nodestack: register ModelExecution class w/in NodeStack
[c11tester.git] / execution.cc
index 828502686d0194e642e8b2335c910c33f86ff049..33c862bac36fd2a7e788cfae00a1a46a774ecd4a 100644 (file)
@@ -4,19 +4,17 @@
 #include <new>
 #include <stdarg.h>
 
-#include "execution.h"
 #include "model.h"
+#include "execution.h"
 #include "action.h"
 #include "nodestack.h"
 #include "schedule.h"
-#include "snapshot-interface.h"
 #include "common.h"
 #include "clockvector.h"
 #include "cyclegraph.h"
 #include "promise.h"
 #include "datarace.h"
 #include "threads-model.h"
-#include "output.h"
 #include "bugmessage.h"
 
 #define INITIAL_THREAD_ID      0
@@ -31,7 +29,6 @@ struct model_snapshot_members {
                used_sequence_numbers(0),
                next_backtrack(NULL),
                bugs(),
-               stats(),
                failed_promise(false),
                too_many_reads(false),
                no_valid_reads(false),
@@ -49,7 +46,6 @@ struct model_snapshot_members {
        modelclock_t used_sequence_numbers;
        ModelAction *next_backtrack;
        SnapVector<bug_message *> bugs;
-       struct execution_stats stats;
        bool failed_promise;
        bool too_many_reads;
        bool no_valid_reads;
@@ -61,54 +57,52 @@ struct model_snapshot_members {
 };
 
 /** @brief Constructor */
-ModelExecution::ModelExecution(struct model_params *params, Scheduler *scheduler, NodeStack *node_stack) :
+ModelExecution::ModelExecution(ModelChecker *m,
+               struct model_params *params,
+               Scheduler *scheduler,
+               NodeStack *node_stack) :
+       model(m),
        params(params),
        scheduler(scheduler),
-       action_trace(new action_list_t()),
-       thread_map(new HashTable<int, Thread *, int>()),
-       obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
-       condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
-       obj_thrd_map(new HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4 >()),
-       promises(new SnapVector<Promise *>()),
-       futurevalues(new SnapVector<struct PendingFutureValue>()),
-       pending_rel_seqs(new SnapVector<struct release_seq *>()),
-       thrd_last_action(new SnapVector<ModelAction *>(1)),
-       thrd_last_fence_release(new SnapVector<ModelAction *>()),
+       action_trace(),
+       thread_map(2), /* We'll always need at least 2 threads */
+       obj_map(),
+       condvar_waiters_map(),
+       obj_thrd_map(),
+       promises(),
+       futurevalues(),
+       pending_rel_seqs(),
+       thrd_last_action(1),
+       thrd_last_fence_release(),
        node_stack(node_stack),
        priv(new struct model_snapshot_members()),
-       mo_graph(new CycleGraph()),
-       execution_number(1)
+       mo_graph(new CycleGraph())
 {
        /* Initialize a model-checker thread, for special ModelActions */
        model_thread = new Thread(get_next_id());
-       thread_map->put(id_to_int(model_thread->get_id()), model_thread);
+       add_thread(model_thread);
        scheduler->register_engine(this);
+       node_stack->register_engine(this);
 }
 
 /** @brief Destructor */
 ModelExecution::~ModelExecution()
 {
        for (unsigned int i = 0; i < get_num_threads(); i++)
-               delete thread_map->get(i);
-       delete thread_map;
-
-       delete obj_thrd_map;
-       delete obj_map;
-       delete condvar_waiters_map;
-       delete action_trace;
-
-       for (unsigned int i = 0; i < promises->size(); i++)
-               delete (*promises)[i];
-       delete promises;
+               delete get_thread(int_to_id(i));
 
-       delete pending_rel_seqs;
+       for (unsigned int i = 0; i < promises.size(); i++)
+               delete promises[i];
 
-       delete thrd_last_action;
-       delete thrd_last_fence_release;
        delete mo_graph;
        delete priv;
 }
 
