64321acc22
opal does not always define MB. It is recommended that opal_atomic_[rw]mb is called instead. We will need to address the cases where these functions are no-ops on weak-memory ordered cpus. cmr=v1.7.5:ticket=trac:4158 This commit was SVN r30534. The following Trac tickets were found above: Ticket 4158 --> https://svn.open-mpi.org/trac/ompi/ticket/4158
476 строки
16 KiB
C
476 строки
16 KiB
C
/*
|
|
* Copyright (c) 2009-2012 Oak Ridge National Laboratory. All rights reserved.
|
|
* Copyright (c) 2009-2012 Mellanox Technologies. All rights reserved.
|
|
* Copyright (c) 2014 Los Alamos National Security, LLC. All rights
|
|
* reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
|
|
#include "ompi_config.h"
|
|
|
|
#include "ompi/constants.h"
|
|
#include "ompi/datatype/ompi_datatype.h"
|
|
#include "ompi/communicator/communicator.h"
|
|
|
|
#include "bcol_basesmuma.h"
|
|
|
|
#define __TEST_BLOCKING__ 1
|
|
#define __TEST_WAIT__ 0
|
|
#define __TEST_TEST__ 0
|
|
|
|
/* debug
|
|
* #include "opal/sys/timer.h"
|
|
*
|
|
* extern uint64_t timers[7];
|
|
* end debug */
|
|
|
|
/* debug */
|
|
/* end debug */
|
|
int bcol_basesmuma_bcast_init(mca_bcol_base_module_t *super)
|
|
{
|
|
mca_bcol_base_coll_fn_comm_attributes_t comm_attribs;
|
|
mca_bcol_base_coll_fn_invoke_attributes_t inv_attribs;
|
|
|
|
comm_attribs.bcoll_type = BCOL_BCAST;
|
|
comm_attribs.comm_size_min = 0;
|
|
comm_attribs.comm_size_max = 64;
|
|
comm_attribs.data_src = DATA_SRC_KNOWN;
|
|
comm_attribs.waiting_semantics = NON_BLOCKING;
|
|
|
|
inv_attribs.bcol_msg_min = 0;
|
|
inv_attribs.bcol_msg_max = 20000; /* range 1 */
|
|
inv_attribs.datatype_bitmap = 0xffffffff;
|
|
inv_attribs.op_types_bitmap = 0xffffffff;
|
|
|
|
mca_bcol_base_set_attributes(super, &comm_attribs, &inv_attribs,
|
|
bcol_basesmuma_bcast_k_nomial_knownroot,
|
|
bcol_basesmuma_bcast_k_nomial_knownroot);
|
|
|
|
comm_attribs.data_src = DATA_SRC_UNKNOWN;
|
|
mca_bcol_base_set_attributes(super, &comm_attribs, &inv_attribs,
|
|
bcol_basesmuma_bcast_k_nomial_anyroot,
|
|
bcol_basesmuma_bcast_k_nomial_anyroot);
|
|
|
|
comm_attribs.data_src = DATA_SRC_UNKNOWN;
|
|
inv_attribs.bcol_msg_min = 10000000;
|
|
inv_attribs.bcol_msg_max = 10485760; /* range 4 */
|
|
|
|
#ifdef __PORTALS_AVAIL__
|
|
|
|
comm_attribs.waiting_semantics = BLOCKING;
|
|
mca_bcol_base_set_attributes(super, &comm_attribs, &inv_attribs,
|
|
bcol_basesmuma_lmsg_scatter_allgather_portals_bcast,
|
|
bcol_basesmuma_lmsg_scatter_allgather_portals_bcast);
|
|
|
|
|
|
comm_attribs.waiting_semantics = NON_BLOCKING;
|
|
mca_bcol_base_set_attributes(super, &comm_attribs, &inv_attribs,
|
|
bcol_basesmuma_lmsg_scatter_allgather_portals_nb_bcast,
|
|
bcol_basesmuma_lmsg_scatter_allgather_portals_nb_bcast);
|
|
|
|
comm_attribs.data_src = DATA_SRC_KNOWN;
|
|
mca_bcol_base_set_attributes(super, &comm_attribs, &inv_attribs,
|
|
bcol_basesmuma_lmsg_scatter_allgather_portals_nb_knownroot_bcast,
|
|
bcol_basesmuma_lmsg_scatter_allgather_portals_nb_knownroot_bcast);
|
|
|
|
#else
|
|
/*
|
|
if (super->use_hdl) {
|
|
mca_bcol_base_set_attributes(super, &comm_attribs, &inv_attribs,
|
|
bcol_basesmuma_hdl_zerocopy_bcast,
|
|
bcol_basesmuma_hdl_zerocopy_bcast);
|
|
} else { */
|
|
mca_bcol_base_set_attributes(super, &comm_attribs, &inv_attribs,
|
|
NULL,
|
|
NULL);
|
|
/*
|
|
bcol_basesmuma_binary_scatter_allgather_segment,
|
|
bcol_basesmuma_binary_scatter_allgather_segment);
|
|
*/
|
|
/* } */
|
|
#endif
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
/* includes shared memory optimization */
|
|
|
|
/**
|
|
* Shared memory blocking Broadcast - fanin, for small data buffers.
