Merge pull request #7566 from rhc54/topic/up2

Update PMIx and PRRTE
2020-03-31 10:29:19 -07:00 · 2020-03-31 10:29:19 -07:00 · 538d2de860
--- a/config/ompi_setup_prrte.m4
+++ b/config/ompi_setup_prrte.m4
@ -24,7 +24,7 @@
 #

 AC_DEFUN([OMPI_SETUP_PRRTE],[
-    OPAL_VAR_SCOPE_PUSH([opal_prrte_save_CPPFLAGS opal_prrte_save_CFLAGS opal_prrte_save_LDFLAGS opal_prrte_save_LIBS opal_prrte_args opal_prrte_save_enable_dlopen opal_prrte_save_enable_mca_dso opal_prrte_save_enable_mca_static])
+    OPAL_VAR_SCOPE_PUSH([opal_prrte_save_CPPFLAGS opal_prrte_save_CFLAGS opal_prrte_save_LDFLAGS opal_prrte_save_LIBS opal_prrte_args opal_prrte_save_enable_dlopen opal_prrte_save_enable_mca_dso opal_prrte_save_enable_mca_static opal_prrte_extra_libs opal_prrte_extra_ltlibs opal_prrte_extra_ldflags])

    opal_prrte_save_CFLAGS=$CFLAGS
    opal_prrte_save_CPPFLAGS=$CPPFLAGS
@ -59,29 +59,29 @@ AC_DEFUN([OMPI_SETUP_PRRTE],[
    if test "$enable_internal_rte" != "no"; then
        AC_MSG_RESULT([yes])
        ompi_want_prrte=yes
-        if test -z $with_libevent || test "$with_libevent" = "internal" || test "$with_libevent" = "yes"; then
+        opal_prrte_extra_libs=$OMPI_TOP_BUILDDIR/opal/libopen-pal.la
+        opal_prrte_extra_ltlibs=$OMPI_TOP_BUILDDIR/opal/libopen-pal.la
+
+        if test "$opal_event_external_support" = "yes"; then
+            opal_prrte_extra_libs="$opal_prrte_extra_libs $opal_event_external_LIBS"
+            opal_prrte_extra_ltlibs="$opal_prrte_extra_ltlibs $opal_event_external_LIBS"
+        fi
+        # specifying --with-libevent-header causes prrte to ignore the with_libevent and with_libevent_libdir options
        opal_prrte_libevent_arg="--with-libevent-header=$OMPI_TOP_SRCDIR/opal/mca/event/event.h"
-        elif test "$with_libevent" = "external"; then
-            opal_prrte_libevent_arg=""
-        else
-            opal_prrte_libevent_arg="--with-libevent=$with_libevent"
-        fi

-        if test -z $with_hwloc || test "$with_hwloc" = "internal" || test "$with_hwloc" = "yes"; then
+        if test "$opal_hwloc_external_support" = "yes"; then
+            opal_prrte_extra_libs="$opal_prrte_extra_libs $opal_hwloc_external_LIBS"
+            opal_prrte_extra_ltlibs="$opal_prrte_extra_ltlibs $opal_hwloc_external_LIBS"
+        fi
+        # specifying --with-hwloc-header causes prrte to ignore the with_hwloc and with_hwloc_libdir options
        opal_prrte_hwloc_arg="--with-hwloc-header=$OMPI_TOP_SRCDIR/opal/mca/hwloc/hwloc-internal.h"
-        elif test "$with_hwloc" = "external"; then
-            opal_prrte_hwloc_arg=""
-        else
-            opal_prrte_hwloc_arg="--with-hwloc=$with_hwloc"
-        fi

