ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
openmpi/ompi/mca/sbgp/basesmsocket/sbgp_basesmsocket_component.c
Nathan Hjelm 1a021b8f2d coll/ml: add support for blocking and non-blocking allreduce, reduce, and 
allgather.

The new collectives provide a signifigant performance increase over tuned for
small and medium messages. We are initially setting the priority lower than
tuned until this has had some time to soak in the trunk. Please set
coll_ml_priority to 90 for MTT runs.

Credit for this work goes to Manjunath Gorentla Venkata (ORNL), Pavel Shamis (ORNL),
and Nathan Hjelm (LANL).

Commit details (for reference):

Import ORNL's collectives for MPI_Allreduce, MPI_Reduce, and MPI_Allgather.

We need to take the basesmuma header into account when calculating the
ptpcoll small message thresholds. Add a define to bcol.h indicating the
maximum header size so we can take the header into account while not
making ptpcoll dependent on information from basesmuma.

This resolves an issue with allreduce where ptpcoll overwrites the
header of the next buffer in the basesmuma bank.

Fix reduce and make a sequential collective launcher in coll_ml_inlines.h

The root calculation for reduce was wrong for any root != 0. There are
four possibilities for the root:

 - The root is not the current process but is in the current hierarchy. In
   this case the root is the index of the global root as specified in the
   root vector.

 - The root is not the current process and is not in the next level of the
   hierarchy. In this case 0 must be the local root since this process will
   never communicate with the real root.

 - The root is not the current process but will be in next level of the
   hierarchy. In this case the current process must be the root.

 - I am the root. The root is my index.

Tested with IMB which rotates the root on every call to MPI_Reduce. Consider
IMB the reproducer for the issue this commit solves.

Make the bcast algorithm decision an enumerated variable

Resolve various asset failures when destructing coll ml requests.

Two issues:

 - Always reset the request to be invalid before returning it to the
   free list. This will avoid an asset in ompi_request_t's destructor.
   OMPI_REQUEST_FINI does this (and also releases the fortran handle
   index).

 - Never explicitly construct or destruct the superclass of an opal
   object. This screws up the class function tables and will cause
   either an assert failure or a segmentation fault when destructing
   coll ml requests.

Cleanup allgather.

I removed the duplicate non-blocking and blocking functions and modeled
the cleanup after what I found in allreduce. Also cleaned up the code
somewhat.

Don't bother copying from the send to the recieve buffer in
bcol_basesmuma_allreduce_intra_fanin_fanout if the pointers are the
same.

The eliminates a warning about memcpy and aliasing and avoids an
unnecessary call to memcpy.

Alwasy call CHECK_AND_RELEASE on memsync collectives.

There was a call to OBJ_RELEASE on the collective communicator but
because CHECK_AND_RECYLCE was never called there was not matching call
to OBJ_RELEASE. This caused coll ml to leak communicators.

Make allreduce use the sequential collective launcher in coll_ml_inlines.h

Just launch the next collective in the component progress.

I am a little unsure about this patch. There appears to be some sort
of race between collectives that causes buffer exhaustion in some cases
(IMB Allreduce is a reproducer). Changing progress to only launch the
next bcol seems to resolve the issue but might not be the best fix.

Note that I see little-no performance penalty for this change.

Fix allreduce when there are extra sources.

There was an issue with the buffer offset calculation when there are
extra sources. In the case of extra sources == 1 the offset was set
to buffer_size (just past the header of the next buffer). I adjusted
the buffer size to take into accoun the maximum header size (see the
earlier commit that added this) and simplified the offset calculation.

Make reduce/allreduce non-blocking. This is required for MPI_Comm_idup
to work correctly.

This has been tested with various layouts using the ibm testsuite and
imb and appears to have the same performance as the old blocking version.

Fix allgather for non-contiguous layouts and simplify parsing the
topology.

Some things in this patch:

 - There were several comments to the effect that level 0 of the
   hierarchy MUST contain all of the ranks. At least one function
   made this assumption but it was not true. I changed the sbgp
   components and the coll ml initization code to enforce this
   requirement.

 - Ensure that hierarchy level 0 has the ranks in the correct
   scatter gather order. This removes the need for a separate
   sort list and fixes the offset calculation for allgather.

