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

Merge pull request #510 from jithinjosepkl/pr/mtl-opt-pr

Просмотр исходного кода 
Optimizations to PML-CM, MTL-OFI
Этот коммит содержится в:
rhc54 2015-04-04 09:22:55 -07:00
родитель 5f19436cd2 9c937d44ae
Коммит 657490c763
12 изменённых файлов: 955 добавлений и 1065 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

11
ompi/mca/mtl/base/mtl_base_datatype.h
Просмотреть файл

@ -26,7 +26,10 @@
#include "opal/datatype/opal_convertor.h" #include "opal/datatype/opal_convertor.h"
#include "opal/datatype/opal_datatype_internal.h" #include "opal/datatype/opal_datatype_internal.h"
static inline int #ifndef MTL_BASE_DATATYPE_H_INCLUDED
#define MTL_BASE_DATATYPE_H_INCLUDED
__opal_attribute_always_inline__ static inline int
ompi_mtl_datatype_pack(struct opal_convertor_t *convertor, ompi_mtl_datatype_pack(struct opal_convertor_t *convertor,
void **buffer, void **buffer,
size_t *buffer_len, size_t *buffer_len,
@ -57,7 +60,7 @@ ompi_mtl_datatype_pack(struct opal_convertor_t *convertor,
} }
static inline int __opal_attribute_always_inline__ static inline int
ompi_mtl_datatype_recv_buf(struct opal_convertor_t *convertor, ompi_mtl_datatype_recv_buf(struct opal_convertor_t *convertor,
void ** buffer, void ** buffer,
size_t *buffer_len, size_t *buffer_len,
@ -81,7 +84,7 @@ ompi_mtl_datatype_recv_buf(struct opal_convertor_t *convertor,
} }
static inline int __opal_attribute_always_inline__ static inline int
ompi_mtl_datatype_unpack(struct opal_convertor_t *convertor, ompi_mtl_datatype_unpack(struct opal_convertor_t *convertor,
void *buffer, void *buffer,
size_t buffer_len) size_t buffer_len)
@ -100,3 +103,5 @@ ompi_mtl_datatype_unpack(struct opal_convertor_t *convertor,
return OMPI_SUCCESS; return OMPI_SUCCESS;
} }
#endif /* MTL_BASE_DATATYPE_H_INCLUDED */

4
ompi/mca/mtl/ofi/Makefile.am
Просмотреть файл

@ -18,16 +18,12 @@ dist_ompidata_DATA = help-mtl-ofi.txt
mtl_ofi_sources = \ mtl_ofi_sources = \
mtl_ofi.h \ mtl_ofi.h \
mtl_ofi.c \ mtl_ofi.c \
mtl_ofi_cancel.c \
mtl_ofi_component.c \ mtl_ofi_component.c \
mtl_ofi_endpoint.h \ mtl_ofi_endpoint.h \
mtl_ofi_endpoint.c \ mtl_ofi_endpoint.c \
mtl_ofi_message.h \ mtl_ofi_message.h \
mtl_ofi_message.c \ mtl_ofi_message.c \
mtl_ofi_probe.c \
mtl_ofi_recv.c \
mtl_ofi_request.h \ mtl_ofi_request.h \
mtl_ofi_send.c \
mtl_ofi_types.h mtl_ofi_types.h
# Make the output library in this directory, and name it either # Make the output library in this directory, and name it either

63
ompi/mca/mtl/ofi/mtl_ofi.c
Просмотреть файл

@ -13,12 +13,13 @@
#include "ompi/proc/proc.h" #include "ompi/proc/proc.h"
#include "ompi/mca/mtl/mtl.h" #include "ompi/mca/mtl/mtl.h"
#include "opal/class/opal_list.h" #include "opal/class/opal_list.h"
#include "opal/mca/pmix/pmix.h"
#include "mtl_ofi.h" #include "mtl_ofi.h"
#include "mtl_ofi_types.h" #include "mtl_ofi_types.h"
#include "mtl_ofi_endpoint.h" #include "mtl_ofi_endpoint.h"
OMPI_DECLSPEC extern mca_mtl_ofi_component_t mca_mtl_ofi_component;
mca_mtl_ofi_module_t ompi_mtl_ofi = { mca_mtl_ofi_module_t ompi_mtl_ofi = {
{ {
8191, /* max cid - 2^13 - 1 */ 8191, /* max cid - 2^13 - 1 */
@ -47,7 +48,6 @@ mca_mtl_ofi_module_t ompi_mtl_ofi = {
NULL NULL
}; };
int int
ompi_mtl_ofi_add_procs(struct mca_mtl_base_module_t *mtl, ompi_mtl_ofi_add_procs(struct mca_mtl_base_module_t *mtl,
size_t nprocs, size_t nprocs,
@ -137,7 +137,6 @@ bail:
return ret; return ret;
} }
int int
ompi_mtl_ofi_del_procs(struct mca_mtl_base_module_t *mtl, ompi_mtl_ofi_del_procs(struct mca_mtl_base_module_t *mtl,
size_t nprocs, size_t nprocs,
@ -164,61 +163,3 @@ ompi_mtl_ofi_del_procs(struct mca_mtl_base_module_t *mtl,
return OMPI_SUCCESS; return OMPI_SUCCESS;
} }
int
ompi_mtl_ofi_finalize(struct mca_mtl_base_module_t *mtl)
{
opal_progress_unregister(ompi_mtl_ofi_progress);
/**
* Close all the OFI objects
*/
if (fi_close((fid_t)ompi_mtl_ofi.ep)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.mr)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.cq)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.av)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.domain)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.fabric)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
return OMPI_SUCCESS;
}
int
ompi_mtl_ofi_add_comm(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm)
{
return OMPI_SUCCESS;
}
int
ompi_mtl_ofi_del_comm(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm)
{
return OMPI_SUCCESS;
}

