X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/4ff3a4cf6e03165421f65616a22be86b8f528e21..ab4d43a39df33bb5f24c3b2cc6d576130b16a649:/src/smpi/mpi/smpi_request.cpp

diff --git a/src/smpi/mpi/smpi_request.cpp b/src/smpi/mpi/smpi_request.cpp
index b8fa3dbcfa..3b4f02d013 100644
--- a/src/smpi/mpi/smpi_request.cpp
+++ b/src/smpi/mpi/smpi_request.cpp
@@ -16,6 +16,7 @@
 #include "src/mc/mc_replay.hpp"
 #include "src/simix/ActorImpl.hpp"
 #include "xbt/config.hpp"
+#include <xbt/ex.hpp>
 
 #include <algorithm>
 
@@ -67,6 +68,7 @@ Request::Request(void* buf, int count, MPI_Datatype datatype, int src, int dst,
   else
     refcount_ = 0;
   op_   = MPI_REPLACE;
+  cancelled_ = 0;
 }
 
 MPI_Comm Request::comm(){
@@ -93,6 +95,10 @@ int Request::detached(){
   return detached_;
 }
 
+MPI_Datatype Request::type(){
+  return old_type_;
+}
+
 size_t Request::size(){
   return size_;
 }
@@ -105,7 +111,10 @@ void Request::unref(MPI_Request* request)
 {
   if((*request) != MPI_REQUEST_NULL){
     (*request)->refcount_--;
-    if((*request)->refcount_<0) xbt_die("wrong refcount");
+    if((*request)->refcount_ < 0) {
+      (*request)->print_request("wrong refcount");
+      xbt_die("Whoops, wrong refcount");
+    }
     if((*request)->refcount_==0){
         Datatype::unref((*request)->old_type_);
         Comm::unref((*request)->comm_);
@@ -139,6 +148,8 @@ int Request::match_recv(void* a, void* b, simgrid::kernel::activity::CommImpl* i
       ref->truncated_ = 1;
     if(req->detached_==1)
       ref->detached_sender_=req; //tie the sender to the receiver, as it is detached and has to be freed in the receiver
+    if(req->cancelled_==0)
+      req->cancelled_=-1;//mark as uncancellable
     XBT_DEBUG("match succeeded");
     return 1;
   }else return 0;
@@ -162,6 +173,8 @@ int Request::match_send(void* a, void* b, simgrid::kernel::activity::CommImpl* i
       req->truncated_ = 1;
     if(ref->detached_==1)
       req->detached_sender_=ref; //tie the sender to the receiver, as it is detached and has to be freed in the receiver
+    if(req->cancelled_==0)
+      req->cancelled_=-1;//mark as uncancellable
     XBT_DEBUG("match succeeded");
     return 1;
   } else
@@ -179,19 +192,19 @@ void Request::print_request(const char *message)
 MPI_Request Request::send_init(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
 {
 
-  return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::getPid(),
+  return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
                      comm->group()->actor(dst)->get_pid(), tag, comm, PERSISTENT | SEND | PREPARED);
 }
 
 MPI_Request Request::ssend_init(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
 {
-  return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::getPid(),
+  return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
                      comm->group()->actor(dst)->get_pid(), tag, comm, PERSISTENT | SSEND | SEND | PREPARED);
 }
 
 MPI_Request Request::isend_init(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
 {
-  return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::getPid(),
+  return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
                      comm->group()->actor(dst)->get_pid(), tag, comm, PERSISTENT | ISEND | SEND | PREPARED);
 }
 
@@ -217,7 +230,7 @@ MPI_Request Request::recv_init(void *buf, int count, MPI_Datatype datatype, int
 {
   return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype,
                      src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : comm->group()->actor(src)->get_pid(),
-                     simgrid::s4u::this_actor::getPid(), tag, comm, PERSISTENT | RECV | PREPARED);
+                     simgrid::s4u::this_actor::get_pid(), tag, comm, PERSISTENT | RECV | PREPARED);
 }
 
 MPI_Request Request::rma_recv_init(void *buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm,
@@ -240,13 +253,13 @@ MPI_Request Request::irecv_init(void *buf, int count, MPI_Datatype datatype, int
 {
   return new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype,
                      src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : comm->group()->actor(src)->get_pid(),
-                     simgrid::s4u::this_actor::getPid(), tag, comm, PERSISTENT | RECV | PREPARED);
+                     simgrid::s4u::this_actor::get_pid(), tag, comm, PERSISTENT | RECV | PREPARED);
 }
 
