Improve the implementation of __enumerable_thread_local_storage (#2049)

Implement with CRTP.
Store construction arguments by values.
Replace unique_ptr with optional.

Co-authored-by: Dan Hoeflinger <[email protected]>

Author: akukanov

include/oneapi/dpl/pstl/omp/util.h

@@ -152,36 +152,46 @@ __process_chunk(const __chunk_metrics& __metrics, _Iterator __base, _Index __chu
 
 namespace __detail
 {
-struct __get_num_threads
+
+// Workaround for VS 2017: declare an alias to the CRTP base template
+template <typename _ValueType, typename... _Args>
+struct __enumerable_thread_local_storage;
+
+template <typename... _Ts>
+using __etls_base = __utils::__enumerable_thread_local_storage_base<__enumerable_thread_local_storage, _Ts...>;
+
+template <typename _ValueType, typename... _Args>
+struct __enumerable_thread_local_storage : public __etls_base<_ValueType, _Args...>
 {
-    std::size_t
-    operator()() const
+
+    template <typename... _LocalArgs>
+    __enumerable_thread_local_storage(_LocalArgs&&... __args)
+        : __etls_base<_ValueType, _Args...>({std::forward<_LocalArgs>(__args)...})
+    {
+    }
+
+    static std::size_t
+    get_num_threads()
     {
         return omp_in_parallel() ? omp_get_num_threads() : omp_get_max_threads();
     }
-};
 
-struct __get_thread_num
-{
-    std::size_t
-    operator()() const
+    static std::size_t
+    get_thread_num()
     {
         return omp_get_thread_num();
     }
 };
 
 } // namespace __detail
 
-// enumerable thread local storage should only be created from make function
-template <typename _ValueType, typename... Args>
-oneapi::dpl::__utils::__detail::__enumerable_thread_local_storage<
-    _ValueType, oneapi::dpl::__omp_backend::__detail::__get_num_threads,
-    oneapi::dpl::__omp_backend::__detail::__get_thread_num, Args...>
-__make_enumerable_tls(Args&&... __args)
+// enumerable thread local storage should only be created with this make function
+template <typename _ValueType, typename... _Args>
+__detail::__enumerable_thread_local_storage<_ValueType, std::remove_reference_t<_Args>...>
+__make_enumerable_tls(_Args&&... __args)
 {
-    return oneapi::dpl::__utils::__detail::__enumerable_thread_local_storage<
-        _ValueType, oneapi::dpl::__omp_backend::__detail::__get_num_threads,
-        oneapi::dpl::__omp_backend::__detail::__get_thread_num, Args...>(std::forward<Args>(__args)...);
+    return __detail::__enumerable_thread_local_storage<_ValueType, std::remove_reference_t<_Args>...>(
+        std::forward<_Args>(__args)...);
 }
 
 } // namespace __omp_backend

include/oneapi/dpl/pstl/parallel_backend_tbb.h

@@ -1308,35 +1308,46 @@ __parallel_for_each(oneapi::dpl::__internal::__tbb_backend_tag, _ExecutionPolicy
 
 namespace __detail
 {
-struct __get_num_threads
+
+// Workaround for VS 2017: declare an alias to the CRTP base template
+template <typename _ValueType, typename... _Args>
+struct __enumerable_thread_local_storage;
+
+template <typename... _Ts>
+using __etls_base = __utils::__enumerable_thread_local_storage_base<__enumerable_thread_local_storage, _Ts...>;
+
+template <typename _ValueType, typename... _Args>
+struct __enumerable_thread_local_storage : public __etls_base<_ValueType, _Args...>
 {
-    std::size_t
-    operator()() const
+
+    template <typename... _LocalArgs>
+    __enumerable_thread_local_storage(_LocalArgs&&... __args)
+        : __etls_base<_ValueType, _Args...>({std::forward<_LocalArgs>(__args)...})
+    {
+    }
+
+    static std::size_t
+    get_num_threads()
     {
         return tbb::this_task_arena::max_concurrency();
     }
-};
 
