Prepare to use StarPU and check OMP4

CMakeLists.txt

@@ -64,6 +64,7 @@ OPTION( ScalFMM_USE_AVX              "Set to ON to compile with AVX support"
 OPTION( ScalFMM_USE_ASSERT           "Set to ON to enable safe tests during execution" ON  )
 OPTION( ScalFMM_USE_MIC_NATIVE       "Set to ON to compile in native mode for MIC" OFF  )
 OPTION( ScalFMM_BUILD_ONLY_LIB        "Set to ON to compile only the lib (examples are not compiled) " OFF  )
+OPTION( ScalFMM_USE_STARPU 		"Set to ON to build ScaFMM with StarPU" OFF )
 #OPTION( ScalFMM_ONLY_DEVEL             "Set to ON to compile Development tools (only scalfmm team)" OFF  )
 # Set scalfmm to default libraries
 SET(SCALFMM_LIBRARIES "")

Src/Utils/FGlobal.hpp

@@ -95,5 +95,14 @@ typedef long long MortonIndex;
     #endif
 #endif
 
+
+///////////////////////////////////////////////////////
+// Test OMP4
+///////////////////////////////////////////////////////
+
+#if _OPENMP >= 201307
+#define ScalFMM_USE_OMP4
+#endif
+
 #endif //FGLOBAL_HPP
 

Tests/noDist/testBlockedAlgorithm.cpp

+#include "../../Src/Utils/FGlobal.hpp"
+
 #include "../../Src/GroupTree/FGroupTree.hpp"
 
 #include "../../Src/Components/FSimpleLeaf.hpp"
@@ -29,15 +31,23 @@ int main(int argc, char* argv[]){
         "The size of the block of the blocked tree"
     };
     FHelpDescribeAndExit(argc, argv, "Test the blocked tree by counting the particles.",
-                         FParameterDefinitions::OctreeHeight,
+                         FParameterDefinitions::OctreeHeight, FParameterDefinitions::NbThreads,
                          FParameterDefinitions::NbParticles, LocalOptionBlocSize);
     // Initialize the types
     typedef FTestCell                                                       GroupCellClass;
     typedef FGroupTestParticleContainer                                     GroupContainerClass;
     typedef FGroupTree< GroupCellClass, GroupContainerClass, 2, long long int>  GroupOctreeClass;
     typedef FTestKernels< GroupCellClass, GroupContainerClass >                       GroupKernelClass;
+#ifdef ScalFMM_USE_STARPU
+    typedef FGroupStarPUAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#elif defined(ScalFMM_USE_OMP4)
+    // Set the number of threads
+    omp_set_num_threads(FParameters::getValue(argc,argv,FParameterDefinitions::NbThreads.options, omp_get_max_threads()));
+    typedef FGroupTaskDepAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#else
     //typedef FGroupSeqAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
-typedef FGroupTaskDepAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+    typedef FGroupTaskAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#endif
 
     typedef FTestCell                   CellClass;
     typedef FTestParticleContainer      ContainerClass;

Tests/noDist/testBlockedChebyshev.cpp

+#include "../../Src/Utils/FGlobal.hpp"
 
 #include "../../Src/GroupTree/FGroupTree.hpp"
 
@@ -42,8 +43,6 @@ int main(int argc, char* argv[]){
                          FParameterDefinitions::OctreeHeight,FParameterDefinitions::InputFile,
                          FParameterDefinitions::NbParticles, FParameterDefinitions::NbThreads,
                          LocalOptionBlocSize, LocalOptionNoValidate);
-    // Set the number of threads
-    omp_set_num_threads(FParameters::getValue(argc,argv,FParameterDefinitions::NbThreads.options, omp_get_max_threads()));
 
     // Initialize the types
     static const int ORDER = 6;
@@ -52,9 +51,16 @@ int main(int argc, char* argv[]){
     typedef FP2PGroupParticleContainer<>          GroupContainerClass;
     typedef FGroupTree< GroupCellClass, GroupContainerClass, 5, FReal>  GroupOctreeClass;
     typedef FChebSymKernel<GroupCellClass,GroupContainerClass,MatrixKernelClass,ORDER> GroupKernelClass;
-    //typedef FGroupSeqAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
-    //typedef FGroupTaskAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#ifdef ScalFMM_USE_STARPU
+    typedef FGroupStarPUAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#elif defined(ScalFMM_USE_OMP4)
+    // Set the number of threads
+    omp_set_num_threads(FParameters::getValue(argc,argv,FParameterDefinitions::NbThreads.options, omp_get_max_threads()));
     typedef FGroupTaskDepAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#else
+    //typedef FGroupSeqAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+    typedef FGroupTaskAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#endif
 
