Skip to content

Fix some issues with dynamic algorithm selection in coll/tuned #8186

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
rajachan merged 5 commits into from
Nov 12, 2020
Merged

Fix some issues with dynamic algorithm selection in coll/tuned #8186

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
06f605c
coll/tuned: add hint about dynamic rules to mca parameters
devreal Nov 5, 2020
7261255
coll/tuned: Mark global static algorithm as const
devreal Nov 5, 2020
04d198f
coll/tuned: don't select algorithms knowing when it's clear they woul…
devreal Nov 5, 2020
22e289b
coll/tuned: fix minor errors in comments
devreal Nov 5, 2020
a15e5dc
COLL TUNED: remove stray selection of linear algs for alreduce and al…
devreal Nov 11, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 4 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_allgather_decision.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -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
Expand Down Expand Up @@ -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"},
Expand Down Expand Up @@ -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,
Expand Down
7 changes: 4 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_allgatherv_decision.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -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
Expand Down Expand Up @@ -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"},
Expand Down Expand Up @@ -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,
Expand Down
7 changes: 4 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_allreduce_decision.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -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
Expand Down Expand Up @@ -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"},
Expand Down Expand Up @@ -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,
Expand Down
7 changes: 4 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_alltoall_decision.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -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
Expand Down Expand Up @@ -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"},
Expand Down Expand Up @@ -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,
Expand Down
7 changes: 4 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_alltoallv_decision.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -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
Expand Down Expand Up @@ -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"},
Expand Down Expand Up @@ -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,
Expand Down
7 changes: 4 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_barrier_decision.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -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.
* $COPYRIGHT$
Expand All @@ -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"},
Expand Down Expand Up @@ -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,
Expand Down
7 changes: 4 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_bcast_decision.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -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
Expand Down Expand Up @@ -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"},
Expand Down Expand Up @@ -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,
Expand Down
6 changes: 3 additions & 3 deletions ompi/mca/coll/tuned/coll_tuned_decision_dynamic.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -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,
Expand Down Expand Up @@ -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,
Expand Down Expand Up @@ -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)
*/
Expand Down
70 changes: 12 additions & 58 deletions ompi/mca/coll/tuned/coll_tuned_decision_fixed.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -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,
Expand Down Expand Up @@ -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;
Expand Down Expand Up @@ -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;
Expand Down Expand Up @@ -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;
Expand All @@ -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;
Expand Down Expand Up @@ -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) {
Expand Down Expand Up @@ -1164,19 +1128,15 @@ 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;
} else {
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;
Expand All @@ -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;
Expand All @@ -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;
Expand All @@ -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;
Expand Down
Loading