modifications towards adding new local-leader communicators.
This commit was SVN r7760.
Этот коммит содержится в:
Edgar Gabriel
родитель
636b42afff
Коммит
770aeaf97b
@ -37,11 +37,19 @@
|
||||
|
||||
|
||||
/* local functions and data */
|
||||
#define HIER_MAXPROTOCOL 7
|
||||
/* #define HIER_MAXPROTOCOL 7
|
||||
static int mca_coll_hierarch_max_protocol=HIER_MAXPROTOCOL;
|
||||
|
||||
static char hier_prot[HIER_MAXPROTOCOL][7]={"0","tcp","gm","mx","mvapi","openib","sm"};
|
||||
|
||||
*/
|
||||
|
||||
#define HIER_MAXPROTOCOL 4
|
||||
static int mca_coll_hierarch_max_protocol=HIER_MAXPROTOCOL;
|
||||
static char hier_prot[HIER_MAXPROTOCOL][7]={"0","tcp","openib","sm"};
|
||||
|
||||
|
||||
|
||||
static void mca_coll_hierarch_checkfor_component (struct ompi_communicator_t *comm,
|
||||
int component_level,
|
||||
char *component_name,
|
||||
@ -154,6 +162,7 @@ mca_coll_hierarch_comm_query(struct ompi_communicator_t *comm, int *priority,
|
||||
struct mca_coll_base_comm_t *tdata=NULL;
|
||||
int level;
|
||||
int ret=OMPI_SUCCESS;
|
||||
int ignore_sm=0;
|
||||
|
||||
/* Get the priority level attached to this module */
|
||||
if (OMPI_SUCCESS != mca_base_param_lookup_int(mca_coll_hierarch_priority_param,
|
||||
@ -161,6 +170,13 @@ mca_coll_hierarch_comm_query(struct ompi_communicator_t *comm, int *priority,
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* Check whether we should ignore sm. This might be necessary to take advantage
|
||||
of the some ib or gm collectives. */
|
||||
if (OMPI_SUCCESS != mca_base_param_lookup_int(mca_coll_hierarch_ignore_sm_param,
|
||||
&ignore_sm)) {
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* This module only works for intra-communicators at the moment */
|
||||
if ( OMPI_COMM_IS_INTER(comm) ) {
|
||||
return NULL;
|
||||
@ -189,6 +205,9 @@ mca_coll_hierarch_comm_query(struct ompi_communicator_t *comm, int *priority,
|
||||
* is feasable.
|
||||
* Later we start with level=0, and introduce the multi-cell check
|
||||
*/
|
||||
if ( ignore_sm ) {
|
||||
mca_coll_hierarch_max_protocol = HIER_MAXPROTOCOL - 1;
|
||||
}
|
||||
for ( level = 1; level < mca_coll_hierarch_max_protocol; level++) {
|
||||
mca_coll_hierarch_checkfor_component ( comm,
|
||||
level,
|
||||
@ -289,8 +308,8 @@ const struct mca_coll_base_module_1_0_0_t *
|
||||
mca_coll_hierarch_module_init(struct ompi_communicator_t *comm)
|
||||
{
|
||||
int color;
|
||||
int *llr=NULL;
|
||||
int size, rank, ret=OMPI_SUCCESS;
|
||||
int *llr=NULL;
|
||||
|
||||
struct ompi_communicator_t *lcomm=NULL;
|
||||
struct ompi_communicator_t *llcomm=NULL;
|
||||
@ -329,7 +348,13 @@ mca_coll_hierarch_module_init(struct ompi_communicator_t *comm)
|
||||
data->hier_num_llead = 1;
|
||||
|
||||
/* determine how many local leader there are and who they are */
|
||||
data->hier_llead[0].num_lleaders = mca_coll_hierarch_count_lleaders (size, data->hier_colorarr);
|
||||
llr = (int *) malloc ( sizeof(int) * size);
|
||||
if ( NULL == llr ) {
|
||||
goto exit;
|
||||
}
|
||||
data->hier_llead[0].num_lleaders = mca_coll_hierarch_count_lleaders (size,
|
||||
data->hier_colorarr,
|
||||
llr);
|
||||
data->hier_llead[0].lleaders = (int *) malloc (sizeof(int) * data->hier_llead[0].num_lleaders);
|
||||
if ( NULL == data->hier_llead[0].lleaders ) {
|
||||
goto exit;
|
||||
@ -337,7 +362,9 @@ mca_coll_hierarch_module_init(struct ompi_communicator_t *comm)
|
||||
mca_coll_hierarch_get_all_lleaders ( data->hier_num_colorarr,
|
||||
data->hier_colorarr,
|
||||
data->hier_llead[0].num_lleaders,
|
||||
data->hier_llead[0].lleaders );
|
||||
llr,
|
||||
data->hier_llead[0].lleaders,
|
||||
1 );
|
||||
/* determine my lleader, maybe its me */
|
||||
data->hier_llead[0].am_lleader=0; /* false */
|
||||
mca_coll_hierarch_get_lleader ( rank, data, &(data->hier_llead[0].my_lleader) );
|
||||
@ -464,14 +491,26 @@ struct ompi_communicator_t* mca_coll_hierarch_get_llcomm (int rank,
|
||||
}
|
||||
|
||||
if ( !found ) {
|
||||
int offset;
|
||||
/* Here we have to introduce later on the code how to create the new
|
||||
lleader intercommunicators. For the moment, we just return a NULL communicator.
