Skip to content

Commit 25919a1

Browse files
hppritchahppritcha
authored
Merge pull request #3711 from gpaulsen/rel_v2.0.x_pr3500_pr3635
Rel v2.0.x pr3500 pr3635
2 parents 18f1659 + 7dd2bc0 commit 25919a1

File tree

1 file changed

+61
-60
lines changed

1 file changed

+61
-60
lines changed

‎ompi/mpi/c/sendrecv_replace.c

Lines changed: 61 additions & 60 deletions
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
33
* University Research and Technology
44
* Corporation. All rights reserved.
5-
* Copyright (c) 2004-2010 The University of Tennessee and The University
5+
* Copyright (c) 2004-2017 The University of Tennessee and The University
66
* of Tennessee Research Foundation. All rights
77
* reserved.
88
* Copyright (c) 2004-2008 High Performance Computing Center Stuttgart,
@@ -12,6 +12,7 @@
1212
* Copyright (c) 2010-2012 Oracle and/or its affiliates. All rights reserved.
1313
* Copyright (c) 2015 Research Organization for Information Science
1414
* and Technology (RIST). All rights reserved.
15+
* Copyright (c) 2017 IBM Corporation. All rights reserved.
1516
* $COPYRIGHT$
1617
*
1718
* Additional copyrights may follow
@@ -48,10 +49,10 @@ int MPI_Sendrecv_replace(void * buf, int count, MPI_Datatype datatype,
4849
int rc = MPI_SUCCESS;
4950

5051
MEMCHECKER(
51-
memchecker_datatype(datatype);
52-
memchecker_call(&opal_memchecker_base_isdefined, buf, count, datatype);
53-
memchecker_comm(comm);
54-
);
52+
memchecker_datatype(datatype);
53+
memchecker_call(&opal_memchecker_base_isdefined, buf, count, datatype);
54+
memchecker_comm(comm);
55+
);
5556

5657
if ( MPI_PARAM_CHECK ) {
5758
rc = MPI_SUCCESS;
@@ -74,66 +75,66 @@ int MPI_Sendrecv_replace(void * buf, int count, MPI_Datatype datatype,
7475

7576
/* simple case */
7677
if ( source == MPI_PROC_NULL || dest == MPI_PROC_NULL || count == 0 ) {
77-
rc = PMPI_Sendrecv(buf,count,datatype,dest,sendtag,buf,count,datatype,source,recvtag,comm,status);
78+
rc = PMPI_Sendrecv(buf, count, datatype, dest, sendtag, buf, count, datatype, source, recvtag, comm, status);
7879

7980
return rc;
80-
} else {
81-
82-
opal_convertor_t convertor;
83-
struct iovec iov;
84-
unsigned char recv_data[2048];
85-
size_t packed_size, max_data;
86-
uint32_t iov_count;
87-
ompi_status_public_t recv_status;
88-
ompi_proc_t* proc = ompi_comm_peer_lookup(comm,source);
89-
if(proc == NULL) {
90-
rc = MPI_ERR_RANK;
91-
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
92-
}
93-
94-
/* initialize convertor to unpack recv buffer */
95-
OBJ_CONSTRUCT(&convertor, opal_convertor_t);
96-
opal_convertor_copy_and_prepare_for_recv( proc->super.proc_convertor, &(datatype->super),
97-
count, buf, 0, &convertor );
98-
99-
/* setup a buffer for recv */
100-
opal_convertor_get_packed_size( &convertor, &packed_size );
101-
if( packed_size > sizeof(recv_data) ) {
102-
rc = PMPI_Alloc_mem(packed_size, MPI_INFO_NULL, &iov.iov_base);
103-
if(OMPI_SUCCESS != rc) {
104-
OMPI_ERRHANDLER_RETURN(OMPI_ERR_OUT_OF_RESOURCE, comm, MPI_ERR_BUFFER, FUNC_NAME);
105-
}
106-
} else {
107-
iov.iov_base = (caddr_t)recv_data;
108-
}
109-
110-
/* recv into temporary buffer */
111-
rc = PMPI_Sendrecv( buf, count, datatype, dest, sendtag, iov.iov_base, packed_size,
112-
MPI_BYTE, source, recvtag, comm, &recv_status );
113-
if (rc != MPI_SUCCESS) {
114-
if(packed_size > sizeof(recv_data))
115-
PMPI_Free_mem(iov.iov_base);
116-
OBJ_DESTRUCT(&convertor);
117-
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
118-
}
119-
120-
/* unpack into users buffer */
121-
iov.iov_len = recv_status._ucount;
122-
iov_count = 1;
123-
max_data = recv_status._ucount;
124-
opal_convertor_unpack(&convertor, &iov, &iov_count, &max_data );
81+
}
12582

