openucx / ucc

Unified Collective Communication Library
https://openucx.github.io/ucc/
BSD 3-Clause "New" or "Revised" License
177 stars 85 forks source link

TL/UCP: Allgather Sparbit algorithm #940

Closed ikryukov closed 2 months ago

ikryukov commented 3 months ago

What

Implementation on Sparbit algorithm from https://arxiv.org/pdf/2109.08751.pdf

Why ?

How ?

inspired by https://github.com/open-mpi/ompi/blob/main/ompi/mca/coll/base/coll_base_allgather.c implementation

Launch string

mpirun -x UCC_TL_UCP_TUNE=allgather:0-inf:@4 --mca coll ^hcoll --mca coll_ucc_enable 0 -x LD_LIBRARY_PATH=/home/ikryukov/work/ucc/install/lib:/home/ikryukov/work/ucx/install/lib:$LD_LIBRARY_PATH -x UCC_CLS=basic -x UCC_TLS=ucp -x UCC_CL_BASIC_TLS=ucp -x UCC_LOG_LEVEL=info -np 16 ./install/bin/ucc_test_mpi -c allgather --teams world

swx-jenkins3 commented 3 months ago

Can one of the admins verify this patch?

Sergei-Lebedev commented 3 months ago

ok to test