+int ModelExecution::get_execution_number() const
+{
+       return model->get_execution_number();
+}
+
 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
 {
        action_list_t *tmp = hash->get(ptr);
@@ -131,7 +125,7 @@ static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, Sn
 
 action_list_t * ModelExecution::get_actions_on_obj(void * obj, thread_id_t tid) const
 {
-       SnapVector<action_list_t> *wrv=obj_thrd_map->get(obj);
+       SnapVector<action_list_t> *wrv = obj_thrd_map.get(obj);
        if (wrv==NULL)
                return NULL;
        unsigned int thread=id_to_int(tid);
@@ -313,8 +307,8 @@ ModelAction * ModelExecution::get_last_fence_conflict(ModelAction *act) const
                return NULL;
 
        /* Skip past the release */
-       action_list_t *list = action_trace;
-       action_list_t::reverse_iterator rit;
+       const action_list_t *list = &action_trace;
+       action_list_t::const_reverse_iterator rit;
        for (rit = list->rbegin(); rit != list->rend(); rit++)
                if (*rit == last_release)
                        break;
@@ -379,7 +373,7 @@ ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
                ModelAction *ret = NULL;
 
                /* linear search: from most recent to oldest */
-               action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
+               action_list_t *list = obj_map.get(act->get_location());
                action_list_t::reverse_iterator rit;
                for (rit = list->rbegin(); rit != list->rend(); rit++) {
                        ModelAction *prev = *rit;
@@ -401,7 +395,7 @@ ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
        case ATOMIC_LOCK:
        case ATOMIC_TRYLOCK: {
                /* linear search: from most recent to oldest */
-               action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
+               action_list_t *list = obj_map.get(act->get_location());
                action_list_t::reverse_iterator rit;
                for (rit = list->rbegin(); rit != list->rend(); rit++) {
                        ModelAction *prev = *rit;
@@ -412,7 +406,7 @@ ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
        }
        case ATOMIC_UNLOCK: {
                /* linear search: from most recent to oldest */
-               action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
+               action_list_t *list = obj_map.get(act->get_location());
                action_list_t::reverse_iterator rit;
                for (rit = list->rbegin(); rit != list->rend(); rit++) {
                        ModelAction *prev = *rit;
@@ -423,7 +417,7 @@ ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
        }
        case ATOMIC_WAIT: {
                /* linear search: from most recent to oldest */
-               action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
+               action_list_t *list = obj_map.get(act->get_location());
                action_list_t::reverse_iterator rit;
                for (rit = list->rbegin(); rit != list->rend(); rit++) {
                        ModelAction *prev = *rit;
@@ -438,7 +432,7 @@ ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
        case ATOMIC_NOTIFY_ALL:
        case ATOMIC_NOTIFY_ONE: {
                /* linear search: from most recent to oldest */
-               action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
+               action_list_t *list = obj_map.get(act->get_location());
                action_list_t::reverse_iterator rit;
                for (rit = list->rbegin(); rit != list->rend(); rit++) {
                        ModelAction *prev = *rit;
@@ -619,7 +613,7 @@ bool ModelExecution::process_read(ModelAction *curr)
                        struct future_value fv = node->get_future_value();
                        Promise *promise = new Promise(this, curr, fv);
                        curr->set_read_from_promise(promise);
-                       promises->push_back(promise);
+                       promises.push_back(promise);
                        mo_graph->startChanges();
                        updated = r_modification_order(curr, promise);
                        mo_graph->commitChanges();
@@ -698,14 +692,14 @@ bool ModelExecution::process_mutex(ModelAction *curr)
 
