Replace all the lookup tables in smpi_f77.cpp by a single one.

[simgrid.git] / src / smpi / smpi_request.cpp

diff --git a/src/smpi/smpi_request.cpp b/src/smpi/smpi_request.cpp
index d6e0fcc..978e335 100644 (file)
--- a/src/smpi/smpi_request.cpp
+++ b/src/smpi/smpi_request.cpp
@@ -109,25 +109,25 @@ static double smpi_or(size_t size)
 
 namespace simgrid{
 namespace smpi{
+
 Request::Request(){}
 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) 
 {
   void *old_buf = nullptr;
-//  s_smpi_subtype_t *subtype = static_cast<s_smpi_subtype_t*>(datatype->substruct);
-
-//  if((((flags & RECV) != 0) && ((flags & ACCUMULATE) !=0)) || (datatype->sizeof_substruct != 0)){
-//    // This part handles the problem of non-contiguous memory
-//    old_buf = buf;
-//    buf_ = count==0 ? nullptr : xbt_malloc(count*datatype->size());
-//    if ((datatype->sizeof_substruct != 0) && ((flags & SEND) != 0)) {
-//      subtype->serialize(old_buf, buf_, count, datatype->substruct);
-//    }
-//  }
+
+  if((((flags & RECV) != 0) && ((flags & ACCUMULATE) !=0)) || (datatype->flags() & DT_FLAG_DERIVED)){
+    // This part handles the problem of non-contiguous memory
+    old_buf = buf;
+    buf_ = count==0 ? nullptr : xbt_malloc(count*datatype->size());
+    if ((datatype->flags() & DT_FLAG_DERIVED) && ((flags & SEND) != 0)) {
+      datatype->serialize(old_buf, buf_, count);
+    }
+  }
   // This part handles the problem of non-contiguous memory (for the unserialisation at the reception)
   old_buf_  = old_buf;
   size_ = datatype->size() * count;
   datatype->use();
-  comm_->use();
+  comm_->ref();
   action_          = nullptr;
   detached_        = 0;
   detached_sender_ = nullptr;
@@ -173,6 +173,14 @@ int Request::detached(){
   return detached_;
 }
 
+size_t Request::size(){
+  return size_;
+}
+
+size_t Request::real_size(){
+  return real_size_;
+}
+
 
 void Request::unuse(MPI_Request* request)
 {
@@ -182,7 +190,7 @@ void Request::unuse(MPI_Request* request)
 
     if((*request)->refcount_==0){
         (*request)->old_type_->unuse();
-        (*request)->comm_->unuse();
+        Comm::unref((*request)->comm_);
         (*request)->print_request("Destroying");
         delete *request;
         *request = MPI_REQUEST_NULL;
@@ -482,24 +490,24 @@ void Request::start()
     void* buf = buf_;
     if ((flags_ & SSEND) == 0 && ( (flags_ & RMA) != 0
         || static_cast<int>(size_) < xbt_cfg_get_int("smpi/send-is-detached-thresh") ) ) {
-//      void *oldbuf = nullptr;
+      void *oldbuf = nullptr;
       detached_ = 1;
       XBT_DEBUG("Send request %p is detached", this);
       refcount_++;
-//      if(old_type_->sizeof_substruct == 0){
-//        oldbuf = buf_;
-//        if (!smpi_process_get_replaying() && oldbuf != nullptr && size_!=0){
-//          if((smpi_privatize_global_variables != 0)
-//            && (static_cast<char*>(buf_) >= smpi_start_data_exe)
-//            && (static_cast<char*>(buf_) < smpi_start_data_exe + smpi_size_data_exe )){
-//            XBT_DEBUG("Privatization : We are sending from a zone inside global memory. Switch data segment ");
-//            smpi_switch_data_segment(src_);
-//          }
-//          buf = xbt_malloc(size_);
-//          memcpy(buf,oldbuf,size_);
-//          XBT_DEBUG("buf %p copied into %p",oldbuf,buf);
-//        }
-//      }
+      if(!(old_type_->flags() & DT_FLAG_DERIVED)){
+        oldbuf = buf_;
+        if (!smpi_process_get_replaying() && oldbuf != nullptr && size_!=0){
+          if((smpi_privatize_global_variables != 0)
+            && (static_cast<char*>(buf_) >= smpi_start_data_exe)
+            && (static_cast<char*>(buf_) < smpi_start_data_exe + smpi_size_data_exe )){
+            XBT_DEBUG("Privatization : We are sending from a zone inside global memory. Switch data segment ");
+            smpi_switch_data_segment(src_);
+          }
+          buf = xbt_malloc(size_);
+          memcpy(buf,oldbuf,size_);
+          XBT_DEBUG("buf %p copied into %p",oldbuf,buf);
+        }
+      }
     }
 
