UBCL: Add PML/UBCL and OSC/UBCL

config/ompi_check_ubcl.m4

@@ -0,0 +1,57 @@
+# -*- shell-script -*-
+#
+# Copyright (C) 2015-2017 Mellanox Technologies, Inc.
+#                         All rights reserved.
+# Copyright (c) 2015      Research Organization for Information Science
+#                         and Technology (RIST). All rights reserved.
+# Copyright (c) 2016      Los Alamos National Security, LLC. All rights
+#                         reserved.
+# Copyright (c) 2016 Cisco Systems, Inc.  All rights reserved.
+# Copyright (c) 2022      Amazon.com, Inc. or its affiliates.  All Rights reserved.
+# Copyright (c) 2024-2025 Bull S.A.S.  All rights reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+# OMPI_CHECK_UBCL(prefix, [action-if-found], [action-if-not-found])
+# --------------------------------------------------------
+# check if UBCL support can be found.  sets prefix_{CPPFLAGS,
+# as needed and runs action-if-found if there is
+# support, otherwise executes action-if-not-found
+AC_DEFUN([OMPI_CHECK_UBCL],[
+    OPAL_VAR_SCOPE_PUSH([ompi_check_ubcl_dir ompi_check_ubcl_happy])
+
+    m4_ifblank([$1], [m4_fatal([First argument to OMPI_CHECK_UBCL cannot be blank])])
+
+    AC_ARG_WITH([ubcl],
+        [AC_HELP_STRING([--with-ubcl(=DIR)],
+            [Build with UBCL support])])
+
+    # UBCL is dlopen'd to avoid direct link to libubcl.so.
+    # OAC_CHECK_PACKAGE would add this explicit link, so it cannot be used.
+    # OPAL_CHECK_WITHDIR prints an error if the given path is invalid
+    OPAL_CHECK_WITHDIR([ubcl], [$with_ubcl], [include/ubcl_api.h])
+
+    AS_IF([test "$with_ubcl" == "no"],
+          [ompi_check_ubcl_happy="no"],
+
+          [test -z "$with_ubcl"],
+          [ompi_check_ubcl_happy="no"],
+
+          [ompi_check_ubcl_happy="yes"
+           $1_CPPFLAGS="${$1_CPPFLAGS} -I$with_ubcl/include/"
+           AC_MSG_NOTICE([$1_CPPFLAGS is set to: ${$1_CPPFLAGS}])])
+
+
+    OPAL_SUMMARY_ADD([Transports],[UBCL],[],[$ompi_check_ubcl_happy])
+
+    AS_IF([test "$ompi_check_ubcl_happy" = "yes"],
+          [$2],
+          [$3])
+
+    OPAL_VAR_SCOPE_POP
+])
+