 MPI_Request Request::isend(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
 {
   MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
-  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::getPid(),
+  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
                         comm->group()->actor(dst)->get_pid(), tag, comm, NON_PERSISTENT | ISEND | SEND);
   request->start();
   return request;
@@ -255,7 +268,7 @@ MPI_Request Request::isend(void *buf, int count, MPI_Datatype datatype, int dst,
 MPI_Request Request::issend(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
 {
   MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
-  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::getPid(),
+  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
                         comm->group()->actor(dst)->get_pid(), tag, comm, NON_PERSISTENT | ISEND | SSEND | SEND);
   request->start();
   return request;
@@ -267,7 +280,7 @@ MPI_Request Request::irecv(void *buf, int count, MPI_Datatype datatype, int src,
   MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
   request             = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype,
                         src == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : comm->group()->actor(src)->get_pid(),
-                        simgrid::s4u::this_actor::getPid(), tag, comm, NON_PERSISTENT | RECV);
+                        simgrid::s4u::this_actor::get_pid(), tag, comm, NON_PERSISTENT | RECV);
   request->start();
   return request;
 }
@@ -283,7 +296,7 @@ void Request::recv(void *buf, int count, MPI_Datatype datatype, int src, int tag
 void Request::send(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
 {
   MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
-  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::getPid(),
+  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
                         comm->group()->actor(dst)->get_pid(), tag, comm, NON_PERSISTENT | SEND);
 
   request->start();
@@ -294,7 +307,7 @@ void Request::send(void *buf, int count, MPI_Datatype datatype, int dst, int tag
 void Request::ssend(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
 {
   MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
-  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::getPid(),
+  request = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, simgrid::s4u::this_actor::get_pid(),
                         comm->group()->actor(dst)->get_pid(), tag, comm, NON_PERSISTENT | SSEND | SEND);
 
   request->start();
@@ -308,7 +321,7 @@ void Request::sendrecv(void *sendbuf, int sendcount, MPI_Datatype sendtype,int d
 {
   MPI_Request requests[2];
   MPI_Status stats[2];
-  int myid = simgrid::s4u::this_actor::getPid();
+  int myid = simgrid::s4u::this_actor::get_pid();
   if ((comm->group()->actor(dst)->get_pid() == myid) && (comm->group()->actor(src)->get_pid() == myid)) {
     Datatype::copy(sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype);
     if (status != MPI_STATUS_IGNORE) {
@@ -343,9 +356,9 @@ void Request::start()
   if ((flags_ & RECV) != 0) {
     this->print_request("New recv");
 
-    simgrid::smpi::Process* process = smpi_process_remote(simgrid::s4u::Actor::byPid(dst_));
+    simgrid::smpi::Process* process = smpi_process_remote(simgrid::s4u::Actor::by_pid(dst_));
 
-    int async_small_thresh = xbt_cfg_get_int("smpi/async-small-thresh");
+    int async_small_thresh = simgrid::config::get_config<int>("smpi/async-small-thresh");
 
     xbt_mutex_t mut = process->mailboxes_mutex();
     if (async_small_thresh != 0 || (flags_ & RMA) != 0)
@@ -388,14 +401,14 @@ void Request::start()
     // we make a copy here, as the size is modified by simix, and we may reuse the request in another receive later
     real_size_=size_;
     action_   = simcall_comm_irecv(
-        process->process()->getImpl(), mailbox, buf_, &real_size_, &match_recv,
+        process->process()->get_impl(), mailbox, buf_, &real_size_, &match_recv,
         process->replaying() ? &smpi_comm_null_copy_buffer_callback : smpi_comm_copy_data_callback, this, -1.0);
     XBT_DEBUG("recv simcall posted");
 
     if (async_small_thresh != 0 || (flags_ & RMA) != 0 )
       xbt_mutex_release(mut);
   } else { /* the RECV flag was not set, so this is a send */
-    simgrid::smpi::Process* process = smpi_process_remote(simgrid::s4u::Actor::byPid(dst_));
+    simgrid::smpi::Process* process = smpi_process_remote(simgrid::s4u::Actor::by_pid(dst_));
     int rank = src_;
     if (TRACE_smpi_view_internals()) {
       TRACE_smpi_send(rank, rank, dst_, tag_, size_);
@@ -403,8 +416,9 @@ void Request::start()
     this->print_request("New send");
 
