When using starpu number of threads should be given by env variables (and not...

When using starpu number of threads should be given by env variables (and not command line parameter)

Src/GroupTree/Core/FGroupTaskAlgorithm.hpp

@@ -41,8 +41,14 @@ public:
         FAssertLF(inKernels, "kernels cannot be null");
 
         kernels = new KernelClass*[MaxThreads];
+        #pragma omp parallel for schedule(static) num_threads(MaxThreads)
         for(int idxThread = 0 ; idxThread < MaxThreads ; ++idxThread){
-            this->kernels[idxThread] = new KernelClass(*inKernels);
+            // We want to ensure that each thread allocate data close to him
+            // and that only one thread at a time call the copy constructor
+            #pragma omp critical (FGroupTaskDepAlgorithm_InitKernels)
+            {
+                this->kernels[idxThread] = new KernelClass(*inKernels);
+            }
         }
 
         FLOG(FLog::Controller << "FGroupTaskAlgorithm (Max Thread " << MaxThreads << ")\n");
@@ -58,7 +64,7 @@ public:
     void execute(const unsigned operationsToProceed = FFmmNearAndFarFields){
         FLOG( FLog::Controller << "\tStart FGroupTaskAlgorithm\n" );
 
-        #pragma omp parallel
+        #pragma omp parallel num_threads(MaxThreads)
         {
             #pragma omp single nowait
             {
@@ -67,7 +73,7 @@ public:
             }
         }
 
-        #pragma omp parallel
+        #pragma omp parallel num_threads(MaxThreads)
         {
             #pragma omp single nowait
             {

Src/GroupTree/Core/FGroupTaskDepAlgorithm.hpp

@@ -43,7 +43,7 @@ public:
         FAssertLF(inKernels, "kernels cannot be null");
 
         kernels = new KernelClass*[MaxThreads];
-        #pragma omp parallel for schedule(static)
+        #pragma omp parallel for schedule(static) num_threads(MaxThreads)
         for(int idxThread = 0 ; idxThread < MaxThreads ; ++idxThread){
             // We want to ensure that each thread allocate data close to him
             // and that only one thread at a time call the copy constructor
@@ -66,7 +66,7 @@ public:
     void execute(const unsigned operationsToProceed = FFmmNearAndFarFields){
         FLOG( FLog::Controller << "\tStart FGroupTaskDepAlgorithm\n" );
 
-        #pragma omp parallel
+        #pragma omp parallel num_threads(MaxThreads)
         {
             #pragma omp single nowait
             {
@@ -75,7 +75,7 @@ public:
             }
         }
 
-        #pragma omp parallel
+        #pragma omp parallel num_threads(MaxThreads)
         {
             #pragma omp single nowait
             {

Src/GroupTree/Core/FGroupTaskStarpuAlgorithm.hpp

@@ -80,7 +80,6 @@ protected:
     std::vector< std::vector< std::vector<BlockInteractions<CellContainerClass>>>> externalInteractionsAllLevel;
     std::vector< std::vector<BlockInteractions<ParticleGroupClass>>> externalInteractionsLeafLevel;
 
-    int MaxThreads;         //< The number of threads
     OctreeClass*const tree;       //< The Tree
     KernelClass*const originalCpuKernel;
 
@@ -112,8 +111,8 @@ protected:
     FStarPUPtrInterface* wrapperptr;
 
 public:
-    FGroupTaskStarPUAlgorithm(OctreeClass*const inTree, KernelClass* inKernels, const int inMaxThreads = -1)
-        : MaxThreads(inMaxThreads), tree(inTree), originalCpuKernel(inKernels),
+    FGroupTaskStarPUAlgorithm(OctreeClass*const inTree, KernelClass* inKernels)
+        : tree(inTree), originalCpuKernel(inKernels),
           cellHandles(nullptr),
 #ifdef STARPU_USE_CPU
             cpuWrapper(tree->getHeight()),
@@ -127,7 +126,6 @@ public:
             wrapperptr(&wrappers){
         FAssertLF(tree, "tree cannot be null");
         FAssertLF(inKernels, "kernels cannot be null");
-        FAssertLF(MaxThreads <= STARPU_MAXCPUS, "number of threads to high");
 
         struct starpu_conf conf;
         FAssertLF(starpu_conf_init(&conf) == 0);
@@ -164,8 +162,6 @@ public:
 
         starpu_pause();
 
-        MaxThreads = starpu_worker_get_count();//starpu_cpu_worker_get_count();
-
         cellHandles   = new std::vector<CellHandles>[tree->getHeight()];
 
         initCodelet();

Src/GroupTree/Core/FGroupTaskStarpuMpiAlgorithm.hpp

@@ -95,7 +95,6 @@ protected:
     std::vector< std::vector< std::vector<BlockInteractions<CellContainerClass>>>> externalInteractionsAllLevel;
     std::vector< std::vector<BlockInteractions<ParticleGroupClass>>> externalInteractionsLeafLevel;
 
-    int MaxThreads;         //< The number of threads
     OctreeClass*const tree;       //< The Tree
     KernelClass*const originalCpuKernel;
 
@@ -129,8 +128,8 @@ protected:
     FStarPUPtrInterface* wrapperptr;
 
 public:
-    FGroupTaskStarPUMpiAlgorithm(const FMpi::FComm& inComm, OctreeClass*const inTree, KernelClass* inKernels, const int inMaxThreads = -1)
-        :   comm(inComm), MaxThreads(inMaxThreads), tree(inTree), originalCpuKernel(inKernels),
+    FGroupTaskStarPUMpiAlgorithm(const FMpi::FComm& inComm, OctreeClass*const inTree, KernelClass* inKernels)
+        :   comm(inComm), tree(inTree), originalCpuKernel(inKernels),
           cellHandles(nullptr),
 #ifdef STARPU_USE_CPU
             cpuWrapper(tree->getHeight()),
@@ -144,7 +143,6 @@ public:
             wrapperptr(&wrappers){
         FAssertLF(tree, "tree cannot be null");
         FAssertLF(inKernels, "kernels cannot be null");
-        FAssertLF(MaxThreads <= STARPU_MAXCPUS, "number of threads to high");
 
         struct starpu_conf conf;
         FAssertLF(starpu_conf_init(&conf) == 0);
@@ -182,8 +180,6 @@ public:
 
         starpu_pause();
 
-        MaxThreads = starpu_worker_get_count();//starpu_cpu_worker_get_count();
-
         cellHandles   = new std::vector<CellHandles>[tree->getHeight()];
 
         initCodelet();