Cleanups on communicator.

[loba.git] / communicator.cpp

diff --git a/communicator.cpp b/communicator.cpp
index e6653ebda4e9acf52d0c5a40acd8e8c4f09f1d61..96cd598fbaf807574f8870952a9e05c352f1784f 100644 (file)
--- a/communicator.cpp
+++ b/communicator.cpp
@@ -1,14 +1,18 @@
-#include "communicator.h"
-

 #include <algorithm>
 #include <algorithm>

+#include <cstring>

 #include <tr1/functional>
 #include <sstream>
 #include <msg/msg.h>
 #include <xbt/log.h>
 #include <tr1/functional>
 #include <sstream>
 #include <msg/msg.h>
 #include <xbt/log.h>

+
+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"

 
 

-XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(comm);
+#include "communicator.h"

 
 std::string message::to_string()
 {
 
 std::string message::to_string()
 {


@@ -19,133 +23,196 @@ std::string message::to_string()
     return oss.str();
 }
 
     return oss.str();
 }
 

-const int communicator::send_count_before_flush = 16;
-

 communicator::communicator()
 communicator::communicator()

-    : host((hostdata* )MSG_host_get_data(MSG_host_self()))
-    , send_counter(0)
-    , ctrl_task(NULL)
-    , ctrl_comm(NULL)
-    , ctrl_close_is_last(false)
-    , data_task(NULL)
-    , data_comm(NULL)
-    , data_close_is_last(false)
+    : host(static_cast<hostdata*>(MSG_host_get_data(MSG_host_self())))
+    , receiver_mutex(xbt_mutex_init())
+    , receiver_cond(xbt_cond_init())

 {
 {

+    xbt_mutex_acquire(receiver_mutex);
+    receiver_thread =
+        MSG_process_create("receiver", communicator::receiver_wrapper,
+                           this, MSG_host_self());
+    // wait for the receiver to be ready
+    xbt_cond_wait(receiver_cond, receiver_mutex);
+    xbt_mutex_release(receiver_mutex);

 }
 
 communicator::~communicator()
 {
 }
 
 communicator::~communicator()
 {

-    if (ctrl_comm)
-        WARN0("ctrl_comm is pending!");
-    if (data_comm)
-        WARN0("data_comm is pending!");
+    m_task_t task;
+
+    XBT_DEBUG("send finalize to receiver/ctrl");
+    task = MSG_task_create("finalize", 0.0, 0, NULL);
+    MSG_task_send(task, host->get_ctrl_mbox());
+
+    XBT_DEBUG("send finalize to receiver/data");
+    task = MSG_task_create("finalize", 0.0, 0, NULL);
+    MSG_task_send(task, host->get_data_mbox());
+
+    xbt_mutex_acquire(receiver_mutex);
+    while (receiver_thread) {
+        XBT_DEBUG("waiting for receiver to terminate");
+        xbt_cond_wait(receiver_cond, receiver_mutex);
+    }
+    xbt_mutex_release(receiver_mutex);
+
+    if (!received.empty())
+        XBT_WARN("lost %zu received message%s!",
+                 received.size(), ESSE(received.size()));

     if (!sent_comm.empty())
     if (!sent_comm.empty())

-        WARN2("lost %ld send communication%s!",
-              (long )sent_comm.size(), ESSE(sent_comm.size()));
-}
+        XBT_WARN("lost %zu sent message%s!",
+                 sent_comm.size(), ESSE(sent_comm.size()));

 
 

-void communicator::listen()
-{
-    ctrl_comm = MSG_task_irecv(&ctrl_task, get_ctrl_mbox());
-    data_comm = MSG_task_irecv(&data_task, get_data_mbox());
+    xbt_cond_destroy(receiver_cond);
+    xbt_mutex_destroy(receiver_mutex);

 }
 
 void communicator::send(const char* dest, message* msg)
 {
 }
 
 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);    
     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);
     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)

 {
 {

-    bool restart;
-    msg = NULL;
-
-    do {
-        if (ctrl_comm && comm_test_n_destroy(ctrl_comm)) {
-            msg = (message* )MSG_task_get_data(ctrl_task);
-            from = MSG_task_get_source(ctrl_task);
-            MSG_task_destroy(ctrl_task);
-            ctrl_task = NULL;
-            ctrl_comm =
-                (!ctrl_close_is_last || msg->get_type() != message::CTRL_CLOSE)
-                ? MSG_task_irecv(&ctrl_task, get_ctrl_mbox())
-                : NULL;
-
-        } else if (data_comm && comm_test_n_destroy(data_comm)) {
-            msg = (message* )MSG_task_get_data(data_task);
-            from = MSG_task_get_source(data_task);
-            MSG_task_destroy(data_task);
-            data_task = NULL;
-            data_comm =
-                (!data_close_is_last || msg->get_type() != message::DATA_CLOSE)
-                ? MSG_task_irecv(&data_task, get_data_mbox())
-                : NULL;
+    if (timeout != 0) {
+        volatile double deadline =
+            timeout > 0 ? MSG_get_clock() + timeout : 0.0;
+        xbt_mutex_acquire(receiver_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(receiver_cond, receiver_mutex,
+                                       deadline - MSG_get_clock());
+                else
+                    xbt_cond_wait(receiver_cond, receiver_mutex);
+            }
+            CATCH (e) {
+                if (e.category != timeout_error)
+                    RETHROW;
+                xbt_ex_free(e);
+            }