     void* buf = buf_;
-    if ((flags_ & SSEND) == 0 && ( (flags_ & RMA) != 0
-        || static_cast<int>(size_) < xbt_cfg_get_int("smpi/send-is-detached-thresh") ) ) {
+    if ((flags_ & SSEND) == 0 &&
+        ((flags_ & RMA) != 0 ||
+         static_cast<int>(size_) < simgrid::config::get_config<int>("smpi/send-is-detached-thresh"))) {
       void *oldbuf = nullptr;
       detached_ = 1;
       XBT_DEBUG("Send request %p is detached", this);
@@ -416,7 +430,7 @@ void Request::start()
               (static_cast<char*>(buf_) >= smpi_data_exe_start) &&
               (static_cast<char*>(buf_) < smpi_data_exe_start + smpi_data_exe_size)) {
             XBT_DEBUG("Privatization : We are sending from a zone inside global memory. Switch data segment ");
-            smpi_switch_data_segment(simgrid::s4u::Actor::byPid(src_));
+            smpi_switch_data_segment(simgrid::s4u::Actor::by_pid(src_));
           }
           buf = xbt_malloc(size_);
           memcpy(buf,oldbuf,size_);
@@ -439,7 +453,7 @@ void Request::start()
       XBT_DEBUG("sending size of %zu : sleep %f ", size_, sleeptime);
     }
 
-    int async_small_thresh = xbt_cfg_get_int("smpi/async-small-thresh");
+    int async_small_thresh = simgrid::config::get_config<int>("smpi/async-small-thresh");
 
     xbt_mutex_t mut=process->mailboxes_mutex();
 
@@ -476,7 +490,7 @@ void Request::start()
     // we make a copy here, as the size is modified by simix, and we may reuse the request in another receive later
     real_size_=size_;
     action_   = simcall_comm_isend(
-        simgrid::s4u::Actor::byPid(src_)->getImpl(), mailbox, size_, -1.0, buf, real_size_, &match_send,
+        simgrid::s4u::Actor::by_pid(src_)->get_impl(), mailbox, size_, -1.0, buf, real_size_, &match_send,
         &xbt_free_f, // how to free the userdata if a detached send fails
         not process->replaying() ? smpi_comm_copy_data_callback : &smpi_comm_null_copy_buffer_callback, this,
         // detach if msg size < eager/rdv switch limit
@@ -501,6 +515,14 @@ void Request::startall(int count, MPI_Request * requests)
   }
 }
 
+void Request::cancel()
+{
+  if(cancelled_!=-1)
+    cancelled_=1;
+  if (this->action_ != nullptr)
+    (boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(this->action_))->cancel();
+}
+
 int Request::test(MPI_Request * request, MPI_Status * status) {
   //assume that request is not MPI_REQUEST_NULL (filtered in PMPI_Test or testall before)
   // to avoid deadlocks if used as a break condition, such as
@@ -514,14 +536,19 @@ int Request::test(MPI_Request * request, MPI_Status * status) {
   Status::empty(status);
   int flag = 1;
   if (((*request)->flags_ & PREPARED) == 0) {
-    if ((*request)->action_ != nullptr)
-      flag = simcall_comm_test((*request)->action_);
+    if ((*request)->action_ != nullptr){
+      try{
+        flag = simcall_comm_test((*request)->action_);
+      }catch (xbt_ex& e) {
+        return 0;
+      }
+    }
     if (flag) {
       finish_wait(request,status);
       nsleeps=1;//reset the number of sleeps we will do next time
       if (*request != MPI_REQUEST_NULL && ((*request)->flags_ & PERSISTENT) == 0)
         *request = MPI_REQUEST_NULL;
-    } else if (xbt_cfg_get_boolean("smpi/grow-injected-times")){
+    } else if (simgrid::config::get_config<bool>("smpi/grow-injected-times")) {
       nsleeps++;
     }
   }
@@ -576,8 +603,12 @@ int Request::testany(int count, MPI_Request requests[], int *index, MPI_Status *
     static int nsleeps = 1;
     if(smpi_test_sleep > 0)
       simcall_process_sleep(nsleeps*smpi_test_sleep);
-
-    i = simcall_comm_testany(comms.data(), comms.size()); // The i-th element in comms matches!
+    try{
+      i = simcall_comm_testany(comms.data(), comms.size()); // The i-th element in comms matches!
+    }catch (xbt_ex& e) {
+      return 0;
+    }
+    
     if (i != -1) { // -1 is not MPI_UNDEFINED but a SIMIX return code. (nothing matches)
       *index = map[i];
       finish_wait(&requests[*index],status);
@@ -640,11 +671,11 @@ void Request::iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status*
   double maxrate = xbt_cfg_get_double("smpi/iprobe-cpu-usage");
   MPI_Request request = new Request(nullptr, 0, MPI_CHAR,
                                     source == MPI_ANY_SOURCE ? MPI_ANY_SOURCE : comm->group()->actor(source)->get_pid(),
-                                    simgrid::s4u::this_actor::getPid(), tag, comm, PERSISTENT | RECV);
+                                    simgrid::s4u::this_actor::get_pid(), tag, comm, PERSISTENT | RECV);
   if (smpi_iprobe_sleep > 0) {
     smx_activity_t iprobe_sleep = simcall_execution_start(
         "iprobe", /* flops to executek*/ nsleeps * smpi_iprobe_sleep * speed * maxrate, /* priority */ 1.0,
-        /* performance bound */ maxrate * speed, smpi_process()->process()->getImpl()->host);
+        /* performance bound */ maxrate * speed, smpi_process()->process()->get_impl()->host);
     simcall_execution_wait(iprobe_sleep);
   }
   // behave like a receive, but don't do it
@@ -652,10 +683,10 @@ void Request::iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status*
 
