int orte_rmaps_base_compute_bindings(orte_job_t *jdata)
{
hwloc_obj_type_t hwb, hwm;
unsigned clvl=0, clvm=0;
opal_binding_policy_t bind;
orte_mapping_policy_t map;
orte_node_t *node;
int i, rc;
struct hwloc_topology_support *support;
bool force_down = false;
hwloc_cpuset_t totalcpuset;
int bind_depth, map_depth;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: compute bindings for job %s with policy %s",
ORTE_JOBID_PRINT(jdata->jobid),
opal_hwloc_base_print_binding(jdata->map->binding));
map = ORTE_GET_MAPPING_POLICY(jdata->map->mapping);
bind = OPAL_GET_BINDING_POLICY(jdata->map->binding);
if (ORTE_MAPPING_BYUSER == map) {
/* user specified binding by rankfile - nothing for us to do */
return ORTE_SUCCESS;
}
if (OPAL_BIND_TO_CPUSET == bind) {
int rc;
/* cpuset was given - setup the bindings */
if (ORTE_SUCCESS != (rc = bind_to_cpuset(jdata))) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
if (OPAL_BIND_TO_NONE == bind) {
/* no binding requested */
return ORTE_SUCCESS;
}
if (OPAL_BIND_TO_BOARD == bind) {
/* doesn't do anything at this time */
return ORTE_SUCCESS;
}
/* binding requested - convert the binding level to the hwloc obj type */
switch (bind) {
case OPAL_BIND_TO_NUMA:
hwb = HWLOC_OBJ_NODE;
break;
case OPAL_BIND_TO_SOCKET:
hwb = HWLOC_OBJ_SOCKET;
break;
case OPAL_BIND_TO_L3CACHE:
hwb = HWLOC_OBJ_CACHE;
clvl = 3;
break;
case OPAL_BIND_TO_L2CACHE:
hwb = HWLOC_OBJ_CACHE;
clvl = 2;
break;
case OPAL_BIND_TO_L1CACHE:
hwb = HWLOC_OBJ_CACHE;
clvl = 1;
break;
case OPAL_BIND_TO_CORE:
hwb = HWLOC_OBJ_CORE;
break;
case OPAL_BIND_TO_HWTHREAD:
hwb = HWLOC_OBJ_PU;
break;
default:
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
return ORTE_ERR_BAD_PARAM;
}
/* do the same for the mapping policy */
switch (map) {
case ORTE_MAPPING_BYNODE:
case ORTE_MAPPING_BYSLOT:
case ORTE_MAPPING_SEQ:
hwm = HWLOC_OBJ_MACHINE;
break;
case ORTE_MAPPING_BYDIST:
case ORTE_MAPPING_BYNUMA:
hwm = HWLOC_OBJ_NODE;
break;
case ORTE_MAPPING_BYSOCKET:
hwm = HWLOC_OBJ_SOCKET;
break;
case ORTE_MAPPING_BYL3CACHE:
hwm = HWLOC_OBJ_CACHE;
clvm = 3;
break;
case ORTE_MAPPING_BYL2CACHE:
hwm = HWLOC_OBJ_CACHE;
clvm = 2;
break;
case ORTE_MAPPING_BYL1CACHE:
hwm = HWLOC_OBJ_CACHE;
clvm = 1;
break;
case ORTE_MAPPING_BYCORE:
hwm = HWLOC_OBJ_CORE;
break;
case ORTE_MAPPING_BYHWTHREAD:
hwm = HWLOC_OBJ_PU;
break;
default:
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
return ORTE_ERR_BAD_PARAM;
}
/* if the job was mapped by the corresponding target, then
* we bind in place
*
* otherwise, we have to bind either up or down the hwloc
* tree. If we are binding upwards (e.g., mapped to hwthread
* but binding to core), then we just climb the tree to find
* the first matching object.
*
* if we are binding downwards (e.g., mapped to node and bind
* to core), then we have to do a round-robin assigment of
* procs to the resources below.
