TBB support enhancements

CMakeLists.txt

@@ -158,20 +158,15 @@ endif()
 # TBB (optional)
 # Must come before CGAL so it can use TBB properly
 if ( CCCORELIB_USE_TBB )
-	find_package( TBB COMPONENTS tbb CONFIG )
+	find_package( TBB COMPONENTS tbb 2022.0.0 CONFIG )
 
+	set( CCCORELIB_USE_QT_CONCURRENT OFF )
 	if ( TBB_FOUND )
-		if ( ${TBB_VERSION} VERSION_GREATER 2021.0.0 )
-			target_link_libraries( CCCoreLib
-				PUBLIC
-					TBB::tbb
-			)
-		else()
-			target_link_libraries( CCCoreLib
-				PUBLIC
-					${TBB_IMPORTED_TARGETS}
-			)
-		endif()
+		target_link_libraries( CCCoreLib
+			PUBLIC
+				TBB::tbb
+		)
+
 		target_compile_definitions( CCCoreLib
 			PUBLIC
 				CC_CORE_LIB_USES_TBB
@@ -192,14 +187,7 @@ if ( CCCORELIB_USE_CGAL )
 
 	if ( CCCORELIB_USE_TBB )
 		if ( TBB_FOUND )
-			# Once Linux libcgal-dev >= 5.0, target_compile_definitions replaced by:
-			#  CGAL_target_use_TBB( CCCoreLib )
-
-			target_compile_definitions( CCCoreLib
-				PRIVATE
-					CGAL_LINKED_WITH_TBB
-					NOMINMAX
-			)
+			CGAL_target_use_TBB( CCCoreLib )
 		else()
 			message( WARNING "CGAL cannot compile with TBB (TBB not found)" )
 		endif()
@@ -232,7 +220,7 @@ if ( CCCORELIB_USE_QT_CONCURRENT )
 
 	target_link_libraries( CCCoreLib
 		PUBLIC
-            Qt6::Concurrent
+			Qt6::Concurrent
 	)
 
 	target_compile_definitions( CCCoreLib
@@ -268,9 +256,9 @@ install(
 include( CMakePackageConfigHelpers )
 
 configure_package_config_file(${CMAKE_CURRENT_LIST_DIR}/cmake/${PROJECT_NAME}Config.cmake.in
-  ${CMAKE_CURRENT_BINARY_DIR}/${PROJECT_NAME}Config.cmake
-  INSTALL_DESTINATION lib/cmake/${PROJECT_NAME}
-  NO_CHECK_REQUIRED_COMPONENTS_MACRO
+		${CMAKE_CURRENT_BINARY_DIR}/${PROJECT_NAME}Config.cmake
+		INSTALL_DESTINATION lib/cmake/${PROJECT_NAME}
+		NO_CHECK_REQUIRED_COMPONENTS_MACRO
 )
 
 write_basic_package_version_file( CCCoreLibConfigVersion.cmake

include/ParallelSort.h

@@ -10,29 +10,28 @@
 
 #if defined(_MSC_VER) && (_MSC_VER >= 1800)
 
-   //Parallel Patterns Library (for parallel sort)
-   #include <ppl.h>
+	//Parallel Patterns Library (for parallel sort)
+	#include <ppl.h>
 
-   #define ParallelSort Concurrency::parallel_sort
+	#define ParallelSort Concurrency::parallel_sort
 
 #elif CC_CORE_LIB_USES_TBB
 
-   #ifndef Q_MOC_RUN
-   #if defined(emit)
-      #undef emit
-      #include <tbb/parallel_sort.h>
-      #define emit // restore the macro definition of "emit", as it was defined in gtmetamacros.h
-   #else
-   #include <tbb/parallel_sort.h>
-   #endif // defined(emit)
-   #endif // Q_MOC_RUN
+	#ifndef Q_MOC_RUN
+	#if defined(emit)
+		#undef emit
+		#include <oneapi/tbb/parallel_sort.h>
+		#define emit // restore the macro definition of "emit", as it was defined in gtmetamacros.h
+	#else
+	#include <oneapi/tbb/parallel_sort.h>
+	#endif // defined(emit)
+	#endif // Q_MOC_RUN
 
-   #define ParallelSort tbb::parallel_sort
+	#define ParallelSort oneapi::tbb::parallel_sort
 
 #else
 
-   #include <algorithm>
-
-   #define ParallelSort std::sort
+	#include <algorithm>
+	#define ParallelSort std::sort
 