|
|
* This routine assumes that buf (the input buffer) is a single writer
|
|
* multi reader (SWMR) shared memory buffer owned by the calling rank
|
|
* which is the only rank that can write to this buffers.
|
|
* It is also assumed that the buffers are registered and fragmented
|
|
* at the ML level and that buf is sufficiently large to hold the data.
|
|
*
|
|
*
|
|
* @param buf - SWMR shared buffer within a sbgp that the
|
|
* executing rank can write to.
|
|
* @param count - the number of elements in the shared buffer.
|
|
* @param dtype - the datatype of a shared buffer element.
|
|
* @param root - the index within the sbgp of the root.
|
|
* @param module - basesmuma module.
|
|
*/
|
|
int bcol_basesmuma_bcast(bcol_function_args_t *input_args,
|
|
coll_ml_function_t *c_input_args)
|
|
{
|
|
/* local variables */
|
|
int group_size, process_shift, my_node_index;
|
|
int my_rank;
|
|
int rc = OMPI_SUCCESS;
|
|
int my_fanout_parent;
|
|
int leading_dim, buff_idx, idx;
|
|
volatile int8_t ready_flag;
|
|
int count=input_args->count;
|
|
struct ompi_datatype_t* dtype=input_args->dtype;
|
|
int root=input_args->root;
|
|
int64_t sequence_number=input_args->sequence_num;
|
|
mca_bcol_basesmuma_module_t* bcol_module=
|
|
(mca_bcol_basesmuma_module_t *)c_input_args->bcol_module;
|
|
int bcol_id = (int) bcol_module->super.bcol_id;
|
|
volatile mca_bcol_basesmuma_payload_t *data_buffs;
|
|
volatile char* parent_data_pointer;
|
|
mca_bcol_basesmuma_header_t *my_ctl_pointer;
|
|
volatile mca_bcol_basesmuma_header_t *parent_ctl_pointer;
|
|
netpatterns_tree_node_t* my_fanout_read_tree;
|
|
size_t pack_len = 0, dt_size;
|
|
|
|
void *data_addr = (void *)((unsigned char *)input_args->src_desc->data_addr );
|
|
|
|
#if 0
|
|
fprintf(stderr,"Entering sm broadcast input_args->sbuf_offset %d \n",input_args->sbuf_offset);
|
|
fflush(stderr);
|
|
#endif
|
|
|
|
|
|
/* we will work only on packed data - so compute the length*/
|
|
ompi_datatype_type_size(dtype, &dt_size);
|
|
pack_len=count*dt_size;
|
|
|
|
buff_idx = input_args->src_desc->buffer_index;
|
|
|
|
/* Get addressing information */
|
|
my_rank = bcol_module->super.sbgp_partner_module->my_index;
|
|
group_size = bcol_module->colls_no_user_data.size_of_group;
|
|
leading_dim=bcol_module->colls_no_user_data.size_of_group;
|
|
idx=SM_ARRAY_INDEX(leading_dim,buff_idx,0);
|
|
data_buffs=(volatile mca_bcol_basesmuma_payload_t *)
|
|
bcol_module->colls_with_user_data.data_buffs+idx;
|
|
|
|
/* Align node index to around sbgp root */
|
|
process_shift = root;
|
|
my_node_index = my_rank - root;
|
|
if(0 > my_node_index ) {
|
|
my_node_index += group_size;
|
|
}
|
|
|
|
/* get my node for the bcast tree */
|
|
my_fanout_read_tree = &(bcol_module->fanout_read_tree[my_node_index]);
|
|
my_fanout_parent = my_fanout_read_tree->parent_rank + process_shift;
|
|
if(group_size <= my_fanout_parent){
|
|
my_fanout_parent -= group_size;
|
|
}
|
|
|
|
/* Set pointer to current proc ctrl region */
|
|
/*my_ctl_pointer = ctl_structs[my_rank]; */
|
|
my_ctl_pointer = data_buffs[my_rank].ctl_struct;
|
|
|
|
/* setup resource recycling */
|
|
|
|
BASESMUMA_HEADER_INIT(my_ctl_pointer, ready_flag, sequence_number, bcol_id);
|
|
|
|
/*
|
|
* Fan out from root
|
|
*/
|
|
if(ROOT_NODE == my_fanout_read_tree->my_node_type) {
|
|
input_args->result_in_rbuf = false;
|
|
/* Root should only signal it is ready */
|
|
my_ctl_pointer->flags[BCAST_FLAG][bcol_id] = ready_flag;
|
|
|
|
}else if(LEAF_NODE == my_fanout_read_tree->my_node_type) {
|
|
input_args->result_in_rbuf = false;
|
|
/*
|
|
* Get parent payload data and control data.
