ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность

Merge pull request #8198 from devreal/fix-tuned-dynamic-v4.1.x

Просмотр исходного кода 
Fix some issues with dynamic algorithm selection in coll/tuned
Этот коммит содержится в:
Jeff Squyres 2020-11-12 13:57:21 -05:00 коммит произвёл GitHub
родитель ff89195381 3cae9f7874
Коммит 6f21a39372
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: 4AEE18F83AFDEB23
16 изменённых файлов: 78 добавлений и 101 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

7
ompi/mca/coll/tuned/coll_tuned_allgather_decision.c
Просмотреть файл

@ -1,5 +1,5 @@
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
@ -31,7 +31,7 @@ static int coll_tuned_allgather_tree_fanout;
static int coll_tuned_allgather_chain_fanout;
/* valid values for coll_tuned_allgather_forced_algorithm */
static mca_base_var_enum_value_t allgather_algorithms[] = {
static const mca_base_var_enum_value_t allgather_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "bruck"},
@ -77,7 +77,8 @@ ompi_coll_tuned_allgather_intra_check_forced_init(coll_tuned_force_algorithm_mca
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgather_algorithm",
"Which allallgather algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 bruck, 3 recursive doubling, 4 ring, 5 neighbor exchange, 6: two proc only.",
"Which allallgather algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 bruck, 3 recursive doubling, 4 ring, 5 neighbor exchange, 6: two proc only. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_allgatherv_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
@ -31,7 +31,7 @@ static int coll_tuned_allgatherv_tree_fanout;
static int coll_tuned_allgatherv_chain_fanout;
/* valid values for coll_tuned_allgatherv_forced_algorithm */
static mca_base_var_enum_value_t allgatherv_algorithms[] = {
static const mca_base_var_enum_value_t allgatherv_algorithms[] = {
{0, "ignore"},
{1, "default"},
{2, "bruck"},
@ -76,7 +76,8 @@ ompi_coll_tuned_allgatherv_intra_check_forced_init(coll_tuned_force_algorithm_mc
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allgatherv_algorithm",
"Which allallgatherv algorithm is used. Can be locked down to choice of: 0 ignore, 1 default (allgathervv + bcast), 2 bruck, 3 ring, 4 neighbor exchange, 5: two proc only.",
"Which allallgatherv algorithm is used. Can be locked down to choice of: 0 ignore, 1 default (allgathervv + bcast), 2 bruck, 3 ring, 4 neighbor exchange, 5: two proc only. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_CONSTANT,

