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

Yet another paffinity cleanup...sigh.

Просмотр исходного кода 
1. ensure that orte_rmaps_base_schedule_policy does not override cmd line settings

2. when you try to bind to more cores than we have, generate a not-enough-processors error message

3. allow npersocket -bind-to-core combination - because, yes, somebody actually wants to do it.

This commit was SVN r21996.
Этот коммит содержится в:
Ralph Castain 2009-09-22 18:44:53 +00:00
родитель 5406fdfb80
Коммит dff0d01673
4 изменённых файлов: 109 добавлений и 27 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

99
orte/mca/odls/default/odls_default_module.c
Просмотреть файл

@ -342,8 +342,65 @@ static int odls_default_fork_local_proc(orte_app_context_t* context,
"odls-default:invalid-node-rank", true);
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
/* get the local rank */
if (ORTE_LOCAL_RANK_INVALID == (lrank = orte_ess.get_local_rank(child->name))) {
orte_show_help("help-odls-default.txt",
"odls-default:invalid-local-rank", true);
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
/* init the mask */
OPAL_PAFFINITY_CPU_ZERO(mask);
if (ORTE_MAPPING_BYSOCKET & jobdat->policy) {
if (ORTE_MAPPING_NPERXXX & jobdat->policy) {
/* we need to balance the children from this job across the available sockets */
npersocket = jobdat->num_local_procs / orte_odls_globals.num_sockets;
/* determine the socket to use based on those available */
if (npersocket < 2) {
/* if we only have 1/sock, or we have less procs than sockets,
* then just put it on the lrank socket
*/
logical_skt = lrank;
} else if (ORTE_MAPPING_BYSOCKET & jobdat->policy) {
logical_skt = lrank % npersocket;
} else {
logical_skt = lrank / npersocket;
}
if (orte_odls_globals.bound) {
/* if we are bound, use this as an index into our available sockets */
for (target_socket=0; target_socket < opal_bitmap_size(&orte_odls_globals.sockets) && n < logical_skt; target_socket++) {
if (opal_bitmap_is_set_bit(&orte_odls_globals.sockets, target_socket)) {
n++;
}
}
/* if we don't have enough sockets, that is an error */
if (n < logical_skt) {
ORTE_ODLS_IF_BIND_NOT_REQD("bind-to-core");
orte_show_help("help-odls-default.txt",
"odls-default:not-enough-resources", true,
"sockets", orte_process_info.nodename,
"bind-to-core", context->app);
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
} else {
target_socket = opal_paffinity_base_get_physical_socket_id(logical_skt);
if (ORTE_ERR_NOT_SUPPORTED == target_socket) {
/* OS doesn't support providing topology information */
ORTE_ODLS_IF_BIND_NOT_REQD("bind-to-core");
orte_show_help("help-odls-default.txt",
"odls-default:topo-not-supported",
true, orte_process_info.nodename, "bind-to-core", "",
context->app);
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
}
OPAL_OUTPUT_VERBOSE((2, orte_odls_globals.output,
"%s odls:default:fork child %s local rank %d npersocket %d logical socket %d target socket %d",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME), ORTE_NAME_PRINT(child->name), lrank,
npersocket, logical_skt, target_socket));
/* set the starting point */
logical_cpu = (lrank % npersocket) * jobdat->cpus_per_rank;
/* bind to this socket */
goto bind_socket;
} else if (ORTE_MAPPING_BYSOCKET & jobdat->policy) {
/* this corresponds to a mapping policy where
* local rank 0 goes on socket 0, and local
* rank 1 goes on socket 1, etc. - round robin
@ -352,11 +409,6 @@ static int odls_default_fork_local_proc(orte_app_context_t* context,
* NOTE: we already know our number of sockets
* from when we initialized
*/
if (ORTE_LOCAL_RANK_INVALID == (lrank = orte_ess.get_local_rank(child->name))) {
orte_show_help("help-odls-default.txt",
"odls-default:invalid-local-rank", true);
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
target_socket = opal_paffinity_base_get_physical_socket_id(lrank % orte_odls_globals.num_sockets);
if (ORTE_ERR_NOT_SUPPORTED == target_socket) {
/* OS does not support providing topology information */
@ -374,22 +426,25 @@ static int odls_default_fork_local_proc(orte_app_context_t* context,
target_socket));
/* my starting core within this socket has to be offset by cpus_per_rank */
logical_cpu = (lrank / orte_odls_globals.num_sockets) * jobdat->cpus_per_rank;
bind_socket:
/* cycle across the cpus_per_rank */
for (n=0; n < jobdat->cpus_per_rank; n++) {
ncpu = logical_cpu + (n * jobdat->stride);
/* get the physical core within this target socket */
phys_core = opal_paffinity_base_get_physical_core_id(target_socket, ncpu);
phys_core = opal_paffinity_base_get_physical_core_id(target_socket, logical_cpu);
if (0 > phys_core) {
ORTE_ODLS_IF_BIND_NOT_REQD("bind-to-socket");
ORTE_ODLS_IF_BIND_NOT_REQD("bind-to-core");
orte_show_help("help-odls-default.txt",
"odls-default:invalid-phys-cpu", true);
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
/* map this to a physical cpu on this node */
if (ORTE_SUCCESS != opal_paffinity_base_get_map_to_processor_id(target_socket, phys_core, &phys_cpu)) {
ORTE_ODLS_IF_BIND_NOT_REQD("bind-to-socket");
ORTE_ODLS_IF_BIND_NOT_REQD("bind-to-core");
orte_show_help("help-odls-default.txt",
"odls-default:invalid-phys-cpu", true);
"odls-default:not-enough-resources", true,
"processors", orte_process_info.nodename,
"bind-to-core", context->app);
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
/* are we bound? */
@ -405,6 +460,13 @@ static int odls_default_fork_local_proc(orte_app_context_t* context,
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
target_socket, phys_core, phys_cpu));
OPAL_PAFFINITY_CPU_SET(phys_cpu, mask);
/* increment logical cpu */
logical_cpu += jobdat->stride;
}
if (orte_odls_globals.report_bindings) {
opal_output(0, "%s odls:default:fork binding child %s to socket %d cpus %04lx",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(child->name), target_socket, mask.bitmask[0]);
}
} else {
/* my starting core has to be offset by cpus_per_rank */
@ -444,14 +506,15 @@ static int odls_default_fork_local_proc(orte_app_context_t* context,
ORTE_ODLS_ERROR_OUT(ORTE_ERR_FATAL);
}
}
OPAL_PAFFINITY_CPU_SET(phys_cpu, mask);
/* increment logical cpu */
logical_cpu += jobdat->stride;
}
if (orte_odls_globals.report_bindings) {
opal_output(0, "%s odls:default:fork binding child %s to cpus %04lx",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(child->name), mask.bitmask[0]);
}
OPAL_PAFFINITY_CPU_SET(phys_cpu, mask);
logical_cpu += jobdat->stride;
}
if (orte_odls_globals.report_bindings) {
opal_output(0, "%s odls:default:fork binding child %s to cpus %04lx",
ORTE_NAME_PRINT(ORTE_PROC_MY_NAME),
ORTE_NAME_PRINT(child->name), mask.bitmask[0]);
}
if (ORTE_SUCCESS != (rc = opal_paffinity_base_set(mask))) {
ORTE_ODLS_IF_BIND_NOT_REQD("bind-to-core");