126-
/* return status to user */
127-
if(status != MPI_STATUS_IGNORE) {
128-
*status = recv_status;
129-
}
83+
/**
84+
* If we look for an optimal solution, then we should receive the data into a temporary buffer
85+
* and once the send completes we would unpack back into the original buffer. However, if the
86+
* sender is unknown, this approach can only be implementing by receiving with the recv datatype
87+
* (potentially non-contiguous) and thus the allocated memory will be larger than the size of the
88+
* datatype. A simpler, but potentially less efficient approach is to work on the data we have
89+
* control of, aka the sent data, and pack it into a contiguous buffer before posting the receive.
90+
* Once the send completes, we free it.
91+
*/
92+
opal_convertor_t convertor;
93+
unsigned char packed_data[2048];
94+
struct iovec iov = { .iov_base = packed_data, .iov_len = sizeof(packed_data) };
95+
size_t packed_size, max_data;
96+
uint32_t iov_count;
97+
ompi_status_public_t recv_status;
98+
ompi_proc_t* proc = ompi_comm_peer_lookup(comm, dest);
99+
if(proc == NULL) {
100+
rc = MPI_ERR_RANK;
101+
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
102+
}
130103

131-
/* release resources */
132-
if(packed_size > sizeof(recv_data)) {
133-
PMPI_Free_mem(iov.iov_base);
104+
/* initialize convertor to unpack recv buffer */
105+
OBJ_CONSTRUCT(&convertor, opal_convertor_t);
106+
opal_convertor_copy_and_prepare_for_send( proc->super.proc_convertor, &(datatype->super),
107+
count, buf, 0, &convertor );
108+
109+
/* setup a buffer for recv */
110+
opal_convertor_get_packed_size( &convertor, &packed_size );
111+
if( packed_size > sizeof(packed_data) ) {
112+
rc = PMPI_Alloc_mem(packed_size, MPI_INFO_NULL, &iov.iov_base);
113+
if(OMPI_SUCCESS != rc) {
114+
rc = OMPI_ERR_OUT_OF_RESOURCE;
115+
goto cleanup_and_return;
134116
}
135-
OBJ_DESTRUCT(&convertor);
117+
iov.iov_len = packed_size;
118+
}
119+
max_data = packed_size;
120+
iov_count = 1;
121+
rc = opal_convertor_pack(&convertor, &iov, &iov_count, &max_data);
122+
123+
/* recv into temporary buffer */
124+
rc = PMPI_Sendrecv( iov.iov_base, packed_size, MPI_PACKED, dest, sendtag, buf, count,
125+
datatype, source, recvtag, comm, &recv_status );
126+
127+
cleanup_and_return:
128+
/* return status to user */
129+
if(status != MPI_STATUS_IGNORE) {
130+
*status = recv_status;
131+
}
136132

137-
return MPI_SUCCESS;
133+
/* release resources */
134+
if(packed_size > sizeof(packed_data)) {
135+
PMPI_Free_mem(iov.iov_base);
138136
}
137+
OBJ_DESTRUCT(&convertor);
138+
139+
OMPI_ERRHANDLER_RETURN(rc, comm, rc, FUNC_NAME);
139140
}

0 commit comments

Comments
 (0)