part-persist: implement message aggregation

ompi/mca/part/base/part_base_frame.c

@@ -107,9 +107,6 @@ static int mca_part_base_close(void)
     }
     OBJ_DESTRUCT(&mca_part_base_part);
 
-    OBJ_DESTRUCT(&mca_part_base_psend_requests);
-    OBJ_DESTRUCT(&mca_part_base_precv_requests);
-
     /* Close all remaining available components */
     return mca_base_framework_components_close(&ompi_part_base_framework, NULL);
 }
@@ -122,9 +119,6 @@ static int mca_part_base_open(mca_base_open_flag_t flags)
 {
     OBJ_CONSTRUCT(&mca_part_base_part, opal_pointer_array_t);
 
-
-    OBJ_CONSTRUCT(&mca_part_base_psend_requests, opal_free_list_t);
-    OBJ_CONSTRUCT(&mca_part_base_precv_requests, opal_free_list_t);
     /* Open up all available components */
 
     if (OPAL_SUCCESS !=
@@ -137,8 +131,8 @@ static int mca_part_base_open(mca_base_open_flag_t flags)
 
     mca_part_base_selected_component.partm_finalize = NULL;
 
-    /* Currently this uses a default with no selection criteria as there is only 1 module. */
     opal_pointer_array_add(&mca_part_base_part, strdup("persist"));
+    opal_pointer_array_add(&mca_part_base_part, strdup("persist_aggregated"));
 
     return OMPI_SUCCESS;
 }

ompi/mca/part/base/part_base_prequest.c

@@ -24,13 +24,6 @@
 #include "ompi/mca/part/part.h"
 #include "ompi/mca/part/base/part_base_prequest.h"
 
-/**
- * If you wonder why these 2 freelists are declared here read the comment
- * in the part_base_request.h file.
- */
-opal_free_list_t mca_part_base_psend_requests = {{{0}}};
-opal_free_list_t mca_part_base_precv_requests = {{{0}}};
-
 static void mca_part_base_prequest_construct(mca_part_base_prequest_t* req)
 {
     req->req_ompi.req_type = OMPI_REQUEST_PART;

ompi/mca/part/base/part_base_prequest.h

@@ -35,12 +35,6 @@
 #include "opal/datatype/opal_convertor.h"
 BEGIN_C_DECLS
 
-/**
- * External list for the partitioned requests. 
- */
-OMPI_DECLSPEC extern opal_free_list_t mca_part_base_psend_requests;
-OMPI_DECLSPEC extern opal_free_list_t mca_part_base_precv_requests;
-
 typedef enum {
     MCA_PART_REQUEST_NULL,
     MCA_PART_REQUEST_PSEND,

ompi/mca/part/persist/Makefile.am

@@ -6,6 +6,8 @@
 # Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved.
 # Copyright (c) 2017      IBM Corporation.  All rights reserved.
 # Copyright (c) 2020-2021 Sandia National Laboratories. All rights reserved.
+# Copyright (c) 2024      High Performance Computing Center Stuttgart,
+#                         University of Stuttgart.  All rights reserved.
 # $COPYRIGHT$
 #
 # Additional copyrights may follow

ompi/mca/part/persist/part_persist.h

@@ -113,7 +113,7 @@ mca_part_persist_free_req(struct mca_part_persist_request_t* req)
 
 __opal_attribute_always_inline__ static inline void mca_part_persist_init_lists(void)
 {
-    opal_free_list_init (&mca_part_base_precv_requests,
+    opal_free_list_init (&mca_part_persist_precv_requests,
                          sizeof(mca_part_persist_precv_request_t),
                          opal_cache_line_size,
                          OBJ_CLASS(mca_part_persist_precv_request_t),
@@ -122,7 +122,7 @@ __opal_attribute_always_inline__ static inline void mca_part_persist_init_lists(
                          ompi_part_persist.free_list_max,
                          ompi_part_persist.free_list_inc,
                          NULL, 0, NULL, NULL, NULL);
-    opal_free_list_init (&mca_part_base_psend_requests,
+    opal_free_list_init (&mca_part_persist_psend_requests,
                          sizeof(mca_part_persist_psend_request_t),
                          opal_cache_line_size,
                          OBJ_CLASS(mca_part_persist_psend_request_t),

ompi/mca/part/persist/part_persist_component.c

@@ -12,6 +12,8 @@
  * Copyright (c) 2013-2021 Sandia National Laboratories.  All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
  *                         reserved.
+ * Copyright (c) 2024      High Performance Computing Center Stuttgart,
+ *                         University of Stuttgart.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -22,6 +24,7 @@
 #include "ompi_config.h"
 
