Merge pull request #14 from rhc54/topic/sync

Sync to upstream

Author: Ralph Castain

.gitignore

@@ -300,6 +300,12 @@ opal/mca/hwloc/base/static-components.h.new.struct
 
 opal/mca/installdirs/config/install_dirs.h
 
+opal/mca/pmix/ext2x/ext2x.c
+opal/mca/pmix/ext2x/ext2x.h
+opal/mca/pmix/ext2x/ext2x_client.c
+opal/mca/pmix/ext2x/ext2x_component.c
+opal/mca/pmix/ext2x/ext2x_server_north.c
+opal/mca/pmix/ext2x/ext2x_server_south.c
 opal/mca/pmix/pmix*/pmix/include/pmix/autogen/config.h
 opal/mca/pmix/pmix*/pmix/include/pmix/autogen/config.h.in
 opal/mca/pmix/pmix*/pmix/src/include/private/autogen/config.h.in

config/opal_check_libnl.m4

@@ -3,6 +3,7 @@ dnl
 dnl Copyright (c) 2015-2017 Research Organization for Information Science
 dnl                         and Technology (RIST). All rights reserved.
 dnl Copyright (c) 2017      Intel, Inc.  All rights reserved.
+dnl Copyright (c) 2017      Cisco Systems, Inc.  All rights reserved.
 dnl $COPYRIGHT$
 dnl
 dnl Additional copyrights may follow
@@ -19,7 +20,7 @@ dnl libnl-3.so.200 and friends, so if libnl3-devel packages are not
 dnl installed, but libnl-devel are, Open MPI should not try to use libnl.
 dnl
 dnl GROSS: libnl wants us to either use pkg-config (which we
-dnl can't assume is always present) or we need to look in a
+dnl cannot assume is always present) or we need to look in a
 dnl particular directory for the right libnl3 include files.  For
 dnl now, just hard code the special path into this logic.
 dnl
@@ -68,46 +69,128 @@ AC_DEFUN([OPAL_LIBNL_SANITY_INIT], [
     esac
 ])
 