938
ompi/mca/mtl/ofi/mtl_ofi.h
Просмотреть файл

@ -15,6 +15,7 @@
#include "ompi/mca/mtl/mtl.h" #include "ompi/mca/mtl/mtl.h"
#include "ompi/mca/mtl/base/base.h" #include "ompi/mca/mtl/base/base.h"
#include "opal/datatype/opal_convertor.h" #include "opal/datatype/opal_convertor.h"
#include "opal/mca/pmix/pmix.h"
#include <rdma/fabric.h> #include <rdma/fabric.h>
#include <rdma/fi_cm.h> #include <rdma/fi_cm.h>
@ -23,10 +24,26 @@
#include <rdma/fi_errno.h> #include <rdma/fi_errno.h>
#include <rdma/fi_tagged.h> #include <rdma/fi_tagged.h>
#include "ompi_config.h"
#include "ompi/proc/proc.h"
#include "ompi/mca/mtl/mtl.h"
#include "opal/class/opal_list.h"
#include "ompi/communicator/communicator.h"
#include "opal/datatype/opal_convertor.h"
#include "ompi/mca/mtl/base/base.h"
#include "ompi/mca/mtl/base/mtl_base_datatype.h"
#include "ompi/message/message.h"
#include "mtl_ofi.h"
#include "mtl_ofi_types.h"
#include "mtl_ofi_request.h"
#include "mtl_ofi_endpoint.h"
#include "mtl_ofi_message.h"
BEGIN_C_DECLS BEGIN_C_DECLS
/* MTL interface functions */ extern mca_mtl_ofi_module_t ompi_mtl_ofi;
extern int ompi_mtl_ofi_finalize(struct mca_mtl_base_module_t *mtl); extern mca_base_framework_t ompi_mtl_base_framework;
extern int ompi_mtl_ofi_add_procs(struct mca_mtl_base_module_t *mtl, extern int ompi_mtl_ofi_add_procs(struct mca_mtl_base_module_t *mtl,
size_t nprocs, size_t nprocs,
@ -36,62 +53,887 @@ extern int ompi_mtl_ofi_del_procs(struct mca_mtl_base_module_t *mtl,
size_t nprocs, size_t nprocs,
struct ompi_proc_t **procs); struct ompi_proc_t **procs);
extern int ompi_mtl_ofi_send(struct mca_mtl_base_module_t *mtl, __opal_attribute_always_inline__ static inline int
struct ompi_communicator_t *comm, ompi_mtl_ofi_progress(void)
int dest, {
int tag, int ret, count = 0;
struct opal_convertor_t *convertor, struct fi_cq_tagged_entry wc;
mca_pml_base_send_mode_t mode); struct fi_cq_err_entry error;
ompi_mtl_ofi_request_t *ofi_req = NULL;
extern int ompi_mtl_ofi_isend(struct mca_mtl_base_module_t *mtl, /**
struct ompi_communicator_t *comm, * Read the work completions from the CQ.
int dest, * From the completion's op_context, we get the associated OFI request.
int tag, * Call the request's callback.
struct opal_convertor_t *convertor, */
mca_pml_base_send_mode_t mode, while (true) {
bool blocking, memset(&wc, 0, sizeof(wc));
mca_mtl_request_t *mtl_request); ret = fi_cq_read(ompi_mtl_ofi.cq, (void *)&wc, 1);
if (ret > 0) {
count++;
if (NULL != wc.op_context) {
ofi_req = TO_OFI_REQ(wc.op_context);
assert(ofi_req);
ret = ofi_req->event_callback(&wc, ofi_req);
if (OMPI_SUCCESS != ret) {
opal_output(ompi_mtl_base_framework.framework_output,
"Error returned by request event callback: %d",
ret);
abort();
}
}
} else if (ret == -FI_EAVAIL) {
/**
* An error occured and is being reported via the CQ.
* Read the error and forward it to the upper layer.
*/
memset(&error, 0, sizeof(error));
ret = fi_cq_readerr(ompi_mtl_ofi.cq,
&error,
0);
if (ret) {
opal_output(ompi_mtl_base_framework.framework_output,
"Error returned from fi_cq_readerr: %d", ret);
}
extern int ompi_mtl_ofi_irecv(struct mca_mtl_base_module_t *mtl, assert(error.op_context);
struct ompi_communicator_t *comm, ofi_req = TO_OFI_REQ(error.op_context);
int src, assert(ofi_req);
int tag, ret = ofi_req->error_callback(&error, ofi_req);
struct opal_convertor_t *convertor, if (OMPI_SUCCESS != ret) {
mca_mtl_request_t *mtl_request); opal_output(ompi_mtl_base_framework.framework_output,
"Error returned by request error callback: %d",
ret);
abort();
}
} else {
/**
* The CQ is empty. Return.
*/
break;
}
}
return count;
}
extern int ompi_mtl_ofi_iprobe(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int src,
int tag,
int *flag,
struct ompi_status_public_t *status);
extern int ompi_mtl_ofi_imrecv(struct mca_mtl_base_module_t *mtl, /* MTL interface functions */
struct opal_convertor_t *convertor, __opal_attribute_always_inline__ static inline int
struct ompi_message_t **message, ompi_mtl_ofi_finalize(struct mca_mtl_base_module_t *mtl)
struct mca_mtl_request_t *mtl_request); {
opal_progress_unregister(ompi_mtl_ofi_progress);
extern int ompi_mtl_ofi_improbe(struct mca_mtl_base_module_t *mtl, /**
struct ompi_communicator_t *comm, * Close all the OFI objects
int src, */
int tag, if (fi_close((fid_t)ompi_mtl_ofi.ep)) {
int *matched, opal_output(ompi_mtl_base_framework.framework_output,
struct ompi_message_t **message, "fi_close failed: %s", strerror(errno));
struct ompi_status_public_t *status); abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.mr)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.cq)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.av)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.domain)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
if (fi_close((fid_t)ompi_mtl_ofi.fabric)) {
opal_output(ompi_mtl_base_framework.framework_output,
"fi_close failed: %s", strerror(errno));
abort();
}
extern int ompi_mtl_ofi_cancel(struct mca_mtl_base_module_t *mtl, return OMPI_SUCCESS;
mca_mtl_request_t *mtl_request, }
int flag);
extern int ompi_mtl_ofi_add_comm(struct mca_mtl_base_module_t *mtl, __opal_attribute_always_inline__ static inline int
struct ompi_communicator_t *comm); ompi_mtl_ofi_get_error(int error_num)
{
int ret;
extern int ompi_mtl_ofi_del_comm(struct mca_mtl_base_module_t *mtl, switch (error_num) {
struct ompi_communicator_t *comm); case 0:
ret = OMPI_SUCCESS;
break;
default:
ret = OMPI_ERROR;
}
extern int ompi_mtl_ofi_progress(void); return ret;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
assert(ofi_req->completion_count > 0);
ofi_req->completion_count--;
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
switch(error->err) {
case FI_EMSGSIZE:
ofi_req->status.MPI_ERROR = MPI_ERR_TRUNCATE;
break;
default:
ofi_req->status.MPI_ERROR = MPI_ERR_INTERN;
}
return ofi_req->event_callback(NULL, ofi_req);
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_ack_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
ompi_mtl_ofi_request_t *parent_req = ofi_req->parent;
free(ofi_req);
parent_req->event_callback(NULL, parent_req);
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_ack_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
ompi_mtl_ofi_request_t *parent_req = ofi_req->parent;
free(ofi_req);
parent_req->status.MPI_ERROR = MPI_ERR_INTERN;
return parent_req->error_callback(error, parent_req);
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_isend_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
assert(ofi_req->completion_count > 0);
ofi_req->completion_count--;
if (0 == ofi_req->completion_count) {
/* Request completed */
if (OPAL_UNLIKELY(NULL != ofi_req->buffer)) {
free(ofi_req->buffer);
ofi_req->buffer = NULL;
}
ofi_req->super.ompi_req->req_status.MPI_ERROR =
ofi_req->status.MPI_ERROR;
ofi_req->super.completion_callback(&ofi_req->super);
}
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send_start(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode,
ompi_mtl_ofi_request_t *ofi_req)
{
int ret;
void *start;
size_t length;
ssize_t ret_length;
bool free_after;
uint64_t match_bits;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
ompi_mtl_ofi_request_t *ack_req = NULL; /* For synchronous send */
ompi_proc = ompi_comm_peer_lookup(comm, dest);
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
ret = ompi_mtl_datatype_pack(convertor, &start, &length, &free_after);
if (OMPI_SUCCESS != ret) return ret;
ofi_req->buffer = (free_after) ? start : NULL;
ofi_req->length = length;
ofi_req->status.MPI_ERROR = OMPI_SUCCESS;
if (OPAL_UNLIKELY(MCA_PML_BASE_SEND_SYNCHRONOUS == mode)) {
ack_req = malloc(sizeof(ompi_mtl_ofi_request_t));
assert(ack_req);
ack_req->parent = ofi_req;
ack_req->event_callback = ompi_mtl_ofi_send_ack_callback;
ack_req->error_callback = ompi_mtl_ofi_send_ack_error_callback;
