int MPID_Get_universe_size(int *universe_size)
{
int mpi_errno = MPI_SUCCESS;
PMICALL(PMI_Get_universe_size(universe_size));
fn_exit:
return mpi_errno;
/* --- */
fn_fail:
goto fn_exit;
}
bool mca_common_pmi_size(int *size) {
#if !WANT_PMI2_SUPPORT
{
int ret;
if (PMI_SUCCESS != (ret = PMI_Get_universe_size(&mca_common_pmi_init_size))) {
OPAL_PMI_ERROR(ret, "PMI_Get_universe_size");
return false;
}
}
#endif
*size = mca_common_pmi_init_size;
return true;
}
/*
* MPID_Get_universe_size - Get the universe size from the process manager
*
* Notes: The ch3 device requires that the PMI routines are used to
* communicate with the process manager. If a channel wishes to
* bypass the standard PMI implementations, it is the responsibility of the
* channel to provide an implementation of the PMI routines.
*/
int MPID_Get_universe_size(int * universe_size)
{
int mpi_errno = MPI_SUCCESS;
#ifdef USE_PMI2_API
char val[PMI2_MAX_VALLEN];
int found = 0;
char *endptr;
mpi_errno = PMI2_Info_GetJobAttr("universeSize", val, sizeof(val), &found);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
if (!found)
*universe_size = MPIR_UNIVERSE_SIZE_NOT_AVAILABLE;
else {
*universe_size = strtol(val, &endptr, 0);
MPIR_ERR_CHKINTERNAL(endptr - val != strlen(val), mpi_errno, "can't parse universe size");
}
#else
int pmi_errno = PMI_SUCCESS;
pmi_errno = PMI_Get_universe_size(universe_size);
if (pmi_errno != PMI_SUCCESS) {
MPIR_ERR_SETANDJUMP1(mpi_errno, MPI_ERR_OTHER,
"**pmi_get_universe_size",
"**pmi_get_universe_size %d", pmi_errno);
}
if (*universe_size < 0)
{
*universe_size = MPIR_UNIVERSE_SIZE_NOT_AVAILABLE;
}
#endif
fn_exit:
return mpi_errno;
/* --BEGIN ERROR HANDLING-- */
fn_fail:
*universe_size = MPIR_UNIVERSE_SIZE_NOT_AVAILABLE;
goto fn_exit;
/* --END ERROR HANDLING-- */
}
static int s1_init(void)
{
PMI_BOOL initialized;
int spawned;
int rc, ret = OPAL_ERROR;
int i, rank, lrank, nrank;
char *pmix_id, tmp[64];
opal_value_t kv;
char *str;
uint32_t ui32;
opal_process_name_t ldr;
char **localranks=NULL;
if (PMI_SUCCESS != (rc = PMI_Initialized(&initialized))) {
OPAL_PMI_ERROR(rc, "PMI_Initialized");
return OPAL_ERROR;
}
if (PMI_TRUE != initialized && PMI_SUCCESS != (rc = PMI_Init(&spawned))) {
OPAL_PMI_ERROR(rc, "PMI_Init");
return OPAL_ERROR;
}
// setup hash table
opal_pmix_base_hash_init();
// Initialize space demands
rc = PMI_KVS_Get_value_length_max(&pmix_vallen_max);
if (PMI_SUCCESS != rc) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_value_length_max");
goto err_exit;
}
pmix_vallen_threshold = pmix_vallen_max * 3;
pmix_vallen_threshold >>= 2;
rc = PMI_KVS_Get_name_length_max(&pmix_kvslen_max);
if (PMI_SUCCESS != rc) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_name_length_max");
goto err_exit;
}
rc = PMI_KVS_Get_key_length_max(&pmix_keylen_max);
if (PMI_SUCCESS != rc) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_key_length_max");
goto err_exit;
}
// Initialize job environment information
pmix_id = (char*)malloc(pmix_vallen_max);
if (pmix_id == NULL) {
ret = OPAL_ERR_OUT_OF_RESOURCE;
goto err_exit;
}
/* Get domain id */
if (PMI_SUCCESS != (rc = PMI_Get_kvs_domain_id(pmix_id, pmix_vallen_max))) {
free(pmix_id);
goto err_exit;
}
/* get our rank */
ret = PMI_Get_rank(&rank);
if( PMI_SUCCESS != ret ) {
OPAL_PMI_ERROR(ret, "PMI_Get_rank");
free(pmix_id);
goto err_exit;
}
/* Slurm PMI provides the job id as an integer followed
