My configuration
[c11tester.git] / execution.cc
index d6988edf3096008288213af4ff20f8b096d1409b..8853bfeefee1a44a7a05e220999ceb057ecc385e 100644 (file)
@@ -63,13 +63,17 @@ ModelExecution::ModelExecution(ModelChecker *m, Scheduler *scheduler) :
        thrd_last_fence_release(),
        priv(new struct model_snapshot_members ()),
        mo_graph(new CycleGraph()),
+#ifdef NEWFUZZER
        fuzzer(new NewFuzzer()),
+#else
+       fuzzer(new Fuzzer()),
+#endif
        isfinished(false)
 {
        /* Initialize a model-checker thread, for special ModelActions */
        model_thread = new Thread(get_next_id());
        add_thread(model_thread);
-       fuzzer->register_engine(m->get_history(), this);
+       fuzzer->register_engine(this);
        scheduler->register_engine(this);
 #ifdef TLS
        pthread_key_create(&pthreadkey, tlsdestructor);
@@ -129,6 +133,12 @@ modelclock_t ModelExecution::get_next_seq_num()
        return ++priv->used_sequence_numbers;
 }
 
+/** @return a sequence number for a new ModelAction */
+modelclock_t ModelExecution::get_curr_seq_num()
+{
+       return priv->used_sequence_numbers;
+}
+
 /** Restore the last used sequence number when actions of a thread are postponed by Fuzzer */
 void ModelExecution::restore_last_seq_num()
 {
@@ -269,7 +279,9 @@ ModelAction * ModelExecution::convertNonAtomicStore(void * location) {
        add_normal_write_to_lists(act);
        add_write_to_lists(act);
        w_modification_order(act);
+#ifdef NEWFUZZER
        model->get_history()->process_action(act, act->get_tid());
+#endif
        return act;
 }
 
@@ -277,11 +289,11 @@ ModelAction * ModelExecution::convertNonAtomicStore(void * location) {
  * Processes a read model action.
  * @param curr is the read model action to process.
  * @param rf_set is the set of model actions we can possibly read from
- * @return True if processing this read updates the mo_graph.
+ * @return True if the read can be pruned from the thread map list.
  */
 bool ModelExecution::process_read(ModelAction *curr, SnapVector<ModelAction *> * rf_set)
 {
-       SnapVector<const ModelAction *> * priorset = new SnapVector<const ModelAction *>();
+       SnapVector<ModelAction *> * priorset = new SnapVector<ModelAction *>();
        bool hasnonatomicstore = hasNonAtomicStore(curr->get_location());
        if (hasnonatomicstore) {
                ModelAction * nonatomicstore = convertNonAtomicStore(curr->get_location());
@@ -289,15 +301,16 @@ bool ModelExecution::process_read(ModelAction *curr, SnapVector<ModelAction *> *
        }
 
        // Remove writes that violate read modification order
-       uint i = 0;
-       while (i < rf_set->size()) {
-               ModelAction * rf = (*rf_set)[i];
-               if (!r_modification_order(curr, rf, NULL, NULL, true)) {
-                       (*rf_set)[i] = rf_set->back();
-                       rf_set->pop_back();
-               } else
-                       i++;
-       }
+       /*
+          uint i = 0;
+          while (i < rf_set->size()) {
+               ModelAction * rf = (*rf_set)[i];
+               if (!r_modification_order(curr, rf, NULL, NULL, true)) {
+                       (*rf_set)[i] = rf_set->back();
+                       rf_set->pop_back();
+               } else
+                       i++;
+          }*/
 
        while(true) {
                int index = fuzzer->selectWrite(curr, rf_set);
@@ -313,16 +326,8 @@ bool ModelExecution::process_read(ModelAction *curr, SnapVector<ModelAction *> *
                        read_from(curr, rf);
                        get_thread(curr)->set_return_value(curr->get_return_value());
                        delete priorset;
-                       if (canprune && curr->get_type() == ATOMIC_READ) {
-                               int tid = id_to_int(curr->get_tid());
-                               (*obj_thrd_map.get(curr->get_location()))[tid].pop_back();
-                               curr->setThrdMapRef(NULL);
-                       }
-                       return true;
+                       return canprune && (curr->get_type() == ATOMIC_READ);
                }
-
-               ASSERT(false);
-               /* TODO: Following code not needed anymore */
                priorset->clear();
                (*rf_set)[index] = rf_set->back();
                rf_set->pop_back();
@@ -378,18 +383,19 @@ bool ModelExecution::process_mutex(ModelAction *curr)
                break;
        }
        case ATOMIC_WAIT: {
-               /* wake up the other threads */
-               for (unsigned int i = 0;i < get_num_threads();i++) {
-                       Thread *t = get_thread(int_to_id(i));
-                       Thread *curr_thrd = get_thread(curr);
-                       if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
-                               scheduler->wake(t);
-               }
+               //TODO: DOESN'T REALLY IMPLEMENT SPURIOUS WAKEUPS CORRECTLY
+               if (fuzzer->shouldWait(curr)) {
+                       /* wake up the other threads */
+                       for (unsigned int i = 0;i < get_num_threads();i++) {
+                               Thread *t = get_thread(int_to_id(i));
+                               Thread *curr_thrd = get_thread(curr);
+                               if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
+                                       scheduler->wake(t);
+                       }
 
