Bug fix: we really need to consume all messages in proc::receive!

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index 306687f2f8dd68de8d2bf82218adff88e4ff70d8..53cbb59c7434da4a84beb5d2971e8c98ea7ae910 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -39,16 +39,8 @@ process::process(int argc, char* argv[])
     total_load_init += load;
 
     ctrl_close_pending = data_close_pending = neigh.size();
     total_load_init += load;
 
     ctrl_close_pending = data_close_pending = neigh.size();

-    if (neigh.size() == 1) {
-        comm.next_close_on_ctrl_is_last();
-        comm.next_close_on_data_is_last();
-    }

     close_received = false;
     close_received = false;

-    may_receive =  (neigh.size() > 0); // the same as (ctrl_close_pending ||
-                                       //              data_close_pending)

     finalizing = false;
     finalizing = false;

-    if (may_receive)
-        comm.listen();

 
     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
     if (!LOG_ISENABLED(logp))
 
     e_xbt_log_priority_t logp = xbt_log_priority_verbose;
     if (!LOG_ISENABLED(logp))


@@ -75,17 +67,17 @@ int process::run()
 {
     INFO1("Initial load: %g", load);
     VERB0("Starting...");
 {
     INFO1("Initial load: %g", load);
     VERB0("Starting...");

-    iter = 0;
+    comp_iter = lb_iter = 0;

     while (true) {
         if (load > 0.0) {
     while (true) {
         if (load > 0.0) {

-            ++iter;
-            if (opt::log_rate && iter % opt::log_rate == 0) {
+            ++comp_iter;
+            if (opt::log_rate && comp_iter % opt::log_rate == 0) {

                 if (opt::bookkeeping)
                 if (opt::bookkeeping)

-                    INFO3("(%u) current load: %g ; expected: %g",
-                          iter, load, expected_load);
+                    INFO4("(%u:%u) current load: %g ; expected: %g",
+                          comp_iter, lb_iter, load, expected_load);

                 else
                     INFO2("(%u) current load: %g",
                 else
                     INFO2("(%u) current load: %g",

-                          iter, load);
+                          comp_iter, load);

             }
 
             if (opt::bookkeeping)
             }
 
             if (opt::bookkeeping)


@@ -98,31 +90,34 @@ int process::run()
             send();
             compute();
 
             send();
             compute();
 

-            if (opt::maxiter && iter >= opt::maxiter)
-                break;

         } else {
             // send load information, and load when bookkeeping
             send();
         }
 
         } else {
             // send load information, and load when bookkeeping
             send();
         }
 

+        if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter)
+            break;
+        if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter)
+            break;
+

         // block on receiving unless there is something to compute or
         // to send
         // block on receiving unless there is something to compute or
         // to send

-        bool recv_wait = (load == 0 &&
-                          ((opt::bookkeeping ? expected_load : load)
-                           == prev_load_broadcast));
-        DEBUG1("CALL RECEIVE(%s)", recv_wait? "WAIT": "NO_WAIT");
-        receive(recv_wait? WAIT: NO_WAIT);
+        bool wait = (load == 0 &&
+                     ((opt::bookkeeping ? expected_load : load)
+                      == prev_load_broadcast));
+        receive(wait);

 
         // one of our neighbor is finalizing
         if (opt::exit_on_close && close_received)
             break;
 
         // have no load and cannot receive anything
 
         // one of our neighbor is finalizing
         if (opt::exit_on_close && close_received)
             break;
 
         // have no load and cannot receive anything

-        if (load == 0.0 && !may_receive)
+        if (load == 0.0 && !may_receive())

             break;
 
         // fixme: this check should be implemented with a distributed
         // algorithm, and not a shared global variable!
             break;
 
         // fixme: this check should be implemented with a distributed
         // algorithm, and not a shared global variable!

