Skip to content

Commit a42415a

Browse files
janjustjanjust
authored
Merge pull request #12927 from burlen/v5.0.x
MCA variable scope and coll tuned rules file update onto v5.0.x
2 parents 8a5c2ef + 64ae8c2 commit a42415a

File tree

10 files changed

+201
-120
lines changed

10 files changed

+201
-120
lines changed

‎ompi/mca/coll/adapt/coll_adapt_component.c

Lines changed: 11 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,7 @@
22
* Copyright (c) 2014-2020 The University of Tennessee and The University
33
* of Tennessee Research Foundation. All rights
44
* reserved.
5+
* Copyright (c) 2024 NVIDIA CORPORATION. All rights reserved.
56
* $COPYRIGHT$
67
*
78
* Additional copyrights may follow
@@ -114,39 +115,39 @@ static int adapt_register(void)
114115
we should have a high priority */
115116
cs->adapt_priority = 0;
116117
(void) mca_base_component_var_register(c, "priority", "Priority of the adapt coll component",
117-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
118+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
118119
OPAL_INFO_LVL_9,
119-
MCA_BASE_VAR_SCOPE_READONLY, &cs->adapt_priority);
120+
MCA_BASE_VAR_SCOPE_ALL, &cs->adapt_priority);
120121

121122
cs->adapt_verbose = ompi_coll_base_framework.framework_verbose;
122123
(void) mca_base_component_var_register(c, "verbose",
123124
"Verbose level (default set to the collective framework verbosity)",
124-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
125+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
125126
OPAL_INFO_LVL_9,
126-
MCA_BASE_VAR_SCOPE_READONLY, &cs->adapt_verbose);
127+
MCA_BASE_VAR_SCOPE_ALL, &cs->adapt_verbose);
127128

128129
cs->adapt_context_free_list_min = 64;
129130
(void) mca_base_component_var_register(c, "context_free_list_min",
130131
"Minimum number of segments in context free list",
131-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
132+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
132133
OPAL_INFO_LVL_9,
133-
MCA_BASE_VAR_SCOPE_READONLY,
134+
MCA_BASE_VAR_SCOPE_ALL,
134135
&cs->adapt_context_free_list_min);
135136

136137
cs->adapt_context_free_list_max = 1024;
137138
(void) mca_base_component_var_register(c, "context_free_list_max",
138139
"Maximum number of segments in context free list",
139-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
140+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
140141
OPAL_INFO_LVL_9,
141-
MCA_BASE_VAR_SCOPE_READONLY,
142+
MCA_BASE_VAR_SCOPE_ALL,
142143
&cs->adapt_context_free_list_max);
143144

144145
cs->adapt_context_free_list_inc = 32;
145146
(void) mca_base_component_var_register(c, "context_free_list_inc",
146147
"Increasement number of segments in context free list",
147-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
148+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
148149
OPAL_INFO_LVL_9,
149-
MCA_BASE_VAR_SCOPE_READONLY,
150+
MCA_BASE_VAR_SCOPE_ALL,
150151
&cs->adapt_context_free_list_inc);
151152
ompi_coll_adapt_ibcast_register();
152153
ompi_coll_adapt_ireduce_register();

‎ompi/mca/coll/adapt/coll_adapt_ibcast.c

Lines changed: 12 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -3,6 +3,7 @@
33
* of Tennessee Research Foundation. All rights
44
* reserved.
55
* Copyright (c) 2022 IBM Corporation. All rights reserved
6+
* Copyright (c) 2024 NVIDIA CORPORATION. All rights reserved.
67
* $COPYRIGHT$
78
*
89
* Additional copyrights may follow
@@ -34,8 +35,9 @@ int ompi_coll_adapt_ibcast_register(void)
3435

