector=*/std::true_type) noexcept
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _RandomAccessIterator3,
+          class _UnaryPredicate>
+std::pair<_RandomAccessIterator2, _RandomAccessIterator3>
+__brick_partition_copy(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last, _RandomAccessIterator2 __out_true,
+                       _RandomAccessIterator3 __out_false, _UnaryPredicate __pred,
+                       /*is_vector=*/std::true_type) noexcept
 {
-#if (_PSTL_MONOTONIC_PRESENT)
+#if defined(_PSTL_MONOTONIC_PRESENT)
     return __unseq_backend::__simd_partition_copy(__first, __last - __first, __out_true, __out_false, __pred);
 #else
     return std::partition_copy(__first, __last, __out_true, __out_false, __pred);
 #endif
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator1, class _OutputIterator2,
-          class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator1, class _OutputIterator2,
+          class _UnaryPredicate>
 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
+__pattern_partition_copy(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                         _OutputIterator1 __out_true, _OutputIterator2 __out_false, _UnaryPredicate __pred) noexcept
 {
-    return __internal::__brick_partition_copy(__first, __last, __out_true, __out_false, __pred, __is_vector);
+    return __internal::__brick_partition_copy(__first, __last, __out_true, __out_false, __pred,
+                                              typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _RandomAccessIterator3, class _UnaryPredicate>
+std::pair<_RandomAccessIterator2, _RandomAccessIterator3>
+__pattern_partition_copy(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                         _RandomAccessIterator1 __last, _RandomAccessIterator2 __out_true,
+                         _RandomAccessIterator3 __out_false, _UnaryPredicate __pred)
 {
-    typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::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 __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 __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
-                    __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);
-        });
+        return __internal::__except_handler(
+            [&__exec, __n, __first, __out_true, __out_false, __pred, &__mask_buf]()
+            {
+                bool* __mask = __mask_buf.get();
+                _ReturnType __m{};
+                __par_backend::__parallel_strict_scan(
+                    __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __n,
+                    std::make_pair(_DifferenceType(0), _DifferenceType(0)),
+                    [=](_DifferenceType __i, _DifferenceType __len) { // Reduce
+                        return __internal::__brick_calc_mask_1<_DifferenceType>(__first + __i, __first + (__i + __len),
+                                                                                __mask + __i, __pred, _IsVector{});
+                    },
+                    [](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
+                        __internal::__brick_partition_by_mask(
+                            __first + __i, __first + (__i + __len), __out_true + __initial.first,
+                            __out_false + __initial.second, __mask + __i, _IsVector{});
+                    },
+                    [&__m](_ReturnType __total) { __m = __total; });
+                return std::make_pair(__out_true + __m.first, __out_false + __m.second);
+            });
     }
     // trivial sequence - use serial algorithm
-    return __internal::__brick_partition_copy(__first, __last, __out_true, __out_false, __pred, __is_vector);
+    return __internal::__brick_partition_copy(__first, __last, __out_true, __out_false, __pred, _IsVector{});
 }
 
 //------------------------------------------------------------------------
 // sort
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector,
-          class _IsMoveConstructible>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsMoveConstructible>
 void
-__pattern_sort(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
-               _IsVector /*is_vector*/, /*is_parallel=*/std::false_type, _IsMoveConstructible) noexcept
+__pattern_sort(_Tag, _ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
+               _IsMoveConstructible) noexcept
 {
     std::sort(__first, __last, __comp);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 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)
+__pattern_sort(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+               _RandomAccessIterator __last, _Compare __comp, /*is_move_constructible=*/std::true_type)
 {
-    __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); });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_stable_sort(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
+                [](_RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp)
+                { std::sort(__first, __last, __comp); });
+        });
 }
 
 //------------------------------------------------------------------------
 // stable_sort
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_stable_sort(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
-                      _IsVector /*is_vector*/, /*is_parallel=*/std::false_type) noexcept
+__pattern_stable_sort(_Tag, _ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last,
+                      _Compare __comp) noexcept
 {
     std::stable_sort(__first, __last, __comp);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_stable_sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
-                      _Compare __comp, _IsVector /*is_vector*/, /*is_parallel=*/std::true_type)
+__pattern_stable_sort(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                      _RandomAccessIterator __last, _Compare __comp)
 {
-    __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); });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_stable_sort(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
+                [](_RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp)
+                { std::stable_sort(__first, __last, __comp); });
+        });
 }
 
 //------------------------------------------------------------------------
 // partial_sort
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_partial_sort(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __middle,
-                       _RandomAccessIterator __last, _Compare __comp, _IsVector,
-                       /*is_parallel=*/std::false_type) noexcept
+__pattern_partial_sort(_Tag, _ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __middle,
+                       _RandomAccessIterator __last, _Compare __comp) noexcept
 {
     std::partial_sort(__first, __middle, __last, __comp);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_partial_sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __middle,
-                       _RandomAccessIterator __last, _Compare __comp, _IsVector, /*is_parallel=*/std::true_type)
+__pattern_partial_sort(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                       _RandomAccessIterator __middle, _RandomAccessIterator __last, _Compare __comp)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     const auto __n = __middle - __first;
     if (__n == 0)
         return;
 
-    __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);
-            },
-            __n);
-    });
+    __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_stable_sort(
+                __backend_tag{}, 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);
+                },
+                __n);
+        });
 }
 
 //------------------------------------------------------------------------
 // partial_sort_copy
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _RandomAccessIterator, class _Compare>
 _RandomAccessIterator
-__pattern_partial_sort_copy(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
-                            _RandomAccessIterator __d_first, _RandomAccessIterator __d_last, _Compare __comp, _IsVector,
-                            /*is_parallel=*/std::false_type) noexcept
+__pattern_partial_sort_copy(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                            _RandomAccessIterator __d_first, _RandomAccessIterator __d_last, _Compare __comp) noexcept
 {
     return std::partial_sort_copy(__first, __last, __d_first, __d_last, __comp);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _RandomAccessIterator, class _Compare, class _IsVector>
-_RandomAccessIterator
-__pattern_partial_sort_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
-                            _RandomAccessIterator __d_first, _RandomAccessIterator __d_last, _Compare __comp,
-                            _IsVector __is_vector, /*is_parallel=*/std::true_type)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Compare>
+_RandomAccessIterator2
+__pattern_partial_sort_copy(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first,
+                            _RandomAccessIterator1 __last, _RandomAccessIterator2 __d_first,
+                            _RandomAccessIterator2 __d_last, _Compare __comp)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__last == __first || __d_last == __d_first)
     {
         return __d_first;
@@ -2210,18 +2299,18 @@ __pattern_partial_sort_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first,
         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);
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __d_first, __d_first + __n1, __comp,
+                [__first, __d_first](_RandomAccessIterator2 __i, _RandomAccessIterator2 __j, _Compare __comp)
+                {
+                    _RandomAccessIterator1 __i1 = __first + (__i - __d_first);
+                    _RandomAccessIterator1 __j1 = __first + (__j - __d_first);
 
                 // 1. Copy elements from input to output
-#    if !_PSTL_ICC_18_OMP_SIMD_BROKEN
-                    __internal::__brick_copy(__i1, __j1, __i, __is_vector);
-#    else
+#if !defined(_PSTL_ICC_18_OMP_SIMD_BROKEN)
+                    __internal::__brick_copy(__i1, __j1, __i, _IsVector{});
+#else
                     std::copy(__i1, __j1, __i);
-#    endif
+#endif
                     // 2. Sort elements in output sequence
                     std::sort(__i, __j, __comp);
                 },
@@ -2230,38 +2319,38 @@ __pattern_partial_sort_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first,
         }
         else
         {
-            typedef typename std::iterator_traits<_ForwardIterator>::value_type _T1;
-            typedef typename std::iterator_traits<_RandomAccessIterator>::value_type _T2;
+            typedef typename std::iterator_traits<_RandomAccessIterator1>::value_type _T1;
+            typedef typename std::iterator_traits<_RandomAccessIterator2>::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,
-                                                  [__n2, __first, __r](_T1* __i, _T1* __j, _Compare __comp) {
-                                                      _ForwardIterator __it = __first + (__i - __r);
+            __par_backend::__parallel_stable_sort(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __r, __r + __n1, __comp,
+                [__n2, __first, __r](_T1* __i, _T1* __j, _Compare __comp)
+                {
+                    _RandomAccessIterator1 __it = __first + (__i - __r);
 
-                                                      // 1. Copy elements from input to raw memory
-                                                      for (_T1* __k = __i; __k != __j; ++__k, ++__it)
-                                                      {
-                                                          ::new (__k) _T2(*__it);
-                                                      }
+                    // 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);
+                    // 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_destroy()(
-                                                  __i, __j, __d_first + (__i - __r), __is_vector);
-                                          });
-            __par_backend::__parallel_for(
-                std::forward<_ExecutionPolicy>(__exec), __r + __n2, __r + __n1,
-                [__is_vector](_T1* __i, _T1* __j) { __brick_destroy(__i, __j, __is_vector); });
+            __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __r, __r + __n2,
+                                          [__r, __d_first](_T1* __i, _T1* __j)
+                                          { __brick_move_destroy()(__i, __j, __d_first + (__i - __r), _IsVector{}); });
+            __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __r + __n2,
+                                          __r + __n1,
+                                          [](_T1* __i, _T1* __j) { __brick_destroy(__i, __j, _IsVector{}); });
 
             return __d_first + __n2;
         }
@@ -2271,9 +2360,9 @@ __pattern_partial_sort_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first,
 //------------------------------------------------------------------------
 // adjacent_find
 //------------------------------------------------------------------------
-template <class _ForwardIterator, class _BinaryPredicate>
-_ForwardIterator
-__brick_adjacent_find(_ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
+template <class _RandomAccessIterator, class _BinaryPredicate>
+_RandomAccessIterator
+__brick_adjacent_find(_RandomAccessIterator __first, _RandomAccessIterator __last, _BinaryPredicate __pred,
                       /* IsVector = */ std::true_type, bool __or_semantic) noexcept
 {
     return __unseq_backend::__simd_adjacent_find(__first, __last, __pred, __or_semantic);
@@ -2287,78 +2376,79 @@ __brick_adjacent_find(_ForwardIterator __first, _ForwardIterator __last, _Binary
     return std::adjacent_find(__first, __last, __pred);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate>
 _ForwardIterator
-__pattern_adjacent_find(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred,
-                        /* is_parallel */ std::false_type, _IsVector __is_vector, bool __or_semantic) noexcept
+__pattern_adjacent_find(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                        _BinaryPredicate __pred, bool __or_semantic) noexcept
 {
-    return __internal::__brick_adjacent_find(__first, __last, __pred, __is_vector, __or_semantic);
+    return __internal::__brick_adjacent_find(__first, __last, __pred, typename _Tag::__is_vector{}, __or_semantic);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _BinaryPredicate, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _BinaryPredicate>
 _RandomAccessIterator
-__pattern_adjacent_find(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
-                        _BinaryPredicate __pred, /* is_parallel */ std::true_type, _IsVector __is_vector,
-                        bool __or_semantic)
+__pattern_adjacent_find(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                        _RandomAccessIterator __last, _BinaryPredicate __pred, bool __or_semantic)
 {
     if (__last - __first < 2)
         return __last;
 
-    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;
-                }
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
 
-                if (__value > __begin)
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __par_backend::__parallel_reduce(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last, __last,
+                [__last, __pred, __or_semantic](_RandomAccessIterator __begin, _RandomAccessIterator __end,
+                                                _RandomAccessIterator __value) -> _RandomAccessIterator
                 {
-                    // 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 =
-                        __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
-        );
-    });
+                    // 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 =
+                            __internal::__brick_adjacent_find(__begin, __end, __pred, _IsVector{}, __or_semantic);
+                        if (__res < __end)
+                            __value = __res;
+                    }
+                    return __value;
+                },
+                [](_RandomAccessIterator __x, _RandomAccessIterator __y) -> _RandomAccessIterator
+                { return __x < __y ? __x : __y; } //reduce a __value
+            );
+        });
 }
 
 //------------------------------------------------------------------------
 // nth_element
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_nth_element(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __nth,
-                      _RandomAccessIterator __last, _Compare __comp, _IsVector,
-                      /*is_parallel=*/std::false_type) noexcept
+__pattern_nth_element(_Tag, _ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __nth,
+                      _RandomAccessIterator __last, _Compare __comp) noexcept
 {
     std::nth_element(__first, __nth, __last, __comp);
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_nth_element(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __nth,
-                      _RandomAccessIterator __last, _Compare __comp, _IsVector __is_vector,
-                      /*is_parallel=*/std::true_type) noexcept
+__pattern_nth_element(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                      _RandomAccessIterator __nth, _RandomAccessIterator __last, _Compare __comp) noexcept
 {
     if (__first == __last || __nth == __last)
     {
@@ -2370,10 +2460,8 @@ __pattern_nth_element(_ExecutionPolicy&& __exec, _RandomAccessIterator __first,
     _RandomAccessIterator __x;
     do
     {
-        __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 = __internal::__pattern_partition(__tag, std::forward<_ExecutionPolicy>(__exec), __first + 1, __last,
+                                              [&__comp, __first](const _Tp& __x) { return __comp(__x, *__first); });
         --__x;
         if (__x != __first)
         {
@@ -2404,9 +2492,9 @@ __pattern_nth_element(_ExecutionPolicy&& __exec, _RandomAccessIterator __first,
 //------------------------------------------------------------------------
 // fill, fill_n
 //------------------------------------------------------------------------
-template <class _ForwardIterator, class _Tp>
+template <class _RandomAccessIterator, class _Tp>
 void
-__brick_fill(_ForwardIterator __first, _ForwardIterator __last, const _Tp& __value,
+__brick_fill(_RandomAccessIterator __first, _RandomAccessIterator __last, const _Tp& __value,
              /* __is_vector = */ std::true_type) noexcept
 {
     __unseq_backend::__simd_fill_n(__first, __last - __first, __value);
@@ -2420,31 +2508,34 @@ __brick_fill(_ForwardIterator __first, _ForwardIterator __last, const _Tp& __val
     std::fill(__first, __last, __value);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Tp>
 void
-__pattern_fill(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value,
-               /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
+__pattern_fill(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value) noexcept
 {
-    __internal::__brick_fill(__first, __last, __value, __is_vector);
+    __internal::__brick_fill(__first, __last, __value, typename _Tag::__is_vector{});
 }
 
-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 __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) {
-                                          __internal::__brick_fill(__begin, __end, __value, __is_vector);
-                                      });
-        return __last;
-    });
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Tp>
+_RandomAccessIterator
+__pattern_fill(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+               _RandomAccessIterator __last, const _Tp& __value)
+{
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&__exec, __first, __last, &__value]()
+        {
+            __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                          [&__value](_RandomAccessIterator __begin, _RandomAccessIterator __end)
+                                          { __internal::__brick_fill(__begin, __end, __value, _IsVector{}); });
+            return __last;
+        });
 }
 
-template <class _OutputIterator, class _Size, class _Tp>
-_OutputIterator
-__brick_fill_n(_OutputIterator __first, _Size __count, const _Tp& __value, /* __is_vector = */ std::true_type) noexcept
+template <class _RandomAccessIterator, class _Size, class _Tp>
+_RandomAccessIterator
+__brick_fill_n(_RandomAccessIterator __first, _Size __count, const _Tp& __value,
+               /* __is_vector = */ std::true_type) noexcept
 {
     return __unseq_backend::__simd_fill_n(__first, __count, __value);
 }
@@ -2456,21 +2547,20 @@ __brick_fill_n(_OutputIterator __first, _Size __count, const _Tp& __value, /* __
     return std::fill_n(__first, __count, __value);
 }
 
-template <class _ExecutionPolicy, class _OutputIterator, class _Size, class _Tp, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _OutputIterator, class _Size, class _Tp>
 _OutputIterator
-__pattern_fill_n(_ExecutionPolicy&&, _OutputIterator __first, _Size __count, const _Tp& __value,
-                 /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
+__pattern_fill_n(_Tag, _ExecutionPolicy&&, _OutputIterator __first, _Size __count, const _Tp& __value) noexcept
 {
-    return __internal::__brick_fill_n(__first, __count, __value, __is_vector);
+    return __internal::__brick_fill_n(__first, __count, __value, typename _Tag::__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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Tp>
+_RandomAccessIterator
+__pattern_fill_n(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                 _Size __count, const _Tp& __value)
 {
-    return __internal::__pattern_fill(std::forward<_ExecutionPolicy>(__exec), __first, __first + __count, __value,
-                                      std::true_type(), __is_vector);
+    return __internal::__pattern_fill(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __first + __count,
+                                      __value);
 }
 
 //------------------------------------------------------------------------
@@ -2492,59 +2582,61 @@ __brick_generate(_ForwardIterator __first, _ForwardIterator __last, _Generator _
     std::generate(__first, __last, __g);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Generator, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Generator>
 void
-__pattern_generate(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Generator __g,
-                   /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
+__pattern_generate(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Generator __g) noexcept
 {
-    __internal::__brick_generate(__first, __last, __g, __is_vector);
+    __internal::__brick_generate(__first, __last, __g, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Generator>
+_RandomAccessIterator
+__pattern_generate(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                   _RandomAccessIterator __last, _Generator __g)
 {
-    return __internal::__except_handler([&]() {
-        __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                      [__g, __is_vector](_ForwardIterator __begin, _ForwardIterator __end) {
-                                          __internal::__brick_generate(__begin, __end, __g, __is_vector);
-                                      });
-        return __last;
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                          [__g](_RandomAccessIterator __begin, _RandomAccessIterator __end)
+                                          { __internal::__brick_generate(__begin, __end, __g, _IsVector{}); });
+            return __last;
+        });
 }
 
-template <class _OutputIterator, class _Size, class _Generator>
-  _OutputIterator
-__brick_generate_n(_OutputIterator __first, _Size __count, _Generator __g, /* is_vector = */ std::true_type) noexcept
+template <class _RandomAccessIterator, class _Size, class _Generator>
+_RandomAccessIterator
+__brick_generate_n(_RandomAccessIterator __first, _Size __count, _Generator __g,
+                   /* is_vector = */ std::true_type) noexcept
 {
     return __unseq_backend::__simd_generate_n(__first, __count, __g);
 }
 
 template <class _OutputIterator, class _Size, class _Generator>
-  _OutputIterator
+_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>
+template <class _Tag, class _ExecutionPolicy, class _OutputIterator, class _Size, class _Generator>
 _OutputIterator
-__pattern_generate_n(_ExecutionPolicy&&, _OutputIterator __first, _Size __count, _Generator __g,
-                     /*is_parallel=*/std::false_type, _IsVector __is_vector) noexcept
+__pattern_generate_n(_Tag, _ExecutionPolicy&&, _OutputIterator __first, _Size __count, _Generator __g) noexcept
 {
-    return __internal::__brick_generate_n(__first, __count, __g, __is_vector);
+    return __internal::__brick_generate_n(__first, __count, __g, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Size, class _Generator>
+_RandomAccessIterator
+__pattern_generate_n(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                     _Size __count, _Generator __g)
 {
-    static_assert(__is_random_access_iterator<_OutputIterator>::value,
+    static_assert(__are_random_access_iterators<_RandomAccessIterator>::value,
                   "Pattern-brick error. Should be a random access iterator.");
-    return __internal::__pattern_generate(std::forward<_ExecutionPolicy>(__exec), __first, __first + __count, __g,
-                                          std::true_type(), __is_vector);
+    return __internal::__pattern_generate(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __first + __count,
+                                          __g);
 }
 