                /* Should we go to sleep? (simulate spurious failures) */
                if (curr->get_node()->get_misc() == 0) {
-                       get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
+                       get_safe_ptr_action(&condvar_waiters_map, curr->get_location())->push_back(curr);
                        /* disable us */
                        scheduler->sleep(get_thread(curr));
                }
                break;
        }
        case ATOMIC_NOTIFY_ALL: {
-               action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
+               action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
                //activate all the waiting threads
                for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
                        scheduler->wake(get_thread(*rit));
@@ -714,7 +708,7 @@ bool ModelExecution::process_mutex(ModelAction *curr)
                break;
        }
        case ATOMIC_NOTIFY_ONE: {
-               action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
+               action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
                int wakeupthread = curr->get_node()->get_misc();
                action_list_t::iterator it = waiters->begin();
                advance(it, wakeupthread);
@@ -746,10 +740,10 @@ bool ModelExecution::process_mutex(ModelAction *curr)
 bool ModelExecution::promises_may_allow(const ModelAction *writer,
                const ModelAction *reader) const
 {
-       if (promises->empty())
+       if (promises.empty())
                return true;
-       for(int i=promises->size()-1;i>=0;i--) {
-               ModelAction *pr=(*promises)[i]->get_reader(0);
+       for (int i = promises.size() - 1; i >= 0; i--) {
+               ModelAction *pr = promises[i]->get_reader(0);
                //reader is after promise...doesn't cross any promise
                if (*reader > *pr)
                        return true;
@@ -822,17 +816,17 @@ bool ModelExecution::process_write(ModelAction *curr)
                        if (promises_may_allow(curr, read)) {
                                add_future_value(curr, read);
                        } else {
-                               futurevalues->push_back(PendingFutureValue(curr, read));
+                               futurevalues.push_back(PendingFutureValue(curr, read));
                        }
                }
        }
 
        /* Check the pending future values */
-       for (int i = (int)futurevalues->size() - 1; i >= 0; i--) {
-               struct PendingFutureValue pfv = (*futurevalues)[i];
+       for (int i = (int)futurevalues.size() - 1; i >= 0; i--) {
+               struct PendingFutureValue pfv = futurevalues[i];
                if (promises_may_allow(pfv.writer, pfv.reader)) {
                        add_future_value(pfv.writer, pfv.reader);
-                       futurevalues->erase(futurevalues->begin() + i);
+                       futurevalues.erase(futurevalues.begin() + i);
                }
        }
 
@@ -860,7 +854,7 @@ bool ModelExecution::process_fence(ModelAction *curr)
         */
        bool updated = false;
        if (curr->is_acquire()) {
-               action_list_t *list = action_trace;
+               action_list_t *list = &action_trace;
                action_list_t::reverse_iterator rit;
                /* Find X : is_read(X) && X --sb-> curr */
                for (rit = list->rbegin(); rit != list->rend(); rit++) {
@@ -916,8 +910,8 @@ bool ModelExecution::process_thread_action(ModelAction *curr)
                add_thread(th);
                th->set_creation(curr);
                /* Promises can be satisfied by children */
-               for (unsigned int i = 0; i < promises->size(); i++) {
-                       Promise *promise = (*promises)[i];
+               for (unsigned int i = 0; i < promises.size(); i++) {
+                       Promise *promise = promises[i];
                        if (promise->thread_is_available(curr->get_tid()))
                                promise->add_thread(th->get_id());
                }
@@ -941,8 +935,8 @@ bool ModelExecution::process_thread_action(ModelAction *curr)
                }
                th->complete();
                /* Completed thread can't satisfy promises */
-               for (unsigned int i = 0; i < promises->size(); i++) {
-                       Promise *promise = (*promises)[i];
+               for (unsigned int i = 0; i < promises.size(); i++) {
+                       Promise *promise = promises[i];
                        if (promise->thread_is_available(th->get_id()))
                                if (promise->eliminate_thread(th->get_id()))
                                        priv->failed_promise = true;
@@ -977,8 +971,8 @@ bool ModelExecution::process_thread_action(ModelAction *curr)
 void ModelExecution::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
 {
        const ModelAction *write = curr->get_node()->get_relseq_break();
-       struct release_seq *sequence = pending_rel_seqs->back();
-       pending_rel_seqs->pop_back();
+       struct release_seq *sequence = pending_rel_seqs.back();
+       pending_rel_seqs.pop_back();
        ASSERT(sequence);
        ModelAction *acquire = sequence->acquire;
        const ModelAction *rf = sequence->rf;
@@ -1008,7 +1002,7 @@ void ModelExecution::process_relseq_fixup(ModelAction *curr, work_queue_t *work_
                work_queue->push_back(MOEdgeWorkEntry(acquire));
 