 - There were several passes over the hierarchy to determine
   properties of the hierarchy. I eliminated these extra passes
   and the memory allocation associated with them and calculate the
   tree properties on the fly. The same DFS recursion also handles
   the re-order of level 0.

All these changes have been verified with MPI_Allreduce, MPI_Reduce, and
MPI_Allgather. All functions now pass all IBM/Open MPI, and IMB tests.

coll/ml: correct pointer usage for MPI_BOTTOM

Since contiguous datatypes are copied via memcpy (bypassing the convertor) we
need to adjust for the lb of the datatype. This corrects problems found testing
code that uses MPI_BOTTOM (NULL) as the send pointer.

Add fallback collectives for allreduce and reduce.

cmr=v1.7.5:reviewer=pasha

This commit was SVN r30363.
2014-01-22 15:39:19 +00:00

839 строки
24 KiB
C
Исходник Ответственный История

/*
* Copyright (c) 2009-2012 Oak Ridge National Laboratory. All rights reserved.
* Copyright (c) 2009-2012 Mellanox Technologies. All rights reserved.
* Copyright (c) 2013 Cisco Systems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
/**
* @file
*
*/
#include "ompi_config.h"
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#include <sys/types.h>
#ifdef HAVE_SYS_MMAN_H
#include <sys/mman.h>
#endif
#ifdef HAVE_FCNTL_H
#include <fcntl.h>
#endif
#include "opal/mca/hwloc/hwloc.h"
#include "opal/mca/hwloc/base/base.h"
#include "opal/dss/dss_internal.h"
#include "opal/class/opal_object.h"
#include "ompi/constants.h"
#include "ompi/communicator/communicator.h"
#include "sbgp_basesmsocket.h"
#include "ompi/patterns/comm/coll_ops.h"
/*
* Public string showing the coll ompi_sm V2 component version number
*/
const char *mca_sbgp_basesmsocket_component_version_string =
"Open MPI sbgp - basesmsocket collective MCA component version " OMPI_VERSION;
/*
* Local functions
*/
static int basesmsocket_register(void);
static int basesmsocket_open(void);
static int basesmsocket_close(void);
static mca_sbgp_base_module_t *mca_sbgp_basesmsocket_select_procs(struct ompi_proc_t ** procs,
int n_procs_in,
struct ompi_communicator_t *comm,
char *key,
void *output_data
);
static int mca_sbgp_basesmsocket_init_query(bool enable_progress_threads,
bool enable_mpi_threads);
/*----end local functions ----*/
/*
* Instantiate the public struct with all of our public information
* and pointers to our public functions in it
*/
mca_sbgp_basesmsocket_component_t mca_sbgp_basesmsocket_component = {
/* First, fill in the super */
{
/* First, the mca_component_t struct containing meta
information about the component itself */
{
MCA_SBGP_BASE_VERSION_2_0_0,
/* Component name and version */
"basesmsocket",
OMPI_MAJOR_VERSION,
OMPI_MINOR_VERSION,
OMPI_RELEASE_VERSION,
/* Component open and close functions */
basesmsocket_open,
basesmsocket_close,
NULL,
basesmsocket_register
},
mca_sbgp_basesmsocket_init_query,
mca_sbgp_basesmsocket_select_procs,
/* (default) priority */
0
}
};
/*
* Register the component
*/
static int basesmsocket_register(void)
{
mca_sbgp_basesmsocket_component_t *cs = &mca_sbgp_basesmsocket_component;
cs->super.priority = 90;
(void) mca_base_component_var_register(&mca_sbgp_basesmsocket_component.super.sbgp_version,
"priority", "Priority for the sbgp basesmsocket component",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY, &cs->super.priority);
return OMPI_SUCCESS;
}
/*
