Merge branch 'main' into main

Author: mshanthagit

docs/Makefile.am

@@ -881,7 +881,14 @@ ALL_MAN_BUILT = \
 # built HTML and man docs into a separate location that is included in
 # the tarball.  This gives users a fully copy of the docs included in
 # distribution tarballs.
-html: $(ALL_MAN_BUILT)
+#
+# Note: we name this "html-local" because otherwise Automake issues a
+# warning about us overriding its default "html" target.  In our
+# particular use case, the rules generated if we name this target
+# "html-local" instead of "html" are effectively equivalent to when we
+# name this target "html", so we might as well avoid the Automake
+# warning.
+html-local: $(ALL_MAN_BUILT)
 	$(OMPI_V_COPYALL) rm -rf html; cp -rp $(OUTDIR)/html .
 
 man: $(ALL_MAN_BUILT)

docs/conf.py

@@ -37,10 +37,10 @@ def read_version_file(path):
     if not os.path.exists(path):
         print(f"ERROR: Unable to find file {path}")
         exit(1)
-        
+
     with open(path) as fp:
         version_lines = fp.readlines()
-    
+
     data = dict()
     for line in version_lines:
         if '#' in line:
@@ -129,6 +129,15 @@ def get_tarball_version(path, expr):
 if key in os.environ and os.environ[key] == 'True':
     print("OMPI: found ReadTheDocs build environment")
 
+    # Tell Jinja2 templates the build is running on Read the Docs
+    if "html_context" not in globals():
+        html_context = {}
+    html_context["READTHEDOCS"] = True
+
+    # Define the canonical URL if you are using a custom domain on
+    # Read the Docs
+    html_baseurl = os.environ.get("READTHEDOCS_CANONICAL_URL", "")
+
     rtd_v = os.environ['READTHEDOCS_VERSION']
     if os.environ['READTHEDOCS_VERSION_TYPE'] == 'external':
         # Make "release" be shorter than the full "ompi_ver" value.

ompi/communicator/communicator.h

@@ -25,6 +25,7 @@
  * Copyright (c) 2018-2024 Triad National Security, LLC. All rights
  *                         reserved.
  * Copyright (c) 2023      Advanced Micro Devices, Inc. All rights reserved.
+ * Copyright (c) 2024      NVIDIA Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -613,11 +614,6 @@ static inline struct ompi_proc_t* ompi_comm_peer_lookup (const ompi_communicator
     return ompi_group_peer_lookup(comm->c_remote_group,peer_id);
 }
 
-static inline bool ompi_comm_instances_same(const ompi_communicator_t *comm1, const ompi_communicator_t *comm2)
-{
-    return comm1->instance == comm2->instance;
-}
-
 #if OPAL_ENABLE_FT_MPI
 /*
  * Support for MPI_ANY_SOURCE point-to-point operations

ompi/mca/coll/han/Makefile.am

@@ -2,7 +2,7 @@
 # Copyright (c) 2018-2020 The University of Tennessee and The University
 #                         of Tennessee Research Foundation.  All rights
 #                         reserved.
-# Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
+# Copyright (c) 2024      Amazon.com, Inc. or its affiliates.  All Rights reserved.
 # Copyright (c) 2022      BULL S.A.S. All rights reserved.
 # $COPYRIGHT$
 #
@@ -15,6 +15,7 @@ sources = \
 coll_han.h \
 coll_han_trigger.h \
 coll_han_algorithms.h \
+coll_han_alltoall.c \
 coll_han_dynamic.h \
 coll_han_dynamic_file.h \
 coll_han_barrier.c \

ompi/mca/coll/han/coll_han.h

@@ -42,6 +42,7 @@
 #include "mpi.h"
 #include "ompi/mca/mca.h"
 #include "opal/util/output.h"
+#include "opal/mca/smsc/smsc.h"
 #include "ompi/mca/coll/base/coll_base_functions.h"
 #include "coll_han_trigger.h"
 #include "ompi/mca/coll/han/coll_han_dynamic.h"
@@ -197,6 +198,7 @@ typedef struct mca_coll_han_op_module_name_t {
     mca_coll_han_op_up_low_module_name_t gatherv;
     mca_coll_han_op_up_low_module_name_t scatter;
     mca_coll_han_op_up_low_module_name_t scatterv;
+    mca_coll_han_op_up_low_module_name_t alltoall;
 } mca_coll_han_op_module_name_t;
 