 //------------------------------------------------------------------------
@@ -2564,41 +2656,41 @@ _RandomAccessIterator
 __brick_remove_if(_RandomAccessIterator __first, _RandomAccessIterator __last, _UnaryPredicate __pred,
                   /* __is_vector = */ std::true_type) noexcept
 {
-#if _PSTL_MONOTONIC_PRESENT
+#if defined(_PSTL_MONOTONIC_PRESENT)
     return __unseq_backend::__simd_remove_if(__first, __last - __first, __pred);
 #else
     return std::remove_if(__first, __last, __pred);
 #endif
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 _ForwardIterator
-__pattern_remove_if(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred,
-                    _IsVector __is_vector, /*is_parallel*/ std::false_type) noexcept
+__pattern_remove_if(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                    _UnaryPredicate __pred) noexcept
 {
-    return __internal::__brick_remove_if(__first, __last, __pred, __is_vector);
+    return __internal::__brick_remove_if(__first, __last, __pred, typename _Tag::__is_vector{});
 }
 
-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
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _UnaryPredicate>
+_RandomAccessIterator
+__pattern_remove_if(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                    _RandomAccessIterator __last, _UnaryPredicate __pred) noexcept
 {
-    typedef typename std::iterator_traits<_ForwardIterator>::reference _ReferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator>::reference _ReferenceType;
 
     if (__first == __last || __first + 1 == __last)
     {
         // Trivial sequence - use serial algorithm
-        return __internal::__brick_remove_if(__first, __last, __pred, __is_vector);
+        return __internal::__brick_remove_if(__first, __last, __pred, _IsVector{});
     }
 
     return __internal::__remove_elements(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last,
-        [&__pred, __is_vector](bool* __b, bool* __e, _ForwardIterator __it) {
-            __internal::__brick_walk2(__b, __e, __it, [&__pred](bool& __x, _ReferenceType __y) { __x = !__pred(__y); },
-                                      __is_vector);
-        },
-        __is_vector);
+        __tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+        [&__pred](bool* __b, bool* __e, _RandomAccessIterator __it)
+        {
+            __internal::__brick_walk2(
+                __b, __e, __it, [&__pred](bool& __x, _ReferenceType __y) { __x = !__pred(__y); }, _IsVector{});
+        });
 }
 
 //------------------------------------------------------------------------
@@ -2614,39 +2706,42 @@ __brick_merge(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIte
     return std::merge(__first1, __last1, __first2, __last2, __d_first, __comp);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator
-__brick_merge(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-              _ForwardIterator2 __last2, _OutputIterator __d_first, _Compare __comp,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__brick_merge(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+              _RandomAccessIterator2 __last2, _RandomAccessIterator3 __d_first, _Compare __comp,
               /* __is_vector = */ std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::merge(__first1, __last1, __first2, __last2, __d_first, __comp);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _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
+__pattern_merge(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __d_first,
+                _Compare __comp) noexcept
 {
-    return __internal::__brick_merge(__first1, __last1, __first2, __last2, __d_first, __comp, __is_vector);
+    return __internal::__brick_merge(__first1, __last1, __first2, __last2, __d_first, __comp,
+                                     typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__pattern_merge(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                _RandomAccessIterator3 __d_first, _Compare __comp)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     __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 __internal::__brick_merge(__f1, __l1, __f2, __l2, __f3, __comp, __is_vector);
-        });
+        __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __d_first,
+        __comp,
+        [](_RandomAccessIterator1 __f1, _RandomAccessIterator1 __l1, _RandomAccessIterator2 __f2,
+           _RandomAccessIterator2 __l2, _RandomAccessIterator3 __f3, _Compare __comp)
+        { return __internal::__brick_merge(__f1, __l1, __f2, __l2, __f3, __comp, _IsVector{}); });
     return __d_first + (__last1 - __first1) + (__last2 - __first2);
 }
 
@@ -2661,83 +2756,89 @@ __brick_inplace_merge(_BidirectionalIterator __first, _BidirectionalIterator __m
     std::inplace_merge(__first, __middle, __last, __comp);
 }
 
-template <class _BidirectionalIterator, class _Compare>
+template <class _RandomAccessIterator, class _Compare>
 void
-__brick_inplace_merge(_BidirectionalIterator __first, _BidirectionalIterator __middle, _BidirectionalIterator __last,
+__brick_inplace_merge(_RandomAccessIterator __first, _RandomAccessIterator __middle, _RandomAccessIterator __last,
                       _Compare __comp, /* __is_vector = */ std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial")
     std::inplace_merge(__first, __middle, __last, __comp);
 }
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _BidirectionalIterator, class _Compare>
 void
-__pattern_inplace_merge(_ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __middle,
-                        _BidirectionalIterator __last, _Compare __comp, _IsVector __is_vector,
-                        /* is_parallel = */ std::false_type) noexcept
+__pattern_inplace_merge(_Tag, _ExecutionPolicy&&, _BidirectionalIterator __first, _BidirectionalIterator __middle,
+                        _BidirectionalIterator __last, _Compare __comp) noexcept
 {
-    __internal::__brick_inplace_merge(__first, __middle, __last, __comp, __is_vector);
+    __internal::__brick_inplace_merge(__first, __middle, __last, __comp, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _BidirectionalIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 void
-__pattern_inplace_merge(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __middle,
-                        _BidirectionalIterator __last, _Compare __comp, _IsVector __is_vector,
-                        /*is_parallel=*/std::true_type)
+__pattern_inplace_merge(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                        _RandomAccessIterator __middle, _RandomAccessIterator __last, _Compare __comp)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__first == __last || __first == __middle || __middle == __last)
     {
         return;
     }
-    typedef typename std::iterator_traits<_BidirectionalIterator>::value_type _Tp;
+    typedef typename std::iterator_traits<_RandomAccessIterator>::value_type _Tp;
     auto __n = __last - __first;
     __par_backend::__buffer<_Tp> __buf(__n);
     _Tp* __r = __buf.get();
-    __internal::__except_handler([&]() {
-        auto __move_values = [](_BidirectionalIterator __x, _Tp* __z) {
-            __internal::__invoke_if_else(std::is_trivial<_Tp>(), [&]() { *__z = std::move(*__x); },
-                                         [&]() { ::new (std::addressof(*__z)) _Tp(std::move(*__x)); });
-        };
+    __internal::__except_handler(
+        [&]()
+        {
+            auto __move_values = [](_RandomAccessIterator __x, _Tp* __z)
+            {
+                __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 __internal::__brick_uninitialized_move(__first1, __last1, __first2, _IsVector());
-        };
+            auto __move_sequences = [](_RandomAccessIterator __first1, _RandomAccessIterator __last1, _Tp* __first2)
+            { 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) {
-                (__utils::__serial_move_merge(__n))(__f1, __l1, __f2, __l2, __f3, __comp, __move_values, __move_values,
-                                                    __move_sequences, __move_sequences);
-                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_destroy()(__i, __j, __first + (__i - __r), __is_vector);
-            });
-    });
+            __par_backend::__parallel_merge(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __middle, __middle, __last, __r,
+                __comp,
+                [__n, __move_values, __move_sequences](_RandomAccessIterator __f1, _RandomAccessIterator __l1,
+                                                       _RandomAccessIterator __f2, _RandomAccessIterator __l2,
+                                                       _Tp* __f3, _Compare __comp)
+                {
+                    (__utils::__serial_move_merge(__n))(__f1, __l1, __f2, __l2, __f3, __comp, __move_values,
+                                                        __move_values, __move_sequences, __move_sequences);
+                    return __f3 + (__l1 - __f1) + (__l2 - __f2);
+                });
+            __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __r, __r + __n,
+                                          [__r, __first](_Tp* __i, _Tp* __j)
+                                          { __brick_move_destroy()(__i, __j, __first + (__i - __r), _IsVector{}); });
+        });
 }
 
 //------------------------------------------------------------------------
 // includes
 //------------------------------------------------------------------------
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare>
 bool
-__pattern_includes(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
-                   _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp, _IsVector,
-                   /*is_parallel=*/std::false_type) noexcept
+__pattern_includes(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                   _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp) noexcept
 {
     return std::includes(__first1, __last1, __first2, __last2, __comp);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Compare>
 bool
-__pattern_includes(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
-                   _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp, _IsVector,
-                   /*is_parallel=*/std::true_type)
+__pattern_includes(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                   _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                   _Compare __comp)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__first2 >= __last2)
         return true;
 
@@ -2751,51 +2852,55 @@ __pattern_includes(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _Forwa
     if (__last2 - __first2 == 1)
         return !__comp(*__first1, *__first2) && !__comp(*__first2, *__first1);
 
-    return __internal::__except_handler([&]() {
-        return !__internal::__parallel_or(
-            std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
-            [__first1, __last1, __first2, __last2, &__comp](_ForwardIterator2 __i, _ForwardIterator2 __j) {
-                _PSTL_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);
-                };
-
-                //1.1 left bound, case "aaa[aaaxyz...]" - searching "x"
-                if (__i > __first2 && __is_equal(__i, __i - 1))
+    return __internal::__except_handler(
+        [&]()
+        {
+            return !__internal::__parallel_or(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
+                [__first1, __last1, __first2, __last2, &__comp](_RandomAccessIterator2 __i, _RandomAccessIterator2 __j)
                 {
-                    //whole subrange continues to content equal elements - return "no op"
-                    if (__is_equal(__i, __j - 1))
-                        return false;
+                    _PSTL_ASSERT(__j > __i);
+                    //_PSTL_ASSERT(__j - __i > 1);
 
-                    __i = std::upper_bound(__i, __last2, *__i, __comp);
-                }
+                    //1. moving boundaries to "consume" subsequence of equal elements
+                    auto __is_equal = [&__comp](_RandomAccessIterator2 __a, _RandomAccessIterator2 __b) -> bool
+                    { return !__comp(*__a, *__b) && !__comp(*__b, *__a); };
+
+                    //1.1 left bound, case "aaa[aaaxyz...]" - searching "x"
+                    if (__i > __first2 && __is_equal(__i, __i - 1))
+                    {
+                        //whole subrange continues to content equal elements - return "no op"
+                        if (__is_equal(__i, __j - 1))
+                            return false;
 
-                //1.2 right bound, case "[...aaa]aaaxyz" - searching "x"
-                if (__j < __last2 && __is_equal(__j - 1, __j))
-                    __j = std::upper_bound(__j, __last2, *__j, __comp);
+                        __i = std::upper_bound(__i, __last2, *__i, __comp);
+                    }
 
-                //2. testing is __a subsequence of the second range included into the first range
-                auto __b = std::lower_bound(__first1, __last1, *__i, __comp);
+                    //1.2 right bound, case "[...aaa]aaaxyz" - searching "x"
+                    if (__j < __last2 && __is_equal(__j - 1, __j))
+                        __j = std::upper_bound(__j, __last2, *__j, __comp);
 
-                _PSTL_ASSERT(!__comp(*(__last1 - 1), *__b));
-                _PSTL_ASSERT(!__comp(*(__j - 1), *__i));
-                return !std::includes(__b, __last1, __i, __j, __comp);
-            });
-    });
+                    //2. testing is __a subsequence of the second range included into the first range
+                    auto __b = std::lower_bound(__first1, __last1, *__i, __comp);
+
+                    _PSTL_ASSERT(!__comp(*(__last1 - 1), *__b));
+                    _PSTL_ASSERT(!__comp(*(__j - 1), *__i));
+                    return !std::includes(__b, __last1, __i, __j, __comp);
+                });
+        });
 }
 
 constexpr auto __set_algo_cut_off = 1000;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector, class _SizeFunction, class _SetOP>
+template <class _IsVector, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2,
+          class _OutputIterator, class _Compare, class _SizeFunction, class _SetOP>
 _OutputIterator
-__parallel_set_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
-                  _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result, _Compare __comp,
-                  _SizeFunction __size_func, _SetOP __set_op, _IsVector __is_vector)
+__parallel_set_op(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _ForwardIterator1 __first1,
+                  _ForwardIterator1 __last1, _ForwardIterator2 __first2, _ForwardIterator2 __last2,
+                  _OutputIterator __result, _Compare __comp, _SizeFunction __size_func, _SetOP __set_op)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
     typedef typename std::iterator_traits<_OutputIterator>::value_type _Tp;
 
@@ -2814,100 +2919,100 @@ __parallel_set_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _Forwar
 
     __par_backend::__buffer<_Tp> __buf(__size_func(__n1, __n2));
 
-    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_destroy()(__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);
+    return __internal::__except_handler(
+        [&__exec, __n1, __first1, __last1, __first2, __last2, __result, __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_destroy()(__buffer + __s.__buf_pos, __buffer + (__s.__buf_pos + __s.__len),
+                                           __result + __s.__pos, _IsVector{});
+            };
+            __par_backend::__parallel_strict_scan(
+                __backend_tag{}, 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
-                if (__b != __first1)
-                    __b = std::upper_bound(__b, __last1, *__b, __comp);
+                    //try searching for the first element which not equal to *__b
+                    if (__b != __first1)
+                        __b = std::upper_bound(__b, __last1, *__b, __comp);
 
-                //try searching for the first element which not equal to *__e
-                if (__e != __last1)
-                    __e = std::upper_bound(__e, __last1, *__e, __comp);
+                    //try searching for the first element which not equal to *__e
+                    if (__e != __last1)
+                        __e = std::upper_bound(__e, __last1, *__e, __comp);
 
-                //check is [__b; __e) empty
-                if (__e - __b < 1)
-                {
-                    _ForwardIterator2 __bb = __last2;
-                    if (__b != __last1)
-                        __bb = std::lower_bound(__first2, __last2, *__b, __comp);
+                    //check is [__b; __e) empty
+                    if (__e - __b < 1)
+                    {
+                        _ForwardIterator2 __bb = __last2;
+                        if (__b != __last1)
+                            __bb = std::lower_bound(__first2, __last2, *__b, __comp);
 
-                    const _DifferenceType __buf_pos = __size_func((__b - __first1), (__bb - __first2));
-                    return _SetRange{0, 0, __buf_pos};
-                }
+                        const _DifferenceType __buf_pos = __size_func((__b - __first1), (__bb - __first2));
+                        return _SetRange{0, 0, __buf_pos};
+                    }
 
-                //try searching for "corresponding" subrange [__bb; __ee) in the second sequence
-                _ForwardIterator2 __bb = __first2;
-                if (__b != __first1)
-                    __bb = std::lower_bound(__first2, __last2, *__b, __comp);
+                    //try searching for "corresponding" subrange [__bb; __ee) in the second sequence
+                    _ForwardIterator2 __bb = __first2;
+                    if (__b != __first1)
+                        __bb = std::lower_bound(__first2, __last2, *__b, __comp);
 
-                _ForwardIterator2 __ee = __last2;
-                if (__e != __last1)
-                    __ee = std::lower_bound(__bb, __last2, *__e, __comp);
+                    _ForwardIterator2 __ee = __last2;
+                    if (__e != __last1)
+                        __ee = std::lower_bound(__bb, __last2, *__e, __comp);
 
-                const _DifferenceType __buf_pos = __size_func((__b - __first1), (__bb - __first2));
-                auto __buffer_b = __buffer + __buf_pos;
-                auto __res = __set_op(__b, __e, __bb, __ee, __buffer_b, __comp);
+                    const _DifferenceType __buf_pos = __size_func((__b - __first1), (__bb - __first2));
+                    auto __buffer_b = __buffer + __buf_pos;
+                    auto __res = __set_op(__b, __e, __bb, __ee, __buffer_b, __comp);
 
-                return _SetRange{0, __res - __buffer_b, __buf_pos};
-            },
-            [](const _SetRange& __a, const _SetRange& __b) { // Combine
-                if (__b.__buf_pos > __a.__buf_pos || ((__b.__buf_pos == __a.__buf_pos) && !__b.empty()))
-                    return _SetRange{__a.__pos + __a.__len + __b.__pos, __b.__len, __b.__buf_pos};
-                return _SetRange{__b.__pos + __b.__len + __a.__pos, __a.__len, __a.__buf_pos};
-            },
-            __scan,                                     // Scan
-            [&__m, &__scan](const _SetRange& __total) { // Apex
-                //final scan
-                __scan(0, 0, __total);
-                __m = __total.__pos + __total.__len;
-            });
-        return __result + __m;
-    });
+                    return _SetRange{0, __res - __buffer_b, __buf_pos};
+                },
+                [](const _SetRange& __a, const _SetRange& __b) { // Combine
+                    if (__b.__buf_pos > __a.__buf_pos || ((__b.__buf_pos == __a.__buf_pos) && !__b.empty()))
+                        return _SetRange{__a.__pos + __a.__len + __b.__pos, __b.__len, __b.__buf_pos};
+                    return _SetRange{__b.__pos + __b.__len + __a.__pos, __a.__len, __a.__buf_pos};
+                },
+                __scan,                                     // Scan
+                [&__m, &__scan](const _SetRange& __total) { // Apex
+                    //final scan
+                    __scan(0, 0, __total);
+                    __m = __total.__pos + __total.__len;
+                });
+            return __result + __m;
+        });
 }
 
 //a shared parallel pattern for '__pattern_set_union' and '__pattern_set_symmetric_difference'
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _SetUnionOp, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare, class _SetUnionOp>
 _OutputIterator
-__parallel_set_union_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+__parallel_set_union_op(_Tag __tag, _ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                         _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
-                        _Compare __comp, _SetUnionOp __set_union_op, _IsVector __is_vector)
+                        _Compare __comp, _SetUnionOp __set_union_op)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     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 __internal::__brick_copy(__begin, __end, __res, __is_vector);
-    };
-    auto __copy_range2 = [__is_vector](_ForwardIterator2 __begin, _ForwardIterator2 __end, _OutputIterator __res) {
-        return __internal::__brick_copy(__begin, __end, __res, __is_vector);
-    };
+    auto copy_range1 = [](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res)
+    { return __internal::__brick_copy(__begin, __end, __res, typename _Tag::__is_vector{}); };
+    auto copy_range2 = [](_ForwardIterator2 __begin, _ForwardIterator2 __end, _OutputIterator __res)
+    { return __internal::__brick_copy(__begin, __end, __res, typename _Tag::__is_vector{}); };
 
     // {1} {}: parallel copying just first sequence
     if (__n2 == 0)
-        return __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
-                                                 __copy_range1, std::true_type());
+        return __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                                                 __result, copy_range1);
 
     // {} {2}: parallel copying justmake  second sequence
     if (__n1 == 0)
-        return __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2, __result,
-                                                 __copy_range2, std::true_type());
+        return __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
+                                                 __result, copy_range2);
 
