550 linhas
17 KiB
Plaintext
550 linhas
17 KiB
Plaintext
##
|
|
## Copyright (C) by Argonne National Laboratory
|
|
## See COPYRIGHT in top-level directory
|
|
##
|
|
|
|
# This config file is used by maint/gen_coll_cvar.py to generate
|
|
# coll/testlist.collalgo, which include tests that test specific collective
|
|
# algorithms controlled by using CVARs.
|
|
|
|
# The basic format are indentation-based hierarchy. There are 2 top level
|
|
# groups -- "tests" and "algorithms".
|
|
#
|
|
# "tests" lists tests grouped by collective operation then by intra-blocking,
|
|
# intra-nonblocking, inter-blocking, inter-nonblocking, and persistent. The
|
|
# leaf-node specifies tests in the same format used in testlist.
|
|
#
|
|
# "algorithms" lists algorithm names grouped by hierarchy manner. Each
|
|
# algorithm name, e.g. binomial under bcast intra-blocking, refers to
|
|
# MPIR_CVAR_BCAST_INTRA_ALGORITHM=binomial
|
|
# Device layer algorithm are prefixed, e.g. posix:release_gather, refers to
|
|
# MPIR_CVAR_BCAST_POSIX_INTRA_ALGORITHM=release_gather
|
|
# Each algorithm can also have a list of additional cvar variables with a list
|
|
# of values separated by comma. For example,
|
|
# .MPIR_CVAR_BCAST_INTRANODE_NUM_CELLS=2,4
|
|
# .MPIR_CVAR_BCAST_INTRANODE_TREE_KVAL=8,64
|
|
# will generate 4 tests setting each CVAR respectively. Note the leading dot
|
|
# part of the format, not part of the CVAR name.
|
|
#
|
|
# Each algorithm with each additional cvar choice will generate a test against
|
|
# each test listed in the corresponding group. Additionally, each nonblocking
|
|
# algorithm will be tested against each blocking test using the nb algorithm.
|
|
# Each nonblocking algorithm will be tested against each persistent test as
|
|
# well.
|
|
|
|
-------------------------------------------------
|
|
tests:
|
|
allgather:
|
|
intra-blocking:
|
|
allgather2 10
|
|
allgather3 10
|
|
persistent:
|
|
p_allgather 10
|
|
allgatherv:
|
|
intra-blocking:
|
|
allgatherv2 10
|
|
allgatherv3 10
|
|
allgatherv4 4 timeLimit=600
|
|
persistent:
|
|
p_allgatherv 10
|
|
allreduce:
|
|
intra-blocking:
|
|
allred 4 arg=-count=100
|
|
allred 7 arg=-count=10
|
|
allredmany 4
|
|
allred2 4
|
|
allred3 10
|
|
allred5 5
|
|
allred6 4
|
|
allred6 7
|
|
allred_float 10
|
|
persistent:
|
|
p_allred 7
|
|
alltoall:
|
|
intra-blocking:
|
|
alltoall1 8
|
|
persistent:
|
|
p_alltoall 8
|
|
alltoallv:
|
|
intra-blocking:
|
|
alltoallv 8
|
|
alltoallv0 10
|
|
persistent:
|
|
p_alltoallv 10
|
|
alltoallw:
|
|
intra-blocking:
|
|
alltoallw1 10
|
|
alltoallw2 10
|
|
alltoallw_zeros 1
|
|
alltoallw_zeros 2
|
|
alltoallw_zeros 5
|
|
alltoallw_zeros 8
|
|
persistent:
|
|
p_alltoallw 10
|
|
bcast:
|
|
intra-blocking:
|
|
bcasttest 10
|
|
bcastzerotype 5
|
|
persistent:
|
|
p_bcast 4
|
|
p_bcast2 8
|
|