-        if test -z $with_pmix || test "$with_pmix" = "internal" || test "$with_pmix" = "yes"; then
-            opal_prrte_pmix_arg="--with-pmix-header=$OMPI_TOP_SRCDIR/opal/mca/pmix/pmix-internal.h"
-        elif test "$with_pmix" = "external"; then
-            opal_prrte_pmix_arg=""
-        else
-            opal_prrte_pmix_arg="--with-pmix=$with_pmix"
+        if test "$opal_external_pmix_happy" = "yes"; then
+            opal_prrte_extra_libs="$opal_prrte_extra_libs $opal_pmix_external_LIBS"
+            opal_prrte_extra_ltlibs="$opal_prrte_extra_ltlibs $opal_pmix_external_LIBS"
        fi
+        # specifying --with-pmix-header causes prrte to ignore the with_pmix and with_pmix_libdir options
+        opal_prrte_pmix_arg="--with-pmix-header=$OMPI_TOP_SRCDIR/opal/mca/pmix/pmix-internal.h"

        if test -z $enable_prte_prefix_by_default || test "$enable_prte_prefix_by_default" = "yes" ||
           test "$enable_orterun_prefix_given" = "yes"; then
@ -102,7 +102,7 @@ AC_DEFUN([OMPI_SETUP_PRRTE],[
            opal_prrte_args="$opal_prrte_args --with-platform=$with_prrte_platform"
        fi
        # add the extra libs
-        opal_prrte_args="$opal_prrte_args --with-prrte-extra-lib=$OMPI_TOP_BUILDDIR/opal/libopen-pal.la --with-prrte-extra-ltlib=$OMPI_TOP_BUILDDIR/opal/libopen-pal.la"
+        opal_prrte_args="$opal_prrte_args --with-prrte-extra-lib=\"$opal_prrte_extra_libs\" --with-prrte-extra-ltlib=\"$opal_prrte_extra_ltlibs\""

        AC_MSG_CHECKING([final prrte configure args])
        AC_MSG_RESULT([$opal_prrte_args])
--- a/opal/mca/pmix/base/pmix_base_fns.c
+++ b/opal/mca/pmix/base/pmix_base_fns.c
@ -110,10 +110,6 @@ int opal_pmix_convert_jobid(pmix_nspace_t nspace, opal_jobid_t jobid)
    /* zero out the nspace */
    PMIX_LOAD_NSPACE(nspace, NULL);

-    if (opal_process_info.nativelaunch) {
-        opal_snprintf_jobid(nspace, PMIX_MAX_NSLEN, jobid);
-        return OPAL_SUCCESS;
-    } else {
    /* cycle across our list of known jobids */
    OPAL_LIST_FOREACH(nptr, &localnspaces, opal_nptr_t) {
        if (jobid == nptr->jobid) {
@ -121,7 +117,7 @@ int opal_pmix_convert_jobid(pmix_nspace_t nspace, opal_jobid_t jobid)
            return OPAL_SUCCESS;
        }
    }
-    }
+
    return OPAL_ERR_NOT_FOUND;
 }

@ -129,13 +125,26 @@ int opal_pmix_convert_nspace(opal_jobid_t *jobid, pmix_nspace_t nspace)
 {
    opal_nptr_t *nptr;
    opal_jobid_t jid;
+    uint16_t jobfam;
+    uint32_t hash32, localjob = 0;
+    char *p = NULL;

    /* set a default */
    *jobid = OPAL_JOBID_INVALID;