     // testing  whether the sequences are intersected
     _ForwardIterator1 __left_bound_seq_1 = std::lower_bound(__first1, __last1, *__first2, __comp);
@@ -2916,14 +3021,16 @@ __parallel_set_union_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _
     {
         //{1} < {2}: seq2 is wholly greater than seq1, so, do parallel copying seq1 and seq2
         __par_backend::__parallel_invoke(
-            std::forward<_ExecutionPolicy>(__exec),
-            [=] {
-                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
-                                                  __copy_range1, std::true_type());
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec),
+            [=]
+            {
+                __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                                                  __result, copy_range1);
             },
-            [=] {
-                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
-                                                  __result + __n1, __copy_range2, std::true_type());
+            [=]
+            {
+                __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
+                                                  __result + __n1, copy_range2);
             });
         return __result + __n1 + __n2;
     }
@@ -2935,14 +3042,16 @@ __parallel_set_union_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _
     {
         //{2} < {1}: seq2 is wholly greater than seq1, so, do parallel copying seq1 and seq2
         __par_backend::__parallel_invoke(
-            std::forward<_ExecutionPolicy>(__exec),
-            [=] {
-                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __last2, __result,
-                                                  __copy_range2, std::true_type());
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec),
+            [=]
+            {
+                __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first2, __last2,
+                                                  __result, copy_range2);
             },
-            [=] {
-                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
-                                                  __result + __n2, __copy_range1, std::true_type());
+            [=]
+            {
+                __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1,
+                                                  __result + __n2, copy_range1);
             });
         return __result + __n1 + __n2;
     }
@@ -2953,17 +3062,19 @@ __parallel_set_union_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _
         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),
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec),
             //do parallel copying of [first1; left_bound_seq_1)
-            [=] {
-                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first1, __left_bound_seq_1,
-                                                  __res_or, __copy_range1, std::true_type());
+            [=]
+            {
+                __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                                  __left_bound_seq_1, __res_or, copy_range1);
             },
-            [=, &__result] {
+            [=, &__result]
+            {
                 __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);
+                    __tag, std::forward<_ExecutionPolicy>(__exec), __left_bound_seq_1, __last1, __first2, __last2,
+                    __result, __comp, [](_DifferenceType __n, _DifferenceType __m) { return __n + __m; },
+                    __set_union_op);
             });
         return __result;
     }
@@ -2975,24 +3086,26 @@ __parallel_set_union_op(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _
         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),
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec),
             //do parallel copying of [first2; left_bound_seq_2)
-            [=] {
-                __internal::__pattern_walk2_brick(std::forward<_ExecutionPolicy>(__exec), __first2, __left_bound_seq_2,
-                                                  __res_or, __copy_range2, std::true_type());
+            [=]
+            {
+                __internal::__pattern_walk2_brick(__tag, std::forward<_ExecutionPolicy>(__exec), __first2,
+                                                  __left_bound_seq_2, __res_or, copy_range2);
             },
-            [=, &__result] {
+            [=, &__result]
+            {
                 __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);
+                    __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __left_bound_seq_2, __last2,
+                    __result, __comp, [](_DifferenceType __n, _DifferenceType __m) { return __n + __m; },
+                    __set_union_op);
             });
         return __result;
     }
 
     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);
+        __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
+        [](_DifferenceType __n, _DifferenceType __m) { return __n + __m; }, __set_union_op);
 }
 
 //------------------------------------------------------------------------
@@ -3019,33 +3132,33 @@ struct __BrickCopyConstruct
     }
 };
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _OutputIterator, class _Compare>
 _OutputIterator
-__brick_set_union(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                  _ForwardIterator2 __last2, _OutputIterator __result, _Compare __comp,
+__brick_set_union(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+                  _RandomAccessIterator2 __last2, _OutputIterator __result, _Compare __comp,
                   /*__is_vector=*/std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::set_union(__first1, __last1, __first2, __last2, __result, __comp);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _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
+__pattern_set_union(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                    _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
+                    _Compare __comp) noexcept
 {
-    return __internal::__brick_set_union(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
+    return __internal::__brick_set_union(__first1, __last1, __first2, __last2, __result, __comp,
+                                         typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _OutputIterator, class _Compare>
 _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)
+__pattern_set_union(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                    _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                    _OutputIterator __result, _Compare __comp)
 {
 
     const auto __n1 = __last1 - __first1;
@@ -3057,13 +3170,13 @@ __pattern_set_union(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _Forw
 
     typedef typename std::iterator_traits<_OutputIterator>::value_type _Tp;
     return __parallel_set_union_op(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
-        [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2, _ForwardIterator2 __last2,
-           _Tp* __result, _Compare __comp) {
+        __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
+        [](_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+           _RandomAccessIterator2 __last2, _Tp* __result, _Compare __comp)
+        {
             return __pstl::__utils::__set_union_construct(__first1, __last1, __first2, __last2, __result, __comp,
                                                           __BrickCopyConstruct<_IsVector>());
-        },
-        __is_vector);
+        });
 }
 
 //------------------------------------------------------------------------
@@ -3079,35 +3192,37 @@ __brick_set_intersection(_ForwardIterator1 __first1, _ForwardIterator1 __last1,
     return std::set_intersection(__first1, __last1, __first2, __last2, __result, __comp);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator
-__brick_set_intersection(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                         _ForwardIterator2 __last2, _OutputIterator __result, _Compare __comp,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__brick_set_intersection(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
+                         _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                         _RandomAccessIterator3 __result, _Compare __comp,
                          /*__is_vector=*/std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::set_intersection(__first1, __last1, __first2, __last2, __result, __comp);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_set_intersection(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+__pattern_set_intersection(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                            _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
-                           _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+                           _Compare __comp) noexcept
 {
-    return __internal::__brick_set_intersection(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
+    return __internal::__brick_set_intersection(__first1, __last1, __first2, __last2, __result, __comp,
+                                                typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__pattern_set_intersection(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                           _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+                           _RandomAccessIterator2 __last2, _RandomAccessIterator3 __result, _Compare __comp)
 {
-    typedef typename std::iterator_traits<_OutputIterator>::value_type _Tp;
-    typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator3>::value_type _Tp;
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::difference_type _DifferenceType;
 
     const auto __n1 = __last1 - __first1;
     const auto __n2 = __last2 - __first2;
@@ -3117,13 +3232,13 @@ __pattern_set_intersection(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1
         return __result;
 
     // testing  whether the sequences are intersected
-    _ForwardIterator1 __left_bound_seq_1 = std::lower_bound(__first1, __last1, *__first2, __comp);
+    _RandomAccessIterator1 __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 = std::lower_bound(__first2, __last2, *__first1, __comp);
+    _RandomAccessIterator2 __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;
@@ -3133,14 +3248,13 @@ __pattern_set_intersection(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1
     {
         //we know proper offset due to [first1; left_bound_seq_1) < [first2; last2)
         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, _Tp* __result, _Compare __comp) {
+            __tag, std::forward<_ExecutionPolicy>(__exec), __left_bound_seq_1, __last1, __first2, __last2, __result,
+            __comp, [](_DifferenceType __n, _DifferenceType __m) { return std::min(__n, __m); },
+            [](_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+               _RandomAccessIterator2 __last2, _Tp* __result, _Compare __comp) {
                 return __pstl::__utils::__set_intersection_construct(__first1, __last1, __first2, __last2, __result,
                                                                      __comp);
-            },
-            __is_vector);
+            });
     }
 
     const auto __m2 = __last2 - __left_bound_seq_2 + __n1;
@@ -3148,14 +3262,13 @@ __pattern_set_intersection(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1
     {
         //we know proper offset due to [first2; left_bound_seq_2) < [first1; last1)
         __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, _Tp* __result, _Compare __comp) {
+            __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __left_bound_seq_2, __last2, __result,
+            __comp, [](_DifferenceType __n, _DifferenceType __m) { return std::min(__n, __m); },
+            [](_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+               _RandomAccessIterator2 __last2, _Tp* __result, _Compare __comp) {
                 return __pstl::__utils::__set_intersection_construct(__first2, __last2, __first1, __last1, __result,
                                                                      __comp);
-            },
-            __is_vector);
+            });
         return __result;
     }
 
@@ -3176,35 +3289,36 @@ __brick_set_difference(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _F
     return std::set_difference(__first1, __last1, __first2, __last2, __result, __comp);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator
-__brick_set_difference(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                       _ForwardIterator2 __last2, _OutputIterator __result, _Compare __comp,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__brick_set_difference(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+                       _RandomAccessIterator2 __last2, _RandomAccessIterator3 __result, _Compare __comp,
                        /*__is_vector=*/std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::set_difference(__first1, __last1, __first2, __last2, __result, __comp);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_set_difference(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+__pattern_set_difference(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                          _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
-                         _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+                         _Compare __comp) noexcept
 {
-    return __internal::__brick_set_difference(__first1, __last1, __first2, __last2, __result, __comp, __is_vector);
+    return __internal::__brick_set_difference(__first1, __last1, __first2, __last2, __result, __comp,
+                                              typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__pattern_set_difference(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                         _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+                         _RandomAccessIterator2 __last2, _RandomAccessIterator3 __result, _Compare __comp)
 {
-    typedef typename std::iterator_traits<_OutputIterator>::value_type _Tp;
-    typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator3>::value_type _Tp;
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::difference_type _DifferenceType;
 
     const auto __n1 = __last1 - __first1;
     const auto __n2 = __last2 - __first2;
@@ -3216,44 +3330,38 @@ __pattern_set_difference(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1,
     // {1} \ {}: parallel copying just first sequence
     if (__n2 == 0)
         return __internal::__pattern_walk2_brick(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
-            [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res) {
-                return __internal::__brick_copy(__begin, __end, __res, __is_vector);
-            },
-            std::true_type());
+            __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
+            [](_RandomAccessIterator1 __begin, _RandomAccessIterator1 __end, _RandomAccessIterator3 __res)
+            { return __internal::__brick_copy(__begin, __end, __res, _IsVector{}); });
 
     // testing  whether the sequences are intersected
-    _ForwardIterator1 __left_bound_seq_1 = std::lower_bound(__first1, __last1, *__first2, __comp);
+    _RandomAccessIterator1 __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 __internal::__pattern_walk2_brick(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
-            [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res) {
-                return __internal::__brick_copy(__begin, __end, __res, __is_vector);
-            },
-            std::true_type());
+            __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
+            [](_RandomAccessIterator1 __begin, _RandomAccessIterator1 __end, _RandomAccessIterator3 __res)
+            { return __internal::__brick_copy(__begin, __end, __res, _IsVector{}); });
 
     // testing  whether the sequences are intersected
-    _ForwardIterator2 __left_bound_seq_2 = std::lower_bound(__first2, __last2, *__first1, __comp);
+    _RandomAccessIterator2 __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 __internal::__pattern_walk2_brick(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
-            [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _OutputIterator __res) {
-                return __internal::__brick_copy(__begin, __end, __res, __is_vector);
-            },
-            std::true_type());
+            __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __result,
+            [](_RandomAccessIterator1 __begin, _RandomAccessIterator1 __end, _RandomAccessIterator3 __res)
+            { return __internal::__brick_copy(__begin, __end, __res, _IsVector{}); });
 
     if (__n1 + __n2 > __set_algo_cut_off)
-        return __parallel_set_op(std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result,
-                                 __comp, [](_DifferenceType __n, _DifferenceType) { return __n; },
-                                 [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                                    _ForwardIterator2 __last2, _Tp* __result, _Compare __comp) {
-                                     return __pstl::__utils::__set_difference_construct(
-                                         __first1, __last1, __first2, __last2, __result, __comp,
-                                         __BrickCopyConstruct<_IsVector>());
-                                 },
-                                 __is_vector);
+        return __parallel_set_op(
+            __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
+            [](_DifferenceType __n, _DifferenceType) { return __n; },
+            [](_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+               _RandomAccessIterator2 __last2, _Tp* __result, _Compare __comp)
+            {
+                return __pstl::__utils::__set_difference_construct(__first1, __last1, __first2, __last2, __result,
+                                                                   __comp, __BrickCopyConstruct<_IsVector>());
+            });
 
     // use serial algorithm
     return std::set_difference(__first1, __last1, __first2, __last2, __result, __comp);
@@ -3272,33 +3380,35 @@ __brick_set_symmetric_difference(_ForwardIterator1 __first1, _ForwardIterator1 _
     return std::set_symmetric_difference(__first1, __last1, __first2, __last2, __result, __comp);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator, class _Compare>
-_OutputIterator
-__brick_set_symmetric_difference(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                                 _ForwardIterator2 __last2, _OutputIterator __result, _Compare __comp,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__brick_set_symmetric_difference(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
+                                 _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                                 _RandomAccessIterator3 __result, _Compare __comp,
                                  /*__is_vector=*/std::true_type) noexcept
 {
     _PSTL_PRAGMA_MESSAGE("Vectorized algorithm unimplemented, redirected to serial");
     return std::set_symmetric_difference(__first1, __last1, __first2, __last2, __result, __comp);
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
-          class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _OutputIterator,
+          class _Compare>
 _OutputIterator
-__pattern_set_symmetric_difference(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+__pattern_set_symmetric_difference(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                                    _ForwardIterator2 __first2, _ForwardIterator2 __last2, _OutputIterator __result,
-                                   _Compare __comp, _IsVector __is_vector, /*is_parallel=*/std::false_type) noexcept
+                                   _Compare __comp) noexcept
 {
     return __internal::__brick_set_symmetric_difference(__first1, __last1, __first2, __last2, __result, __comp,
-                                                        __is_vector);
+                                                        typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _RandomAccessIterator3, class _Compare>
+_RandomAccessIterator3
+__pattern_set_symmetric_difference(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec,
+                                   _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
+                                   _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                                   _RandomAccessIterator3 __result, _Compare __comp)
 {
 
     const auto __n1 = __last1 - __first1;
@@ -3308,15 +3418,15 @@ __pattern_set_symmetric_difference(_ExecutionPolicy&& __exec, _ForwardIterator1
     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 _Tp;
+    typedef typename std::iterator_traits<_RandomAccessIterator3>::value_type _Tp;
     return __internal::__parallel_set_union_op(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
-        [](_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2, _ForwardIterator2 __last2,
-           _Tp* __result, _Compare __comp) {
+        __tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
+        [](_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+           _RandomAccessIterator2 __last2, _Tp* __result, _Compare __comp)
+        {
             return __pstl::__utils::__set_symmetric_difference_construct(__first1, __last1, __first2, __last2, __result,
                                                                          __comp, __BrickCopyConstruct<_IsVector>());
-        },
-        __is_vector);
+        });
 }
 
 //------------------------------------------------------------------------
@@ -3344,12 +3454,12 @@ __brick_is_heap_until(_RandomAccessIterator __first, _RandomAccessIterator __las
         [&__comp](_RandomAccessIterator __it, _SizeType __i) { return __comp(__it[(__i - 1) / 2], __it[__i]); });
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 _RandomAccessIterator
-__pattern_is_heap_until(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last,
-                        _Compare __comp, _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
+__pattern_is_heap_until(_Tag, _ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last,
+                        _Compare __comp) noexcept
 {
-    return __internal::__brick_is_heap_until(__first, __last, __comp, __is_vector);
+    return __internal::__brick_is_heap_until(__first, __last, __comp, typename _Tag::__is_vector{});
 }
 
 template <class _RandomAccessIterator, class _DifferenceType, class _Compare>
@@ -3378,22 +3488,27 @@ __is_heap_until_local(_RandomAccessIterator __first, _DifferenceType __begin, _D
         [&__comp](_RandomAccessIterator __it, _DifferenceType __i) { return __comp(__it[(__i - 1) / 2], __it[__i]); });
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 _RandomAccessIterator
-__pattern_is_heap_until(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
-                        _Compare __comp, _IsVector __is_vector, /* is_parallel = */ std::true_type) noexcept
+__pattern_is_heap_until(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                        _RandomAccessIterator __last, _Compare __comp) noexcept
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__last - __first < 2)
         return __last;
 
-    return __internal::__except_handler([&]() {
-        return __parallel_find(
-            std::forward<_ExecutionPolicy>(__exec), __first, __last,
-            [__first, __comp, __is_vector](_RandomAccessIterator __i, _RandomAccessIterator __j) {
-                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);
-    });
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __parallel_find(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                [__first, __comp](_RandomAccessIterator __i, _RandomAccessIterator __j) {
+                    return __internal::__is_heap_until_local(__first, __i - __first, __j - __first, __comp,
+                                                             _IsVector{});
+                },
+                std::less<typename std::iterator_traits<_RandomAccessIterator>::difference_type>(), /*is_first=*/true);
+        });
 }
 
 //------------------------------------------------------------------------
@@ -3408,47 +3523,51 @@ __brick_min_element(_ForwardIterator __first, _ForwardIterator __last, _Compare
     return std::min_element(__first, __last, __comp);
 }
 
-template <typename _ForwardIterator, typename _Compare>
-_ForwardIterator
-__brick_min_element(_ForwardIterator __first, _ForwardIterator __last, _Compare __comp,
+template <typename _RandomAccessIterator, typename _Compare>
+_RandomAccessIterator
+__brick_min_element(_RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
                     /* __is_vector = */ std::true_type) noexcept
 {
-#if _PSTL_UDR_PRESENT
+#if defined(_PSTL_UDR_PRESENT)
     return __unseq_backend::__simd_min_element(__first, __last - __first, __comp);
 #else
     return std::min_element(__first, __last, __comp);
 #endif
 }
 
-template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
+template <typename _Tag, typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare>
 _ForwardIterator
-__pattern_min_element(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp,
-                      _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
+__pattern_min_element(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                      _Compare __comp) noexcept
 {
-    return __internal::__brick_min_element(__first, __last, __comp, __is_vector);
+    return __internal::__brick_min_element(__first, __last, __comp, typename _Tag::__is_vector{});
 }
 
-template <typename _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare, typename _IsVector>
+template <typename _IsVector, typename _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare>
 _RandomAccessIterator
-__pattern_min_element(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last,
-                      _Compare __comp, _IsVector __is_vector, /* is_parallel = */ std::true_type)
+__pattern_min_element(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                      _RandomAccessIterator __last, _Compare __comp)
 {
     if (__first == __last)
         return __last;
 
-    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 =
-                    __internal::__brick_min_element(__begin, __end, __comp, __is_vector);
-                return __internal::__cmp_iterators_by_values(__init, subresult, __comp);
-            },
-            [=](_RandomAccessIterator __it1, _RandomAccessIterator __it2) -> _RandomAccessIterator {
-                return __internal::__cmp_iterators_by_values(__it1, __it2, __comp);
-            });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __par_backend::__parallel_reduce(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first + 1, __last, __first,
+                [=](_RandomAccessIterator __begin, _RandomAccessIterator __end,
+                    _RandomAccessIterator __init) -> _RandomAccessIterator
+                {
+                    const _RandomAccessIterator subresult =
+                        __internal::__brick_min_element(__begin, __end, __comp, _IsVector{});
+                    return __internal::__cmp_iterators_by_values(__init, subresult, __comp);
+                },
+                [=](_RandomAccessIterator __it1, _RandomAccessIterator __it2) -> _RandomAccessIterator
+                { return __internal::__cmp_iterators_by_values(__it1, __it2, __comp); });
+        });
 }
 
 //------------------------------------------------------------------------
@@ -3463,46 +3582,51 @@ __brick_minmax_element(_ForwardIterator __first, _ForwardIterator __last, _Compa
     return std::minmax_element(__first, __last, __comp);
 }
 
-template <typename _ForwardIterator, typename _Compare>
-std::pair<_ForwardIterator, _ForwardIterator>
-__brick_minmax_element(_ForwardIterator __first, _ForwardIterator __last, _Compare __comp,
+template <typename _RandomAccessIterator, typename _Compare>
+std::pair<_RandomAccessIterator, _RandomAccessIterator>
+__brick_minmax_element(_RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
                        /* __is_vector = */ std::true_type) noexcept
 {
-#if _PSTL_UDR_PRESENT
+#if defined(_PSTL_UDR_PRESENT)
     return __unseq_backend::__simd_minmax_element(__first, __last - __first, __comp);
 #else
     return std::minmax_element(__first, __last, __comp);
 #endif
 }
 
-template <typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare, typename _IsVector>
+template <typename _Tag, typename _ExecutionPolicy, typename _ForwardIterator, typename _Compare>
 std::pair<_ForwardIterator, _ForwardIterator>
-__pattern_minmax_element(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp,
-                         _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
+__pattern_minmax_element(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                         _Compare __comp) noexcept
 {
-    return __internal::__brick_minmax_element(__first, __last, __comp, __is_vector);
+    return __internal::__brick_minmax_element(__first, __last, __comp, typename _Tag::__is_vector{});
 }
 
