X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/299c52751d8bf7ed4f9fc40f668025a5cb425901..35f18859ce88a96b7e07de4d7d70867424feb8df:/src/smpi/mpi/smpi_request.cpp

diff --git a/src/smpi/mpi/smpi_request.cpp b/src/smpi/mpi/smpi_request.cpp
index 451a57c489..05cdc18c1e 100644
--- a/src/smpi/mpi/smpi_request.cpp
+++ b/src/smpi/mpi/smpi_request.cpp
@@ -36,14 +36,14 @@ extern void (*smpi_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*
 namespace simgrid{
 namespace smpi{
 
-Request::Request(void* buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm, unsigned flags)
-    : buf_(buf), old_type_(datatype), src_(src), dst_(dst), tag_(tag), comm_(comm), flags_(flags)
+Request::Request(const void* buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm, unsigned flags, MPI_Op op)
+    : buf_(const_cast<void*>(buf)), old_type_(datatype), src_(src), dst_(dst), tag_(tag), comm_(comm), flags_(flags), op_(op)
 {
   void *old_buf = nullptr;
 // FIXME Handle the case of a partial shared malloc.
   if ((((flags & MPI_REQ_RECV) != 0) && ((flags & MPI_REQ_ACCUMULATE) != 0)) || (datatype->flags() & DT_FLAG_DERIVED)) {
     // This part handles the problem of non-contiguous memory
-    old_buf = buf;
+    old_buf = const_cast<void*>(buf);
     if (count==0){
       buf_ = nullptr;
     }else {
@@ -58,18 +58,19 @@ Request::Request(void* buf, int count, MPI_Datatype datatype, int src, int dst,
   size_ = datatype->size() * count;
   datatype->ref();
   comm_->ref();
+  if(op != MPI_REPLACE && op != MPI_OP_NULL)
+    op_->ref();
   action_          = nullptr;
-  detached_        = 0;
+  detached_        = false;
   detached_sender_ = nullptr;
   real_src_        = 0;
-  truncated_       = 0;
+  truncated_       = false;
   real_size_       = 0;
   real_tag_        = 0;
   if (flags & MPI_REQ_PERSISTENT)
     refcount_ = 1;
   else
     refcount_ = 0;
-  op_   = MPI_REPLACE;
   cancelled_ = 0;
   generalized_funcs=nullptr;
   nbc_requests_=nullptr;
@@ -91,10 +92,14 @@ void Request::unref(MPI_Request* request)
     if((*request)->refcount_==0){
       if ((*request)->flags_ & MPI_REQ_GENERALIZED){
         ((*request)->generalized_funcs)->free_fn(((*request)->generalized_funcs)->extra_state);
+        delete (*request)->generalized_funcs;
       }else{
         Comm::unref((*request)->comm_);
         Datatype::unref((*request)->old_type_);
       }
+      if ((*request)->op_!=MPI_REPLACE && (*request)->op_!=MPI_OP_NULL)
+        Op::unref(&(*request)->op_);
+
       (*request)->print_request("Destroying");
       delete *request;
       *request = MPI_REQUEST_NULL;
@@ -110,11 +115,12 @@ int Request::match_recv(void* a, void* b, simgrid::kernel::activity::CommImpl*)
 {
   MPI_Request ref = static_cast<MPI_Request>(a);
   MPI_Request req = static_cast<MPI_Request>(b);
-  XBT_DEBUG("Trying to match a recv of src %d against %d, tag %d against %d",ref->src_,req->src_, ref->tag_, req->tag_);
+  XBT_DEBUG("Trying to match a recv of src %d against %d, tag %d against %d, id %d against %d",ref->src_,req->src_, ref->tag_, req->tag_,ref->comm_->id(),req->comm_->id());
 
