RCCL (pronounced "Rickle") is a stand-alone library of standard collective communication routines for GPUs, implementing all-reduce, all-gather, reduce, broadcast, reduce-scatter, gather, scatter, and all-to-all. There is also initial support for direct GPU-to-GPU send and receive operations.
... part of T2, get it here
URL: https://github.com/ROCm/rccl
Author: Advanced Micro Devices, Inc.
Maintainer: The T2 Project <t2 [at] t2-project [dot] org>
License: MIT
Status: Stable
Version: 6.3.3
Remark: Does cross compile (as setup and patched in T2).
Download: https://github.com/ROCm/rccl/ rccl-rocm-6.3.3.tar.gz
T2 source: rccl.cache
T2 source: rccl.desc
Build time (on reference hardware): 980% (relative to binutils)2
Installed size (on reference hardware): 83.21 MB, 39 files
Dependencies (build time detected): bash cmake coreutils diffutils gawk grep gzip hip-rocclr hipcc hipify linux-header make perl python rocm-cmake rocm-core rocm-device-libs rocm-llvm rocm_smi_lib rocprofiler-register rocr-runtime sed sysfiles tar tbb
Installed files (on reference hardware): n.a.
1) This page was automatically generated from the T2 package source. Corrections, such as dead links, URL changes or typos need to be performed directly on that source.
2) Compatible with Linux From Scratch's "Standard Build Unit" (SBU).