* Open the component
*/
static int basesmsocket_open(void)
{
return OMPI_SUCCESS;
}
/*
* Close the component
*/
static int basesmsocket_close(void)
{
return OMPI_SUCCESS;
}
/* query to see if the component is available for use, and can
* satisfy the thread and progress requirements
*/
int mca_sbgp_basesmsocket_init_query(bool enable_progress_threads,
bool enable_mpi_threads)
{
/* at this stage there is no reason to disaulify this component */
/* done */
return OMPI_SUCCESS;
}
static int mca_sbgp_map_to_logical_socket_id(int *socket)
{
int ret = OMPI_SUCCESS;
hwloc_obj_t obj;
hwloc_obj_t first_pu_object;
hwloc_bitmap_t good;
int pu_os_index = -1, my_logical_socket_id = -1;
int this_pus_logical_socket_id = -1;
*socket = my_logical_socket_id;
/* bozo check */
if (NULL == opal_hwloc_topology) {
return OPAL_ERR_NOT_INITIALIZED;
}
good = hwloc_bitmap_alloc();
if (NULL == good) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
/* get this process' CPU binding */
if( 0 != hwloc_get_cpubind(opal_hwloc_topology,good, 0)){
/* report some error */
BASESMSOCKET_VERBOSE(10, ("The global variable opal_hwloc_topology appears not to have been initialized\n"));
return OMPI_ERROR;
}
/* find the first logical PU object in the hwloc tree */
first_pu_object = hwloc_get_obj_by_type(opal_hwloc_topology, HWLOC_OBJ_PU, 0);
/* get the next bit in the bitmap (note: if pu_os_index == -1, then the
* first bit is returned
*/
/* traverse the hwloc tree */
while( -1 != (pu_os_index = hwloc_bitmap_next(good, pu_os_index) ) ) {
/* Traverse all PUs in the machine in logical order, in the simple case
* there should only be a single PU that this process is bound to, right?
*
*/
for( obj = first_pu_object; obj != NULL; obj = obj->next_cousin ) {/* WTF is a "next_cousin" ? */
/* is this PU the same as the bit I pulled off the mask? */
if( obj->os_index == (unsigned int) pu_os_index) {
/* Then I found it, break out of for loop */
break;
}
}
if( NULL != obj) {
/* if we found the PU, then go upward in the tree
* looking for the enclosing socket
*/
while( (NULL != obj) && ( HWLOC_OBJ_SOCKET != obj->type) ){
obj = obj->parent;
}
if( NULL == obj ) {
/* then we couldn't find an enclosing socket, report this */
} else {
/* We found the enclosing socket */
if( -1 == my_logical_socket_id ){
/* this is the first PU that I'm bound to */
this_pus_logical_socket_id = obj->logical_index;
my_logical_socket_id = this_pus_logical_socket_id;
} else {
/* this is not the first PU that I'm bound to.
* Seems I'm bound to more than a single PU. Question
* is, am I bound to the same socket??
*/
/* in order to get rid of the compiler warning, I had to cast
* "this_pus_logical_socket_id", at a glance this seems ok,
* but if subgrouping problems arise, maybe look here. I shall
* tag this line with the "mark of the beast" for grepability
* 666
*/
if( (unsigned int) this_pus_logical_socket_id != obj->logical_index ){
/* 666 */
/* Then we're bound to more than one socket...fail */
this_pus_logical_socket_id = -1;
my_logical_socket_id = -1;
break;
}
}
}
}
/* end while */
}
*socket = my_logical_socket_id;
return ret;
}
/* This routine is used to find the list of procs that run on the
** same host as the calling process.
*/
static mca_sbgp_base_module_t *mca_sbgp_basesmsocket_select_procs(struct ompi_proc_t ** procs,
int n_procs_in,
struct ompi_communicator_t *comm,
char *key,
void *output_data
)
