1
1
openmpi/ompi/mca/coll/tuned/coll_tuned_barrier.c
Rolf vandeVaart 74d0259480 Add new implentation of barrier. This shows better performance on some clusters.
However, no decision logic is changed by this commit so default behavior has not changed.  This
is only selectable by runtime parameters.

This commit was SVN r18464.
2008-05-20 17:37:41 +00:00

496 строки
18 KiB
C

/*
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2006 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2008 Sun Microsystems, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/op/op.h"
#include "coll_tuned.h"
#include "coll_tuned_topo.h"
#include "coll_tuned_util.h"
/*
* Barrier is ment to be a synchronous operation, as some BTLs can mark
* a request done before its passed to the NIC and progress might not be made
* elsewhere we cannot allow a process to exit the barrier until its last
* [round of] sends are completed.
*
* It is last round of sends rather than 'last' individual send as each pair of
* peers can use different channels/devices/btls and the receiver of one of
* these sends might be forced to wait as the sender
* leaves the collective and does not make progress until the next mpi call
*
*/
/*
* Simple double ring version of barrier
*
* synchronous gurantee made by last ring of sends are synchronous
*
*/
int ompi_coll_tuned_barrier_intra_doublering(struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module)
{
int rank, size;
int err=0, line=0;
int left, right;
rank = ompi_comm_rank(comm);
size = ompi_comm_size(comm);
ORTE_OUTPUT((ompi_coll_tuned_stream,"ompi_coll_tuned_barrier_intra_doublering rank %d", rank));
left = ((rank-1)%size);
right = ((rank+1)%size);
if (rank > 0) { /* receive message from the left */
err = MCA_PML_CALL(recv((void*)NULL, 0, MPI_BYTE, left,
MCA_COLL_BASE_TAG_BARRIER, comm,
MPI_STATUS_IGNORE));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl; }
}
/* Send message to the right */
err = MCA_PML_CALL(send((void*)NULL, 0, MPI_BYTE, right,
MCA_COLL_BASE_TAG_BARRIER,
MCA_PML_BASE_SEND_STANDARD, comm));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl; }
/* root needs to receive from the last node */
if (rank == 0) {
err = MCA_PML_CALL(recv((void*)NULL, 0, MPI_BYTE, left,
MCA_COLL_BASE_TAG_BARRIER, comm,
MPI_STATUS_IGNORE));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl; }
}
/* Allow nodes to exit */
if (rank > 0) { /* post Receive from left */
err = MCA_PML_CALL(recv((void*)NULL, 0, MPI_BYTE, left,
MCA_COLL_BASE_TAG_BARRIER, comm,
MPI_STATUS_IGNORE));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl; }
}
/* send message to the right one */
err = MCA_PML_CALL(send((void*)NULL, 0, MPI_BYTE, right,
MCA_COLL_BASE_TAG_BARRIER,
MCA_PML_BASE_SEND_SYNCHRONOUS, comm));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl; }
/* rank 0 post receive from the last node */
if (rank == 0) {
err = MCA_PML_CALL(recv((void*)NULL, 0, MPI_BYTE, left,
MCA_COLL_BASE_TAG_BARRIER, comm,
MPI_STATUS_IGNORE));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl; }
}
return MPI_SUCCESS;
err_hndl:
ORTE_OUTPUT((ompi_coll_tuned_stream,"%s:%4d\tError occurred %d, rank %2d",
__FILE__, line, err, rank));
return err;
}
/*
* To make synchronous, uses sync sends and sync sendrecvs
*/
int ompi_coll_tuned_barrier_intra_recursivedoubling(struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module)
{
int rank, size, adjsize;
int err, line;
int mask, remote;
rank = ompi_comm_rank(comm);
size = ompi_comm_size(comm);
ORTE_OUTPUT((ompi_coll_tuned_stream,
"ompi_coll_tuned_barrier_intra_recursivedoubling rank %d",
rank));
/* do nearest power of 2 less than size calc */