-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)
+template <typename _IsVector, typename _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare>
+std::pair<_RandomAccessIterator, _RandomAccessIterator>
+__pattern_minmax_element(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                         _RandomAccessIterator __last, _Compare __comp)
 {
     if (__first == __last)
         return std::make_pair(__first, __first);
 
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     return __internal::__except_handler([&]() {
-        typedef std::pair<_ForwardIterator, _ForwardIterator> _Result;
+        typedef std::pair<_RandomAccessIterator, _RandomAccessIterator> _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 = __internal::__brick_minmax_element(__begin, __end, __comp, __is_vector);
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first + 1, __last,
+            std::make_pair(__first, __first),
+            [=](_RandomAccessIterator __begin, _RandomAccessIterator __end, _Result __init) -> _Result
+            {
+                const _Result __subresult = __internal::__brick_minmax_element(__begin, __end, __comp, _IsVector{});
                 return std::make_pair(
                     __internal::__cmp_iterators_by_values(__subresult.first, __init.first, __comp),
                     __internal::__cmp_iterators_by_values(__init.second, __subresult.second, std::not_fn(__comp)));
             },
-            [=](_Result __p1, _Result __p2) -> _Result {
+            [=](_Result __p1, _Result __p2) -> _Result
+            {
                 return std::make_pair(
                     __internal::__cmp_iterators_by_values(__p1.first, __p2.first, __comp),
                     __internal::__cmp_iterators_by_values(__p2.second, __p1.second, std::not_fn(__comp)));
@@ -3518,7 +3642,7 @@ std::pair<_ForwardIterator1, _ForwardIterator2>
 __mismatch_serial(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
                   _ForwardIterator2 __last2, _BinaryPredicate __pred)
 {
-#if _PSTL_CPP14_2RANGE_MISMATCH_EQUAL_PRESENT
+#if defined(_PSTL_CPP14_2RANGE_MISMATCH_EQUAL_PRESENT)
     return std::mismatch(__first1, __last1, __first2, __last2, __pred);
 #else
     for (; __first1 != __last1 && __first2 != __last2 && __pred(*__first1, *__first2); ++__first1, ++__first2)
@@ -3536,38 +3660,40 @@ __brick_mismatch(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _Forward
     return __mismatch_serial(__first1, __last1, __first2, __last2, __pred);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
-std::pair<_ForwardIterator1, _ForwardIterator2>
-__brick_mismatch(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                 _ForwardIterator2 __last2, _Predicate __pred, /* __is_vector = */ std::true_type) noexcept
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Predicate>
+std::pair<_RandomAccessIterator1, _RandomAccessIterator2>
+__brick_mismatch(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2,
+                 _RandomAccessIterator2 __last2, _Predicate __pred, /* __is_vector = */ std::true_type) noexcept
 {
     auto __n = std::min(__last1 - __first1, __last2 - __first2);
     return __unseq_backend::__simd_first(__first1, __n, __first2, std::not_fn(__pred));
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
 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
+__pattern_mismatch(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                   _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Predicate __pred) noexcept
 {
-    return __internal::__brick_mismatch(__first1, __last1, __first2, __last2, __pred, __is_vector);
+    return __internal::__brick_mismatch(__first1, __last1, __first2, __last2, __pred, typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Predicate,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Predicate>
 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
+__pattern_mismatch(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                   _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                   _Predicate __pred) noexcept
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     return __internal::__except_handler([&]() {
         auto __n = std::min(__last1 - __first1, __last2 - __first2);
         auto __result = __internal::__parallel_find(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
-            [__first1, __first2, __pred, __is_vector](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j) {
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
+            [__first1, __first2, __pred](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+            {
                 return __internal::__brick_mismatch(__i, __j, __first2 + (__i - __first1), __first2 + (__j - __first1),
-                                                    __pred, __is_vector)
+                                                    __pred, _IsVector{})
                     .first;
             },
             std::less<typename std::iterator_traits<_RandomAccessIterator1>::difference_type>(), /*is_first=*/true);
@@ -3588,10 +3714,11 @@ __brick_lexicographical_compare(_ForwardIterator1 __first1, _ForwardIterator1 __
     return std::lexicographical_compare(__first1, __last1, __first2, __last2, __comp);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Compare>
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Compare>
 bool
-__brick_lexicographical_compare(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
-                                _ForwardIterator2 __last2, _Compare __comp, /* __is_vector = */ std::true_type) noexcept
+__brick_lexicographical_compare(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
+                                _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2, _Compare __comp,
+                                /* __is_vector = */ std::true_type) noexcept
 {
     if (__first2 == __last2)
     { // if second sequence is empty
@@ -3603,12 +3730,12 @@ __brick_lexicographical_compare(_ForwardIterator1 __first1, _ForwardIterator1 __
     }
     else
     {
-        typedef typename std::iterator_traits<_ForwardIterator1>::reference ref_type1;
-        typedef typename std::iterator_traits<_ForwardIterator2>::reference ref_type2;
+        typedef typename std::iterator_traits<_RandomAccessIterator1>::reference ref_type1;
+        typedef typename std::iterator_traits<_RandomAccessIterator2>::reference ref_type2;
         --__last1;
         --__last2;
         auto __n = std::min(__last1 - __first1, __last2 - __first2);
-        std::pair<_ForwardIterator1, _ForwardIterator2> __result = __unseq_backend::__simd_first(
+        std::pair<_RandomAccessIterator1, _RandomAccessIterator2> __result = __unseq_backend::__simd_first(
             __first1, __n, __first2, [__comp](const ref_type1 __x, const ref_type2 __y) mutable {
                 return __comp(__x, __y) || __comp(__y, __x);
             });
@@ -3624,21 +3751,25 @@ __brick_lexicographical_compare(_ForwardIterator1 __first1, _ForwardIterator1 __
     }
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare>
 bool
-__pattern_lexicographical_compare(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
-                                  _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp,
-                                  _IsVector __is_vector, /* is_parallel = */ std::false_type) noexcept
+__pattern_lexicographical_compare(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+                                  _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp) noexcept
 {
-    return __internal::__brick_lexicographical_compare(__first1, __last1, __first2, __last2, __comp, __is_vector);
+    return __internal::__brick_lexicographical_compare(__first1, __last1, __first2, __last2, __comp,
+                                                       typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Compare, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Compare>
 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
+__pattern_lexicographical_compare(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec,
+                                  _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
+                                  _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                                  _Compare __comp) noexcept
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     if (__first2 == __last2)
     { // if second sequence is empty
         return false;
@@ -3649,22 +3780,23 @@ __pattern_lexicographical_compare(_ExecutionPolicy&& __exec, _ForwardIterator1 _
     }
     else
     {
-        typedef typename std::iterator_traits<_ForwardIterator1>::reference _RefType1;
-        typedef typename std::iterator_traits<_ForwardIterator2>::reference _RefType2;
+        typedef typename std::iterator_traits<_RandomAccessIterator1>::reference _RefType1;
+        typedef typename std::iterator_traits<_RandomAccessIterator2>::reference _RefType2;
         --__last1;
         --__last2;
         auto __n = std::min(__last1 - __first1, __last2 - __first2);
         auto __result = __internal::__parallel_find(
-            std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
-            [__first1, __first2, &__comp, __is_vector](_ForwardIterator1 __i, _ForwardIterator1 __j) {
-                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)
+            __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first1, __first1 + __n,
+            [__first1, __first2, &__comp](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j)
+            {
+                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); },
+                           _IsVector{})
                     .first;
             },
-            std::less<typename std::iterator_traits<_ForwardIterator1>::difference_type>(), /*is_first=*/true);
+            std::less<typename std::iterator_traits<_RandomAccessIterator1>::difference_type>(), /*is_first=*/true);
 
         if (__result == __last1 && __first2 + (__result - __first1) != __last2)
         { // if first sequence shorter than second
diff --git a/libstdc++-v3/include/pstl/execution_defs.h b/libstdc++-v3/include/pstl/execution_defs.h
index 3eca558dac2..522518c4116 100644
--- a/libstdc++-v3/include/pstl/execution_defs.h
+++ b/libstdc++-v3/include/pstl/execution_defs.h
@@ -22,88 +22,20 @@ inline namespace v1
 // 2.4, Sequential execution policy
 class sequenced_policy
 {
-  public:
-    // For internal use only
-    static constexpr std::false_type
-    __allow_unsequenced()
-    {
-        return std::false_type{};
-    }
-    static constexpr std::false_type
-    __allow_vector()
-    {
-        return std::false_type{};
-    }
-    static constexpr std::false_type
-    __allow_parallel()
-    {
-        return std::false_type{};
-    }
 };
 
 // 2.5, Parallel execution policy
 class parallel_policy
 {
-  public:
-    // For internal use only
-    static constexpr std::false_type
-    __allow_unsequenced()
-    {
-        return std::false_type{};
-    }
-    static constexpr std::false_type
-    __allow_vector()
-    {
-        return std::false_type{};
-    }
-    static constexpr std::true_type
-    __allow_parallel()
-    {
-        return std::true_type{};
-    }
 };
 
 // 2.6, Parallel+Vector execution policy
 class parallel_unsequenced_policy
 {
-  public:
-    // For internal use only
-    static constexpr std::true_type
-    __allow_unsequenced()
-    {
-        return std::true_type{};
-    }
-    static constexpr std::true_type
-    __allow_vector()
-    {
-        return std::true_type{};
-    }
-    static constexpr std::true_type
-    __allow_parallel()
-    {
-        return std::true_type{};
-    }
 };
 
 class unsequenced_policy
 {
-  public:
-    // For internal use only
-    static constexpr std::true_type
-    __allow_unsequenced()
-    {
-        return std::true_type{};
-    }
-    static constexpr std::true_type
-    __allow_vector()
-    {
-        return std::true_type{};
-    }
-    static constexpr std::false_type
-    __allow_parallel()
-    {
-        return std::false_type{};
-    }
 };
 
 // 2.8, Execution policy objects
@@ -135,7 +67,7 @@ struct is_execution_policy<__pstl::execution::unsequenced_policy> : std::true_ty
 {
 };
 
-#if _PSTL_CPP14_VARIABLE_TEMPLATES_PRESENT
+#if defined (_PSTL_CPP14_VARIABLE_TEMPLATES_PRESENT)
 template <class _Tp>
 constexpr bool is_execution_policy_v = __pstl::execution::is_execution_policy<_Tp>::value;
 #endif
@@ -155,6 +87,12 @@ using __enable_if_execution_policy =
     typename std::enable_if<__pstl::execution::is_execution_policy<typename std::decay<_ExecPolicy>::type>::value,
                             _Tp>::type;
 #endif
+
+template <class _IsVector>
+struct __serial_tag;
+template <class _IsVector>
+struct __parallel_tag;
+
 } // namespace __internal
 
 } // namespace __pstl
diff --git a/libstdc++-v3/include/pstl/execution_impl.h b/libstdc++-v3/include/pstl/execution_impl.h
index e0969a1db9c..64f6cc4357a 100644
--- a/libstdc++-v3/include/pstl/execution_impl.h
+++ b/libstdc++-v3/include/pstl/execution_impl.h
@@ -20,140 +20,79 @@ namespace __pstl
 namespace __internal
 {
 
-using namespace __pstl::execution;
+template <typename _IteratorTag, typename... _IteratorTypes>
+using __are_iterators_of = std::conjunction<
+    std::is_base_of<_IteratorTag, typename std::iterator_traits<std::decay_t<_IteratorTypes>>::iterator_category>...>;
 
-/* predicate */
+template <typename... _IteratorTypes>
+using __are_random_access_iterators = __are_iterators_of<std::random_access_iterator_tag, _IteratorTypes...>;
 
-template <typename _Tp>
-std::false_type __lazy_and(_Tp, std::false_type)
-{
-    return std::false_type{};
-}
-
-template <typename _Tp>
-inline _Tp
-__lazy_and(_Tp __a, std::true_type)
-{
-    return __a;
-}
-
-template <typename _Tp>
-std::true_type __lazy_or(_Tp, std::true_type)
-{
-    return std::true_type{};
-}
-
-template <typename _Tp>
-inline _Tp
-__lazy_or(_Tp __a, std::false_type)
-{
-    return __a;
-}
-
-/* iterator */
-template <typename _IteratorType, typename... _OtherIteratorTypes>
-struct __is_random_access_iterator
-{
-    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>
+struct __serial_backend_tag
 {
 };
-
-/* policy */
-template <typename _Policy>
-struct __policy_traits
+struct __tbb_backend_tag
 {
 };
-
-template <>
-struct __policy_traits<sequenced_policy>
+struct __openmp_backend_tag
 {
-    typedef std::false_type allow_parallel;
-    typedef std::false_type allow_unsequenced;
-    typedef std::false_type allow_vector;
 };
 
-template <>
-struct __policy_traits<unsequenced_policy>
-{
-    typedef std::false_type allow_parallel;
-    typedef std::true_type allow_unsequenced;
-    typedef std::true_type allow_vector;
-};
+#if defined(_PSTL_PAR_BACKEND_TBB)
+using __par_backend_tag = __tbb_backend_tag;
+#elif defined(_PSTL_PAR_BACKEND_OPENMP)
+using __par_backend_tag = __openmp_backend_tag;
+#elif defined(_PSTL_PAR_BACKEND_SERIAL)
+using __par_backend_tag = __serial_backend_tag;
+#else
+#    error "A parallel backend must be specified";
+#endif
 
-template <>
-struct __policy_traits<parallel_policy>
+template <class _IsVector>
+struct __serial_tag
 {
-    typedef std::true_type allow_parallel;
-    typedef std::false_type allow_unsequenced;
-    typedef std::false_type allow_vector;
+    using __is_vector = _IsVector;
 };
 
-template <>
-struct __policy_traits<parallel_unsequenced_policy>
+template <class _IsVector>
+struct __parallel_tag
 {
-    typedef std::true_type allow_parallel;
-    typedef std::true_type allow_unsequenced;
-    typedef std::true_type allow_vector;
+    using __is_vector = _IsVector;
+    // backend tag can be change depending on
+    // TBB availability in the environment
+    using __backend_tag = __par_backend_tag;
 };
 
-template <typename _ExecutionPolicy>
-using __collector_t =
-    typename __internal::__policy_traits<typename std::decay<_ExecutionPolicy>::type>::__collector_type;
-
-template <typename _ExecutionPolicy>
-using __allow_vector =
-    typename __internal::__policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_vector;
+template <class _IsVector, class... _IteratorTypes>
+using __tag_type = typename std::conditional<__internal::__are_random_access_iterators<_IteratorTypes...>::value,
+                                             __parallel_tag<_IsVector>, __serial_tag<_IsVector>>::type;
 
-template <typename _ExecutionPolicy>
-using __allow_unsequenced =
-    typename __internal::__policy_traits<typename std::decay<_ExecutionPolicy>::type>::__allow_unsequenced;
-
-template <typename _ExecutionPolicy>
-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(__internal::__lazy_and(__exec.__allow_vector(),
-                                       typename __internal::__is_random_access_iterator<_IteratorTypes...>::type()))
+template <class... _IteratorTypes>
+__serial_tag</*_IsVector = */ std::false_type>
+__select_backend(__pstl::execution::sequenced_policy, _IteratorTypes&&...)
 {
-    return __internal::__lazy_and(__exec.__allow_vector(),
-                                  typename __internal::__is_random_access_iterator<_IteratorTypes...>::type());
+    return {};
 }
 
-template <typename _ExecutionPolicy, typename... _IteratorTypes>
-auto
-__is_parallelization_preferred(_ExecutionPolicy&& __exec)
-    -> decltype(__internal::__lazy_and(__exec.__allow_parallel(),
-                                       typename __internal::__is_random_access_iterator<_IteratorTypes...>::type()))
+template <class... _IteratorTypes>
+__serial_tag<__internal::__are_random_access_iterators<_IteratorTypes...>>
+__select_backend(__pstl::execution::unsequenced_policy, _IteratorTypes&&...)
 {
-    return __internal::__lazy_and(__exec.__allow_parallel(),
-                                  typename __internal::__is_random_access_iterator<_IteratorTypes...>::type());
+    return {};
 }
 
-template <typename __policy, typename... _IteratorTypes>
-struct __prefer_unsequenced_tag
+template <class... _IteratorTypes>
+__tag_type</*_IsVector = */ std::false_type, _IteratorTypes...>
+__select_backend(__pstl::execution::parallel_policy, _IteratorTypes&&...)
 {
-    static constexpr bool value = __internal::__allow_unsequenced<__policy>::value &&
-                                  __internal::__is_random_access_iterator<_IteratorTypes...>::value;
-    typedef std::integral_constant<bool, value> type;
-};
+    return {};
+}
 
-template <typename __policy, typename... _IteratorTypes>
-struct __prefer_parallel_tag
+template <class... _IteratorTypes>
+__tag_type<__internal::__are_random_access_iterators<_IteratorTypes...>, _IteratorTypes...>
+__select_backend(__pstl::execution::parallel_unsequenced_policy, _IteratorTypes&&...)
 {
-    static constexpr bool value = __internal::__allow_parallel<__policy>::value &&
-                                  __internal::__is_random_access_iterator<_IteratorTypes...>::value;
-    typedef std::integral_constant<bool, value> type;
-};
+    return {};
+}
 
 } // namespace __internal
 } // namespace __pstl
diff --git a/libstdc++-v3/include/pstl/glue_algorithm_impl.h b/libstdc++-v3/include/pstl/glue_algorithm_impl.h
index b16306f2e99..5be45ce57fb 100644
--- a/libstdc++-v3/include/pstl/glue_algorithm_impl.h
+++ b/libstdc++-v3/include/pstl/glue_algorithm_impl.h
@@ -26,10 +26,10 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, bool>
 any_of(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred)
 {
-    return __pstl::__internal::__pattern_any_of(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_any_of(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                __pred);
 }
 
 // [alg.all_of]
@@ -56,20 +56,19 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Function>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 for_each(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Function __f)
 {
-    __pstl::__internal::__pattern_walk1(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __f,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_walk1(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last, __f);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Function>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 for_each_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __n, _Function __f)
 {
-    return __pstl::__internal::__pattern_walk1_n(
-        std::forward<_ExecutionPolicy>(__exec), __first, __n, __f,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_walk1_n(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n,
+                                                 __f);
 }
 
 // [alg.find]
@@ -78,10 +77,10 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 find_if(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred)
 {
-    return __pstl::__internal::__pattern_find_if(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_find_if(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                 __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
@@ -105,12 +104,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 find_end(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
          _ForwardIterator2 __s_last, _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_find_end(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __s_first, __s_last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __s_first);
+
+    return __pstl::__internal::__pattern_find_end(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                  __last, __s_first, __s_last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -128,12 +125,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 find_first_of(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last,
               _ForwardIterator2 __s_first, _ForwardIterator2 __s_last, _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_find_first_of(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __s_first, __s_last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __s_first);
+
+    return __pstl::__internal::__pattern_find_first_of(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                       __last, __s_first, __s_last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -150,23 +145,20 @@ template <class _ExecutionPolicy, class _ForwardIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 adjacent_find(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last)
 {
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
-    return __pstl::__internal::__pattern_adjacent_find(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, std::equal_to<_ValueType>(),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        /*first_semantic*/ false);
+    return __pstl::__internal::__pattern_adjacent_find(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                       __last, std::equal_to<_ValueType>(), /*first_semantic*/ false);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 adjacent_find(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_adjacent_find(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        /*first_semantic*/ false);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_adjacent_find(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                       __last, __pred, /*first_semantic*/ false);
 }
 