7
ompi/mca/coll/tuned/coll_tuned_allreduce_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015-2018 Research Organization for Information Science
@ -34,7 +34,7 @@ static int coll_tuned_allreduce_tree_fanout;
static int coll_tuned_allreduce_chain_fanout;
/* valid values for coll_tuned_allreduce_forced_algorithm */
static mca_base_var_enum_value_t allreduce_algorithms[] = {
static const mca_base_var_enum_value_t allreduce_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "nonoverlapping"},
@ -77,7 +77,8 @@ int ompi_coll_tuned_allreduce_intra_check_forced_init (coll_tuned_force_algorith
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"allreduce_algorithm",
"Which allreduce algorithm is used. Can be locked down to any of: 0 ignore, 1 basic linear, 2 nonoverlapping (tuned reduce + tuned bcast), 3 recursive doubling, 4 ring, 5 segmented ring",
"Which allreduce algorithm is used. Can be locked down to any of: 0 ignore, 1 basic linear, 2 nonoverlapping (tuned reduce + tuned bcast), 3 recursive doubling, 4 ring, 5 segmented ring. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_alltoall_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
@ -32,7 +32,7 @@ static int coll_tuned_alltoall_tree_fanout;
static int coll_tuned_alltoall_chain_fanout;
/* valid values for coll_tuned_alltoall_forced_algorithm */
static mca_base_var_enum_value_t alltoall_algorithms[] = {
static const mca_base_var_enum_value_t alltoall_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "pairwise"},
@ -74,7 +74,8 @@ int ompi_coll_tuned_alltoall_intra_check_forced_init (coll_tuned_force_algorithm
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"alltoall_algorithm",
"Which alltoall algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 pairwise, 3: modified bruck, 4: linear with sync, 5:two proc only.",
"Which alltoall algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 pairwise, 3: modified bruck, 4: linear with sync, 5:two proc only. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_alltoallv_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
@ -29,7 +29,7 @@
static int coll_tuned_alltoallv_forced_algorithm = 0;
/* valid values for coll_tuned_alltoallv_forced_algorithm */
static mca_base_var_enum_value_t alltoallv_algorithms[] = {
static const mca_base_var_enum_value_t alltoallv_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "pairwise"},
@ -71,7 +71,8 @@ int ompi_coll_tuned_alltoallv_intra_check_forced_init(coll_tuned_force_algorithm
"alltoallv_algorithm",
"Which alltoallv algorithm is used. "
"Can be locked down to choice of: 0 ignore, "
"1 basic linear, 2 pairwise.",
"1 basic linear, 2 pairwise. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_barrier_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2015 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
@ -27,7 +27,7 @@
static int coll_tuned_barrier_forced_algorithm = 0;
/* valid values for coll_tuned_barrier_forced_algorithm */
static mca_base_var_enum_value_t barrier_algorithms[] = {
static const mca_base_var_enum_value_t barrier_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "double_ring"},
@ -72,7 +72,8 @@ int ompi_coll_tuned_barrier_intra_check_forced_init (coll_tuned_force_algorithm_
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"barrier_algorithm",
"Which barrier algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 double ring, 3: recursive doubling 4: bruck, 5: two proc only, 6: tree",
"Which barrier algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 double ring, 3: recursive doubling 4: bruck, 5: two proc only, 6: tree. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_bcast_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
@ -34,7 +34,7 @@ static int coll_tuned_bcast_chain_fanout;
static int coll_tuned_bcast_knomial_radix = 4;
/* valid values for coll_tuned_bcast_forced_algorithm */
static mca_base_var_enum_value_t bcast_algorithms[] = {
static const mca_base_var_enum_value_t bcast_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "chain"},
@ -80,7 +80,8 @@ int ompi_coll_tuned_bcast_intra_check_forced_init (coll_tuned_force_algorithm_mc
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"bcast_algorithm",
"Which bcast algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 chain, 3: pipeline, 4: split binary tree, 5: binary tree, 6: binomial tree, 7: knomial tree, 8: scatter_allgather, 9: scatter_allgather_ring.",
"Which bcast algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 chain, 3: pipeline, 4: split binary tree, 5: binary tree, 6: binomial tree, 7: knomial tree, 8: scatter_allgather, 9: scatter_allgather_ring. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

6
ompi/mca/coll/tuned/coll_tuned_decision_dynamic.c
Просмотреть файл

@ -2,7 +2,7 @@
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2015 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@ -97,7 +97,7 @@ ompi_coll_tuned_allreduce_intra_dec_dynamic (const void *sbuf, void *rbuf, int c
*
* Function: - seletects alltoall algorithm to use
* Accepts: - same arguments as MPI_Alltoall()
* Returns: - MPI_SUCCESS or error code (passed from the bcast implementation)
* Returns: - MPI_SUCCESS or error code (passed from the alltoall implementation)
*/
int ompi_coll_tuned_alltoall_intra_dec_dynamic(const void *sbuf, int scount,
@ -238,7 +238,7 @@ int ompi_coll_tuned_barrier_intra_dec_dynamic(struct ompi_communicator_t *comm,
/*
* bcast_intra_dec
*
* Function: - seletects broadcast algorithm to use
* Function: - selects broadcast algorithm to use
* Accepts: - same arguments as MPI_Bcast()
* Returns: - MPI_SUCCESS or error code (passed from the bcast implementation)
*/

