communicator: make c_name a dynamic array and reorder struct

edgargabriel · edgargabriel · commit 2d68804b21cd · 2023-02-14T20:34:02.000Z
make the c_name element of the communicator structure a dynamic element. This allows us to reduce the size of PREDEFINED_COMMUNICATOR_PAD back to 512 to maintain backwards compatibility with the ompi 4.1.x release series. Reorder the communicator fields to reduce the struct size. This brings the communicator size at 536 bytes with FT, PERUSE enabled and compiled in debug mode. Fixes issue #11373 Signed-off-by: Edgar Gabriel <edgar.gabriel@amd.com> Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
diff --git a/ompi/communicator/comm.c b/ompi/communicator/comm.c
@@ -26,6 +26,7 @@
  * Copyright (c) 2021      Nanook Consulting.  All rights reserved.
  * Copyright (c) 2018-2022 Triad National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2023      Advanced Micro Devices, Inc. All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -215,6 +216,11 @@ int ompi_comm_set_nb (ompi_communicator_t **ncomm, ompi_communicator_t *oldcomm,
     if (NULL == newcomm) {
         return OMPI_ERR_OUT_OF_RESOURCE;
     }
+    newcomm->c_name = (char*) malloc (OPAL_MAX_OBJECT_NAME);
+    if (NULL == newcomm->c_name) {
+        return OMPI_ERR_OUT_OF_RESOURCE;
+    }
+    newcomm->c_name[0]    = '\0';
     newcomm->super.s_info = NULL;
     /* fill in the inscribing hyper-cube dimensions */
     newcomm->c_cube_dim = opal_cube_dim(local_size);
diff --git a/ompi/communicator/comm_init.c b/ompi/communicator/comm_init.c
@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2020 The University of Tennessee and The University
+ * Copyright (c) 2004-2023 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -25,6 +25,7 @@
  * Copyright (c) 2016-2017 IBM Corporation. All rights reserved.
  * Copyright (c) 2018-2022 Triad National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2023      Advanced Micro Devices, Inc. All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -169,8 +170,7 @@ int ompi_comm_init(void)
     (void)opal_pointer_array_test_and_set_item(&ompi_mpi_communicators, 0, &ompi_mpi_comm_null);
     (void)opal_pointer_array_test_and_set_item(&ompi_mpi_communicators, 1, &ompi_mpi_comm_null);
 
-    opal_string_copy(ompi_mpi_comm_null.comm.c_name, "MPI_COMM_NULL",
-                     sizeof(ompi_mpi_comm_null.comm.c_name));
+    ompi_mpi_comm_null.comm.c_name = strdup ("MPI_COMM_NULL");
     ompi_mpi_comm_null.comm.c_flags |= OMPI_COMM_NAMEISSET | OMPI_COMM_INTRINSIC |
         OMPI_COMM_GLOBAL_INDEX;
 
@@ -221,8 +221,7 @@ int ompi_comm_init_mpi3 (void)
     OMPI_COMM_SET_PML_ADDED(&ompi_mpi_comm_world.comm);
     opal_pointer_array_set_item (&ompi_mpi_communicators, 0, &ompi_mpi_comm_world);
 
-    opal_string_copy(ompi_mpi_comm_world.comm.c_name, "MPI_COMM_WORLD",
-                     sizeof(ompi_mpi_comm_world.comm.c_name));
+    ompi_mpi_comm_world.comm.c_name = strdup("MPI_COMM_WORLD");
     ompi_mpi_comm_world.comm.c_flags |= OMPI_COMM_NAMEISSET | OMPI_COMM_INTRINSIC |
         OMPI_COMM_GLOBAL_INDEX;
     ompi_mpi_comm_world.comm.instance = group->grp_instance;
@@ -280,8 +279,7 @@ int ompi_comm_init_mpi3 (void)
     OMPI_COMM_SET_PML_ADDED(&ompi_mpi_comm_self.comm);
     opal_pointer_array_set_item (&ompi_mpi_communicators, 1, &ompi_mpi_comm_self);
 
