Experimental tracing support.

[loba.git] / process.cpp

diff --git a/process.cpp b/process.cpp
index 0afbf2950781c4ac33139c2cbfbb3f2bf0a70ebe..03c173432cbfee93576d6dfc6ca65909b6ab0da0 100644 (file)
--- a/process.cpp
+++ b/process.cpp
@@ -11,6 +11,7 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(proc);
 
 #include "misc.h"
 #include "options.h"
+#include "tracing.h"
 
 #include "process.h"
 
@@ -33,6 +34,8 @@ 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;
@@ -70,49 +73,45 @@ int process::run()
     VERB0("Starting...");
     comp_iter = lb_iter = 0;
     while (true) {
-        if (opt::min_iter_duration) {
+        double ld = lb_load();
+        if (ld > 0.0) {
             double now = MSG_get_clock();
-            if (now < next_iter_after_date){
-                double delay = next_iter_after_date - now;
-                DEBUG1("sleeping for %g s", delay);
+            if (now < next_iter_after_date)
                 MSG_process_sleep(next_iter_after_date - now);
-            }
             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) {
+            ++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);
             }
 
-            double ld = lb_load();
             ld -= load_balance(ld);
-            lb_load() = 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;
@@ -120,8 +119,14 @@ int process::run()
 
         // block on receiving unless there is something to compute or
         // to send
-        bool wait = (load == 0 && lb_load() == prev_load_broadcast);
-        receive(wait);
+        double timeout;
+        if (load != 0 || lb_load() != prev_load_broadcast)
+            timeout = 0.0;
+        else if (opt::min_iter_duration)
+            timeout = opt::min_iter_duration;
+        else
+            timeout = 1.0;
+        receive(timeout);
 
         // one of our neighbor is finalizing
         if (opt::exit_on_close && close_received) {
@@ -142,6 +147,8 @@ 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...");
@@ -156,13 +163,13 @@ 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;
 }
@@ -188,10 +195,12 @@ 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);
+        TRACE_msg_set_task_category(task, TRACE_CAT_COMP);
+        DEBUG2("compute %g flop%s", flops, ESSE(flops));
         MSG_task_execute(task);
+        comp += flops;
         MSG_task_destroy(task);
     } else {
         DEBUG0("nothing to compute !");
@@ -255,12 +264,13 @@ void process::send()
     comm.flush(false);
 }
 
-void process::receive(bool wait)
+void process::receive(double timeout)
 {
     message* msg;
     m_host_t from;
 
-    while (may_receive() && comm.recv(msg, from, wait)) {
+    DEBUG2("%sblocking receive (%g)", "\0non-" + !timeout, timeout);
+    while (may_receive() && comm.recv(msg, from, timeout)) {
         switch (msg->get_type()) {
         case message::INFO: {
             neighbor* n = rev_neigh[from];
@@ -287,7 +297,7 @@ void process::receive(bool wait)
             break;
         }
         delete msg;
-        wait = false;           // only wait on first recv
+        timeout = 0.0;          // only wait on first recv
     }
     comm.flush(false);
 }
@@ -315,7 +325,7 @@ void process::finalize()
            (unsigned long )neigh.size(), ESSE(neigh.size()));
     while (may_receive()) {
         comm.flush(false);
-        receive(true);
+        receive(-1.0);
     }
 
     comm.flush(true);