-               /* unlock the lock - after checking who was waiting on it */
-               state->locked = NULL;
+                       /* unlock the lock - after checking who was waiting on it */
+                       state->locked = NULL;
 
-               if (fuzzer->shouldWait(curr)) {
                        /* disable this thread */
                        get_safe_ptr_action(&condvar_waiters_map, curr->get_location())->push_back(curr);
                        scheduler->sleep(get_thread(curr));
@@ -399,7 +405,14 @@ bool ModelExecution::process_mutex(ModelAction *curr)
        }
        case ATOMIC_TIMEDWAIT:
        case ATOMIC_UNLOCK: {
-               //TODO: FIX WAIT SITUATION...WAITS CAN SPURIOUSLY FAIL...TIMED WAITS SHOULD PROBABLY JUST BE THE SAME AS NORMAL WAITS...THINK ABOUT PROBABILITIES THOUGH....AS IN TIMED WAIT MUST FAIL TO GUARANTEE PROGRESS...NORMAL WAIT MAY FAIL...SO NEED NORMAL WAIT TO WORK CORRECTLY IN THE CASE IT SPURIOUSLY FAILS AND IN THE CASE IT DOESN'T...  TIMED WAITS MUST EVENMTUALLY RELEASE...
+               //TODO: FIX WAIT SITUATION...WAITS CAN SPURIOUSLY
+               //FAIL...TIMED WAITS SHOULD PROBABLY JUST BE THE SAME
+               //AS NORMAL WAITS...THINK ABOUT PROBABILITIES
+               //THOUGH....AS IN TIMED WAIT MUST FAIL TO GUARANTEE
+               //PROGRESS...NORMAL WAIT MAY FAIL...SO NEED NORMAL
+               //WAIT TO WORK CORRECTLY IN THE CASE IT SPURIOUSLY
+               //FAILS AND IN THE CASE IT DOESN'T...  TIMED WAITS
+               //MUST EVENMTUALLY RELEASE...
 
                /* wake up the other threads */
                for (unsigned int i = 0;i < get_num_threads();i++) {
@@ -716,15 +729,17 @@ ModelAction * ModelExecution::check_current_action(ModelAction *curr)
        if (newly_explored && curr->is_read())
                rf_set = build_may_read_from(curr);
 
+       bool canprune = false;
+
        if (curr->is_read() && !second_part_of_rmw) {
-               process_read(curr, rf_set);
+               canprune = process_read(curr, rf_set);
                delete rf_set;
        } else
                ASSERT(rf_set == NULL);
 
        /* Add the action to lists */
        if (!second_part_of_rmw)
-               add_action_to_lists(curr);
+               add_action_to_lists(curr, canprune);
 
        if (curr->is_write())
                add_write_to_lists(curr);
@@ -774,19 +789,27 @@ ModelAction * ModelExecution::process_rmw(ModelAction *act) {
  */
 
 bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *rf,
-                                                                                                                                                                       SnapVector<const ModelAction *> * priorset, bool * canprune, bool check_only)
+                                                                                                                                                                       SnapVector<ModelAction *> * priorset, bool * canprune, bool check_only)
 {
        SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
-       unsigned int i;
        ASSERT(curr->is_read());
 