*/
if (ORTE_MAPPING_BYDIST == map) {
int rc = ORTE_SUCCESS;
if (OPAL_BIND_TO_NUMA == bind) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: bindings for job %s - dist to numa",
ORTE_JOBID_PRINT(jdata->jobid));
if (ORTE_SUCCESS != (rc = bind_in_place(jdata, HWLOC_OBJ_NODE, 0))) {
ORTE_ERROR_LOG(rc);
}
} else if (OPAL_BIND_TO_NUMA < bind) {
/* bind every proc downwards */
force_down = true;
goto execute;
}
/* if the binding policy is less than numa, then we are unbound - so
* just ignore this and return (should have been caught in prior
* tests anyway as only options meeting that criteria are "none"
* and "board")
*/
return rc;
}
/* now deal with the remaining binding policies based on hardware */
if (bind == map) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: bindings for job %s - bind in place",
ORTE_JOBID_PRINT(jdata->jobid));
if (ORTE_SUCCESS != (rc = bind_in_place(jdata, hwb, clvl))) {
ORTE_ERROR_LOG(rc);
}
return rc;
}
/* we need to handle the remaining binding options on a per-node
* basis because different nodes could potentially have different
* topologies, with different relative depths for the two levels
*/
execute:
/* initialize */
totalcpuset = hwloc_bitmap_alloc();
for (i=0; i < jdata->map->nodes->size; i++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(jdata->map->nodes, i))) {
continue;
}
if (!orte_do_not_launch) {
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
* interprets the current behavior. So check both flags to be sure.
*/
if (!support->cpubind->set_thisproc_cpubind &&
!support->cpubind->set_thisthread_cpubind) {
if (!OPAL_BINDING_REQUIRED(jdata->map->binding) ||
!OPAL_BINDING_POLICY_IS_SET(jdata->map->binding)) {
/* we are not required to bind, so ignore this */
continue;
}
orte_show_help("help-orte-rmaps-base.txt", "rmaps:cpubind-not-supported", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
/* check if topology supports membind - have to be careful here
* as hwloc treats this differently than I (at least) would have
* expected. Per hwloc, Linux memory binding is at the thread,
* and not process, level. Thus, hwloc sets the "thisproc" flag
* to "false" on all Linux systems, and uses the "thisthread" flag
* to indicate binding capability - don't warn if the user didn't
* specifically request binding
*/
if (!support->membind->set_thisproc_membind &&
!support->membind->set_thisthread_membind &&
OPAL_BINDING_POLICY_IS_SET(jdata->map->binding)) {
if (OPAL_HWLOC_BASE_MBFA_WARN == opal_hwloc_base_mbfa && !membind_warned) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported", true, node->name);
membind_warned = true;
} else if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported-fatal", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
}
}
/* some systems do not report cores, and so we can get a situation where our
* default binding policy will fail for no necessary reason. So if we are
* computing a binding due to our default policy, and no cores are found
* on this node, just silently skip it - we will not bind
*/
if (!OPAL_BINDING_POLICY_IS_SET(jdata->map->binding) &&
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology, HWLOC_OBJ_CORE)) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"Unable to bind-to core by default on node %s as no cores detected",
node->name);
continue;
}
/* we share topologies in order
* to save space, so we need to reset the usage info to reflect
* our own current state
*/
reset_usage(node, jdata->jobid);
if (force_down) {
if (ORTE_SUCCESS != (rc = bind_downwards(jdata, node, hwb, clvl))) {
ORTE_ERROR_LOG(rc);
return rc;
}
} else {
/* determine the relative depth on this node */
if (HWLOC_OBJ_CACHE == hwb) {
/* must use a unique function because blasted hwloc
* just doesn't deal with caches very well...sigh
*/
bind_depth = hwloc_get_cache_type_depth(node->topology, clvl, -1);
} else {
bind_depth = hwloc_get_type_depth(node->topology, hwb);
}
if (0 > bind_depth) {
/* didn't find such an object */
orte_show_help("help-orte-rmaps-base.txt", "orte-rmaps-base:no-objects",
true, hwloc_obj_type_string(hwb), node->name);
return ORTE_ERR_SILENT;
}
if (HWLOC_OBJ_CACHE == hwm) {
/* must use a unique function because blasted hwloc