* by a '.', followed by essentially a stepid. The first integer
* defines an overall job number. The second integer is the number of
* individual jobs we have run within that allocation. */
s1_pname.jobid = strtoul(pmix_id, &str, 10);
s1_pname.jobid = (s1_pname.jobid << 16) & 0xffff0000;
if (NULL != str) {
ui32 = strtoul(str, NULL, 10);
s1_pname.jobid |= (ui32 & 0x0000ffff);
}
ldr.jobid = s1_pname.jobid;
s1_pname.vpid = rank;
/* store our name in the opal_proc_t so that
* debug messages will make sense - an upper
* layer will eventually overwrite it, but that
* won't do any harm */
opal_proc_set_name(&s1_pname);
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s pmix:s1: assigned tmp name",
OPAL_NAME_PRINT(s1_pname));
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_JOBID);
kv.type = OPAL_UINT32;
kv.data.uint32 = s1_pname.jobid;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* save it */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_RANK);
kv.type = OPAL_UINT32;
kv.data.uint32 = rank;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
pmix_kvs_name = (char*)malloc(pmix_kvslen_max);
if (pmix_kvs_name == NULL) {
ret = OPAL_ERR_OUT_OF_RESOURCE;
goto err_exit;
}
rc = PMI_KVS_Get_my_name(pmix_kvs_name, pmix_kvslen_max);
if (PMI_SUCCESS != rc) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_my_name");
goto err_exit;
}
/* get our local proc info to find our local rank */
if (PMI_SUCCESS != (rc = PMI_Get_clique_size(&nlranks))) {
OPAL_PMI_ERROR(rc, "PMI_Get_clique_size");
return rc;
}
/* save the local size */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCAL_SIZE);
kv.type = OPAL_UINT32;
kv.data.uint32 = nlranks;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
lrank = 0;
nrank = 0;
ldr.vpid = rank;
if (0 < nlranks) {
/* now get the specific ranks */
lranks = (int*)calloc(nlranks, sizeof(int));
if (NULL == lranks) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
OPAL_ERROR_LOG(rc);
return rc;
}
if (PMI_SUCCESS != (rc = PMI_Get_clique_ranks(lranks, nlranks))) {
OPAL_PMI_ERROR(rc, "PMI_Get_clique_ranks");
free(lranks);
return rc;
}
/* note the local ldr */
ldr.vpid = lranks[0];
/* save this */
memset(tmp, 0, 64);
for (i=0; i < nlranks; i++) {
(void)snprintf(tmp, 64, "%d", lranks[i]);
opal_argv_append_nosize(&localranks, tmp);
if (rank == lranks[i]) {
lrank = i;
nrank = i;
}
}
str = opal_argv_join(localranks, ',');
opal_argv_free(localranks);
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCAL_PEERS);
kv.type = OPAL_STRING;
kv.data.string = str;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
}
/* save the local leader */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCALLDR);
kv.type = OPAL_UINT64;
kv.data.uint64 = *(uint64_t*)&ldr;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* save our local rank */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_LOCAL_RANK);
kv.type = OPAL_UINT16;
kv.data.uint16 = lrank;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* and our node rank */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_NODE_RANK);
kv.type = OPAL_UINT16;
kv.data.uint16 = nrank;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* get universe size */
ret = PMI_Get_universe_size(&i);
if (PMI_SUCCESS != ret) {
OPAL_PMI_ERROR(ret, "PMI_Get_universe_size");
goto err_exit;
}