ompi/mca/common/ubcl/Makefile.am

@@ -0,0 +1,94 @@
+# Copyright (c) 2025      Bull SAS.  All rights reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+#AM_CPPFLAGS = $(common_ubcl_CPPFLAGS)
+
+common_ubcl_sources  = \
+	common_ubcl.c \
+	common_ubcl.h
+
+lib_LTLIBRARIES =
+noinst_LTLIBRARIES =
+
+#Common component naming is forced by MCA_PROCESS_COMPONENT in config/opal_mca.m4
+# to lib${<PROJECT>_LIB_NAME}mca_common_ubcl.la but OMPI_LIB_NAME does not exist
+# so let's hope that no other project name is empty or there are no other common
+comp_inst = libmca_common_ubcl.la
+comp_noinst = libmca_common_ubcl_noinst.la
+
+if MCA_BUILD_ompi_common_ubcl_DSO
+lib_LTLIBRARIES += $(comp_inst)
+else
+noinst_LTLIBRARIES += $(comp_noinst)
+endif
+
+libmca_common_ubcl_la_SOURCES = $(common_ubcl_sources)
+libmca_common_ubcl_la_CFLAGS = $(common_ubcl_CFLAGS)
+libmca_common_ubcl_la_CPPFLAGS = $(common_ubcl_CPPFLAGS)
+libmca_common_ubcl_la_LDFLAGS = $(common_ubcl_LDFLAGS)
+libmca_common_ubcl_la_LIBADD = $(common_ubcl_LIBS) \
+	$(OPAL_TOP_BUILDDIR)/opal/mca/common/ubcl/lib@OPAL_LIB_NAME@mca_common_ubcl.la
+
+libmca_common_ubcl_noinst_la_SOURCES = $(common_ubcl_sources)
+
+# Conditionally install the header files
+
+if WANT_INSTALL_HEADERS
+ompidir = $(ompiincludedir)/$(subdir)
+ompi_HEADERS = common_ubcl.h
+endif
+
+
+# This library is linked against various MCA components.
+# There's two cases:
+#
+# 1. libmca_common_ubcl.la is a shared library.  By linking that shared
+# library to all components that need it, the OS linker will
+# automatically load it into the process as necessary, and there will
+# only be one copy (i.e., all the components will share *one* copy of
+# the code and data).
+#
+# 2. libmca_common_ubcl.la is a static library.  In this case, it will
+# be rolled up into the top-level libmpi.la.  It will also be rolled
+# into each component, but then the component will also be rolled up
+# into the upper-level libmpi.la.  Linkers universally know how to
+# "figure this out" so that we end up with only one copy of the code
+# and data.
+#
+# As per above, we'll either have an installable or noinst result.
+# The installable one should follow the same MCA prefix naming rules
+# (i.e., libmca_<type>_<name>.la).  The noinst one can be named
+# whatever it wants, although libmca_<type>_<name>_noinst.la is
+# recommended.
+
+# To simplify components that link to this library, we will *always*
+# have an output libtool library named libmca_<type>_<name>.la -- even
+# for case 2) described above (i.e., so there's no conditional logic
+# necessary in component Makefile.am's that link to this library).
+# Hence, if we're creating a noinst version of this library (i.e.,
+# case 2), we sym link it to the libmca_<type>_<name>.la name
+# (libtool will do the Right Things under the covers).  See the
+# all-local and clean-local rules, below, for how this is effected.
+# These two rules will sym link the "noinst" libtool library filename
+# to the installable libtool library filename in the case where we are
+# compiling this component statically (case 2), described above).
+V=0
+OMPI_V_LN_SCOMP = $(ompi__v_LN_SCOMP_$V)
+ompi__v_LN_SCOMP_ = $(ompi__v_LN_SCOMP_$AM_DEFAULT_VERBOSITY)
+ompi__v_LN_SCOMP_0 = @echo "  LN_S    " `basename $(comp_inst)`;
+
+all-local:
+	$(OMPI_V_LN_SCOMP) if test -z "$(lib_LTLIBRARIES)"; then \
+	  rm -f "$(comp_inst)"; \
+	  $(LN_S) "$(comp_noinst)" "$(comp_inst)"; \
+	fi
+
+clean-local:
+	if test -z "$(lib_LTLIBRARIES)"; then \
+	  rm -f "$(comp_inst)"; \
+	fi