for( adjsize = 1; adjsize <= size; adjsize <<= 1 );
adjsize >>= 1;
/* if size is not exact power of two, perform an extra step */
if (adjsize != size) {
if (rank >= adjsize) {
/* send message to lower ranked node */
remote = rank - adjsize;
err = ompi_coll_tuned_sendrecv_actual(NULL, 0, MPI_BYTE, remote,
MCA_COLL_BASE_TAG_BARRIER,
NULL, 0, MPI_BYTE, remote,
MCA_COLL_BASE_TAG_BARRIER,
comm, MPI_STATUS_IGNORE);
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl;}
} else if (rank < (size - adjsize)) {
/* receive message from high level rank */
err = MCA_PML_CALL(recv((void*)NULL, 0, MPI_BYTE, rank+adjsize,
MCA_COLL_BASE_TAG_BARRIER, comm,
MPI_STATUS_IGNORE));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl;}
}
}
/* exchange messages */
if ( rank < adjsize ) {
mask = 0x1;
while ( mask < adjsize ) {
remote = rank ^ mask;
mask <<= 1;
if (remote >= adjsize) continue;
/* post receive from the remote node */
err = ompi_coll_tuned_sendrecv_actual(NULL, 0, MPI_BYTE, remote,
MCA_COLL_BASE_TAG_BARRIER,
NULL, 0, MPI_BYTE, remote,
MCA_COLL_BASE_TAG_BARRIER,
comm, MPI_STATUS_IGNORE);
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl;}
}
}
/* non-power of 2 case */
if (adjsize != size) {
if (rank < (size - adjsize)) {
/* send enter message to higher ranked node */
remote = rank + adjsize;
err = MCA_PML_CALL(send((void*)NULL, 0, MPI_BYTE, remote,
MCA_COLL_BASE_TAG_BARRIER,
MCA_PML_BASE_SEND_SYNCHRONOUS, comm));
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl;}
}
}
return MPI_SUCCESS;
err_hndl:
ORTE_OUTPUT((ompi_coll_tuned_stream,"%s:%4d\tError occurred %d, rank %2d",
__FILE__, line, err, rank));
return err;
}
/*
* To make synchronous, uses sync sends and sync sendrecvs
*/
int ompi_coll_tuned_barrier_intra_bruck(struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module)
{
int rank, size;
int distance, to, from;
int err, line = 0;
rank = ompi_comm_rank(comm);
size = ompi_comm_size(comm);
ORTE_OUTPUT((ompi_coll_tuned_stream,
"ompi_coll_tuned_barrier_intra_bruck rank %d", rank));
/* exchange data with rank-2^k and rank+2^k */
for (distance = 1; distance < size; distance <<= 1) {
from = (rank + size - distance) % size;
to = (rank + distance) % size;
/* send message to lower ranked node */
err = ompi_coll_tuned_sendrecv_actual(NULL, 0, MPI_BYTE, to,
MCA_COLL_BASE_TAG_BARRIER,
NULL, 0, MPI_BYTE, from,
MCA_COLL_BASE_TAG_BARRIER,
comm, MPI_STATUS_IGNORE);
if (err != MPI_SUCCESS) { line = __LINE__; goto err_hndl;}
}
return MPI_SUCCESS;
err_hndl:
ORTE_OUTPUT((ompi_coll_tuned_stream,"%s:%4d\tError occurred %d, rank %2d",
__FILE__, line, err, rank));
return err;
}
/*
* To make synchronous, uses sync sends and sync sendrecvs
*/
/* special case for two processes */
int ompi_coll_tuned_barrier_intra_two_procs(struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module)
{
int remote, err;
remote = ompi_comm_rank(comm);
ORTE_OUTPUT((ompi_coll_tuned_stream,
"ompi_coll_tuned_barrier_intra_two_procs rank %d", remote));
remote = (remote + 1) & 0x1;
err = ompi_coll_tuned_sendrecv_actual(NULL, 0, MPI_BYTE, remote,
MCA_COLL_BASE_TAG_BARRIER,
NULL, 0, MPI_BYTE, remote,
MCA_COLL_BASE_TAG_BARRIER,
comm, MPI_STATUS_IGNORE);
return (err);
}
/*
* Linear functions are copied from the BASIC coll module
* they do not segment the message and are simple implementations
* but for some small number of nodes and/or small data sizes they
* are just as fast as tuned/tree based segmenting operations
* and as such may be selected by the decision functions
* These are copied into this module due to the way we select modules
* in V1. i.e. in V2 we will handle this differently and so will not
* have to duplicate code.
