From: Augustin Degomme <adegomme@users.noreply.github.com>
Date: Sun, 20 Mar 2022 23:15:01 +0000 (+0100)
Subject: Have MPI collectives display their root and MPI_Op to handle mismatches lazily
X-Git-Tag: v3.31~17
X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/commitdiff_plain/cdbe41fd5a9d7501206af7e3fe0080b661e62e76

Have MPI collectives display their root and MPI_Op to handle mismatches lazily
---

diff --git a/src/smpi/bindings/smpi_pmpi_coll.cpp b/src/smpi/bindings/smpi_pmpi_coll.cpp
index 4549c06970..7df4731f6d 100644
--- a/src/smpi/bindings/smpi_pmpi_coll.cpp
+++ b/src/smpi/bindings/smpi_pmpi_coll.cpp
@@ -67,7 +67,9 @@ int PMPI_Ibcast(void* buf, int count, MPI_Datatype datatype, int root, MPI_Comm
   CHECK_BUFFER(1, buf, count, datatype)
   CHECK_ROOT(4)
   CHECK_REQUEST(6)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Bcast" : "PMPI_Ibcast")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Bcast" : "PMPI_Ibcast");
+  name += " with root " + std::to_string(root);
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
   aid_t pid = simgrid::s4u::this_actor::get_pid();
@@ -116,7 +118,9 @@ int PMPI_Igather(const void* sendbuf, int sendcount, MPI_Datatype sendtype, void
   }
   CHECK_ROOT(7)
   CHECK_REQUEST(9)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Gather" : "PMPI_Igather")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Gather" : "PMPI_Igather");
+  name += " with root " + std::to_string(root);
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const void* real_sendbuf   = sendbuf;
   int real_sendcount         = sendcount;
@@ -177,7 +181,9 @@ int PMPI_Igatherv(const void* sendbuf, int sendcount, MPI_Datatype sendtype, voi
   }
   CHECK_ROOT(8)
   CHECK_REQUEST(10)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Gatherv" : "PMPI_Igatherv")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Gatherv" : "PMPI_Igatherv");
+  name += " with root " + std::to_string(root);
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   if (rank == root){
     for (int i = 0; i < comm->size(); i++) {
@@ -356,7 +362,9 @@ int PMPI_Iscatter(const void* sendbuf, int sendcount, MPI_Datatype sendtype, voi
   }
   CHECK_ROOT(8)
   CHECK_REQUEST(9)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Scatter" : "PMPI_Iscatter")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Scatter" : "PMPI_Iscatter");
+  name += " with root " + std::to_string(root);
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   if (recvbuf == MPI_IN_PLACE) {
     recvtype  = sendtype;
@@ -421,7 +429,9 @@ int PMPI_Iscatterv(const void* sendbuf, const int* sendcounts, const int* displs
   } else {
     CHECK_NOT_IN_PLACE_ROOT(4, recvbuf)
   }
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Scatterv" : "PMPI_Iscatterv")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Scatterv" : "PMPI_Iscatterv");
+  name += " with root " + std::to_string(root);
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
 
@@ -471,7 +481,10 @@ int PMPI_Ireduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype dat
   CHECK_OP(5, op, datatype)
   CHECK_ROOT(7)
   CHECK_REQUEST(8)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce" : "PMPI_Ireduce")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Reduce" : "PMPI_Ireduce");
+  name += " with op " + op->name();
+  name += " and root " + std::to_string(root);
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
   aid_t pid = simgrid::s4u::this_actor::get_pid();
@@ -522,7 +535,9 @@ int PMPI_Iallreduce(const void *sendbuf, void *recvbuf, int count, MPI_Datatype
   CHECK_BUFFER(1, sendbuf, count, datatype)
   CHECK_BUFFER(2, recvbuf, count, datatype)
   CHECK_REQUEST(7)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Allreduce" : "PMPI_Iallreduce")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Alleduce" : "PMPI_Iallreduce");
+  name += " with op " + op->name();
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
   std::vector<unsigned char> tmp_sendbuf;
@@ -560,7 +575,9 @@ int PMPI_Iscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype datat
   CHECK_BUFFER(2,recvbuf,count, datatype)
   CHECK_REQUEST(7)
   CHECK_OP(5, op, datatype)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Scan" : "PMPI_Iscan")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Scan" : "PMPI_Iscan");
+  name += " with op " + op->name();
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
   aid_t pid = simgrid::s4u::this_actor::get_pid();
@@ -596,7 +613,9 @@ int PMPI_Iexscan(const void *sendbuf, void *recvbuf, int count, MPI_Datatype dat
   CHECK_BUFFER(2, recvbuf, count, datatype)
   CHECK_REQUEST(7)
   CHECK_OP(5, op, datatype)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Exscan" : "PMPI_Iexscan");
+  name += " with op " + op->name();
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
   aid_t pid = simgrid::s4u::this_actor::get_pid();
@@ -638,7 +657,9 @@ int PMPI_Ireduce_scatter(const void *sendbuf, void *recvbuf, const int *recvcoun
     CHECK_BUFFER(1, sendbuf, recvcounts[i], datatype)
     CHECK_BUFFER(2, recvbuf, recvcounts[i], datatype)
   }
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter" : "PMPI_Ireduce_scatter")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter" : "PMPI_Ireduce_scatter");
+  name += " with op " + op->name();
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
   aid_t pid                          = simgrid::s4u::this_actor::get_pid();
@@ -685,7 +706,9 @@ int PMPI_Ireduce_scatter_block(const void* sendbuf, void* recvbuf, int recvcount
   CHECK_BUFFER(2, recvbuf, recvcount, datatype)
   CHECK_REQUEST(7)
   CHECK_OP(5, op, datatype)
-  CHECK_COLLECTIVE(comm, request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter_block" : "PMPI_Ireduce_scatter_block")
+  std::string name = (request == MPI_REQUEST_IGNORED ? "PMPI_Reduce_scatter_block" : "PMPI_Ireduce_scatter_block");
+  name += " with op " + op->name();
+  CHECK_COLLECTIVE(comm, name.c_str())
 