{
/* local variables */
mca_sbgp_basesmsocket_module_t *module;
int ret;
int my_socket_index;
int proc, cnt, local, n_local_peers, my_rank;
ompi_proc_t* my_proc;
int *local_ranks_in_comm=NULL;
int *socket_info=NULL, my_socket_info;
int i_cnt, lp_cnt, my_local_index = -1, comm_size=ompi_comm_size(comm);
/* initialize data */
output_data=NULL;
my_rank=ompi_comm_rank(comm);
my_proc=ompi_comm_peer_lookup(comm,my_rank);
/*create a new module*/
module=OBJ_NEW(mca_sbgp_basesmsocket_module_t);
if (!module ) {
return NULL;
}
module->super.group_size=0;
module->super.group_comm = comm;
module->super.group_list = NULL;
module->super.group_net = OMPI_SBGP_SOCKET;
/* test to see if process is bound */
/* this needs to change. Process may have been bound by an
* some entity other than OPAL. Just because the binding
* policy isn't set doesn't mean it's not bound
*/
if( OPAL_BIND_TO_NONE == OPAL_GET_BINDING_POLICY(opal_hwloc_binding_policy) ) {
/* pa affinity not set, so socket index will be set to -1 */
my_socket_index=-1;
/*debug print*/
/* */
BASESMSOCKET_VERBOSE(10, ("[%d] FAILED to set basesmsocket group, processes are not bound!!!\n",my_rank));
/*end debug*/
goto NoLocalPeers;
} else {
my_socket_index=-1;
/* this should find my logical socket id which is the socket id we want
* physical socket ids are not necessarily unique, logical ones, as defined
* by the hwloc API are unique.
*/
if( OMPI_SUCCESS != mca_sbgp_map_to_logical_socket_id(&my_socket_index)){
BASESMSOCKET_VERBOSE(10, ("[%d] FAILED to set basesmsocket group !!!\n",my_rank));
goto NoLocalPeers;
}
}
/* Debug prints */
/*
{
fprintf(stderr,"Number of processors per node: %d\n",num_processors);
fprintf(stderr,"I am rank %d and my socket index is %d\n and my core index is %d\n",my_rank,my_socket_index,core_index);
fprintf(stderr,"n_proc_in = %d\n",n_procs_in);
fprintf(stderr,"\n");
fflush(stderr);
}
end debug prints */
/*get my socket index*/
cnt=0;
for( proc=0 ; proc < n_procs_in ; proc++) {
local=OPAL_PROC_ON_LOCAL_NODE(procs[proc]->proc_flags);
if( local ) {
cnt++;
}
}
/*debug print */
/*
fprintf(stderr,"Number of local processors %d\n",cnt);
end debug print*/
/* if no other local procs found skip to end */
if( 1 > cnt ) {
goto NoLocalPeers;
}
/* allocate structure to hold the list of local ranks */
local_ranks_in_comm=(int *)malloc(sizeof(int)*cnt);
if(NULL == local_ranks_in_comm ){
goto Error;
}
/* figure out which ranks from the input communicator - comm - will
* particiapte in the local socket determination.
*/
n_local_peers=0;
i_cnt=0;
for( proc = 0; proc < n_procs_in; proc++){
local = OPAL_PROC_ON_LOCAL_NODE(procs[proc]->proc_flags);
if ( local ) {
/* set the rank within the on-host ranks - this will be used for tha
* allgather
*/
if( my_proc == procs[proc] ) {
my_local_index=n_local_peers;
}
/* find the rank of the current proc in comm. We take advantage
* of the fact that ranks in a group have the same relative
* ordering as they do within the communicator.
*/
for( lp_cnt=proc; lp_cnt < comm_size ; lp_cnt++ ) {
if(procs[proc] == ompi_comm_peer_lookup(comm,lp_cnt) ){
local_ranks_in_comm[i_cnt]=lp_cnt;
/* lp_cnt has alrady been checked */
i_cnt++;
/* found the corresponding rank in comm, so don't need
* to search any more */
break;
}
/*i_cnt++;*/
/*fprintf(stderr,"QQQ i_cnt %d \n",i_cnt);*/
}
n_local_peers++;
}
}
/*fprintf(stderr,"YYY n_local_peers %d\n",n_local_peers);*/