        /* Last SC fence in the current thread */
        ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
 
        int tid = curr->get_tid();
+
+       /* Need to ensure thrd_lists is big enough because we have not added the curr actions yet.  */
+       if ((int)thrd_lists->size() <= tid) {
+               uint oldsize = thrd_lists->size();
+               thrd_lists->resize(priv->next_thread_id);
+               for(uint i = oldsize;i < priv->next_thread_id;i++)
+                       new (&(*thrd_lists)[i]) action_list_t();
+       }
+
        ModelAction *prev_same_thread = NULL;
        /* Iterate over all threads */
-       for (i = 0;i < thrd_lists->size();i++, tid = (((unsigned int)(tid+1)) == thrd_lists->size()) ? 0 : tid + 1) {
+       for (unsigned int i = 0;i < thrd_lists->size();i++, tid = (((unsigned int)(tid+1)) == thrd_lists->size()) ? 0 : tid + 1) {
                /* Last SC fence in thread tid */
                ModelAction *last_sc_fence_thread_local = NULL;
                if (i != 0)
@@ -868,7 +891,7 @@ bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *
                                        if (!check_only)
                                                priorset->push_back(act);
                                } else {
-                                       const ModelAction *prevrf = act->get_reads_from();
+                                       ModelAction *prevrf = act->get_reads_from();
                                        if (!prevrf->equals(rf)) {
                                                if (mo_graph->checkReachable(rf, prevrf))
                                                        return false;
@@ -1098,7 +1121,7 @@ ClockVector * ModelExecution::get_hb_from_write(ModelAction *rf) const {
  *
  * @param act is the ModelAction to add.
  */
-void ModelExecution::add_action_to_lists(ModelAction *act)
+void ModelExecution::add_action_to_lists(ModelAction *act, bool canprune)
 {
        int tid = id_to_int(act->get_tid());
        if ((act->is_fence() && act->is_seqcst()) || act->is_unlock()) {
@@ -1118,7 +1141,8 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
                for(uint i = oldsize;i < priv->next_thread_id;i++)
                        new (&(*vec)[i]) action_list_t();
        }
-       act->setThrdMapRef((*vec)[tid].add_back(act));
+       if (!canprune && (act->is_read() || act->is_write()))
+               act->setThrdMapRef((*vec)[tid].add_back(act));
 
        // Update thrd_last_action, the last action taken by each thread
        if ((int)thrd_last_action.size() <= tid)
@@ -1135,30 +1159,21 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
        if (act->is_wait()) {
                void *mutex_loc = (void *) act->get_value();
                act->setActionRef(get_safe_ptr_action(&obj_map, mutex_loc)->add_back(act));
-
-               SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, mutex_loc);
-               if ((int)vec->size() <= tid) {
-                       uint oldsize = vec->size();
-                       vec->resize(priv->next_thread_id);
-                       for(uint i = oldsize;i < priv->next_thread_id;i++)
-                               new (&(*vec)[i]) action_list_t();
-               }
-               act->setThrdMapRef((*vec)[tid].add_back(act));
        }
 }
 
 sllnode<ModelAction *>* insertIntoActionList(action_list_t *list, ModelAction *act) {
        sllnode<ModelAction*> * rit = list->end();
        modelclock_t next_seq = act->get_seq_number();
-       if (rit == NULL || (rit->getVal()->get_seq_number() == next_seq))
+       if (rit == NULL || (rit->getVal()->get_seq_number() <= next_seq))
                return list->add_back(act);
        else {
                for(;rit != NULL;rit=rit->getPrev()) {
-                       if (rit->getVal()->get_seq_number() == next_seq) {
+                       if (rit->getVal()->get_seq_number() <= next_seq) {
                                return list->insertAfter(rit, act);
                        }
                }
-               return NULL;
+               return list->add_front(act);
        }
 }
 
@@ -1167,18 +1182,18 @@ sllnode<ModelAction *>* insertIntoActionListAndSetCV(action_list_t *list, ModelA
        modelclock_t next_seq = act->get_seq_number();
        if (rit == NULL) {
                act->create_cv(NULL);
-               return NULL;
-       } else if (rit->getVal()->get_seq_number() == next_seq) {
+               return list->add_back(act);
+       } else if (rit->getVal()->get_seq_number() <= next_seq) {
                act->create_cv(rit->getVal());
                return list->add_back(act);
        } else {
                for(;rit != NULL;rit=rit->getPrev()) {
-                       if (rit->getVal()->get_seq_number() == next_seq) {
+                       if (rit->getVal()->get_seq_number() <= next_seq) {
                                act->create_cv(rit->getVal());
                                return list->insertAfter(rit, act);
                        }
                }
-               return NULL;
+               return list->add_front(act);
        }
 }
 
@@ -1205,8 +1220,9 @@ void ModelExecution::add_normal_write_to_lists(ModelAction *act)
        }
        act->setThrdMapRef(insertIntoActionList(&(*vec)[tid],act));
 