* GEF Oct05 after asking Jeff.
*/
/* copied function (with appropriate renaming) starts here */
static int ompi_coll_tuned_barrier_intra_basic_linear(struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module)
{
int i, err;
int size = ompi_comm_size(comm);
int rank = ompi_comm_rank(comm);
/* All non-root send & receive zero-length message. */
if (rank > 0) {
err = MCA_PML_CALL(send (NULL, 0, MPI_BYTE, 0,
MCA_COLL_BASE_TAG_BARRIER,
MCA_PML_BASE_SEND_STANDARD, comm));
if (MPI_SUCCESS != err) {
return err;
}
err = MCA_PML_CALL(recv (NULL, 0, MPI_BYTE, 0,
MCA_COLL_BASE_TAG_BARRIER,
comm, MPI_STATUS_IGNORE));
if (MPI_SUCCESS != err) {
return err;
}
}
/* The root collects and broadcasts the messages. */
else {
ompi_request_t** requests;
requests = (ompi_request_t**)malloc( size * sizeof(ompi_request_t*) );
for (i = 1; i < size; ++i) {
err = MCA_PML_CALL(irecv(NULL, 0, MPI_BYTE, MPI_ANY_SOURCE,
MCA_COLL_BASE_TAG_BARRIER, comm,
&(requests[i])));
if (MPI_SUCCESS != err) {
return err;
}
}
ompi_request_wait_all( size-1, requests+1, MPI_STATUSES_IGNORE );
for (i = 1; i < size; ++i) {
err = MCA_PML_CALL(isend(NULL, 0, MPI_BYTE, i,
MCA_COLL_BASE_TAG_BARRIER,
MCA_PML_BASE_SEND_STANDARD, comm,
&(requests[i])));
if (MPI_SUCCESS != err) {
return err;
}
}
ompi_request_wait_all( size-1, requests+1, MPI_STATUSES_IGNORE );
free( requests );
}
/* All done */
return MPI_SUCCESS;
}
/* copied function (with appropriate renaming) ends here */
/*
* Another recursive doubling type algorithm, but in this case
* we go up the tree and back down the tree.
*/
int ompi_coll_tuned_barrier_intra_tree(struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module)
{
int rank, size, depth;
int err, line;
int jump, partner;
rank = ompi_comm_rank(comm);
size = ompi_comm_size(comm);
OPAL_OUTPUT((ompi_coll_tuned_stream,
"ompi_coll_tuned_barrier_intra_tree %d",
rank));
/* Find the nearest power of 2 of the communicator size. */
for(depth = 1; depth < size; depth <<= 1 );
for (jump=1; jump<depth; jump<<=1) {
partner = rank ^ jump;
if (!(partner & (jump-1)) && partner < size) {
if (partner > rank) {
err = MCA_PML_CALL(recv (NULL, 0, MPI_BYTE, partner,
MCA_COLL_BASE_TAG_BARRIER, comm,
MPI_STATUS_IGNORE));
if (MPI_SUCCESS != err)
return err;
} else if (partner < rank) {
err = MCA_PML_CALL(send (NULL, 0, MPI_BYTE, partner,
MCA_COLL_BASE_TAG_BARRIER,
MCA_PML_BASE_SEND_STANDARD, comm));
if (MPI_SUCCESS != err)
return err;
}
}
}
depth>>=1;
for (jump = depth; jump>0; jump>>=1) {
partner = rank ^ jump;
if (!(partner & (jump-1)) && partner < size) {
if (partner > rank) {
err = MCA_PML_CALL(send (NULL, 0, MPI_BYTE, partner,
MCA_COLL_BASE_TAG_BARRIER,
MCA_PML_BASE_SEND_STANDARD, comm));
if (MPI_SUCCESS != err)
return err;
} else if (partner < rank) {
err = MCA_PML_CALL(recv (NULL, 0, MPI_BYTE, partner,
MCA_COLL_BASE_TAG_BARRIER, comm,
MPI_STATUS_IGNORE));
if (MPI_SUCCESS != err)
return err;
}
}
}
return MPI_SUCCESS;
}
/* The following are used by dynamic and forced rules */
/* publish details of each algorithm and if its forced/fixed/locked in */
/* as you add methods/algorithms you must update this and the query/map */
/* routines */
/* this routine is called by the component only */
/* this makes sure that the mca parameters are set to their initial values */
/* and perms */