-struct __get_thread_num
-{
-    std::size_t
-    operator()() const
+    static std::size_t
+    get_thread_num()
     {
         return tbb::this_task_arena::current_thread_index();
     }
 };
-} //namespace __detail
-
-// enumerable thread local storage should only be created from make function
-template <typename _ValueType, typename... Args>
-oneapi::dpl::__utils::__detail::__enumerable_thread_local_storage<
-    _ValueType, oneapi::dpl::__tbb_backend::__detail::__get_num_threads,
-    oneapi::dpl::__tbb_backend::__detail::__get_thread_num, Args...>
-__make_enumerable_tls(Args&&... __args)
+
+} // namespace __detail
+
+// enumerable thread local storage should only be created with this make function
+template <typename _ValueType, typename... _Args>
+__detail::__enumerable_thread_local_storage<_ValueType, std::remove_reference_t<_Args>...>
+__make_enumerable_tls(_Args&&... __args)
 {
-    return oneapi::dpl::__utils::__detail::__enumerable_thread_local_storage<
-        _ValueType, oneapi::dpl::__tbb_backend::__detail::__get_num_threads,
-        oneapi::dpl::__tbb_backend::__detail::__get_thread_num, Args...>(std::forward<Args>(__args)...);
+    return __detail::__enumerable_thread_local_storage<_ValueType, std::remove_reference_t<_Args>...>(
+        std::forward<_Args>(__args)...);
 }
 
 } // namespace __tbb_backend

include/oneapi/dpl/pstl/parallel_backend_utils.h

@@ -19,7 +19,7 @@
 #include <atomic>
 #include <cstddef>
 #include <iterator>
-#include <memory>
+#include <optional>
 #include <tuple>
 #include <utility>
 #include <vector>
@@ -306,16 +306,13 @@ __set_symmetric_difference_construct(_ForwardIterator1 __first1, _ForwardIterato
     return __cc_range(__first2, __last2, __result);
 }
 
-namespace __detail
+template <template <typename, typename...> typename _Concrete, typename _ValueType, typename... _Args>
+struct __enumerable_thread_local_storage_base
 {
+    using _Derived = _Concrete<_ValueType, _Args...>;
 
-template <typename _ValueType, typename _GetNumThreads, typename _GetThreadNum, typename... _Args>
-struct __enumerable_thread_local_storage
-{
-
-    template <typename... _LocalArgs>
-    __enumerable_thread_local_storage(_LocalArgs&&... __args)
-        : __thread_specific_storage(_GetNumThreads{}()), __num_elements(0), __args(std::forward<_LocalArgs>(__args)...)
+    __enumerable_thread_local_storage_base(std::tuple<_Args...> __tp)
+        : __thread_specific_storage(_Derived::get_num_threads()), __num_elements(0), __args(__tp)
     {
     }
 
@@ -359,24 +356,22 @@ struct __enumerable_thread_local_storage
     _ValueType&
     get_for_current_thread()
     {
-        const std::size_t __i = _GetThreadNum{}();
-        std::unique_ptr<_ValueType>& __thread_local_storage = __thread_specific_storage[__i];
-        if (!__thread_local_storage)
+        const std::size_t __i = _Derived::get_thread_num();
+        std::optional<_ValueType>& __local = __thread_specific_storage[__i];
+        if (!__local)
         {
             // create temporary storage on first usage to avoid extra parallel region and unnecessary instantiation
-            __thread_local_storage =
-                std::apply([](_Args... __arg_pack) { return std::make_unique<_ValueType>(__arg_pack...); }, __args);
+            std::apply([&__local](_Args... __arg_pack) { __local.emplace(__arg_pack...); }, __args);
             __num_elements.fetch_add(1, std::memory_order_relaxed);
         }
-        return *__thread_local_storage;
+        return *__local;
     }
 
-    std::vector<std::unique_ptr<_ValueType>> __thread_specific_storage;
+    std::vector<std::optional<_ValueType>> __thread_specific_storage;
     std::atomic_size_t __num_elements;
     const std::tuple<_Args...> __args;
 };
 
-} // namespace __detail
 } // namespace __utils
 } // namespace dpl
 } // namespace oneapi