* just doesn't deal with caches very well...sigh
*/
map_depth = hwloc_get_cache_type_depth(node->topology, clvm, -1);
} else {
map_depth = hwloc_get_type_depth(node->topology, hwm);
}
if (0 > map_depth) {
/* didn't find such an object */
orte_show_help("help-orte-rmaps-base.txt", "orte-rmaps-base:no-objects",
true, hwloc_obj_type_string(hwm), node->name);
return ORTE_ERR_SILENT;
}
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s bind_depth: %d map_depth %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
bind_depth, map_depth);
if (bind_depth > map_depth) {
if (ORTE_SUCCESS != (rc = bind_downwards(jdata, node, hwb, clvl))) {
ORTE_ERROR_LOG(rc);
return rc;
}
} else {
if (ORTE_SUCCESS != (rc = bind_upwards(jdata, node, hwb, clvl))) {
ORTE_ERROR_LOG(rc);
return rc;
}
}
}
}
return ORTE_SUCCESS;
}
static int bind_in_place(orte_job_t *jdata,
hwloc_obj_type_t target,
unsigned cache_level)
{
/* traverse the hwloc topology tree on each node downwards
* until we find an unused object of type target - and then bind
* the process to that target
*/
int i, j;
orte_job_map_t *map;
orte_node_t *node;
orte_proc_t *proc;
hwloc_cpuset_t cpus;
unsigned int idx, ncpus;
struct hwloc_topology_support *support;
opal_hwloc_obj_data_t *data;
hwloc_obj_t locale, sib;
char *cpu_bitmap;
bool found;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: bind in place for job %s with bindings %s",
ORTE_JOBID_PRINT(jdata->jobid),
opal_hwloc_base_print_binding(jdata->map->binding));
/* initialize */
map = jdata->map;
for (i=0; i < map->nodes->size; i++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(map->nodes, i))) {
continue;
}
if (!orte_do_not_launch) {
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
* interprets the current behavior. So check both flags to be sure.
*/
if (!support->cpubind->set_thisproc_cpubind &&
!support->cpubind->set_thisthread_cpubind) {
if (!OPAL_BINDING_REQUIRED(map->binding) ||
!OPAL_BINDING_POLICY_IS_SET(map->binding)) {
/* we are not required to bind, so ignore this */
continue;
}
orte_show_help("help-orte-rmaps-base.txt", "rmaps:cpubind-not-supported", true, node->name);
return ORTE_ERR_SILENT;
}
/* check if topology supports membind - have to be careful here
* as hwloc treats this differently than I (at least) would have
* expected. Per hwloc, Linux memory binding is at the thread,
* and not process, level. Thus, hwloc sets the "thisproc" flag
* to "false" on all Linux systems, and uses the "thisthread" flag
* to indicate binding capability - don't warn if the user didn't
* specifically request binding
*/
if (!support->membind->set_thisproc_membind &&
!support->membind->set_thisthread_membind &&
OPAL_BINDING_POLICY_IS_SET(map->binding)) {
if (OPAL_HWLOC_BASE_MBFA_WARN == opal_hwloc_base_mbfa && !membind_warned) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported", true, node->name);
membind_warned = true;
} else if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported-fatal", true, node->name);
return ORTE_ERR_SILENT;
}
}
}
/* some systems do not report cores, and so we can get a situation where our
* default binding policy will fail for no necessary reason. So if we are
* computing a binding due to our default policy, and no cores are found
* on this node, just silently skip it - we will not bind
*/
if (!OPAL_BINDING_POLICY_IS_SET(map->binding) &&
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology, HWLOC_OBJ_CORE)) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"Unable to bind-to core by default on node %s as no cores detected",
node->name);
continue;
}
/* we share topologies in order
* to save space, so we need to reset the usage info to reflect
* our own current state
*/
reset_usage(node, jdata->jobid);
/* cycle thru the procs */
for (j=0; j < node->procs->size; j++) {
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(node->procs, j))) {
continue;
}
/* ignore procs from other jobs */
if (proc->name.jobid != jdata->jobid) {
continue;
}
/* bozo check */
if (!orte_get_attribute(&proc->attributes, ORTE_PROC_HWLOC_LOCALE, (void**)&locale, OPAL_PTR)) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-locale", true, ORTE_NAME_PRINT(&proc->name));