                /* propagate synchronization to later actions */
-               action_list_t::reverse_iterator rit = action_trace->rbegin();
+               action_list_t::reverse_iterator rit = action_trace.rbegin();
                for (; (*rit) != acquire; rit++) {
                        ModelAction *propagate = *rit;
                        if (acquire->happens_before(propagate)) {
@@ -1093,7 +1087,7 @@ bool ModelExecution::initialize_curr_action(ModelAction **curr)
                else if (newcurr->is_wait())
                        newcurr->get_node()->set_misc_max(2);
                else if (newcurr->is_notify_one()) {
-                       newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
+                       newcurr->get_node()->set_misc_max(get_safe_ptr_action(&condvar_waiters_map, newcurr->get_location())->size());
                }
                return true; /* This was a new ModelAction */
        }
@@ -1160,8 +1154,8 @@ bool ModelExecution::synchronize(const ModelAction *first, ModelAction *second)
  */
 void ModelExecution::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
 {
-       for (unsigned int i = 0; i < promises->size(); i++) {
-               Promise *promise = (*promises)[i];
+       for (unsigned int i = 0; i < promises.size(); i++) {
+               Promise *promise = promises[i];
                if (!promise->thread_is_available(waiting->get_id()))
                        continue;
                for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
@@ -1332,8 +1326,8 @@ void ModelExecution::check_curr_backtracking(ModelAction *curr)
 
 bool ModelExecution::promises_expired() const
 {
-       for (unsigned int i = 0; i < promises->size(); i++) {
-               Promise *promise = (*promises)[i];
+       for (unsigned int i = 0; i < promises.size(); i++) {
+               Promise *promise = promises[i];
                if (promise->get_expiration() < priv->used_sequence_numbers)
                        return true;
        }
@@ -1347,7 +1341,7 @@ bool ModelExecution::promises_expired() const
  */
 bool ModelExecution::isfeasibleprefix() const
 {
-       return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
+       return pending_rel_seqs.size() == 0 && is_feasible_prefix_ignore_relseq();
 }
 
 /**
@@ -1371,7 +1365,7 @@ void ModelExecution::print_infeasibility(const char *prefix) const
                ptr += sprintf(ptr, "[bad sw ordering]");
        if (promises_expired())
                ptr += sprintf(ptr, "[promise expired]");
-       if (promises->size() != 0)
+       if (promises.size() != 0)
                ptr += sprintf(ptr, "[unresolved promise]");
        if (ptr != buf)
                model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
@@ -1383,7 +1377,7 @@ void ModelExecution::print_infeasibility(const char *prefix) const
  */
 bool ModelExecution::is_feasible_prefix_ignore_relseq() const
 {
-       return !is_infeasible() && promises->size() == 0;
+       return !is_infeasible() && promises.size() == 0;
 }
 
 /**
@@ -1438,7 +1432,7 @@ bool ModelExecution::should_read_instead(const ModelAction *curr, const T *rf, c
        if (!mo_graph->checkReachable(rf, other_rf))
                return false;
 
-       SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
        action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
        action_list_t::reverse_iterator rit = list->rbegin();
        ASSERT((*rit) == curr);
@@ -1481,7 +1475,7 @@ bool ModelExecution::check_recency(ModelAction *curr, const T *rf) const
                        curr->get_node()->get_read_from_promise_size() <= 1)
                return true;
 
-       SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
        int tid = id_to_int(curr->get_tid());
        ASSERT(tid < (int)thrd_lists->size());
        action_list_t *list = &(*thrd_lists)[tid];
@@ -1538,7 +1532,7 @@ bool ModelExecution::check_recency(ModelAction *curr, const T *rf) const
 template <typename rf_type>
 bool ModelExecution::r_modification_order(ModelAction *curr, const rf_type *rf)
 {
-       SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
        unsigned int i;
        bool added = false;
        ASSERT(curr->is_read());
@@ -1631,9 +1625,9 @@ bool ModelExecution::r_modification_order(ModelAction *curr, const rf_type *rf)
         * All compatible, thread-exclusive promises must be ordered after any
         * concrete loads from the same thread
         */
-       for (unsigned int i = 0; i < promises->size(); i++)
-               if ((*promises)[i]->is_compatible_exclusive(curr))
-                       added = mo_graph->addEdge(rf, (*promises)[i]) || added;
+       for (unsigned int i = 0; i < promises.size(); i++)
+               if (promises[i]->is_compatible_exclusive(curr))
+                       added = mo_graph->addEdge(rf, promises[i]) || added;
 
