1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
|
/*
* Copyright (c) 2014-2017 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2014-2015 NVIDIA Corporation. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "coll_cuda.h"
#include <stdio.h>
#include "ompi/op/op.h"
#include "opal/datatype/opal_convertor.h"
#include "opal/datatype/opal_datatype_cuda.h"
/*
* allreduce_intra
*
* Function: - allreduce using other MPI collectives
* Accepts: - same as MPI_Allreduce()
* Returns: - MPI_SUCCESS or error code
*/
int
mca_coll_cuda_allreduce(const void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_cuda_module_t *s = (mca_coll_cuda_module_t*) module;
ptrdiff_t gap;
char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
size_t bufsize;
int rc;
bufsize = opal_datatype_span(&dtype->super, count, &gap);
if ((MPI_IN_PLACE != sbuf) && (opal_cuda_check_bufs((char *)sbuf, NULL))) {
sbuf1 = (char*)malloc(bufsize);
if (NULL == sbuf1) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
opal_cuda_memcpy_sync(sbuf1, sbuf, bufsize);
sbuf = sbuf1 - gap;
}
if (opal_cuda_check_bufs(rbuf, NULL)) {
rbuf1 = (char*)malloc(bufsize);
if (NULL == rbuf1) {
if (NULL != sbuf1) free(sbuf1);
return OMPI_ERR_OUT_OF_RESOURCE;
}
opal_cuda_memcpy_sync(rbuf1, rbuf, bufsize);
rbuf2 = rbuf; /* save away original buffer */
rbuf = rbuf1 - gap;
}
rc = s->c_coll.coll_allreduce(sbuf, rbuf, count, dtype, op, comm, s->c_coll.coll_allreduce_module);
if (NULL != sbuf1) {
free(sbuf1);
}
if (NULL != rbuf1) {
rbuf = rbuf2;
opal_cuda_memcpy_sync(rbuf, rbuf1, bufsize);
free(rbuf1);
}
return rc;
}
|