 #include "ompi/mca/part/persist/part_persist.h"
+#include "ompi/mca/part/persist/part_persist_request.h"
 
 #include "ompi/mca/part/persist/part_persist_sendreq.h"
 #include "ompi/mca/part/persist/part_persist_recvreq.h"
@@ -96,6 +99,9 @@ mca_part_persist_component_open(void)
 
     ompi_part_persist.next_send_tag = 0;                /**< This is a counter for send tags for the actual data transfer. */
     ompi_part_persist.next_recv_tag = 0; 
+
+    OBJ_CONSTRUCT(&mca_part_persist_psend_requests, opal_free_list_t);
+    OBJ_CONSTRUCT(&mca_part_persist_precv_requests, opal_free_list_t);
 
     mca_part_persist_init_lists(); 
 
@@ -113,6 +119,9 @@ mca_part_persist_component_open(void)
 static int
 mca_part_persist_component_close(void)
 {
+    OBJ_DESTRUCT(&mca_part_persist_psend_requests);
+    OBJ_DESTRUCT(&mca_part_persist_precv_requests);
+
     OBJ_DESTRUCT(&ompi_part_persist.lock);
     return OMPI_SUCCESS; 
 }

ompi/mca/part/persist/part_persist_component.h

@@ -8,6 +8,8 @@
  * Copyright (c) 2004-2025 High Performance Computing Center Stuttgart,
  *                         University of Stuttgart.  All rights reserved.
  * Copyright (c) 2020-2021 Sandia National Laboratories. All rights reserved.
+ * Copyright (c) 2024      High Performance Computing Center Stuttgart,
+ *                         University of Stuttgart.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow

ompi/mca/part/persist/part_persist_recvreq.h

@@ -42,7 +42,7 @@ OBJ_CLASS_DECLARATION(mca_part_persist_precv_request_t);
 #define MCA_PART_PERSIST_PRECV_REQUEST_ALLOC(precvreq)                           \
 do {                                                                         \
     precvreq = (mca_part_persist_precv_request_t*)                               \
-      opal_free_list_get (&mca_part_base_precv_requests);                    \
+      opal_free_list_get (&mca_part_persist_precv_requests);                    \
     precvreq->req_base.req_type = MCA_PART_PERSIST_REQUEST_PRECV;            \
  } while (0)
 
@@ -94,7 +94,7 @@ do {                                                                  \
     OMPI_DATATYPE_RELEASE((recvreq)->req_datatype);            \
     OMPI_REQUEST_FINI(&(recvreq)->req_ompi);                   \
     opal_convertor_cleanup( &((recvreq)->req_convertor) );     \
-    opal_free_list_return ( &mca_part_base_precv_requests,               \
+    opal_free_list_return ( &mca_part_persist_precv_requests,               \
                            (opal_free_list_item_t*)(recvreq));          \
 }
 

ompi/mca/part/persist/part_persist_request.c

@@ -22,6 +22,10 @@
 #include "ompi/mca/part/persist/part_persist.h"
 #include "ompi/mca/part/persist/part_persist_request.h"
 
+/**
+ */
+opal_free_list_t mca_part_persist_psend_requests = {{{0}}};
+opal_free_list_t mca_part_persist_precv_requests = {{{0}}};
 
 static void mca_part_persist_request_construct( mca_part_persist_request_t* req) {
     OBJ_CONSTRUCT(&req->req_convertor, opal_convertor_t);

ompi/mca/part/persist/part_persist_request.h

@@ -45,6 +45,13 @@ struct ompi_mca_persist_setup_t {
 };
 
 
+/**
+ * External list for the partitioned requests. 
+ */
+OMPI_DECLSPEC extern opal_free_list_t mca_part_persist_psend_requests;
+OMPI_DECLSPEC extern opal_free_list_t mca_part_persist_precv_requests;
+
+
 /**
  *  Base type for PART PERSIST requests
  */

ompi/mca/part/persist/part_persist_sendreq.h

@@ -42,7 +42,7 @@ OBJ_CLASS_DECLARATION(mca_part_persist_psend_request_t);
                                            ompi_proc)                 \
 do {                                                                  \
     sendreq = (mca_part_persist_psend_request_t*)                         \
-        opal_free_list_wait (&mca_part_base_psend_requests);          \
+        opal_free_list_wait (&mca_part_persist_psend_requests);          \
     sendreq->req_base.req_type = MCA_PART_PERSIST_REQUEST_PSEND;          \
 } while(0)
 
