Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
[mc] Log some information for the prefiltering phase of state comparator
[simgrid.git] / src / mc / mc_dpor.c
index 4209fac..dec62f4 100644 (file)
@@ -13,9 +13,161 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dpor, mc,
 
 xbt_dynar_t visited_states;
 xbt_dict_t first_enabled_state;
+xbt_dynar_t initial_communications_pattern;
+xbt_dynar_t communications_pattern;
+int nb_comm_pattern;
 
 /********** Static functions ***********/
 
+static void comm_pattern_free(mc_comm_pattern_t p){
+  xbt_free(p->rdv);
+  xbt_free(p->data);
+  xbt_free(p);
+  p = NULL;
+}
+
+static void comm_pattern_free_voidp( void *p){
+  comm_pattern_free((mc_comm_pattern_t) * (void **)p);
+}
+
+static mc_comm_pattern_t get_comm_pattern_from_idx(xbt_dynar_t pattern, unsigned int *idx, e_smx_comm_type_t type, unsigned long proc){
+  mc_comm_pattern_t current_comm;
+  while(*idx < xbt_dynar_length(pattern)){
+    current_comm = (mc_comm_pattern_t)xbt_dynar_get_as(pattern, *idx, mc_comm_pattern_t);
+    if(current_comm->type == type && type == SIMIX_COMM_SEND){
+      if(current_comm->src_proc == proc)
+        return current_comm;
+    }else if(current_comm->type == type && type == SIMIX_COMM_RECEIVE){
+      if(current_comm->dst_proc == proc)
+        return current_comm;
+    }
+    (*idx)++;
+  }
+  return NULL;
+}
+
+static int compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2){
+  if(strcmp(comm1->rdv, comm2->rdv) != 0)
+    return 1;
+  if(comm1->src_proc != comm2->src_proc)
+    return 1;
+  if(comm1->dst_proc != comm2->dst_proc)
+    return 1;
+  if(comm1->data_size != comm2->data_size)
+    return 1;
+  if(memcmp(comm1->data, comm2->data, comm1->data_size) != 0)
+    return 1;
+  return 0;
+}
+
+static void deterministic_pattern(xbt_dynar_t initial_pattern, xbt_dynar_t pattern){
+  unsigned int cursor = 0, send_index = 0, recv_index = 0;
+  mc_comm_pattern_t comm1, comm2;
+  int comm_comparison = 0;
+  int current_process = 0;
+  while(current_process < simix_process_maxpid){
+    while(cursor < xbt_dynar_length(initial_pattern)){
+      comm1 = (mc_comm_pattern_t)xbt_dynar_get_as(initial_pattern, cursor, mc_comm_pattern_t);
+      if(comm1->type == SIMIX_COMM_SEND && comm1->src_proc == current_process){
+        comm2 = get_comm_pattern_from_idx(pattern, &send_index, comm1->type, current_process);
+        comm_comparison = compare_comm_pattern(comm1, comm2);
+        if(comm_comparison == 1){
+          initial_state_safety->send_deterministic = 0;
+          initial_state_safety->comm_deterministic = 0;
+          return;
+        }
+        send_index++;
+      }else if(comm1->type == SIMIX_COMM_RECEIVE && comm1->dst_proc == current_process){
+        comm2 = get_comm_pattern_from_idx(pattern, &recv_index, comm1->type, current_process);
+        comm_comparison = compare_comm_pattern(comm1, comm2);
+        if(comm_comparison == 1){
+          initial_state_safety->comm_deterministic = 0;
+        }
+        recv_index++;
+      }
+      cursor++;
+    }
+    cursor = 0;
+    send_index = 0;
+    recv_index = 0;
+    current_process++;
+  }
+}
+
+static int complete_comm_pattern(xbt_dynar_t list, mc_comm_pattern_t pattern){
+  mc_comm_pattern_t current_pattern;
+  unsigned int cursor = 0;
+  xbt_dynar_foreach(list, cursor, current_pattern){
+    if(current_pattern->comm == pattern->comm){
+      if(!current_pattern->completed){