   const SmpiBenchGuard suspend_bench;
   int count = comm->size();
diff --git a/src/smpi/include/smpi_op.hpp b/src/smpi/include/smpi_op.hpp
index ff1c1283f7..1409e8a302 100644
--- a/src/smpi/include/smpi_op.hpp
+++ b/src/smpi/include/smpi_op.hpp
@@ -19,14 +19,15 @@ class Op : public F2C{
   int refcount_ = 1;
   bool is_predefined_;
   int types_; //bitmask of the allowed datatypes flags
+  std::string name_;
 
 public:
-  Op(MPI_User_function* function, bool commutative, bool predefined=false, int types=0) : func_(function), is_commutative_(commutative), is_predefined_(predefined), types_(types) {if(not predefined) this->add_f();}
+  Op(MPI_User_function* function, bool commutative, bool predefined=false, int types=0, std::string name="MPI_Op") : func_(function), is_commutative_(commutative), is_predefined_(predefined), types_(types), name_(name) {if(not predefined) this->add_f();}
   bool is_commutative() const { return is_commutative_; }
   bool is_predefined() const { return is_predefined_; }
   bool is_fortran_op() const { return is_fortran_op_; }
   int allowed_types() const { return types_; }
-  std::string name() const override {return std::string("MPI_Op");}
+  std::string name() const override {return name_;}
   // tell that we were created from fortran, so we need to translate the type to fortran when called
   void set_fortran_op() { is_fortran_op_ = true; }
   void apply(const void* invec, void* inoutvec, const int* len, MPI_Datatype datatype) const;
diff --git a/src/smpi/internals/smpi_utils.cpp b/src/smpi/internals/smpi_utils.cpp
index 686bc14605..8d1fec4204 100644
--- a/src/smpi/internals/smpi_utils.cpp
+++ b/src/smpi/internals/smpi_utils.cpp
@@ -361,7 +361,7 @@ int check_collectives_ordering(MPI_Comm comm, std::string call){
         vec->second.push_back(call);
       } else if (vec->second.size() > count){
         if (vec->second[count] != call){
-          XBT_WARN("Collective communication mismatch. For process %ld, expected %s, got %s", simgrid::s4u::this_actor::get_pid(), vec->second[count].c_str(), call.c_str());
+          XBT_WARN("Collective operation mismatch. For process %ld, expected %s, got %s", simgrid::s4u::this_actor::get_pid(), vec->second[count].c_str(), call.c_str());
           return MPI_ERR_OTHER;
         }
       } else {
diff --git a/src/smpi/mpi/smpi_op.cpp b/src/smpi/mpi/smpi_op.cpp
index 3e4e34822f..6679911632 100644
--- a/src/smpi/mpi/smpi_op.cpp
+++ b/src/smpi/mpi/smpi_op.cpp
@@ -240,7 +240,7 @@ static void no_func(void*, void*, int*, MPI_Datatype*)
 
 
 #define CREATE_MPI_OP(name, func, types)                                                                                      \
-  SMPI_Op _XBT_CONCAT(smpi_MPI_, name)(&(func) /* func */, true, true, types);
+  SMPI_Op _XBT_CONCAT(smpi_MPI_, name)(&(func) /* func */, true, true, types, std::string(_XBT_STRINGIFY(MPI_##name)));
 
 #define MAX_TYPES DT_FLAG_C_INTEGER|DT_FLAG_F_INTEGER|DT_FLAG_FP|DT_FLAG_MULTILANG
 #define LAND_TYPES DT_FLAG_C_INTEGER|DT_FLAG_FP|DT_FLAG_LOGICAL|DT_FLAG_MULTILANG