Added depend in update from clauses

Author: iomaganaris

coreneuron/network/netcvode.cpp

@@ -571,7 +571,8 @@ void NetCvode::check_thresh(NrnThread* nt) {  // for default method
 #endif
         nrn_pragma_acc(update host(nsbuffer [0:nt->_net_send_buffer_cnt]) async(nt->streams[nt->stream_id]))
         nrn_pragma_acc(wait async(nt->streams[nt->stream_id]))
-        nrn_pragma_omp(target update from(nsbuffer [0:nt->_net_send_buffer_cnt]))
+        nrn_pragma_omp(target update from(nsbuffer [0:nt->_net_send_buffer_cnt]) depend(inout: nt->streams[nt->stream_id]) nowait)
+        nrn_pragma_omp(taskwait)
     }
 
     // on CPU...

coreneuron/network/partrans.cpp

@@ -63,7 +63,7 @@ void nrnmpi_v_transfer() {
         }
         nrn_pragma_acc(update host(src_gather [0:n_src_gather]) if (nt->compute_gpu)
                            async(nt->streams[nt->stream_id]))
-        nrn_pragma_omp(target update from(src_gather [0:n_src_gather]) if (nt->compute_gpu))
+        nrn_pragma_omp(target update from(src_gather [0:n_src_gather]) if (nt->compute_gpu) depend(inout: nt->streams[nt->stream_id]) nowait)
     }
 
     // copy gathered source values to outsrc_buf_

coreneuron/sim/fadvance_core.cpp

@@ -317,10 +317,9 @@ void nrncore2nrn_send_values(NrnThread* nth) {
                 double* gather_i = tr->gather[i];
                 nrn_pragma_acc(update self(gather_i [0:1]) if (nth->compute_gpu)
                                    async(nth->streams[nth->stream_id]))
-                nrn_pragma_omp(target update from(gather_i [0:1]) if (nth->compute_gpu))
+                nrn_pragma_omp(target update from(gather_i [0:1]) if (nth->compute_gpu) depend(inout: nth->streams[nth->stream_id]) nowait)
             }
             nrn_pragma_acc(wait async(nth->streams[nth->stream_id]))
-            nrn_pragma_omp(taskwait)
             for (int i = 0; i < tr->n_trajec; ++i) {
                 *(tr->scatter[i]) = *(tr->gather[i]);
             }
@@ -345,7 +344,6 @@ static void* nrn_fixed_step_thread(NrnThread* nth) {
         nrn_pragma_acc(update device(nth->_t) if (nth->compute_gpu) async(nth->streams[nth->stream_id]))
         nrn_pragma_acc(wait async(nth->streams[nth->stream_id]))
         nrn_pragma_omp(target update to(nth->_t) if (nth->compute_gpu))
-        nrn_pragma_omp(taskwait)
         fixed_play_continuous(nth);
 
         {
@@ -382,7 +380,6 @@ void* nrn_fixed_step_lastpart(NrnThread* nth) {
         nrn_pragma_acc(update device(nth->_t) if (nth->compute_gpu) async(nth->streams[nth->stream_id]))
         nrn_pragma_acc(wait async(nth->streams[nth->stream_id]))
         nrn_pragma_omp(target update to(nth->_t) if (nth->compute_gpu))
-        nrn_pragma_omp(taskwait)
         fixed_play_continuous(nth);
         nonvint(nth);
         nrncore2nrn_send_values(nth);