 // [alg.count]
@@ -179,12 +171,11 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy,
                                                  typename iterator_traits<_ForwardIterator>::difference_type>
 count(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value)
 {
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
-    return __pstl::__internal::__pattern_count(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last,
-        [&__value](const _ValueType& __x) { return __value == __x; },
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    return __pstl::__internal::__pattern_count(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                               [&__value](const _ValueType& __x) { return __value == __x; });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Predicate>
@@ -192,10 +183,9 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy,
                                                  typename iterator_traits<_ForwardIterator>::difference_type>
 count_if(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Predicate __pred)
 {
-    return __pstl::__internal::__pattern_count(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_count(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                               __pred);
 }
 
 // [alg.search]
@@ -205,12 +195,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 search(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __s_first,
        _ForwardIterator2 __s_last, _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_search(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __s_first, __s_last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __s_first);
+
+    return __pstl::__internal::__pattern_search(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                __s_first, __s_last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -226,10 +214,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 search_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Size __count,
          const _Tp& __value, _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_search_n(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __count, __value, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_search_n(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                  __last, __count, __value, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Tp>
@@ -247,34 +235,28 @@ template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterato
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator2>
 copy(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result)
 {
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__pattern_walk2_brick(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
-        [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _ForwardIterator2 __res) {
-            return __pstl::__internal::__brick_copy(__begin, __end, __res, __is_vector);
-        },
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+        __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+        [](_ForwardIterator1 __begin, _ForwardIterator1 __end, _ForwardIterator2 __res)
+        { return __pstl::__internal::__brick_copy(__begin, __end, __res, __is_vector{}); });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _Size, class _ForwardIterator2>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator2>
 copy_n(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _Size __n, _ForwardIterator2 __result)
 {
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__pattern_walk2_brick_n(
-        std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
-        [__is_vector](_ForwardIterator1 __begin, _Size __sz, _ForwardIterator2 __res) {
-            return __pstl::__internal::__brick_copy_n(__begin, __sz, __res, __is_vector);
-        },
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+        __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
+        [](_ForwardIterator1 __begin, _Size __sz, _ForwardIterator2 __res)
+        { return __pstl::__internal::__brick_copy_n(__begin, __sz, __res, __is_vector{}); });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Predicate>
@@ -282,12 +264,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 copy_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result,
         _Predicate __pred)
 {
-    return __pstl::__internal::__pattern_copy_if(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    return __pstl::__internal::__pattern_copy_if(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                 __last, __result, __pred);
 }
 
 // [alg.swap]
@@ -299,16 +279,16 @@ swap_ranges(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardItera
 {
     typedef typename iterator_traits<_ForwardIterator1>::reference _ReferenceType1;
     typedef typename iterator_traits<_ForwardIterator2>::reference _ReferenceType2;
-    return __pstl::__internal::__pattern_walk2(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2,
-        [](_ReferenceType1 __x, _ReferenceType2 __y) {
-            using std::swap;
-            swap(__x, __y);
-        },
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+
+    return __pstl::__internal::__pattern_walk2(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                               __last1, __first2,
+                                               [](_ReferenceType1 __x, _ReferenceType2 __y)
+                                               {
+                                                   using std::swap;
+                                                   swap(__x, __y);
+                                               });
 }
 
 // [alg.transform]
@@ -320,13 +300,12 @@ transform(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator
 {
     typedef typename iterator_traits<_ForwardIterator1>::reference _InputType;
     typedef typename iterator_traits<_ForwardIterator2>::reference _OutputType;
-    return __pstl::__internal::__pattern_walk2(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
-        [__op](_InputType __x, _OutputType __y) mutable { __y = __op(__x); },
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    return __pstl::__internal::__pattern_walk2(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                               __result,
+                                               [__op](_InputType __x, _OutputType __y) mutable { __y = __op(__x); });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator,
@@ -338,13 +317,12 @@ transform(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterato
     typedef typename iterator_traits<_ForwardIterator1>::reference _Input1Type;
     typedef typename iterator_traits<_ForwardIterator2>::reference _Input2Type;
     typedef typename iterator_traits<_ForwardIterator>::reference _OutputType;
+
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2, __result);
+
     return __pstl::__internal::__pattern_walk3(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __result,
-        [__op](_Input1Type __x, _Input2Type __y, _OutputType __z) mutable { __z = __op(__x, __y); },
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                         _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                           _ForwardIterator>(__exec));
+        __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __result,
+        [__op](_Input1Type __x, _Input2Type __y, _OutputType __z) mutable { __z = __op(__x, __y); });
 }
 
 // [alg.replace]
@@ -355,16 +333,17 @@ replace_if(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator
            const _Tp& __new_value)
 {
     typedef typename iterator_traits<_ForwardIterator>::reference _ElementType;
-    __pstl::__internal::__pattern_walk1(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last,
-        [&__pred, &__new_value](_ElementType __elem) {
-            if (__pred(__elem))
-            {
-                __elem = __new_value;
-            }
-        },
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_walk1(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                        [&__pred, &__new_value](_ElementType __elem)
+                                        {
+                                            if (__pred(__elem))
+                                            {
+                                                __elem = __new_value;
+                                            }
+                                        });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp>
@@ -383,13 +362,12 @@ replace_copy_if(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIt
 {
     typedef typename iterator_traits<_ForwardIterator1>::reference _InputType;
     typedef typename iterator_traits<_ForwardIterator2>::reference _OutputType;
+
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
     return __pstl::__internal::__pattern_walk2(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
-        [__pred, &__new_value](_InputType __x, _OutputType __y) mutable { __y = __pred(__x) ? __new_value : __x; },
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+        __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+        [__pred, &__new_value](_InputType __x, _OutputType __y) mutable { __y = __pred(__x) ? __new_value : __x; });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Tp>
@@ -407,10 +385,10 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Tp>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 fill(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, const _Tp& __value)
 {
-    __pstl::__internal::__pattern_fill(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __value,
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_fill(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                       __value);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Tp>
@@ -420,10 +398,10 @@ fill_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __count, const
     if (__count <= 0)
         return __first;
 
-    return __pstl::__internal::__pattern_fill_n(
-        std::forward<_ExecutionPolicy>(__exec), __first, __count, __value,
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_fill_n(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                __count, __value);
 }
 
 // [alg.generate]
@@ -431,10 +409,10 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Generator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 generate(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Generator __g)
 {
-    __pstl::__internal::__pattern_generate(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __g,
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_generate(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                           __g);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size, class _Generator>
@@ -444,10 +422,10 @@ generate_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __count, _
     if (__count <= 0)
         return __first;
 
-    return __pstl::__internal::__pattern_generate_n(
-        std::forward<_ExecutionPolicy>(__exec), __first, __count, __g,
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_generate_n(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                    __count, __g);
 }
 
 // [alg.remove]
@@ -473,10 +451,10 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 remove_if(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_remove_if(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_remove_if(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                   __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp>
@@ -493,10 +471,10 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _BinaryPredicate
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 unique(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_unique(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_unique(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator>
@@ -511,12 +489,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 unique_copy(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __result,
             _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_unique_copy(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    return __pstl::__internal::__pattern_unique_copy(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                     __last, __result, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -532,10 +508,9 @@ template <class _ExecutionPolicy, class _BidirectionalIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 reverse(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __last)
 {
-    __pstl::__internal::__pattern_reverse(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _BidirectionalIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _BidirectionalIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_reverse(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last);
 }
 
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _ForwardIterator>
@@ -543,12 +518,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 reverse_copy(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __last,
              _ForwardIterator __d_first)
 {
-    return __pstl::__internal::__pattern_reverse_copy(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __d_first,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _BidirectionalIterator, _ForwardIterator>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _BidirectionalIterator, _ForwardIterator>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __d_first);
+
+    return __pstl::__internal::__pattern_reverse_copy(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                      __last, __d_first);
 }
 
 // [alg.rotate]
@@ -557,10 +530,10 @@ template <class _ExecutionPolicy, class _ForwardIterator>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 rotate(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __middle, _ForwardIterator __last)
 {
-    return __pstl::__internal::__pattern_rotate(
-        std::forward<_ExecutionPolicy>(__exec), __first, __middle, __last,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_rotate(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                __middle, __last);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -568,12 +541,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 rotate_copy(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __middle, _ForwardIterator1 __last,
             _ForwardIterator2 __result)
 {
-    return __pstl::__internal::__pattern_rotate_copy(
-        std::forward<_ExecutionPolicy>(__exec), __first, __middle, __last, __result,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    return __pstl::__internal::__pattern_rotate_copy(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                     __middle, __last, __result);
 }
 
 // [alg.partitions]
@@ -582,20 +553,19 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, bool>
 is_partitioned(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_is_partitioned(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_is_partitioned(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                        __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _UnaryPredicate>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 partition(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _UnaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_partition(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_partition(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                   __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _BidirectionalIterator, class _UnaryPredicate>
@@ -603,10 +573,9 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _Bidirectiona
 stable_partition(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __last,
                  _UnaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_stable_partition(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _BidirectionalIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _BidirectionalIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_stable_partition(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                          __first, __last, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _ForwardIterator1, class _ForwardIterator2,
@@ -615,12 +584,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, std::pair<_Fo
 partition_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
                _ForwardIterator1 __out_true, _ForwardIterator2 __out_false, _UnaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_partition_copy(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __out_true, __out_false, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator, _ForwardIterator1,
-                                                         _ForwardIterator2>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator, _ForwardIterator1,
-                                                           _ForwardIterator2>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __out_true, __out_false);
+
+    return __pstl::__internal::__pattern_partition_copy(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                        __last, __out_true, __out_false, __pred);
 }
 
 // [alg.sort]
@@ -629,12 +596,11 @@ template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp)
 {
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
     typedef typename iterator_traits<_RandomAccessIterator>::value_type _InputType;
-    return __pstl::__internal::__pattern_sort(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec),
-        typename std::is_move_constructible<_InputType>::type());
+    return __pstl::__internal::__pattern_sort(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                              __comp, typename std::is_move_constructible<_InputType>::type());
 }
 
 template <class _ExecutionPolicy, class _RandomAccessIterator>
@@ -651,10 +617,10 @@ template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 stable_sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_stable_sort(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_stable_sort(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                     __last, __comp);
 }
 
 template <class _ExecutionPolicy, class _RandomAccessIterator>
@@ -672,12 +638,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, std::pair<_Fo
 mismatch(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
          _ForwardIterator2 __last2, _BinaryPredicate __pred)
 {
-    return __pstl::__internal::__pattern_mismatch(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __pred,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+
+    return __pstl::__internal::__pattern_mismatch(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                                  __last1, __first2, __last2, __pred);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _BinaryPredicate>
@@ -714,10 +678,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, bool>
 equal(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
       _BinaryPredicate __p)
 {
-    return __pstl::__internal::__pattern_equal(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __p,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+
+    return __pstl::__internal::__pattern_equal(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                               __last1, __first2, __p);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -732,10 +696,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, bool>
 equal(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
       _ForwardIterator2 __last2, _BinaryPredicate __p)
 {
-    return __pstl::__internal::__pattern_equal(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __p,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+
+    return __pstl::__internal::__pattern_equal(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                               __last1, __first2, __last2, __p);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -751,17 +715,14 @@ template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterato
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator2>
 move(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __d_first)
 {
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __d_first);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__pattern_walk2_brick(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __d_first,
-        [__is_vector](_ForwardIterator1 __begin, _ForwardIterator1 __end, _ForwardIterator2 __res) {
-            return __pstl::__internal::__brick_move(__begin, __end, __res, __is_vector);
-        },
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+        __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last, __d_first,
+        [](_ForwardIterator1 __begin, _ForwardIterator1 __end, _ForwardIterator2 __res)
+        { return __pstl::__internal::__brick_move(__begin, __end, __res, __is_vector{}); });
 }
 
 // [partial.sort]
@@ -771,10 +732,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 partial_sort(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __middle,
              _RandomAccessIterator __last, _Compare __comp)
 {
-    __pstl::__internal::__pattern_partial_sort(
-        std::forward<_ExecutionPolicy>(__exec), __first, __middle, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_partial_sort(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                               __middle, __last, __comp);
 }
 
 template <class _ExecutionPolicy, class _RandomAccessIterator>
@@ -793,12 +754,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _RandomAccess
 partial_sort_copy(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last,
                   _RandomAccessIterator __d_first, _RandomAccessIterator __d_last, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_partial_sort_copy(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __d_first, __d_last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator, _RandomAccessIterator>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator, _RandomAccessIterator>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __d_first);
+
+    return __pstl::__internal::__pattern_partial_sort_copy(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                           __first, __last, __d_first, __d_last, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _RandomAccessIterator>
@@ -815,11 +774,11 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Compare>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 is_sorted_until(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp)
 {
-    const _ForwardIterator __res = __pstl::__internal::__pattern_adjacent_find(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __pstl::__internal::__reorder_pred<_Compare>(__comp),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        /*first_semantic*/ false);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    const _ForwardIterator __res =
+        __pstl::__internal::__pattern_adjacent_find(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                    __last, __pstl::__internal::__reorder_pred<_Compare>(__comp),
+                                                    /*first_semantic*/ false);
     return __res == __last ? __last : std::next(__res);
 }
 
@@ -835,12 +794,10 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Compare>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, bool>
 is_sorted(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_adjacent_find(
-               std::forward<_ExecutionPolicy>(__exec), __first, __last,
-               __pstl::__internal::__reorder_pred<_Compare>(__comp),
-               __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-               __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-               /*or_semantic*/ true) == __last;
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_adjacent_find(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                       __last, __pstl::__internal::__reorder_pred<_Compare>(__comp),
+                                                       /*or_semantic*/ true) == __last;
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator>
@@ -858,12 +815,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 merge(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
       _ForwardIterator2 __last2, _ForwardIterator __d_first, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_merge(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __d_first, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                         _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                           _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2, __d_first);
+
+    return __pstl::__internal::__pattern_merge(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                               __last1, __first2, __last2, __d_first, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator>
@@ -880,10 +835,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 inplace_merge(_ExecutionPolicy&& __exec, _BidirectionalIterator __first, _BidirectionalIterator __middle,
               _BidirectionalIterator __last, _Compare __comp)
 {
-    __pstl::__internal::__pattern_inplace_merge(
-        std::forward<_ExecutionPolicy>(__exec), __first, __middle, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _BidirectionalIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _BidirectionalIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_inplace_merge(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                __middle, __last, __comp);
 }
 
 template <class _ExecutionPolicy, class _BidirectionalIterator>
@@ -902,12 +857,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, bool>
 includes(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
          _ForwardIterator2 __last2, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_includes(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+
+    return __pstl::__internal::__pattern_includes(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                                  __last1, __first2, __last2, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
@@ -926,12 +879,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 set_union(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2,
           _ForwardIterator2 __last2, _ForwardIterator __result, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_set_union(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                         _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                           _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2, __result);
+
+    return __pstl::__internal::__pattern_set_union(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1,
+                                                   __last1, __first2, __last2, __result, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator>
@@ -951,12 +902,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 set_intersection(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                  _ForwardIterator2 __first2, _ForwardIterator2 __last2, _ForwardIterator __result, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_set_intersection(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                         _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                           _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2, __result);
+
+    return __pstl::__internal::__pattern_set_intersection(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                          __first1, __last1, __first2, __last2, __result, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator>
@@ -976,12 +925,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 set_difference(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                _ForwardIterator2 __first2, _ForwardIterator2 __last2, _ForwardIterator __result, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_set_difference(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                         _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                           _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2, __result);
+
+    return __pstl::__internal::__pattern_set_difference(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                        __first1, __last1, __first2, __last2, __result, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator>
@@ -1002,12 +949,10 @@ set_symmetric_difference(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1,
                          _ForwardIterator2 __first2, _ForwardIterator2 __last2, _ForwardIterator __result,
                          _Compare __comp)
 {
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2, __result);
+
     return __pstl::__internal::__pattern_set_symmetric_difference(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                         _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2,
-                                                           _ForwardIterator>(__exec));
+        __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __result, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _ForwardIterator>
@@ -1024,10 +969,10 @@ template <class _ExecutionPolicy, class _RandomAccessIterator, class _Compare>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _RandomAccessIterator>
 is_heap_until(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_is_heap_until(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    return __pstl::__internal::__pattern_is_heap_until(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                       __last, __comp);
 }
 
 template <class _ExecutionPolicy, class _RandomAccessIterator>
@@ -1059,10 +1004,9 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Compare>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardIterator>
 min_element(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_min_element(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_min_element(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                     __last, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator>
@@ -1094,10 +1038,9 @@ template <class _ExecutionPolicy, class _ForwardIterator, class _Compare>
 __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, std::pair<_ForwardIterator, _ForwardIterator>>
 minmax_element(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_minmax_element(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_minmax_element(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                        __last, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator>
@@ -1115,10 +1058,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, void>
 nth_element(_ExecutionPolicy&& __exec, _RandomAccessIterator __first, _RandomAccessIterator __nth,
             _RandomAccessIterator __last, _Compare __comp)
 {
-    __pstl::__internal::__pattern_nth_element(
-        std::forward<_ExecutionPolicy>(__exec), __first, __nth, __last, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _RandomAccessIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    __pstl::__internal::__pattern_nth_element(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __nth,
+                                              __last, __comp);
 }
 
 template <class _ExecutionPolicy, class _RandomAccessIterator>
@@ -1137,12 +1080,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, bool>
 lexicographical_compare(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                         _ForwardIterator2 __first2, _ForwardIterator2 __last2, _Compare __comp)
 {
-    return __pstl::__internal::__pattern_lexicographical_compare(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __last2, __comp,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+
+    return __pstl::__internal::__pattern_lexicographical_compare(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                                 __first1, __last1, __first2, __last2, __comp);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
diff --git a/libstdc++-v3/include/pstl/glue_execution_defs.h b/libstdc++-v3/include/pstl/glue_execution_defs.h
index 24ede331ac1..e7b36c7dd10 100644
--- a/libstdc++-v3/include/pstl/glue_execution_defs.h
+++ b/libstdc++-v3/include/pstl/glue_execution_defs.h
@@ -18,10 +18,10 @@ namespace std
 {
 // Type trait
 using __pstl::execution::is_execution_policy;
-#if _PSTL_CPP14_VARIABLE_TEMPLATES_PRESENT
-#    if __INTEL_COMPILER
-template <class T>
-constexpr bool is_execution_policy_v = is_execution_policy<T>::value;
+#if defined(_PSTL_CPP14_VARIABLE_TEMPLATES_PRESENT)
+#    if defined(__INTEL_COMPILER)
+template <class _Tp>
+constexpr bool is_execution_policy_v = is_execution_policy<_Tp>::value;
 #    else
 using __pstl::execution::is_execution_policy_v;
 #    endif
diff --git a/libstdc++-v3/include/pstl/glue_memory_impl.h b/libstdc++-v3/include/pstl/glue_memory_impl.h
index ea462e9ccdd..7850605f94a 100644
--- a/libstdc++-v3/include/pstl/glue_memory_impl.h
+++ b/libstdc++-v3/include/pstl/glue_memory_impl.h
@@ -13,6 +13,8 @@
 #include "utils.h"
 #include "algorithm_fwd.h"
 