-    opal_string_copy(ompi_mpi_comm_self.comm.c_name, "MPI_COMM_SELF",
-                     sizeof(ompi_mpi_comm_self.comm.c_name));
+    ompi_mpi_comm_self.comm.c_name = strdup("MPI_COMM_SELF");
     ompi_mpi_comm_self.comm.c_flags |= OMPI_COMM_NAMEISSET | OMPI_COMM_INTRINSIC |
         OMPI_COMM_GLOBAL_INDEX;
     ompi_mpi_comm_self.comm.instance = group->grp_instance;
@@ -412,7 +410,7 @@ static int ompi_comm_finalize (void)
 static void ompi_comm_construct(ompi_communicator_t* comm)
 {
     int idx;
-    comm->c_name[0]      = '\0';
+    comm->c_name         = NULL;
     comm->c_index        = MPI_UNDEFINED;
     comm->c_flags        = 0;
     comm->c_my_rank      = 0;
@@ -444,7 +442,7 @@ static void ompi_comm_construct(ompi_communicator_t* comm)
        this communicator */
     comm->c_keyhash      = NULL;
 
-    comm->errhandler_type  = OMPI_ERRHANDLER_TYPE_COMM;
+    comm->error_handler  = &ompi_mpi_errors_are_fatal.eh;
 #ifdef OMPI_WANT_PERUSE
     comm->c_peruse_handles = NULL;
 #endif
@@ -520,6 +518,11 @@ static void ompi_comm_destruct(ompi_communicator_t* comm)
         comm->error_handler = NULL;
     }
 
+    if (NULL != comm->c_name) {
+        free (comm->c_name);
+        comm->c_name = NULL;
+    }
+
 #if OPAL_ENABLE_FT_MPI
     if( NULL != comm->agreement_specific ) {
         OBJ_RELEASE( comm->agreement_specific );
diff --git a/ompi/communicator/communicator.h b/ompi/communicator/communicator.h
@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2020 The University of Tennessee and The University
+ * Copyright (c) 2004-2023 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -24,6 +24,7 @@
  * Copyright (c) 2016-2017 IBM Corporation. All rights reserved.
  * Copyright (c) 2018-2022 Triad National Security, LLC. All rights
  *                         reserved.
+ * Copyright (c) 2023      Advanced Micro Devices, Inc. All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -261,20 +262,30 @@ struct ompi_communicator_t {
     opal_infosubscriber_t      super;
     opal_mutex_t               c_lock; /* mutex for name and potentially
                                           attributes */
-    char  c_name[MPI_MAX_OBJECT_NAME];
+    char                      *c_name;
     ompi_comm_extended_cid_t      c_contextid;
     ompi_comm_extended_cid_block_t c_contextidb;
     uint32_t                      c_index;
     int                           c_my_rank;
     uint32_t                      c_flags; /* flags, e.g. intercomm,
                                               topology, etc. */
     uint32_t                      c_assertions; /* info assertions */
-    int c_id_available; /* the currently available Cid for allocation
-               to a child*/
-    int c_id_start_index; /* the starting index of the block of cids
-                 allocated to this communicator*/
+#if OPAL_ENABLE_FT_MPI
     uint32_t c_epoch;  /* Identifier used to differentiate between two communicators
                           using the same c_contextid (not at the same time, obviously) */
+#endif
+    /* Non-blocking collective tag. These tags might be shared between
+     * all non-blocking collective modules (to avoid message collision
+     * between them in the case where multiple outstanding non-blocking
+     * collective coexists using multiple backends).
+     */
+    opal_atomic_int32_t c_nbc_tag;
+
+    /**< inscribing cube dimension */
+    int c_cube_dim;
+
+    /* index in Fortran <-> C translation array */
+    int c_f_to_c_index;
 
     ompi_group_t        *c_local_group;
     ompi_group_t       *c_remote_group;
@@ -287,16 +298,10 @@ struct ompi_communicator_t {
     /* Attributes */
     struct opal_hash_table_t       *c_keyhash;
 