gather:
|
|
intra-blocking:
|
|
gather 4
|
|
gather2 4
|
|
persistent:
|
|
p_gather 4
|
|
gatherv:
|
|
intra-blocking:
|
|
gatherv 5
|
|
persistent:
|
|
p_gatherv 5
|
|
neighbor_allgather:
|
|
intra-blocking:
|
|
neighb_allgather 4
|
|
persistent:
|
|
p_neighb_allgather 4
|
|
neighbor_allgatherv:
|
|
intra-blocking:
|
|
neighb_allgatherv 4
|
|
persistent:
|
|
p_neighb_allgatherv 4
|
|
neighbor_alltoall:
|
|
intra-blocking:
|
|
neighb_alltoall 4
|
|
persistent:
|
|
p_neighb_alltoall 4
|
|
neighbor_alltoallv:
|
|
intra-blocking:
|
|
neighb_alltoallv 4
|
|
persistent:
|
|
p_neighb_alltoallv 4
|
|
neighbor_alltoallw:
|
|
intra-blocking:
|
|
neighb_alltoallw 4
|
|
persistent:
|
|
p_neighb_alltoallw 4
|
|
reduce:
|
|
intra-blocking:
|
|
reduce 5
|
|
reduce 10
|
|
red3 10
|
|
red4 10
|
|
persistent:
|
|
p_red 5
|
|
reduce_scatter:
|
|
intra-blocking:
|
|
redscat 4
|
|
redscat 6
|
|
redscat2 4
|
|
redscat2 5
|
|
redscat2 10
|
|
redscat3 8
|
|
persistent:
|
|
p_redscat 4
|
|
reduce_scatter_block:
|
|
intra-blocking:
|
|
red_scat_block 4
|
|
red_scat_block 5
|
|
red_scat_block 8
|
|
red_scat_block2 4
|
|
red_scat_block2 5
|
|
red_scat_block2 10
|
|
redscatblk3 8
|
|
redscatblk3 10
|
|
persistent:
|
|
p_red_scat_block 4
|
|
scan:
|
|
intra-blocking:
|
|
scantst 4
|
|
persistent:
|
|
p_scan 4
|
|
scatter:
|
|
intra-blocking:
|
|
scattern 4
|
|
scatter2 4
|
|
scatter3 4
|
|
persistent:
|
|
p_scatter 4
|
|
scatterv:
|
|
intra-blocking:
|
|
scatterv 4
|
|
persistent:
|
|
p_scatterv 4
|
|
|
|
--------------------------------------------------
|
|
algorithms:
|
|
allgather:
|
|
inter-blocking:
|
|
local_gather_remote_bcast
|
|
inter-nonblocking:
|
|
sched_local_gather_remote_bcast
|
|
intra-blocking:
|
|
composition:1
|
|
composition:2
|
|
brucks
|
|
k_brucks
|
|
.MPIR_CVAR_ALLGATHER_BRUCKS_KVAL=2,3,4
|
|
recursive_doubling
|
|
ring
|
|
recexch_doubling
|
|
.MPIR_CVAR_ALLGATHER_RECEXCH_KVAL=2,3,4
|
|
.MPIR_CVAR_ALLGATHER_RECEXCH_SINGLE_PHASE_RECV=0,1
|
|
recexch_halving
|
|
.MPIR_CVAR_ALLGATHER_RECEXCH_KVAL=2,3,4
|
|
.MPIR_CVAR_ALLGATHER_RECEXCH_SINGLE_PHASE_RECV=0,1
|
|
intra-nonblocking:
|
|
sched_ring
|
|
sched_brucks
|
|
sched_recursive_doubling
|
|
tsp_ring
|
|
tsp_brucks
|
|
.MPIR_CVAR_IALLGATHER_BRUCKS_KVAL=2,3,4
|
|
tsp_recexch_doubling
|
|
.MPIR_CVAR_IALLGATHER_RECEXCH_KVAL=2,3,4
|
|
tsp_recexch_halving
|
|
.MPIR_CVAR_IALLGATHER_RECEXCH_KVAL=2,3,4
|
|
allgatherv:
|
|
inter-blocking:
|
|
remote_gather_local_bcast
|
|
inter-nonblocking:
|
|
sched_remote_gather_local_bcast
|
|
intra-blocking:
|
|
brucks
|
|
recursive_doubling
|
|
ring
|
|
intra-nonblocking:
|
|
sched_brucks
|
|
sched_recursive_doubling
|
|
sched_ring
|
|
tsp_recexch_doubling
|
|
.MPIR_CVAR_IALLGATHERV_RECEXCH_KVAL=2,3,4
|
|
tsp_recexch_halving
|
|
.MPIR_CVAR_IALLGATHERV_RECEXCH_KVAL=2,3,4
|
|
tsp_ring
|
