Added stream id vector in NrnThread

Author: iomaganaris

coreneuron/io/phase2.cpp

@@ -875,7 +875,9 @@ void Phase2::populate(NrnThread& nt, const UserParams& userParams) {
  * greater than number of omp threads.
  */
 #if defined(_OPENMP)
-    nt.stream_id = omp_get_thread_num();
+    const auto omp_thread_num = omp_get_thread_num();
+    nt.stream_id = omp_thread_num;
+    nt.streams[nt.stream_id] = omp_thread_num;
 #endif
 
     int shadow_rhs_cnt = 0;

coreneuron/sim/multicore.cpp

@@ -106,6 +106,9 @@ void nrn_threads_create(int n) {
             for (int i = 0; i < nrn_nthread; ++i) {
                 NrnThread& nt = nrn_threads[i];
                 nt.id = i;
+#if defined(_OPENMP)
+                nt.streams.reserve(omp_get_num_threads());
+#endif
                 for (int j = 0; j < BEFORE_AFTER_SIZE; ++j) {
                     nt.tbl[j] = nullptr;
                 }

coreneuron/sim/multicore.hpp

@@ -130,9 +130,10 @@ struct NrnThread: public MemoryManaged {
 
     NrnThreadBAList* tbl[BEFORE_AFTER_SIZE]; /* wasteful since almost all empty */
 
-    int shadow_rhs_cnt = 0; /* added to facilitate the NrnThread transfer to GPU */
-    int compute_gpu = 0;    /* define whether to compute with gpus */
-    int stream_id = 0;      /* define where the kernel will be launched on GPU stream */
+    int shadow_rhs_cnt = 0;    /* added to facilitate the NrnThread transfer to GPU */
+    int compute_gpu = 0;       /* define whether to compute with gpus */
+    int stream_id = 0;         /* define where the kernel will be launched on GPU stream */
+    std::vector<int> streams;  /* vector of stream ids needed for async execution of OpenMP in multiple streams */
     int _net_send_buffer_size = 0;
     int _net_send_buffer_cnt = 0;
     int* _net_send_buffer = nullptr;