   request->print_request("New iprobe");
   // We have to test both mailboxes as we don't know if we will receive one one or another
-  if (xbt_cfg_get_int("smpi/async-small-thresh") > 0){
-      mailbox = smpi_process()->mailbox_small();
-      XBT_DEBUG("Trying to probe the perm recv mailbox");
-      request->action_ = simcall_comm_iprobe(mailbox, 0, &match_recv, static_cast<void*>(request));
+  if (simgrid::config::get_config<int>("smpi/async-small-thresh") > 0) {
+    mailbox = smpi_process()->mailbox_small();
+    XBT_DEBUG("Trying to probe the perm recv mailbox");
+    request->action_ = simcall_comm_iprobe(mailbox, 0, &match_recv, static_cast<void*>(request));
   }
 
   if (request->action_ == nullptr){
@@ -679,7 +710,7 @@ void Request::iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status*
   }
   else {
     *flag = 0;
-    if (xbt_cfg_get_boolean("smpi/grow-injected-times"))
+    if (simgrid::config::get_config<bool>("smpi/grow-injected-times"))
       nsleeps++;
   }
   unref(&request);
@@ -689,6 +720,12 @@ void Request::finish_wait(MPI_Request* request, MPI_Status * status)
 {
   MPI_Request req = *request;
   Status::empty(status);
+  
+  if (req->cancelled_==1){
+    if (status!=MPI_STATUS_IGNORE)
+      status->cancelled=1;
+    return;
+  }
 
   if (not((req->detached_ != 0) && ((req->flags_ & SEND) != 0)) && ((req->flags_ & PREPARED) == 0)) {
     if(status != MPI_STATUS_IGNORE) {
@@ -730,7 +767,7 @@ void Request::finish_wait(MPI_Request* request, MPI_Status * status)
   }
 
   if (TRACE_smpi_view_internals() && ((req->flags_ & RECV) != 0)){
-    int rank       = simgrid::s4u::this_actor::getPid();
+    int rank       = simgrid::s4u::this_actor::get_pid();
     int src_traced = (req->src_ == MPI_ANY_SOURCE ? req->real_src_ : req->src_);
     TRACE_smpi_recv(src_traced, rank,req->tag_);
   }
@@ -758,9 +795,15 @@ void Request::wait(MPI_Request * request, MPI_Status * status)
     return;
   }
 
-  if ((*request)->action_ != nullptr)
-    // this is not a detached send
-    simcall_comm_wait((*request)->action_, -1.0);
+  if ((*request)->action_ != nullptr){
+      try{
+        // this is not a detached send
+        simcall_comm_wait((*request)->action_, -1.0);
+      }catch (xbt_ex& e) {
+        XBT_VERB("Request cancelled");
+      }
+  }
+
 
   finish_wait(request,status);
   if (*request != MPI_REQUEST_NULL && (((*request)->flags_ & NON_PERSISTENT)!=0))
@@ -802,7 +845,14 @@ int Request::waitany(int count, MPI_Request requests[], MPI_Status * status)
     }
     if (size > 0) {
       XBT_DEBUG("Enter waitany for %lu comms", xbt_dynar_length(&comms));
-      int i = simcall_comm_waitany(&comms, -1);
+      int i=MPI_UNDEFINED;
+      try{
+        // this is not a detached send
+        i = simcall_comm_waitany(&comms, -1);
+      }catch (xbt_ex& e) {
+      XBT_INFO("request %d cancelled ",i);
+        return i;
+      }
 
       // not MPI_UNDEFINED, as this is a simix return code
       if (i != -1) {
@@ -856,6 +906,7 @@ int Request::waitall(int count, MPI_Request requests[], MPI_Status status[])
       index = c;
     } else {
       index = waitany(count, (MPI_Request*)requests, pstat);
+      
       if (index == MPI_UNDEFINED)
         break;