socket_info=(int *)malloc(sizeof(int)*n_local_peers);
/*fprintf(stderr,"XXX got socket info\n");*/
if(NULL == socket_info ){
goto Error;
}
my_socket_info=my_socket_index;
/* Allgather data over the communicator */
ret=comm_allgather_pml(&my_socket_info, socket_info, 1,
MPI_INT, my_local_index, n_local_peers, local_ranks_in_comm,comm);
if (OMPI_SUCCESS != ret ) {
BASESMSOCKET_VERBOSE(10, ("comm_allgather_pml returned error %d\n",ret));
return NULL;
}
/*allocate memory to the group_list probably an overestimation
of the necessary resources */
module->super.group_list=(int *)malloc(sizeof(int)*cnt);
if(NULL == module->super.group_list){
goto Error;
}
/* figure out who is sharing the same socket */
cnt=0;
for (proc = 0; proc < n_local_peers; proc++) {
int rem_rank=local_ranks_in_comm[proc];
int rem_socket_index=socket_info[proc];
/*Populate the list*/
if (rem_socket_index == my_socket_index) {
module->super.group_list[cnt]=rem_rank;
cnt++;
}
}
module->super.group_size=cnt;
#if 0
/*debug print*/
/*
{
int ii;
fprintf(stderr,"Ranks per socket: %d\n",cnt);
fprintf(stderr,"Socket %d owns ranks: ", my_socket_index);
for (ii=0; ii < cnt; ii++)
fprintf(stderr,"%d ",module->super.group_list[ii]);
fprintf(stderr,"\n");
fflush(stderr);
}
*/
#endif
/* end debug*/
/*Free resources*/
free(local_ranks_in_comm);
free(socket_info);
/*Return the module*/
return (mca_sbgp_base_module_t *) module;
NoLocalPeers:
/* nothing to store, so just free the module and return */
/*fprintf(stderr,"No local socket peers\n");*/
/*free(module);*/
if(socket_info) {
free(socket_info);
socket_info=NULL;
}
if(local_ranks_in_comm){
free(local_ranks_in_comm);
}
OBJ_RELEASE(module);
return NULL;
Error:
/*clean up*/
if( NULL != module->super.group_list){
free(module->super.group_list);
module->super.group_list=NULL;
}
if(socket_info) {
free(socket_info);
socket_info=NULL;
}
if(local_ranks_in_comm){
free(local_ranks_in_comm);
}
OBJ_RELEASE(module);
return NULL;
}
#if 0
static int mca_sbgp_map_to_socket_core(int processor_id, int *socket, int *core)
{
int ret = OPAL_ERR_NOT_FOUND;
hwloc_obj_t obj;
hwloc_topology_t *t;
hwloc_bitmap_t good;
/* bozo check */
if (NULL == opal_hwloc_topology) {
return OPAL_ERR_NOT_INITIALIZED;
}
t = &opal_hwloc_topology;
good = hwloc_bitmap_alloc();
if (NULL == good) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
/* Iterate through every core and find one that contains the
processor_id. Then find the corresponding socket. */
for (obj = hwloc_get_next_obj_by_type(*t, HWLOC_OBJ_CORE, NULL);
NULL != obj;
obj = hwloc_get_next_obj_by_type(*t, HWLOC_OBJ_CORE, obj)) {
hwloc_bitmap_and(good, obj->online_cpuset,
obj->allowed_cpuset);
/* Does this core contain the processor_id in question? */
if (hwloc_bitmap_isset(good, processor_id)) {
*core = obj->os_index;
/* Go upward from the core object until we find its parent
socket. */
while (HWLOC_OBJ_SOCKET != obj->type) {
if (NULL == obj->parent) {
/* If we get to the root without finding a socket,
er.. Hmm. Error! */
ret = OPAL_ERR_NOT_FOUND;
goto out;
}
obj = obj->parent;
}
*socket = obj->os_index;
ret = OPAL_SUCCESS;
goto out;
}
}
/* If we didn't even find the right core, we didn't find it. Fall
through. */
ret = OPAL_ERR_NOT_FOUND;
out:
hwloc_bitmap_free(good);
return ret;
}
#endif
#if 0
static mca_sbgp_base_module_t *mca_sbgp_basesmsocket_select_procs(struct ompi_proc_t ** procs,
int n_procs_in,