   xbt_assert(ref, "Cannot match recv against null reference");
   xbt_assert(req, "Cannot match recv against null request");
-  if((ref->src_ == MPI_ANY_SOURCE || req->src_ == ref->src_)
+  if((ref->comm_->id()==MPI_UNDEFINED || req->comm_->id() == MPI_UNDEFINED || (ref->comm_->id()==req->comm_->id()))
+    && ((ref->src_ == MPI_ANY_SOURCE  && (ref->comm_->group()->rank(req->src_) != MPI_UNDEFINED)) || req->src_ == ref->src_)
     && ((ref->tag_ == MPI_ANY_TAG && req->tag_ >=0) || req->tag_ == ref->tag_)){
     //we match, we can transfer some values
     if(ref->src_ == MPI_ANY_SOURCE)
@@ -122,8 +128,8 @@ int Request::match_recv(void* a, void* b, simgrid::kernel::activity::CommImpl*)
     if(ref->tag_ == MPI_ANY_TAG)
       ref->real_tag_ = req->tag_;
     if(ref->real_size_ < req->real_size_)
-      ref->truncated_ = 1;
-    if(req->detached_==1)
+      ref->truncated_ = true;
+    if (req->detached_)
       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
@@ -136,19 +142,20 @@ int Request::match_send(void* a, void* b, simgrid::kernel::activity::CommImpl*)
 {
   MPI_Request ref = static_cast<MPI_Request>(a);
   MPI_Request req = static_cast<MPI_Request>(b);
-  XBT_DEBUG("Trying to match a send of src %d against %d, tag %d against %d",ref->src_,req->src_, ref->tag_, req->tag_);
+  XBT_DEBUG("Trying to match a send of src %d against %d, tag %d against %d, id %d against %d",ref->src_,req->src_, ref->tag_, req->tag_,ref->comm_->id(),req->comm_->id());
   xbt_assert(ref, "Cannot match send against null reference");
   xbt_assert(req, "Cannot match send against null request");
 
-  if((req->src_ == MPI_ANY_SOURCE || req->src_ == ref->src_)
+  if((ref->comm_->id()==MPI_UNDEFINED || req->comm_->id() == MPI_UNDEFINED || (ref->comm_->id()==req->comm_->id()))
+      && ((req->src_ == MPI_ANY_SOURCE  && (req->comm_->group()->rank(ref->src_) != MPI_UNDEFINED)) || req->src_ == ref->src_)
       && ((req->tag_ == MPI_ANY_TAG && ref->tag_ >=0)|| req->tag_ == ref->tag_)){
     if(req->src_ == MPI_ANY_SOURCE)
       req->real_src_ = ref->src_;
     if(req->tag_ == MPI_ANY_TAG)
       req->real_tag_ = ref->tag_;
     if(req->real_size_ < ref->real_size_)
-      req->truncated_ = 1;
-    if(ref->detached_==1)
+      req->truncated_ = true;
+    if (ref->detached_)
       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
@@ -166,7 +173,7 @@ void Request::print_request(const char *message)
 
 
 /* factories, to hide the internal flags from the caller */
-MPI_Request Request::send_init(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
+MPI_Request Request::send_init(const 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::get_pid(),
@@ -174,14 +181,14 @@ MPI_Request Request::send_init(void *buf, int count, MPI_Datatype datatype, int
                      MPI_REQ_PERSISTENT | MPI_REQ_SEND | MPI_REQ_PREPARED);
 }
 
-MPI_Request Request::ssend_init(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
+MPI_Request Request::ssend_init(const 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::get_pid(),
                      comm->group()->actor(dst)->get_pid(), tag, comm,
                      MPI_REQ_PERSISTENT | MPI_REQ_SSEND | MPI_REQ_SEND | MPI_REQ_PREPARED);
 }
 
-MPI_Request Request::isend_init(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
+MPI_Request Request::isend_init(const 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::get_pid(),
                      comm->group()->actor(dst)->get_pid(), tag, comm,
@@ -189,7 +196,7 @@ MPI_Request Request::isend_init(void *buf, int count, MPI_Datatype datatype, int
 }
 
 
-MPI_Request Request::rma_send_init(void *buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm,
+MPI_Request Request::rma_send_init(const void *buf, int count, MPI_Datatype datatype, int src, int dst, int tag, MPI_Comm comm,
                                MPI_Op op)
 {
   MPI_Request request = nullptr; /* MC needs the comm to be set to nullptr during the call */
@@ -201,8 +208,7 @@ MPI_Request Request::rma_send_init(void *buf, int count, MPI_Datatype datatype,
     request      = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, comm->group()->actor(src)->get_pid(),
                           comm->group()->actor(dst)->get_pid(), tag, comm,
                           MPI_REQ_RMA | MPI_REQ_NON_PERSISTENT | MPI_REQ_ISEND | MPI_REQ_SEND | MPI_REQ_PREPARED |
-                              MPI_REQ_ACCUMULATE);
-    request->op_ = op;
+                              MPI_REQ_ACCUMULATE, op);
   }
   return request;
 }
@@ -226,8 +232,7 @@ MPI_Request Request::rma_recv_init(void *buf, int count, MPI_Datatype datatype,
   }else{
     request      = new Request(buf == MPI_BOTTOM ? nullptr : buf, count, datatype, comm->group()->actor(src)->get_pid(),
                           comm->group()->actor(dst)->get_pid(), tag, comm,
-                          MPI_REQ_RMA | MPI_REQ_NON_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PREPARED | MPI_REQ_ACCUMULATE);
-    request->op_ = op;
+                          MPI_REQ_RMA | MPI_REQ_NON_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PREPARED | MPI_REQ_ACCUMULATE, op);
   }
   return request;
 }
@@ -240,7 +245,7 @@ MPI_Request Request::irecv_init(void *buf, int count, MPI_Datatype datatype, int
                      MPI_REQ_PERSISTENT | MPI_REQ_RECV | MPI_REQ_PREPARED);
 }
 
