If we ingnore the viria, it looks like we are here

plumed · Feb 21, 2025 · 8f15f0d · 8f15f0d
1 parent ab63e35
commit 8f15f0d
Showing 1 changed file with 16 additions and 9 deletions.
diff --git a/src/core/ParallelTaskManager.h b/src/core/ParallelTaskManager.h
@@ -389,7 +389,7 @@ void ParallelTaskManager<T, D>::applyForces( std::vector<double>& forcesForApply
     }
   }
 
-  if( useacc &&false ) {
+  if( useacc ) {
 #ifdef __PLUMED_HAS_OPENACC
     omp_forces[0].assign( omp_forces[0].size(), 0.0 );
     auto value_stash_data = value_stash.data();
@@ -405,8 +405,14 @@ void ParallelTaskManager<T, D>::applyForces( std::vector<double>& forcesForApply
     const auto ndev_per_task = nderivatives_per_task;
     const auto ncomponents = myinput.ncomponents;
 
-    // std::vector<double> derivatives( nderivs );
-    double * derivatives = nullptr;// derivatives.data();
+    //To future me/you:
+    // I need to allocate this on the host to create a bigger temporay data array
+    // on the device 
+    // by trying with double* x=nullptr, you will get a failure
+    // antother solution is acc_malloc and then device_ptr in the pragma 
+    // (but you have to remember the acc_free)
+    std::vector<double> derivative(0);
+    double * derivatives = derivative.data();
 
     ParallelActionsInput input = myinput;
     auto myinput_acc = fromToDataHelper(input);
@@ -429,16 +435,17 @@ void ParallelTaskManager<T, D>::applyForces( std::vector<double>& forcesForApply
         std::vector<double> fake_valstmp( input.ncomponents );
         std::size_t task_number = partialTaskList_data[i];
         std::size_t val_pos = task_number*input.ncomponents;
-        ParallelActionsOutput myout( input.ncomponents, fake_valstmp.data(), nderivs,derivatives );
+        // std::vector<double> derivative( nderivs );
+        ParallelActionsOutput myout( input.ncomponents, fake_valstmp.data(), nderivs,derivatives);
         // Calculate the stuff in the loop for this action
         T::performTask( task_number, t_actiondata, input, myout );
 
         // Gather the forces from the values
-        // T::gatherForces<false>( task_number, t_actiondata, input,
-        //                         ForceInput( input.ncomponents,
-        //                                     value_stash_data+input.ncomponents*task_number,
-        //                                     ndev_per_task, derivatives),
-        //                         ForceOutput { omp_forces_data,of_size, forcesForApply_data,ffa_size } );
+        T::gatherForces<false>( task_number, t_actiondata, input,
+                                ForceInput( input.ncomponents,
+                                            value_stash_data+input.ncomponents*task_number,
+                                            ndev_per_task, derivatives),
+                                ForceOutput { omp_forces_data,of_size, forcesForApply_data,ffa_size } );
 
       }
       // #pragma acc parallel loop