struct ompi_communicator_t *comm,
char *key,
void *output_data
)
{
/* local variables */
mca_sbgp_basesmsocket_module_t *module;
/*
opal_buffer_t* sbuffer = OBJ_NEW(opal_buffer_t);
opal_buffer_t* rbuffer = OBJ_NEW(opal_buffer_t);
*/
opal_paffinity_base_cpu_set_t my_cpu_set;
bool bound;
int ret;
int num_processors;
int socket_tmp;
int my_socket_index;
int core_index=-1;
int proc, cnt, local, n_local_peers, my_index, my_rank;
ompi_proc_t* my_proc;
int *local_ranks_in_comm=NULL;
int *socket_info=NULL, my_socket_info;
int i_cnt, lp_cnt, my_local_index, comm_size=ompi_comm_size(comm);
/* initialize data */
output_data=NULL;
my_rank=ompi_comm_rank(comm);
my_proc=ompi_comm_peer_lookup(comm,my_rank);
for( proc=0 ; proc < n_procs_in ; proc++) {
if( procs[proc]==my_proc){
my_index=proc;
}
}
/*create a new module*/
module=OBJ_NEW(mca_sbgp_basesmsocket_module_t);
if (!module ) {
return NULL;
}
module->super.group_size=0;
module->super.group_comm = comm;
module->super.group_list = NULL;
module->super.group_net = OMPI_SBGP_SOCKET;
/*
** get my process affinity information
** */
/* get the number of processors on this node */
ret=opal_paffinity_base_get_processor_info(&num_processors);
/* get process affinity mask */
OPAL_PAFFINITY_CPU_ZERO(my_cpu_set);
ret=opal_paffinity_base_get(&my_cpu_set);
OPAL_PAFFINITY_PROCESS_IS_BOUND(my_cpu_set,&bound);
/*debug process affinity*/
/*
{
ret=opal_paffinity_base_get_socket_info(&num_socket);
fprintf(stderr,"Number of sockets %d\n",num_socket);
fprintf(stderr,"Test if rank %d is bound %d\n", my_rank, bound);
fprintf(stderr,"return from opal_paffinity_base_get: %d\n\n",ret);
fprintf(stderr,"bitmask elements: ");
unsigned int long jj;
for(jj=0; jj < OPAL_PAFFINITY_BITMASK_NUM_ELEMENTS; jj++)
fprintf(stderr," %d ",my_cpu_set.bitmask[jj]);
fprintf(stderr,"\n");
fflush(stderr);
}
end debug process affinity*/
if( !bound ) {
/* pa affinity not set, so socket index will be set to -1 */
my_socket_index=-1;
/*debug print*/
/* */
fprintf(stderr,"[%d]FAILED to set basesmsocket group !!!\n",my_rank);
fflush(stderr);
/*end debug*/
goto NoLocalPeers;
} else {
my_socket_index=-1;
/* loop over number of processors */
for ( proc=0 ; proc < num_processors ; proc++ ) {
if (OPAL_PAFFINITY_CPU_ISSET(proc,my_cpu_set)) {
ret=opal_paffinity_base_get_map_to_socket_core(proc,&socket_tmp,&core_index);
if( my_socket_index != socket_tmp ) {
my_socket_index=socket_tmp;
break;
}
}
} /* end of proc loop */
}
/* Debug prints */
/*
{
fprintf(stderr,"Number of processors per node: %d\n",num_processors);
fprintf(stderr,"I am rank %d and my socket index is %d\n and my core index is %d\n",my_rank,my_socket_index,core_index);
fprintf(stderr,"n_proc_in = %d\n",n_procs_in);
fprintf(stderr,"\n");
fflush(stderr);
}
end debug prints */
/*get my socket index*/
cnt=0;
for( proc=0 ; proc < n_procs_in ; proc++) {
local=OPAL_PROC_ON_LOCAL_NODE(procs[proc]->proc_flags);
if( local ) {
cnt++;
}
}
/*debug print */
/*
fprintf(stderr,"Number of local processors %d\n",cnt);
end debug print*/
/* if no other local procs found skip to end */
if( 1 >= cnt ) {
goto NoLocalPeers;
}
#if 0
int *local_ranks_in_comm;
int32_t *socket_info, *my_socket_info;
int my_local_index;
#endif
/* allocate structure to hold the list of local ranks */
local_ranks_in_comm=(int *)malloc(sizeof(int)*cnt);
if(NULL == local_ranks_in_comm ){
goto Error;
}
