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
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
|
/*
* Copyright (c) 2004-2023 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2014-2015 NVIDIA Corporation. All rights reserved.
* Copyright (c) 2022 Amazon.com, Inc. or its affiliates. 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"
/*
* reduce_log_inter
*
* Function: - reduction using O(N) algorithm
* Accepts: - same as MPI_Reduce()
* Returns: - MPI_SUCCESS or error code
*/
int
mca_coll_cuda_reduce(const void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
int root, struct ompi_communicator_t *comm,
mca_coll_base_module_t *module)
{
mca_coll_cuda_module_t *s = (mca_coll_cuda_module_t*) module;
int rank = ompi_comm_rank(comm);
ptrdiff_t gap;
char *rbuf1 = NULL, *sbuf1 = NULL, *rbuf2 = NULL;
size_t bufsize;
int rc;
bufsize = opal_datatype_span(&dtype->super, count, &gap);
rc = mca_coll_cuda_check_buf((void *)sbuf);
if (rc < 0) {
return rc;
}
if ((MPI_IN_PLACE != sbuf) && (rc > 0)) {
sbuf1 = (char*)malloc(bufsize);
if (NULL == sbuf1) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
mca_coll_cuda_memcpy(sbuf1, sbuf, bufsize);
sbuf = sbuf1 - gap;
}
rc = mca_coll_cuda_check_buf(rbuf);
if (rc < 0) {
return rc;
}
if ((rank == root) && (rc > 0)) {
rbuf1 = (char*)malloc(bufsize);
if (NULL == rbuf1) {
if (NULL != sbuf1) free(sbuf1);
return OMPI_ERR_OUT_OF_RESOURCE;
}
mca_coll_cuda_memcpy(rbuf1, rbuf, bufsize);
rbuf2 = rbuf; /* save away original buffer */
rbuf = rbuf1 - gap;
}
rc = s->c_coll.coll_reduce((void *) sbuf, rbuf, count,
dtype, op, root, comm,
s->c_coll.coll_reduce_module);
if (NULL != sbuf1) {
free(sbuf1);
}
if (NULL != rbuf1) {
rbuf = rbuf2;
mca_coll_cuda_memcpy(rbuf, rbuf1, bufsize);
free(rbuf1);
}
return rc;
}
int
mca_coll_cuda_reduce_local(const void *sbuf, void *rbuf, int count,
struct ompi_datatype_t *dtype,
struct ompi_op_t *op,
mca_coll_base_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);
rc = mca_coll_cuda_check_buf((void *)sbuf);
if (rc < 0) {
return rc;
}
if ((MPI_IN_PLACE != sbuf) && (rc > 0)) {
sbuf1 = (char*)malloc(bufsize);
if (NULL == sbuf1) {
return OMPI_ERR_OUT_OF_RESOURCE;
}
mca_coll_cuda_memcpy(sbuf1, sbuf, bufsize);
sbuf = sbuf1 - gap;
}
rc = mca_coll_cuda_check_buf(rbuf);
if (rc < 0) {
return rc;
}
if (rc > 0) {
rbuf1 = (char*)malloc(bufsize);
if (NULL == rbuf1) {
if (NULL != sbuf1) free(sbuf1);
return OMPI_ERR_OUT_OF_RESOURCE;
}
mca_coll_cuda_memcpy(rbuf1, rbuf, bufsize);
rbuf2 = rbuf; /* save away original buffer */
rbuf = rbuf1 - gap;
}
ompi_op_reduce(op, (void *)sbuf, rbuf, count, dtype);
rc = OMPI_SUCCESS;
if (NULL != sbuf1) {
free(sbuf1);
}
if (NULL != rbuf1) {
rbuf = rbuf2;
mca_coll_cuda_memcpy(rbuf, rbuf1, bufsize);
free(rbuf1);
}
return rc;
}
|