+#include "execution_impl.h"
+
 namespace std
 {
 
@@ -27,28 +29,25 @@ uninitialized_copy(_ExecutionPolicy&& __exec, _InputIterator __first, _InputIter
     typedef typename iterator_traits<_InputIterator>::reference _ReferenceType1;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType2;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__invoke_if_else(
         std::integral_constant < bool, std::is_trivial<_ValueType1>::value&& std::is_trivial<_ValueType2>::value > (),
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk2_brick(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
-                [__is_vector](_InputIterator __begin, _InputIterator __end, _ForwardIterator __res) {
-                    return __pstl::__internal::__brick_copy(__begin, __end, __res, __is_vector);
-                },
-                __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+                [](_InputIterator __begin, _InputIterator __end, _ForwardIterator __res)
+                { return __pstl::__internal::__brick_copy(__begin, __end, __res, __is_vector{}); });
         },
-        [&]() {
-            return __pstl::__internal::__pattern_walk2(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                                       __result,
-                                                       [](_ReferenceType1 __val1, _ReferenceType2 __val2) {
-                                                           ::new (std::addressof(__val2)) _ValueType2(__val1);
-                                                       },
-                                                       __is_vector, __is_parallel);
+        [&]()
+        {
+            return __pstl::__internal::__pattern_walk2(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                       __last, __result,
+                                                       [](_ReferenceType1 __val1, _ReferenceType2 __val2)
+                                                       { ::new (std::addressof(__val2)) _ValueType2(__val1); });
         });
 }
 
@@ -61,27 +60,25 @@ uninitialized_copy_n(_ExecutionPolicy&& __exec, _InputIterator __first, _Size __
     typedef typename iterator_traits<_InputIterator>::reference _ReferenceType1;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType2;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__invoke_if_else(
         std::integral_constant < bool, std::is_trivial<_ValueType1>::value&& std::is_trivial<_ValueType2>::value > (),
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk2_brick_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
-                [__is_vector](_InputIterator __begin, _Size __sz, _ForwardIterator __res) {
-                    return __pstl::__internal::__brick_copy_n(__begin, __sz, __res, __is_vector);
-                },
-                __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
+                [](_InputIterator __begin, _Size __sz, _ForwardIterator __res)
+                { return __pstl::__internal::__brick_copy_n(__begin, __sz, __res, __is_vector{}); });
         },
-        [&]() {
-            return __pstl::__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);
+        [&]()
+        {
+            return __pstl::__internal::__pattern_walk2_n(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                         __first, __n, __result,
+                                                         [](_ReferenceType1 __val1, _ReferenceType2 __val2)
+                                                         { ::new (std::addressof(__val2)) _ValueType2(__val1); });
         });
 }
 
@@ -96,28 +93,25 @@ uninitialized_move(_ExecutionPolicy&& __exec, _InputIterator __first, _InputIter
     typedef typename iterator_traits<_InputIterator>::reference _ReferenceType1;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType2;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__invoke_if_else(
         std::integral_constant < bool, std::is_trivial<_ValueType1>::value&& std::is_trivial<_ValueType2>::value > (),
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk2_brick(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
-                [__is_vector](_InputIterator __begin, _InputIterator __end, _ForwardIterator __res) {
-                    return __pstl::__internal::__brick_copy(__begin, __end, __res, __is_vector);
-                },
-                __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+                [](_InputIterator __begin, _InputIterator __end, _ForwardIterator __res)
+                { return __pstl::__internal::__brick_copy(__begin, __end, __res, __is_vector{}); });
         },
-        [&]() {
+        [&]()
+        {
             return __pstl::__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);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last, __result,
+                [](_ReferenceType1 __val1, _ReferenceType2 __val2)
+                { ::new (std::addressof(__val2)) _ValueType2(std::move(__val1)); });
         });
 }
 
@@ -130,28 +124,25 @@ uninitialized_move_n(_ExecutionPolicy&& __exec, _InputIterator __first, _Size __
     typedef typename iterator_traits<_InputIterator>::reference _ReferenceType1;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType2;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _InputIterator, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__invoke_if_else(
         std::integral_constant < bool, std::is_trivial<_ValueType1>::value&& std::is_trivial<_ValueType2>::value > (),
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk2_brick_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
-                [__is_vector](_InputIterator __begin, _Size __sz, _ForwardIterator __res) {
-                    return __pstl::__internal::__brick_copy_n(__begin, __sz, __res, __is_vector);
-                },
-                __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
+                [](_InputIterator __begin, _Size __sz, _ForwardIterator __res)
+                { return __pstl::__internal::__brick_copy_n(__begin, __sz, __res, __is_vector{}); });
         },
-        [&]() {
-            return __pstl::__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);
+        [&]()
+        {
+            return __pstl::__internal::__pattern_walk2_n(
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n, __result,
+                [](_ReferenceType1 __val1, _ReferenceType2 __val2)
+                { ::new (std::addressof(__val2)) _ValueType2(std::move(__val1)); });
         });
 }
 
@@ -164,26 +155,24 @@ uninitialized_fill(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Forward
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     __pstl::__internal::__invoke_if_else(
         std::is_arithmetic<_ValueType>(),
-        [&]() {
+        [&]()
+        {
             __pstl::__internal::__pattern_walk_brick(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                [&__value, &__is_vector](_ForwardIterator __begin, _ForwardIterator __end) {
-                    __pstl::__internal::__brick_fill(__begin, __end, _ValueType(__value), __is_vector);
-                },
-                __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                [&__value](_ForwardIterator __begin, _ForwardIterator __end)
+                { __pstl::__internal::__brick_fill(__begin, __end, _ValueType(__value), __is_vector{}); });
         },
-        [&]() {
-            __pstl::__internal::__pattern_walk1(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                [&__value](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(__value); }, __is_vector,
-                __is_parallel);
+        [&]()
+        {
+            __pstl::__internal::__pattern_walk1(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                [&__value](_ReferenceType __val)
+                                                { ::new (std::addressof(__val)) _ValueType(__value); });
         });
 }
 
@@ -194,26 +183,24 @@ uninitialized_fill_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__invoke_if_else(
         std::is_arithmetic<_ValueType>(),
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk_brick_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n,
-                [&__value, &__is_vector](_ForwardIterator __begin, _Size __count) {
-                    return __pstl::__internal::__brick_fill_n(__begin, __count, _ValueType(__value), __is_vector);
-                },
-                __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n,
+                [&__value](_ForwardIterator __begin, _Size __count)
+                { return __pstl::__internal::__brick_fill_n(__begin, __count, _ValueType(__value), __is_vector{}); });
         },
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk1_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n,
-                [&__value](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(__value); }, __is_vector,
-                __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n,
+                [&__value](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(__value); });
         });
 }
 
@@ -226,16 +213,15 @@ destroy(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
 
-    __pstl::__internal::__invoke_if_not(std::is_trivially_destructible<_ValueType>(), [&]() {
-        __pstl::__internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                            [](_ReferenceType __val) { __val.~_ValueType(); }, __is_vector,
-                                            __is_parallel);
-    });
+    __pstl::__internal::__invoke_if_not(std::is_trivially_destructible<_ValueType>(),
+                                        [&]()
+                                        {
+                                            __pstl::__internal::__pattern_walk1(
+                                                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                [](_ReferenceType __val) { __val.~_ValueType(); });
+                                        });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size>
@@ -245,17 +231,15 @@ destroy_n(_ExecutionPolicy&& __exec, _ForwardIterator __first, _Size __n)
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
 
     return __pstl::__internal::__invoke_if_else(
         std::is_trivially_destructible<_ValueType>(), [&]() { return std::next(__first, __n); },
-        [&]() {
-            return __pstl::__internal::__pattern_walk1_n(std::forward<_ExecutionPolicy>(__exec), __first, __n,
-                                                         [](_ReferenceType __val) { __val.~_ValueType(); }, __is_vector,
-                                                         __is_parallel);
+        [&]()
+        {
+            return __pstl::__internal::__pattern_walk1_n(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                         __first, __n,
+                                                         [](_ReferenceType __val) { __val.~_ValueType(); });
         });
 }
 
@@ -268,16 +252,15 @@ uninitialized_default_construct(_ExecutionPolicy&& __exec, _ForwardIterator __fi
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
 
-    __pstl::__internal::__invoke_if_not(std::is_trivial<_ValueType>(), [&]() {
-        __pstl::__internal::__pattern_walk1(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                            [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; },
-                                            __is_vector, __is_parallel);
-    });
+    __pstl::__internal::__invoke_if_not(std::is_trivial<_ValueType>(),
+                                        [&]()
+                                        {
+                                            __pstl::__internal::__pattern_walk1(
+                                                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; });
+                                        });
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Size>
@@ -287,17 +270,15 @@ uninitialized_default_construct_n(_ExecutionPolicy&& __exec, _ForwardIterator __
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
 
     return __pstl::__internal::__invoke_if_else(
         std::is_trivial<_ValueType>(), [&]() { return std::next(__first, __n); },
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk1_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n,
-                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; }, __is_vector, __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n,
+                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType; });
         });
 }
 
@@ -310,25 +291,24 @@ uninitialized_value_construct(_ExecutionPolicy&& __exec, _ForwardIterator __firs
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     __pstl::__internal::__invoke_if_else(
         std::is_trivial<_ValueType>(),
-        [&]() {
-            __pstl::__internal::__pattern_walk_brick(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                                     [__is_vector](_ForwardIterator __begin, _ForwardIterator __end) {
-                                                         __pstl::__internal::__brick_fill(__begin, __end, _ValueType(),
-                                                                                          __is_vector);
-                                                     },
-                                                     __is_parallel);
+        [&]()
+        {
+            __pstl::__internal::__pattern_walk_brick(
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                [](_ForwardIterator __begin, _ForwardIterator __end)
+                { __pstl::__internal::__brick_fill(__begin, __end, _ValueType(), __is_vector{}); });
         },
-        [&]() {
-            __pstl::__internal::__pattern_walk1(
-                std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(); }, __is_vector, __is_parallel);
+        [&]()
+        {
+            __pstl::__internal::__pattern_walk1(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                [](_ReferenceType __val)
+                                                { ::new (std::addressof(__val)) _ValueType(); });
         });
 }
 
@@ -339,25 +319,24 @@ uninitialized_value_construct_n(_ExecutionPolicy&& __exec, _ForwardIterator __fi
     typedef typename iterator_traits<_ForwardIterator>::value_type _ValueType;
     typedef typename iterator_traits<_ForwardIterator>::reference _ReferenceType;
 
-    const auto __is_parallel =
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
-    const auto __is_vector =
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec);
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+
+    using __is_vector = typename decltype(__dispatch_tag)::__is_vector;
 
     return __pstl::__internal::__invoke_if_else(
         std::is_trivial<_ValueType>(),
-        [&]() {
-            return __pstl::__internal::__pattern_walk_brick_n(std::forward<_ExecutionPolicy>(__exec), __first, __n,
-                                                              [__is_vector](_ForwardIterator __begin, _Size __count) {
-                                                                  return __pstl::__internal::__brick_fill_n(
-                                                                      __begin, __count, _ValueType(), __is_vector);
-                                                              },
-                                                              __is_parallel);
+        [&]()
+        {
+            return __pstl::__internal::__pattern_walk_brick_n(
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n,
+                [](_ForwardIterator __begin, _Size __count)
+                { return __pstl::__internal::__brick_fill_n(__begin, __count, _ValueType(), __is_vector{}); });
         },
-        [&]() {
+        [&]()
+        {
             return __pstl::__internal::__pattern_walk1_n(
-                std::forward<_ExecutionPolicy>(__exec), __first, __n,
-                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(); }, __is_vector, __is_parallel);
+                __dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __n,
+                [](_ReferenceType __val) { ::new (std::addressof(__val)) _ValueType(); });
         });
 }
 
diff --git a/libstdc++-v3/include/pstl/glue_numeric_impl.h b/libstdc++-v3/include/pstl/glue_numeric_impl.h
index 9db64b513a4..490175c8b83 100644
--- a/libstdc++-v3/include/pstl/glue_numeric_impl.h
+++ b/libstdc++-v3/include/pstl/glue_numeric_impl.h
@@ -54,14 +54,12 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _Tp>
 transform_reduce(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                  _ForwardIterator2 __first2, _Tp __init)
 {
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+
     typedef typename iterator_traits<_ForwardIterator1>::value_type _InputType;
-    return __pstl::__internal::__pattern_transform_reduce(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __init, std::plus<_InputType>(),
-        std::multiplies<_InputType>(),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    return __pstl::__internal::__pattern_transform_reduce(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                          __first1, __last1, __first2, __init, std::plus<_InputType>(),
+                                                          std::multiplies<_InputType>());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Tp, class _BinaryOperation1,
@@ -70,12 +68,10 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _Tp>
 transform_reduce(_ExecutionPolicy&& __exec, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                  _ForwardIterator2 __first2, _Tp __init, _BinaryOperation1 __binary_op1, _BinaryOperation2 __binary_op2)
 {
-    return __pstl::__internal::__pattern_transform_reduce(
-        std::forward<_ExecutionPolicy>(__exec), __first1, __last1, __first2, __init, __binary_op1, __binary_op2,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first1, __first2);
+    return __pstl::__internal::__pattern_transform_reduce(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                          __first1, __last1, __first2, __init, __binary_op1,
+                                                          __binary_op2);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation>
@@ -83,10 +79,9 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _Tp>
 transform_reduce(_ExecutionPolicy&& __exec, _ForwardIterator __first, _ForwardIterator __last, _Tp __init,
                  _BinaryOperation __binary_op, _UnaryOperation __unary_op)
 {
-    return __pstl::__internal::__pattern_transform_reduce(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __init, __binary_op, __unary_op,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator>(__exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first);
+    return __pstl::__internal::__pattern_transform_reduce(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                          __first, __last, __init, __binary_op, __unary_op);
 }
 
 // [exclusive.scan]
@@ -96,12 +91,12 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 exclusive_scan(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last,
                _ForwardIterator2 __result, _Tp __init)
 {
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
     using namespace __pstl;
-    return __internal::__pattern_transform_scan(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result, __pstl::__internal::__no_op(), __init,
-        std::plus<_Tp>(), /*inclusive=*/std::false_type(),
-        __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(__exec),
-        __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(__exec));
+    return __internal::__pattern_transform_scan(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                __result, __pstl::__internal::__no_op(), __init, std::plus<_Tp>(),
+                                                /*inclusive=*/std::false_type());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Tp, class _BinaryOperation>
@@ -109,12 +104,12 @@ __pstl::__internal::__enable_if_execution_policy<_ExecutionPolicy, _ForwardItera
 exclusive_scan(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _ForwardIterator1 __last,
                _ForwardIterator2 __result, _Tp __init, _BinaryOperation __binary_op)
 {
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
     using namespace __pstl;
-    return __internal::__pattern_transform_scan(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result, __pstl::__internal::__no_op(), __init,
-        __binary_op, /*inclusive=*/std::false_type(),
-        __internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(__exec),
-        __internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(__exec));
+    return __internal::__pattern_transform_scan(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                                __result, __pstl::__internal::__no_op(), __init, __binary_op,
+                                                /*inclusive=*/std::false_type());
 }
 
 // [inclusive.scan]
@@ -156,13 +151,11 @@ transform_exclusive_scan(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _
                          _ForwardIterator2 __result, _Tp __init, _BinaryOperation __binary_op,
                          _UnaryOperation __unary_op)
 {
-    return __pstl::__internal::__pattern_transform_scan(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result, __unary_op, __init, __binary_op,
-        /*inclusive=*/std::false_type(),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    return __pstl::__internal::__pattern_transform_scan(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                        __last, __result, __unary_op, __init, __binary_op,
+                                                        /*inclusive=*/std::false_type());
 }
 
 // [transform.inclusive.scan]
@@ -174,13 +167,11 @@ transform_inclusive_scan(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _
                          _ForwardIterator2 __result, _BinaryOperation __binary_op, _UnaryOperation __unary_op,
                          _Tp __init)
 {
-    return __pstl::__internal::__pattern_transform_scan(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __result, __unary_op, __init, __binary_op,
-        /*inclusive=*/std::true_type(),
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __result);
+
+    return __pstl::__internal::__pattern_transform_scan(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec), __first,
+                                                        __last, __result, __unary_op, __init, __binary_op,
+                                                        /*inclusive=*/std::true_type());
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _UnaryOperation,
@@ -213,12 +204,10 @@ adjacent_difference(_ExecutionPolicy&& __exec, _ForwardIterator1 __first, _Forwa
     if (__first == __last)
         return __d_first;
 
-    return __pstl::__internal::__pattern_adjacent_difference(
-        std::forward<_ExecutionPolicy>(__exec), __first, __last, __d_first, __op,
-        __pstl::__internal::__is_vectorization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec),
-        __pstl::__internal::__is_parallelization_preferred<_ExecutionPolicy, _ForwardIterator1, _ForwardIterator2>(
-            __exec));
+    auto __dispatch_tag = __pstl::__internal::__select_backend(__exec, __first, __d_first);
+
+    return __pstl::__internal::__pattern_adjacent_difference(__dispatch_tag, std::forward<_ExecutionPolicy>(__exec),
+                                                             __first, __last, __d_first, __op);
 }
 
 template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2>
diff --git a/libstdc++-v3/include/pstl/memory_impl.h b/libstdc++-v3/include/pstl/memory_impl.h
index be5a103f93b..8cb32d043e5 100644
--- a/libstdc++-v3/include/pstl/memory_impl.h
+++ b/libstdc++-v3/include/pstl/memory_impl.h
@@ -36,13 +36,13 @@ __brick_uninitialized_move(_ForwardIterator __first, _ForwardIterator __last, _O
     return __result;
 }
 
-template <typename _ForwardIterator, typename _OutputIterator>
+template <typename _RandomAccessIterator, typename _OutputIterator>
 _OutputIterator
-__brick_uninitialized_move(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
+__brick_uninitialized_move(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
                            /*vector=*/std::true_type) noexcept
 {
     using __ValueType = typename std::iterator_traits<_OutputIterator>::value_type;
-    using _ReferenceType1 = typename std::iterator_traits<_ForwardIterator>::reference;
+    using _ReferenceType1 = typename std::iterator_traits<_RandomAccessIterator>::reference;
     using _ReferenceType2 = typename std::iterator_traits<_OutputIterator>::reference;
 
     return __unseq_backend::__simd_walk_2(
@@ -60,12 +60,12 @@ __brick_destroy(_Iterator __first, _Iterator __last, /*vector*/ std::false_type)
         __first->~_ValueType();
 }
 
-template <typename _Iterator>
+template <typename _RandomAccessIterator>
 void
-__brick_destroy(_Iterator __first, _Iterator __last, /*vector*/ std::true_type) noexcept
+__brick_destroy(_RandomAccessIterator __first, _RandomAccessIterator __last, /*vector*/ std::true_type) noexcept
 {
-    using _ValueType = typename std::iterator_traits<_Iterator>::value_type;
-    using _ReferenceType = typename std::iterator_traits<_Iterator>::reference;
+    using _ValueType = typename std::iterator_traits<_RandomAccessIterator>::value_type;
+    using _ReferenceType = typename std::iterator_traits<_RandomAccessIterator>::reference;
 
     __unseq_backend::__simd_walk_1(__first, __last - __first, [](_ReferenceType __x) { __x.~_ValueType(); });
 }
@@ -87,13 +87,13 @@ __brick_uninitialized_copy(_ForwardIterator __first, _ForwardIterator __last, _O
     return __result;
 }
 