+       ModelAction * lastact = thrd_last_action[tid];
        // Update thrd_last_action, the last action taken by each thrad
-       if (thrd_last_action[tid]->get_seq_number() == act->get_seq_number())
+       if (lastact == NULL || lastact->get_seq_number() == act->get_seq_number())
                thrd_last_action[tid] = act;
 }
 
@@ -1380,46 +1396,47 @@ SnapVector<ModelAction *> *  ModelExecution::build_may_read_from(ModelAction *cu
        SnapVector<ModelAction *> * rf_set = new SnapVector<ModelAction *>();
 
        /* Iterate over all threads */
-       for (i = 0;i < thrd_lists->size();i++) {
-               /* Iterate over actions in thread, starting from most recent */
-               action_list_t *list = &(*thrd_lists)[i];
-               sllnode<ModelAction *> * rit;
-               for (rit = list->end();rit != NULL;rit=rit->getPrev()) {
-                       ModelAction *act = rit->getVal();
-
-                       if (act == curr)
-                               continue;
+       if (thrd_lists != NULL)
+               for (i = 0;i < thrd_lists->size();i++) {
+                       /* Iterate over actions in thread, starting from most recent */
+                       action_list_t *list = &(*thrd_lists)[i];
+                       sllnode<ModelAction *> * rit;
+                       for (rit = list->end();rit != NULL;rit=rit->getPrev()) {
+                               ModelAction *act = rit->getVal();
+
+                               if (act == curr)
+                                       continue;
 
-                       /* Don't consider more than one seq_cst write if we are a seq_cst read. */
-                       bool allow_read = true;
-
-                       if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
-                               allow_read = false;
-
-                       /* Need to check whether we will have two RMW reading from the same value */
-                       if (curr->is_rmwr()) {
-                               /* It is okay if we have a failing CAS */
-                               if (!curr->is_rmwrcas() ||
-                                               valequals(curr->get_value(), act->get_value(), curr->getSize())) {
-                                       //Need to make sure we aren't the second RMW
-                                       CycleNode * node = mo_graph->getNode_noCreate(act);
-                                       if (node != NULL && node->getRMW() != NULL) {
-                                               //we are the second RMW
-                                               allow_read = false;
+                               /* Don't consider more than one seq_cst write if we are a seq_cst read. */
+                               bool allow_read = true;
+
+                               if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
+                                       allow_read = false;
+
+                               /* Need to check whether we will have two RMW reading from the same value */
+                               if (curr->is_rmwr()) {
+                                       /* It is okay if we have a failing CAS */
+                                       if (!curr->is_rmwrcas() ||
+                                                       valequals(curr->get_value(), act->get_value(), curr->getSize())) {
+                                               //Need to make sure we aren't the second RMW
+                                               CycleNode * node = mo_graph->getNode_noCreate(act);
+                                               if (node != NULL && node->getRMW() != NULL) {
+                                                       //we are the second RMW
+                                                       allow_read = false;
+                                               }
                                        }
                                }
-                       }
 
-                       if (allow_read) {
-                               /* Only add feasible reads */
-                               rf_set->push_back(act);
-                       }
+                               if (allow_read) {
+                                       /* Only add feasible reads */
+                                       rf_set->push_back(act);
+                               }
 
-                       /* Include at most one act per-thread that "happens before" curr */
-                       if (act->happens_before(curr))
-                               break;
+                               /* Include at most one act per-thread that "happens before" curr */
+                               if (act->happens_before(curr))
+                                       break;
+                       }
                }
-       }
 
        if (DBG_ENABLED()) {
                model_print("Reached read action:\n");
@@ -1632,14 +1649,17 @@ Thread * ModelExecution::take_step(ModelAction *curr)
        ASSERT(curr);
 
        /* Process this action in ModelHistory for records */
+#ifdef NEWFUZZER
        model->get_history()->process_action( curr, curr->get_tid() );
-
+#endif
        if (curr_thrd->is_blocked() || curr_thrd->is_complete())
                scheduler->remove_thread(curr_thrd);
 