-    /**< inscribing cube dimension */
-    int c_cube_dim;
-
     /* Standard information about the selected topology module (or NULL
        if this is not a cart, graph or dist graph communicator) */
     struct mca_topo_base_module_t* c_topo;
 
-    /* index in Fortran <-> C translation array */
-    int c_f_to_c_index;
-
 #ifdef OMPI_WANT_PERUSE
     /*
      * Place holder for the PERUSE events.
@@ -307,9 +312,7 @@ struct ompi_communicator_t {
     /* Error handling.  This field does not have the "c_" prefix so
        that the OMPI_ERRHDL_* macros can find it, regardless of whether
        it's a comm, window, or file. */
-
     ompi_errhandler_t                  *error_handler;
-    ompi_errhandler_type_t             errhandler_type;
 
     /* Hooks for PML to hang things */
     struct mca_pml_comm_t  *c_pml_comm;
@@ -320,21 +323,14 @@ struct ompi_communicator_t {
     /* Collectives module interface and data */
     mca_coll_base_comm_coll_t *c_coll;
 
-    /* Non-blocking collective tag. These tags might be shared between
-     * all non-blocking collective modules (to avoid message collision
-     * between them in the case where multiple outstanding non-blocking
-     * collective coexists using multiple backends).
-     */
-    opal_atomic_int32_t c_nbc_tag;
-
     /* instance that this comm belongs to */
     ompi_instance_t* instance;
 
 #if OPAL_ENABLE_FT_MPI
-    /** MPI_ANY_SOURCE Failed Group Offset - OMPI_Comm_failure_get_acked */
-    int                      any_source_offset;
     /** agreement caching info for topology and previous returned decisions */
     opal_object_t           *agreement_specific;
+    /** MPI_ANY_SOURCE Failed Group Offset - OMPI_Comm_failure_get_acked */
+    int                      any_source_offset;
     /** Are MPI_ANY_SOURCE operations enabled? - OMPI_Comm_failure_ack */
     bool                     any_source_enabled;
     /** Has this communicator been revoked - OMPI_Comm_revoke() */
@@ -437,7 +433,7 @@ typedef struct ompi_communicator_t ompi_communicator_t;
  * the PREDEFINED_COMMUNICATOR_PAD macro?
  * A: Most likely not, but it would be good to check.
  */
-#define PREDEFINED_COMMUNICATOR_PAD 1024
+#define PREDEFINED_COMMUNICATOR_PAD 512
 
 struct ompi_predefined_communicator_t {
     struct ompi_communicator_t comm;
diff --git a/ompi/debuggers/predefined_gap_test.c b/ompi/debuggers/predefined_gap_test.c
@@ -1,7 +1,7 @@
 /*
  * Copyright (c) 2009      Sun Microsystems, Inc  All rights reserved.
  * Copyright (c) 2009-2013 Cisco Systems, Inc.  All rights reserved.
- * Copyright (c) 2012-2013 The University of Tennessee and The University
+ * Copyright (c) 2012-2023 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2012-2013 Inria.  All rights reserved.
@@ -59,8 +59,6 @@ int main(int argc, char **argv) {
     GAP_CHECK("c_contextid", test_comm, c_contextid, c_name, 1);
     GAP_CHECK("c_my_rank", test_comm, c_my_rank, c_contextid, 1);
     GAP_CHECK("c_flags", test_comm, c_flags, c_my_rank, 1);
-    GAP_CHECK("c_id_available", test_comm, c_id_available, c_flags, 1);
-    GAP_CHECK("c_id_start_index", test_comm, c_id_start_index, c_id_available,  1);
     GAP_CHECK("c_remote_group", test_comm, c_remote_group, c_local_group, 1);
     GAP_CHECK("c_local_comm", test_comm, c_local_comm,  c_remote_group,  1);
     GAP_CHECK("c_keyhash", test_comm, c_keyhash, c_local_comm,  1);
@@ -73,8 +71,6 @@ int main(int argc, char **argv) {
 #else
     GAP_CHECK("error_handler", test_comm, error_handler, c_f_to_c_index, 1);
 #endif
-    GAP_CHECK("errhandler_type", test_comm, errhandler_type, error_handler, 1);
-    GAP_CHECK("c_pml_comm", test_comm, c_pml_comm, errhandler_type, 1);
     GAP_CHECK("c_coll", test_comm, c_coll, c_pml_comm, 1);
 