+        current_pattern->src_proc = pattern->comm->comm.src_proc->pid;
+        current_pattern->src_host = simcall_host_get_name(pattern->comm->comm.src_proc->smx_host);
+        current_pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
+        current_pattern->dst_host = simcall_host_get_name(pattern->comm->comm.dst_proc->smx_host);
+        current_pattern->data_size = pattern->comm->comm.src_buff_size;
+        current_pattern->data = xbt_malloc0(current_pattern->data_size);
+        current_pattern->matched_comm = pattern->num;
+        memcpy(current_pattern->data, current_pattern->comm->comm.src_buff, current_pattern->data_size);
+        current_pattern->completed = 1;
+        return current_pattern->num;
+      }
+    }
+  }
+  return -1;
+}
+
+void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, int call){
+  mc_comm_pattern_t pattern = NULL;
+  pattern = xbt_new0(s_mc_comm_pattern_t, 1);
+  pattern->num = ++nb_comm_pattern;
+  pattern->completed = 0;
+  if(call == 1){ // ISEND
+    pattern->comm = simcall_comm_isend__get__result(request);
+    pattern->type = SIMIX_COMM_SEND;
+    if(pattern->comm->comm.dst_proc != NULL){ 
+      pattern->matched_comm = complete_comm_pattern(list, pattern);
+      pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
+      pattern->completed = 1;
+      pattern->dst_host = simcall_host_get_name(pattern->comm->comm.dst_proc->smx_host);
+    }
+    pattern->src_proc = pattern->comm->comm.src_proc->pid;
+    pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
+    pattern->data_size = pattern->comm->comm.src_buff_size;
+    pattern->data=xbt_malloc0(pattern->data_size);
+    memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
+  }else{ // IRECV
+    pattern->comm = simcall_comm_irecv__get__result(request);
+    pattern->type = SIMIX_COMM_RECEIVE;
+    if(pattern->comm->comm.src_proc != NULL){
+      pattern->matched_comm = complete_comm_pattern(list, pattern);
+      pattern->src_proc = pattern->comm->comm.src_proc->pid;
+      pattern->src_host = simcall_host_get_name(pattern->comm->comm.src_proc->smx_host);
+      pattern->completed = 1;
+      pattern->data_size = pattern->comm->comm.src_buff_size;
+      pattern->data=xbt_malloc0(pattern->data_size);
+      memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
+    }
+    pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
+    pattern->dst_host = simcall_host_get_name(request->issuer->smx_host);
+  }
+  if(pattern->comm->comm.rdv != NULL)
+    pattern->rdv = strdup(pattern->comm->comm.rdv->name);
+  else
+    pattern->rdv = strdup(pattern->comm->comm.rdv_cpy->name);
+  xbt_dynar_push(list, &pattern);
+}
+
+static void print_communications_pattern(xbt_dynar_t comms_pattern){
+  unsigned int cursor = 0;
+  mc_comm_pattern_t current_comm;
+  xbt_dynar_foreach(comms_pattern, cursor, current_comm){
+    if(current_comm->type == SIMIX_COMM_SEND)
+      XBT_INFO("[(%lu) %s -> %s] %s ", current_comm->src_proc, current_comm->src_host, current_comm->dst_host, "iSend");
+    else
+      XBT_INFO("[(%lu) %s <- %s] %s ", current_comm->dst_proc, current_comm->dst_host, current_comm->src_host, "iRecv");
+  }
+}
+
 static void visited_state_free(mc_visited_state_t state){
   if(state){
     MC_free_snapshot(state->system_state);
@@ -27,13 +179,17 @@ static void visited_state_free_voidp(void *s){
   visited_state_free((mc_visited_state_t) * (void **) s);
 }
 