-MPI_Request Request::isend(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
+MPI_Request Request::isend(const 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::get_pid(),
@@ -250,7 +255,7 @@ MPI_Request Request::isend(void *buf, int count, MPI_Datatype datatype, int dst,
   return request;
 }
 
-MPI_Request Request::issend(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
+MPI_Request Request::issend(const 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::get_pid(),
@@ -279,7 +284,7 @@ void Request::recv(void *buf, int count, MPI_Datatype datatype, int src, int tag
   request = nullptr;
 }
 
-void Request::send(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
+void Request::send(const 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::get_pid(),
@@ -290,7 +295,7 @@ void Request::send(void *buf, int count, MPI_Datatype datatype, int dst, int tag
   request = nullptr;
 }
 
-void Request::ssend(void *buf, int count, MPI_Datatype datatype, int dst, int tag, MPI_Comm comm)
+void Request::ssend(const 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::get_pid(),
@@ -302,7 +307,7 @@ void Request::ssend(void *buf, int count, MPI_Datatype datatype, int dst, int ta
   request = nullptr;
 }
 
-void Request::sendrecv(void *sendbuf, int sendcount, MPI_Datatype sendtype,int dst, int sendtag,
+void Request::sendrecv(const void *sendbuf, int sendcount, MPI_Datatype sendtype,int dst, int sendtag,
                        void *recvbuf, int recvcount, MPI_Datatype recvtype, int src, int recvtag,
                        MPI_Comm comm, MPI_Status * status)
 {
@@ -407,7 +412,7 @@ void Request::start()
         ((flags_ & MPI_REQ_RMA) != 0 ||
          static_cast<int>(size_) < simgrid::config::get_value<int>("smpi/send-is-detached-thresh"))) {
       void *oldbuf = nullptr;
-      detached_ = 1;
+      detached_    = true;
       XBT_DEBUG("Send request %p is detached", this);
       this->ref();
       if (not(old_type_->flags() & DT_FLAG_DERIVED)) {
@@ -428,7 +433,7 @@ void Request::start()
 
     //if we are giving back the control to the user without waiting for completion, we have to inject timings
     double sleeptime = 0.0;
-    if (detached_ != 0 || ((flags_ & (MPI_REQ_ISEND | MPI_REQ_SSEND)) != 0)) { // issend should be treated as isend
+    if (detached_ || ((flags_ & (MPI_REQ_ISEND | MPI_REQ_SSEND)) != 0)) { // issend should be treated as isend
       // isend and send timings may be different
       sleeptime = ((flags_ & MPI_REQ_ISEND) != 0)
                       ? simgrid::s4u::Actor::self()->get_host()->extension<simgrid::smpi::Host>()->oisend(size_)
@@ -488,8 +493,8 @@ void Request::start()
 
     /* FIXME: detached sends are not traceable (action_ == nullptr) */
     if (action_ != nullptr) {
-      std::string category = smpi_process()->get_tracing_category();
-      simgrid::simix::simcall([this, category] { this->action_->set_category(category); });
+      boost::static_pointer_cast<kernel::activity::CommImpl>(action_)->set_tracing_category(
+          smpi_process()->get_tracing_category());
     }
 
     if (async_small_thresh != 0 || ((flags_ & MPI_REQ_RMA) != 0))
@@ -539,14 +544,13 @@ int Request::test(MPI_Request * request, MPI_Status * status, int* flag) {
   if(smpi_test_sleep > 0)
     simcall_process_sleep(nsleeps*smpi_test_sleep);
 