-dnl OPAL_LIBNL_SANITY_CHECK(lib, function, LIBS)
+dnl OPAL_LIBNL_SANITY_FAIL_MSG(lib)
+dnl
+dnl Helper to pring a big warning message when we detect a libnl conflict.
+dnl
+dnl --------------------------------------------------------------------
+AC_DEFUN([OPAL_LIBNL_SANITY_FAIL_MSG], [
+    AC_MSG_WARN([This is a configuration that is *known* to cause run-time crashes.])
+    AC_MSG_WARN([This is an error in lib$1 (not Open MPI).])
+    AC_MSG_WARN([Open MPI will therefore skip using lib$1.])
+])
+
+dnl OPAL_LIBNL_SANITY_CHECK(lib, function, LIBS, libnl_check_ok)
+dnl
+dnl This macro is invoked from OPAL_CHECK_PACKAGE to make sure that
+dnl new libraries that are added to LIBS do not pull in conflicting
+dnl versions of libnl.  E.g., if we already have a library in LIBS
+dnl that pulls in libnl v3, if OPAL_CHECK_PACKAGE is later called that
+dnl pulls in a library that pulls in libnl v1, this macro will detect
+dnl the conflict and will abort configure.
+dnl
+dnl We abort rather than silently ignore this library simply because
+dnl we are now multiple levels deep in the M4 "call stack", and this
+dnl layer does not know the intent of the user.  Hence, all we can do
+dnl is abort with a hopefully helpful error message (that we aborted
+dnl because Open MPI would have been built in a configuration that is
+dnl known to crash).
+dnl
 dnl --------------------------------------------------------------------
 AC_DEFUN([OPAL_LIBNL_SANITY_CHECK], [
+    OPAL_VAR_SCOPE_PUSH([opal_libnl_sane])
+    opal_libnl_sane=1
     case $host in
         *linux*)
-            OPAL_VAR_SCOPE_PUSH([ldd_output libnl_version])
-            AC_LANG_PUSH(C)
-            cat > conftest_c.$ac_ext << EOF
+            OPAL_LIBNL_SANITY_CHECK_LINUX($1, $2, $3, opal_libnl_sane)
+            ;;
+    esac
+
+    $4=$opal_libnl_sane
+    OPAL_VAR_SCOPE_POP([opal_libnl_sane])
+])
+
+dnl
+dnl Simple helper for OPAL_LIBNL_SANITY_CHECK
+dnl $1: library name
+dnl $2: function
+dnl $3: LIBS
+dnl $4: output variable (1=ok, 0=not ok)
+dnl
+AC_DEFUN([OPAL_LIBNL_SANITY_CHECK_LINUX], [
+    OPAL_VAR_SCOPE_PUSH([this_requires_v1 libnl_sane this_requires_v3 ldd_output result_msg])
+
+    AC_LANG_PUSH(C)
+
+    AC_MSG_CHECKING([if lib$1 requires libnl v1 or v3])
+    cat > conftest_c.$ac_ext << EOF
 extern void $2 (void);
 int main(int argc, char *argv[[]]) {
     $2 ();
     return 0;
 }
 EOF
-            OPAL_LOG_COMMAND([$CC -o conftest $CFLAGS $CPPFLAGS conftest_c.$ac_ext $LDFLAGS -l$1 $LIBS $3],
-                             [ldd_output=`ldd conftest`
-                              libnl_version=0
-                              AS_IF([echo $ldd_output | grep -q libnl.so],
-                                    [AS_IF([test $opal_libnl_version -eq 3],
-                                           [AC_MSG_WARN([lib nl version conflict: $opal_libnlv3_libs requires libnl-3 whereas $1 requires libnl])],
-                                           [opal_libnlv1_libs="$opal_libnlv1_libs $1"
-                                            OPAL_UNIQ([opal_libnlv1_libs])
-                                            opal_libnl_version=1])
-                                     libnl_version=1])
-                              AS_IF([echo $ldd_output | grep -q libnl-3.so],
-                                    [AS_IF([test $libnl_version -eq 1],
-                                           [AC_MSG_WARN([lib $1 requires both libnl v1 and libnl v3 -- yoinks!])
-                                            AC_MSG_WARN([This is a configuration that is known to cause run-time crashes])
-                                            AC_MSG_ERROR([Cannot continue])])
-                                     AS_IF([test $opal_libnl_version -eq 1],
-                                           [AC_MSG_WARN([lib nl version conflict: $opal_libnlv1_libs requires libnl whereas $1 requires libnl-3])],
-                                           [opal_libnlv3_libs="$opal_libnlv3_libs $1"
-                                            OPAL_UNIQ([opal_libnlv3_libs])
-                                            opal_libnl_version=3])])
-                              rm -f conftest conftest_c.$ac_ext],
-                             [AC_MSG_WARN([Could not link a simple program with lib $1])])
-            AC_LANG_POP(C)
-            OPAL_VAR_SCOPE_POP([ldd_output libnl_version])
-            ;;
-        esac
+
+    this_requires_v1=0
+    this_requires_v3=0
+    result_msg=
+    OPAL_LOG_COMMAND([$CC -o conftest $CFLAGS $CPPFLAGS conftest_c.$ac_ext $LDFLAGS -l$1 $LIBS $3],
+        [ldd_output=`ldd conftest`
+         AS_IF([echo $ldd_output | grep -q libnl-3.so],
+               [this_requires_v3=1
+                result_msg="v3"])
+         AS_IF([echo $ldd_output | grep -q libnl.so],
+               [this_requires_v1=1
+                result_msg="v1 $result_msg"])
+         AC_MSG_RESULT([$result_msg])
+         ],
+        [AC_MSG_WARN([Could not link a simple program with lib $1])
+        ])
+
+    # Assume that our configuration is sane; this may get reset below
+    libnl_sane=1
+
+    # Note: in all the checks below, only add this library to the list
+    # of libraries (for v1 or v3 as relevant) if we do not fail.
+    # I.e., assume that a higher level will refuse to use this library
+    # if we return failure.
+
+    # Does this library require both v1 and v3?  If so, fail.
+    AS_IF([test $this_requires_v1 -eq 1 && test $this_requires_v3 -eq 1],
+          [AC_MSG_WARN([Unfortunately, lib$1 links to both libnl and libnl-3.])
+           OPAL_LIBNL_SANITY_FAIL_MSG($1)
+           libnl_sane=0])
+
+    # Does this library require v1, but some prior library required
+    # v3?  If so, fail.
+    AS_IF([test $libnl_sane -eq 1 && test $this_requires_v1 -eq 1],
+          [AS_IF([test $opal_libnl_version -eq 3],
+                 [AC_MSG_WARN([libnl version conflict: $opal_libnlv3_libs requires libnl-3 whereas $1 requires libnl])
+                  OPAL_LIBNL_SANITY_FAIL_MSG($1)
+                  libnl_sane=0],
+                 [opal_libnlv1_libs="$opal_libnlv1_libs $1"
+                  OPAL_UNIQ([opal_libnlv1_libs])
+                  opal_libnl_version=1])
+           ])
+
+    # Does this library require v3, but some prior library required
+    # v1?  If so, fail.
+    AS_IF([test $libnl_sane -eq 1 && test $this_requires_v3 -eq 1],
+          [AS_IF([test $opal_libnl_version -eq 1],
+                 [AC_MSG_WARN([libnl version conflict: $opal_libnlv1_libs requires libnl whereas lib$1 requires libnl-3])
+                  OPAL_LIBNL_SANITY_FAIL_MSG($1)
+                  libnl_sane=0],
+                 [opal_libnlv3_libs="$opal_libnlv3_libs $1"
+                  OPAL_UNIQ([opal_libnlv3_libs])
+                  opal_libnl_version=3])
+          ])
+
+    AC_LANG_POP(C)
+    rm -f conftest conftest_c.$ac_ext
+
+    $4=$libnl_sane
+
+    OPAL_VAR_SCOPE_POP([ldd_output libnl_sane this_requires_v1 this_requires_v3 result_msg])
 ])
 