|
||||
*/
|
||||
llcomm = MPI_COMM_NULL;
|
||||
}
|
||||
/* ompi_group_decrement_proc_count (group);
|
||||
OBJ_RELEASE(group); */
|
||||
#if 0
|
||||
offset = mca_coll_hierarch_get_offset ( root, colorarr );
|
||||
|
||||
/* determine what our offset of root is in the colorarr */
|
||||
/* generate the list of lleaders with this offset */
|
||||
/* allocate a new llead element */
|
||||
/* create new lleader subcommunicator */
|
||||
/* set the structure */
|
||||
#endif
|
||||
|
||||
}
|
||||
|
||||
/* ompi_group_decrement_proc_count (group);
|
||||
OBJ_RELEASE(group);
|
||||
*/
|
||||
|
||||
return llcomm;
|
||||
}
|
||||
|
||||
@ -37,6 +37,7 @@ extern int mca_coll_hierarch_priority_param;
|
||||
extern int mca_coll_hierarch_verbose_param;
|
||||
extern int mca_coll_hierarch_verbose;
|
||||
extern int mca_coll_hierarch_use_rdma_param;
|
||||
extern int mca_coll_hierarch_ignore_sm_param;
|
||||
|
||||
|
||||
#define HIER_DEFAULT_NUM_LLEAD 5
|
||||
@ -115,15 +116,10 @@ extern int mca_coll_hierarch_use_rdma_param;
|
||||
#define MCA_COLL_HIERARCH_COLORARR_STRIDE2 2
|
||||
#define MCA_COLL_HIERARCH_COLORARR_STRIDE4 3
|
||||
|
||||
static inline int mca_coll_hierarch_count_lleaders ( int size, int *carr )
|
||||
static inline int mca_coll_hierarch_count_lleaders ( int size, int *carr, int *llr )
|
||||
{
|
||||
int cnt, i, j, found;
|
||||
int *llr=NULL;
|
||||
|
||||
llr = (int *) calloc (1, sizeof(int) * size);
|
||||
if (NULL == llr ) {
|
||||
return -1;
|
||||
}
|
||||
|
||||
for ( i=0; i<size; i++ ) {
|
||||
if ( carr[i] != MPI_UNDEFINED ) {
|
||||
@ -145,33 +141,30 @@ static inline int mca_coll_hierarch_count_lleaders ( int size, int *carr )
|
||||
}
|
||||
}
|
||||
|
||||
free (llr);
|
||||
return cnt;
|
||||
}
|
||||
|
||||
static inline void mca_coll_hierarch_get_all_lleaders ( int size, int *carr, int lsize,
|
||||
int *larr)
|
||||
int *llr, int *larr, int offset)
|
||||
{
|
||||
int i, j, cnt, found;
|
||||
int i, j, k;
|
||||
int *cntarr=NULL;
|
||||
|
||||
for ( i=0; i<size; i++ ) {
|
||||
if ( carr[i] != MPI_UNDEFINED ) {
|
||||
larr[0] = carr[i];
|
||||
cntarr = (int *)calloc (1, sizeof (int)*lsize );
|
||||
if ( NULL == cntarr ) {
|
||||
return;
|
||||
}
|
||||
|
||||
for ( k=0, i=0;i<size;i++) {
|
||||
for ( j=0; j<lsize; j++) {
|
||||
if ( carr[i] == llr[j] && cntarr[j] < offset) {
|
||||
cntarr[j]++;
|
||||
larr[j] = i;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
for (cnt=1, i=0; i<size; i++ ) {
|
||||
for ( found=0, j=0; j<cnt; j++ ) {
|
||||
if ( carr[i] == larr[j] ) {
|
||||
found = 1;
|
||||
break;
|
||||
}
|
||||
}
|
||||
if ( !found && (MPI_UNDEFINED != carr[i]) ) {
|
||||
larr[cnt++] = carr[i];
|
||||
}
|
||||
}
|
||||
|
||||
return;
|
||||
}
|
||||
|
||||
@ -39,6 +39,7 @@ int mca_coll_hierarch_priority_param = -1;
|
||||
int mca_coll_hierarch_verbose_param = -1;
|
||||
int mca_coll_hierarch_verbose = 0;
|
||||
int mca_coll_hierarch_use_rdma_param=-1;
|
||||
int mca_coll_hierarch_ignore_sm_param=-1;
|
||||
|
||||
|
||||
/*
|
||||
@ -101,5 +102,8 @@ static int hierarch_open(void)
|
||||
mca_coll_hierarch_use_rdma_param =
|
||||
mca_base_param_register_int("coll", "hierarch", "use_rdma", NULL, 0);
|
||||
|
||||
mca_coll_hierarch_ignore_sm_param =
|
||||
mca_base_param_register_int("coll", "hierarch", "ignore_sm", NULL, 0);
|
||||
|
||||
return OMPI_SUCCESS;
|
||||
}
|
||||
|
||||
Загрузка…
Ссылка в новой задаче
Block a user