3536
mca_coll_adapt_component.adapt_ibcast_algorithm = 1;
3637
mca_base_component_var_register(c, "bcast_algorithm",
37-
"Algorithm of broadcast, 0: tuned, 1: binomial, 2: in_order_binomial, 3: binary, 4: pipeline, 5: chain, 6: linear", MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
38-
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_READONLY,
38+
"Algorithm of broadcast, 0: tuned, 1: binomial, 2: in_order_binomial, 3: binary, 4: pipeline, 5: chain, 6: linear",
39+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
40+
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_ALL,
3941
&mca_coll_adapt_component.adapt_ibcast_algorithm);
4042
if( (mca_coll_adapt_component.adapt_ibcast_algorithm < 0) ||
4143
(mca_coll_adapt_component.adapt_ibcast_algorithm >= OMPI_COLL_ADAPT_ALGORITHM_COUNT) ) {
@@ -45,33 +47,33 @@ int ompi_coll_adapt_ibcast_register(void)
4547
mca_coll_adapt_component.adapt_ibcast_segment_size = 0;
4648
mca_base_component_var_register(c, "bcast_segment_size",
4749
"Segment size in bytes used by default for bcast algorithms. Only has meaning if algorithm is forced and supports segmenting. 0 bytes means no segmentation.",
48-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
50+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
4951
OPAL_INFO_LVL_5,
50-
MCA_BASE_VAR_SCOPE_READONLY,
52+
MCA_BASE_VAR_SCOPE_ALL,
5153
&mca_coll_adapt_component.adapt_ibcast_segment_size);
5254

5355
mca_coll_adapt_component.adapt_ibcast_max_send_requests = 2;
5456
mca_base_component_var_register(c, "bcast_max_send_requests",
5557
"Maximum number of send requests",
56-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
58+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
5759
OPAL_INFO_LVL_5,
58-
MCA_BASE_VAR_SCOPE_READONLY,
60+
MCA_BASE_VAR_SCOPE_ALL,
5961
&mca_coll_adapt_component.adapt_ibcast_max_send_requests);
6062

6163
mca_coll_adapt_component.adapt_ibcast_max_recv_requests = 3;
6264
mca_base_component_var_register(c, "bcast_max_recv_requests",
6365
"Maximum number of receive requests",
64-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
66+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
6567
OPAL_INFO_LVL_5,
66-
MCA_BASE_VAR_SCOPE_READONLY,
68+
MCA_BASE_VAR_SCOPE_ALL,
6769
&mca_coll_adapt_component.adapt_ibcast_max_recv_requests);
6870

6971
mca_coll_adapt_component.adapt_ibcast_synchronous_send = true;
7072
(void) mca_base_component_var_register(c, "bcast_synchronous_send",
7173
"Whether to use synchronous send operations during setup of bcast operations",
72-
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
74+
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
7375
OPAL_INFO_LVL_9,
74-
MCA_BASE_VAR_SCOPE_READONLY,
76+
MCA_BASE_VAR_SCOPE_ALL,
7577
&mca_coll_adapt_component.adapt_ibcast_synchronous_send);
7678

7779
mca_coll_adapt_component.adapt_ibcast_context_free_list = NULL;

‎ompi/mca/coll/adapt/coll_adapt_ireduce.c

Lines changed: 18 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@
55
* Copyright (c) 2020 Cisco Systems, Inc. All rights reserved.
66
* Copyright (c) 2022 IBM Corporation. All rights reserved
77
* Copyright (c) 2023 Jeffrey M. Squyres. All rights reserved.
8+
* Copyright (c) 2024 NVIDIA CORPORATION. All rights reserved.
89
* $COPYRIGHT$
910
*
1011
* Additional copyrights may follow
@@ -38,8 +39,9 @@ int ompi_coll_adapt_ireduce_register(void)
3839