 /**
@@ -252,6 +254,13 @@ typedef struct mca_coll_han_component_t {
     uint32_t han_scatterv_up_module;
     /* low level module for scatterv */
     uint32_t han_scatterv_low_module;
+
+    /* low level module for alltoall */
+    uint32_t han_alltoall_low_module;
+    /* alltoall: parallel stages */
+    int32_t han_alltoall_pstages;
+
+
     /* name of the modules */
     mca_coll_han_op_module_name_t han_op_module_name;
     /* whether we need reproducible results
@@ -287,6 +296,7 @@ typedef struct mca_coll_han_single_collective_fallback_s
 {
     union
     {
+        mca_coll_base_module_alltoall_fn_t alltoall;
         mca_coll_base_module_allgather_fn_t allgather;
         mca_coll_base_module_allgatherv_fn_t allgatherv;
         mca_coll_base_module_allreduce_fn_t allreduce;
@@ -308,6 +318,7 @@ typedef struct mca_coll_han_single_collective_fallback_s
  */
 typedef struct mca_coll_han_collectives_fallback_s
 {
+    mca_coll_han_single_collective_fallback_t alltoall;
     mca_coll_han_single_collective_fallback_t allgather;
     mca_coll_han_single_collective_fallback_t allgatherv;
     mca_coll_han_single_collective_fallback_t allreduce;
@@ -370,6 +381,9 @@ OBJ_CLASS_DECLARATION(mca_coll_han_module_t);
  * Some defines to stick to the naming used in the other components in terms of
  * fallback routines
  */
+#define previous_alltoall           fallback.alltoall.alltoall
+#define previous_alltoall_module    fallback.alltoall.module
+
 #define previous_allgather          fallback.allgather.allgather
 #define previous_allgather_module   fallback.allgather.module
 
@@ -425,6 +439,7 @@ OBJ_CLASS_DECLARATION(mca_coll_han_module_t);
         HAN_UNINSTALL_COLL_API(COMM, HANM, allreduce);                 \
         HAN_UNINSTALL_COLL_API(COMM, HANM, allgather);                 \
         HAN_UNINSTALL_COLL_API(COMM, HANM, allgatherv);                \
+        HAN_UNINSTALL_COLL_API(COMM, HANM, alltoall);                  \
         han_module->enabled = false;  /* entire module set to pass-through from now on */ \
     } while(0)
 
@@ -485,6 +500,9 @@ mca_coll_han_get_all_coll_modules(struct ompi_communicator_t *comm,
                                   mca_coll_han_module_t *han_module);
 
 int
+mca_coll_han_alltoall_intra_dynamic(ALLTOALL_BASE_ARGS,
+                                    mca_coll_base_module_t *module);
+int
 mca_coll_han_allgather_intra_dynamic(ALLGATHER_BASE_ARGS,
                                      mca_coll_base_module_t *module);
 int
@@ -532,4 +550,20 @@ coll_han_utils_gcd(const uint64_t *numerators, const size_t size);
 int
 coll_han_utils_create_contiguous_datatype(size_t count, const ompi_datatype_t *oldType,
                                           ompi_datatype_t **newType);
+
+static inline struct mca_smsc_endpoint_t *mca_coll_han_get_smsc_endpoint (struct ompi_proc_t *proc) {
+    extern opal_mutex_t mca_coll_han_lock;
+    if (NULL == proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_SMSC]) {
+        if (NULL == proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_SMSC]) {
+            OPAL_THREAD_LOCK(&mca_coll_han_lock);
+            if (NULL == proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_SMSC]) {
+                proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_SMSC] = mca_smsc->get_endpoint(&proc->super);
+            }
+            OPAL_THREAD_UNLOCK(&mca_coll_han_lock);
+        }
+    }
+
+    return (struct mca_smsc_endpoint_t *) proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_SMSC];
+}
+
 #endif                          /* MCA_COLL_HAN_EXPORT_H */

ompi/mca/coll/han/coll_han_algorithms.c

@@ -1,6 +1,7 @@
 /* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
  * Copyright (c) 2020-2022 Bull S.A.S. All rights reserved.
+ * Copyright (c) 2024      Amazon.com, Inc. or its affiliates. All Rights Reserved.
  *
  * $COPYRIGHT$
  *
@@ -77,6 +78,10 @@ mca_coll_han_algorithm_value_t*  mca_coll_han_available_algorithms[COLLCOUNT] =
         {"simple", (fnptr_t)&mca_coll_han_allgather_intra_simple}, // 2-level
         { 0 }
     },
+    [ALLTOALL] = (mca_coll_han_algorithm_value_t[]){
+        {"smsc", (fnptr_t)&mca_coll_han_alltoall_using_smsc}, // 2-level
+        { 0 }
+    },
 };
 
 int

ompi/mca/coll/han/coll_han_algorithms.h

@@ -1,6 +1,7 @@
 /* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
 /*
  * Copyright (c) 2020-2022 Bull S.A.S. All rights reserved.
+ * Copyright (c) 2024      Amazon.com, Inc. or its affiliates. All Rights Reserved.
  *
  * $COPYRIGHT$
  *
@@ -208,4 +209,10 @@ mca_coll_han_allgather_intra_simple(const void *sbuf, size_t scount,
                                     struct ompi_communicator_t *comm,
                                     mca_coll_base_module_t *module);
 
+/* Alltoall */
+int
+mca_coll_han_alltoall_using_smsc(ALLTOALL_BASE_ARGS,
+                                    mca_coll_base_module_t *module);
+
+
 #endif