        return added;
 }
@@ -1664,7 +1658,7 @@ bool ModelExecution::r_modification_order(ModelAction *curr, const rf_type *rf)
  */
 bool ModelExecution::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
 {
-       SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
        unsigned int i;
        bool added = false;
        ASSERT(curr->is_write());
@@ -1770,9 +1764,9 @@ bool ModelExecution::w_modification_order(ModelAction *curr, ModelVector<ModelAc
         * concrete stores to the same thread, or else they can be merged with
         * this store later
         */
-       for (unsigned int i = 0; i < promises->size(); i++)
-               if ((*promises)[i]->is_compatible_exclusive(curr))
-                       added = mo_graph->addEdge(curr, (*promises)[i]) || added;
+       for (unsigned int i = 0; i < promises.size(); i++)
+               if (promises[i]->is_compatible_exclusive(curr))
+                       added = mo_graph->addEdge(curr, promises[i]) || added;
 
        return added;
 }
@@ -1808,7 +1802,7 @@ bool ModelExecution::thin_air_constraint_may_allow(const ModelAction *writer, co
  */
 bool ModelExecution::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
 {
-       SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(reader->get_location());
        unsigned int i;
        /* Iterate over all threads */
        for (i = 0; i < thrd_lists->size(); i++) {
@@ -1909,7 +1903,7 @@ bool ModelExecution::release_seq_heads(const ModelAction *rf,
                release_heads->push_back(fence_release);
 
        int tid = id_to_int(rf->get_tid());
-       SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(rf->get_location());
        action_list_t *list = &(*thrd_lists)[tid];
        action_list_t::const_reverse_iterator rit;
 
@@ -2030,7 +2024,7 @@ void ModelExecution::get_release_seq_heads(ModelAction *acquire,
 
        if (!release_seq_heads(rf, release_heads, sequence)) {
                /* add act to 'lazy checking' list */
-               pending_rel_seqs->push_back(sequence);
+               pending_rel_seqs.push_back(sequence);
        } else {
                snapshot_free(sequence);
        }
@@ -2052,8 +2046,8 @@ void ModelExecution::get_release_seq_heads(ModelAction *acquire,
 bool ModelExecution::resolve_release_sequences(void *location, work_queue_t *work_queue)
 {
        bool updated = false;
-       SnapVector<struct release_seq *>::iterator it = pending_rel_seqs->begin();
-       while (it != pending_rel_seqs->end()) {
+       SnapVector<struct release_seq *>::iterator it = pending_rel_seqs.begin();
+       while (it != pending_rel_seqs.end()) {
                struct release_seq *pending = *it;
                ModelAction *acquire = pending->acquire;
                const ModelAction *read = pending->read;
@@ -2081,7 +2075,7 @@ bool ModelExecution::resolve_release_sequences(void *location, work_queue_t *wor
                                work_queue->push_back(MOEdgeWorkEntry(acquire));
 
                        /* propagate synchronization to later actions */
-                       action_list_t::reverse_iterator rit = action_trace->rbegin();
+                       action_list_t::reverse_iterator rit = action_trace.rbegin();
                        for (; (*rit) != acquire; rit++) {
                                ModelAction *propagate = *rit;
                                if (acquire->happens_before(propagate)) {
@@ -2092,7 +2086,7 @@ bool ModelExecution::resolve_release_sequences(void *location, work_queue_t *wor
                        }
                }
                if (complete) {
-                       it = pending_rel_seqs->erase(it);
+                       it = pending_rel_seqs.erase(it);
                        snapshot_free(pending);
                } else {
                        it++;
@@ -2117,7 +2111,7 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
        int tid = id_to_int(act->get_tid());
        ModelAction *uninit = NULL;
        int uninit_id = -1;
-       action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
+       action_list_t *list = get_safe_ptr_action(&obj_map, act->get_location());
        if (list->empty() && act->is_atomic_var()) {
                uninit = get_uninitialized_action(act);
                uninit_id = id_to_int(uninit->get_tid());
@@ -2125,34 +2119,34 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
        }
        list->push_back(act);
 