-  MPI_Status* mystatus;
   Status::empty(status);
   *flag = 1;
   if (((*request)->flags_ & MPI_REQ_PREPARED) == 0) {
     if ((*request)->action_ != nullptr){
       try{
         *flag = simcall_comm_test((*request)->action_);
-      }catch (xbt_ex& e) {
+      } catch (const Exception&) {
         *flag = 0;
         return ret;
       }
@@ -558,6 +562,7 @@ int Request::test(MPI_Request * request, MPI_Status * status, int* flag) {
     if (*flag) {
       finish_wait(request,status);
       if (*request != MPI_REQUEST_NULL && ((*request)->flags_ & MPI_REQ_GENERALIZED)){
+        MPI_Status* mystatus;
         if(status==MPI_STATUS_IGNORE){
           mystatus=new MPI_Status();
           Status::empty(mystatus);
@@ -620,7 +625,6 @@ int Request::testany(int count, MPI_Request requests[], int *index, int* flag, M
   int i;
   *flag = 0;
   int ret = MPI_SUCCESS;
-  MPI_Status* mystatus;
   *index = MPI_UNDEFINED;
 
   std::vector<int> map; /** Maps all matching comms back to their location in requests **/
@@ -637,7 +641,8 @@ int Request::testany(int count, MPI_Request requests[], int *index, int* flag, M
       simcall_process_sleep(nsleeps*smpi_test_sleep);
     try{
       i = simcall_comm_testany(comms.data(), comms.size()); // The i-th element in comms matches!
-    }catch (xbt_ex& e) {
+    } catch (const Exception&) {
+      XBT_DEBUG("Exception in testany");
       return 0;
     }
     
@@ -650,6 +655,7 @@ int Request::testany(int count, MPI_Request requests[], int *index, int* flag, M
       } else {
         finish_wait(&requests[*index],status);
       if (requests[*index] != MPI_REQUEST_NULL && (requests[*index]->flags_ & MPI_REQ_GENERALIZED)){
+        MPI_Status* mystatus;
         if(status==MPI_STATUS_IGNORE){
           mystatus=new MPI_Status();
           Status::empty(mystatus);
@@ -663,6 +669,7 @@ int Request::testany(int count, MPI_Request requests[], int *index, int* flag, M
 
         if (requests[*index] != MPI_REQUEST_NULL && (requests[*index]->flags_ & MPI_REQ_NON_PERSISTENT)) 
           requests[*index] = MPI_REQUEST_NULL;
+        XBT_DEBUG("Testany - returning with index %d", *index);
         *flag=1;
       }
       nsleeps = 1;
@@ -670,8 +677,10 @@ int Request::testany(int count, MPI_Request requests[], int *index, int* flag, M
       nsleeps++;
     }
   } else {
+      XBT_DEBUG("Testany on inactive handles, returning flag=1 but empty status");
       //all requests are null or inactive, return true
       *flag = 1;
+      *index = MPI_UNDEFINED;
       Status::empty(status);
   }
 
@@ -682,7 +691,8 @@ int Request::testall(int count, MPI_Request requests[], int* outflag, MPI_Status
 {
   MPI_Status stat;
   MPI_Status *pstat = status == MPI_STATUSES_IGNORE ? MPI_STATUS_IGNORE : &stat;
-  int flag, error=0;
+  int flag;
+  int error = 0;
   int ret=MPI_SUCCESS;
   *outflag = 1;
   for(int i=0; i<count; i++){
@@ -779,6 +789,7 @@ void Request::iprobe(int source, int tag, MPI_Comm comm, int* flag, MPI_Status*
       nsleeps++;
   }
   unref(&request);
+  xbt_assert(request == MPI_REQUEST_NULL);
 }
 
 void Request::finish_wait(MPI_Request* request, MPI_Status * status)
@@ -795,45 +806,43 @@ void Request::finish_wait(MPI_Request* request, MPI_Status * status)
     return;
   }
 