|
|
* Get the pointer to the base address of the parent's payload buffer.
|
|
* Get the parent's control buffer.
|
|
*/
|
|
parent_data_pointer = data_buffs[my_fanout_parent].payload;
|
|
parent_ctl_pointer = data_buffs[my_fanout_parent].ctl_struct;
|
|
|
|
/* Wait until parent signals that data is ready */
|
|
/* The order of conditions checked in this loop is important, as it can
|
|
* result in a race condition.
|
|
*/
|
|
while (!IS_PEER_READY(parent_ctl_pointer, ready_flag, sequence_number, BCAST_FLAG, bcol_id)){
|
|
opal_progress();
|
|
}
|
|
|
|
/* Copy the rank to a shared buffer writable by the current rank */
|
|
memcpy(data_addr, (void *)parent_data_pointer, pack_len);
|
|
|
|
if( 0 != rc ) {
|
|
return OMPI_ERROR;
|
|
}
|
|
|
|
}else{
|
|
input_args->result_in_rbuf = false;
|
|
/* Interior node */
|
|
|
|
/* Get parent payload data and control data */
|
|
parent_data_pointer = data_buffs[my_fanout_parent].payload;
|
|
parent_ctl_pointer = data_buffs[my_fanout_parent].ctl_struct;
|
|
|
|
|
|
/* Wait until parent signals that data is ready */
|
|
/* The order of conditions checked in this loop is important, as it can
|
|
* result in a race condition.
|
|
*/
|
|
while (!IS_PEER_READY(parent_ctl_pointer, ready_flag, sequence_number, BCAST_FLAG, bcol_id)){
|
|
opal_progress();
|
|
}
|
|
|
|
/* Copy the rank to a shared buffer writable by the current rank */
|
|
memcpy(data_addr, (void *)parent_data_pointer,pack_len);
|
|
|
|
/* Signal to children that they may read the data from my shared buffer */
|
|
opal_atomic_wmb ();
|
|
my_ctl_pointer->flags[BCAST_FLAG][bcol_id] = ready_flag;
|
|
}
|
|
|
|
/* if I am the last instance of a basesmuma function in this collectie,
|
|
* release the resrouces */
|
|
|
|
my_ctl_pointer->starting_flag_value[bcol_id]++;
|
|
|
|
return rc;
|
|
}
|
|
|
|
|
|
/*zero-copy large massage communication methods*/
|
|
#if 0
|
|
int bcol_basesmuma_hdl_zerocopy_bcast(bcol_function_args_t *input_args,
|
|
coll_ml_function_t *c_input_args)
|
|
{
|
|
/* local variables */
|
|
int group_size, process_shift, my_node_index;
|
|
int my_rank, first_instance=0, flag_offset;
|
|
int rc = OMPI_SUCCESS;
|
|
int my_fanout_parent;
|
|
int leading_dim, buff_idx, idx;
|
|
volatile int64_t ready_flag;
|
|
int count=input_args->count;
|
|
struct ompi_datatype_t* dtype=input_args->dtype;
|
|
int root=input_args->root;
|
|
int64_t sequence_number=input_args->sequence_num;
|
|
mca_bcol_basesmuma_module_t* bcol_module=
|
|
(mca_bcol_basesmuma_module_t *)c_input_args->bcol_module;
|
|
|
|
netpatterns_tree_node_t* my_fanout_read_tree;
|
|
size_t pack_len = 0, dt_size;
|
|
|
|
void *data_addr = (void *)((unsigned char *)input_args->src_desc->data_addr);
|
|
|
|
struct mca_hdl_base_descriptor_t *hdl_desc;
|
|
struct mca_hdl_base_segment_t *hdl_seg;
|
|
int ret, completed, ridx/*remote rank index*/;
|
|
bool status;
|
|
volatile mca_bcol_basesmuma_ctl_struct_t **ctl_structs;
|
|
mca_bcol_basesmuma_ctl_struct_t *my_ctl_pointer= NULL;
|
|
volatile mca_bcol_basesmuma_ctl_struct_t *parent_ctl_pointer= NULL;
|
|
volatile mca_bcol_basesmuma_ctl_struct_t *child_ctl_pointer= NULL;
|
|
struct mca_hdl_base_module_t* hdl = bcol_module->hdl_module[0];
|
|
|
|
|
|
/* we will work only on packed data - so compute the length*/
