par_unseq in most cases

Signed-off-by: Hari Hara Naveen S <[email protected]>
STEllAR-GROUP · Aug 9, 2023 · bfb7da1 · bfb7da1
1 parent 215ae4f
commit bfb7da1
Show file tree

Hide file tree

Showing 4 changed files with 49 additions and 1 deletion.
diff --git a/libs/core/algorithms/include/hpx/parallel/algorithms/detail/find.hpp b/libs/core/algorithms/include/hpx/parallel/algorithms/detail/find.hpp
@@ -9,6 +9,7 @@
 #pragma once
 
 #include <hpx/config.hpp>
+#include <hpx/execution/traits/is_execution_policy.hpp>
 #include <hpx/functional/detail/tag_fallback_invoke.hpp>
 #include <hpx/functional/invoke.hpp>
 #include <hpx/parallel/algorithms/detail/advance_to_sentinel.hpp>
@@ -34,6 +35,8 @@ namespace hpx::parallel::detail {
             sequential_find_t<ExPolicy>, Iterator first, Sentinel last,
             T const& value, Proj proj = Proj())
         {
+            static_assert(hpx::is_sequenced_execution_policy<ExPolicy>::value ||
+                hpx::is_unsequenced_execution_policy<ExPolicy>::value);
             return util::loop_pred<ExPolicy>(
                 first, last, [&value, &proj](auto const& curr) {
                     return HPX_INVOKE(proj, *curr) == value;

diff --git a/libs/core/algorithms/include/hpx/parallel/unseq/simd_helpers.hpp b/libs/core/algorithms/include/hpx/parallel/unseq/simd_helpers.hpp
@@ -25,6 +25,7 @@ namespace hpx::parallel::util {
         Compiler and Hardware should also support vector operations for IterDiff,
         else we see slower performance when compared to sequential version
     */
+    // f(Iter) -> bool
     template <typename Iter, typename IterDiff, typename F>
     Iter unseq_first_n(Iter const first, IterDiff const n, F&& f) noexcept
     {

diff --git a/libs/core/algorithms/include/hpx/parallel/util/loop.hpp b/libs/core/algorithms/include/hpx/parallel/util/loop.hpp
@@ -900,6 +900,20 @@ namespace hpx::parallel::util {
 
                 return call(base_idx, it, num, HPX_FORWARD(F, f));
             }
+
+            template <typename Iter, typename F>
+            HPX_HOST_DEVICE HPX_FORCEINLINE static constexpr Iter unseq_call(
+                std::size_t base_idx, Iter it, std::size_t num, F&& f)
+            {
+                // clang-format off
+                HPX_VECTORIZE
+                for (std::size_t i = 0; i < num; i++)    // -V112
+                {
+                    HPX_INVOKE(f, *(it + i), base_idx++);
+                }
+                // clang-format on
+                return it + num;
+            }
         };
     }    // namespace detail
 
@@ -931,6 +945,36 @@ namespace hpx::parallel::util {
         }
     };
 
+    template <typename Iter, typename CancelToken, typename F,
+        HPX_CONCEPT_REQUIRES_(hpx::traits::is_random_access_iterator_v<Iter>)>
+    HPX_HOST_DEVICE HPX_FORCEINLINE static constexpr Iter tag_invoke(
+        hpx::parallel::util::loop_idx_n_t<hpx::execution::unsequenced_policy>,
+        std::size_t base_idx, Iter it, std::size_t count, CancelToken& tok,
+        F&& f)
+    {
+        if (tok.was_cancelled(base_idx))
+            return it;
+
+        return detail::loop_idx_n<std::random_access_iterator_tag,
+            hpx::execution::unsequenced_policy>::unseq_call(base_idx, it, count,
+            HPX_FORWARD(F, f));
+    }
+
+    template <typename Iter, typename CancelToken, typename F,
+        HPX_CONCEPT_REQUIRES_(hpx::traits::is_random_access_iterator_v<Iter>)>
+    HPX_HOST_DEVICE HPX_FORCEINLINE static constexpr Iter tag_invoke(
+        hpx::parallel::util::loop_idx_n_t<hpx::execution::unsequenced_task_policy>,
+        std::size_t base_idx, Iter it, std::size_t count, CancelToken& tok,
+        F&& f)
+    {
+        if (tok.was_cancelled(base_idx))
+            return it;
+
+        return detail::loop_idx_n<std::random_access_iterator_tag,
+            hpx::execution::unsequenced_task_policy>::unseq_call(base_idx, it, count,
+            HPX_FORWARD(F, f));
+    }
+
 #if !defined(HPX_COMPUTE_DEVICE_CODE)
     template <typename ExPolicy>
     inline constexpr loop_idx_n_t<ExPolicy> loop_idx_n =

diff --git a/libs/core/algorithms/tests/unit/algorithms/util/test_simd_helpers.cpp b/libs/core/algorithms/tests/unit/algorithms/util/test_simd_helpers.cpp
@@ -23,7 +23,7 @@ void test_unseq_first_n1_dispatch2(std::size_t length, std::size_t first_index)
 {
     first_index = first_index % length;
 
-    std::vector<T> v(length, static_cast<T>(0));
+    std::vector<T> v(length);
     std::size_t i = 0;
 
     std::for_each(v.begin(), v.end(), [&](T& t) {