ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Add debug, cleanup cpus/rank

Просмотр исходного кода 
Signed-off-by: Ralph Castain <rhc@open-mpi.org>
Этот коммит содержится в:
Ralph Castain 2016-12-27 21:25:52 -08:00
родитель 1e4bffd937
Коммит 7866bb1119
1 изменённых файлов: 39 добавлений и 30 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

69
orte/mca/rmaps/base/rmaps_base_map_job.c
Просмотреть файл

@ -191,6 +191,19 @@ void orte_rmaps_base_map_job(int fd, short args, void *cbdata)
* MCA param, then we use it - this can include a directive
* to overload */
jdata->map->binding = opal_hwloc_binding_policy;
} else if (1 < jdata->map->cpus_per_rank) {
/* bind to cpus */
if (opal_hwloc_use_hwthreads_as_cpus) {
/* if we are using hwthread cpus, then bind to those */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using byhwthread", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_HWTHREAD);
} else {
/* bind to core */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bycore", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_CORE);
}
} else {
/* if the user explicitly mapped-by some object, then we default
* to binding to that object */
@ -198,55 +211,51 @@ void orte_rmaps_base_map_job(int fd, short args, void *cbdata)
mpol = ORTE_GET_MAPPING_POLICY(jdata->map->mapping);
if (ORTE_MAPPING_GIVEN & ORTE_GET_MAPPING_DIRECTIVE(jdata->map->mapping)) {
if (ORTE_MAPPING_BYHWTHREAD == mpol) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using byhwthread", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_HWTHREAD);
} else if (ORTE_MAPPING_BYCORE == mpol) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bycore", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_CORE);
} else if (ORTE_MAPPING_BYL1CACHE == mpol) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using byL1", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_L1CACHE);
} else if (ORTE_MAPPING_BYL2CACHE == mpol) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using byL2", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_L2CACHE);
} else if (ORTE_MAPPING_BYL3CACHE == mpol) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using byL3", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_L3CACHE);
} else if (ORTE_MAPPING_BYSOCKET == mpol) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bysocket", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_SOCKET);
} else if (ORTE_MAPPING_BYNUMA == mpol) {
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bynuma", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_NUMA);
}
} else if (nprocs <= 2) {
/* if nothing was specified, then we default to a policy
* based on number of procs and cpus_per_rank */
if (1 < orte_rmaps_base.cpus_per_rank) {
/* assigning multiple cpus to a rank implies threading,
* so we only bind to the NUMA level */
if (opal_hwloc_use_hwthreads_as_cpus) {
/* if we are using hwthread cpus, then bind to those */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bynuma", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_NUMA);
"mca:rmaps[%d] binding not given - using byhwthread", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_HWTHREAD);
} else {
if (opal_hwloc_use_hwthreads_as_cpus) {
/* if we are using hwthread cpus, then bind to those */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using byhwthread", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_HWTHREAD);
} else {
/* for performance, bind to core */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bycore", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_CORE);
}
/* for performance, bind to core */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bycore", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_CORE);
}
} else {
if (1 < orte_rmaps_base.cpus_per_rank) {
/* assigning multiple cpus to a rank implies threading,
* so we only bind to the NUMA level */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bynuma", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_NUMA);
} else {
/* for performance, bind to socket */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bysocket", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_SOCKET);
}
/* for performance, bind to NUMA */
opal_output_verbose(5, orte_rmaps_base_framework.framework_output,
"mca:rmaps[%d] binding not given - using bynuma", __LINE__);
OPAL_SET_DEFAULT_BINDING_POLICY(jdata->map->binding, OPAL_BIND_TO_NUMA);
}
if (OPAL_BIND_OVERLOAD_ALLOWED(opal_hwloc_binding_policy)) {
jdata->map->binding |= OPAL_BIND_ALLOW_OVERLOAD;