-template <typename _ForwardIterator, typename _OutputIterator>
+template <typename _RandomAccessIterator, typename _OutputIterator>
 _OutputIterator
-__brick_uninitialized_copy(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
+__brick_uninitialized_copy(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
                            /*vector=*/std::true_type) noexcept
 {
     using __ValueType = typename std::iterator_traits<_OutputIterator>::value_type;
-    using _ReferenceType1 = typename std::iterator_traits<_ForwardIterator>::reference;
+    using _ReferenceType1 = typename std::iterator_traits<_RandomAccessIterator>::reference;
     using _ReferenceType2 = typename std::iterator_traits<_OutputIterator>::reference;
 
     return __unseq_backend::__simd_walk_2(
diff --git a/libstdc++-v3/include/pstl/numeric_fwd.h b/libstdc++-v3/include/pstl/numeric_fwd.h
index d92cacb7644..e98e3b1e12b 100644
--- a/libstdc++-v3/include/pstl/numeric_fwd.h
+++ b/libstdc++-v3/include/pstl/numeric_fwd.h
@@ -22,9 +22,10 @@ namespace __internal
 // transform_reduce (version with two binary functions, according to draft N4659)
 //------------------------------------------------------------------------
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Tp, class _BinaryOperation1, class _BinaryOperation2>
-_Tp __brick_transform_reduce(_ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Tp, _BinaryOperation1,
-                             _BinaryOperation2,
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Tp, class _BinaryOperation1,
+          class _BinaryOperation2>
+_Tp __brick_transform_reduce(_RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2, _Tp,
+                             _BinaryOperation1, _BinaryOperation2,
                              /*__is_vector=*/std::true_type) noexcept;
 
 template <class _ForwardIterator1, class _ForwardIterator2, class _Tp, class _BinaryOperation1, class _BinaryOperation2>
@@ -32,45 +33,41 @@ _Tp __brick_transform_reduce(_ForwardIterator1, _ForwardIterator1, _ForwardItera
                              _BinaryOperation2,
                              /*__is_vector=*/std::false_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Tp, class _BinaryOperation1,
-          class _BinaryOperation2, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Tp,
+          class _BinaryOperation1, class _BinaryOperation2>
 _Tp
-__pattern_transform_reduce(_ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Tp,
-                           _BinaryOperation1, _BinaryOperation2, _IsVector,
-                           /*is_parallel=*/std::false_type) noexcept;
+__pattern_transform_reduce(_Tag, _ExecutionPolicy&&, _ForwardIterator1, _ForwardIterator1, _ForwardIterator2, _Tp,
+                           _BinaryOperation1, _BinaryOperation2) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Tp,
-          class _BinaryOperation1, class _BinaryOperation2, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Tp, class _BinaryOperation1, class _BinaryOperation2>
 _Tp
-__pattern_transform_reduce(_ExecutionPolicy&&, _RandomAccessIterator1, _RandomAccessIterator1, _RandomAccessIterator2,
-                           _Tp, _BinaryOperation1, _BinaryOperation2, _IsVector __is_vector,
-                           /*is_parallel=*/std::true_type);
+__pattern_transform_reduce(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator1,
+                           _RandomAccessIterator1, _RandomAccessIterator2, _Tp, _BinaryOperation1, _BinaryOperation2);
 
 //------------------------------------------------------------------------
 // transform_reduce (version with unary and binary functions)
 //------------------------------------------------------------------------
 
-template <class _ForwardIterator, class _Tp, class _UnaryOperation, class _BinaryOperation>
-_Tp __brick_transform_reduce(_ForwardIterator, _ForwardIterator, _Tp, _BinaryOperation, _UnaryOperation,
+template <class _RandomAccessIterator, class _Tp, class _UnaryOperation, class _BinaryOperation>
+_Tp __brick_transform_reduce(_RandomAccessIterator, _RandomAccessIterator, _Tp, _BinaryOperation, _UnaryOperation,
                              /*is_vector=*/std::true_type) noexcept;
 
 template <class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation>
 _Tp __brick_transform_reduce(_ForwardIterator, _ForwardIterator, _Tp, _BinaryOperation, _UnaryOperation,
                              /*is_vector=*/std::false_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation,
+          class _UnaryOperation>
 _Tp
-__pattern_transform_reduce(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Tp, _BinaryOperation,
-                           _UnaryOperation, _IsVector,
-                           /*is_parallel=*/std::false_type) noexcept;
+__pattern_transform_reduce(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Tp, _BinaryOperation,
+                           _UnaryOperation) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Tp, class _BinaryOperation,
+          class _UnaryOperation>
 _Tp
-__pattern_transform_reduce(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _Tp, _BinaryOperation,
-                           _UnaryOperation, _IsVector,
-                           /*is_parallel=*/std::true_type);
+__pattern_transform_reduce(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                           _Tp, _BinaryOperation, _UnaryOperation);
 
 //------------------------------------------------------------------------
 // transform_exclusive_scan
@@ -83,29 +80,28 @@ std::pair<_OutputIterator, _Tp> __brick_transform_scan(_ForwardIterator, _Forwar
                                                        _UnaryOperation, _Tp, _BinaryOperation,
                                                        /*Inclusive*/ std::false_type) noexcept;
 
-template <class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp, class _BinaryOperation>
-std::pair<_OutputIterator, _Tp> __brick_transform_scan(_ForwardIterator, _ForwardIterator, _OutputIterator,
+template <class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp, class _BinaryOperation>
+std::pair<_OutputIterator, _Tp> __brick_transform_scan(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator,
                                                        _UnaryOperation, _Tp, _BinaryOperation,
                                                        /*Inclusive*/ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
-          class _BinaryOperation, class _Inclusive, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryOperation,
+          class _Tp, class _BinaryOperation, class _Inclusive>
 _OutputIterator
-__pattern_transform_scan(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator, _UnaryOperation, _Tp,
-                         _BinaryOperation, _Inclusive, _IsVector,
-                         /*is_parallel=*/std::false_type) noexcept;
+__pattern_transform_scan(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator, _UnaryOperation,
+                         _Tp, _BinaryOperation, _Inclusive) noexcept;
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
-          class _BinaryOperation, class _Inclusive, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator,
+          class _UnaryOperation, class _Tp, class _BinaryOperation, class _Inclusive>
 typename std::enable_if<!std::is_floating_point<_Tp>::value, _OutputIterator>::type
-__pattern_transform_scan(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _OutputIterator,
-                         _UnaryOperation, _Tp, _BinaryOperation, _Inclusive, _IsVector, /*is_parallel=*/std::true_type);
+__pattern_transform_scan(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&&, _RandomAccessIterator,
+                         _RandomAccessIterator, _OutputIterator, _UnaryOperation, _Tp, _BinaryOperation, _Inclusive);
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
-          class _BinaryOperation, class _Inclusive, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator,
+          class _UnaryOperation, class _Tp, class _BinaryOperation, class _Inclusive>
 typename std::enable_if<std::is_floating_point<_Tp>::value, _OutputIterator>::type
-__pattern_transform_scan(_ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator, _OutputIterator,
-                         _UnaryOperation, _Tp, _BinaryOperation, _Inclusive, _IsVector, /*is_parallel=*/std::true_type);
+__pattern_transform_scan(__parallel_tag<_IsVector>, _ExecutionPolicy&&, _RandomAccessIterator, _RandomAccessIterator,
+                         _OutputIterator, _UnaryOperation, _Tp, _BinaryOperation, _Inclusive);
 
 //------------------------------------------------------------------------
 // adjacent_difference
@@ -115,20 +111,20 @@ template <class _ForwardIterator, class _OutputIterator, class _BinaryOperation>
 _OutputIterator __brick_adjacent_difference(_ForwardIterator, _ForwardIterator, _OutputIterator, _BinaryOperation,
                                             /*is_vector*/ std::false_type) noexcept;
 
-template <class _ForwardIterator, class _OutputIterator, class _BinaryOperation>
-_OutputIterator __brick_adjacent_difference(_ForwardIterator, _ForwardIterator, _OutputIterator, _BinaryOperation,
+template <class _RandomAccessIterator, class _OutputIterator, class _BinaryOperation>
+_OutputIterator __brick_adjacent_difference(_RandomAccessIterator, _RandomAccessIterator, _OutputIterator,
+                                            _BinaryOperation,
                                             /*is_vector*/ std::true_type) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryOperation,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryOperation>
 _OutputIterator
-__pattern_adjacent_difference(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator, _BinaryOperation,
-                              _IsVector, /*is_parallel*/ std::false_type) noexcept;
+__pattern_adjacent_difference(_Tag, _ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator,
+                              _BinaryOperation) noexcept;
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryOperation,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator,
+          class _BinaryOperation>
 _OutputIterator
-__pattern_adjacent_difference(_ExecutionPolicy&&, _ForwardIterator, _ForwardIterator, _OutputIterator, _BinaryOperation,
+__pattern_adjacent_difference(_ExecutionPolicy&&, _OutputIterator, _OutputIterator, _OutputIterator, _BinaryOperation,
                               _IsVector, /*is_parallel*/ std::true_type);
 
 } // namespace __internal
diff --git a/libstdc++-v3/include/pstl/numeric_impl.h b/libstdc++-v3/include/pstl/numeric_impl.h
index 0b02bcf0789..7ba83eeb714 100644
--- a/libstdc++-v3/include/pstl/numeric_impl.h
+++ b/libstdc++-v3/include/pstl/numeric_impl.h
@@ -38,50 +38,56 @@ __brick_transform_reduce(_ForwardIterator1 __first1, _ForwardIterator1 __last1,
     return std::inner_product(__first1, __last1, __first2, __init, __binary_op1, __binary_op2);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _Tp, class _BinaryOperation1, class _BinaryOperation2>
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _Tp, class _BinaryOperation1,
+          class _BinaryOperation2>
 _Tp
-__brick_transform_reduce(_ForwardIterator1 __first1, _ForwardIterator1 __last1, _ForwardIterator2 __first2, _Tp __init,
-                         _BinaryOperation1 __binary_op1, _BinaryOperation2 __binary_op2,
+__brick_transform_reduce(_RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
+                         _RandomAccessIterator2 __first2, _Tp __init, _BinaryOperation1 __binary_op1,
+                         _BinaryOperation2 __binary_op2,
                          /*is_vector=*/std::true_type) noexcept
 {
-    typedef typename std::iterator_traits<_ForwardIterator1>::difference_type _DifferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::difference_type _DifferenceType;
     return __unseq_backend::__simd_transform_reduce(
         __last1 - __first1, __init, __binary_op1,
         [=, &__binary_op2](_DifferenceType __i) { return __binary_op2(__first1[__i], __first2[__i]); });
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Tp, class _BinaryOperation1,
-          class _BinaryOperation2, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator1, class _ForwardIterator2, class _Tp,
+          class _BinaryOperation1, class _BinaryOperation2>
 _Tp
-__pattern_transform_reduce(_ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
+__pattern_transform_reduce(_Tag, _ExecutionPolicy&&, _ForwardIterator1 __first1, _ForwardIterator1 __last1,
                            _ForwardIterator2 __first2, _Tp __init, _BinaryOperation1 __binary_op1,
-                           _BinaryOperation2 __binary_op2, _IsVector __is_vector,
-                           /*is_parallel=*/std::false_type) noexcept
+                           _BinaryOperation2 __binary_op2) noexcept
 {
-    return __brick_transform_reduce(__first1, __last1, __first2, __init, __binary_op1, __binary_op2, __is_vector);
+    return __brick_transform_reduce(__first1, __last1, __first2, __init, __binary_op1, __binary_op2,
+                                    typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2, class _Tp,
-          class _BinaryOperation1, class _BinaryOperation2, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _Tp, class _BinaryOperation1, class _BinaryOperation2>
 _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)
+__pattern_transform_reduce(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator1 __first1,
+                           _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _Tp __init,
+                           _BinaryOperation1 __binary_op1, _BinaryOperation2 __binary_op2)
 {
-    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 __internal::__brick_transform_reduce(__i, __j, __first2 + (__i - __first1), __init, __binary_op1,
-                                                            __binary_op2, __is_vector);
-            });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __par_backend::__parallel_transform_reduce(
+                __backend_tag{}, 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](_RandomAccessIterator1 __i, _RandomAccessIterator1 __j,
+                                                                 _Tp __init) -> _Tp
+                {
+                    return __internal::__brick_transform_reduce(__i, __j, __first2 + (__i - __first1), __init,
+                                                                __binary_op1, __binary_op2, _IsVector{});
+                });
+        });
 }
 
 //------------------------------------------------------------------------
@@ -96,42 +102,47 @@ __brick_transform_reduce(_ForwardIterator __first, _ForwardIterator __last, _Tp
     return std::transform_reduce(__first, __last, __init, __binary_op, __unary_op);
 }
 
-template <class _ForwardIterator, class _Tp, class _UnaryOperation, class _BinaryOperation>
+template <class _RandomAccessIterator, class _Tp, class _UnaryOperation, class _BinaryOperation>
 _Tp
-__brick_transform_reduce(_ForwardIterator __first, _ForwardIterator __last, _Tp __init, _BinaryOperation __binary_op,
-                         _UnaryOperation __unary_op, /*is_vector=*/std::true_type) noexcept
+__brick_transform_reduce(_RandomAccessIterator __first, _RandomAccessIterator __last, _Tp __init,
+                         _BinaryOperation __binary_op, _UnaryOperation __unary_op,
+                         /*is_vector=*/std::true_type) noexcept
 {
-    typedef typename std::iterator_traits<_ForwardIterator>::difference_type _DifferenceType;
+    typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
     return __unseq_backend::__simd_transform_reduce(
         __last - __first, __init, __binary_op,
         [=, &__unary_op](_DifferenceType __i) { return __unary_op(__first[__i]); });
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation,
+          class _UnaryOperation>
 _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
+__pattern_transform_reduce(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last, _Tp __init,
+                           _BinaryOperation __binary_op, _UnaryOperation __unary_op) noexcept
 {
-    return __internal::__brick_transform_reduce(__first, __last, __init, __binary_op, __unary_op, __is_vector);
+    return __internal::__brick_transform_reduce(__first, __last, __init, __binary_op, __unary_op,
+                                                typename _Tag::__is_vector{});
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _Tp, class _BinaryOperation, class _UnaryOperation,
-          class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _Tp, class _BinaryOperation,
+          class _UnaryOperation>
 _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)
+__pattern_transform_reduce(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                           _RandomAccessIterator __last, _Tp __init, _BinaryOperation __binary_op,
+                           _UnaryOperation __unary_op)
 {
-    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 __internal::__brick_transform_reduce(__i, __j, __init, __binary_op, __unary_op, __is_vector);
-            });
-    });
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
+    return __internal::__except_handler(
+        [&]()
+        {
+            return __par_backend::__parallel_transform_reduce(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                [__unary_op](_RandomAccessIterator __i) mutable { return __unary_op(*__i); }, __init, __binary_op,
+                [__unary_op, __binary_op](_RandomAccessIterator __i, _RandomAccessIterator __j, _Tp __init) {
+                    return __internal::__brick_transform_reduce(__i, __j, __init, __binary_op, __unary_op, _IsVector{});
+                });
+        });
 }
 
 //------------------------------------------------------------------------
@@ -157,9 +168,9 @@ __brick_transform_scan(_ForwardIterator __first, _ForwardIterator __last, _Outpu
 }
 
 // Inclusive form
-template <class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp, class _BinaryOperation>
+template <class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp, class _BinaryOperation>
 std::pair<_OutputIterator, _Tp>
-__brick_transform_scan(_ForwardIterator __first, _ForwardIterator __last, _OutputIterator __result,
+__brick_transform_scan(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
                        _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op,
                        /*Inclusive*/ std::true_type, /*is_vector=*/std::false_type) noexcept
 {
@@ -179,14 +190,14 @@ using is_arithmetic_udop = std::integral_constant<bool, std::is_arithmetic<_Tp>:
 
 // [restriction] - T shall be DefaultConstructible.
 // [violation] - default ctor of T shall set the identity value for binary_op.
-template <class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp, class _BinaryOperation,
+template <class _RandomAccessIterator, 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,
+__brick_transform_scan(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
                        _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op, _Inclusive,
                        /*is_vector=*/std::true_type) noexcept
 {
-#if (_PSTL_UDS_PRESENT)
+#if defined(_PSTL_UDS_PRESENT)
     return __unseq_backend::__simd_scan(__first, __last - __first, __result, __unary_op, __init, __binary_op,
                                         _Inclusive());
 #else
@@ -196,10 +207,10 @@ __brick_transform_scan(_ForwardIterator __first, _ForwardIterator __last, _Outpu
 #endif
 }
 
-template <class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp, class _BinaryOperation,
+template <class _RandomAccessIterator, 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,
+__brick_transform_scan(_RandomAccessIterator __first, _RandomAccessIterator __last, _OutputIterator __result,
                        _UnaryOperation __unary_op, _Tp __init, _BinaryOperation __binary_op, _Inclusive,
                        /*is_vector=*/std::true_type) noexcept
 {
@@ -207,55 +218,62 @@ __brick_transform_scan(_ForwardIterator __first, _ForwardIterator __last, _Outpu
                                               /*is_vector=*/std::false_type());
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
-          class _BinaryOperation, class _Inclusive, class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _UnaryOperation,
+          class _Tp, class _BinaryOperation, class _Inclusive>
 _OutputIterator
-__pattern_transform_scan(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+__pattern_transform_scan(_Tag, _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
+                         _Inclusive) noexcept
 {
     return __internal::__brick_transform_scan(__first, __last, __result, __unary_op, __init, __binary_op, _Inclusive(),
-                                              __is_vector)
+                                              typename _Tag::__is_vector{})
         .first;
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
-          class _BinaryOperation, class _Inclusive, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator,
+          class _UnaryOperation, class _Tp, class _BinaryOperation, class _Inclusive>
 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)
+__pattern_transform_scan(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                         _RandomAccessIterator __last, _OutputIterator __result, _UnaryOperation __unary_op, _Tp __init,
+                         _BinaryOperation __binary_op, _Inclusive)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
 
-    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 __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 __internal::__brick_transform_scan(__first + __i, __first + __j, __result + __i, __unary_op,
-                                                          __init, __binary_op, _Inclusive(), __is_vector)
-                    .second;
-            });
-        return __result + (__last - __first);
-    });
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_transform_scan(
+                __backend_tag{}, 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 __internal::__brick_transform_reduce(__first + __i, __first + __j, __init, __binary_op,
+                                                                __unary_op,
+                                                                /*__is_vector*/ std::false_type());
+                },
+                [__first, __unary_op, __binary_op, __result](_DifferenceType __i, _DifferenceType __j, _Tp __init)
+                {
+                    return __internal::__brick_transform_scan(__first + __i, __first + __j, __result + __i, __unary_op,
+                                                              __init, __binary_op, _Inclusive(), _IsVector{})
+                        .second;
+                });
+            return __result + (__last - __first);
+        });
 }
 
-template <class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator, class _UnaryOperation, class _Tp,
-          class _BinaryOperation, class _Inclusive, class _IsVector>
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator, class _OutputIterator,
+          class _UnaryOperation, class _Tp, class _BinaryOperation, class _Inclusive>
 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)
+__pattern_transform_scan(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec, _RandomAccessIterator __first,
+                         _RandomAccessIterator __last, _OutputIterator __result, _UnaryOperation __unary_op, _Tp __init,
+                         _BinaryOperation __binary_op, _Inclusive)
 {
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
+
     typedef typename std::iterator_traits<_RandomAccessIterator>::difference_type _DifferenceType;
     _DifferenceType __n = __last - __first;
 
@@ -263,26 +281,31 @@ __pattern_transform_scan(_ExecutionPolicy&& __exec, _RandomAccessIterator __firs
     {
         return __result;
     }
-    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 __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);
-                                        }) -
-                         1);
-            },
-            [](_Tp) {});
-        return __result + (__last - __first);
-    });
+    return __internal::__except_handler(
+        [&]()
+        {
+            __par_backend::__parallel_strict_scan(
+                __backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __n, __init,
+                [__first, __unary_op, __binary_op, __result](_DifferenceType __i, _DifferenceType __len)
+                {
+                    return __internal::__brick_transform_scan(__first + __i, __first + (__i + __len), __result + __i,
+                                                              __unary_op, _Tp{}, __binary_op, _Inclusive(), _IsVector{})
+                        .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);
+                                            }) -
+                             1);
+                },
+                [](_Tp) {});
+            return __result + (__last - __first);
+        });
 }
 
