]> AND Private Git Repository - loba.git/blobdiff - communicator.cpp
Logo AND Algorithmique Numérique Distribuée

Private GIT Repository
Define opt::min_comp_iter_duration.
[loba.git] / communicator.cpp
index 94cdb04160337353a0a1bab8b2797d88c0ce415d..0b06d6c12d444b033b5e2940ef16f07fb7b74f89 100644 (file)
@@ -9,6 +9,8 @@ XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(comm);
 
 #include "misc.h"
 #include "options.h"
+#include "simgrid_features.h"
+#include "tracing.h"
 
 #include "communicator.h"
 
@@ -21,13 +23,10 @@ std::string message::to_string()
     return oss.str();
 }
 
-const int communicator::send_count_before_flush = 4;
-
 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())
-    , send_counter(0)
     , ctrl_task(NULL)
     , ctrl_comm(NULL)
     , data_task(NULL)
@@ -37,6 +36,7 @@ communicator::communicator()
     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);
 }
 
@@ -44,30 +44,31 @@ communicator::~communicator()
 {
     m_task_t task;
 
-    DEBUG0("send finalize to receiver/ctrl");
+    XBT_DEBUG("send finalize to receiver/ctrl");
     task = MSG_task_create("finalize", 0.0, 0, NULL);
     MSG_task_send(task, get_ctrl_mbox());
 
-    DEBUG0("send finalize to receiver/data");
+    XBT_DEBUG("send finalize to receiver/data");
     task = MSG_task_create("finalize", 0.0, 0, NULL);
     MSG_task_send(task, get_data_mbox());
 
-    DEBUG0("wait for receiver to terminate");
     xbt_mutex_acquire(mutex);
-    while (receiver_process)
+    while (receiver_process) {
+        XBT_DEBUG("waiting for receiver to terminate");
         xbt_cond_wait(cond, mutex);
+    }
     xbt_mutex_release(mutex);
 
     if (ctrl_comm)
-        WARN0("ctrl_comm is pending!");
+        XBT_WARN("ctrl_comm is pending!");
     if (data_comm)
-        WARN0("data_comm is pending!");
+        XBT_WARN("data_comm is pending!");
     if (!received.empty())
-        WARN2("lost %lu received message%s!",
-              (unsigned long )received.size(), ESSE(received.size()));
+        XBT_WARN("lost %zu received message%s!",
+                 received.size(), ESSE(received.size()));
     if (!sent_comm.empty())
-        WARN2("lost %lu sent message%s!",
-              (unsigned long )sent_comm.size(), ESSE(sent_comm.size()));
+        XBT_WARN("lost %zu sent message%s!",
+                 sent_comm.size(), ESSE(sent_comm.size()));
 
     xbt_cond_destroy(cond);
     xbt_mutex_destroy(mutex);
@@ -75,27 +76,39 @@ communicator::~communicator()
 
 void communicator::send(const char* dest, message* msg)
 {
-    DEBUG2("send %s to %s", msg->to_string().c_str(), dest);
+    XBT_DEBUG("send %s to %s", msg->to_string().c_str(), dest);
     double msg_size = sizeof *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);
-
-    if (++send_counter >= send_count_before_flush) {
-        flush(false);
-        send_counter = 0;
-    }
 }
 
-bool communicator::recv(message*& msg, m_host_t& from, bool wait)
+bool communicator::recv(message*& msg, m_host_t& from, double timeout)
 {
-    if (wait) {
-        DEBUG0("suspend main process on recv");
+    if (timeout != 0) {
+        volatile double deadline =
+            timeout > 0 ? MSG_get_clock() + timeout : 0.0;
         xbt_mutex_acquire(mutex);
-        while (received.empty())
-            xbt_cond_wait(cond, mutex);
+        while (received.empty() && (!deadline || deadline > MSG_get_clock())) {
+            xbt_ex_t e;
+            XBT_DEBUG("waiting for a message to come");
+            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);
     }
 
@@ -104,11 +117,11 @@ bool communicator::recv(message*& msg, m_host_t& from, bool wait)
 
     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);
 
-    DEBUG2("received %s from %s",
+    XBT_DEBUG("received %s from %s",
            msg->to_string().c_str(), MSG_host_get_name(from));
 
     return true;
@@ -116,21 +129,14 @@ bool communicator::recv(message*& msg, m_host_t& from, 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()) {
-        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();
     }
 }
 
@@ -146,10 +152,10 @@ bool communicator::comm_test_n_destroy(msg_comm_t 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");
+    XBT_DEBUG("terminate");
     xbt_mutex_acquire(comm->mutex);
     comm->receiver_process = NULL;
     xbt_cond_signal(comm->cond);
@@ -158,10 +164,34 @@ int communicator::receiver_wrapper(int, char* [])
     return result;
 }
 
+void communicator::receiver1(msg_comm_t& comm, m_task_t& task, const char* mbox)
+{
+    MSG_comm_destroy(comm);
+    if (strcmp(MSG_task_get_name(task), "finalize")) {
+        XBT_DEBUG("received message on %s", mbox);
+        xbt_mutex_acquire(mutex);
+        received.push(task);
+        xbt_cond_signal(cond);
+        xbt_mutex_release(mutex);
+        task = NULL;
+        comm = MSG_task_irecv(&task, mbox);
+    } else {
+        XBT_DEBUG("received finalize on %s", mbox);
+        MSG_task_destroy(task);
+        task = NULL;
+        comm = NULL;
+    }
+}
+
 int communicator::receiver()
 {
     ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
     data_comm = MSG_task_irecv(&data_task, get_data_mbox());
+    XBT_DEBUG("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) {
 
@@ -169,39 +199,18 @@ int communicator::receiver()
             xbt_dynar_push(comms, &ctrl_comm);
         if (data_comm)
             xbt_dynar_push(comms, &data_comm);
-        MSG_comm_waitany(comms);
+        int recvd = MSG_comm_waitany(comms);
+        msg_comm_t comm = xbt_dynar_get_as(comms, recvd, msg_comm_t);
         xbt_dynar_reset(comms);
 
-        if (ctrl_comm && comm_test_n_destroy(ctrl_comm)) {
-            if (strcmp(MSG_task_get_name(ctrl_task), "finalize")) {
-                DEBUG0("received message from ctrl");
-                received.push(ctrl_task);
-                ctrl_task = NULL;
-                ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
-            } else {
-                DEBUG0("received finalize from ctrl");
-                MSG_task_destroy(ctrl_task);
-                ctrl_task = NULL;
-                ctrl_comm = NULL;
-            }
+        if (comm == ctrl_comm)
+            receiver1(ctrl_comm, ctrl_task, get_ctrl_mbox());
+        else if (comm == data_comm)
+            receiver1(data_comm, data_task, get_data_mbox());
+        else {
+            XBT_ERROR("Handling unknown comm -- %p", comm);
+            MSG_comm_destroy(comm);
         }
-
-        if (data_comm && comm_test_n_destroy(data_comm)) {
-            if (strcmp(MSG_task_get_name(data_task), "finalize")) {
-                DEBUG0("received message from data");
-                received.push(data_task);
-                data_task = NULL;
-                data_comm = MSG_task_irecv(&data_task, get_data_mbox());
-            } else {
-                DEBUG0("received finalize from data");
-                MSG_task_destroy(data_task);
-                data_task = NULL;
-                data_comm = NULL;
-            }
-        }
-        xbt_mutex_acquire(mutex);
-        xbt_cond_signal(cond);
-        xbt_mutex_release(mutex);
     }
     xbt_dynar_free(&comms);
     return 0;