There is a bug...

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index ac9ab244dfd65103da9cfd2705f2948ad65372d3..0544b21d9e5b62b08a6a09a25618386913c0d377 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -33,6 +33,8 @@ process::process(int argc, char* argv[])
         rev_neigh.insert(std::make_pair(host, ptr));
     }
 
+    comp = 0.0;
+
     prev_load_broadcast = -1;   // force sending of load on first send()
     expected_load = load;
     total_load_running += load;
@@ -75,45 +77,45 @@ int process::run()
             if (now < next_iter_after_date){
                 double delay = next_iter_after_date - now;
                 DEBUG1("sleeping for %g s", delay);
-                MSG_process_sleep(next_iter_after_date - now);
+                MSG_process_sleep(delay);
             }
             next_iter_after_date = MSG_get_clock() + opt::min_iter_duration;
         }
 
-        if (load > 0.0) {
-            ++comp_iter;
-            if (opt::log_rate && comp_iter % opt::log_rate == 0) {
+        double ld = lb_load();
+        if (ld > 0.0) {
+            ++lb_iter;
+
+            if (opt::log_rate && lb_iter % opt::log_rate == 0) {
                 if (opt::bookkeeping)
                     INFO4("(%u:%u) current load: %g ; expected: %g",
-                          comp_iter, lb_iter, load, expected_load);
+                          lb_iter, comp_iter, load, expected_load);
                 else
                     INFO2("(%u) current load: %g",
-                          comp_iter, load);
+                          lb_iter, load);
             }
 
-            if (opt::bookkeeping)
-                expected_load -= load_balance(expected_load);
-            else
-                load -= load_balance(load);
+            ld -= load_balance(ld);
 
             print_loads(true, xbt_log_priority_debug);
+        }
+        lb_load() = ld;
 
-            send();
+        // send load information, and load (data) if any
+        send();
+        if (load > 0.0) {
+            ++comp_iter;
             compute();
-
-        } else {
-            // send load information, and load when bookkeeping
-            send();
         }
 
-        if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
-            VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
-            break;
-        }
         if (opt::lb_maxiter && lb_iter >= opt::lb_maxiter) {
             VERB2("Reached lb_maxiter: %d/%d", lb_iter, opt::lb_maxiter);
             break;
         }
+        if (opt::comp_maxiter && comp_iter >= opt::comp_maxiter) {
+            VERB2("Reached comp_maxiter: %d/%d", comp_iter, opt::comp_maxiter);
+            break;
+        }
         if (opt::time_limit && MSG_get_clock() >= opt::time_limit) {
             VERB2("Reached time limit: %g/%g", MSG_get_clock(), opt::time_limit);
             break;
@@ -121,9 +123,7 @@ int process::run()
 
         // block on receiving unless there is something to compute or
         // to send
-        bool wait = (load == 0 &&
-                     ((opt::bookkeeping ? expected_load : load)
-                      == prev_load_broadcast));
+        bool wait = (load == 0 && lb_load() == prev_load_broadcast);
         receive(wait);
 
         // one of our neighbor is finalizing
@@ -145,6 +145,8 @@ int process::run()
             opt::load_ratio_threshold) {
             VERB0("No more load to balance in system.");
             break;
+        } else {
+            DEBUG1("still %g load to balance, continuing...", total_load_running);
         }
     }
     VERB0("Going to finalize...");
@@ -159,13 +161,13 @@ int process::run()
     VERB0("Done.");
     if (opt::bookkeeping) {
         INFO4("Final load after %d:%d iterations: %g ; expected: %g",
-              comp_iter, lb_iter, load, expected_load);
+              lb_iter, comp_iter, load, expected_load);
     } else {
         INFO2("Final load after %d iterations: %g",
-              comp_iter, load);
-        if (comp_iter != lb_iter)
-            WARN2("comp_iter (%d) and lb_iter (%d) differ!",
-                  comp_iter, lb_iter);
+              lb_iter, load);
+        if (lb_iter != comp_iter)
+            WARN2("lb_iter (%d) and comp_iter (%d) differ!",
+                  lb_iter, comp_iter);
     }
     return 0;
 }
@@ -191,10 +193,11 @@ double process::load_balance(double /*my_load*/)
 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);
-        DEBUG2("compute %g flop%s", duration, ESSE(duration));
+        double flops = opt::comp_cost(load);
+        m_task_t task = MSG_task_create("computation", flops, 0.0, NULL);
+        DEBUG2("compute %g flop%s", flops, ESSE(flops));
         MSG_task_execute(task);
+        comp += flops;
         MSG_task_destroy(task);
     } else {
         DEBUG0("nothing to compute !");
@@ -263,6 +266,7 @@ void process::receive(bool wait)
     message* msg;
     m_host_t from;
 
+    DEBUG1("%sblocking receive", "\0non-" + !wait);
     while (may_receive() && comm.recv(msg, from, wait)) {
         switch (msg->get_type()) {
         case message::INFO: {