     /* Test Predefined group sizes */
@@ -129,8 +125,6 @@ int main(int argc, char **argv) {
     GAP_CHECK("w_keyhash", test_win, w_keyhash, w_flags, 1);
     GAP_CHECK("w_f_to_c_index", test_win, w_f_to_c_index, w_keyhash, 1);
     GAP_CHECK("error_handler", test_win, error_handler, w_f_to_c_index, 1);
-    GAP_CHECK("errhandler_type", test_win, errhandler_type, error_handler, 1);
-    GAP_CHECK("w_osc_module", test_win, w_osc_module, errhandler_type, 1);
 
     /* Test Predefined info sizes */
     printf("=============================================\n");
@@ -151,8 +145,6 @@ int main(int argc, char **argv) {
     GAP_CHECK("f_flags", test_file, f_flags, f_amode,  1);
     GAP_CHECK("f_f_to_c_index", test_file, f_f_to_c_index, f_flags, 1);
     GAP_CHECK("error_handler", test_file, error_handler, f_f_to_c_index, 1);
-    GAP_CHECK("errhandler_type", test_file, errhandler_type, error_handler, 1);
-    GAP_CHECK("f_io_version", test_file, f_io_version, errhandler_type, 1);
     GAP_CHECK("f_io_selected_component", test_file, f_io_selected_component, f_io_version, 1);
     GAP_CHECK("f_io_selected_module", test_file, f_io_selected_module, f_io_selected_component, 1);
     GAP_CHECK("f_io_selected_data", test_file, f_io_selected_data, f_io_selected_module, 1);
diff --git a/ompi/errhandler/errhandler.c b/ompi/errhandler/errhandler.c
@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2022 The University of Tennessee and The University
+ * Copyright (c) 2004-2023 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -388,10 +388,10 @@ int ompi_errhandler_proc_failed_internal(ompi_proc_t* ompi_proc, int status, boo
                              OMPI_NAME_PRINT(&ompi_proc->super.proc_name),
                              ompi_comm_print_cid(comm),
                              proc_rank,
-                             (OMPI_ERRHANDLER_TYPE_PREDEFINED == comm->errhandler_type ? "P" :
-                              (OMPI_ERRHANDLER_TYPE_COMM == comm->errhandler_type ? "C" :
-                               (OMPI_ERRHANDLER_TYPE_WIN == comm->errhandler_type ? "W" :
-                                (OMPI_ERRHANDLER_TYPE_FILE == comm->errhandler_type ? "F" : "U") ) ) )
+                             (OMPI_ERRHANDLER_TYPE_PREDEFINED == comm->error_handler->eh_mpi_object_type ? "P" :
+                              (OMPI_ERRHANDLER_TYPE_COMM == comm->error_handler->eh_mpi_object_type ? "C" :
+                               (OMPI_ERRHANDLER_TYPE_WIN == comm->error_handler->eh_mpi_object_type ? "W" :
+                                (OMPI_ERRHANDLER_TYPE_FILE == comm->error_handler->eh_mpi_object_type ? "F" : "U") ) ) )
                              ));
     }
 
diff --git a/ompi/errhandler/errhandler.h b/ompi/errhandler/errhandler.h
@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2022 The University of Tennessee and The University
+ * Copyright (c) 2004-2023 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -238,7 +238,7 @@ extern opal_atomic_int32_t ompi_instance_count;
 #define OMPI_ERRHANDLER_INVOKE(mpi_object, err_code, message) \
   ompi_errhandler_invoke((mpi_object)->error_handler, \
                          (mpi_object), \
-                         (int)(mpi_object)->errhandler_type, \
+                         (int)(mpi_object)->error_handler->eh_mpi_object_type, \
                          ompi_errcode_get_mpi_code(err_code), \
                          (message));
 