70
ompi/mca/coll/tuned/coll_tuned_decision_fixed.c
Просмотреть файл

@ -3,7 +3,7 @@
* Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2015 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
@ -92,30 +92,10 @@ ompi_coll_tuned_allreduce_intra_dec_fixed(const void *sbuf, void *rbuf, int coun
} else {
alg = 2;
}
} else if (communicator_size < 32) {
alg = 3;
} else if (communicator_size < 64) {
if (total_dsize < 131072) {
alg = 1;
} else {
alg = 3;
}
} else if (communicator_size < 128) {
if (total_dsize < 128) {
alg = 1;
} else if (total_dsize < 512) {
alg = 3;
} else if (total_dsize < 8192) {
alg = 1;
} else {
alg = 3;
}
alg = 3;
} else if (communicator_size < 256) {
if (total_dsize < 2048) {
alg = 2;
} else if (total_dsize < 16384) {
alg = 1;
} else if (total_dsize < 131072) {
if (total_dsize < 131072) {
alg = 2;
} else if (total_dsize < 524288) {
alg = 3;
@ -183,23 +163,13 @@ ompi_coll_tuned_allreduce_intra_dec_fixed(const void *sbuf, void *rbuf, int coun
alg = 6;
}
} else if (communicator_size < 128) {
if (total_dsize < 128) {
alg = 1;
} else if (total_dsize < 512) {
alg = 3;
} else if (total_dsize < 8192) {
alg = 1;
} else if (total_dsize < 262144) {
if (total_dsize < 262144) {
alg = 3;
} else {
alg = 6;
}
} else if (communicator_size < 256) {
if (total_dsize < 2048) {
alg = 2;
} else if (total_dsize < 16384) {
alg = 1;
} else if (total_dsize < 131072) {
if (total_dsize < 131072) {
alg = 2;
} else if (total_dsize < 262144) {
alg = 3;
@ -567,9 +537,7 @@ int ompi_coll_tuned_bcast_intra_dec_fixed(void *buff, int count,
* {9, "scatter_allgather_ring"},
*/
if (communicator_size < 4) {
if (total_dsize < 2) {
alg = 9;
} else if (total_dsize < 32) {
if (total_dsize < 32) {
alg = 3;
} else if (total_dsize < 256) {
alg = 5;
@ -591,9 +559,7 @@ int ompi_coll_tuned_bcast_intra_dec_fixed(void *buff, int count,
alg = 5;
}
} else if (communicator_size < 8) {
if (total_dsize < 2) {
alg = 8;
} else if (total_dsize < 64) {
if (total_dsize < 64) {
alg = 5;
} else if (total_dsize < 128) {
alg = 6;
@ -639,8 +605,6 @@ int ompi_coll_tuned_bcast_intra_dec_fixed(void *buff, int count,
} else if (communicator_size < 256) {
if (total_dsize < 2) {
alg = 6;
} else if (total_dsize < 128) {
alg = 8;
} else if (total_dsize < 16384) {
alg = 5;
} else if (total_dsize < 32768) {
@ -1164,9 +1128,7 @@ int ompi_coll_tuned_allgather_intra_dec_fixed(const void *sbuf, int scount,
alg = 4;
}
} else if (communicator_size < 128) {
if (total_dsize < 64) {
alg = 1;
} else if (total_dsize < 512) {
if (total_dsize < 512) {
alg = 3;
} else if (total_dsize < 65536) {
alg = 5;
@ -1174,9 +1136,7 @@ int ompi_coll_tuned_allgather_intra_dec_fixed(const void *sbuf, int scount,
alg = 4;
}
} else if (communicator_size < 256) {
if (total_dsize < 32) {
alg = 1;
} else if (total_dsize < 512) {
if (total_dsize < 512) {
alg = 3;
} else if (total_dsize < 131072) {
alg = 5;
@ -1188,9 +1148,7 @@ int ompi_coll_tuned_allgather_intra_dec_fixed(const void *sbuf, int scount,
alg = 4;
}
} else if (communicator_size < 512) {
if (total_dsize < 16) {
alg = 1;
} else if (total_dsize < 32) {
if (total_dsize < 32) {
alg = 3;
} else if (total_dsize < 128) {
alg = 2;
@ -1206,9 +1164,7 @@ int ompi_coll_tuned_allgather_intra_dec_fixed(const void *sbuf, int scount,
alg = 4;
}
} else if (communicator_size < 1024) {
if (total_dsize < 4) {
alg = 1;
} else if (total_dsize < 64) {
if (total_dsize < 64) {
alg = 3;
} else if (total_dsize < 256) {
alg = 2;
@ -1218,9 +1174,7 @@ int ompi_coll_tuned_allgather_intra_dec_fixed(const void *sbuf, int scount,
alg = 5;
}
} else if (communicator_size < 2048) {
if (total_dsize < 2) {
alg = 1;
} else if (total_dsize < 4) {
if (total_dsize < 4) {
alg = 3;
} else if (total_dsize < 8) {
alg = 2;