-       action_trace->push_back(act);
+       action_trace.push_back(act);
        if (uninit)
-               action_trace->push_front(uninit);
+               action_trace.push_front(uninit);
 
-       SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
+       SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, act->get_location());
        if (tid >= (int)vec->size())
                vec->resize(priv->next_thread_id);
        (*vec)[tid].push_back(act);
        if (uninit)
                (*vec)[uninit_id].push_front(uninit);
 
-       if ((int)thrd_last_action->size() <= tid)
-               thrd_last_action->resize(get_num_threads());
-       (*thrd_last_action)[tid] = act;
+       if ((int)thrd_last_action.size() <= tid)
+               thrd_last_action.resize(get_num_threads());
+       thrd_last_action[tid] = act;
        if (uninit)
-               (*thrd_last_action)[uninit_id] = uninit;
+               thrd_last_action[uninit_id] = uninit;
 
        if (act->is_fence() && act->is_release()) {
-               if ((int)thrd_last_fence_release->size() <= tid)
-                       thrd_last_fence_release->resize(get_num_threads());
-               (*thrd_last_fence_release)[tid] = act;
+               if ((int)thrd_last_fence_release.size() <= tid)
+                       thrd_last_fence_release.resize(get_num_threads());
+               thrd_last_fence_release[tid] = act;
        }
 
        if (act->is_wait()) {
                void *mutex_loc = (void *) act->get_value();
-               get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
+               get_safe_ptr_action(&obj_map, mutex_loc)->push_back(act);
 
-               SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
+               SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, mutex_loc);
                if (tid >= (int)vec->size())
                        vec->resize(priv->next_thread_id);
                (*vec)[tid].push_back(act);
@@ -2167,8 +2161,8 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
 ModelAction * ModelExecution::get_last_action(thread_id_t tid) const
 {
        int threadid = id_to_int(tid);
-       if (threadid < (int)thrd_last_action->size())
-               return (*thrd_last_action)[id_to_int(tid)];
+       if (threadid < (int)thrd_last_action.size())
+               return thrd_last_action[id_to_int(tid)];
        else
                return NULL;
 }
@@ -2181,8 +2175,8 @@ ModelAction * ModelExecution::get_last_action(thread_id_t tid) const
 ModelAction * ModelExecution::get_last_fence_release(thread_id_t tid) const
 {
        int threadid = id_to_int(tid);
-       if (threadid < (int)thrd_last_fence_release->size())
-               return (*thrd_last_fence_release)[id_to_int(tid)];
+       if (threadid < (int)thrd_last_fence_release.size())
+               return thrd_last_fence_release[id_to_int(tid)];
        else
                return NULL;
 }
@@ -2198,7 +2192,7 @@ ModelAction * ModelExecution::get_last_fence_release(thread_id_t tid) const
 ModelAction * ModelExecution::get_last_seq_cst_write(ModelAction *curr) const
 {
        void *location = curr->get_location();
-       action_list_t *list = get_safe_ptr_action(obj_map, location);
+       action_list_t *list = obj_map.get(location);
        /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
        action_list_t::reverse_iterator rit;
        for (rit = list->rbegin(); (*rit) != curr; rit++)
@@ -2220,8 +2214,12 @@ ModelAction * ModelExecution::get_last_seq_cst_write(ModelAction *curr) const
  */
 ModelAction * ModelExecution::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
 {
-       /* All fences should have NULL location */
-       action_list_t *list = get_safe_ptr_action(obj_map, NULL);
+       /* All fences should have location FENCE_LOCATION */
+       action_list_t *list = obj_map.get(FENCE_LOCATION);
+
+       if (!list)
+               return NULL;
+
        action_list_t::reverse_iterator rit = list->rbegin();
 