ofi_req->completion_count = 2;
MTL_OFI_SET_SEND_BITS(match_bits, comm->c_contextid,
comm->c_my_rank, tag, MTL_OFI_SYNC_SEND);
ret_length = fi_trecv(ompi_mtl_ofi.ep,
NULL,
0,
ompi_mtl_ofi.mr,
endpoint->peer_fiaddr,
match_bits | MTL_OFI_SYNC_SEND_ACK,
0, /* Exact match, no ignore bits */
(void *) &ack_req->ctx);
if (OPAL_UNLIKELY(ret_length < 0)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_trecv failed: %s(%zd)",
__FILE__, __LINE__,
strerror(errno), ret_length);
return ompi_mtl_ofi_get_error(ret);
}
} else {
ofi_req->completion_count = 1;
MTL_OFI_SET_SEND_BITS(match_bits, comm->c_contextid,
comm->c_my_rank, tag, 0);
}
ret_length = fi_tsend(ompi_mtl_ofi.ep,
start,
length,
ompi_mtl_ofi.mr,
endpoint->peer_fiaddr,
match_bits,
(void *) &ofi_req->ctx);
if (OPAL_UNLIKELY(0 > ret_length)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsend failed: %zd",
__FILE__, __LINE__, ret_length);
return ompi_mtl_ofi_get_error(ret);
}
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_send(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode)
{
int ret = OMPI_SUCCESS;
ompi_mtl_ofi_request_t ofi_req;
/**
* Create a send request, start it and wait until it completes.
*/
ofi_req.event_callback = ompi_mtl_ofi_send_callback;
ofi_req.error_callback = ompi_mtl_ofi_send_error_callback;
ret = ompi_mtl_ofi_send_start(mtl, comm, dest, tag,
convertor, mode, &ofi_req);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
if (NULL != ofi_req.buffer) {
free(ofi_req.buffer);
}
return ret;
}
/**
* Wait until the request is completed.
* ompi_mtl_ofi_send_callback() updates this variable.
*/
while (0 < ofi_req.completion_count) {
ompi_mtl_ofi_progress();
}
if (OPAL_UNLIKELY(NULL != ofi_req.buffer)) {
free(ofi_req.buffer);
}
return ofi_req.status.MPI_ERROR;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_isend(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode,
bool blocking,
mca_mtl_request_t *mtl_request)
{
int ret = OMPI_SUCCESS;
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
ofi_req->event_callback = ompi_mtl_ofi_isend_callback;
ofi_req->error_callback = ompi_mtl_ofi_send_error_callback;
ret = ompi_mtl_ofi_send_start(mtl, comm, dest, tag,
convertor, mode, ofi_req);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret && NULL != ofi_req->buffer)) {
free(ofi_req->buffer);
}
return ret;
}
/**
* Called when a completion for SYNC ACK send is received.
* This completes the synchronous recv operation. Thus, we
* call the upper layer's completion function.
*/
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_sync_recv_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
ofi_req->super.completion_callback(&ofi_req->super);
return OMPI_SUCCESS;
}
/**
* Called when a completion for a posted recv is received.
*/
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_recv_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
int ret;
ssize_t ret_length;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
int src;
ompi_status_public_t *status = NULL;
assert(ofi_req->super.ompi_req);
status = &ofi_req->super.ompi_req->req_status;
/**
* Any event associated with a request starts it.
* This prevents a started request from being cancelled.
*/
ofi_req->req_started = true;
status->MPI_SOURCE = MTL_OFI_GET_SOURCE(wc->tag);
status->MPI_TAG = MTL_OFI_GET_TAG(wc->tag);
status->_ucount = wc->len;
if (OPAL_UNLIKELY(wc->len > ofi_req->length)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"truncate expected: %ld %ld",
wc->len, ofi_req->length);
status->MPI_ERROR = MPI_ERR_TRUNCATE;
}
/**
* Unpack data into recv buffer if necessary.
*/
if (OPAL_UNLIKELY(ofi_req->buffer)) {
ret = ompi_mtl_datatype_unpack(ofi_req->convertor,
ofi_req->buffer,
wc->len);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: ompi_mtl_datatype_unpack failed: %d",
__FILE__, __LINE__, ret);
status->MPI_ERROR = ret;
}
}
/**
* We do not want any SYNC_SEND_ACK here!
* See mtl_ofi_send.c for details.
*/
assert(!MTL_OFI_IS_SYNC_SEND_ACK(wc->tag));
/**
* If this recv is part of an MPI_Ssend operation, then we send an
* acknowledgment back to the sender. The fi_context can be
* re-used safely because the previous operation has completed.
* This recv request will complete once we get a completion for
* this send. See ompi_mtl_ofi_sync_recv_callback().
* Otherwise, this request is now complete.
*/
if (OPAL_UNLIKELY(MTL_OFI_IS_SYNC_SEND(wc->tag))) {
ofi_req->event_callback = ompi_mtl_ofi_sync_recv_callback;
/**
* If the recv request was posted for any source,
* we need to extract the source's actual address.
*/
if (!ofi_req->remote_addr) {
src = MTL_OFI_GET_SOURCE(wc->tag);
ompi_proc = ompi_comm_peer_lookup(ofi_req->comm, src );
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
ofi_req->remote_addr = endpoint->peer_fiaddr;
}
ret_length = fi_tsend(ompi_mtl_ofi.ep,
NULL,
0,
ompi_mtl_ofi.mr,
ofi_req->remote_addr,
wc->tag | MTL_OFI_SYNC_SEND_ACK,
(void *) &ofi_req->ctx);
if (OPAL_UNLIKELY(ret_length < 0)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsend failed: %zd",
__FILE__, __LINE__, ret_length);
status->MPI_ERROR = OMPI_ERROR;
}
} else {
ofi_req->super.completion_callback(&ofi_req->super);
}
return OMPI_SUCCESS;
}
/**
* Called when an error occured on a recv request.
*/
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_recv_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
ompi_status_public_t *status;
assert(ofi_req->super.ompi_req);
status = &ofi_req->super.ompi_req->req_status;
status->MPI_TAG = MTL_OFI_GET_TAG(ofi_req->match_bits);
status->MPI_SOURCE = MTL_OFI_GET_SOURCE(ofi_req->match_bits);
/* FIXME: This could be done on a single line... */
switch (error->err) {
case FI_EMSGSIZE:
status->MPI_ERROR = MPI_ERR_TRUNCATE;
break;
default:
status->MPI_ERROR = MPI_ERR_INTERN;
}
ofi_req->super.completion_callback(&ofi_req->super);
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_irecv(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int src,
int tag,
struct opal_convertor_t *convertor,
mca_mtl_request_t *mtl_request)
{
int ret = OMPI_SUCCESS;
ssize_t ret_length;
uint64_t match_bits, mask_bits;
fi_addr_t remote_addr;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
void *start;
size_t length;
bool free_after;
if (MPI_ANY_SOURCE != src) {
ompi_proc = ompi_comm_peer_lookup(comm, src);
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
remote_addr = endpoint->peer_fiaddr;
} else {
remote_addr = ompi_mtl_ofi.any_addr;
}
MTL_OFI_SET_RECV_BITS(match_bits, mask_bits, comm->c_contextid, src, tag);
ret = ompi_mtl_datatype_recv_buf(convertor, &start, &length, &free_after);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
return ret;
}
ofi_req->type = OMPI_MTL_OFI_RECV;
ofi_req->event_callback = ompi_mtl_ofi_recv_callback;
ofi_req->error_callback = ompi_mtl_ofi_recv_error_callback;
ofi_req->comm = comm;
ofi_req->buffer = (free_after) ? start : NULL;
ofi_req->length = length;
ofi_req->convertor = convertor;
ofi_req->req_started = false;
ofi_req->status.MPI_ERROR = OMPI_SUCCESS;
ofi_req->remote_addr = remote_addr;
ofi_req->match_bits = match_bits;
ret_length = fi_trecv(ompi_mtl_ofi.ep,
start,
length,
ompi_mtl_ofi.mr,
remote_addr,
match_bits,
mask_bits,
(void *)&ofi_req->ctx);
if (OPAL_UNLIKELY(ret_length < 0)) {
if (NULL != ofi_req->buffer) {
free(ofi_req->buffer);
}
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_trecv failed: %s(%zd)",
__FILE__, __LINE__, strerror(errno), ret_length);
return ompi_mtl_ofi_get_error(ret);
}
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_imrecv(struct mca_mtl_base_module_t *mtl,
struct opal_convertor_t *convertor,
struct ompi_message_t **message,
struct mca_mtl_request_t *mtl_request)
{
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
void *start;
size_t length;
bool free_after;
int ret;
ompi_mtl_ofi_message_t *ofi_message =