+/** \brief Save the current state
+ *
+ *  \return Snapshot of the current state.
+ */
 static mc_visited_state_t visited_state_new(){
 
   mc_visited_state_t new_state = NULL;
   new_state = xbt_new0(s_mc_visited_state_t, 1);
   new_state->heap_bytes_used = mmalloc_get_bytes_used(std_heap);
   new_state->nb_processes = xbt_swag_size(simix_global->process_list);
-  new_state->system_state = MC_take_snapshot();
+  new_state->system_state = MC_take_snapshot(mc_stats->expanded_states);
   new_state->num = mc_stats->expanded_states;
   new_state->other_num = -1;
 
@@ -41,7 +197,25 @@ static mc_visited_state_t visited_state_new(){
   
 }
 
+/** \brief Find a suitable subrange of candidate duplicates for a given state
+ *
+ *  \param all_ pairs dynamic array of states with candidate duplicates of the current state;
+ *  \param pair current state;
+ *  \param min (output) index of the beginning of the the subrange
+ *  \param max (output) index of the enf of the subrange
+ *
+ *  Given a suitably ordered array of state, this function extracts a subrange
+ *  (with index *min <= i <= *max) with candidate duplicates of the given state.
+ *  This function uses only fast discriminating criterions and does not use the
+ *  full state comparison algorithms.
+ *
+ *  The states in all_pairs MUST be ordered using a (given) weak order
+ *  (based on nb_processes and heap_bytes_used).
+ *  The subrange is the subrange of "equivalence" of the given state.
+ */
 static int get_search_interval(xbt_dynar_t all_states, mc_visited_state_t state, int *min, int *max){
+  XBT_VERB("Searching interval for state %i: nd_processes=%zu heap_bytes_used=%zu",
+    state->num, (size_t)state->nb_processes, (size_t)state->heap_bytes_used);
 
   int raw_mem_set = (mmalloc_get_current_heap() == raw_heap);
 
@@ -95,6 +269,10 @@ static int get_search_interval(xbt_dynar_t all_states, mc_visited_state_t state,
   return cursor;
 }
 
+/** \brief Take a snapshot the current state and process it.
+ *
+ *  \return number of the duplicate state or -1 (not visited)
+ */
 static int is_visited_state(){
 
   if(_sg_mc_visited == 0)
@@ -118,13 +296,16 @@ static int is_visited_state(){
   }else{
 
     int min = -1, max = -1, index;
-    int res;
+    //int res;
     mc_visited_state_t state_test;
+    int cursor;
 
     index = get_search_interval(visited_states, new_state, &min, &max);
 
     if(min != -1 && max != -1){
-      res = xbt_parmap_mc_apply(parmap, snapshot_compare, xbt_dynar_get_ptr(visited_states, min), (max-min)+1, new_state);
+
+      // Parallell implementation
+      /*res = xbt_parmap_mc_apply(parmap, snapshot_compare, xbt_dynar_get_ptr(visited_states, min), (max-min)+1, new_state);
       if(res != -1){
         state_test = (mc_visited_state_t)xbt_dynar_get_as(visited_states, (min+res)-1, mc_visited_state_t);
         if(state_test->other_num == -1)
@@ -140,9 +321,40 @@ static int is_visited_state(){
         if(!raw_mem_set)
           MC_UNSET_RAW_MEM;
         return new_state->other_num;
-      } 
+        }*/
+
+      cursor = min;
+      while(cursor <= max){
+        state_test = (mc_visited_state_t)xbt_dynar_get_as(visited_states, cursor, mc_visited_state_t);
+        if(snapshot_compare(state_test, new_state) == 0){
+          // The state has been visited:
+
+          if(state_test->other_num == -1)
+            new_state->other_num = state_test->num;
+          else
+            new_state->other_num = state_test->other_num;
+          if(dot_output == NULL)
+            XBT_DEBUG("State %d already visited ! (equal to state %d)", new_state->num, state_test->num);
+          else
+            XBT_DEBUG("State %d already visited ! (equal to state %d (state %d in dot_output))", new_state->num, state_test->num, new_state->other_num);
+
+          // Replace the old state with the new one (why?):
+          xbt_dynar_remove_at(visited_states, cursor, NULL);
+          xbt_dynar_insert_at(visited_states, cursor, &new_state);
+
+          if(!raw_mem_set)
+            MC_UNSET_RAW_MEM;
+          return new_state->other_num;
+        }
+        cursor++;
+      }
+
+      // The state has not been visited, add it to the list:
       xbt_dynar_insert_at(visited_states, min, &new_state);
+
     }else{
+
+      // The state has not been visited: insert the state in the dynamic array.
       state_test = (mc_visited_state_t)xbt_dynar_get_as(visited_states, index, mc_visited_state_t);
       if(state_test->nb_processes < new_state->nb_processes){
         xbt_dynar_insert_at(visited_states, index+1, &new_state);
@@ -152,9 +364,13 @@ static int is_visited_state(){
         else
           xbt_dynar_insert_at(visited_states, index, &new_state);
       }
+
     }
 
+    // We have reached the maximum number of stored states;
     if(xbt_dynar_length(visited_states) > _sg_mc_visited){
+
+      // Find the (index of the) older state:
       int min2 = mc_stats->expanded_states;
       unsigned int cursor2 = 0;
       unsigned int index2 = 0;
@@ -164,6 +380,8 @@ static int is_visited_state(){
           min2 = state_test->num;
         }
       }
+
+      // and drop it:
       xbt_dynar_remove_at(visited_states, index2, NULL);
     }
 
