Add support for inter-node communication using sockets and InfiniBand/RoCE. Improve latency. Add support for aggregation. Improve LL/regular tuning. Remove tests as those are now at github.com/nvidia/nccl-tests .
25 lines
652 B
Bash
25 lines
652 B
Bash
#!/bin/bash
|
|
#
|
|
# Copyright (c) 2017-2018, NVIDIA CORPORATION. All rights reserved.
|
|
#
|
|
# See LICENSE.txt for license information
|
|
#
|
|
|
|
# To run from $BUILDDIR/
|
|
|
|
BUILDDIR=`basename $PWD`
|
|
|
|
cd ..
|
|
NCCL_MAJOR=${nccl:Major}
|
|
NCCL_MINOR=${nccl:Minor}
|
|
NCCL_PATCH=${nccl:Patch}
|
|
NCCL_SUFFIX=${nccl:Suffix}
|
|
CUDA_MAJOR=${cuda:Major}
|
|
CUDA_MINOR=${cuda:Minor}
|
|
PKG_REVISION=${pkg:Revision}
|
|
PKG_ARCH=${pkg:Arch}
|
|
|
|
NCCLNAME="nccl_${NCCL_MAJOR}.${NCCL_MINOR}.${NCCL_PATCH}${NCCL_SUFFIX}-${PKG_REVISION}+cuda${CUDA_MAJOR}.${CUDA_MINOR}_${PKG_ARCH}"
|
|
|
|
tar --transform "s/^$BUILDDIR/$NCCLNAME/" -Jcf $NCCLNAME.txz --owner=0 --group=0 $BUILDDIR/include $BUILDDIR/lib $BUILDDIR/*.txt
|