(ompi_mtl_ofi_message_t*) (*message)->req_ptr;
ret = ompi_mtl_datatype_recv_buf(convertor, &start, &length, &free_after);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
return ret;
}
ofi_req->type = OMPI_MTL_OFI_RECV;
ofi_req->event_callback = ompi_mtl_ofi_recv_callback;
ofi_req->error_callback = ompi_mtl_ofi_recv_error_callback;
ofi_req->buffer = (free_after) ? start : NULL;
ofi_req->length = length;
ofi_req->convertor = convertor;
ofi_req->status.MPI_ERROR = OMPI_SUCCESS;
(*message) = MPI_MESSAGE_NULL;
return ompi_mtl_ofi_recv_callback(&(ofi_message->wc), ofi_req);
}
/**
* Called when a probe request completes. Read fi_cq_tagged_entry's
* data field to determine whether or not a matching message was found.
*/
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_probe_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
if (wc->data > 0) {
ofi_req->match_state = 1;
ofi_req->status.MPI_SOURCE = MTL_OFI_GET_SOURCE(wc->tag);
ofi_req->status.MPI_TAG = MTL_OFI_GET_TAG(wc->tag);
ofi_req->status.MPI_ERROR = MPI_SUCCESS;
ofi_req->status._ucount = wc->len;
} else {
ofi_req->match_state = 0;
}
ofi_req->completion_count--;
return OMPI_SUCCESS;
}
/**
* Called when a probe request encounters an error.
*/
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_probe_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
ofi_req->super.ompi_req->req_status.MPI_ERROR = MPI_ERR_INTERN;
ofi_req->completion_count--;
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_iprobe(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int src,
int tag,
int *flag,
struct ompi_status_public_t *status)
{
struct ompi_mtl_ofi_request_t ofi_req;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
fi_addr_t remote_proc = 0;
size_t length = 0;
uint64_t match_bits, mask_bits;
int ret;
/**
* If the source is known, use its peer_fiaddr.
*/
if (MPI_ANY_SOURCE != src) {
ompi_proc = ompi_comm_peer_lookup( comm, src );
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
remote_proc = endpoint->peer_fiaddr;
}
MTL_OFI_SET_RECV_BITS(match_bits, mask_bits, comm->c_contextid, src, tag);
ofi_req.type = OMPI_MTL_OFI_PROBE;
ofi_req.event_callback = ompi_mtl_ofi_probe_callback;
ofi_req.error_callback = ompi_mtl_ofi_probe_error_callback;
ofi_req.completion_count = 1;
ofi_req.match_state = 0;
ret = fi_tsearch(ompi_mtl_ofi.ep,
&match_bits,
mask_bits,
0,
&remote_proc,
&length,
(void *)&ofi_req.ctx);
/**
* Probe is a blocking operation. fi_tsearch() is non-blocking.
* We inspect the return code and decide what to do.
* The request can either:
* - be queued successfully,
* - return no matching message, or
* - return a matching message.
*/
if (0 == ret) {
/**
* The search request was queued successfully. Wait until complete.
*/
while (0 < ofi_req.completion_count) {
opal_progress();
}
*flag = ofi_req.match_state;
if (1 == *flag) {
*status = ofi_req.status;
}
} else if (1 == ret) {
/**
* The search request completed and a matching message was found.
*/
ofi_req.match_state = 1;
ofi_req.status.MPI_SOURCE = MTL_OFI_GET_SOURCE(match_bits);
ofi_req.status.MPI_TAG = MTL_OFI_GET_TAG(match_bits);
ofi_req.status.MPI_ERROR = MPI_SUCCESS;
ofi_req.status._ucount = length;
*flag = 1;
*status = ofi_req.status;
} else if (ret < 0 && -FI_ENOMSG == ret) {
/**
* The search request completed but no matching message was found.
*/
*flag = 0;
} else if (ret < 0 && ret != -FI_ENOMSG) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsearch failed: %d (%s)",
__FILE__, __LINE__, ret, fi_strerror(ret));
return ompi_mtl_ofi_get_error(-ret);
} else {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: unexpected return code from fi_tsearch: %d",
__FILE__, __LINE__, ret);
return ompi_mtl_ofi_get_error(-ret);
}
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_improbe(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int src,
int tag,
int *matched,
struct ompi_message_t **message,
struct ompi_status_public_t *status)
{
struct ompi_mtl_ofi_request_t ofi_req;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
fi_addr_t remote_proc = 0;
size_t length = 0;
uint64_t match_bits, mask_bits;
int ret;
/**
* If the source is known, use its peer_fiaddr.
*/
if (MPI_ANY_SOURCE != src) {
ompi_proc = ompi_comm_peer_lookup( comm, src );
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
remote_proc = endpoint->peer_fiaddr;
}
MTL_OFI_SET_RECV_BITS(match_bits, mask_bits, comm->c_contextid, src, tag);
ofi_req.type = OMPI_MTL_OFI_PROBE;
ofi_req.event_callback = ompi_mtl_ofi_probe_callback;
ofi_req.error_callback = ompi_mtl_ofi_probe_error_callback;
ofi_req.completion_count = 1;
ofi_req.match_state = 0;
ret = fi_tsearch(ompi_mtl_ofi.ep,
&match_bits,
mask_bits,
FI_CLAIM,
&remote_proc,
&length,
(void *)&ofi_req.ctx);
/**
* Probe is a blocking operation. fi_tsearch() is non-blocking.
* We inspect the return code and decide what to do.
* The request can either:
* - be queued successfully,
* - return no matching message, or
* - return a matching message.
*/
if (ret == 0) {
/**
* The search request was queued successfully. Wait until complete.
*/
while (0 < ofi_req.completion_count) {
opal_progress();
}
*matched = ofi_req.match_state;
if (1 == *matched) {
*status = ofi_req.status;
(*message) = ompi_message_alloc();
if (NULL == (*message)) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
(*message)->comm = comm;
(*message)->req_ptr = ofi_req.message;
(*message)->peer = status->MPI_SOURCE;
(*message)->count = status->_ucount;
if (NULL == (*message)->req_ptr) {
ompi_message_return(*message);
*message = NULL;
return OMPI_ERR_OUT_OF_RESOURCE;
}
} else {
(*message) = MPI_MESSAGE_NULL;
}
} else if (1 == ret) {
/**
* The search request completed and a matching message was found.
*/
*matched = 1;
*status = ofi_req.status;
ofi_req.match_state = 1;
(*message) = ompi_message_alloc();
if (NULL == (*message)) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
(*message)->comm = comm;
(*message)->req_ptr = ofi_req.message;
(*message)->peer = status->MPI_SOURCE;
(*message)->count = status->_ucount;
if (NULL == (*message)->req_ptr) {
ompi_message_return(*message);
*message = NULL;
return OMPI_ERR_OUT_OF_RESOURCE;
}
} else if (ret < 0 && ret == -FI_ENOMSG) {
/**
* The search request completed but no matching message was found.
*/
*matched = 0;
} else if (ret < 0 && ret != -FI_ENOMSG) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsearch failed: %d (%s)",
__FILE__, __LINE__, ret, fi_strerror(ret));
return ompi_mtl_ofi_get_error(-ret);
} else {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: unexpected return code from fi_tsearch: %d",
__FILE__, __LINE__, ret);
return ompi_mtl_ofi_get_error(-ret);
}
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_cancel(struct mca_mtl_base_module_t *mtl,
mca_mtl_request_t *mtl_request,
int flag)
{
int ret;
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
switch (ofi_req->type) {
case OMPI_MTL_OFI_SEND:
/**
* Cannot cancel sends yet
*/
break;
case OMPI_MTL_OFI_RECV:
/**
* Cancel a receive request only if it hasn't been matched yet.
* The event queue needs to be drained to make sure there isn't
* any pending receive completion event.
*/
ompi_mtl_ofi_progress();
if (!ofi_req->req_started) {
ret = fi_cancel((fid_t)ompi_mtl_ofi.ep, &ofi_req->ctx);
if (0 == ret) {
/**
* The request was successfully cancelled.
*/
ofi_req->super.ompi_req->req_status._cancelled = true;
ofi_req->super.completion_callback(&ofi_req->super);
}
}
break;
default:
return OMPI_ERROR;
}
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_add_comm(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm)
{
return OMPI_SUCCESS;
}
__opal_attribute_always_inline__ static inline int
ompi_mtl_ofi_del_comm(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm)
{
return OMPI_SUCCESS;
}
extern int ompi_mtl_ofi_get_error(int fi_error);
END_C_DECLS END_C_DECLS