 dnl
@@ -219,7 +302,11 @@ AC_DEFUN([OPAL_CHECK_LIBNL_V3],[
     AS_IF([test $opal_libnlv3_happy -eq 1],
           [$2_LIBS="-lnl-3 -lnl-route-3"
            OPAL_HAVE_LIBNL3=1],
-          [$2_LIBS=""])
+          [# OPAL_CHECK_PACKAGE(...,nl_recvmsgs_report,...) might have set the variables below
+           # so reset them if libnl v3 cannot be used
+           $2_CPPFLAGS=""
+           $2_LDFLAGS=""
+           $2_LIBS=""])
 
    OPAL_VAR_SCOPE_POP
 ])

config/opal_check_package.m4

@@ -10,9 +10,9 @@ dnl Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
 dnl                         University of Stuttgart.  All rights reserved.
 dnl Copyright (c) 2004-2005 The Regents of the University of California.
 dnl                         All rights reserved.
-dnl Copyright (c) 2012-2015 Cisco Systems, Inc.  All rights reserved.
+dnl Copyright (c) 2012-2017 Cisco Systems, Inc.  All rights reserved.
 dnl Copyright (c) 2012      Oracle and/or its affiliates.  All rights reserved.
-dnl Copyright (c) 2014      Intel, Inc. All rights reserved.
+dnl Copyright (c) 2014-2017 Intel, Inc. All rights reserved.
 dnl Copyright (c) 2015-2016 Research Organization for Information Science
 dnl                         and Technology (RIST). All rights reserved.
 dnl $COPYRIGHT$
@@ -132,16 +132,22 @@ AC_DEFUN([_OPAL_CHECK_PACKAGE_LIB], [
                           $1_LDFLAGS="$opal_check_package_$1_orig_LDFLAGS"
                           unset opal_Lib])])])])
 