-  if (not((req->detached_ != 0) && ((req->flags_ & MPI_REQ_SEND) != 0)) 
-  && ((req->flags_ & MPI_REQ_PREPARED) == 0)
-  && ((req->flags_ & MPI_REQ_GENERALIZED) == 0)) {
+  if ((req->flags_ & (MPI_REQ_PREPARED | MPI_REQ_GENERALIZED | MPI_REQ_FINISHED)) == 0) {
     if(status != MPI_STATUS_IGNORE) {
       int src = req->src_ == MPI_ANY_SOURCE ? req->real_src_ : req->src_;
       status->MPI_SOURCE = req->comm_->group()->rank(src);
       status->MPI_TAG = req->tag_ == MPI_ANY_TAG ? req->real_tag_ : req->tag_;
-      status->MPI_ERROR = req->truncated_ != 0 ? MPI_ERR_TRUNCATE : MPI_SUCCESS;
+      status->MPI_ERROR  = req->truncated_ ? MPI_ERR_TRUNCATE : MPI_SUCCESS;
       // this handles the case were size in receive differs from size in send
       status->count = req->real_size_;
-//      int flag;
-//      Request::get_status(req,&flag,status);
     }
+    //detached send will be finished at the other end
+    if (not(req->detached_ && ((req->flags_ & MPI_REQ_SEND) != 0))) {
+      req->print_request("Finishing");
+      MPI_Datatype datatype = req->old_type_;
+
+      // FIXME Handle the case of a partial shared malloc.
+      if (((req->flags_ & MPI_REQ_ACCUMULATE) != 0) ||
+          (datatype->flags() & DT_FLAG_DERIVED)) { // && (not smpi_is_shared(req->old_buf_))){
+
+        if (not smpi_process()->replaying() && smpi_privatize_global_variables != SmpiPrivStrategies::NONE &&
+            static_cast<char*>(req->old_buf_) >= smpi_data_exe_start &&
+            static_cast<char*>(req->old_buf_) < smpi_data_exe_start + smpi_data_exe_size) {
+          XBT_VERB("Privatization : We are unserializing to a zone in global memory  Switch data segment ");
+          smpi_switch_data_segment(simgrid::s4u::Actor::self());
+        }
 
-    req->print_request("Finishing");
-    MPI_Datatype datatype = req->old_type_;
-
-// FIXME Handle the case of a partial shared malloc.
-    if (((req->flags_ & MPI_REQ_ACCUMULATE) != 0) ||
-        (datatype->flags() & DT_FLAG_DERIVED)) { // && (not smpi_is_shared(req->old_buf_))){
-
-      if (not smpi_process()->replaying() && smpi_privatize_global_variables != SmpiPrivStrategies::NONE &&
-          static_cast<char*>(req->old_buf_) >= smpi_data_exe_start &&
-          static_cast<char*>(req->old_buf_) < smpi_data_exe_start + smpi_data_exe_size) {
-        XBT_VERB("Privatization : We are unserializing to a zone in global memory  Switch data segment ");
-        smpi_switch_data_segment(simgrid::s4u::Actor::self());
-      }
-
-      if(datatype->flags() & DT_FLAG_DERIVED){
-        // This part handles the problem of non-contignous memory the unserialization at the reception
-        if ((req->flags_ & MPI_REQ_RECV) && datatype->size() != 0)
-          datatype->unserialize(req->buf_, req->old_buf_, req->real_size_/datatype->size() , req->op_);
-        xbt_free(req->buf_);
-      } else if (req->flags_ & MPI_REQ_RECV) { // apply op on contiguous buffer for accumulate
-        if (datatype->size() != 0) {
-          int n = req->real_size_ / datatype->size();
-          req->op_->apply(req->buf_, req->old_buf_, &n, datatype);
+        if(datatype->flags() & DT_FLAG_DERIVED){
+          // This part handles the problem of non-contignous memory the unserialization at the reception
+          if ((req->flags_ & MPI_REQ_RECV) && datatype->size() != 0)
+            datatype->unserialize(req->buf_, req->old_buf_, req->real_size_/datatype->size() , req->op_);
+          xbt_free(req->buf_);
+        } else if (req->flags_ & MPI_REQ_RECV) { // apply op on contiguous buffer for accumulate
+          if (datatype->size() != 0) {
+            int n = req->real_size_ / datatype->size();
+            req->op_->apply(req->buf_, req->old_buf_, &n, datatype);
+          }
+          xbt_free(req->buf_);
         }
-        xbt_free(req->buf_);
       }
     }
   }
@@ -867,12 +876,25 @@ int Request::wait(MPI_Request * request, MPI_Status * status)
   if ((*request)->nbc_requests_size_>0){
     ret = waitall((*request)->nbc_requests_size_, (*request)->nbc_requests_, MPI_STATUSES_IGNORE);
     for (int i = 0; i < (*request)->nbc_requests_size_; i++) {
+      if((*request)->buf_!=nullptr && (*request)->nbc_requests_[i]!=MPI_REQUEST_NULL){//reduce case
+        void * buf=(*request)->nbc_requests_[i]->buf_;
+        if((*request)->old_type_->flags() & DT_FLAG_DERIVED)
+          buf=(*request)->nbc_requests_[i]->old_buf_;
+        if((*request)->nbc_requests_[i]->flags_ & MPI_REQ_RECV ){
+          if((*request)->op_!=MPI_OP_NULL){
+            int count=(*request)->size_/ (*request)->old_type_->size();
+            (*request)->op_->apply(buf, (*request)->buf_, &count, (*request)->old_type_);
+          }
+          smpi_free_tmp_buffer(static_cast<unsigned char*>(buf));
+        }
+      }
       if((*request)->nbc_requests_[i]!=MPI_REQUEST_NULL)
         Request::unref(&((*request)->nbc_requests_[i]));
     }
     delete[] (*request)->nbc_requests_;
     (*request)->nbc_requests_size_=0;
     unref(request);
+    (*request)=MPI_REQUEST_NULL;
     return ret;
   }
 