58
ompi/mca/mtl/ofi/mtl_ofi_cancel.c
Просмотреть файл

@ -1,58 +0,0 @@
/*
* Copyright (c) 2013-2014 Intel, Inc. All rights reserved
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mtl_ofi.h"
#include "mtl_ofi_types.h"
#include "mtl_ofi_request.h"
int
ompi_mtl_ofi_cancel(struct mca_mtl_base_module_t *mtl,
mca_mtl_request_t *mtl_request,
int flag)
{
int ret;
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
switch (ofi_req->type) {
case OMPI_MTL_OFI_SEND:
/**
* Cannot cancel sends yet
*/
break;
case OMPI_MTL_OFI_RECV:
/**
* Cancel a receive request only if it hasn't been matched yet.
* The event queue needs to be drained to make sure there isn't
* any pending receive completion event.
*/
ompi_mtl_ofi_progress();
if (!ofi_req->req_started) {
ret = fi_cancel((fid_t)ompi_mtl_ofi.ep, &ofi_req->ctx);
if (0 == ret) {
/**
* The request was successfully cancelled.
*/
ofi_req->super.ompi_req->req_status._cancelled = true;
ofi_req->super.completion_callback(&ofi_req->super);
}
}
break;
default:
return OMPI_ERROR;
}
return OMPI_SUCCESS;
}

78
ompi/mca/mtl/ofi/mtl_ofi_component.c
Просмотреть файл

@ -416,82 +416,4 @@ error:
} }
int
ompi_mtl_ofi_get_error(int error_num)
{
int ret;
switch (error_num) {
case 0:
ret = OMPI_SUCCESS;
break;
default:
ret = OMPI_ERROR;
}
return ret;
}
int
ompi_mtl_ofi_progress(void)
{
int ret, count = 0;
struct fi_cq_tagged_entry wc;
struct fi_cq_err_entry error;
ompi_mtl_ofi_request_t *ofi_req = NULL;
/**
* Read the work completions from the CQ.
* From the completion's op_context, we get the associated OFI request.
* Call the request's callback.
*/
while (true) {
memset(&wc, 0, sizeof(wc));
ret = fi_cq_read(ompi_mtl_ofi.cq, (void *)&wc, 1);
if (ret > 0) {
count++;
if (NULL != wc.op_context) {
ofi_req = TO_OFI_REQ(wc.op_context);
assert(ofi_req);
ret = ofi_req->event_callback(&wc, ofi_req);
if (OMPI_SUCCESS != ret) {
opal_output(ompi_mtl_base_framework.framework_output,
"Error returned by request event callback: %d",
ret);
abort();
}
}
} else if (ret == -FI_EAVAIL) {
/**
* An error occured and is being reported via the CQ.
* Read the error and forward it to the upper layer.
*/
memset(&error, 0, sizeof(error));
ret = fi_cq_readerr(ompi_mtl_ofi.cq,
&error,
0);
if (ret) {
opal_output(ompi_mtl_base_framework.framework_output,
"Error returned from fi_cq_readerr: %d", ret);
}
assert(error.op_context);
ofi_req = TO_OFI_REQ(error.op_context);
assert(ofi_req);
ret = ofi_req->error_callback(&error, ofi_req);
if (OMPI_SUCCESS != ret) {
opal_output(ompi_mtl_base_framework.framework_output,
"Error returned by request error callback: %d",
ret);
abort();
}
} else {
/**
* The CQ is empty. Return.
*/
break;
}
}
return count;
}

269
ompi/mca/mtl/ofi/mtl_ofi_probe.c
Просмотреть файл

@ -1,269 +0,0 @@
/*
* Copyright (c) 2013-2014 Intel, Inc. All rights reserved
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "ompi/communicator/communicator.h"
#include "ompi/message/message.h"
#include "mtl_ofi.h"
#include "mtl_ofi_types.h"
#include "mtl_ofi_request.h"
#include "mtl_ofi_endpoint.h"
#include "mtl_ofi_message.h"
/**
* Called when a probe request completes. Read fi_cq_tagged_entry's
* data field to determine whether or not a matching message was found.
*/
static int
ompi_mtl_ofi_probe_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
if (wc->data > 0) {
ofi_req->match_state = 1;
ofi_req->status.MPI_SOURCE = MTL_OFI_GET_SOURCE(wc->tag);
ofi_req->status.MPI_TAG = MTL_OFI_GET_TAG(wc->tag);
ofi_req->status.MPI_ERROR = MPI_SUCCESS;
ofi_req->status._ucount = wc->len;
} else {
ofi_req->match_state = 0;
}
ofi_req->completion_count--;
return OMPI_SUCCESS;
}
/**
* Called when a probe request encounters an error.
*/
static int
ompi_mtl_ofi_probe_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
ofi_req->super.ompi_req->req_status.MPI_ERROR = MPI_ERR_INTERN;
ofi_req->completion_count--;
return OMPI_SUCCESS;
}
int
ompi_mtl_ofi_iprobe(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int src,
int tag,
int *flag,
struct ompi_status_public_t *status)
{
struct ompi_mtl_ofi_request_t ofi_req;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
fi_addr_t remote_proc = 0;
size_t length = 0;
uint64_t match_bits, mask_bits;
int ret;
/**
* If the source is known, use its peer_fiaddr.
*/
if (MPI_ANY_SOURCE != src) {
ompi_proc = ompi_comm_peer_lookup( comm, src );
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
remote_proc = endpoint->peer_fiaddr;
}
MTL_OFI_SET_RECV_BITS(match_bits, mask_bits, comm->c_contextid, src, tag);
ofi_req.type = OMPI_MTL_OFI_PROBE;
ofi_req.event_callback = ompi_mtl_ofi_probe_callback;
ofi_req.error_callback = ompi_mtl_ofi_probe_error_callback;
ofi_req.completion_count = 1;
ofi_req.match_state = 0;
ret = fi_tsearch(ompi_mtl_ofi.ep,
&match_bits,
mask_bits,
0,
&remote_proc,
&length,
(void *)&ofi_req.ctx);
/**
* Probe is a blocking operation. fi_tsearch() is non-blocking.
* We inspect the return code and decide what to do.
* The request can either:
* - be queued successfully,
* - return no matching message, or
* - return a matching message.
*/
if (0 == ret) {
/**
* The search request was queued successfully. Wait until complete.
*/
while (0 < ofi_req.completion_count) {
opal_progress();
}
*flag = ofi_req.match_state;
if (1 == *flag) {
*status = ofi_req.status;
}
} else if (1 == ret) {
/**
* The search request completed and a matching message was found.
*/
ofi_req.match_state = 1;
ofi_req.status.MPI_SOURCE = MTL_OFI_GET_SOURCE(match_bits);
ofi_req.status.MPI_TAG = MTL_OFI_GET_TAG(match_bits);
ofi_req.status.MPI_ERROR = MPI_SUCCESS;
ofi_req.status._ucount = length;
*flag = 1;
*status = ofi_req.status;
} else if (ret < 0 && -FI_ENOMSG == ret) {
/**
* The search request completed but no matching message was found.
*/
*flag = 0;
} else if (ret < 0 && ret != -FI_ENOMSG) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsearch failed: %d (%s)",
__FILE__, __LINE__, ret, fi_strerror(ret));
return ompi_mtl_ofi_get_error(-ret);
} else {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: unexpected return code from fi_tsearch: %d",
__FILE__, __LINE__, ret);
return ompi_mtl_ofi_get_error(-ret);
}
return OMPI_SUCCESS;
}
int
ompi_mtl_ofi_improbe(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int src,
int tag,
int *matched,
struct ompi_message_t **message,
struct ompi_status_public_t *status)
{
struct ompi_mtl_ofi_request_t ofi_req;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
fi_addr_t remote_proc = 0;
size_t length = 0;
uint64_t match_bits, mask_bits;
int ret;
/**
* If the source is known, use its peer_fiaddr.
*/
if (MPI_ANY_SOURCE != src) {
ompi_proc = ompi_comm_peer_lookup( comm, src );
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
remote_proc = endpoint->peer_fiaddr;
}
MTL_OFI_SET_RECV_BITS(match_bits, mask_bits, comm->c_contextid, src, tag);
ofi_req.type = OMPI_MTL_OFI_PROBE;
ofi_req.event_callback = ompi_mtl_ofi_probe_callback;
ofi_req.error_callback = ompi_mtl_ofi_probe_error_callback;
ofi_req.completion_count = 1;
ofi_req.match_state = 0;
ret = fi_tsearch(ompi_mtl_ofi.ep,
&match_bits,
mask_bits,
FI_CLAIM,
&remote_proc,
&length,
(void *)&ofi_req.ctx);
/**
* Probe is a blocking operation. fi_tsearch() is non-blocking.
* We inspect the return code and decide what to do.
* The request can either:
* - be queued successfully,
* - return no matching message, or
* - return a matching message.
*/
if (ret == 0) {
/**
* The search request was queued successfully. Wait until complete.
*/
while (0 < ofi_req.completion_count) {
opal_progress();
}
*matched = ofi_req.match_state;
if (1 == *matched) {
*status = ofi_req.status;
(*message) = ompi_message_alloc();
if (NULL == (*message)) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
(*message)->comm = comm;
(*message)->req_ptr = ofi_req.message;
(*message)->peer = status->MPI_SOURCE;
(*message)->count = status->_ucount;
if (NULL == (*message)->req_ptr) {
ompi_message_return(*message);
*message = NULL;
return OMPI_ERR_OUT_OF_RESOURCE;
}
} else {
(*message) = MPI_MESSAGE_NULL;
}
} else if (1 == ret) {
/**
* The search request completed and a matching message was found.
*/
*matched = 1;
*status = ofi_req.status;
ofi_req.match_state = 1;
(*message) = ompi_message_alloc();
if (NULL == (*message)) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
(*message)->comm = comm;
(*message)->req_ptr = ofi_req.message;
(*message)->peer = status->MPI_SOURCE;
(*message)->count = status->_ucount;
if (NULL == (*message)->req_ptr) {
ompi_message_return(*message);
*message = NULL;
return OMPI_ERR_OUT_OF_RESOURCE;
}
} else if (ret < 0 && ret == -FI_ENOMSG) {
/**
* The search request completed but no matching message was found.
*/
*matched = 0;
} else if (ret < 0 && ret != -FI_ENOMSG) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsearch failed: %d (%s)",
__FILE__, __LINE__, ret, fi_strerror(ret));
return ompi_mtl_ofi_get_error(-ret);
} else {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: unexpected return code from fi_tsearch: %d",
__FILE__, __LINE__, ret);
return ompi_mtl_ofi_get_error(-ret);
}
return OMPI_SUCCESS;
}