|
tsp_brucks
|
|
.MPIR_CVAR_IALLGATHERV_BRUCKS_KVAL=2,3,4
|
|
allreduce:
|
|
inter-blocking:
|
|
reduce_exchange_bcast
|
|
inter-nonblocking:
|
|
sched_remote_reduce_local_bcast
|
|
intra-blocking:
|
|
composition:1
|
|
composition:2
|
|
composition:3
|
|
composition:4
|
|
smp
|
|
recursive_doubling
|
|
reduce_scatter_allgather
|
|
tree
|
|
.MPIR_CVAR_ALLREDUCE_TREE_TYPE=knomial_1,knomial_2,kary
|
|
.MPIR_CVAR_ALLREDUCE_TREE_KVAL=2,3,4
|
|
.MPIR_CVAR_ALLREDUCE_TREE_PIPELINE_CHUNK_SIZE=4096,8192
|
|
recexch
|
|
.MPIR_CVAR_ALLREDUCE_RECEXCH_KVAL=2,3,4,5,9
|
|
.MPIR_CVAR_ALLREDUCE_RECEXCH_SINGLE_PHASE_RECV=0,1
|
|
ring
|
|
k_reduce_scatter_allgather
|
|
.MPIR_CVAR_ALLREDUCE_RECEXCH_KVAL=2,3,4,5,9
|
|
.MPIR_CVAR_ALLREDUCE_RECEXCH_SINGLE_PHASE_RECV=0,1
|
|
posix:release_gather
|
|
.MPIR_CVAR_COLL_SHM_LIMIT_PER_NODE=131072
|
|
.MPIR_CVAR_REDUCE_INTRANODE_BUFFER_TOTAL_SIZE=16384
|
|
.MPIR_CVAR_REDUCE_INTRANODE_TREE_KVAL=4,8
|
|
intra-nonblocking:
|
|
sched_naive
|
|
sched_smp
|
|
sched_recursive_doubling
|
|
sched_reduce_scatter_allgather
|
|
tsp_recexch_single_buffer
|
|
tsp_recexch_multiple_buffer
|
|
tsp_tree
|
|
.MPIR_CVAR_IALLREDUCE_TREE_TYPE=kary,knomial_1,knomial_2
|
|
.MPIR_CVAR_IALLREDUCE_TREE_KVAL=2,3,4
|
|
.MPIR_CVAR_IALLREDUCE_TREE_PIPELINE_CHUNK_SIZE=4096
|
|
tsp_ring
|
|
tsp_recexch_reduce_scatter_recexch_allgatherv
|
|
.MPIR_CVAR_IALLREDUCE_RECEXCH_KVAL=2,3,4
|
|
alltoall:
|
|
inter-blocking:
|
|
pairwise_exchange
|
|
inter-nonblocking:
|
|
sched_pairwise_exchange
|
|
intra-blocking:
|
|
composition:1
|
|
composition:2
|
|
brucks
|
|
k_brucks
|
|
.MPIR_CVAR_ALLTOALL_BRUCKS_KVAL=2,3,4
|
|
pairwise
|
|
pairwise_sendrecv_replace
|
|
scattered
|
|
intra-nonblocking:
|
|
sched_brucks
|
|
sched_inplace
|
|
sched_pairwise
|
|
sched_permuted_sendrecv
|
|
tsp_ring
|
|
tsp_brucks
|
|
.MPIR_CVAR_IALLTOALL_BRUCKS_KVAL=2,3,4
|
|
.MPIR_CVAR_IALLTOALL_BRUCKS_BUFFER_PER_NBR=0,1
|
|
tsp_scattered
|
|
.MPIR_CVAR_IALLTOALL_SCATTERED_BATCH_SIZE=1,2,4
|
|
.MPIR_CVAR_IALLTOALL_SCATTERED_OUTSTANDING_TASKS=4,8
|
|
alltoallv:
|
|
inter-blocking:
|
|
pairwise_exchange
|
|
inter-nonblocking:
|
|
sched_pairwise_exchange
|
|
intra-blocking:
|
|
pairwise_sendrecv_replace
|
|
scattered
|
|
intra-nonblocking:
|
|
sched_blocked
|
|
sched_inplace
|
|
tsp_scattered
|
|
.MPIR_CVAR_IALLTOALLV_SCATTERED_BATCH_SIZE=1,2,4
|
|
.MPIR_CVAR_IALLTOALLV_SCATTERED_OUTSTANDING_TASKS=4,8
|
|
tsp_blocked
|
|
tsp_inplace
|
|
alltoallw:
|
|
inter-blocking:
|
|
pairwise_exchange
|
|
inter-nonblocking:
|
|
sched_pairwise_exchange
|
|
intra-blocking:
|
|
pairwise_sendrecv_replace
|
|
scattered
|
|
intra-nonblocking:
|
|
sched_blocked
|
|
sched_inplace
|
|
tsp_blocked
|
|
tsp_inplace
|
|
barrier:
|
|
inter-blocking:
|
|
bcast
|
|
inter-nonblocking:
|
|
sched_bcast
|
|
intra-blocking:
|
|
composition:1
|
|