return ORTE_ERR_SILENT;
}
/* get the index of this location */
if (UINT_MAX == (idx = opal_hwloc_base_get_obj_idx(node->topology, locale, OPAL_HWLOC_AVAILABLE))) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
return ORTE_ERR_SILENT;
}
data = (opal_hwloc_obj_data_t*)locale->userdata;
/* get the number of cpus under this location */
if (0 == (ncpus = opal_hwloc_base_get_npus(node->topology, locale))) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
return ORTE_ERR_SILENT;
}
/* if we don't have enough cpus to support this additional proc, try
* shifting the location to a cousin that can support it - the important
* thing is that we maintain the same level in the topology */
if (ncpus < (data->num_bound+1)) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s bind_in_place: searching right",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
sib = locale;
found = false;
while (NULL != (sib = sib->next_cousin)) {
data = (opal_hwloc_obj_data_t*)sib->userdata;
ncpus = opal_hwloc_base_get_npus(node->topology, sib);
if (data->num_bound < ncpus) {
found = true;
locale = sib;
break;
}
}
if (!found) {
/* try the other direction */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s bind_in_place: searching left",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME));
sib = locale;
while (NULL != (sib = sib->prev_cousin)) {
data = (opal_hwloc_obj_data_t*)sib->userdata;
ncpus = opal_hwloc_base_get_npus(node->topology, sib);
if (data->num_bound < ncpus) {
found = true;
locale = sib;
break;
}
}
}
if (!found) {
/* no place to put this - see if overload is allowed */
if (!OPAL_BIND_OVERLOAD_ALLOWED(jdata->map->binding)) {
if (OPAL_BINDING_POLICY_IS_SET(jdata->map->binding)) {
/* if the user specified a binding policy, then we cannot meet
* it since overload isn't allowed, so error out - have the
* message indicate that setting overload allowed will remove
* this restriction */
orte_show_help("help-orte-rmaps-base.txt", "rmaps:binding-overload", true,
opal_hwloc_base_print_binding(map->binding), node->name,
data->num_bound, ncpus);
return ORTE_ERR_SILENT;
} else {
/* if we have the default binding policy, then just don't bind */
OPAL_SET_BINDING_POLICY(map->binding, OPAL_BIND_TO_NONE);
unbind_procs(jdata);
return ORTE_SUCCESS;
}
}
}
}
/* track the number bound */
data = (opal_hwloc_obj_data_t*)locale->userdata; // just in case it changed
data->num_bound++;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"BINDING PROC %s TO %s NUMBER %u",
ORTE_NAME_PRINT(&proc->name),
hwloc_obj_type_string(locale->type), idx);
/* bind the proc here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, locale);
hwloc_bitmap_list_asprintf(&cpu_bitmap, cpus);
orte_set_attribute(&proc->attributes, ORTE_PROC_CPU_BITMAP, ORTE_ATTR_GLOBAL, cpu_bitmap, OPAL_STRING);
/* update the location, in case it changed */
orte_set_attribute(&proc->attributes, ORTE_PROC_HWLOC_BOUND, ORTE_ATTR_LOCAL, locale, OPAL_PTR);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s BOUND PROC %s TO %s[%s:%u] on node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name),
cpu_bitmap, hwloc_obj_type_string(locale->type),
idx, node->name);
if (NULL != cpu_bitmap) {
free(cpu_bitmap);
}
}
}
return ORTE_SUCCESS;
}
static int bind_in_place(orte_job_t *jdata,
hwloc_obj_type_t target,
unsigned cache_level)
{
/* traverse the hwloc topology tree on each node downwards
* until we find an unused object of type target - and then bind
* the process to that target
*/
int i, j;
orte_job_map_t *map;
orte_node_t *node;
orte_proc_t *proc;
hwloc_cpuset_t cpus;
unsigned int idx, ncpus;
struct hwloc_topology_support *support;
opal_hwloc_obj_data_t *data;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: bind in place for job %s with bindings %s",
ORTE_JOBID_PRINT(jdata->jobid),
opal_hwloc_base_print_binding(jdata->map->binding));
/* initialize */
map = jdata->map;
for (i=0; i < map->nodes->size; i++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(map->nodes, i))) {
continue;
}
if (!orte_do_not_launch) {
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
* interprets the current behavior. So check both flags to be sure.