-    if (opal_process_info.nativelaunch) {
-        return opal_convert_string_to_jobid(jobid, nspace);
-    } else {
+    /* if the nspace is empty, there is nothing more to do */
+    if (0 == strlen(nspace)) {
+        return OPAL_SUCCESS;
+    }
+    if (NULL != strstr(nspace, "JOBID_WILDCARD")) {
+        *jobid = OPAL_JOBID_WILDCARD;
+        return OPAL_SUCCESS;
+    }
+    if (NULL != strstr(nspace, "JOBID_INVALID")) {
+        *jobid = OPAL_JOBID_INVALID;
+        return OPAL_SUCCESS;
+    }
+
    /* cycle across our list of known jobids */
    OPAL_LIST_FOREACH(nptr, &localnspaces, opal_nptr_t) {
        if (PMIX_CHECK_NSPACE(nspace, nptr->nspace)) {
@ -144,14 +153,27 @@ int opal_pmix_convert_nspace(opal_jobid_t *jobid, pmix_nspace_t nspace)
        }
    }
    /* if we get here, we don't know this nspace */
-        OPAL_HASH_STR(nspace, jid);
-        jid &= ~(0x8000);
+    /* find the "." at the end that indicates the child job */
+    if (NULL != (p = strrchr(nspace, '.'))) {
+        *p = '\0';
+    }
+    OPAL_HASH_STR(nspace, hash32);
+    if (NULL != p) {
+        *p = '.';
+        ++p;
+        localjob = strtoul(p, NULL, 10);
+    }
+
+    /* now compress to 16-bits */
+    jobfam = (uint16_t)(((0x0000ffff & (0xffff0000 & hash32) >> 16)) ^ (0x0000ffff & hash32));
+    jid = (0xffff0000 & ((uint32_t)jobfam << 16)) | (0x0000ffff & localjob);
    *jobid = jid;
+    /* save this jobid/nspace pair */
    nptr = OBJ_NEW(opal_nptr_t);
    nptr->jobid = jid;
    PMIX_LOAD_NSPACE(nptr->nspace, nspace);
    opal_list_append(&localnspaces, &nptr->super);
-    }
+
    return OPAL_SUCCESS;
 }

--- a/opal/mca/pmix/pmix-internal.h
+++ b/opal/mca/pmix/pmix-internal.h
@ -599,9 +599,11 @@ OPAL_DECLSPEC int opal_pmix_convert_nspace(opal_jobid_t *jobid, pmix_nspace_t ns
 OPAL_DECLSPEC void opal_pmix_setup_nspace_tracker(void);
 OPAL_DECLSPEC void opal_pmix_finalize_nspace_tracker(void);

+/* convert jobid to nspace */
 #define OPAL_PMIX_CONVERT_JOBID(n, j) \
    opal_pmix_convert_jobid((n), (j))

+/* convert vpid to rank */
 #define OPAL_PMIX_CONVERT_VPID(r, v)        \
    do {                                    \
        if (OPAL_VPID_WILDCARD == (v)) {    \
@ -611,6 +613,7 @@ OPAL_DECLSPEC void opal_pmix_finalize_nspace_tracker(void);
        }                                   \
    } while(0)

+/* convert opal_process_name_t to pmix_proc_t */
 #define OPAL_PMIX_CONVERT_NAME(p, n)                        \
    do {                                                    \
        OPAL_PMIX_CONVERT_JOBID((p)->nspace, (n)->jobid);   \
@ -618,9 +621,11 @@ OPAL_DECLSPEC void opal_pmix_finalize_nspace_tracker(void);
    } while(0)


+/* convert nspace to jobid */
 #define OPAL_PMIX_CONVERT_NSPACE(r, j, n)       \
    (r) = opal_pmix_convert_nspace((j), (n))

+/* convert pmix rank to opal vpid */
 #define OPAL_PMIX_CONVERT_RANK(v, r)            \
    do {                                        \
        if (PMIX_RANK_WILDCARD == (r)) {        \
@ -632,6 +637,7 @@ OPAL_DECLSPEC void opal_pmix_finalize_nspace_tracker(void);
        }                                       \
    } while(0)

+/* convert pmix_proc_t to opal_process_name_t */
 #define OPAL_PMIX_CONVERT_PROCT(r, n, p)                            \
    do {                                                            \
        OPAL_PMIX_CONVERT_NSPACE((r), &(n)->jobid, (p)->nspace);    \
--- a/opal/mca/pmix/pmix4x/openpmix
+++ b/opal/mca/pmix/pmix4x/openpmix
@ -1 +1 @@
-Subproject commit a18e53138298d61a01fec4471518140304539e8c
+Subproject commit 8c565209c21f93d11e2156c0d53d73c3f6f9aaab
--- a/2
+++ b/2
@ -1 +1 @@
-Subproject commit cdea5231171b2fdea11269033de9e265fc7f3a63
+Subproject commit d879d5669379ffbe093d60b76b46cc9b2aae20e1