+    AS_IF([test "$opal_check_package_lib_happy" = "yes"],
+          [ # libnl v1 and libnl3 are known to *not* coexist
+            # harmoniously in the same process.  Check to see if this
+            # new package will introduce such a conflict.
+           OPAL_LIBNL_SANITY_CHECK([$2], [$3], [$$1_LIBS],
+                                   [opal_check_package_libnl_check_ok])
+           AS_IF([test $opal_check_package_libnl_check_ok -eq 0],
+                 [opal_check_package_lib_happy=no])
+           ])
+
     AS_IF([test "$opal_check_package_lib_happy" = "yes"],
           [ # The result of AC SEARCH_LIBS is cached in $ac_cv_search_[function]
            AS_IF([test "$ac_cv_search_$3" != "no" &&
                   test "$ac_cv_search_$3" != "none required"],
                  [$1_LIBS="$ac_cv_search_$3 $4"],
                  [$1_LIBS="$4"])
-            # libnl v1 and libnl3 are known *not* to coexist
-            # for each library, figure out whether it depends on libnl or libnl3 or none
-            # so conflicts can be reported and/or prevented
-            OPAL_LIBNL_SANITY_CHECK([$2], [$3], [$$1_LIBS])
            $7],
           [$8])
 

config/opal_configure_options.m4

@@ -264,8 +264,8 @@ AC_ARG_ENABLE([dlopen],
                      Disabling dlopen implies --disable-mca-dso.
                      (default: enabled)])])
 if test "$enable_dlopen" = "no" ; then
-    enable_mca_dso="no"
-    enable_mca_static="yes"
+    enable_mca_dso=no
+    enable_mca_static=yes
     OPAL_ENABLE_DLOPEN_SUPPORT=0
     AC_MSG_RESULT([no])
 else

config/opal_mca.m4

@@ -11,7 +11,7 @@ dnl                         University of Stuttgart.  All rights reserved.
 dnl Copyright (c) 2004-2005 The Regents of the University of California.
 dnl                         All rights reserved.
 dnl Copyright (c) 2010-2016 Cisco Systems, Inc.  All rights reserved.
-dnl Copyright (c) 2013-2014 Intel, Inc.  All rights reserved.
+dnl Copyright (c) 2013-2017 Intel, Inc. All rights reserved.
 dnl $COPYRIGHT$
 dnl
 dnl Additional copyrights may follow
@@ -173,6 +173,7 @@ AC_DEFUN([OPAL_MCA],[
     elif test "$enable_mca_dso" = "no"; then
         DSO_all=0
         msg=none
+        enable_dlopen=no
     else
         DSO_all=0
         ifs_save="$IFS"

contrib/platform/intel/bend/linux

@@ -18,6 +18,7 @@ enable_cxx_exceptions=no
 enable_mpi_java=no
 enable_io_romio=no
 enable_contrib_no_build=libnbc
+enable_install_libpmix=yes
 with_memory_manager=no
 with_tm=no
 with_psm=no

contrib/scaling/scaling.pl

@@ -17,17 +17,17 @@
 my $useaprun = 0;
 my $useaprun = 0;
 my $myapp;
-my $runall = 0;
+my $runall = 1;
 my $rawoutput = 0;
 my $myresults = "myresults";
 my $ppn = 1;
 my @csvrow;
 
 my @tests = qw(/bin/true ./orte_no_op ./mpi_no_op ./mpi_no_op ./mpi_no_op);
-my @options = ("", "", "", "--fwd-mpirun-port -mca mpi_add_procs_cutoff 0 -mca pmix_base_async_modex 1", "--fwd-mpirun-port -mca mpi_add_procs_cutoff 0 -mca pmix_base_async_modex 1 -mca async_mpi_init 1 -mca async_mpi_finalize 1");
-my @starterlist = qw(mpirun orterun srun aprun);
+my @options = ("", "", "", "-mca mpi_add_procs_cutoff 0 -mca pmix_base_async_modex 1", "-mca mpi_add_procs_cutoff 0 -mca pmix_base_async_modex 1 -mca async_mpi_init 1 -mca async_mpi_finalize 1");
+my @starterlist = qw(mpirun prun srun aprun);
 my @starteroptionlist = ("--novm",
-                         "--hnp file:dvm_uri",
+                         "",
                          "--distribution=cyclic -N",
                          "-N");
 