composition:2
|
|
k_dissemination
|
|
.MPIR_CVAR_BARRIER_DISSEM_KVAL=2,3,4,8
|
|
recexch
|
|
.MPIR_CVAR_BARRIER_RECEXCH_KVAL=2,3,4,8
|
|
.MPIR_CVAR_BARRIER_RECEXCH_SINGLE_PHASE_RECV=0,1
|
|
intra-nonblocking:
|
|
sched_recursive_doubling
|
|
tsp_recexch
|
|
.MPIR_CVAR_IBARRIER_RECEXCH_KVAL=2,3,4,8
|
|
tsp_k_dissemination
|
|
.MPIR_CVAR_IBARRIER_DISSEM_KVAL=2,3,4,8
|
|
bcast:
|
|
inter-blocking:
|
|
remote_send_local_bcast
|
|
inter-nonblocking:
|
|
sched_flat
|
|
intra-blocking:
|
|
composition:1
|
|
composition:2
|
|
composition:3
|
|
binomial
|
|
smp
|
|
scatter_recursive_doubling_allgather
|
|
scatter_ring_allgather
|
|
tree
|
|
.MPIR_CVAR_BCAST_TREE_TYPE=kary,knomial_1,knomial_2
|
|
.MPIR_CVAR_BCAST_IS_NON_BLOCKING=1
|
|
pipelined_tree
|
|
.MPIR_CVAR_BCAST_TREE_TYPE=kary,knomial_1,knomial_2
|
|
.MPIR_CVAR_BCAST_IS_NON_BLOCKING=1
|
|
.MPIR_CVAR_BCAST_TREE_PIPELINE_CHUNK_SIZE=4096,8192
|
|
posix:release_gather
|
|
.MPIR_CVAR_COLL_SHM_LIMIT_PER_NODE=131072
|
|
.MPIR_CVAR_BCAST_INTRANODE_BUFFER_TOTAL_SIZE=16384,32768
|
|
.MPIR_CVAR_BCAST_INTRANODE_NUM_CELLS=2,4
|
|
.MPIR_CVAR_BCAST_INTRANODE_TREE_KVAL=8,64
|
|
.MPIR_CVAR_BCAST_INTRANODE_TREE_TYPE=knomial_1,knomial_2
|
|
ofi:trigger_tree_tagged
|
|
.MPIR_CVAR_CH4_OFI_ENABLE_TRIGGERED=1
|
|
.FI_PROVIDER=sockets
|
|
.MPIR_CVAR_BCAST_TREE_TYPE=kary,knomial_1,knomial_2
|
|
.MPIR_CVAR_BCAST_TREE_KVAL=2,3,4,8
|
|
ofi:trigger_tree_rma
|
|
.MPIR_CVAR_CH4_OFI_ENABLE_TRIGGERED=1
|
|
.FI_PROVIDER=sockets
|
|
.MPIR_CVAR_BCAST_TREE_TYPE=kary,knomial_1,knomial_2
|
|
.MPIR_CVAR_BCAST_TREE_KVAL=2,3,4,8
|
|
intra-nonblocking:
|
|
sched_binomial
|
|
sched_smp
|
|
sched_scatter_recursive_doubling_allgather
|
|
sched_scatter_ring_allgather
|
|
tsp_tree
|
|
.MPIR_CVAR_IBCAST_TREE_TYPE=kary,knomial_1,knomial_2
|
|
.MPIR_CVAR_IBCAST_TREE_KVAL=3
|
|
.MPIR_CVAR_IBCAST_TREE_PIPELINE_CHUNK_SIZE=4096
|
|
tsp_scatterv_recexch_allgatherv
|
|
.MPIR_CVAR_IBCAST_SCATTERV_KVAL=3
|
|
.MPIR_CVAR_IBCAST_ALLGATHER_RECEXCH=3
|
|
tsp_scatterv_ring_allgatherv
|
|
.MPIR_CVAR_IBCAST_SCATTERV_KVAL=3
|
|
tsp_ring
|
|
.MPIR_CVAR_IBCAST_RING_CHUNK_SIZE=4096
|
|
exscan:
|
|
intra-blocking:
|
|
recursive_doubling
|
|
intra-nonblocking:
|
|
sched_recursive_doubling
|
|
gather:
|
|
inter-blocking:
|
|
linear
|
|
local_gather_remote_send
|
|
inter-nonblocking:
|
|
sched_long
|
|
sched_short
|
|
intra-blocking:
|
|
binomial
|
|
intra-nonblocking:
|
|
sched_binomial
|
|
tsp_tree
|
|
.MPIR_CVAR_IGATHER_TREE_KVAL=2,3,4
|
|
gatherv:
|
|
inter-blocking:
|
|
linear
|
|
inter-nonblocking:
|
|
sched_linear
|
|
intra-blocking:
|
|
linear
|
|
intra-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
neighbor_allgather:
|
|
inter-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
intra-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
neighbor_allgatherv:
|
|