*/
if (!support->cpubind->set_thisproc_cpubind &&
!support->cpubind->set_thisthread_cpubind) {
if (!OPAL_BINDING_REQUIRED(map->binding) ||
!OPAL_BINDING_POLICY_IS_SET(map->binding)) {
/* we are not required to bind, so ignore this */
continue;
}
orte_show_help("help-orte-rmaps-base.txt", "rmaps:cpubind-not-supported", true, node->name);
return ORTE_ERR_SILENT;
}
/* check if topology supports membind - have to be careful here
* as hwloc treats this differently than I (at least) would have
* expected. Per hwloc, Linux memory binding is at the thread,
* and not process, level. Thus, hwloc sets the "thisproc" flag
* to "false" on all Linux systems, and uses the "thisthread" flag
* to indicate binding capability - don't warn if the user didn't
* specifically request binding
*/
if (!support->membind->set_thisproc_membind &&
!support->membind->set_thisthread_membind &&
OPAL_BINDING_POLICY_IS_SET(map->binding)) {
if (OPAL_HWLOC_BASE_MBFA_WARN == opal_hwloc_base_mbfa && !membind_warned) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported", true, node->name);
membind_warned = true;
} else if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported-fatal", true, node->name);
return ORTE_ERR_SILENT;
}
}
}
/* some systems do not report cores, and so we can get a situation where our
* default binding policy will fail for no necessary reason. So if we are
* computing a binding due to our default policy, and no cores are found
* on this node, just silently skip it - we will not bind
*/
if (!OPAL_BINDING_POLICY_IS_SET(map->binding) &&
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology, HWLOC_OBJ_CORE)) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"Unable to bind-to core by default on node %s as no cores detected",
node->name);
continue;
}
/* we share topologies in order
* to save space, so we need to reset the usage info to reflect
* our own current state
*/
reset_usage(node, jdata->jobid);
/* cycle thru the procs */
for (j=0; j < node->procs->size; j++) {
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(node->procs, j))) {
continue;
}
/* ignore procs from other jobs */
if (proc->name.jobid != jdata->jobid) {
continue;
}
/* ignore procs that have already been bound - should
* never happen, but safer
*/
if (NULL != proc->cpu_bitmap) {
continue;
}
/* get the index of this location */
if (UINT_MAX == (idx = opal_hwloc_base_get_obj_idx(node->topology, proc->locale, OPAL_HWLOC_AVAILABLE))) {
ORTE_ERROR_LOG(ORTE_ERR_BAD_PARAM);
return ORTE_ERR_SILENT;
}
/* track the number bound */
data = (opal_hwloc_obj_data_t*)proc->locale->userdata;
data->num_bound++;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"BINDING PROC %s TO %s NUMBER %u",
ORTE_NAME_PRINT(&proc->name),
hwloc_obj_type_string(proc->locale->type), idx);
/* get the number of cpus under this location */
if (0 == (ncpus = opal_hwloc_base_get_npus(node->topology, proc->locale))) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
return ORTE_ERR_SILENT;
}
/* error out if adding a proc would cause overload and that wasn't allowed,
* and it wasn't a default binding policy (i.e., the user requested it)
*/
if (ncpus < data->num_bound &&
!OPAL_BIND_OVERLOAD_ALLOWED(jdata->map->binding) &&
OPAL_BINDING_POLICY_IS_SET(jdata->map->binding)) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:binding-overload", true,
opal_hwloc_base_print_binding(map->binding), node->name,
data->num_bound, ncpus);
return ORTE_ERR_SILENT;
}
/* bind the proc here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, proc->locale);
hwloc_bitmap_list_asprintf(&proc->cpu_bitmap, cpus);
/* record the location */
proc->bind_location = proc->locale;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s BOUND PROC %s TO %s[%s:%u] on node %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name),
proc->cpu_bitmap,
hwloc_obj_type_string(proc->locale->type),
idx, node->name);