ompi/mca/common/ubcl/common_ubcl.c

@@ -0,0 +1,173 @@
+/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
+/*
+ * Copyright (c) 2025      Bull SAS.  All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#include <execinfo.h>
+#include <stdint.h>
+#include <stdio.h>
+
+#include "ompi/communicator/communicator.h"
+#include "ompi/errhandler/errhandler.h"
+#include "ompi/include/mpi.h"
+#include "ompi/runtime/mpiruntime.h"
+#include "ompi/mca/common/ubcl/common_ubcl.h"
+#include "ompi/mca/pml/ubcl/pml_ubcl.h"
+#include "ompi/mca/pml/pml_constants.h"
+#include "opal/mca/common/ubcl/common_ubcl.h"
+#include "opal/util/output.h"
+
+/* Default ompi_common_ubcl values */
+mca_ompi_common_ubcl_component_t mca_ompi_common_ubcl_component = {
+    .n_addr = 32,
+};
+
+static int mca_common_ubcl_find_rank(const struct ompi_communicator_t *comm, const uint64_t wrank)
+{
+    mca_pml_ubcl_comm_t *pml_comm = comm->c_pml_comm;
+
+    if (NULL == comm->c_pml_comm) {
+        common_ubcl_error("UBCL error: no translation array in comm");
+        abort();
+    }
+
+    for (uint32_t i = 0; i < pml_comm->size; i++) {
+        if (pml_comm->array[i] == wrank) {
+            return i;
+        }
+    }
+
+    common_ubcl_error("UBCL error irank translation");
+
+    return 0;
+}
+
+int mca_common_ubcl_get_mpi_rank(const int rank, const struct ompi_communicator_t *comm,
+                                 const uint64_t ubcl_rank)
+{
+    if (OMPI_ANY_SOURCE == rank) {
+        return mca_common_ubcl_find_rank(comm, ubcl_rank);
+    } else {
+        return rank;
+    }
+}
+
+void mca_common_ubcl_status_to_ompi(ompi_status_public_t *status,
+                                    ubcl_status_t ubcl_status,
+                                    struct ompi_communicator_t *comm, int rank)
+{
+    if (MPI_STATUS_IGNORE != status) {
+        status->_cancelled = 0;     //TODO output the information of cancel
+        status->_ucount = ubcl_status.size;
+        status->MPI_TAG = (int) ubcl_status.tag;
+        status->MPI_SOURCE = mca_common_ubcl_get_mpi_rank(rank, comm, ubcl_status.remote);
+    }
+}
+
+int ubcl_error_to_ompi(ubcl_error_t code)
+{
+    int ret;
+    switch (code) {
+    case UBCL_SUCCESS:
+        ret = OPAL_SUCCESS;
+        break;
+    case UBCL_ERROR:
+        ret = OPAL_ERROR;
+        break;
+    case UBCL_ERR_RESOURCE_BUSY:
+        ret = OPAL_ERR_RESOURCE_BUSY;
+        break;
+    case UBCL_ERR_OUT_OF_RESOURCE:
+        ret = OPAL_ERR_OUT_OF_RESOURCE;
+        break;
+    case UBCL_ERR_NOT_IMPLEMENTED:
+        ret = OPAL_ERR_NOT_IMPLEMENTED;
+        break;
+    case UBCL_ERR_NOT_AVAILABLE:
+        ret = OPAL_ERR_NOT_AVAILABLE;
+        break;
+    case UBCL_ERR_TEMP_OUT_OF_RESOURCE:
+        ret = OPAL_ERR_TEMP_OUT_OF_RESOURCE;
+        break;
+    case UBCL_ERR_ARG_INVALID:
+        ret = OPAL_ERR_BAD_PARAM;
+        break;
+    case UBCL_ERR_TOO_LATE:
+        ret = OPAL_ERR_TIMEOUT;
+        break;
+    case UBCL_ERR_TRUNCATE:
+        ret = MPI_ERR_TRUNCATE;
+        break;
+    default:
+        ret = OPAL_ERROR;
+        break;
+    }
+
+    return ret;
+}
+
+void _mca_common_ubcl_error(char *filename, int line, int err,
+                                       char abort, int verbose, int output,
+                                       int is_init, int comp_verbose,
+                                       char *comp_name, char *format, ...)
+{
+    int n_addr   = 0;
+    void **stack_buffer = NULL;
+    char **stack = NULL;
+
+    stack_buffer = malloc(sizeof(void *) * mca_ompi_common_ubcl_component.n_addr);
+    n_addr       = backtrace(stack_buffer, mca_ompi_common_ubcl_component.n_addr);
+    stack        = backtrace_symbols(stack_buffer, n_addr);
+
+    int char_per_line = 256;
+    int n_char = char_per_line * n_addr + 1024;
+    char *msg = malloc(n_char * sizeof(char));
+
+    if (NULL == stack || NULL == msg) {
+        /* Output small error */
+        opal_output_verbose(verbose, output,
+                            "========\n== ERROR: Not enough memory while outputting error...\n== "
+                            "%s encountered an error (%d) at %s:%d\n========\n",
+                            comp_name,  err, filename, line);
+    } else {
+        /* Output full error */
+        int current = 0;
+        current += snprintf(msg + current, n_char - current,
+                            "========\n== %s encountered an error (%d) at %s:%d\n== %s:\n\t",
+                            comp_name, err, filename, line, abort ? "ERROR" : "WARNING");
+        va_list arglist;
+        va_start(arglist, format);
+        current += vsnprintf(msg + current, n_char - current, format, arglist);
+        va_end(arglist);
+
+        current += snprintf(msg + current, n_char - current, "\n== STACK:\n");
+
+        for (int i = 0; i < n_addr; i++) {
+            size_t min_char = char_per_line < (n_char - current) ? char_per_line : n_char - current;
+            current += snprintf(msg + current, min_char, "= [%2d] %s\n", i,
+                                stack[i]);
+        }
+
+        if (is_init && output > 0) {
+            opal_output_verbose(verbose, output,
+                                "%s========", msg);
+        } else if (abort || comp_verbose >= verbose) {
+            fprintf(stderr, "%s\n", msg);
+            fflush(stderr);
+        }
+    }
+
+    if (abort) {
+        OMPI_ERRHANDLER_INVOKE(&ompi_mpi_comm_world.comm, err, stack[0]);
+        ompi_mpi_abort(&ompi_mpi_comm_world.comm, err);
+    }
+
+    free(stack_buffer);
+    free(stack);
+    free(msg);
+}

