Document SG 3.5 bugs, and workarounds taken to circumvent them.

[loba.git] / communicator.cpp

diff --git a/communicator.cpp b/communicator.cpp
index 753b04a98ed4e529ae31dd38e6d57b85962fd54e..e30d73187dc8e184ece920423ed33cca03e16454 100644 (file)
--- a/communicator.cpp
+++ b/communicator.cpp
@@ -9,6 +9,8 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(comm);
 
 #include "misc.h"
 #include "options.h"
 
 #include "misc.h"
 #include "options.h"

+#include "simgrid_features.h"
+#include "tracing.h"

 
 #include "communicator.h"
 
 
 #include "communicator.h"
 


@@ -22,7 +24,7 @@ std::string message::to_string()
 }
 
 communicator::communicator()
 }
 
 communicator::communicator()

-    : host((hostdata* )MSG_host_get_data(MSG_host_self()))
+    : host(static_cast<hostdata*>(MSG_host_get_data(MSG_host_self())))

     , mutex(xbt_mutex_init())
     , cond(xbt_cond_init())
     , ctrl_task(NULL)
     , mutex(xbt_mutex_init())
     , cond(xbt_cond_init())
     , ctrl_task(NULL)


@@ -34,6 +36,7 @@ communicator::communicator()
     receiver_process =
         MSG_process_create("receiver", communicator::receiver_wrapper,
                            this, MSG_host_self());
     receiver_process =
         MSG_process_create("receiver", communicator::receiver_wrapper,
                            this, MSG_host_self());

+    xbt_cond_wait(cond, mutex); // wait for the receiver to be ready

     xbt_mutex_release(mutex);
 }
 
     xbt_mutex_release(mutex);
 }
 


@@ -78,17 +81,33 @@ void communicator::send(const char* dest, message* msg)
     if (msg->get_type() == message::LOAD)
         msg_size += opt::comm_cost(msg->get_amount());
     m_task_t task = MSG_task_create("message", 0.0, msg_size, msg);    
     if (msg->get_type() == message::LOAD)
         msg_size += opt::comm_cost(msg->get_amount());
     m_task_t task = MSG_task_create("message", 0.0, msg_size, msg);    

+    TRACE_msg_set_task_category(task,
+                                msg->get_type() == message::LOAD ?
+                                TRACE_CAT_DATA : TRACE_CAT_CTRL);

     msg_comm_t comm = MSG_task_isend(task, dest);
     sent_comm.push_back(comm);
 }
 
     msg_comm_t comm = MSG_task_isend(task, dest);
     sent_comm.push_back(comm);
 }
 

-bool communicator::recv(message*& msg, m_host_t& from, bool wait)
+bool communicator::recv(message*& msg, m_host_t& from, double timeout)

 {
 {

-    if (wait) {
+    if (timeout != 0) {
+        volatile double deadline =
+            timeout > 0 ? MSG_get_clock() + timeout : 0.0;

         xbt_mutex_acquire(mutex);
         xbt_mutex_acquire(mutex);

-        while (received.empty()) {
+        while (received.empty() && (!deadline || deadline > MSG_get_clock())) {
+            xbt_ex_t e;

             DEBUG0("waiting for a message to come");
             DEBUG0("waiting for a message to come");

-            xbt_cond_wait(cond, mutex);
+            TRY {
+                if (deadline)
+                    xbt_cond_timedwait(cond, mutex, deadline - MSG_get_clock());
+                else
+                    xbt_cond_wait(cond, mutex);
+            }
+            CATCH (e) {
+                if (e.category != timeout_error)
+                    RETHROW;
+                xbt_ex_free(e);
+            }

         }
         xbt_mutex_release(mutex);
     }
         }
         xbt_mutex_release(mutex);
     }


@@ -98,7 +117,7 @@ bool communicator::recv(message*& msg, m_host_t& from, bool wait)
 
     m_task_t task = received.front();
     received.pop();
 
     m_task_t task = received.front();
     received.pop();

-    msg = (message* )MSG_task_get_data(task);
+    msg = static_cast<message*>(MSG_task_get_data(task));

     from = MSG_task_get_source(task);
     MSG_task_destroy(task);
 
     from = MSG_task_get_source(task);
     MSG_task_destroy(task);
 