         }
         }

+        xbt_mutex_release(receiver_mutex);
+    }

 
 

-        restart = wait && !msg && (ctrl_comm || data_comm);
-        if (restart) {
-            xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
-            if (ctrl_comm)
-                xbt_dynar_push(comms, &ctrl_comm);
-            if (data_comm)
-                xbt_dynar_push(comms, &data_comm);
-            MSG_comm_waitany(comms);
-            xbt_dynar_free(&comms);
-        }
-    } while (restart);
+    if (received.empty())
+        return false;
+
+    m_task_t task = received.front();
+    received.pop();
+    msg = static_cast<message*>(MSG_task_get_data(task));
+    from = MSG_task_get_source(task);
+    MSG_task_destroy(task);

 
 

-    if (msg)
-        DEBUG2("received %s from %s",
-               msg->to_string().c_str(), MSG_host_get_name(from));
+    XBT_DEBUG("received %s from %s",
+           msg->to_string().c_str(), MSG_host_get_name(from));

 
 

-    return msg != NULL;
+    return true;

 }
 
 void communicator::flush(bool wait)
 {
 }
 
 void communicator::flush(bool wait)
 {

-    using namespace std::tr1;
-    using namespace std::tr1::placeholders;
-

     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()) {
+        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(),
             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);
+                          comm_check_n_destroy);
+        sent_comm.clear();

     }
 }
 
     }
 }
 

-void communicator::next_close_on_ctrl_is_last()
+int communicator::receiver_wrapper(int, char* [])
+{
+    communicator* comm;
+    comm = static_cast<communicator*>(MSG_process_get_data(MSG_process_self()));
+    comm->receiver();
+
+    XBT_DEBUG("terminate");
+    xbt_mutex_acquire(comm->receiver_mutex);
+    comm->receiver_thread = NULL;
+    xbt_cond_signal(comm->receiver_cond);
+    xbt_mutex_release(comm->receiver_mutex);
+
+    return 0;
+}
+
+void communicator::receiver()

 {
 {

-    ctrl_close_is_last = true;
+    xbt_dynar_t comms = xbt_dynar_new(sizeof(msg_comm_t), NULL);
+    struct channel {
+        msg_comm_t comm;
+        m_task_t task;
+        const char* mbox;
+    };
+    channel chan[] = { { NULL, NULL, host->get_ctrl_mbox() },
+                       { NULL, NULL, host->get_data_mbox() } };
+    const int chan_size = (sizeof chan) / (sizeof chan[0]);
+
+    for (int i = 0 ; i < chan_size ; ++i) {
+        chan[i].comm = MSG_task_irecv(&chan[i].task, chan[i].mbox);
+        xbt_dynar_push(comms, &chan[i].comm);
+    }
+
+    XBT_DEBUG("receiver ready");
+    xbt_mutex_acquire(receiver_mutex);
+    xbt_cond_signal(receiver_cond); // signal master that we are ready
+    xbt_mutex_release(receiver_mutex);
+
+    while (!xbt_dynar_is_empty(comms)) {
+
+        int index = MSG_comm_waitany(comms);
+        msg_comm_t finished_comm = xbt_dynar_get_as(comms, index, msg_comm_t);
+        channel* ch;
+
+        for (ch = chan ; ch->comm != finished_comm ; ++ch)
+            /* nop */;
+
+        comm_check_n_destroy(ch->comm);
+        if (strcmp(MSG_task_get_name(ch->task), "finalize")) {
+            XBT_DEBUG("received message on %s", ch->mbox);
+            xbt_mutex_acquire(receiver_mutex);
+            received.push(ch->task);
+            xbt_cond_signal(receiver_cond);
+            xbt_mutex_release(receiver_mutex);
+            ch->task = NULL;
+            ch->comm = MSG_task_irecv(&ch->task, ch->mbox);
+            xbt_dynar_set_as(comms, index, msg_comm_t, ch->comm);
+        } else {
+            XBT_DEBUG("received finalize on %s", ch->mbox);
+            MSG_task_destroy(ch->task);
+            ch->task = NULL;
+            ch->comm = NULL;
+            xbt_dynar_remove_at(comms, index, NULL);
+        }
+
+    }
+    xbt_dynar_free(&comms);

 }
 
 }
 

-void communicator::next_close_on_data_is_last()
+void communicator::comm_check_n_destroy(msg_comm_t comm)

 {
 {

-    data_close_is_last = true;
+    xbt_assert(MSG_comm_get_status(comm) == MSG_OK);
+    MSG_comm_destroy(comm);

 }
 
 bool communicator::comm_test_n_destroy(msg_comm_t comm)
 {
     if (MSG_comm_test(comm)) {
 }
 
 bool communicator::comm_test_n_destroy(msg_comm_t comm)
 {
     if (MSG_comm_test(comm)) {

-        MSG_comm_destroy(comm);
+        comm_check_n_destroy(comm);

         return true;
     } else
         return false;
         return true;
     } else
         return false;