Remove dependency on parallel stl algorithms and only depend on openmp for parallelism

Author: digitalillusions

CMakeLists.txt

@@ -8,16 +8,14 @@ set_property(GLOBAL PROPERTY USE_FOLDERS on)
 set(CMAKE_CXX_STANDARD 11)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 
-if (UNIX)
-find_package(OpenMP)
+find_package(OpenMP REQUIRED)
 if (OPENMP_FOUND)
 	if (CMAKE_VERSION VERSION_GREATER "3.8")
 		link_libraries(OpenMP::OpenMP_CXX)
 	else()
 		set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
 	endif()
 endif(OPENMP_FOUND)
-endif (UNIX)
 
 OPTION(BUILD_AS_SHARED_LIBS "Build all the libraries as shared" OFF)
 if (BUILD_AS_SHARED_LIBS)
@@ -49,11 +47,6 @@ endif ()
 
 target_include_directories(CompactNSearch PUBLIC include)
 
-if (APPLE AND CMAKE_CXX_COMPILER_ID MATCHES "Clang")
-	find_package(oneDPL REQUIRED)
-	target_link_libraries(CompactNSearch PUBLIC oneDPL)
-endif()
-
 install(FILES "include/CompactNSearch" ${HEADER_FILES}
 	DESTINATION include/)
 

demo/main.cpp

@@ -8,6 +8,7 @@
 #include <limits>
 #include <chrono>
 #include <random>
+#include <algorithm>
 
 #include <omp.h>
 
@@ -151,21 +152,15 @@ enright_velocity_field(std::array<Real, 3> const& x)
 void
 advect()
 {
-#ifdef _MSC_VER
-	concurrency::parallel_for_each(
-#elif defined(__APPLE__) && defined(__clang__)
-	std::for_each(oneapi::dpl::execution::par,
-#else
-	__gnu_parallel::for_each(
-#endif
-	positions.begin(), positions.end(), [&](std::array<Real, 3>& x)
+#pragma omp parallel for
+	for (int _i = 0; _i < positions.size(); _i++)
 	{
+		std::array<Real, 3>& x = positions[_i];
 		std::array<Real, 3> v = enright_velocity_field(x);
 		x[0] += static_cast<Real>(0.005) * v[0];
 		x[1] += static_cast<Real>(0.005) * v[1];
 		x[2] += static_cast<Real>(0.005) * v[1];
 	}
-	);
 }
 
 int main(int argc, char* argv[])

include/Config.h

@@ -11,12 +11,3 @@ namespace CompactNSearch
 
 #define INITIAL_NUMBER_OF_INDICES   50
 #define INITIAL_NUMBER_OF_NEIGHBORS 50
-
-#ifdef _MSC_VER
-	#include <ppl.h>
-#elif defined(__APPLE__) && defined(__clang__)
-	#include <oneapi/dpl/execution>
-	#include <oneapi/dpl/algorithm>
-#else
-	#include <parallel/algorithm>
-#endif

src/CompactNSearch.cpp

@@ -234,15 +234,9 @@ NeighborhoodSearch::update_point_sets()
 	}
 
 	// Precompute cell indices.
-#ifdef _MSC_VER
-	concurrency::parallel_for_each(
-#elif defined(__APPLE__) && defined(__clang__)
-	std::for_each(oneapi::dpl::execution::par,
-#else
-	__gnu_parallel::for_each(
-#endif
-	m_point_sets.begin(), m_point_sets.end(), [&](PointSet& d)
-	{
+#pragma omp parallel for
+	for (int j = 0; j < m_point_sets.size(); j++){
+		PointSet& d = m_point_sets[j];
 		if (d.is_dynamic()) 
 		{
 			d.m_keys.swap(d.m_old_keys);
@@ -251,7 +245,7 @@ NeighborhoodSearch::update_point_sets()
 				d.m_keys[i] = cell_index(d.point(i));
 			}
 		}
-	});
+	}
 
 	std::vector<unsigned int> to_delete;
 	if (m_erase_empty_cells)
@@ -315,15 +309,9 @@ NeighborhoodSearch::erase_empty_entries(std::vector<unsigned int> const& to_dele
 	});
 
 	// Perform neighborhood search.
-#ifdef _MSC_VER
-	concurrency::parallel_for_each(
-#elif defined(__APPLE__) && defined(__clang__)
-	std::for_each(oneapi::dpl::execution::par,
-#else
-	__gnu_parallel::for_each(
-#endif
-	kvps.begin(), kvps.end(), [&](std::pair<HashKey const, unsigned int>* kvp_)
-	{
+#pragma omp parallel for
+	for (int j =0; j < kvps.size(); j++){
+		std::pair<HashKey const, unsigned int>* kvp_ = kvps[j];
 		auto& kvp = *kvp_;
 
 		for (unsigned int i = 0; i < to_delete.size(); ++i)
@@ -334,7 +322,7 @@ NeighborhoodSearch::erase_empty_entries(std::vector<unsigned int> const& to_dele
 				break;
 			}
 		}
-	});
+	}
 }
 
 void
@@ -415,22 +403,16 @@ NeighborhoodSearch::query()
 	});
 
 	// Perform neighborhood search.
-#ifdef _MSC_VER
-	concurrency::parallel_for_each(
-#elif defined(__APPLE__) && defined(__clang__)
-	std::for_each(oneapi::dpl::execution::par,
-#else
-	__gnu_parallel::for_each(
-#endif
-	kvps.begin(), kvps.end(), [&](std::pair<HashKey const, unsigned int> const* kvp_)
-	{
+#pragma omp parallel for
+	for (int _i = 0; _i < kvps.size(); _i++){
+		std::pair<HashKey const, unsigned int> const* kvp_ = kvps[_i];
 		auto const& kvp = *kvp_;
 		HashEntry const& entry = m_entries[kvp.second];
 		HashKey const& key = kvp.first;
 
 		if (entry.n_searching_points == 0u)
 		{
-			return;
+			continue;
 		}
 
 		for (unsigned int a = 0; a < entry.n_indices(); ++a)
@@ -471,26 +453,20 @@ NeighborhoodSearch::query()
 			}
 		}
 	}
-	);
 
 
 	std::vector<std::array<bool, 27>> visited(m_entries.size(), {false});
 	std::vector<Spinlock> entry_locks(m_entries.size());
 
-#ifdef _MSC_VER
-	concurrency::parallel_for_each(
-#elif defined(__APPLE__) && defined(__clang__)
-	std::for_each(oneapi::dpl::execution::par,
-#else
-	__gnu_parallel::for_each(
-#endif
-	kvps.begin(), kvps.end(), [&](std::pair<HashKey const, unsigned int> const* kvp_)
-	{
+#pragma omp parallel for
+	for (int _i = 0; _i < kvps.size(); _i++){
+		std::pair<HashKey const, unsigned int> const* kvp_ = kvps[_i];
+
 		auto const& kvp = *kvp_;
 		HashEntry const& entry = m_entries[kvp.second];
 
 		if (entry.n_searching_points == 0u)
-			return;
+			continue;
 		HashKey const& key = kvp.first;
 
 		for (int dj = -1; dj <= 1; dj++)
@@ -579,8 +555,7 @@ NeighborhoodSearch::query()
 				}
 			}
 		}
-	});
-
+	}
 }