@@ -269,7 +269,7 @@ extern opal_atomic_int32_t ompi_instance_count;
     int __mpi_err_code = ompi_errcode_get_mpi_code(err_code);         \
     ompi_errhandler_invoke((mpi_object)->error_handler, \
                            (mpi_object), \
-                           (int) (mpi_object)->errhandler_type, \
+                           (int) (mpi_object)->error_handler->eh_mpi_object_type, \
                            (__mpi_err_code), \
                            (message)); \
     return (__mpi_err_code); \
@@ -307,7 +307,7 @@ extern opal_atomic_int32_t ompi_instance_count;
     int __mpi_err_code = ompi_errcode_get_mpi_code(err_code);         \
     ompi_errhandler_invoke((mpi_object)->error_handler, \
                            (mpi_object), \
-                           (int)(mpi_object)->errhandler_type, \
+                           (int)(mpi_object)->error_handler->eh_mpi_object_type, \
                            (__mpi_err_code), \
                            (message)); \
     return (__mpi_err_code); \
diff --git a/ompi/errhandler/errhandler_invoke.c b/ompi/errhandler/errhandler_invoke.c
@@ -3,7 +3,7 @@
  * Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
  *                         University Research and Technology
  *                         Corporation.  All rights reserved.
- * Copyright (c) 2004-2020 The University of Tennessee and The University
+ * Copyright (c) 2004-2023 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@@ -183,19 +183,19 @@ int ompi_errhandler_request_invoke(int count,
     case OMPI_REQUEST_COLL:
         return ompi_errhandler_invoke(mpi_object.comm->error_handler,
                                       mpi_object.comm,
-                                      mpi_object.comm->errhandler_type,
+                                      mpi_object.comm->error_handler->eh_mpi_object_type,
                                       ec, message);
         break;
     case OMPI_REQUEST_IO:
         return ompi_errhandler_invoke(mpi_object.file->error_handler,
                                       mpi_object.file,
-                                      mpi_object.file->errhandler_type,
+                                      mpi_object.file->error_handler->eh_mpi_object_type,
                                       ec, message);
         break;
     case OMPI_REQUEST_WIN:
         return ompi_errhandler_invoke(mpi_object.win->error_handler,
                                       mpi_object.win,
-                                      mpi_object.win->errhandler_type,
+                                      mpi_object.win->error_handler->eh_mpi_object_type,
                                       ec, message);
         break;
     default:
diff --git a/ompi/include/ompi/memchecker.h b/ompi/include/ompi/memchecker.h
@@ -1,7 +1,7 @@
 /*
  * Copyright (c) 2004-2008 High Performance Computing Center Stuttgart,
  *                         University of Stuttgart.  All rights reserved.
- * Copyright (c) 2010-2017 The University of Tennessee and The University
+ * Copyright (c) 2010-2023 The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
  * Copyright (c) 2009      Oak Ridge National Labs.  All rights reserved.
@@ -220,8 +220,6 @@ static inline int memchecker_comm(MPI_Comm comm)
     opal_memchecker_base_isdefined (&comm->c_name, MPI_MAX_OBJECT_NAME);
     opal_memchecker_base_isdefined (&comm->c_my_rank, sizeof(int));
     opal_memchecker_base_isdefined (&comm->c_flags, sizeof(uint32_t));
-    opal_memchecker_base_isdefined (&comm->c_id_available, sizeof(int));
-    opal_memchecker_base_isdefined (&comm->c_id_start_index, sizeof(int));
     opal_memchecker_base_isdefined (&comm->c_local_group, sizeof(ompi_group_t *));
     opal_memchecker_base_isdefined (&comm->c_remote_group, sizeof(ompi_group_t *));
     opal_memchecker_base_isdefined (&comm->c_keyhash, sizeof(struct opal_hash_table_t *));