8
ompi/mca/coll/tuned/coll_tuned_exscan_decision.c
Просмотреть файл

@ -2,6 +2,9 @@
/*
* Copyright (c) 2018 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -26,7 +29,7 @@
static int coll_tuned_exscan_forced_algorithm = 0;
/* valid values for coll_tuned_exscan_forced_algorithm */
static mca_base_var_enum_value_t exscan_algorithms[] = {
static const mca_base_var_enum_value_t exscan_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "recursive_doubling"},
@ -68,7 +71,8 @@ int ompi_coll_tuned_exscan_intra_check_forced_init (coll_tuned_force_algorithm_m
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"exscan_algorithm",
"Which exscan algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 recursive_doubling",
"Which exscan algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 recursive_doubling. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_gather_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015-2017 Research Organization for Information Science
@ -32,7 +32,7 @@ static int coll_tuned_gather_tree_fanout;
static int coll_tuned_gather_chain_fanout;
/* valid values for coll_tuned_gather_forced_algorithm */
static mca_base_var_enum_value_t gather_algorithms[] = {
static const mca_base_var_enum_value_t gather_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "binomial"},
@ -75,7 +75,8 @@ ompi_coll_tuned_gather_intra_check_forced_init(coll_tuned_force_algorithm_mca_pa
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"gather_algorithm",
"Which gather algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 binomial, 3 linear with synchronization.",
"Which gather algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 binomial, 3 linear with synchronization. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_reduce_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015-2018 Research Organization for Information Science
@ -33,7 +33,7 @@ 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[] = {
static const mca_base_var_enum_value_t reduce_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "chain"},
@ -80,7 +80,8 @@ int ompi_coll_tuned_reduce_intra_check_forced_init (coll_tuned_force_algorithm_m
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, 7 rabenseifner",
"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, 7 rabenseifner. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

10
ompi/mca/coll/tuned/coll_tuned_reduce_scatter_block_decision.c
Просмотреть файл

@ -4,6 +4,9 @@
* and Information Sciences. All rights reserved.
* Copyright (c) 2018 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -31,7 +34,7 @@ static int coll_tuned_reduce_scatter_block_segment_size = 0;
static int coll_tuned_reduce_scatter_block_tree_fanout;
/* valid values for coll_tuned_reduce_scatter_blokc_forced_algorithm */
static mca_base_var_enum_value_t reduce_scatter_block_algorithms[] = {
static const mca_base_var_enum_value_t reduce_scatter_block_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "recursive_doubling"},
@ -77,7 +80,8 @@ int ompi_coll_tuned_reduce_scatter_block_intra_check_forced_init (coll_tuned_for
"reduce_scatter_block_algorithm",
"Which reduce reduce_scatter_block algorithm is used. "
"Can be locked down to choice of: 0 ignore, 1 basic_linear, 2 recursive_doubling, "
"3 recursive_halving, 4 butterfly",
"3 recursive_halving, 4 butterfly. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,
@ -130,7 +134,7 @@ int ompi_coll_tuned_reduce_scatter_block_intra_do_this(const void *sbuf, void *r
dtype, op, comm, module);
case (3): return ompi_coll_base_reduce_scatter_block_intra_recursivehalving(sbuf, rbuf, rcount,
dtype, op, comm, module);
case (4): return ompi_coll_base_reduce_scatter_block_intra_butterfly(sbuf, rbuf, rcount, dtype, op, comm,
case (4): return ompi_coll_base_reduce_scatter_block_intra_butterfly(sbuf, rbuf, rcount, dtype, op, comm,
module);
} /* switch */
OPAL_OUTPUT((ompi_coll_tuned_stream, "coll:tuned:reduce_scatter_block_intra_do_this attempt to select algorithm %d when only 0-%d is valid?",