@@ -87,7 +87,7 @@ do {                                                                  \
         OBJ_RELEASE(sendreq->req_comm);               \
         OMPI_REQUEST_FINI(&sendreq->req_ompi);        \
         opal_convertor_cleanup( &(sendreq->req_convertor) ); \
-        opal_free_list_return ( &mca_part_base_psend_requests,            \
+        opal_free_list_return ( &mca_part_persist_psend_requests,            \
                                (opal_free_list_item_t*)sendreq);        \
     }
 

ompi/mca/part/persist_aggregated/Makefile.am

@@ -0,0 +1,55 @@
+#
+# Copyright (c) 2004-2006 The Regents of the University of California.
+#                         All rights reserved.
+# Copyright (c) 2009      High Performance Computing Center Stuttgart,
+#                         University of Stuttgart.  All rights reserved.
+# Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved.
+# Copyright (c) 2017      IBM Corporation.  All rights reserved.
+# Copyright (c) 2020-2021 Sandia National Laboratories. All rights reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+# Make the output library in this directory, and name it either
+# mca_<type>_<name>.la (for DSO builds) or libmca_<type>_<name>.la
+# (for static builds).
+
+# EXTRA_DIST = post_configure.sh
+
+if MCA_BUILD_ompi_part_persist_aggregated_DSO
+component_noinst =
+component_install = mca_part_persist_aggregated.la
+else
+component_noinst = libmca_part_persist_aggregated.la
+component_install =
+endif
+
+local_sources = \
+	part_persist_aggregated.c \
+	part_persist_aggregated.h \
+	part_persist_aggregated_component.c \
+	part_persist_aggregated_component.h \
+	part_persist_aggregated_recvreq.h \
+	part_persist_aggregated_recvreq.c \
+	part_persist_aggregated_request.h \
+	part_persist_aggregated_request.c \
+	part_persist_aggregated_sendreq.h \
+	part_persist_aggregated_sendreq.c \
+	schemes/part_persist_aggregated_scheme_regular.h \
+	schemes/part_persist_aggregated_scheme_regular.c
+
+mcacomponentdir = $(ompilibdir)
+mcacomponent_LTLIBRARIES = $(component_install)
+mca_part_persist_aggregated_la_SOURCES = $(local_sources)
+mca_part_persist_aggregated_la_LIBADD = $(top_builddir)/ompi/lib@[email protected] \
+	$(part_persist_aggregated_LIBS)
+mca_part_persist_aggregated_la_LDFLAGS = -module -avoid-version $(part_persist_aggregated_LDFLAGS)
+
+noinst_LTLIBRARIES = $(component_noinst)
+libmca_part_persist_aggregated_la_SOURCES = $(local_sources)
+libmca_part_persist_aggregated_la_LIBADD = $(part_persist_aggregated_LIBS)
+libmca_part_persist_aggregated_la_LDFLAGS = -module -avoid-version $(part_persist_aggregated_LDFLAGS)
+

ompi/mca/part/persist_aggregated/part_persist_aggregated.c

@@ -0,0 +1,49 @@
+/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
+/*
+ * Copyright (c) 2006-2007 The Trustees of Indiana University and Indiana
+ *                         University Research and Technology
+ *                         Corporation.  All rights reserved.
+ * Copyright (c) 2004-2007 The University of Tennessee and The University
+ *                         of Tennessee Research Foundation.  All rights
+ *                         reserved.
+ * Copyright (c) 2004-2006 The Regents of the University of California.
+ *                         All rights reserved.
+ * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
+ *                         reserved.
+ * Copyright (c) 2011-2021 Sandia National Laboratories. All rights reserved.
+ * Copyright (c) 2024      High Performance Computing Center Stuttgart,
+ *                         University of Stuttgart.  All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#include "ompi_config.h"
+
+#include "ompi/communicator/communicator.h"
+#include "ompi/mca/part/base/part_base_prequest.h"
+#include "ompi/mca/part/base/base.h"
+
+#include "ompi/mca/part/persist_aggregated/part_persist_aggregated.h"
+#include "ompi/mca/part/persist_aggregated/part_persist_aggregated_sendreq.h"
+#include "ompi/mca/part/persist_aggregated/part_persist_aggregated_recvreq.h"
+
+ompi_part_persist_aggregated_t ompi_part_persist_aggregated = {
+    .super = {
+        .part_progress = mca_part_persist_aggregated_progress,
+        .part_precv_init = mca_part_persist_aggregated_precv_init,
+        .part_psend_init = mca_part_persist_aggregated_psend_init,
+        .part_start = mca_part_persist_aggregated_start,
+        .part_pready = mca_part_persist_aggregated_pready,
+        .part_parrived = mca_part_persist_aggregated_parrived, 
+    }
+};
+
+
+OBJ_CLASS_INSTANCE(mca_part_persist_aggregated_list_t,
+                   opal_list_item_t,
+                   NULL,
+                   NULL);
+