     // Get params
     const int NbLevels      = FParameters::getValue(argc,argv,FParameterDefinitions::OctreeHeight.options, 5);

Tests/noDist/testBlockedRotation.cpp

+#include "../../Src/Utils/FGlobal.hpp"
+
 #include "../../Src/GroupTree/FGroupTree.hpp"
 
 #include "../../Src/Components/FSimpleLeaf.hpp"
@@ -41,19 +43,23 @@ int main(int argc, char* argv[]){
                           FParameterDefinitions::NbThreads,
                          FParameterDefinitions::NbParticles, LocalOptionBlocSize, LocalOptionNoValidate);
 
-    // Set the number of threads
-    omp_set_num_threads(FParameters::getValue(argc,argv,FParameterDefinitions::NbThreads.options, omp_get_max_threads()));
-
-
     // Initialize the types
     static const int P = 9;
     typedef FRotationCell<P>               GroupCellClass;
     typedef FP2PGroupParticleContainer<>          GroupContainerClass;
     typedef FGroupTree< GroupCellClass, GroupContainerClass, 5, FReal>  GroupOctreeClass;
     typedef FRotationKernel< GroupCellClass, GroupContainerClass , P>   GroupKernelClass;
-    //typedef FGroupSeqAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
-    //typedef FGroupTaskAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#ifdef ScalFMM_USE_STARPU
+    typedef FGroupStarPUAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#elif defined(ScalFMM_USE_OMP4)
+    // Set the number of threads
+    omp_set_num_threads(FParameters::getValue(argc,argv,FParameterDefinitions::NbThreads.options, omp_get_max_threads()));
     typedef FGroupTaskDepAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#else
+    //typedef FGroupSeqAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+    typedef FGroupTaskAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, GroupCellClass, GroupKernelClass, typename GroupOctreeClass::ParticleGroupClass, GroupContainerClass > GroupAlgorithm;
+#endif
+
 
     // Get params
     const int NbLevels      = FParameters::getValue(argc,argv,FParameterDefinitions::OctreeHeight.options, 5);

Tests/noDist/testBlockedTree.cpp

+#include "../../Src/Utils/FGlobal.hpp"
 
 #include "../../Src/GroupTree/FGroupTree.hpp"
 
@@ -24,6 +25,8 @@
 #include "../../Src/Files/FFmaGenericLoader.hpp"
 
 #include "../../Src/GroupTree/FGroupSeqAlgorithm.hpp"
+#include "../../Src/GroupTree/FGroupTaskAlgorithm.hpp"
+#include "../../Src/GroupTree/FGroupTaskDepAlgorithm.hpp"
 #include "../../Src/GroupTree/FP2PGroupParticleContainer.hpp"
 
 #include "../../Src/Utils/FParameterNames.hpp"
@@ -84,7 +87,14 @@ int main(int argc, char* argv[]){
 
 
     typedef FRotationKernel< CellClass, FP2PGroupParticleContainer<> , P>   KernelClass;
-    typedef FGroupSeqAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, CellClass, KernelClass, typename GroupOctreeClass::ParticleGroupClass, FP2PGroupParticleContainer<> > GroupAlgorithm;
+#ifdef ScalFMM_USE_STARPU
+    typedef FGroupStarPUAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, CellClass, KernelClass, typename GroupOctreeClass::ParticleGroupClass, FP2PGroupParticleContainer<> > GroupAlgorithm;
+#elif defined(ScalFMM_USE_OMP4)
+    typedef FGroupTaskDepAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, CellClass, KernelClass, typename GroupOctreeClass::ParticleGroupClass, FP2PGroupParticleContainer<> > GroupAlgorithm;
+#else
+    //typedef FGroupSeqAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, CellClass, KernelClass, typename GroupOctreeClass::ParticleGroupClass, FP2PGroupParticleContainer<> > GroupAlgorithm;
+    typedef FGroupTaskAlgorithm<GroupOctreeClass, typename GroupOctreeClass::CellGroupClass, CellClass, KernelClass, typename GroupOctreeClass::ParticleGroupClass, FP2PGroupParticleContainer<> > GroupAlgorithm;
+#endif
 
     KernelClass kernel(NbLevels, loader.getBoxWidth(), loader.getCenterOfBox());
     GroupAlgorithm algo(&groupedTree2,&kernel);