/* figure out which ranks from the input communicator - comm - will
* particiapte in the local socket determination.
*/
n_local_peers=0;
i_cnt=0;
for( proc = 0; proc < n_procs_in; proc++){
local = OPAL_PROC_ON_LOCAL_NODE(procs[proc]->proc_flags);
if ( local ) {
/* set the rank within the on-host ranks - this will be used for tha
* allgather
*/
if( my_proc == procs[proc] ) {
my_local_index=n_local_peers;
}
/* find the rank of the current proc in comm. We take advantage
* of the fact that ranks in a group have the same relative
* ordering as they do within the communicator.
*/
#if 1
/*for( lp_cnt=i_cnt; lp_cnt < comm_size ; lp_cnt++ ) {*/
for( lp_cnt=proc; lp_cnt < comm_size ; lp_cnt++ ) {
if(procs[proc] == ompi_comm_peer_lookup(comm,lp_cnt) ){
local_ranks_in_comm[i_cnt]=lp_cnt;
/* lp_cnt has alrady been checked */
i_cnt++;
/* found the corresponding rank in comm, so don't need
* to search any more */
break;
}
/*i_cnt++;*/
/*fprintf(stderr,"QQQ i_cnt %d \n",i_cnt);*/
}
#endif
n_local_peers++;
}
}
/*fprintf(stderr,"YYY n_local_peers %d\n",n_local_peers);*/
socket_info=(int *)malloc(sizeof(int)*n_local_peers);
/*fprintf(stderr,"XXX got socket info\n");*/
if(NULL == socket_info ){
goto Error;
}
my_socket_info=my_socket_index;
/* Allgather data over the communicator */
ret=comm_allgather_pml(&my_socket_info, socket_info, 1,
MPI_INT, my_local_index, n_local_peers, local_ranks_in_comm,comm);
if (OMPI_SUCCESS != ret ) {
fprintf(stderr," comm_allgather_pml returned error %d \n", ret);
fflush(stderr);
return NULL;
}
/*allocate memory to the group_list probably an overestimation
of the necessary resources */
module->super.group_list=(int *)malloc(sizeof(int)*cnt);
if(NULL == module->super.group_list){
goto Error;
}
/* figure out who is sharing the same socket */
cnt=0;
for (proc = 0; proc < n_local_peers; proc++) {
int rem_rank=local_ranks_in_comm[proc];
int rem_socket_index=socket_info[proc];
/*Populate the list*/
if (rem_socket_index == my_socket_index) {
module->super.group_list[cnt]=rem_rank;
cnt++;
}
}
module->super.group_size=cnt;
/*debug print*/
/*
{
int ii;
fprintf(stderr,"Ranks per socket: %d\n",cnt);
fprintf(stderr,"Socket %d owns ranks: ", my_socket_index);
for (ii=0; ii < cnt; ii++)
fprintf(stderr,"%d ",module->super.group_list[ii]);
fprintf(stderr,"\n");
fflush(stderr);
}
{
cpu_set_t set;
unsigned int len = sizeof(set);
int i;
unsigned long mask = 0;
CPU_ZERO(&set);
if (sched_getaffinity(0, len, &set) < 0) {
perror("sched_getaffinity");
return -1;
}
for (i = 0; i < CPU_SETSIZE; i++) {
int cpu = CPU_ISSET(i, &set);
if (cpu) {
mask |= 1<< i;
}
}
opal_output(0,"%d: my affinity mask is: %08lx\n", my_local_index,mask);
}
end debug*/
/*Free resources*/
free(local_ranks_in_comm);
free(socket_info);
/*Return the module*/
return (mca_sbgp_base_module_t *) module;
NoLocalPeers:
/* nothing to store, so just free the module and return */
/*fprintf(stderr,"No local socket peers\n");*/
/*free(module);*/
if(socket_info) {
free(socket_info);
socket_info=NULL;
}
if(local_ranks_in_comm){
free(local_ranks_in_comm);
}
OBJ_RELEASE(module);
return NULL;
Error:
/*clean up*/
if( NULL != module->super.group_list){
free(module->super.group_list);
module->super.group_list=NULL;
}
if(socket_info) {
free(socket_info);
socket_info=NULL;
}
if(local_ranks_in_comm){
free(local_ranks_in_comm);
}
OBJ_RELEASE(module);
return NULL;
}
#endif
Показать git blame Копировать постоянную ссылку