 #endif

src/DgmOctree.cpp

@@ -35,12 +35,15 @@
 #ifndef Q_MOC_RUN
 #if defined(emit)
 	#undef emit
-	#include <tbb/parallel_for.h>
+	#include <oneapi/tbb/parallel_for.h>
+	#include <oneapi/tbb/task_arena.h>
 	#define emit // restore the macro definition of "emit", as it was defined in gtmetamacros.h
 #else
-	#include <tbb/parallel_for.h>
+	#include <oneapi/tbb/parallel_for.h>
+	#include <oneapi/tbb/task_arena.h>
 #endif // defined(emit)
 #endif // Q_MOC_RUN
+using namespace oneapi;
 #endif
 //#endif // #ifndef CC_DEBUG
 
@@ -3111,7 +3114,7 @@ unsigned DgmOctree::executeFunctionForAllCellsAtLevel(	unsigned char level,
 			progressCb->update(0);
 			progressCb->start();
 		}
-		
+
 		NormalizedProgress nprogress(progressCb, m_theAssociatedCloud->size());
 
 #ifdef COMPUTE_NN_SEARCH_STATISTICS
@@ -3250,8 +3253,10 @@ unsigned DgmOctree::executeFunctionForAllCellsAtLevel(	unsigned char level,
 			progressCb->update(0);
 			m_MT_wrapper.normProgressCb = new NormalizedProgress(progressCb, m_theAssociatedCloud->size());
 			progressCb->start();
+#if defined(CC_CORE_LIB_USES_QT_CONCURRENT)
 #ifndef __APPLE__
 			QCoreApplication::processEvents(QEventLoop::EventLoopExec); // to allow the GUI to refresh itself
+#endif
 #endif
 		}
 
@@ -3266,15 +3271,17 @@ unsigned DgmOctree::executeFunctionForAllCellsAtLevel(	unsigned char level,
 		QThreadPool::globalInstance()->setMaxThreadCount(maxThreadCount);
 		QtConcurrent::blockingMap(cells, [this](const octreeCellDesc& desc) { m_MT_wrapper.launchOctreeCellFunc(desc); });
 #elif defined(CC_CORE_LIB_USES_TBB)
-		tbb::task_scheduler_init init(maxThreadCount != 0 ? maxThreadCount : tbb::task_scheduler_init::automatic);
-		tbb::parallel_for(tbb::blocked_range<int>(0, cells.size()),
-			[&](tbb::blocked_range<int> r)
+		tbb::task_arena local_arena(maxThreadCount != 0 ? maxThreadCount : tbb::info::default_concurrency());
+		local_arena.execute([&] {
+			tbb::parallel_for(tbb::blocked_range<std::size_t>(0, cells.size()),
+			[&](tbb::blocked_range<std::size_t> r)
 			{
-				for (auto i = r.begin(); i < r.end(); ++i)
-				{
-					m_MT_wrapper.launchOctreeCellFunc(cells[i]);
-				}
+					for (auto i = r.begin(); i < r.end(); ++i)
+					{
+						m_MT_wrapper.launchOctreeCellFunc(cells[i]);
+					}
 			});
+		});
 #endif
 #ifdef COMPUTE_NN_SEARCH_STATISTICS
 		FILE* fp = fopen("octree_log.txt", "at");
@@ -3873,15 +3880,18 @@ unsigned DgmOctree::executeFunctionForAllCellsStartingAtLevel(unsigned char star
 		QtConcurrent::blockingMap(cells, [this](const octreeCellDesc& desc) { m_MT_wrapper.launchOctreeCellFunc(desc); } );
 #elif defined(CC_CORE_LIB_USES_TBB)
 		// Otherwise we use TBB if we can
-		tbb::task_scheduler_init init(maxThreadCount != 0 ? maxThreadCount : tbb::task_scheduler_init::automatic);
-		tbb::parallel_for(tbb::blocked_range<int>(0, cells.size()),
-			[&](tbb::blocked_range<int> r)
+		tbb::task_arena local_arena(maxThreadCount != 0 ? maxThreadCount : tbb::info::default_concurrency());
+		local_arena.execute([&]
+		{
+			tbb::parallel_for(tbb::blocked_range<std::size_t>(0, cells.size()),
+			[&](tbb::blocked_range<std::size_t> r)
 			{
 				for (auto i = r.begin(); i < r.end(); ++i)
 				{
 					m_MT_wrapper.launchOctreeCellFunc(cells[i]);
 				}
 			});
+		});
 #endif
 