@@ -110,21 +129,14 @@ bool communicator::recv(message*& msg, m_host_t& from, bool wait)
 
 void communicator::flush(bool wait)
 {
 
 void communicator::flush(bool wait)
 {

-    using std::tr1::bind;
-    using std::tr1::placeholders::_1;
-

     sent_comm.remove_if(comm_test_n_destroy);
     if (wait && !sent_comm.empty()) {
     sent_comm.remove_if(comm_test_n_destroy);
     if (wait && !sent_comm.empty()) {

-        xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
-        while (!sent_comm.empty()) {
-            std::for_each(sent_comm.begin(), sent_comm.end(),
-                          bind(xbt_dynar_push,
-                               comms, bind(misc::address<msg_comm_t>(), _1)));
-            MSG_comm_waitany(comms);
-            xbt_dynar_reset(comms);
-            sent_comm.remove_if(comm_test_n_destroy);
-        }
-        xbt_dynar_free(&comms);
+        msg_comm_t comms[sent_comm.size()];
+        std::copy(sent_comm.begin(), sent_comm.end(), comms);
+        MSG_comm_waitall(comms, sent_comm.size(), -1.0);
+        if (!MSG_WAIT_DESTROYS_COMMS)
+            std::for_each(sent_comm.begin(), sent_comm.end(), MSG_comm_destroy);
+        sent_comm.clear();

     }
 }
 
     }
 }
 


@@ -140,7 +152,7 @@ bool communicator::comm_test_n_destroy(msg_comm_t comm)
 int communicator::receiver_wrapper(int, char* [])
 {
     communicator* comm;
 int communicator::receiver_wrapper(int, char* [])
 {
     communicator* comm;

-    comm = (communicator* )MSG_process_get_data(MSG_process_self());
+    comm = static_cast<communicator*>(MSG_process_get_data(MSG_process_self()));

     int result = comm->receiver();
 
     DEBUG0("terminate");
     int result = comm->receiver();
 
     DEBUG0("terminate");


@@ -156,6 +168,11 @@ int communicator::receiver()
 {
     ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
     data_comm = MSG_task_irecv(&data_task, get_data_mbox());
 {
     ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
     data_comm = MSG_task_irecv(&data_task, get_data_mbox());

+    DEBUG0("receiver ready");
+    xbt_mutex_acquire(mutex);
+    xbt_cond_signal(cond);      // signal master that we are ready
+    xbt_mutex_release(mutex);
+

     xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
     while (ctrl_comm || data_comm) {
 
     xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
     while (ctrl_comm || data_comm) {
 


@@ -169,7 +186,9 @@ int communicator::receiver()
         if (ctrl_comm && comm_test_n_destroy(ctrl_comm)) {
             if (strcmp(MSG_task_get_name(ctrl_task), "finalize")) {
                 DEBUG0("received message from ctrl");
         if (ctrl_comm && comm_test_n_destroy(ctrl_comm)) {
             if (strcmp(MSG_task_get_name(ctrl_task), "finalize")) {
                 DEBUG0("received message from ctrl");

+                xbt_mutex_acquire(mutex);

                 received.push(ctrl_task);
                 received.push(ctrl_task);

+                xbt_mutex_release(mutex);

                 ctrl_task = NULL;
                 ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
             } else {
                 ctrl_task = NULL;
                 ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
             } else {


@@ -183,7 +202,9 @@ int communicator::receiver()
         if (data_comm && comm_test_n_destroy(data_comm)) {
             if (strcmp(MSG_task_get_name(data_task), "finalize")) {
                 DEBUG0("received message from data");
         if (data_comm && comm_test_n_destroy(data_comm)) {
             if (strcmp(MSG_task_get_name(data_task), "finalize")) {
                 DEBUG0("received message from data");

+                xbt_mutex_acquire(mutex);

                 received.push(data_task);
                 received.push(data_task);

+                xbt_mutex_release(mutex);

                 data_task = NULL;
                 data_comm = MSG_task_irecv(&data_task, get_data_mbox());
             } else {
                 data_task = NULL;
                 data_comm = MSG_task_irecv(&data_task, get_data_mbox());
             } else {


@@ -194,7 +215,8 @@ int communicator::receiver()
             }
         }
         xbt_mutex_acquire(mutex);
             }
         }
         xbt_mutex_acquire(mutex);

-        xbt_cond_signal(cond);
+        if (!received.empty())
+            xbt_cond_signal(cond);

         xbt_mutex_release(mutex);
     }
     xbt_dynar_free(&comms);
         xbt_mutex_release(mutex);
     }
     xbt_dynar_free(&comms);