261
ompi/mca/mtl/ofi/mtl_ofi_recv.c
Просмотреть файл

@ -1,261 +0,0 @@
/*
* Copyright (c) 2013-2014 Intel, Inc. All rights reserved
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "opal/class/opal_list.h"
#include "ompi/communicator/communicator.h"
#include "ompi/datatype/ompi_datatype.h"
#include "opal/datatype/opal_convertor.h"
#include "ompi/mca/mtl/base/base.h"
#include "ompi/mca/mtl/base/mtl_base_datatype.h"
#include "ompi/message/message.h"
#include "mtl_ofi.h"
#include "mtl_ofi_types.h"
#include "mtl_ofi_endpoint.h"
#include "mtl_ofi_request.h"
#include "mtl_ofi_message.h"
/**
* Called when a completion for SYNC ACK send is received.
* This completes the synchronous recv operation. Thus, we
* call the upper layer's completion function.
*/
static int
ompi_mtl_ofi_sync_recv_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
ofi_req->super.completion_callback(&ofi_req->super);
return OMPI_SUCCESS;
}
/**
* Called when a completion for a posted recv is received.
*/
static int
ompi_mtl_ofi_recv_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
int ret;
ssize_t ret_length;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
int src;
ompi_status_public_t *status = NULL;
assert(ofi_req->super.ompi_req);
status = &ofi_req->super.ompi_req->req_status;
/**
* Any event associated with a request starts it.
* This prevents a started request from being cancelled.
*/
ofi_req->req_started = true;
status->MPI_SOURCE = MTL_OFI_GET_SOURCE(wc->tag);
status->MPI_TAG = MTL_OFI_GET_TAG(wc->tag);
status->_ucount = wc->len;
if (OPAL_UNLIKELY(wc->len > ofi_req->length)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"truncate expected: %ld %ld",
wc->len, ofi_req->length);
status->MPI_ERROR = MPI_ERR_TRUNCATE;
}
/**
* Unpack data into recv buffer if necessary.
*/
if (OPAL_UNLIKELY(ofi_req->buffer)) {
ret = ompi_mtl_datatype_unpack(ofi_req->convertor,
ofi_req->buffer,
wc->len);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: ompi_mtl_datatype_unpack failed: %d",
__FILE__, __LINE__, ret);
status->MPI_ERROR = ret;
}
}
/**
* We do not want any SYNC_SEND_ACK here!
* See mtl_ofi_send.c for details.
*/
assert(!MTL_OFI_IS_SYNC_SEND_ACK(wc->tag));
/**
* If this recv is part of an MPI_Ssend operation, then we send an
* acknowledgment back to the sender. The fi_context can be
* re-used safely because the previous operation has completed.
* This recv request will complete once we get a completion for
* this send. See ompi_mtl_ofi_sync_recv_callback().
* Otherwise, this request is now complete.
*/
if (OPAL_UNLIKELY(MTL_OFI_IS_SYNC_SEND(wc->tag))) {
ofi_req->event_callback = ompi_mtl_ofi_sync_recv_callback;
/**
* If the recv request was posted for any source,
* we need to extract the source's actual address.
*/
if (ompi_mtl_ofi.any_addr == ofi_req->remote_addr) {
src = MTL_OFI_GET_SOURCE(wc->tag);
ompi_proc = ompi_comm_peer_lookup(ofi_req->comm, src );
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
ofi_req->remote_addr = endpoint->peer_fiaddr;
}
ret_length = fi_tsend(ompi_mtl_ofi.ep,
NULL,
0,
ompi_mtl_ofi.mr,
ofi_req->remote_addr,
wc->tag | MTL_OFI_SYNC_SEND_ACK,
(void *) &ofi_req->ctx);
if (OPAL_UNLIKELY(ret_length < 0)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsend failed: %zd",
__FILE__, __LINE__, ret_length);
status->MPI_ERROR = OMPI_ERROR;
}
} else {
ofi_req->super.completion_callback(&ofi_req->super);
}
return OMPI_SUCCESS;
}
/**
* Called when an error occured on a recv request.
*/
static int
ompi_mtl_ofi_recv_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
ompi_status_public_t *status;
assert(ofi_req->super.ompi_req);
status = &ofi_req->super.ompi_req->req_status;
status->MPI_TAG = MTL_OFI_GET_TAG(ofi_req->match_bits);
status->MPI_SOURCE = MTL_OFI_GET_SOURCE(ofi_req->match_bits);
/* FIXME: This could be done on a single line... */
switch (error->err) {
case FI_EMSGSIZE:
status->MPI_ERROR = MPI_ERR_TRUNCATE;
break;
default:
status->MPI_ERROR = MPI_ERR_INTERN;
}
ofi_req->super.completion_callback(&ofi_req->super);
return OMPI_SUCCESS;
}
int
ompi_mtl_ofi_irecv(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int src,
int tag,
struct opal_convertor_t *convertor,
mca_mtl_request_t *mtl_request)
{
int ret = OMPI_SUCCESS;
ssize_t ret_length;
uint64_t match_bits, mask_bits;
fi_addr_t remote_addr;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
void *start;
size_t length;
bool free_after;
if (MPI_ANY_SOURCE != src) {
ompi_proc = ompi_comm_peer_lookup(comm, src);
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
remote_addr = endpoint->peer_fiaddr;
} else {
remote_addr = ompi_mtl_ofi.any_addr;
}
MTL_OFI_SET_RECV_BITS(match_bits, mask_bits, comm->c_contextid, src, tag);
ret = ompi_mtl_datatype_recv_buf(convertor, &start, &length, &free_after);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
return ret;
}
ofi_req->type = OMPI_MTL_OFI_RECV;
ofi_req->event_callback = ompi_mtl_ofi_recv_callback;
ofi_req->error_callback = ompi_mtl_ofi_recv_error_callback;
ofi_req->comm = comm;
ofi_req->buffer = (free_after) ? start : NULL;
ofi_req->length = length;
ofi_req->convertor = convertor;
ofi_req->req_started = false;
ofi_req->status.MPI_ERROR = OMPI_SUCCESS;
ofi_req->remote_addr = remote_addr;
ofi_req->match_bits = match_bits;
ret_length = fi_trecv(ompi_mtl_ofi.ep,
start,
length,
ompi_mtl_ofi.mr,
remote_addr,
match_bits,
mask_bits,
(void *)&ofi_req->ctx);
if (OPAL_UNLIKELY(ret_length < 0)) {
if (NULL != ofi_req->buffer) {
free(ofi_req->buffer);
}
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_trecv failed: %s(%zd)",
__FILE__, __LINE__, strerror(errno), ret_length);
return ompi_mtl_ofi_get_error(ret);
}
return OMPI_SUCCESS;
}
int
ompi_mtl_ofi_imrecv(struct mca_mtl_base_module_t *mtl,
struct opal_convertor_t *convertor,
struct ompi_message_t **message,
struct mca_mtl_request_t *mtl_request)
{
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
void *start;
size_t length;
bool free_after;
int ret;
ompi_mtl_ofi_message_t *ofi_message =
(ompi_mtl_ofi_message_t*) (*message)->req_ptr;
ret = ompi_mtl_datatype_recv_buf(convertor, &start, &length, &free_after);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
return ret;
}
ofi_req->type = OMPI_MTL_OFI_RECV;
ofi_req->event_callback = ompi_mtl_ofi_recv_callback;
ofi_req->error_callback = ompi_mtl_ofi_recv_error_callback;
ofi_req->buffer = (free_after) ? start : NULL;
ofi_req->length = length;
ofi_req->convertor = convertor;
ofi_req->status.MPI_ERROR = OMPI_SUCCESS;
(*message) = MPI_MESSAGE_NULL;
return ompi_mtl_ofi_recv_callback(&(ofi_message->wc), ofi_req);
}