}
}
return ORTE_SUCCESS;
}
static int bind_downwards(orte_job_t *jdata,
hwloc_obj_type_t target,
unsigned cache_level)
{
int i, j;
orte_job_map_t *map;
orte_node_t *node;
orte_proc_t *proc;
hwloc_obj_t trg_obj, nxt_obj;
hwloc_cpuset_t cpus;
unsigned int ncpus;
struct hwloc_topology_support *support;
opal_hwloc_obj_data_t *data;
int total_cpus;
hwloc_cpuset_t totalcpuset;
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps: bind downward for job %s with bindings %s",
ORTE_JOBID_PRINT(jdata->jobid),
opal_hwloc_base_print_binding(jdata->map->binding));
/* initialize */
map = jdata->map;
totalcpuset = hwloc_bitmap_alloc();
for (i=0; i < map->nodes->size; i++) {
if (NULL == (node = (orte_node_t*)opal_pointer_array_get_item(map->nodes, i))) {
continue;
}
if (!orte_do_not_launch) {
/* if we don't want to launch, then we are just testing the system,
* so ignore questions about support capabilities
*/
support = (struct hwloc_topology_support*)hwloc_topology_get_support(node->topology);
/* check if topology supports cpubind - have to be careful here
* as Linux doesn't currently support thread-level binding. This
* may change in the future, though, and it isn't clear how hwloc
* interprets the current behavior. So check both flags to be sure.
*/
if (!support->cpubind->set_thisproc_cpubind &&
!support->cpubind->set_thisthread_cpubind) {
if (!OPAL_BINDING_REQUIRED(opal_hwloc_binding_policy) ||
!(OPAL_BIND_GIVEN & opal_hwloc_binding_policy)) {
/* we are not required to bind, so ignore this */
continue;
}
orte_show_help("help-orte-rmaps-base.txt", "rmaps:cpubind-not-supported", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
/* check if topology supports membind - have to be careful here
* as hwloc treats this differently than I (at least) would have
* expected. Per hwloc, Linux memory binding is at the thread,
* and not process, level. Thus, hwloc sets the "thisproc" flag
* to "false" on all Linux systems, and uses the "thisthread" flag
* to indicate binding capability - don't warn if the user didn't
* specifically request binding
*/
if (!support->membind->set_thisproc_membind &&
!support->membind->set_thisthread_membind &&
(OPAL_BIND_GIVEN & opal_hwloc_binding_policy)) {
if (OPAL_HWLOC_BASE_MBFA_WARN == opal_hwloc_base_mbfa && !membind_warned) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported", true, node->name);
membind_warned = true;
} else if (OPAL_HWLOC_BASE_MBFA_ERROR == opal_hwloc_base_mbfa) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:membind-not-supported-fatal", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
}
}
/* some systems do not report cores, and so we can get a situation where our
* default binding policy will fail for no necessary reason. So if we are
* computing a binding due to our default policy, and no cores are found
* on this node, just silently skip it - we will not bind
*/
if (!(OPAL_BIND_GIVEN & opal_hwloc_binding_policy) &&
HWLOC_TYPE_DEPTH_UNKNOWN == hwloc_get_type_depth(node->topology, HWLOC_OBJ_CORE)) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"Unable to bind-to core by default on node %s as no cores detected",
node->name);
continue;
}
/* we share topologies in order
* to save space, so we need to reset the usage info to reflect
* our own current state
*/
reset_usage(node, jdata->jobid);
/* cycle thru the procs */
for (j=0; j < node->procs->size; j++) {
if (NULL == (proc = (orte_proc_t*)opal_pointer_array_get_item(node->procs, j))) {
continue;
}
/* ignore procs from other jobs */
if (proc->name.jobid != jdata->jobid) {
continue;
}
/* ignore procs that have already been bound - should
* never happen, but safer
*/
if (NULL != proc->cpu_bitmap) {
continue;
}
/* we don't know if the target is a direct child of this locale,
* or if it is some depth below it, so we have to conduct a bit
* of a search. Let hwloc find the min usage one for us.