+        // fixme: should this chunk be moved before call to receive() ?

         if (100.0 * total_load_running / total_load_init <=
             opt::load_ratio_threshold) {
             VERB0("No more load to balance in system, stopping.");
         if (100.0 * total_load_running / total_load_init <=
             opt::load_ratio_threshold) {
             VERB0("No more load to balance in system, stopping.");


@@ -140,7 +135,8 @@ int process::run()
      */
 
     VERB0("Done.");
      */
 
     VERB0("Done.");

-    INFO3("Final load after %d iteration%s: %g", iter, ESSE(iter), load);
+    INFO3("Final load after %d iteration%s: %g",
+          comp_iter, ESSE(comp_iter), load);

     if (opt::bookkeeping)
         INFO1("Expected load: %g", expected_load);
     return 0;
     if (opt::bookkeeping)
         INFO1("Expected load: %g", expected_load);
     return 0;


@@ -159,7 +155,7 @@ double process::sum_of_to_send() const
 
 double process::load_balance(double /*my_load*/)
 {
 
 double process::load_balance(double /*my_load*/)
 {

-    if (iter == 1)
+    if (lb_iter == 1)           // warn only once

         WARN0("process::load_balance is a no-op!");
     return 0.0;
 }
         WARN0("process::load_balance is a no-op!");
     return 0.0;
 }


@@ -169,7 +165,7 @@ void process::compute()
     if (load > 0.0) {
         double duration = opt::comp_cost(load);
         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);
     if (load > 0.0) {
         double duration = opt::comp_cost(load);
         m_task_t task = MSG_task_create("computation", duration, 0.0, NULL);

-        DEBUG2("compute %g flop%s.", duration, ESSE(duration));
+        DEBUG2("compute %g flop%s", duration, ESSE(duration));

         MSG_task_execute(task);
         MSG_task_destroy(task);
     } else {
         MSG_task_execute(task);
         MSG_task_destroy(task);
     } else {


@@ -233,14 +229,12 @@ void process::send()
     }
 }
 
     }
 }
 

-void process::receive(recv_wait_mode wait)
+void process::receive(bool wait)

 {
 {

-    // DEBUG1("go for receive(%s)",
-    //        "NO_WAIT\0WAIT\0\0\0\0WAIT_FOR_CLOSE" + 8 * wait);

     message* msg;
     m_host_t from;
     message* msg;
     m_host_t from;

-    bool do_wait = (wait != NO_WAIT);
-    while (may_receive && comm.recv(msg, from, do_wait)) {
+
+    while (may_receive() && comm.recv(msg, from, wait)) {

         switch (msg->get_type()) {
         case message::INFO: {
             neighbor* n = rev_neigh[from];
         switch (msg->get_type()) {
         case message::INFO: {
             neighbor* n = rev_neigh[from];


@@ -258,21 +252,16 @@ void process::receive(recv_wait_mode wait)
             break;
         }
         case message::CTRL_CLOSE:
             break;
         }
         case message::CTRL_CLOSE:

-            if (--ctrl_close_pending == 1)
-                comm.next_close_on_ctrl_is_last();
-            // DEBUG1("ctrl_close_pending = %d", ctrl_close_pending);
+            ctrl_close_pending--;

             close_received = true;
             break;
         case message::DATA_CLOSE:
             close_received = true;
             break;
         case message::DATA_CLOSE:

-            if (--data_close_pending == 1)
-                comm.next_close_on_data_is_last();
-            // DEBUG1("data_close_pending = %d", data_close_pending);
+            data_close_pending--;

             close_received = true;
             break;
         }
         delete msg;
             close_received = true;
             break;
         }
         delete msg;

-        may_receive = (ctrl_close_pending || data_close_pending);
-        do_wait = (wait == WAIT_FOR_CLOSE);
+        wait = false;           // only wait on first recv

     }
 }
 
     }
 }
 


@@ -290,14 +279,15 @@ void process::finalize()
     finalizing = true;
     total_load_running -= load;
 
     finalizing = true;
     total_load_running -= load;
 

-    DEBUG2("send CLOSE to %d neighbor%s.",
-           (int )neigh.size(), ESSE(neigh.size()));
+    DEBUG2("send CLOSE to %lu neighbor%s",
+           (unsigned long )neigh.size(), ESSE(neigh.size()));

     std::for_each(neigh.begin(), neigh.end(),
                   bind(&process::finalize1, this, _1));
 
     std::for_each(neigh.begin(), neigh.end(),
                   bind(&process::finalize1, this, _1));
 

-    DEBUG2("wait for CLOSE from %d neighbor%s.",
-           (int )neigh.size(), ESSE(neigh.size()));
-    receive(WAIT_FOR_CLOSE);
+    DEBUG2("wait for CLOSE from %lu neighbor%s",
+           (unsigned long )neigh.size(), ESSE(neigh.size()));
+    while (may_receive())
+        receive(true);

 
     comm.flush(true);
 }
 
     comm.flush(true);
 }