7
ompi/mca/coll/tuned/coll_tuned_reduce_scatter_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
@ -33,7 +33,7 @@ static int coll_tuned_reduce_scatter_tree_fanout;
static int coll_tuned_reduce_scatter_chain_fanout;
/* valid values for coll_tuned_reduce_scatter_forced_algorithm */
static mca_base_var_enum_value_t reduce_scatter_algorithms[] = {
static const mca_base_var_enum_value_t reduce_scatter_algorithms[] = {
{0, "ignore"},
{1, "non-overlapping"},
{2, "recursive_halving"},
@ -77,7 +77,8 @@ int ompi_coll_tuned_reduce_scatter_intra_check_forced_init (coll_tuned_force_alg
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"reduce_scatter_algorithm",
"Which reduce reduce_scatter algorithm is used. Can be locked down to choice of: 0 ignore, 1 non-overlapping (Reduce + Scatterv), 2 recursive halving, 3 ring, 4 butterfly",
"Which reduce reduce_scatter algorithm is used. Can be locked down to choice of: 0 ignore, 1 non-overlapping (Reduce + Scatterv), 2 recursive halving, 3 ring, 4 butterfly. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

8
ompi/mca/coll/tuned/coll_tuned_scan_decision.c
Просмотреть файл

@ -2,6 +2,9 @@
/*
* Copyright (c) 2018 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
@ -26,7 +29,7 @@
static int coll_tuned_scan_forced_algorithm = 0;
/* valid values for coll_tuned_scan_forced_algorithm */
static mca_base_var_enum_value_t scan_algorithms[] = {
static const mca_base_var_enum_value_t scan_algorithms[] = {
{0, "ignore"},
{1, "linear"},
{2, "recursive_doubling"},
@ -68,7 +71,8 @@ int ompi_coll_tuned_scan_intra_check_forced_init (coll_tuned_force_algorithm_mca
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"scan_algorithm",
"Which scan algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 recursive_doubling",
"Which scan algorithm is used. Can be locked down to choice of: 0 ignore, 1 linear, 2 recursive_doubling. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,

7
ompi/mca/coll/tuned/coll_tuned_scatter_decision.c
Просмотреть файл

@ -1,6 +1,6 @@
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2017 The University of Tennessee and The University
* Copyright (c) 2004-2020 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
@ -33,7 +33,7 @@ static int coll_tuned_scatter_tree_fanout;
static int coll_tuned_scatter_chain_fanout;
/* valid values for coll_tuned_scatter_forced_algorithm */
static mca_base_var_enum_value_t scatter_algorithms[] = {
static const mca_base_var_enum_value_t scatter_algorithms[] = {
{0, "ignore"},
{1, "basic_linear"},
{2, "binomial"},
@ -76,7 +76,8 @@ ompi_coll_tuned_scatter_intra_check_forced_init(coll_tuned_force_algorithm_mca_p
mca_param_indices->algorithm_param_index =
mca_base_component_var_register(&mca_coll_tuned_component.super.collm_version,
"scatter_algorithm",
"Which scatter algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 binomial, 3 non-blocking linear.",
"Which scatter algorithm is used. Can be locked down to choice of: 0 ignore, 1 basic linear, 2 binomial, 3 non-blocking linear. "
"Only relevant if coll_tuned_use_dynamic_rules is true.",
MCA_BASE_VAR_TYPE_INT, new_enum, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5,
MCA_BASE_VAR_SCOPE_ALL,