/* module does not call this they call the forced_getvalues routine instead */
int ompi_coll_tuned_barrier_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
int rc, max_alg = 6, requested_alg;
ompi_coll_tuned_forced_max_algorithms[BARRIER] = max_alg;
rc = mca_base_param_reg_int (&mca_coll_tuned_component.super.collm_version,
"barrier_algorithm_count",
"Number of barrier algorithms available",
false, true, max_alg, NULL);
mca_param_indices->algorithm_param_index =
mca_base_param_reg_int(&mca_coll_tuned_component.super.collm_version,
"barrier_algorithm",
"Which barrier algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 double ring, 3: recursive doubling 4: bruck, 5: two proc only, 6: tree",
false, false, 0, NULL);
mca_base_param_lookup_int(mca_param_indices->algorithm_param_index,
&(requested_alg));
if( requested_alg > max_alg ) {
if( 0 == ompi_comm_rank( MPI_COMM_WORLD ) ) {
orte_output( 0, "Barrier algorithm #%d is not available (range [0..%d]). Switching back to ignore(0)\n",
requested_alg, max_alg );
}
mca_base_param_set_int( mca_param_indices->algorithm_param_index, 0);
}
return (MPI_SUCCESS);
}
int ompi_coll_tuned_barrier_intra_do_forced(struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module)
{
mca_coll_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
mca_coll_tuned_comm_t *data = tuned_module->tuned_data;
ORTE_OUTPUT((ompi_coll_tuned_stream,
"coll:tuned:barrier_intra_do_forced selected algorithm %d",
data->user_forced[BARRIER].algorithm));
switch (data->user_forced[BARRIER].algorithm) {
case (0): return ompi_coll_tuned_barrier_intra_dec_fixed (comm, module);
case (1): return ompi_coll_tuned_barrier_intra_basic_linear (comm, module);
case (2): return ompi_coll_tuned_barrier_intra_doublering (comm, module);
case (3): return ompi_coll_tuned_barrier_intra_recursivedoubling (comm, module);
case (4): return ompi_coll_tuned_barrier_intra_bruck (comm, module);
case (5): return ompi_coll_tuned_barrier_intra_two_procs (comm, module);
case (6): return ompi_coll_tuned_barrier_intra_tree (comm, module);
default:
ORTE_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:barrier_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
data->user_forced[BARRIER].algorithm,
ompi_coll_tuned_forced_max_algorithms[BARRIER]));
return (MPI_ERR_ARG);
} /* switch */
}
int ompi_coll_tuned_barrier_intra_do_this (struct ompi_communicator_t *comm,
struct mca_coll_base_module_1_1_0_t *module,
int algorithm, int faninout, int segsize)
{
ORTE_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:barrier_intra_do_this selected algorithm %d topo fanin/out%d", algorithm, faninout));
switch (algorithm) {
case (0): return ompi_coll_tuned_barrier_intra_dec_fixed (comm, module);
case (1): return ompi_coll_tuned_barrier_intra_basic_linear (comm, module);
case (2): return ompi_coll_tuned_barrier_intra_doublering (comm, module);
case (3): return ompi_coll_tuned_barrier_intra_recursivedoubling (comm, module);
case (4): return ompi_coll_tuned_barrier_intra_bruck (comm, module);
case (5): return ompi_coll_tuned_barrier_intra_two_procs (comm, module);
case (6): return ompi_coll_tuned_barrier_intra_tree (comm, module);
default:
ORTE_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:barrier_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[BARRIER]));
return (MPI_ERR_ARG);
} /* switch */
}