|
|
ompi_datatype_type_size(dtype, &dt_size);
|
|
pack_len = count * dt_size;
|
|
|
|
buff_idx = input_args->src_desc->buffer_index;
|
|
|
|
/* Get addressing information */
|
|
my_rank = bcol_module->super.sbgp_partner_module->my_index;
|
|
group_size = bcol_module->colls_no_user_data.size_of_group;
|
|
leading_dim=bcol_module->colls_no_user_data.size_of_group;
|
|
idx=SM_ARRAY_INDEX(leading_dim,buff_idx,0);
|
|
ctl_structs = (volatile mca_bcol_basesmuma_ctl_struct_t **)
|
|
bcol_module->colls_with_user_data.ctl_buffs+idx;
|
|
my_ctl_pointer = ctl_structs[my_rank];
|
|
|
|
/* Align node index to around sbgp root */
|
|
process_shift = root;
|
|
my_node_index = my_rank - root;
|
|
if(0 > my_node_index ) {
|
|
my_node_index += group_size;
|
|
}
|
|
|
|
/* get my node for the bcast tree */
|
|
my_fanout_read_tree = &(bcol_module->fanout_read_tree[my_node_index]);
|
|
my_fanout_parent = my_fanout_read_tree->parent_rank + process_shift;
|
|
if(group_size <= my_fanout_parent){
|
|
my_fanout_parent -= group_size;
|
|
}
|
|
|
|
/* setup resource recycling */
|
|
if( my_ctl_pointer->sequence_number < sequence_number ) {
|
|
first_instance = 1;
|
|
}
|
|
|
|
if( first_instance ) {
|
|
/* Signal arrival */
|
|
my_ctl_pointer->flag = -1;
|
|
my_ctl_pointer->index = 1;
|
|
/* this does not need to use any flag values , so only need to
|
|
* set the value for subsequent values that may need this */
|
|
my_ctl_pointer->starting_flag_value = 0;
|
|
flag_offset = 0;
|
|
} else {
|
|
/* only one thread at a time will be making progress on this
|
|
* collective, so no need to make this atomic */
|
|
my_ctl_pointer->index++;
|
|
}
|
|
|
|
|
|
/* increment the starting flag by one and return */
|
|
flag_offset = my_ctl_pointer->starting_flag_value;
|
|
ready_flag = flag_offset + sequence_number + 1;
|
|
my_ctl_pointer->sequence_number = sequence_number;
|
|
|
|
hdl_desc = (mca_hdl_base_descriptor_t *)
|
|
malloc (sizeof (mca_hdl_base_descriptor_t) * 1);
|
|
|
|
/*prepare a hdl data segment*/
|
|
hdl_seg = (mca_hdl_base_segment_t*)
|
|
malloc ( sizeof (mca_hdl_base_segment_t) * 1);
|
|
hdl_seg->seg_addr.pval = input_args->sbuf;
|
|
hdl_seg->seg_len = pack_len;
|
|
|
|
|
|
hdl->endpoint->ready_flag = ready_flag;
|
|
hdl->endpoint->local_ctrl = my_ctl_pointer;
|
|
hdl->endpoint->sbgp_contextid =
|
|
bcol_module->super.sbgp_partner_module->group_comm->c_contextid;
|
|
|
|
/*
|
|
* Fan out from root
|
|
*/
|
|
if(ROOT_NODE == my_fanout_read_tree->my_node_type) {
|
|
input_args->result_in_rbuf = false;
|
|
|
|
hdl_desc->des_src = hdl_seg;
|
|
hdl_desc->des_src_cnt = 1;
|
|
hdl_desc->isroot = true;
|
|
|
|
/*As the general semantics, there might multiple pairs of send/recv
|
|
*on the topology tree*/
|
|
for (ridx = 0; ridx < my_fanout_read_tree->n_children; ridx++) {
|
|
child_ctl_pointer =
|
|
ctl_structs[my_fanout_read_tree->children_ranks[ridx]];
|
|
hdl->endpoint->remote_ctrl = child_ctl_pointer;
|
|
ret = hdl->hdl_send(hdl, hdl->endpoint, hdl_desc);
|
|
if (ret != OMPI_SUCCESS) {
|
|
BASESMUMA_VERBOSE(1, ("send eror on rank %d ........", my_rank));
|
|
goto exit_ERROR;
|
|
}
|
|
}
|
|
}else if(LEAF_NODE == my_fanout_read_tree->my_node_type) {
|
|
input_args->result_in_rbuf = false;
|
|
/*
|
|
* Get parent payload data and control data.