24
orte/mca/rmaps/base/rmaps_base_open.c
Просмотреть файл

@ -94,15 +94,21 @@ int orte_rmaps_base_open(void)
/* Are we scheduling by node or by slot? */
param = mca_base_param_reg_string_name("rmaps", "base_schedule_policy",
"Scheduling Policy for RMAPS. [slot (default) | socket | board | node]",
"Scheduling Policy for RMAPS. [slot (alias:core) | socket | board | node]",
false, false, "unspec", &policy);
if (0 == strcmp(policy, "socket")) {
ORTE_SET_MAPPING_POLICY(ORTE_MAPPING_BYSOCKET);
} else if (0 == strcmp(policy, "board")) {
ORTE_SET_MAPPING_POLICY(ORTE_MAPPING_BYBOARD);
} else if (0 == strcmp(policy, "node")) {
ORTE_SET_MAPPING_POLICY(ORTE_MAPPING_BYNODE);
/* if something is specified, do not override what may already
* be present - could have been given on cmd line
*/
if (0 == strcasecmp(policy, "slot") ||
0 == strcasecmp(policy, "core")) {
ORTE_XSET_MAPPING_POLICY(ORTE_MAPPING_BYSLOT);
} else if (0 == strcasecmp(policy, "socket")) {
ORTE_XSET_MAPPING_POLICY(ORTE_MAPPING_BYSOCKET);
} else if (0 == strcasecmp(policy, "board")) {
ORTE_XSET_MAPPING_POLICY(ORTE_MAPPING_BYBOARD);
} else if (0 == strcasecmp(policy, "node")) {
ORTE_XSET_MAPPING_POLICY(ORTE_MAPPING_BYNODE);
}
/* if nothing was specified, leave it alone - we already set it
* in orterun
@ -138,8 +144,8 @@ int orte_rmaps_base_open(void)
false, false, -1, &orte_rmaps_base.npersocket);
if (0 < orte_rmaps_base.npersocket) {
ORTE_ADD_MAPPING_POLICY(ORTE_MAPPING_NPERXXX);
/* force bind to socket */
ORTE_SET_BINDING_POLICY(ORTE_BIND_TO_SOCKET);
/* force bind to socket if not overridden by user */
ORTE_XSET_BINDING_POLICY(ORTE_BIND_TO_SOCKET);
}
/* Do we want to loadbalance the job */

10
orte/runtime/orte_globals.h
Просмотреть файл

@ -293,6 +293,16 @@ typedef uint16_t orte_mapping_policy_t;
/* nice macro for setting these */
#define ORTE_SET_MAPPING_POLICY(pol) \
orte_default_mapping_policy = (orte_default_mapping_policy & 0x00ff) | (pol);
/* macro to detect if some other policy has been set */
#define ORTE_XSET_MAPPING_POLICY(pol) \
do { \
orte_mapping_policy_t tmp; \
tmp = (orte_default_mapping_policy & 0xff00) & ~(pol); \
if (0 == tmp) { \
ORTE_SET_MAPPING_POLICY((pol)); \
} \
} while(0);
/* macro to add another mapping policy */
#define ORTE_ADD_MAPPING_POLICY(pol) \
orte_default_mapping_policy |= (pol);

3
orte/tools/orterun/orterun.c
Просмотреть файл

@ -263,6 +263,9 @@ static opal_cmd_line_init_t cmd_line_init[] = {
{ NULL, NULL, NULL, '\0', "byslot", "byslot", 0,
&orterun_globals.by_slot, OPAL_CMD_LINE_TYPE_BOOL,
"Whether to assign processes round-robin by slot (the default)" },
{ NULL, NULL, NULL, '\0', "bycore", "bycore", 0,
&orterun_globals.by_slot, OPAL_CMD_LINE_TYPE_BOOL,
"Alias for byslot" },
{ NULL, NULL, NULL, '\0', "bysocket", "bysocket", 0,
&orterun_globals.by_socket, OPAL_CMD_LINE_TYPE_BOOL,
"Whether to assign processes round-robin by socket" },