240
ompi/mca/mtl/ofi/mtl_ofi_send.c
Просмотреть файл

@ -1,240 +0,0 @@
/*
* Copyright (c) 2013-2014 Intel, Inc. All rights reserved
*
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/communicator/communicator.h"
#include "opal/datatype/opal_convertor.h"
#include "ompi/mca/mtl/base/base.h"
#include "ompi/mca/mtl/base/mtl_base_datatype.h"
#include "mtl_ofi.h"
#include "mtl_ofi_types.h"
#include "mtl_ofi_request.h"
#include "mtl_ofi_endpoint.h"
static int
ompi_mtl_ofi_send_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
assert(ofi_req->completion_count > 0);
ofi_req->completion_count--;
return OMPI_SUCCESS;
}
static int
ompi_mtl_ofi_send_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
switch(error->err) {
case FI_EMSGSIZE:
ofi_req->status.MPI_ERROR = MPI_ERR_TRUNCATE;
break;
default:
ofi_req->status.MPI_ERROR = MPI_ERR_INTERN;
}
return ofi_req->event_callback(NULL, ofi_req);
}
static int
ompi_mtl_ofi_send_ack_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
ompi_mtl_ofi_request_t *parent_req = ofi_req->parent;
free(ofi_req);
parent_req->event_callback(NULL, parent_req);
return OMPI_SUCCESS;
}
static int
ompi_mtl_ofi_send_ack_error_callback(struct fi_cq_err_entry *error,
ompi_mtl_ofi_request_t *ofi_req)
{
ompi_mtl_ofi_request_t *parent_req = ofi_req->parent;
free(ofi_req);
parent_req->status.MPI_ERROR = MPI_ERR_INTERN;
return parent_req->error_callback(error, parent_req);
}
static int
ompi_mtl_ofi_isend_callback(struct fi_cq_tagged_entry *wc,
ompi_mtl_ofi_request_t *ofi_req)
{
assert(ofi_req->completion_count > 0);
ofi_req->completion_count--;
if (0 == ofi_req->completion_count) {
/* Request completed */
if (OPAL_UNLIKELY(NULL != ofi_req->buffer)) {
free(ofi_req->buffer);
ofi_req->buffer = NULL;
}
ofi_req->super.ompi_req->req_status.MPI_ERROR =
ofi_req->status.MPI_ERROR;
ofi_req->super.completion_callback(&ofi_req->super);
}
return OMPI_SUCCESS;
}
static inline int
ompi_mtl_ofi_send_start(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode,
ompi_mtl_ofi_request_t *ofi_req)
{
int ret;
void *start;
size_t length;
ssize_t ret_length;
bool free_after;
uint64_t match_bits;
ompi_proc_t *ompi_proc = NULL;
mca_mtl_ofi_endpoint_t *endpoint = NULL;
ompi_mtl_ofi_request_t *ack_req = NULL; /* For synchronous send */
ompi_proc = ompi_comm_peer_lookup(comm, dest);
endpoint = ompi_proc->proc_endpoints[OMPI_PROC_ENDPOINT_TAG_MTL];
ret = ompi_mtl_datatype_pack(convertor, &start, &length, &free_after);
if (OMPI_SUCCESS != ret) return ret;
ofi_req->buffer = (free_after) ? start : NULL;
ofi_req->length = length;
ofi_req->status.MPI_ERROR = OMPI_SUCCESS;
if (OPAL_UNLIKELY(MCA_PML_BASE_SEND_SYNCHRONOUS == mode)) {
ack_req = malloc(sizeof(ompi_mtl_ofi_request_t));
assert(ack_req);
ack_req->parent = ofi_req;
ack_req->event_callback = ompi_mtl_ofi_send_ack_callback;
ack_req->error_callback = ompi_mtl_ofi_send_ack_error_callback;
ofi_req->completion_count = 2;
MTL_OFI_SET_SEND_BITS(match_bits, comm->c_contextid,
comm->c_my_rank, tag, MTL_OFI_SYNC_SEND);
ret_length = fi_trecv(ompi_mtl_ofi.ep,
NULL,
0,
ompi_mtl_ofi.mr,
endpoint->peer_fiaddr,
match_bits | MTL_OFI_SYNC_SEND_ACK,
0, /* Exact match, no ignore bits */
(void *) &ack_req->ctx);
if (OPAL_UNLIKELY(ret_length < 0)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_trecv failed: %s(%zd)",
__FILE__, __LINE__,
strerror(errno), ret_length);
return ompi_mtl_ofi_get_error(ret);
}
} else {
ofi_req->completion_count = 1;
MTL_OFI_SET_SEND_BITS(match_bits, comm->c_contextid,
comm->c_my_rank, tag, 0);
}
ret_length = fi_tsend(ompi_mtl_ofi.ep,
start,
length,
ompi_mtl_ofi.mr,
endpoint->peer_fiaddr,
match_bits,
(void *) &ofi_req->ctx);
if (OPAL_UNLIKELY(0 > ret_length)) {
opal_output_verbose(1, ompi_mtl_base_framework.framework_output,
"%s:%d: fi_tsend failed: %zd",
__FILE__, __LINE__, ret_length);
return ompi_mtl_ofi_get_error(ret);
}
return OMPI_SUCCESS;
}
int
ompi_mtl_ofi_send(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode)
{
int ret = OMPI_SUCCESS;
ompi_mtl_ofi_request_t ofi_req;
/**
* Create a send request, start it and wait until it completes.
*/
ofi_req.event_callback = ompi_mtl_ofi_send_callback;
ofi_req.error_callback = ompi_mtl_ofi_send_error_callback;
ret = ompi_mtl_ofi_send_start(mtl, comm, dest, tag,
convertor, mode, &ofi_req);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
if (NULL != ofi_req.buffer) {
free(ofi_req.buffer);
}
return ret;
}
/**
* Wait until the request is completed.
* ompi_mtl_ofi_send_callback() updates this variable.
*/
while (0 < ofi_req.completion_count) {
ompi_mtl_ofi_progress();
}
if (OPAL_UNLIKELY(NULL != ofi_req.buffer)) {
free(ofi_req.buffer);
}
return ofi_req.status.MPI_ERROR;
}
int
ompi_mtl_ofi_isend(struct mca_mtl_base_module_t *mtl,
struct ompi_communicator_t *comm,
int dest,
int tag,
struct opal_convertor_t *convertor,
mca_pml_base_send_mode_t mode,
bool blocking,
mca_mtl_request_t *mtl_request)
{
int ret = OMPI_SUCCESS;
ompi_mtl_ofi_request_t *ofi_req = (ompi_mtl_ofi_request_t*) mtl_request;
ofi_req->event_callback = ompi_mtl_ofi_isend_callback;
ofi_req->error_callback = ompi_mtl_ofi_send_error_callback;
ret = ompi_mtl_ofi_send_start(mtl, comm, dest, tag,
convertor, mode, ofi_req);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret && NULL != ofi_req->buffer)) {
free(ofi_req->buffer);
}
return ret;
}