 //------------------------------------------------------------------------
@@ -297,15 +320,16 @@ __brick_adjacent_difference(_ForwardIterator __first, _ForwardIterator __last, _
     return std::adjacent_difference(__first, __last, __d_first, __op);
 }
 
-template <class _ForwardIterator1, class _ForwardIterator2, class _BinaryOperation>
-_ForwardIterator2
-__brick_adjacent_difference(_ForwardIterator1 __first, _ForwardIterator1 __last, _ForwardIterator2 __d_first,
-			    _BinaryOperation __op, /*is_vector=*/std::true_type) noexcept
+template <class _RandomAccessIterator1, class _RandomAccessIterator2, class _BinaryOperation>
+_RandomAccessIterator2
+__brick_adjacent_difference(_RandomAccessIterator1 __first, _RandomAccessIterator1 __last,
+                            _RandomAccessIterator2 __d_first, _BinaryOperation __op,
+                            /*is_vector=*/std::true_type) noexcept
 {
     _PSTL_ASSERT(__first != __last);
 
-    typedef typename std::iterator_traits<_ForwardIterator1>::reference _ReferenceType1;
-    typedef typename std::iterator_traits<_ForwardIterator2>::reference _ReferenceType2;
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::reference _ReferenceType1;
+    typedef typename std::iterator_traits<_RandomAccessIterator2>::reference _ReferenceType2;
 
     auto __n = __last - __first;
     *__d_first = *__first;
@@ -314,37 +338,38 @@ __brick_adjacent_difference(_ForwardIterator1 __first, _ForwardIterator1 __last,
         [&__op](_ReferenceType1 __x, _ReferenceType1 __y, _ReferenceType2 __z) { __z = __op(__x, __y); });
 }
 
-template <class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryOperation,
-          class _IsVector>
+template <class _Tag, class _ExecutionPolicy, class _ForwardIterator, class _OutputIterator, class _BinaryOperation>
 _OutputIterator
-__pattern_adjacent_difference(_ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
-                              _OutputIterator __d_first, _BinaryOperation __op, _IsVector __is_vector,
-                              /*is_parallel*/ std::false_type) noexcept
+__pattern_adjacent_difference(_Tag, _ExecutionPolicy&&, _ForwardIterator __first, _ForwardIterator __last,
+                              _OutputIterator __d_first, _BinaryOperation __op) noexcept
 {
-    return __internal::__brick_adjacent_difference(__first, __last, __d_first, __op, __is_vector);
+    return __internal::__brick_adjacent_difference(__first, __last, __d_first, __op, typename _Tag::__is_vector{});
 }
 
-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)
+template <class _IsVector, class _ExecutionPolicy, class _RandomAccessIterator1, class _RandomAccessIterator2,
+          class _BinaryOperation>
+_RandomAccessIterator2
+__pattern_adjacent_difference(__parallel_tag<_IsVector> __tag, _ExecutionPolicy&& __exec,
+                              _RandomAccessIterator1 __first, _RandomAccessIterator1 __last,
+                              _RandomAccessIterator2 __d_first, _BinaryOperation __op)
 {
     _PSTL_ASSERT(__first != __last);
-    typedef typename std::iterator_traits<_ForwardIterator1>::reference _ReferenceType1;
-    typedef typename std::iterator_traits<_ForwardIterator2>::reference _ReferenceType2;
+    typedef typename std::iterator_traits<_RandomAccessIterator1>::reference _ReferenceType1;
+    typedef typename std::iterator_traits<_RandomAccessIterator2>::reference _ReferenceType2;
+
+    using __backend_tag = typename decltype(__tag)::__backend_tag;
 
     *__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);
-            __internal::__brick_walk3(
-                __b, __e, __b + 1, __d_b + 1,
-                [&__op](_ReferenceType1 __x, _ReferenceType1 __y, _ReferenceType2 __z) { __z = __op(__y, __x); },
-                __is_vector);
-        });
+    __par_backend::__parallel_for(__backend_tag{}, std::forward<_ExecutionPolicy>(__exec), __first, __last - 1,
+                                  [&__op, __d_first, __first](_RandomAccessIterator1 __b, _RandomAccessIterator1 __e)
+                                  {
+                                      _RandomAccessIterator2 __d_b = __d_first + (__b - __first);
+                                      __internal::__brick_walk3(
+                                          __b, __e, __b + 1, __d_b + 1,
+                                          [&__op](_ReferenceType1 __x, _ReferenceType1 __y, _ReferenceType2 __z)
+                                          { __z = __op(__y, __x); },
+                                          _IsVector{});
+                                  });
     return __d_first + (__last - __first);
 }
 
diff --git a/libstdc++-v3/include/pstl/parallel_backend.h b/libstdc++-v3/include/pstl/parallel_backend.h
index 4caa4a8e412..e9a90e11dba 100644
--- a/libstdc++-v3/include/pstl/parallel_backend.h
+++ b/libstdc++-v3/include/pstl/parallel_backend.h
@@ -22,6 +22,12 @@ namespace __pstl
 {
 namespace __par_backend = __tbb_backend;
 }
+#elif defined(_PSTL_PAR_BACKEND_OPENMP)
+#    include "parallel_backend_omp.h"
+namespace __pstl
+{
+namespace __par_backend = __omp_backend;
+}
 #else
 _PSTL_PRAGMA_MESSAGE("Parallel backend was not specified");
 #endif
diff --git a/libstdc++-v3/include/pstl/parallel_backend_serial.h b/libstdc++-v3/include/pstl/parallel_backend_serial.h
index 16e773d411a..db2da43eb06 100644
--- a/libstdc++-v3/include/pstl/parallel_backend_serial.h
+++ b/libstdc++-v3/include/pstl/parallel_backend_serial.h
@@ -50,15 +50,15 @@ __cancel_execution()
 
 template <class _ExecutionPolicy, class _Index, class _Fp>
 void
-__parallel_for(_ExecutionPolicy&&, _Index __first, _Index __last, _Fp __f)
+__parallel_for(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last, _Fp __f)
 {
     __f(__first, __last);
 }
 
 template <class _ExecutionPolicy, class _Value, class _Index, typename _RealBody, typename _Reduction>
 _Value
-__parallel_reduce(_ExecutionPolicy&&, _Index __first, _Index __last, const _Value& __identity,
-                  const _RealBody& __real_body, const _Reduction&)
+__parallel_reduce(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last,
+                  const _Value& __identity, const _RealBody& __real_body, const _Reduction&)
 {
     if (__first == __last)
     {
@@ -72,16 +72,16 @@ __parallel_reduce(_ExecutionPolicy&&, _Index __first, _Index __last, const _Valu
 
 template <class _ExecutionPolicy, class _Index, class _UnaryOp, class _Tp, class _BinaryOp, class _Reduce>
 _Tp
-__parallel_transform_reduce(_ExecutionPolicy&&, _Index __first, _Index __last, _UnaryOp, _Tp __init, _BinaryOp,
-                            _Reduce __reduce)
+__parallel_transform_reduce(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last,
+                            _UnaryOp, _Tp __init, _BinaryOp, _Reduce __reduce)
 {
     return __reduce(__first, __last, __init);
 }
 
 template <class _ExecutionPolicy, typename _Index, typename _Tp, typename _Rp, typename _Cp, typename _Sp, typename _Ap>
 void
-__parallel_strict_scan(_ExecutionPolicy&&, _Index __n, _Tp __initial, _Rp __reduce, _Cp __combine, _Sp __scan,
-                       _Ap __apex)
+__parallel_strict_scan(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _Index __n, _Tp __initial,
+                       _Rp __reduce, _Cp __combine, _Sp __scan, _Ap __apex)
 {
     _Tp __sum = __initial;
     if (__n)
@@ -93,15 +93,16 @@ __parallel_strict_scan(_ExecutionPolicy&&, _Index __n, _Tp __initial, _Rp __redu
 
 template <class _ExecutionPolicy, class _Index, class _UnaryOp, class _Tp, class _BinaryOp, class _Reduce, class _Scan>
 _Tp
-__parallel_transform_scan(_ExecutionPolicy&&, _Index __n, _UnaryOp, _Tp __init, _BinaryOp, _Reduce, _Scan __scan)
+__parallel_transform_scan(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _Index __n, _UnaryOp,
+                          _Tp __init, _BinaryOp, _Reduce, _Scan __scan)
 {
     return __scan(_Index(0), __n, __init);
 }
 
 template <class _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare, typename _LeafSort>
 void
-__parallel_stable_sort(_ExecutionPolicy&&, _RandomAccessIterator __first, _RandomAccessIterator __last, _Compare __comp,
-                       _LeafSort __leaf_sort, std::size_t = 0)
+__parallel_stable_sort(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _RandomAccessIterator __first,
+                       _RandomAccessIterator __last, _Compare __comp, _LeafSort __leaf_sort, std::size_t = 0)
 {
     __leaf_sort(__first, __last, __comp);
 }
@@ -109,16 +110,16 @@ __parallel_stable_sort(_ExecutionPolicy&&, _RandomAccessIterator __first, _Rando
 template <class _ExecutionPolicy, typename _RandomAccessIterator1, typename _RandomAccessIterator2,
           typename _RandomAccessIterator3, typename _Compare, typename _LeafMerge>
 void
-__parallel_merge(_ExecutionPolicy&&, _RandomAccessIterator1 __first1, _RandomAccessIterator1 __last1,
-                 _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2, _RandomAccessIterator3 __outit,
-                 _Compare __comp, _LeafMerge __leaf_merge)
+__parallel_merge(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _RandomAccessIterator1 __first1,
+                 _RandomAccessIterator1 __last1, _RandomAccessIterator2 __first2, _RandomAccessIterator2 __last2,
+                 _RandomAccessIterator3 __outit, _Compare __comp, _LeafMerge __leaf_merge)
 {
     __leaf_merge(__first1, __last1, __first2, __last2, __outit, __comp);
 }
 
 template <class _ExecutionPolicy, typename _F1, typename _F2>
 void
-__parallel_invoke(_ExecutionPolicy&&, _F1&& __f1, _F2&& __f2)
+__parallel_invoke(__pstl::__internal::__serial_backend_tag, _ExecutionPolicy&&, _F1&& __f1, _F2&& __f2)
 {
     std::forward<_F1>(__f1)();
     std::forward<_F2>(__f2)();
diff --git a/libstdc++-v3/include/pstl/parallel_backend_tbb.h b/libstdc++-v3/include/pstl/parallel_backend_tbb.h
index 3fd1a06ebd8..3ff55237bff 100644
--- a/libstdc++-v3/include/pstl/parallel_backend_tbb.h
+++ b/libstdc++-v3/include/pstl/parallel_backend_tbb.h
@@ -99,7 +99,7 @@ class __parallel_for_body
 // wrapper over tbb::parallel_for
 template <class _ExecutionPolicy, class _Index, class _Fp>
 void
-__parallel_for(_ExecutionPolicy&&, _Index __first, _Index __last, _Fp __f)
+__parallel_for(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last, _Fp __f)
 {
     tbb::this_task_arena::isolate([=]() {
         tbb::parallel_for(tbb::blocked_range<_Index>(__first, __last), __parallel_for_body<_Index, _Fp>(__f));
@@ -110,8 +110,8 @@ __parallel_for(_ExecutionPolicy&&, _Index __first, _Index __last, _Fp __f)
 // wrapper over tbb::parallel_reduce
 template <class _ExecutionPolicy, class _Value, class _Index, typename _RealBody, typename _Reduction>
 _Value
-__parallel_reduce(_ExecutionPolicy&&, _Index __first, _Index __last, const _Value& __identity,
-                  const _RealBody& __real_body, const _Reduction& __reduction)
+__parallel_reduce(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last,
+                  const _Value& __identity, const _RealBody& __real_body, const _Reduction& __reduction)
 {
     return tbb::this_task_arena::isolate([__first, __last, &__identity, &__real_body, &__reduction]() -> _Value {
         return tbb::parallel_reduce(
@@ -191,8 +191,8 @@ 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)
+__parallel_transform_reduce(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _Index __first, _Index __last,
+                            _Up __u, _Tp __init, _Cp __combine, _Rp __brick_reduce)
 {
     __tbb_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
@@ -354,8 +354,8 @@ __downsweep(_Index __i, _Index __m, _Index __tilesize, _Tp* __r, _Index __lastsi
 // T must have a trivial constructor and destructor.
 template <class _ExecutionPolicy, typename _Index, typename _Tp, typename _Rp, typename _Cp, typename _Sp, typename _Ap>
 void
-__parallel_strict_scan(_ExecutionPolicy&&, _Index __n, _Tp __initial, _Rp __reduce, _Cp __combine, _Sp __scan,
-                       _Ap __apex)
+__parallel_strict_scan(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _Index __n, _Tp __initial,
+                       _Rp __reduce, _Cp __combine, _Sp __scan, _Ap __apex)
 {
     tbb::this_task_arena::isolate([=, &__combine]() {
         if (__n > 1)
@@ -394,8 +394,8 @@ __parallel_strict_scan(_ExecutionPolicy&&, _Index __n, _Tp __initial, _Rp __redu
 
 template <class _ExecutionPolicy, class _Index, class _Up, class _Tp, class _Cp, class _Rp, class _Sp>
 _Tp
-__parallel_transform_scan(_ExecutionPolicy&&, _Index __n, _Up __u, _Tp __init, _Cp __combine, _Rp __brick_reduce,
-                          _Sp __scan)
+__parallel_transform_scan(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _Index __n, _Up __u, _Tp __init,
+                          _Cp __combine, _Rp __brick_reduce, _Sp __scan)
 {
     __trans_scan_body<_Index, _Up, _Tp, _Cp, _Rp, _Sp> __body(__u, __init, __combine, __brick_reduce, __scan);
     auto __range = tbb::blocked_range<_Index>(0, __n);
@@ -1154,8 +1154,8 @@ __stable_sort_func<_RandomAccessIterator1, _RandomAccessIterator2, _Compare, _Le
 
 template <class _ExecutionPolicy, typename _RandomAccessIterator, typename _Compare, typename _LeafSort>
 void
-__parallel_stable_sort(_ExecutionPolicy&&, _RandomAccessIterator __xs, _RandomAccessIterator __xe, _Compare __comp,
-                       _LeafSort __leaf_sort, std::size_t __nsort = 0)
+__parallel_stable_sort(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _RandomAccessIterator __xs,
+                       _RandomAccessIterator __xe, _Compare __comp, _LeafSort __leaf_sort, std::size_t __nsort = 0)
 {
     tbb::this_task_arena::isolate([=, &__nsort]() {
         //sorting based on task tree and parallel merge
@@ -1248,9 +1248,9 @@ operator()(__task* __self)
 template <class _ExecutionPolicy, typename _RandomAccessIterator1, typename _RandomAccessIterator2,
           typename _RandomAccessIterator3, typename _Compare, typename _LeafMerge>
 void
-__parallel_merge(_ExecutionPolicy&&, _RandomAccessIterator1 __xs, _RandomAccessIterator1 __xe,
-                 _RandomAccessIterator2 __ys, _RandomAccessIterator2 __ye, _RandomAccessIterator3 __zs, _Compare __comp,
-                 _LeafMerge __leaf_merge)
+__parallel_merge(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _RandomAccessIterator1 __xs,
+                 _RandomAccessIterator1 __xe, _RandomAccessIterator2 __ys, _RandomAccessIterator2 __ye,
+                 _RandomAccessIterator3 __zs, _Compare __comp, _LeafMerge __leaf_merge)
 {
     typedef typename std::iterator_traits<_RandomAccessIterator1>::difference_type _DifferenceType1;
     typedef typename std::iterator_traits<_RandomAccessIterator2>::difference_type _DifferenceType2;
@@ -1279,7 +1279,7 @@ __parallel_merge(_ExecutionPolicy&&, _RandomAccessIterator1 __xs, _RandomAccessI
 //------------------------------------------------------------------------
 template <class _ExecutionPolicy, typename _F1, typename _F2>
 void
-__parallel_invoke(_ExecutionPolicy&&, _F1&& __f1, _F2&& __f2)
+__parallel_invoke(__pstl::__internal::__tbb_backend_tag, _ExecutionPolicy&&, _F1&& __f1, _F2&& __f2)
 {
     //TODO: a version of tbb::this_task_arena::isolate with variadic arguments pack should be added in the future
     tbb::this_task_arena::isolate([&]() { tbb::parallel_invoke(std::forward<_F1>(__f1), std::forward<_F2>(__f2)); });
diff --git a/libstdc++-v3/include/pstl/parallel_impl.h b/libstdc++-v3/include/pstl/parallel_impl.h
index 46560c642ea..4cf2abde650 100644
--- a/libstdc++-v3/include/pstl/parallel_impl.h
+++ b/libstdc++-v3/include/pstl/parallel_impl.h
@@ -24,17 +24,19 @@ namespace __internal
 //-----------------------------------------------------------------------
 /** Return extremum value returned by brick f[i,j) for subranges [i,j) of [first,last)
 Each f[i,j) must return a value in [i,j). */
-template <class _ExecutionPolicy, class _Index, class _Brick, class _Compare>
+template <class _BackendTag, class _ExecutionPolicy, class _Index, class _Brick, class _Compare>
 _Index
-__parallel_find(_ExecutionPolicy&& __exec, _Index __first, _Index __last, _Brick __f, _Compare __comp, bool __b_first)
+__parallel_find(_BackendTag __tag, _ExecutionPolicy&& __exec, _Index __first, _Index __last, _Brick __f,
+                _Compare __comp, bool __b_first)
 {
     typedef typename std::iterator_traits<_Index>::difference_type _DifferenceType;
     const _DifferenceType __n = __last - __first;
     _DifferenceType __initial_dist = __b_first ? __n : -1;
     std::atomic<_DifferenceType> __extremum(__initial_dist);
     // TODO: find out what is better here: parallel_for or parallel_reduce
-    __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                  [__comp, __f, __first, &__extremum](_Index __i, _Index __j) {
+    __par_backend::__parallel_for(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                  [__comp, __f, __first, &__extremum](_Index __i, _Index __j)
+                                  {
                                       // See "Reducing Contention Through Priority Updates", PPoPP '13, for discussion of
                                       // why using a shared variable scales fairly well in this situation.
                                       if (__comp(__i - __first, __extremum))
@@ -59,13 +61,14 @@ __parallel_find(_ExecutionPolicy&& __exec, _Index __first, _Index __last, _Brick
 // parallel_or
 //------------------------------------------------------------------------
 //! Return true if brick f[i,j) returns true for some subrange [i,j) of [first,last)
-template <class _ExecutionPolicy, class _Index, class _Brick>
+template <class _BackendTag, class _ExecutionPolicy, class _Index, class _Brick>
 bool
-__parallel_or(_ExecutionPolicy&& __exec, _Index __first, _Index __last, _Brick __f)
+__parallel_or(_BackendTag __tag, _ExecutionPolicy&& __exec, _Index __first, _Index __last, _Brick __f)
 {
     std::atomic<bool> __found(false);
-    __par_backend::__parallel_for(std::forward<_ExecutionPolicy>(__exec), __first, __last,
-                                  [__f, &__found](_Index __i, _Index __j) {
+    __par_backend::__parallel_for(__tag, std::forward<_ExecutionPolicy>(__exec), __first, __last,
+                                  [__f, &__found](_Index __i, _Index __j)
+                                  {
                                       if[...]

[diff truncated at 524288 bytes]

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2023-06-26 18:35 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-06-26 18:35 [gcc r14-2109] libstdc++: Synchronize PSTL with upstream Thomas Rodgers

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).