        return action_select_next_thread(curr);
 }
 
+/** This method removes references to an Action before we delete it. */
+
 void ModelExecution::removeAction(ModelAction *act) {
        {
                sllnode<ModelAction *> * listref = act->getTraceRef();
@@ -1666,14 +1686,242 @@ void ModelExecution::removeAction(ModelAction *act) {
                        void *mutex_loc = (void *) act->get_value();
                        get_safe_ptr_action(&obj_map, mutex_loc)->erase(listref);
                }
-       } else if (act->is_write()) {
+       } else if (act->is_free()) {
                sllnode<ModelAction *> * listref = act->getActionRef();
                if (listref != NULL) {
                        SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_wr_thrd_map, act->get_location());
                        (*vec)[act->get_tid()].erase(listref);
                }
+               //Clear it from last_sc_map
+               if (obj_last_sc_map.get(act->get_location()) == act) {
+                       obj_last_sc_map.remove(act->get_location());
+               }
+
+               //Remove from Cyclegraph
+               mo_graph->freeAction(act);
+       }
+}
+
+/** Computes clock vector that all running threads have already synchronized to.  */
+
+ClockVector * ModelExecution::computeMinimalCV() {
+       ClockVector *cvmin = NULL;
+       //Thread 0 isn't a real thread, so skip it..
+       for(unsigned int i = 1;i < thread_map.size();i++) {
+               Thread * t = thread_map[i];
+               if (t->get_state() == THREAD_COMPLETED)
+                       continue;
+               thread_id_t tid = int_to_id(i);
+               ClockVector * cv = get_cv(tid);
+               if (cvmin == NULL)
+                       cvmin = new ClockVector(cv, NULL);
+               else
+                       cvmin->minmerge(cv);
+       }
+       return cvmin;
+}
+
+
+/** Sometimes we need to remove an action that is the most recent in the thread.  This happens if it is mo before action in other threads.  In that case we need to create a replacement latest ModelAction */
+
+void ModelExecution::fixupLastAct(ModelAction *act) {
+       ModelAction *newact = new ModelAction(ATOMIC_NOP, std::memory_order_seq_cst, get_thread(act->get_tid()));
+       newact->set_seq_number(get_next_seq_num());
+       newact->create_cv(act);
+       newact->set_last_fence_release(act->get_last_fence_release());
+       add_action_to_lists(newact, false);
+}
+
+/** Compute which actions to free.  */
+
+void ModelExecution::collectActions() {
+       //Compute minimal clock vector for all live threads
+       ClockVector *cvmin = computeMinimalCV();
+       SnapVector<CycleNode *> * queue = new SnapVector<CycleNode *>();
+       modelclock_t maxtofree = priv->used_sequence_numbers - params->traceminsize;
+
+       //Next walk action trace...  When we hit an action, see if it is
+       //invisible (e.g., earlier than the first before the minimum
+       //clock for the thread...  if so erase it and all previous
+       //actions in cyclegraph
+       sllnode<ModelAction*> * it;
+       for (it = action_trace.begin();it != NULL;it=it->getNext()) {
+               ModelAction *act = it->getVal();
+               modelclock_t actseq = act->get_seq_number();
+
+               //See if we are done
+               if (actseq > maxtofree)
+                       break;
+
+               thread_id_t act_tid = act->get_tid();
+               modelclock_t tid_clock = cvmin->getClock(act_tid);
+
+               //Free if it is invisible or we have set a flag to remove visible actions.
+               if (actseq <= tid_clock || params->removevisible) {
+                       ModelAction * write;
+                       if (act->is_write()) {
+                               write = act;
+                       } else if (act->is_read()) {
+                               write = act->get_reads_from();
+                       } else
+                               continue;
+
+                       //Mark everything earlier in MO graph to be freed
+                       CycleNode * cn = mo_graph->getNode_noCreate(write);
+                       if (cn != NULL) {
+                               queue->push_back(cn);
+                               while(!queue->empty()) {
+                                       CycleNode * node = queue->back();
+                                       queue->pop_back();
+                                       for(unsigned int i=0;i<node->getNumInEdges();i++) {
+                                               CycleNode * prevnode = node->getInEdge(i);
+                                               ModelAction * prevact = prevnode->getAction();
+                                               if (prevact->get_type() != READY_FREE) {
+                                                       prevact->set_free();
+                                                       queue->push_back(prevnode);
+                                               }
+                                       }
+                               }
+                       }
+               }
+       }
+
+       //We may need to remove read actions in the window we don't delete to preserve correctness.
+
+       for (sllnode<ModelAction*> * it2 = action_trace.end();it2 != it;) {
+               ModelAction *act = it2->getVal();
+               //Do iteration early in case we delete the act
+               it2=it2->getPrev();
+               bool islastact = false;
+               ModelAction *lastact = get_last_action(act->get_tid());
+               if (act == lastact) {
+                       Thread * th = get_thread(act);
+                       islastact = !th->is_complete();
+               }
+
+               if (act->is_read()) {
+                       if (act->get_reads_from()->is_free()) {
+                               if (act->is_rmw()) {
+                                       //Weaken a RMW from a freed store to a write
