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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
|
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "mpi.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/communicator/communicator.h"
#include "ompi/mca/coll/coll.h"
#include "ompi/mca/coll/base/coll_base_topo.h"
#include "ompi/mca/coll/base/coll_tags.h"
#include "ompi/mca/pml/pml.h"
#include "ompi/op/op.h"
#include "coll_tuned.h"
/* reduce algorithm variables */
static int coll_tuned_reduce_forced_algorithm = 0;
static int coll_tuned_reduce_segment_size = 0;
static int coll_tuned_reduce_max_requests;
static int coll_tuned_reduce_tree_fanout;
static int coll_tuned_reduce_chain_fanout;
/* valid values for coll_tuned_reduce_forced_algorithm */
static mca_base_var_enum_value_t reduce_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "chain"},
{3, "pipeline"},
{4, "binary"},
{5, "binomial"},
{6, "in-order_binary"},
{0, NULL}
};
/**
* The following are used by dynamic and forced rules
*
* publish details of each algorithm and if its forced/fixed/locked in
* as you add methods/algorithms you must update this and the query/map routines
*
* this routine is called by the component only
* this makes sure that the mca parameters are set to their initial values and
* perms module does not call this they call the forced_getvalues routine
* instead.
*/
int ompi_coll_tuned_reduce_intra_check_forced_init (coll_tuned_force_algorithm_mca_param_indices_t *mca_param_indices)
{
mca_base_var_enum_t*new_enum;
int cnt;
for( cnt = 0; NULL != reduce_algorithms[cnt].string; cnt++ );
ompi_coll_tuned_forced_max_algorithms[REDUCE] = cnt;
(void) mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_count",
"Number of reduce algorithms available",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_DEFAULT_ONLY,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,
&cnt);
/* MPI_T: This variable should eventually be bound to a communicator */
coll_tuned_reduce_forced_algorithm = 0;
(void) mca_base_var_enum_create("coll_tuned_reduce_algorithms", reduce_algorithms, &new_enum);
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm",
"Which reduce algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 chain, 3 pipeline, 4 binary, 5 binomial, 6 in-order binary",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_forced_algorithm);
OBJ_RELEASE(new_enum);
if (mca_param_indices->algorithm_param_index < 0) {
return mca_param_indices->algorithm_param_index;
}
coll_tuned_reduce_segment_size = 0;
mca_param_indices->segsize_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_segmentsize",
"Segment size in bytes used by default for reduce algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_segment_size);
coll_tuned_reduce_tree_fanout = ompi_coll_tuned_init_tree_fanout; /* get system wide default */
mca_param_indices->tree_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_tree_fanout",
"Fanout for n-tree used for reduce algorithms. Only has meaning if algorithm is forced and supports n-tree topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_tree_fanout);
coll_tuned_reduce_chain_fanout = ompi_coll_tuned_init_chain_fanout; /* get system wide default */
mca_param_indices->chain_fanout_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_chain_fanout",
"Fanout for chains used for reduce algorithms. Only has meaning if algorithm is forced and supports chain topo based operation.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_chain_fanout);
coll_tuned_reduce_max_requests = 0; /* no limit for reduce by default */
mca_param_indices->max_requests_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_algorithm_max_requests",
"Maximum number of outstanding send requests on leaf nodes. 0 means no limit.",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_READONLY,
&coll_tuned_reduce_max_requests);
if (mca_param_indices->max_requests_param_index < 0) {
return mca_param_indices->max_requests_param_index;
}
if (coll_tuned_reduce_max_requests < 0) {
if( 0 == ompi_comm_rank( MPI_COMM_WORLD ) ) {
opal_output( 0, "Maximum outstanding requests must be positive number or 0. Initializing to 0 (no limit).\n" );
}
coll_tuned_reduce_max_requests = 0;
}
return (MPI_SUCCESS);
}
int ompi_coll_tuned_reduce_intra_do_forced(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_tuned_module_t *tuned_module = (mca_coll_tuned_module_t*) module;
const int segsize = tuned_module->user_forced[REDUCE].segsize;
const int chain_fanout = tuned_module->user_forced[REDUCE].chain_fanout;
const int max_requests = tuned_module->user_forced[REDUCE].max_requests;
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_forced selected algorithm %d",
tuned_module->user_forced[REDUCE].algorithm));
switch (tuned_module->user_forced[REDUCE].algorithm) {
case (0): return ompi_coll_tuned_reduce_intra_dec_fixed(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (1): return ompi_coll_base_reduce_intra_basic_linear(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (2): return ompi_coll_base_reduce_intra_chain(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, chain_fanout, max_requests);
case (3): return ompi_coll_base_reduce_intra_pipeline(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (4): return ompi_coll_base_reduce_intra_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (5): return ompi_coll_base_reduce_intra_binomial(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (6): return ompi_coll_base_reduce_intra_in_order_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_forced attempt to select algorithm %d when only 0-%d is valid?",
tuned_module->user_forced[REDUCE].algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCE]));
return (MPI_ERR_ARG);
}
int ompi_coll_tuned_reduce_intra_do_this(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,
int algorithm, int faninout,
int segsize, int max_requests )
{
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_this selected algorithm %d topo faninout %d segsize %d",
algorithm, faninout, segsize));
switch (algorithm) {
case (0): return ompi_coll_tuned_reduce_intra_dec_fixed(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (1): return ompi_coll_base_reduce_intra_basic_linear(sbuf, rbuf, count, dtype,
op, root, comm, module);
case (2): return ompi_coll_base_reduce_intra_chain(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, faninout, max_requests);
case (3): return ompi_coll_base_reduce_intra_pipeline(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (4): return ompi_coll_base_reduce_intra_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (5): return ompi_coll_base_reduce_intra_binomial(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
case (6): return ompi_coll_base_reduce_intra_in_order_binary(sbuf, rbuf, count, dtype,
op, root, comm, module,
segsize, max_requests);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream,"coll:tuned:reduce_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",
algorithm, ompi_coll_tuned_forced_max_algorithms[REDUCE]));
return (MPI_ERR_ARG);
}
|