|
|
* Get the pointer to the base address of the parent's payload buffer.
|
|
* Get the parent's control buffer.
|
|
*/
|
|
parent_ctl_pointer = ctl_structs[my_fanout_parent];
|
|
|
|
hdl_desc->des_dst = hdl_seg;
|
|
hdl_desc->des_dst_cnt = 1;
|
|
hdl_desc->isroot = false;
|
|
hdl->endpoint->remote_ctrl = parent_ctl_pointer;
|
|
|
|
#if __TEST_BLOCKING__
|
|
ret = hdl->hdl_recv(hdl, hdl->endpoint, hdl_desc);
|
|
#else
|
|
ret = hdl->hdl_recvi(hdl, hdl->endpoint, NULL, 0, 0, &hdl_desc);
|
|
#endif
|
|
|
|
#if __TEST_WAIT__
|
|
ret = hdl->hdl_wait(hdl, hdl->endpoint, hdl_desc);
|
|
BASESMUMA_VERBOSE(1,("wait on rank %d is done!", my_rank));
|
|
#endif
|
|
if (OMPI_SUCCESS != ret) {
|
|
BASESMUMA_VERBOSE(1, ("recvi eror on rank %d ........", my_rank));
|
|
goto exit_ERROR;
|
|
}
|
|
|
|
status = false;
|
|
#if __TEST_TEST__
|
|
while (!status) {
|
|
hdl->hdl_test(&hdl_desc, &completed, &status);
|
|
opal_progress();
|
|
BASESMUMA_VERBOSE(1, ("test on rank %d ........", my_rank));
|
|
}
|
|
#endif
|
|
|
|
goto Release;
|
|
|
|
}else{
|
|
input_args->result_in_rbuf = false;
|
|
/* Interior node */
|
|
|
|
/* Get parent payload data and control data */
|
|
parent_ctl_pointer = ctl_structs[my_fanout_parent];
|
|
|
|
hdl_desc->des_dst = hdl_seg;
|
|
hdl_desc->des_dst_cnt = 1;
|
|
hdl_desc->isroot = false;
|
|
|
|
hdl->endpoint->remote_ctrl = parent_ctl_pointer;
|
|
|
|
ret = hdl->hdl_recv(hdl, hdl->endpoint, hdl_desc);
|
|
if (OMPI_SUCCESS != ret) {
|
|
goto exit_ERROR;
|
|
}
|
|
if (OMPI_SUCCESS != ret) {
|
|
BASESMUMA_VERBOSE(1, ("recvi eror on rank %d ........", my_rank));
|
|
goto exit_ERROR;
|
|
}
|
|
|
|
/* Signal to children that they may read the data from my shared buffer */
|
|
opal_atomic_wmb ();
|
|
hdl_desc->des_src = hdl_seg;
|
|
hdl_desc->des_src_cnt = 1;
|
|
for (ridx = 0; ridx < my_fanout_read_tree->n_children; ridx++) {
|
|
child_ctl_pointer =
|
|
ctl_structs[my_fanout_read_tree->children_ranks[ridx]];
|
|
hdl->endpoint->remote_ctrl = child_ctl_pointer;
|
|
|
|
ret = hdl->hdl_send(hdl, hdl->endpoint, hdl_desc);
|
|
if (ret != OMPI_SUCCESS) {
|
|
BASESMUMA_VERBOSE(1, ("send eror on rank %d ........", my_rank));
|
|
goto exit_ERROR;
|
|
}
|
|
}
|
|
goto Release;
|
|
}
|
|
|
|
Release:
|
|
/* if I am the last instance of a basesmuma function in this collectie,
|
|
* release the resrouces */
|
|
if (IS_LAST_BCOL_FUNC(c_input_args)) {
|
|
rc = bcol_basesmuma_free_buff(
|
|
&(bcol_module->colls_with_user_data),
|
|
sequence_number);
|
|
}
|
|
|
|
my_ctl_pointer->starting_flag_value += 1;
|
|
|
|
return BCOL_FN_COMPLETE;
|
|
exit_ERROR:
|
|
return OMPI_ERROR;
|
|
}
|
|
#endif
|