*/
trg_obj = opal_hwloc_base_find_min_bound_target_under_obj(node->topology,
proc->locale,
target, cache_level);
if (NULL == trg_obj) {
/* there aren't any such targets under this object */
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
/* record the location */
proc->bind_location = trg_obj;
/* start with a clean slate */
hwloc_bitmap_zero(totalcpuset);
total_cpus = 0;
nxt_obj = trg_obj;
do {
if (NULL == nxt_obj) {
/* could not find enough cpus to meet request */
orte_show_help("help-orte-rmaps-base.txt", "rmaps:no-available-cpus", true, node->name);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
}
trg_obj = nxt_obj;
/* get the number of cpus under this location */
ncpus = opal_hwloc_base_get_npus(node->topology, trg_obj);
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"%s GOT %d CPUS",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ncpus);
/* track the number bound */
if (NULL == (data = (opal_hwloc_obj_data_t*)trg_obj->userdata)) {
data = OBJ_NEW(opal_hwloc_obj_data_t);
trg_obj->userdata = data;
}
data->num_bound++;
/* error out if adding a proc would cause overload and that wasn't allowed,
* and it wasn't a default binding policy (i.e., the user requested it)
*/
if (ncpus < data->num_bound &&
!OPAL_BIND_OVERLOAD_ALLOWED(jdata->map->binding)) {
if (OPAL_BIND_GIVEN & opal_hwloc_binding_policy) {
orte_show_help("help-orte-rmaps-base.txt", "rmaps:binding-overload", true,
opal_hwloc_base_print_binding(map->binding), node->name,
data->num_bound, ncpus);
hwloc_bitmap_free(totalcpuset);
return ORTE_ERR_SILENT;
} else {
/* if this is the default binding policy, then just don't
* bind this proc
*/
data->num_bound--; // maintain count
/* show the proc as not bound */
proc->bind_location = NULL;
hwloc_bitmap_zero(totalcpuset);
break;
}
}
/* bind the proc here */
cpus = opal_hwloc_base_get_available_cpus(node->topology, trg_obj);
hwloc_bitmap_or(totalcpuset, totalcpuset, cpus);
/* track total #cpus */
total_cpus += ncpus;
/* move to the next location, in case we need it */
nxt_obj = trg_obj->next_cousin;
} while (total_cpus < orte_rmaps_base.cpus_per_rank);
hwloc_bitmap_list_asprintf(&proc->cpu_bitmap, totalcpuset);
if (4 < opal_output_get_verbosity(orte_rmaps_base_framework.framework_output)) {
char tmp1[1024], tmp2[1024];
if (OPAL_ERR_NOT_BOUND == opal_hwloc_base_cset2str(tmp1, sizeof(tmp1), totalcpuset)) {
opal_output(orte_rmaps_base_framework.framework_output,
"%s PROC %s ON %s IS NOT BOUND",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name);
} else {
opal_hwloc_base_cset2mapstr(tmp2, sizeof(tmp2), totalcpuset);
opal_output(orte_rmaps_base_framework.framework_output,
"%s BOUND PROC %s[%s] TO %s: %s",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(&proc->name), node->name,
tmp1, tmp2);
}
}
}
}
hwloc_bitmap_free(totalcpuset);
return ORTE_SUCCESS;
}