        if (before_fence) {
@@ -2250,7 +2248,7 @@ ModelAction * ModelExecution::get_last_seq_cst_fence(thread_id_t tid, const Mode
 ModelAction * ModelExecution::get_last_unlock(ModelAction *curr) const
 {
        void *location = curr->get_location();
-       action_list_t *list = get_safe_ptr_action(obj_map, location);
+       action_list_t *list = obj_map.get(location);
        /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
        action_list_t::reverse_iterator rit;
        for (rit = list->rbegin(); rit != list->rend(); rit++)
@@ -2285,10 +2283,10 @@ ClockVector * ModelExecution::get_cv(thread_id_t tid) const
  */
 Promise * ModelExecution::pop_promise_to_resolve(const ModelAction *curr)
 {
-       for (unsigned int i = 0; i < promises->size(); i++)
+       for (unsigned int i = 0; i < promises.size(); i++)
                if (curr->get_node()->get_promise(i)) {
-                       Promise *ret = (*promises)[i];
-                       promises->erase(promises->begin() + i);
+                       Promise *ret = promises[i];
+                       promises.erase(promises.begin() + i);
                        return ret;
                }
        return NULL;
@@ -2343,8 +2341,8 @@ bool ModelExecution::resolve_promise(ModelAction *write, Promise *promise)
  */
 void ModelExecution::compute_promises(ModelAction *curr)
 {
-       for (unsigned int i = 0; i < promises->size(); i++) {
-               Promise *promise = (*promises)[i];
+       for (unsigned int i = 0; i < promises.size(); i++) {
+               Promise *promise = promises[i];
                if (!promise->is_compatible(curr) || !promise->same_value(curr))
                        continue;
 
@@ -2365,8 +2363,8 @@ void ModelExecution::compute_promises(ModelAction *curr)
 /** Checks promises in response to change in ClockVector Threads. */
 void ModelExecution::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
 {
-       for (unsigned int i = 0; i < promises->size(); i++) {
-               Promise *promise = (*promises)[i];
+       for (unsigned int i = 0; i < promises.size(); i++) {
+               Promise *promise = promises[i];
                if (!promise->thread_is_available(tid))
                        continue;
                for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
@@ -2385,8 +2383,8 @@ void ModelExecution::check_promises(thread_id_t tid, ClockVector *old_cv, ClockV
 
 void ModelExecution::check_promises_thread_disabled()
 {
-       for (unsigned int i = 0; i < promises->size(); i++) {
-               Promise *promise = (*promises)[i];
+       for (unsigned int i = 0; i < promises.size(); i++) {
+               Promise *promise = promises[i];
                if (promise->has_failed()) {
                        priv->failed_promise = true;
                        return;
@@ -2412,8 +2410,8 @@ void ModelExecution::mo_check_promises(const ModelAction *act, bool is_read_chec
 {
        const ModelAction *write = is_read_check ? act->get_reads_from() : act;
 
-       for (unsigned int i = 0; i < promises->size(); i++) {
-               Promise *promise = (*promises)[i];
+       for (unsigned int i = 0; i < promises.size(); i++) {
+               Promise *promise = promises[i];
 
                // Is this promise on the same location?
                if (!promise->same_location(write))
@@ -2453,10 +2451,10 @@ void ModelExecution::mo_check_promises(const ModelAction *act, bool is_read_chec
  */
 void ModelExecution::compute_relseq_breakwrites(ModelAction *curr)
 {
-       if (pending_rel_seqs->empty())
+       if (pending_rel_seqs.empty())
                return;
 
-       struct release_seq *pending = pending_rel_seqs->back();
+       struct release_seq *pending = pending_rel_seqs.back();
        for (unsigned int i = 0; i < pending->writes.size(); i++) {
                const ModelAction *write = pending->writes[i];
                curr->get_node()->add_relseq_break(write);
@@ -2475,7 +2473,7 @@ void ModelExecution::compute_relseq_breakwrites(ModelAction *curr)
  */
 void ModelExecution::build_may_read_from(ModelAction *curr)
 {
-       SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
        unsigned int i;
        ASSERT(curr->is_read());
 
@@ -2520,8 +2518,8 @@ void ModelExecution::build_may_read_from(ModelAction *curr)
        }
 
        /* Inherit existing, promised future values */
-       for (i = 0; i < promises->size(); i++) {
-               const Promise *promise = (*promises)[i];
+       for (i = 0; i < promises.size(); i++) {
+               const Promise *promise = promises[i];
                const ModelAction *promise_read = promise->get_reader(0);
                if (promise_read->same_var(curr)) {
                        /* Only add feasible future-values */
@@ -2585,9 +2583,9 @@ ModelAction * ModelExecution::get_uninitialized_action(const ModelAction *curr)
        return act;
 }
 