/* push this into the dstore for subsequent fetches */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_UNIV_SIZE);
kv.type = OPAL_UINT32;
kv.data.uint32 = i;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* get job size */
ret = PMI_Get_size(&i);
if (PMI_SUCCESS != ret) {
OPAL_PMI_ERROR(ret, "PMI_Get_size");
goto err_exit;
}
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_JOB_SIZE);
kv.type = OPAL_UINT32;
kv.data.uint32 = i;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* get appnum */
ret = PMI_Get_appnum(&i);
if (PMI_SUCCESS != ret) {
OPAL_PMI_ERROR(ret, "PMI_Get_appnum");
goto err_exit;
}
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_PMIX_APPNUM);
kv.type = OPAL_UINT32;
kv.data.uint32 = i;
if (OPAL_SUCCESS != (ret = opal_pmix_base_store(&OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* increment the init count */
++pmix_init_count;
return OPAL_SUCCESS;
err_exit:
PMI_Finalize();
return ret;
}
static bool cray_get_attr(const char *attr, opal_value_t **kv)
{
int rc, i;
opal_value_t *kp;
if (0 == strcmp(PMIX_JOBID, attr)) {
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = pmix_jobid;
*kv = kp;
return true;
}
if (0 == strcmp(PMIX_RANK, attr)) {
rc = PMI_Get_rank(&i);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_rank");
return false;
}
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = i;
*kv = kp;
return true;
}
if (0 == strcmp(PMIX_UNIV_SIZE, attr)) {
rc = PMI_Get_universe_size(&i);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_universe_size");
return false;
}
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = i;
*kv = kp;
return true;
}
if (0 == strcmp(PMIX_JOB_SIZE, attr)) {
rc = PMI_Get_size(&i);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_size");
return false;
}
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = i;
*kv = kp;
return true;
}
if (0 == strcmp(PMIX_APPNUM, attr)) {
rc = PMI_Get_appnum(&i);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_appnum");
return false;
}
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = i;
*kv = kp;
return true;
}
if (0 == strcmp(PMIX_LOCAL_RANK, attr)) {
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = pmix_lrank;
*kv = kp;
return true;
}
if (0 == strcmp(PMIX_NODE_RANK, attr)) {
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = pmix_nrank;
*kv = kp;
return true;
}
if (0 == strcmp(PMIX_LOCAL_SIZE, attr)) {
kp = OBJ_NEW(opal_value_t);
kp->key = strdup(attr);
kp->type = OPAL_UINT32;
kp->data.uint32 = pmix_nlranks;
*kv = kp;
return true;
}
return OPAL_ERR_NOT_IMPLEMENTED;
}
static int test_item1(void)
{
int rc = 0;
int val = 0;
log_assert(spawned == PMI_FALSE || spawned == PMI_TRUE, "");
if (PMI_SUCCESS != (rc = PMI_Get_size(&size))) {
log_fatal("PMI_Get_Size failed: %d\n", rc);
return rc;
}
log_assert(size >= 0, "");
if (PMI_SUCCESS != (rc = PMI_Get_rank(&rank))) {
log_fatal("PMI_Get_Rank failed: %d\n", rc);
return rc;
}
log_assert(rank >= 0, "");
log_assert(rank < size, "");
if (PMI_SUCCESS != (rc = PMI_Get_appnum(&appnum))) {
log_fatal("PMI_Get_appnum failed: %d\n", rc);
return rc;
}
log_info("spawned=%d size=%d rank=%d appnum=%d\n", spawned, size, rank, appnum);
val = random_value(10, 100);
if (PMI_SUCCESS != (rc = PMI_Get_universe_size(&val))) {
log_fatal("PMI_Get_universe_size failed: %d\n", rc);
return rc;
}
log_assert(size == val, "");
val = random_value(10, 100);
if (PMI_SUCCESS != (rc = PMI_Get_id_length_max(&val))) {