ompi/mca/common/ubcl/common_ubcl.h

@@ -0,0 +1,44 @@
+/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
+/*
+ * Copyright (c) 2025      Bull SAS.  All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#ifndef OMPI_MCA_COMMON_UBCL_H
+#define OMPI_MCA_COMMON_UBCL_H
+
+#include <stddef.h>
+
+#include "ompi/communicator/communicator.h"
+#include "ompi/include/mpi.h"
+#include "opal/mca/common/ubcl/common_ubcl.h"
+
+/* Holds common variable used in multiple UBCL components */
+struct mca_ompi_common_ubcl_component_s {
+    int n_addr; /**< Max number of void * addresses in printed stack*/
+};
+typedef struct mca_ompi_common_ubcl_component_s mca_ompi_common_ubcl_component_t;
+extern mca_ompi_common_ubcl_component_t mca_ompi_common_ubcl_component;
+
+int mca_common_ubcl_get_mpi_rank(const int rank, const struct ompi_communicator_t *comm,
+                                 const uint64_t ubcl_rank);
+void mca_common_ubcl_status_to_ompi(ompi_status_public_t *status,
+                                    ubcl_status_t ubcl_status,
+                                    struct ompi_communicator_t *comm, int rank);
+int ubcl_error_to_ompi(ubcl_error_t code);
+/* UBCL rank is on 61 bits, ompi jobid is 32bits, vpid must be truncated to 29bits */
+#define COMMON_UBCL_VPID_MAX    (((1 << 29) - 1)) /* We need 3 bits for UBCL rank */
+#define PML_UBCL_JOBID_MAX  (OPAL_JOBID_MAX)
+
+/* Error and warning output function used by UBCL components */
+void _mca_common_ubcl_error(char *filename, int line, int err, char abort, int verbose,
+                            int output, int is_init, int comp_verbose, char *comp_name,
+                            char *format, ...);
+
+
+#endif /* OMPI_MCA_COMMON_UBCL_H */
+