inter-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
intra-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
neighbor_alltoall:
|
|
inter-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
intra-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
neighbor_alltoallv:
|
|
inter-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
intra-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
neighbor_alltoallw:
|
|
inter-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
intra-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|
|
reduce:
|
|
inter-blocking:
|
|
local_reduce_remote_send
|
|
inter-nonblocking:
|
|
sched_local_reduce_remote_send
|
|
intra-blocking:
|
|
composition:1
|
|
composition:2
|
|
composition:3
|
|
binomial
|
|
smp
|
|
reduce_scatter_gather
|
|
posix:release_gather
|
|
.MPIR_CVAR_COLL_SHM_LIMIT_PER_NODE=131072
|
|
.MPIR_CVAR_REDUCE_INTRANODE_BUFFER_TOTAL_SIZE=16384,32768
|
|
.MPIR_CVAR_REDUCE_INTRANODE_NUM_CELLS=2,4
|
|
.MPIR_CVAR_REDUCE_INTRANODE_TREE_KVAL=4,8
|
|
.MPIR_CVAR_REDUCE_INTRANODE_TREE_TYPE=knomial_1,knomial_2
|
|
intra-nonblocking:
|
|
sched_smp
|
|
sched_binomial
|
|
sched_reduce_scatter_gather
|
|
tsp_tree
|
|
.MPIR_CVAR_IREDUCE_TREE_TYPE=kary,knomial_1,knomial_2
|
|
.MPIR_CVAR_IREDUCE_TREE_KVAL=3
|
|
.MPIR_CVAR_IREDUCE_TREE_PIPELINE_CHUNK_SIZE=4096
|
|
tsp_ring
|
|
.MPIR_CVAR_IREDUCE_RING_CHUNK_SIZE=4096
|
|
reduce_scatter:
|
|
inter-blocking:
|
|
remote_reduce_local_scatter
|
|
inter-nonblocking:
|
|
sched_remote_reduce_local_scatterv
|
|
intra-blocking:
|
|
pairwise
|
|
recursive_halving
|
|
recursive_doubling
|
|
intra-nonblocking:
|
|
sched_noncommutative
|
|
sched_pairwise
|
|
sched_recursive_halving
|
|
sched_recursive_doubling
|
|
tsp_recexch
|
|
.MPIR_CVAR_IREDUCE_SCATTER_RECEXCH_KVAL=2,3,4
|
|
reduce_scatter_block:
|
|
inter-blocking:
|
|
remote_reduce_local_scatter
|
|
inter-nonblocking:
|
|
sched_remote_reduce_local_scatterv
|
|
intra-blocking:
|
|
noncommutative
|
|
pairwise
|
|
recursive_halving
|
|
recursive_doubling
|
|
intra-nonblocking:
|
|
sched_noncommutative
|
|
sched_pairwise
|
|
sched_recursive_halving
|
|
sched_recursive_doubling
|
|
tsp_recexch
|
|
.MPIR_CVAR_IREDUCE_SCATTER_BLOCK_RECEXCH_KVAL=2,3,4
|
|
scan:
|
|
intra-blocking:
|
|
smp
|
|
recursive_doubling
|
|
intra-nonblocking:
|
|
sched_smp
|
|
sched_recursive_doubling
|
|
tsp_recursive_doubling
|
|
scatter:
|
|
inter-blocking:
|
|
linear
|
|
remote_send_local_scatter
|
|
inter-nonblocking:
|
|
sched_linear
|
|
sched_remote_send_local_scatter
|
|
intra-blocking:
|
|
binomial
|
|
intra-nonblocking:
|
|
sched_binomial
|
|
tsp_tree
|
|
.MPIR_CVAR_ISCATTER_TREE_KVAL=2,3,4
|
|
scatterv:
|
|
inter-blocking:
|
|
linear
|
|
inter-nonblocking:
|
|
sched_linear
|
|
intra-blocking:
|
|
linear
|
|
intra-nonblocking:
|
|
sched_linear
|
|
tsp_linear
|