log_fatal("PMI_Get_id_length_max failed: %d\n", rc);
return rc;
}
log_info("PMI_Get_id_length_max=%d\n", val);
if (!_legacy) {
log_assert(sizeof(jobid) == val, "Check PMIX_MAX_NSLEN value in pmix_common.h");
}
sprintf(jobid, "%s", __func__);
if (PMI_SUCCESS != (rc = PMI_Get_id(jobid, sizeof(jobid)))) {
log_fatal("PMI_Get_id failed: %d\n", rc);
return rc;
}
log_info("jobid=%s\n", jobid);
log_assert(memcmp(jobid, __func__, sizeof(__func__)), "");
sprintf(jobid, "%s", __func__);
if (PMI_SUCCESS != (rc = PMI_Get_kvs_domain_id(jobid, sizeof(jobid)))) {
log_fatal("PMI_Get_kvs_domain_id failed: %d\n", rc);
return rc;
}
log_info("PMI_Get_kvs_domain_id=%s\n", jobid);
log_assert(memcmp(jobid, __func__, sizeof(__func__)), "");
sprintf(jobid, "%s", __func__);
if (PMI_SUCCESS != (rc = PMI_KVS_Get_my_name(jobid, sizeof(jobid)))) {
log_fatal("PMI_KVS_Get_my_name failed: %d\n", rc);
return rc;
}
log_info("PMI_KVS_Get_my_name=%s\n", jobid);
log_assert(memcmp(jobid, __func__, sizeof(__func__)), "");
return rc;
}
static int s1_init(void)
{
PMI_BOOL initialized;
int spawned;
int rc, ret = OPAL_ERROR;
int i;
char *pmix_id, *tmp;
uint32_t jobfam, stepid;
opal_value_t kv;
if (PMI_SUCCESS != (rc = PMI_Initialized(&initialized))) {
OPAL_PMI_ERROR(rc, "PMI_Initialized");
return OPAL_ERROR;
}
if( PMI_TRUE != initialized && PMI_SUCCESS != (rc = PMI_Init(&spawned)) ) {
OPAL_PMI_ERROR(rc, "PMI_Init");
return OPAL_ERROR;
}
// Initialize space demands
rc = PMI_KVS_Get_value_length_max(&pmix_vallen_max);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_value_length_max");
goto err_exit;
}
rc = PMI_KVS_Get_name_length_max(&pmix_kvslen_max);
if (PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_name_length_max");
goto err_exit;
}
rc = PMI_KVS_Get_key_length_max(&pmix_keylen_max);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_key_length_max");
goto err_exit;
}
// Initialize job environment information
pmix_id = (char*)malloc(pmix_vallen_max);
if( pmix_id == NULL ){
ret = OPAL_ERR_OUT_OF_RESOURCE;
goto err_exit;
}
/* Get domain id */
if (PMI_SUCCESS != (rc = PMI_Get_kvs_domain_id(pmix_id, pmix_vallen_max))) {
free(pmix_id);
goto err_exit;
}
/* Slurm PMI provides the job id as an integer followed
* by a '.', followed by essentially a stepid. The first integer
* defines an overall job number. The second integer is the number of
* individual jobs we have run within that allocation. So we translate
* this as the overall job number equating to our job family, and
* the individual number equating to our local jobid
*/
jobfam = strtoul(pmix_id, &tmp, 10);
if (NULL == tmp) {
/* hmmm - no '.', so let's just use zero */
stepid = 0;
} else {
tmp++; /* step over the '.' */
stepid = strtoul(tmp, NULL, 10);
}
/* now build the jobid */
s1_jobid = (jobfam << 16) | stepid;
free(pmix_id);
/* get our rank */
ret = PMI_Get_rank(&s1_rank);
if( PMI_SUCCESS != ret ) {
OPAL_PMI_ERROR(ret, "PMI_Get_rank");
goto err_exit;
}
/* store our name in the opal_proc_t so that
* debug messages will make sense - an upper
* layer will eventually overwrite it, but that
* won't do any harm */
s1_pname.jid = s1_jobid;
s1_pname.vid = s1_rank;
opal_proc_set_name((opal_process_name_t*)&s1_pname);
opal_output_verbose(2, opal_pmix_base_framework.framework_output,
"%s pmix:s1: assigned tmp name",
OPAL_NAME_PRINT(*(opal_process_name_t*)&s1_pname));