2
ompi/mca/mtl/ofi/mtl_ofi_types.h
Просмотреть файл

@ -71,8 +71,6 @@ typedef struct mca_mtl_ofi_component_t {
} mca_mtl_ofi_component_t; } mca_mtl_ofi_component_t;
OMPI_DECLSPEC extern mca_mtl_ofi_component_t mca_mtl_ofi_component;
/* match/ignore bit manipulation /* match/ignore bit manipulation
* *
* 0 123 4567 01234567 01234567 01234567 01234567 01234567 01234567 01234567 * 0 123 4567 01234567 01234567 01234567 01234567 01234567 01234567 01234567

67
ompi/mca/pml/cm/pml_cm_recv.c
Просмотреть файл

@ -82,6 +82,13 @@ mca_pml_cm_irecv(void *addr,
} }
void mca_pml_cm_recv_fast_completion(struct mca_mtl_request_t *mtl_request)
{
// Do nothing!
ompi_request_complete(mtl_request->ompi_req, true);
return;
}
int int
mca_pml_cm_recv(void *addr, mca_pml_cm_recv(void *addr,
size_t count, size_t count,
@ -92,35 +99,53 @@ mca_pml_cm_recv(void *addr,
ompi_status_public_t * status) ompi_status_public_t * status)
{ {
int ret; int ret;
mca_pml_cm_thin_recv_request_t *recvreq; ompi_proc_t *ompi_proc;
ompi_proc_t* ompi_proc; opal_convertor_t convertor;
mca_pml_cm_request_t req;
MCA_PML_CM_THIN_RECV_REQUEST_ALLOC(recvreq); mca_mtl_request_t *req_mtl =
if( OPAL_UNLIKELY(NULL == recvreq) ) return OMPI_ERR_OUT_OF_RESOURCE; alloca(sizeof(mca_mtl_request_t) + ompi_mtl->mtl_request_size);
MCA_PML_CM_THIN_RECV_REQUEST_INIT(recvreq, req_mtl->ompi_req = &req.req_ompi;
ompi_proc, req_mtl->completion_callback = mca_pml_cm_recv_fast_completion;
comm,
src, req.req_pml_type = MCA_PML_CM_REQUEST_RECV_THIN;
datatype, req.req_free_called = false;
addr, req.req_ompi.req_complete = false;
count); req.req_ompi.req_complete_cb = NULL;
req.req_ompi.req_state = OMPI_REQUEST_ACTIVE;
req.req_ompi.req_status.MPI_TAG = OMPI_ANY_TAG;
MCA_PML_CM_THIN_RECV_REQUEST_START(recvreq, comm, tag, src, ret); req.req_ompi.req_status.MPI_ERROR = OMPI_SUCCESS;
req.req_ompi.req_status._cancelled = 0;
if( MPI_ANY_SOURCE == src ) {
ompi_proc = ompi_proc_local_proc;
} else {
ompi_proc = ompi_comm_peer_lookup( comm, src );
}
opal_convertor_copy_and_prepare_for_recv(
ompi_proc->super.proc_convertor,
&(datatype->super),
count,
addr,
0,
&convertor );
ret = OMPI_MTL_CALL(irecv(ompi_mtl,
comm,
src,
tag,
&convertor,
req_mtl));
if( OPAL_UNLIKELY(OMPI_SUCCESS != ret) ) { if( OPAL_UNLIKELY(OMPI_SUCCESS != ret) ) {
/* BWB - XXX - need cleanup of request here */
MCA_PML_CM_THIN_RECV_REQUEST_RETURN(recvreq);
return ret; return ret;
} }
ompi_request_wait_completion(&recvreq->req_base.req_ompi); ompi_request_wait_completion(&req.req_ompi);
if (NULL != status) { /* return status */ if (NULL != status) { /* return status */
*status = recvreq->req_base.req_ompi.req_status; *status = req.req_ompi.req_status;
} }
ret = recvreq->req_base.req_ompi.req_status.MPI_ERROR; ret = req.req_ompi.req_status.MPI_ERROR;
ompi_request_free( (ompi_request_t**)&recvreq );
return ret; return ret;
} }

29
ompi/mca/pml/cm/pml_cm_send.c
Просмотреть файл

@ -150,31 +150,20 @@ mca_pml_cm_send(void *buf,
ompi_request_free( (ompi_request_t**)&sendreq ); ompi_request_free( (ompi_request_t**)&sendreq );
} else { } else {
mca_pml_cm_thin_send_request_t *sendreq; opal_convertor_t convertor;
ompi_proc_t * ompi_proc; ompi_proc_t *ompi_proc = ompi_comm_peer_lookup(comm, dst);
MCA_PML_CM_THIN_SEND_REQUEST_ALLOC(sendreq, comm, dst, ompi_proc);
if (OPAL_UNLIKELY(NULL == sendreq)) return OMPI_ERR_OUT_OF_RESOURCE; opal_convertor_copy_and_prepare_for_send(
ompi_proc->super.proc_convertor,
MCA_PML_CM_THIN_SEND_REQUEST_INIT(sendreq, &datatype->super, count, buf, 0,
ompi_proc, &convertor);
comm,
tag,
dst,
datatype,
sendmode,
buf,
count);
MCA_PML_CM_SEND_REQUEST_START_SETUP((&sendreq->req_send));
ret = OMPI_MTL_CALL(send(ompi_mtl, ret = OMPI_MTL_CALL(send(ompi_mtl,
comm, comm,
dst, dst,
tag, tag,
&sendreq->req_send.req_base.req_convertor, &convertor,
sendmode)); sendmode));
/* Allow a quick path for the request return */
sendreq->req_send.req_base.req_free_called = true;
MCA_PML_CM_THIN_SEND_REQUEST_PML_COMPLETE(sendreq);
} }
return ret; return ret;