@@ -189,12 +407,17 @@ void MC_dpor_init()
   /* Create the initial state and push it into the exploration stack */
   MC_SET_RAW_MEM;
 
-  initial_state = MC_state_new();
   if(_sg_mc_visited > 0)
     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 
   first_enabled_state = xbt_dict_new_homogeneous(&xbt_free_f);
 
+  initial_communications_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
+  communications_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
+  nb_comm_pattern = 0;
+
+  initial_state = MC_state_new();
+
   MC_UNSET_RAW_MEM;
 
   XBT_DEBUG("**************************************************");
@@ -203,6 +426,9 @@ void MC_dpor_init()
   /* Wait for requests (schedules processes) */
   MC_wait_for_requests();
 
+  MC_ignore_heap(simix_global->process_to_run->data, 0);
+  MC_ignore_heap(simix_global->process_that_ran->data, 0);
+
   MC_SET_RAW_MEM;
  
   /* Get an enabled process and insert it in the interleave set of the initial state */
@@ -239,9 +465,8 @@ void MC_dpor_init()
 }
 
 
-/**
- *   \brief Perform the model-checking operation using a depth-first search exploration
- *         with Dynamic Partial Order Reductions
+/** \brief Model-check the application using a DFS exploration
+ *         with DPOR (Dynamic Partial Order Reductions)
  */
 void MC_dpor(void)
 {
@@ -249,12 +474,15 @@ void MC_dpor(void)
   char *req_str = NULL;
   int value;
   smx_simcall_t req = NULL, prev_req = NULL;
-  mc_state_t state = NULL, prev_state = NULL, next_state = NULL, restore_state=NULL;
+  mc_state_t state = NULL, prev_state = NULL, next_state = NULL, restored_state=NULL;
   smx_process_t process = NULL;
   xbt_fifo_item_t item = NULL;
+  mc_state_t state_test = NULL;
   int pos;
   int visited_state = -1;
   int enabled = 0;
+  int interleave_size = 0;
+  int comm_pattern = 0;
 
   while (xbt_fifo_size(mc_stack_safety) > 0) {
 
@@ -263,10 +491,12 @@ void MC_dpor(void)
       xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack_safety));
 
     XBT_DEBUG("**************************************************");
-    XBT_DEBUG("Exploration depth=%d (state=%p, num %d)(%u interleave, user_max_depth %d)",
+    XBT_DEBUG("Exploration depth=%d (state=%p, num %d)(%u interleave, user_max_depth %d, first_enabled_state size : %d)",
               xbt_fifo_size(mc_stack_safety), state, state->num,
-              MC_state_interleave_size(state), user_max_depth_reached);
+              MC_state_interleave_size(state), user_max_depth_reached, xbt_dict_size(first_enabled_state));
       
+    interleave_size = MC_state_interleave_size(state);
+
     /* Update statistics */
     mc_stats->visited_states++;
 
@@ -295,10 +525,29 @@ void MC_dpor(void)
       xbt_dict_remove(first_enabled_state, key); 
       xbt_free(key);
       MC_UNSET_RAW_MEM;
+      
+      if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
+        if(req->call == SIMCALL_COMM_ISEND)
+          comm_pattern = 1;
+        else if(req->call == SIMCALL_COMM_IRECV)
+          comm_pattern = 2;
+      }
 
       /* Answer the request */
       SIMIX_simcall_pre(req, value); /* After this call req is no longer usefull */
 
+      if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
+        MC_SET_RAW_MEM;
+        if(comm_pattern != 0){
+          if(!initial_state_safety->initial_communications_pattern_done)
+            get_comm_pattern(initial_communications_pattern, req, comm_pattern);
+          else
+            get_comm_pattern(communications_pattern, req, comm_pattern);
+        }
+        MC_UNSET_RAW_MEM; 
+        comm_pattern = 0;
+      }
+
       /* Wait for requests (schedules processes) */
       MC_wait_for_requests();
 