pmix_kvs_name = (char*)malloc(pmix_kvslen_max);
if( pmix_kvs_name == NULL ){
ret = OPAL_ERR_OUT_OF_RESOURCE;
goto err_exit;
}
rc = PMI_KVS_Get_my_name(pmix_kvs_name, pmix_kvslen_max);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_my_name");
goto err_exit;
}
/* get our local proc info to find our local rank */
if (PMI_SUCCESS != (rc = PMI_Get_clique_size(&s1_nlranks))) {
OPAL_PMI_ERROR(rc, "PMI_Get_clique_size");
return rc;
}
/* now get the specific ranks */
s1_lranks = (int*)calloc(s1_nlranks, sizeof(int));
if (NULL == s1_lranks) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
OPAL_ERROR_LOG(rc);
return rc;
}
if (PMI_SUCCESS != (rc = PMI_Get_clique_ranks(s1_lranks, s1_nlranks))) {
OPAL_PMI_ERROR(rc, "PMI_Get_clique_ranks");
free(s1_lranks);
return rc;
}
/* find ourselves */
for (i=0; i < s1_nlranks; i++) {
if (s1_rank == s1_lranks[i]) {
s1_lrank = i;
s1_nrank = i;
break;
}
}
/* get universe size */
ret = PMI_Get_universe_size(&s1_usize);
if (PMI_SUCCESS != ret) {
OPAL_PMI_ERROR(ret, "PMI_Get_universe_size");
goto err_exit;
}
/* push this into the dstore for subsequent fetches */
OBJ_CONSTRUCT(&kv, opal_value_t);
kv.key = strdup(OPAL_DSTORE_UNIV_SIZE);
kv.type = OPAL_UINT32;
kv.data.uint32 = s1_usize;
if (OPAL_SUCCESS != (ret = opal_dstore.store(opal_dstore_internal, &OPAL_PROC_MY_NAME, &kv))) {
OPAL_ERROR_LOG(ret);
OBJ_DESTRUCT(&kv);
goto err_exit;
}
OBJ_DESTRUCT(&kv);
/* get job size */
ret = PMI_Get_size(&s1_jsize);
if (PMI_SUCCESS != ret) {
OPAL_PMI_ERROR(ret, "PMI_Get_size");
goto err_exit;
}
/* get appnum */
ret = PMI_Get_appnum(&s1_appnum);
if (PMI_SUCCESS != ret) {
OPAL_PMI_ERROR(ret, "PMI_Get_appnum");
goto err_exit;
}
return OPAL_SUCCESS;
err_exit:
PMI_Finalize();
return ret;
}
static int rte_init(void)
{
int ret, i, j;
char *error = NULL, *localj;
int32_t jobfam, stepid;
char *envar, *ev1, *ev2;
uint64_t unique_key[2];
char *cs_env, *string_key;
char *pmi_id=NULL;
int *ranks;
char *tmp;
orte_jobid_t jobid;
orte_process_name_t proc;
orte_local_rank_t local_rank;
orte_node_rank_t node_rank;
/* run the prolog */
if (ORTE_SUCCESS != (ret = orte_ess_base_std_prolog())) {
error = "orte_ess_base_std_prolog";
goto error;
}
#if OPAL_HAVE_HWLOC
/* get the topology */
if (NULL == opal_hwloc_topology) {
if (OPAL_SUCCESS != opal_hwloc_base_get_topology()) {
error = "topology discovery";
goto error;
}
}
#endif
if (ORTE_PROC_IS_DAEMON) { /* I am a daemon, launched by mpirun */
/* we had to be given a jobid */
mca_base_param_reg_string_name("orte", "ess_jobid", "Process jobid",
true, false, NULL, &tmp);
if (NULL == tmp) {
error = "missing jobid";
ret = ORTE_ERR_FATAL;
goto error;
}
if (ORTE_SUCCESS != (ret = orte_util_convert_string_to_jobid(&jobid, tmp))) {
ORTE_ERROR_LOG(ret);
error = "convert jobid";
goto error;
}
free(tmp);
ORTE_PROC_MY_NAME->jobid = jobid;
/* get our rank from PMI */
if (PMI_SUCCESS != (ret = PMI_Get_rank(&i))) {
ORTE_PMI_ERROR(ret, "PMI_Get_rank");
error = "could not get PMI rank";
goto error;
}
ORTE_PROC_MY_NAME->vpid = i + 1; /* compensate for orterun */
/* get the number of procs from PMI */
if (PMI_SUCCESS != (ret = PMI_Get_universe_size(&i))) {
ORTE_PMI_ERROR(ret, "PMI_Get_universe_size");
error = "could not get PMI universe size";
goto error;
}
orte_process_info.num_procs = i + 1; /* compensate for orterun */
/* complete setup */
if (ORTE_SUCCESS != (ret = orte_ess_base_orted_setup(NULL))) {