3940
mca_coll_adapt_component.adapt_ireduce_algorithm = 1;
4041
mca_base_component_var_register(c, "reduce_algorithm",
41-
"Algorithm of reduce, 1: binomial, 2: in_order_binomial, 3: binary, 4: pipeline, 5: chain, 6: linear", MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
42-
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_READONLY,
42+
"Algorithm of reduce, 1: binomial, 2: in_order_binomial, 3: binary, 4: pipeline, 5: chain, 6: linear",
43+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
44+
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_ALL,
4345
&mca_coll_adapt_component.adapt_ireduce_algorithm);
4446
if( (mca_coll_adapt_component.adapt_ireduce_algorithm < 0) ||
4547
(mca_coll_adapt_component.adapt_ireduce_algorithm > OMPI_COLL_ADAPT_ALGORITHM_COUNT) ) {
@@ -49,58 +51,58 @@ int ompi_coll_adapt_ireduce_register(void)
4951
mca_coll_adapt_component.adapt_ireduce_segment_size = 163740;
5052
mca_base_component_var_register(c, "reduce_segment_size",
5153
"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.",
52-
MCA_BASE_VAR_TYPE_SIZE_T, NULL, 0, 0,
54+
MCA_BASE_VAR_TYPE_SIZE_T, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
5355
OPAL_INFO_LVL_5,
54-
MCA_BASE_VAR_SCOPE_READONLY,
56+
MCA_BASE_VAR_SCOPE_ALL,
5557
&mca_coll_adapt_component.adapt_ireduce_segment_size);
5658

5759
mca_coll_adapt_component.adapt_ireduce_max_send_requests = 2;
5860
mca_base_component_var_register(c, "reduce_max_send_requests",
5961
"Maximum number of send requests",
60-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
62+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
6163
OPAL_INFO_LVL_5,
62-
MCA_BASE_VAR_SCOPE_READONLY,
64+
MCA_BASE_VAR_SCOPE_ALL,
6365
&mca_coll_adapt_component.adapt_ireduce_max_send_requests);
6466

6567
mca_coll_adapt_component.adapt_ireduce_max_recv_requests = 3;
6668
mca_base_component_var_register(c, "reduce_max_recv_requests",
6769
"Maximum number of receive requests per peer",
68-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
70+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
6971
OPAL_INFO_LVL_5,
70-
MCA_BASE_VAR_SCOPE_READONLY,
72+
MCA_BASE_VAR_SCOPE_ALL,
7173
&mca_coll_adapt_component.adapt_ireduce_max_recv_requests);
7274

7375
mca_coll_adapt_component.adapt_inbuf_free_list_min = 10;
7476
mca_base_component_var_register(c, "inbuf_free_list_min",
7577
"Minimum number of segment in inbuf free list",
76-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
78+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
7779
OPAL_INFO_LVL_5,
78-
MCA_BASE_VAR_SCOPE_READONLY,
80+
MCA_BASE_VAR_SCOPE_ALL,
7981
&mca_coll_adapt_component.adapt_inbuf_free_list_min);
8082

8183
mca_coll_adapt_component.adapt_inbuf_free_list_max = 10000;
8284
mca_base_component_var_register(c, "inbuf_free_list_max",
8385
"Maximum number of segment in inbuf free list",
84-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
86+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
8587
OPAL_INFO_LVL_5,
86-
MCA_BASE_VAR_SCOPE_READONLY,
88+
MCA_BASE_VAR_SCOPE_ALL,
8789
&mca_coll_adapt_component.adapt_inbuf_free_list_max);
8890

8991

9092
mca_coll_adapt_component.adapt_inbuf_free_list_inc = 10;
9193
mca_base_component_var_register(c, "inbuf_free_list_inc",
9294
"Number of segments to allocate when growing the inbuf free list",
93-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
95+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
9496
OPAL_INFO_LVL_5,
95-
MCA_BASE_VAR_SCOPE_READONLY,
97+
MCA_BASE_VAR_SCOPE_ALL,
9698
&mca_coll_adapt_component.adapt_inbuf_free_list_inc);
9799

98100
mca_coll_adapt_component.adapt_ireduce_synchronous_send = true;
99101
(void) mca_base_component_var_register(c, "reduce_synchronous_send",
100102
"Whether to use synchronous send operations during setup of reduce operations",
101-
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, 0,
103+
MCA_BASE_VAR_TYPE_BOOL, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
102104
OPAL_INFO_LVL_9,
103-
MCA_BASE_VAR_SCOPE_READONLY,
105+
MCA_BASE_VAR_SCOPE_ALL,
104106
&mca_coll_adapt_component.adapt_ireduce_synchronous_send);
105107