-static void print_list(action_list_t *list)
+static void print_list(const action_list_t *list)
 {
-       action_list_t::iterator it;
+       action_list_t::const_iterator it;
 
        model_print("---------------------------------------------------------------------\n");
 
@@ -2613,7 +2611,7 @@ void ModelExecution::dumpGraph(char *filename) const
        mo_graph->dumpNodes(file);
        ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
 
-       for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
+       for (action_list_t::iterator it = action_trace.begin(); it != action_trace.end(); it++) {
                ModelAction *act = *it;
                if (act->is_read()) {
                        mo_graph->dot_print_node(file, act);
@@ -2648,26 +2646,26 @@ void ModelExecution::print_summary() const
 {
 #if SUPPORT_MOD_ORDER_DUMP
        char buffername[100];
-       sprintf(buffername, "exec%04u", execution_number);
+       sprintf(buffername, "exec%04u", get_execution_number());
        mo_graph->dumpGraphToFile(buffername);
-       sprintf(buffername, "graph%04u", execution_number);
+       sprintf(buffername, "graph%04u", get_execution_number());
        dumpGraph(buffername);
 #endif
 
-       model_print("Execution %d:", execution_number);
+       model_print("Execution %d:", get_execution_number());
        if (isfeasibleprefix()) {
                if (scheduler->all_threads_sleeping())
                        model_print(" SLEEP-SET REDUNDANT");
                model_print("\n");
        } else
                print_infeasibility(" INFEASIBLE");
-       print_list(action_trace);
+       print_list(&action_trace);
        model_print("\n");
-       if (!promises->empty()) {
+       if (!promises.empty()) {
                model_print("Pending promises:\n");
-               for (unsigned int i = 0; i < promises->size(); i++) {
+               for (unsigned int i = 0; i < promises.size(); i++) {
                        model_print(" [P%u] ", i);
-                       (*promises)[i]->print();
+                       promises[i]->print();
                }
                model_print("\n");
        }
@@ -2680,7 +2678,10 @@ void ModelExecution::print_summary() const
  */
 void ModelExecution::add_thread(Thread *t)
 {
-       thread_map->put(id_to_int(t->get_id()), t);
+       unsigned int i = id_to_int(t->get_id());
+       if (i >= thread_map.size())
+               thread_map.resize(i + 1);
+       thread_map[i] = t;
        if (!t->is_model_thread())
                scheduler->add_thread(t);
 }
@@ -2692,7 +2693,10 @@ void ModelExecution::add_thread(Thread *t)
  */
 Thread * ModelExecution::get_thread(thread_id_t tid) const
 {
-       return thread_map->get(id_to_int(tid));
+       unsigned int i = id_to_int(tid);
+       if (i < thread_map.size())
+               return thread_map[i];
+       return NULL;
 }
 
 /**
@@ -2718,8 +2722,8 @@ Thread * ModelExecution::get_thread(const ModelAction *act) const
  */
 int ModelExecution::get_promise_number(const Promise *promise) const
 {
-       for (unsigned int i = 0; i < promises->size(); i++)
-               if ((*promises)[i] == promise)
+       for (unsigned int i = 0; i < promises.size(); i++)
+               if (promises[i] == promise)
                        return i;
        /* Not found */
        return -1;
@@ -2805,12 +2809,12 @@ Thread * ModelExecution::take_step(ModelAction *curr)
  */
 void ModelExecution::fixup_release_sequences()
 {
-       while (!pending_rel_seqs->empty() &&
+       while (!pending_rel_seqs.empty() &&
                        is_feasible_prefix_ignore_relseq() &&
                        !unrealizedraces.empty()) {
                model_print("*** WARNING: release sequence fixup action "
                                "(%zu pending release seuqence(s)) ***\n",
-                               pending_rel_seqs->size());
+                               pending_rel_seqs.size());
                ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
                                std::memory_order_seq_cst, NULL, VALUE_NONE,
                                model_thread);