ORTE_ERROR_LOG(ret);
error = "orte_ess_base_orted_setup";
goto error;
}
} else { /* we are a direct-launched MPI process */
/* get our PMI id length */
if (PMI_SUCCESS != (ret = PMI_Get_id_length_max(&pmi_maxlen))) {
error = "PMI_Get_id_length_max";
goto error;
}
pmi_id = malloc(pmi_maxlen);
if (PMI_SUCCESS != (ret = PMI_Get_kvs_domain_id(pmi_id, pmi_maxlen))) {
free(pmi_id);
error = "PMI_Get_kvs_domain_id";
goto error;
}
/* PMI is very nice to us - the domain id is an integer followed
* by a '.', followed by essentially a stepid. The first integer
* defines an overall job number. The second integer is the number of
* individual jobs we have run within that allocation. So we translate
* this as the overall job number equating to our job family, and
* the individual number equating to our local jobid
*/
jobfam = strtol(pmi_id, &localj, 10);
if (NULL == localj) {
/* hmmm - no '.', so let's just use zero */
stepid = 0;
} else {
localj++; /* step over the '.' */
stepid = strtol(localj, NULL, 10) + 1; /* add one to avoid looking like a daemon */
}
free(pmi_id);
/* now build the jobid */
ORTE_PROC_MY_NAME->jobid = ORTE_CONSTRUCT_LOCAL_JOBID(jobfam << 16, stepid);
/* get our rank */
if (PMI_SUCCESS != (ret = PMI_Get_rank(&i))) {
ORTE_PMI_ERROR(ret, "PMI_Get_rank");
error = "could not get PMI rank";
goto error;
}
ORTE_PROC_MY_NAME->vpid = i;
/* get the number of procs from PMI */
if (PMI_SUCCESS != (ret = PMI_Get_universe_size(&i))) {
ORTE_PMI_ERROR(ret, "PMI_Get_universe_size");
error = "could not get PMI universe size";
goto error;
}
orte_process_info.num_procs = i;
/* push into the environ for pickup in MPI layer for
* MPI-3 required info key
*/
asprintf(&ev1, "OMPI_MCA_orte_ess_num_procs=%d", i);
putenv(ev1);
asprintf(&ev2, "OMPI_APP_CTX_NUM_PROCS=%d", i);
putenv(ev2);
/* setup transport keys in case the MPI layer needs them -
* we can use the jobfam and stepid as unique keys
* because they are unique values assigned by the RM
*/
unique_key[0] = (uint64_t)jobfam;
unique_key[1] = (uint64_t)stepid;
if (NULL == (string_key = orte_pre_condition_transports_print(unique_key))) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
if (NULL == (cs_env = mca_base_param_environ_variable("orte_precondition_transports",NULL,NULL))) {
ORTE_ERROR_LOG(ORTE_ERR_OUT_OF_RESOURCE);
return ORTE_ERR_OUT_OF_RESOURCE;
}
asprintf(&envar, "%s=%s", cs_env, string_key);
putenv(envar);
/* cannot free the envar as that messes up our environ */
free(cs_env);
free(string_key);
/* our app_context number can only be 0 as we don't support
* dynamic spawns
*/
orte_process_info.app_num = 0;
/* setup my daemon's name - arbitrary, since we don't route
* messages
*/
ORTE_PROC_MY_DAEMON->jobid = 0;
ORTE_PROC_MY_DAEMON->vpid = 0;
/* ensure we pick the correct critical components */
putenv("OMPI_MCA_grpcomm=pmi");
putenv("OMPI_MCA_routed=direct");
/* now use the default procedure to finish my setup */
if (ORTE_SUCCESS != (ret = orte_ess_base_app_setup())) {
ORTE_ERROR_LOG(ret);
error = "orte_ess_base_app_setup";
goto error;
}
/* store our info into the database */
if (ORTE_SUCCESS != (ret = orte_db.store(ORTE_PROC_MY_NAME, ORTE_DB_HOSTNAME, orte_process_info.nodename, OPAL_STRING))) {
error = "db store daemon vpid";
goto error;
}
/* get our local proc info to find our local rank */
if (PMI_SUCCESS != (ret = PMI_Get_clique_size(&i))) {