 #ifdef COMPUTE_NN_SEARCH_STATISTICS

src/DistanceComputationTools.cpp

@@ -30,16 +30,20 @@
 #elif defined(CC_CORE_LIB_USES_TBB)
 //enables multi-threading handling with TBB
 #define ENABLE_CLOUD2MESH_DIST_MT
-#include <mutex>
 #ifndef Q_MOC_RUN
 #if defined(emit)
 	#undef emit
-	#include <tbb/parallel_for.h>
+	#include <oneapi/tbb/parallel_for.h>
+	#include <oneapi/tbb/task_arena.h>
+	#include <oneapi/tbb/mutex.h>
 	#define emit // restore the macro definition of "emit", as it was defined in gtmetamacros.h
 #else
-	#include <tbb/parallel_for.h>
+	#include <oneapi/tbb/parallel_for.h>
+	#include <oneapi/tbb/task_arena.h>
+	#include <oneapi/tbb/mutex.h>
 #endif // defined(emit)
 #endif // Q_MOC_RUN
+using namespace oneapi;
 #else
 //Note that there is the case CC_DEBUG=OFF and neither TBB nor Qt
 #undef ENABLE_CLOUD2MESH_DIST_MT
@@ -934,7 +938,7 @@ namespace CCCoreLib
 #if defined(CC_CORE_LIB_USES_QT_CONCURRENT)
 		QMutex currentBitMaskMutex;
 #elif defined(CC_CORE_LIB_USES_TBB)
-		std::mutex currentBitMaskMutex;
+		tbb::mutex currentBitMaskMutex;
 #endif
 	};
 }
@@ -956,8 +960,9 @@ static void CloudMeshDistCellFunc_MT(const DgmOctree::IndexAndCode& desc)
 
 	if (s_multiThreadingWrapper.normProgressCb)
 	{
+#if defined(CC_CORE_LIB_USES_QT_CONCURRENT)
 		QCoreApplication::processEvents(QEventLoop::EventLoopExec); // to allow the GUI to refresh itself
-
+#endif
 		if (!s_multiThreadingWrapper.normProgressCb->oneStep())
 		{
 			s_multiThreadingWrapper.cellFunc_success = false;
@@ -1565,7 +1570,7 @@ int DistanceComputationTools::computePoint2MeshDistancesWithOctree(	const CCVect
 	return DISTANCE_COMPUTATION_RESULTS::SUCCESS;
 }
 
-int DistanceComputationTools::computeCloud2MeshDistancesWithOctree(	const DgmOctree* octree, 
+int DistanceComputationTools::computeCloud2MeshDistancesWithOctree(	const DgmOctree* octree,
 																	const GridAndMeshIntersection& intersection,
 																	Cloud2MeshDistancesComputationParams& params,
 																	GenericProgressCallback* progressCb/*=nullptr*/)
@@ -1819,15 +1824,18 @@ int DistanceComputationTools::computeCloud2MeshDistancesWithOctree(	const DgmOct
 		QThreadPool::globalInstance()->setMaxThreadCount(params.maxThreadCount);
 		QtConcurrent::blockingMap(cellsDescs, CloudMeshDistCellFunc_MT);
 #elif defined(CC_CORE_LIB_USES_TBB)
-		tbb::parallel_for(tbb::blocked_range<int>(0, cellsDescs.size()),
-			[&](tbb::blocked_range<int> r)
+		tbb::task_arena local_arena(params.maxThreadCount != 0 ? params.maxThreadCount : tbb::info::default_concurrency());
+		local_arena.execute([&]
+		{
+			tbb::parallel_for(tbb::blocked_range<std::size_t>(0, cellsDescs.size()),
+			[&](tbb::blocked_range<std::size_t> r)
 			{
-				for (auto i = r.begin(); r.end(); ++i)
+				for (auto i = r.begin(); i < r.end(); ++i)
 				{
 					CloudMeshDistCellFunc_MT(cellsDescs[i]);
 				}
-			}
-		);
+			});
+		});
 #endif
 
 		s_multiThreadingWrapper.octree = nullptr;
@@ -1859,7 +1867,7 @@ int DistanceComputationTools::computeCloud2MeshDistances(	GenericIndexedCloudPer
 															DgmOctree* cloudOctree/*=nullptr*/)
 {
 	//check the input
-	if (!pointCloud) 
+	if (!pointCloud)
 	{
 		assert(false);
 		return DISTANCE_COMPUTATION_RESULTS::ERROR_NULL_COMPAREDCLOUD;
@@ -1921,7 +1929,7 @@ int DistanceComputationTools::computeCloud2MeshDistances(	GenericIndexedCloudPer
 			maxBB.u[k] = std::max(meshMaxBB.u[k], cloudMaxBB.u[k]);
 		}
 
-		//max cubical bounding-box 
+		//max cubical bounding-box
 		cubicalMinBB = minBB;
 		cubicalMaxBB = maxBB;