     //if we are giving back the control to the user without waiting for completion, we have to inject timings
@@ -792,7 +800,7 @@ void Request::finish_wait(MPI_Request* request, MPI_Status * status)
     req->print_request("Finishing");
     MPI_Datatype datatype = req->old_type_;
 
-    if(((req->flags_ & ACCUMULATE) != 0)/* || (datatype->sizeof_substruct != 0)*/){
+    if(((req->flags_ & ACCUMULATE) != 0) || (datatype->flags() & DT_FLAG_DERIVED)){
       if (!smpi_process_get_replaying()){
         if( smpi_privatize_global_variables != 0 && (static_cast<char*>(req->old_buf_) >= smpi_start_data_exe)
             && ((char*)req->old_buf_ < smpi_start_data_exe + smpi_size_data_exe )){
@@ -801,15 +809,12 @@ void Request::finish_wait(MPI_Request* request, MPI_Status * status)
         }
       }
 
-//      if(datatype->sizeof_substruct != 0){
-//        // This part handles the problem of non-contignous memory the unserialization at the reception
-//        s_smpi_subtype_t *subtype = static_cast<s_smpi_subtype_t*>(datatype->substruct);
-//        if(req->flags_ & RECV)
-//          subtype->unserialize(req->buf_, req->old_buf_, req->real_size_/datatype->size() ,
-//                               datatype->substruct, req->op_);
-//        xbt_free(req->buf_);
-//      }else 
-    if(req->flags_ & RECV){//apply op on contiguous buffer for accumulate
+      if(datatype->flags() & DT_FLAG_DERIVED){
+        // This part handles the problem of non-contignous memory the unserialization at the reception
+        if((req->flags_ & 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_ & RECV){//apply op on contiguous buffer for accumulate
           int n =req->real_size_/datatype->size();
           req->op_->apply(req->buf_, req->old_buf_, &n, datatype);
           xbt_free(req->buf_);
@@ -996,6 +1001,30 @@ int Request::waitsome(int incount, MPI_Request requests[], int *indices, MPI_Sta
   return count;
 }
 
+MPI_Request Request::f2c(int id) {
+  char key[KEY_SIZE];
+  if(id==MPI_FORTRAN_REQUEST_NULL)
+    return static_cast<MPI_Request>(MPI_REQUEST_NULL);
+  return static_cast<MPI_Request>(xbt_dict_get(Request::f2c_lookup_, get_key_id(key, id)));
+}
+
+int Request::add_f() {
+  if(Request::f2c_lookup_==nullptr){
+    Request::f2c_lookup_=xbt_dict_new_homogeneous(nullptr);
+  }
+  char key[KEY_SIZE];
+  xbt_dict_set(Request::f2c_lookup_, get_key_id(key, Request::f2c_id_), this, nullptr);
+  Request::f2c_id_++;
+  return Request::f2c_id_-1;
+}
+
+void Request::free_f(int id) {
+  char key[KEY_SIZE];
+  if(id!=MPI_FORTRAN_REQUEST_NULL)
+    xbt_dict_remove(Request::f2c_lookup_, get_key_id(key, id));
+}
+
+
 
 }
 }