+                                       act->set_type(ATOMIC_WRITE);
+                               } else {
+                                       removeAction(act);
+                                       if (islastact) {
+                                               fixupLastAct(act);
+                                       }
+                                       delete act;
+                                       continue;
+                               }
+                       }
+               }
+               //If we don't delete the action, we should remove references to release fences
+
+               const ModelAction *rel_fence =act->get_last_fence_release();
+               if (rel_fence != NULL) {
+                       modelclock_t relfenceseq = rel_fence->get_seq_number();
+                       thread_id_t relfence_tid = rel_fence->get_tid();
+                       modelclock_t tid_clock = cvmin->getClock(relfence_tid);
+                       //Remove references to irrelevant release fences
+                       if (relfenceseq <= tid_clock)
+                               act->set_last_fence_release(NULL);
+               }
+       }
+       //Now we are in the window of old actions that we remove if possible
+       for (;it != NULL;) {
+               ModelAction *act = it->getVal();
+               //Do iteration early since we may delete node...
+               it=it->getPrev();
+               bool islastact = false;
+               ModelAction *lastact = get_last_action(act->get_tid());
+               if (act == lastact) {
+                       Thread * th = get_thread(act);
+                       islastact = !th->is_complete();
+               }
+
+               if (act->is_read()) {
+                       if (act->get_reads_from()->is_free()) {
+                               if (act->is_rmw()) {
+                                       act->set_type(ATOMIC_WRITE);
+                               } else {
+                                       removeAction(act);
+                                       if (islastact) {
+                                               fixupLastAct(act);
+                                       }
+                                       delete act;
+                                       continue;
+                               }
+                       }
+               } else if (act->is_free()) {
+                       removeAction(act);
+                       if (islastact) {
+                               fixupLastAct(act);
+                       }
+                       delete act;
+                       continue;
+               } else if (act->is_write()) {
+                       //Do nothing with write that hasn't been marked to be freed
+               } else if (islastact) {
+                       //Keep the last action for non-read/write actions
+               } else if (act->is_fence()) {
+                       //Note that acquire fences can always be safely
+                       //removed, but could incur extra overheads in
+                       //traversals.  Removing them before the cvmin seems
+                       //like a good compromise.
+
+                       //Release fences before the cvmin don't do anything
+                       //because everyone has already synchronized.
+
+                       //Sequentially fences before cvmin are redundant
+                       //because happens-before will enforce same
+                       //orderings.
+
+                       modelclock_t actseq = act->get_seq_number();
+                       thread_id_t act_tid = act->get_tid();
+                       modelclock_t tid_clock = cvmin->getClock(act_tid);
+                       if (actseq <= tid_clock) {
+                               removeAction(act);
+                               delete act;
+                               continue;
+                       }
+               } else {
+                       //need to deal with lock, annotation, wait, notify, thread create, start, join, yield, finish, nops
+                       //lock, notify thread create, thread finish, yield, finish are dead as soon as they are in the trace
+                       //need to keep most recent unlock/wait for each lock
+                       if(act->is_unlock() || act->is_wait()) {
+                               ModelAction * lastlock = get_last_unlock(act);
+                               if (lastlock != act) {
+                                       removeAction(act);
+                                       delete act;
+                                       continue;
+                               }
+                       } else if (act->is_create()) {
+                               if (act->get_thread_operand()->is_complete()) {
+                                       removeAction(act);
+                                       delete act;
+                                       continue;
+                               }
+                       } else {
+                               removeAction(act);
+                               delete act;
+                               continue;
+                       }
+
+                       //If we don't delete the action, we should remove references to release fences
+                       const ModelAction *rel_fence =act->get_last_fence_release();
+                       if (rel_fence != NULL) {
+                               modelclock_t relfenceseq = rel_fence->get_seq_number();
+                               thread_id_t relfence_tid = rel_fence->get_tid();
+                               modelclock_t tid_clock = cvmin->getClock(relfence_tid);
+                               //Remove references to irrelevant release fences
+                               if (relfenceseq <= tid_clock)
+                                       act->set_last_fence_release(NULL);
+                       }
+
+               }
        }
 
+       delete cvmin;
+       delete queue;
 }
 
 Fuzzer * ModelExecution::getFuzzer() {