ORTE_PMI_ERROR(ret, "PMI_Get_clique_size");
error = "could not get PMI clique size";
goto error;
}
/* store that info - remember, we want the number of peers that
* share the node WITH ME, so we have to subtract ourselves from
* that number
*/
orte_process_info.num_local_peers = i - 1;
/* now get the specific ranks */
ranks = (int*)malloc(i * sizeof(int));
if (PMI_SUCCESS != (ret = PMI_Get_clique_ranks(ranks, i))) {
ORTE_PMI_ERROR(ret, "PMI_Get_clique_ranks");
error = "could not get clique ranks";
goto error;
}
/* The clique ranks are returned in rank order, so
* cycle thru the array and update the local/node
* rank info
*/
proc.jobid = ORTE_PROC_MY_NAME->jobid;
for (j=0; j < i; j++) {
proc.vpid = ranks[j];
local_rank = j;
node_rank = j;
if (ranks[j] == (int)ORTE_PROC_MY_NAME->vpid) {
orte_process_info.my_local_rank = local_rank;
orte_process_info.my_node_rank = node_rank;
}
if (ORTE_SUCCESS != (ret = orte_db.store(&proc, ORTE_DB_LOCALRANK, &local_rank, ORTE_LOCAL_RANK))) {
error = "db store local rank";
goto error;
}
if (ORTE_SUCCESS != (ret = orte_db.store(&proc, ORTE_DB_NODERANK, &node_rank, ORTE_NODE_RANK))) {
error = "db store node rank";
goto error;
}
}
free(ranks);
/* setup process binding */
if (ORTE_SUCCESS != (ret = orte_ess_base_proc_binding())) {
error = "proc_binding";
goto error;
}
}
/* set max procs */
if (orte_process_info.max_procs < orte_process_info.num_procs) {
orte_process_info.max_procs = orte_process_info.num_procs;
}
/* flag that we completed init */
app_init_complete = true;
return ORTE_SUCCESS;
error:
if (ORTE_ERR_SILENT != ret && !orte_report_silent_errors) {
orte_show_help("help-orte-runtime.txt",
"orte_init:startup:internal-failure",
true, error, ORTE_ERROR_NAME(ret), ret);
}
return ret;
}
static int mca_initialize_pmi_v1(void)
{
PMI_BOOL initialized;
int spawned;
int rc, ret = OPAL_ERROR;
if (PMI_SUCCESS != (rc = PMI_Initialized(&initialized))) {
OPAL_PMI_ERROR(rc, "PMI_Initialized");
return OPAL_ERROR;
}
if( PMI_TRUE != initialized && PMI_SUCCESS != (rc = PMI_Init(&spawned)) ) {
OPAL_PMI_ERROR(rc, "PMI_Init");
return OPAL_ERROR;
}
// Initialize space demands
rc = PMI_KVS_Get_value_length_max(&pmi_vallen_max);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_value_length_max");
goto err_exit;
}
rc = PMI_KVS_Get_name_length_max(&pmi_kvslen_max);
if (PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_name_length_max");
goto err_exit;
}
rc = PMI_KVS_Get_key_length_max(&pmi_keylen_max);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_KVS_Get_key_length_max");
goto err_exit;
}
// Initialize job environment information
rc = PMI_Get_rank(&pmi_rank);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_rank");
return OPAL_ERROR;
}
rc = PMI_Get_universe_size(&pmi_usize);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_universe_size");
goto err_exit;
}
rc = PMI_Get_size(&pmi_size);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_size");
goto err_exit;
}
rc = PMI_Get_appnum(&pmi_appnum);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI_Get_appnum");
goto err_exit;
}
pmi_kvs_name = (char*)malloc(pmi_kvslen_max);
if( pmi_kvs_name == NULL ){
ret = OPAL_ERR_OUT_OF_RESOURCE;
goto err_exit;
}
rc = PMI_KVS_Get_my_name(pmi_kvs_name,pmi_kvslen_max);
if( PMI_SUCCESS != rc ) {
OPAL_PMI_ERROR(rc, "PMI2_Job_GetId");
goto err_exit;
}
return OPAL_SUCCESS;
err_exit:
PMI_Finalize();
return ret;
}