@@ -319,7 +568,7 @@ void MC_dpor(void)
         }
 
         if(_sg_mc_checkpoint && ((xbt_fifo_size(mc_stack_safety) + 1) % _sg_mc_checkpoint == 0)){
-          next_state->system_state = MC_take_snapshot();
+          next_state->system_state = MC_take_snapshot(next_state->num);
         }
 
         if(dot_output != NULL)
@@ -371,8 +620,6 @@ void MC_dpor(void)
             char *key = bprintf("%lu", process->pid);
             enabled = (int)strtoul(xbt_dict_get_or_null(first_enabled_state, key), 0, 10);
             xbt_free(key);
-            mc_state_t state_test = NULL;
-            xbt_fifo_item_t item = NULL;
             int cursor = xbt_fifo_size(mc_stack_safety);
             xbt_fifo_foreach(mc_stack_safety, item, state_test, mc_state_t){
               if(cursor-- == enabled){ 
@@ -382,7 +629,7 @@ void MC_dpor(void)
                   break;
                 }
               }
-            } 
+            }
           }
         }
 
@@ -392,11 +639,45 @@ void MC_dpor(void)
 
       }
 
-      /* Trash the current state, no longer needed */
       MC_SET_RAW_MEM;
+
+      if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
+        if(initial_state_safety->initial_communications_pattern_done){
+          if(interleave_size == 0){ /* if (interleave_size > 0), process interleaved but not enabled => "incorrect" path, determinism not evaluated */
+            //print_communications_pattern(communications_pattern);
+            deterministic_pattern(initial_communications_pattern, communications_pattern);
+            if(initial_state_safety->comm_deterministic == 0 && _sg_mc_comms_determinism){
+              XBT_INFO("****************************************************");
+              XBT_INFO("***** Non-deterministic communications pattern *****");
+              XBT_INFO("****************************************************");
+              XBT_INFO("Initial communications pattern:");
+              print_communications_pattern(initial_communications_pattern);
+              XBT_INFO("Communications pattern counter-example:");
+              print_communications_pattern(communications_pattern);
+              MC_print_statistics(mc_stats);
+              return;
+            }else if(initial_state_safety->send_deterministic == 0 && _sg_mc_send_determinism){
+              XBT_INFO("****************************************************");
+              XBT_INFO("***** Non-send-deterministic communications pattern *****");
+              XBT_INFO("****************************************************");
+              XBT_INFO("Initial communications pattern:");
+              print_communications_pattern(initial_communications_pattern);
+              XBT_INFO("Communications pattern counter-example:");
+              print_communications_pattern(communications_pattern);
+              MC_print_statistics(mc_stats);
+              return;
+            }
+          }
+        }else{
+          initial_state_safety->initial_communications_pattern_done = 1;
+        }
+      }
+
+      /* Trash the current state, no longer needed */
       xbt_fifo_shift(mc_stack_safety);
       MC_state_delete(state);
       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack_safety) + 1);
+
       MC_UNSET_RAW_MEM;        
       
       /* Check for deadlocks */
@@ -462,15 +743,15 @@ void MC_dpor(void)
               pos = xbt_fifo_size(mc_stack_safety);
               item = xbt_fifo_get_first_item(mc_stack_safety);
               while(pos>0){
-                restore_state = (mc_state_t) xbt_fifo_get_item_content(item);
-                if(restore_state->system_state != NULL){
+                restored_state = (mc_state_t) xbt_fifo_get_item_content(item);
+                if(restored_state->system_state != NULL){
                   break;
                 }else{
                   item = xbt_fifo_get_next_item(item);
                   pos--;
                 }
               }
-              MC_restore_snapshot(restore_state->system_state);
+              MC_restore_snapshot(restored_state->system_state);
               xbt_fifo_unshift(mc_stack_safety, state);
               MC_UNSET_RAW_MEM;
               MC_replay(mc_stack_safety, pos);
@@ -480,9 +761,16 @@ void MC_dpor(void)
             MC_UNSET_RAW_MEM;
             MC_replay(mc_stack_safety, -1);
           }
-          XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack_safety));
+          XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack_safety));
           break;
         } else {
+          req = MC_state_get_internal_request(state);
+          if(_sg_mc_comms_determinism){
+            if(req->call == SIMCALL_COMM_ISEND || req->call == SIMCALL_COMM_IRECV){
+              if(!xbt_dynar_is_empty(communications_pattern))
+                xbt_dynar_remove_at(communications_pattern, xbt_dynar_length(communications_pattern) - 1, NULL);
+            }
+          }
           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack_safety) + 1); 
           MC_state_delete(state);
         }