@@ -49,7 +49,6 @@
     "aprun" => \$useaprun,
     "mpirun" => \$usempirun,
     "myapp=s" => \$myapp,
-    "all" => \$runall,
     "results=s" => \$myresults,
     "rawout" => \$rawoutput,
     "ppn=s" => \$ppn,
@@ -67,8 +66,7 @@
 --srun               Use srun (if available) to execute the test
 --arpun              Use aprun (if available) to execute the test
 --myapp=s            In addition to the standard tests, run this specific application (including any args)
---all                Use all available start commands [default]
---results=file       File where results are to stored in comma-separated value format
+--results=file       File where results are to be stored in comma-separated value format
 --rawout             Provide raw timing output to the file
 --ppn=n              Run n procs/node
 ";
@@ -90,7 +88,13 @@
 my @starters;
 my @starteroptions;
 
-# if they asked for all, then set all starters to requested
+# if they explicitly requested specific starters, then
+# only use those
+if ($useaprun || $usempirun || $usesrun || $usedvm) {
+    $runall = 0
+}
+
+# if they didn't specify something, then set all starters to requested
 if ($runall) {
     $useaprun = 1;
     $usempirun = 1;
@@ -112,7 +116,7 @@
         }
     }
     if ($exists) {
-        if ($usedvm && $starter eq "orterun") {
+        if ($usedvm && $starter eq "prun") {
             push @starters, $starter;
             $opt = $starteroptionlist[$idx] . " --npernode " . $ppn;
             push @starteroptions, $opt;
@@ -261,23 +265,18 @@ ()
 foreach $starter (@starters) {
     print "STARTER: $starter\n";
     # if we are going to use the dvm, then we
-    if ($starter eq "orterun") {
+    if ($starter eq "prun") {
         # need to start it
-        if (-e "dvm_uri") {
-            system("rm -f dvm_uri");
-        }
-        $cmd = "orte-dvm --report-uri dvm_uri 2>&1 &";
+        $cmd = "orte-dvm -mca pmix_system_server 1 2>&1 &";
         if ($myresults) {
             print FILE "\n\n$cmd\n";
         }
         if (!$SHOWME) {
             system($cmd);
-            # wait for the rendezvous file to appear
-            while (! -e "dvm_uri") {
-                sleep(1);
-            }
             $havedvm = 1;
         }
+        # give it a couple of seconds to start
+        sleep 2;
     }
 
     if ($myresults) {
@@ -298,7 +297,11 @@ ()
             $n = 1;
             while ($n <= $num_nodes) {
                 push @csvrow,$n;
-                $cmd = "time " . $starter . " " . $starteroptions[$index] . " $option $test 2>&1";
+                if ($starter eq "prun" or $starter eq "mpirun") {
+                    $cmd = "time " . $starter . " " . $starteroptions[$index] . " $option -n $n $test 2>&1";
+                } else {
+                    $cmd = "time " . $starter . " " . $starteroptions[$index] . " $option -N $n $test 2>&1";
+                }
                 print $cmd . "\n";
                 if (!$SHOWME) {
                     runcmd();
@@ -321,12 +324,9 @@ ()
     }
     if ($havedvm) {
         if (!$SHOWME) {
-            $cmd = "orterun --hnp file:dvm_uri --terminate";
+            $cmd = "prun --terminate";
             system($cmd);
         }
-        if (-e "dvm_uri") {
-            system("rm -f dvm_uri");
-        }
     }
     $index = $index + 1;
 }