@@ -886,7 +908,7 @@ int Request::wait(MPI_Request * request, MPI_Status * status)
       try{
         // this is not a detached send
         simcall_comm_wait((*request)->action_, -1.0);
-      }catch (xbt_ex& e) {
+      } catch (const Exception&) {
         XBT_VERB("Request cancelled");
       }
   }
@@ -949,8 +971,8 @@ int Request::waitany(int count, MPI_Request requests[], MPI_Status * status)
       try{
         // this is not a detached send
         i = simcall_comm_waitany(comms.data(), comms.size(), -1);
-      }catch (xbt_ex& e) {
-      XBT_INFO("request %d cancelled ",i);
+      } catch (const Exception&) {
+        XBT_INFO("request %d cancelled ", i);
         return i;
       }
 
@@ -1047,7 +1069,9 @@ int Request::waitsome(int incount, MPI_Request requests[], int *indices, MPI_Sta
   indices[count] = index;
   count++;
   for (int i = 0; i < incount; i++) {
-    if((requests[i] != MPI_REQUEST_NULL)) {
+    if (i==index)
+      continue;
+    if (requests[i] != MPI_REQUEST_NULL) {
       test(&requests[i], pstat,&flag);
       if (flag==1){
         indices[count] = i;
@@ -1095,7 +1119,7 @@ int Request::get_status(MPI_Request req, int* flag, MPI_Status * status){
 
   if(req != MPI_REQUEST_NULL && req->action_ != nullptr) {
     req->iprobe(req->src_, req->tag_, req->comm_, flag, status);
-    if(flag)
+    if(*flag)
       return MPI_SUCCESS;
   }
   if (req != MPI_REQUEST_NULL && 
@@ -1106,7 +1130,8 @@ int Request::get_status(MPI_Request req, int* flag, MPI_Status * status){
   }
 
   *flag=1;
-  if(status != MPI_STATUS_IGNORE) {
+  if(req != MPI_REQUEST_NULL &&
+     status != MPI_STATUS_IGNORE) {
     int src = req->src_ == MPI_ANY_SOURCE ? req->real_src_ : req->src_;
     status->MPI_SOURCE = req->comm_->group()->rank(src);
     status->MPI_TAG = req->tag_ == MPI_ANY_TAG ? req->real_tag_ : req->tag_;
@@ -1122,7 +1147,7 @@ int Request::grequest_start( MPI_Grequest_query_function *query_fn, MPI_Grequest
   (*request)->flags_ |= MPI_REQ_GENERALIZED;
   (*request)->flags_ |= MPI_REQ_PERSISTENT;
   (*request)->refcount_ = 1;
-  ((*request)->generalized_funcs)=xbt_new0(s_smpi_mpi_generalized_request_funcs_t ,1);
+  ((*request)->generalized_funcs) = new s_smpi_mpi_generalized_request_funcs_t;
   ((*request)->generalized_funcs)->query_fn=query_fn;
   ((*request)->generalized_funcs)->free_fn=free_fn;
   ((*request)->generalized_funcs)->cancel_fn=cancel_fn;
@@ -1143,8 +1168,13 @@ int Request::grequest_complete( MPI_Request request){
 }
 
 void Request::set_nbc_requests(MPI_Request* reqs, int size){
-  nbc_requests_=reqs;
-  nbc_requests_size_=size;
+  nbc_requests_size_ = size;
+  if (size > 0) {
+    nbc_requests_ = reqs;
+  } else {
+    delete[] reqs;
+    nbc_requests_ = nullptr;
+  }
 }
 
 int Request::get_nbc_requests_size(){