106108
mca_coll_adapt_component.adapt_ireduce_context_free_list = NULL;

‎ompi/mca/coll/base/coll_base_util.c

Lines changed: 23 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -12,6 +12,9 @@
1212
* Copyright (c) 2014-2020 Research Organization for Information Science
1313
* and Technology (RIST). All rights reserved.
1414
* Copyright (c) 2023 Jeffrey M. Squyres. All rights reserved.
15+
*
16+
* Copyright (c) 2024 NVIDIA CORPORATION. All rights reserved.
17+
*
1518
* $COPYRIGHT$
1619
*
1720
* Additional copyrights may follow
@@ -482,6 +485,26 @@ int ompi_coll_base_file_peek_next_char_is(FILE *fptr, int *fileline, int expecte
482485
} while (1);
483486
}
484487

488+
/**
489+
* return non-zero if the next non-space to read on the current line is a digit.
490+
* otherwise return 0.
491+
*/
492+
int ompi_coll_base_file_peek_next_char_isdigit(FILE *fptr)
493+
{
494+
do {
495+
int next = fgetc(fptr);
496+
497+
if ((' ' == next) || ('\t' == next)) {
498+
continue; /* discard space and tab. keep everything else */
499+
}
500+
501+
ungetc(next, fptr); /* put the char back into the stream */
502+
503+
return isdigit(next); /* report back whether or not next is a digit */
504+
505+
} while (1);
506+
}
507+
485508
/**
486509
* There are certainly simpler implementation for this function when performance
487510
* is not a critical point. But, as this function is used during the collective

‎ompi/mca/coll/base/coll_base_util.h

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -11,6 +11,7 @@
1111
* All rights reserved.
1212
* Copyright (c) 2014-2020 Research Organization for Information Science
1313
* and Technology (RIST). All rights reserved.
14+
* Copyright (c) 2024 NVIDIA CORPORATION. All rights reserved.
1415
* $COPYRIGHT$
1516
*
1617
* Additional copyrights may follow
@@ -195,6 +196,7 @@ int ompi_coll_base_file_getnext_string(FILE *fptr, int *fileline, char** val);
195196
* eat the value, otherwise put it back into the file.
196197
*/
197198
int ompi_coll_base_file_peek_next_char_is(FILE *fptr, int *fileline, int expected);
199+
int ompi_coll_base_file_peek_next_char_isdigit(FILE *fptr);
198200

199201
/* Miscellaneous function */
200202
const char* mca_coll_base_colltype_to_str(int collid);

‎ompi/mca/coll/basic/coll_basic_component.c

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -13,6 +13,7 @@
1313
* Copyright (c) 2008 Cisco Systems, Inc. All rights reserved.
1414
* Copyright (c) 2015 Los Alamos National Security, LLC. All rights
1515
* reserved.
16+
* Copyright (c) 2024 NVIDIA CORPORATION. All rights reserved.
1617
* $COPYRIGHT$
1718
*
1819
* Additional copyrights may follow
@@ -91,16 +92,16 @@ basic_register(void)
9192
mca_coll_basic_priority = 10;
9293
(void) mca_base_component_var_register(&mca_coll_basic_component.collm_version, "priority",
9394
"Priority of the basic coll component",
94-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
95+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
9596
OPAL_INFO_LVL_9,
96-
MCA_BASE_VAR_SCOPE_READONLY,
97+
MCA_BASE_VAR_SCOPE_ALL,
9798
&mca_coll_basic_priority);
9899
mca_coll_basic_crossover = 4;
99100
(void) mca_base_component_var_register(&mca_coll_basic_component.collm_version, "crossover",
100101
"Minimum number of processes in a communicator before using the logarithmic algorithms",
101-
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
102+
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
102103
OPAL_INFO_LVL_9,
103-
MCA_BASE_VAR_SCOPE_READONLY,
104+
MCA_BASE_VAR_SCOPE_ALL,
104105
&mca_coll_basic_crossover);
105106

106107
return OMPI_SUCCESS;

0 commit comments

Comments
 (0)