rocm-systems/projects/rccl/docs/source/library.rst

.. toctree::
   :maxdepth: 4
   :caption: Contents:

======
RCCL
======

Introduction
------------

The ROCm Collective Communication Library (RCCL) is a stand-alone library which provides multi-GPU and multi-node collective communication primitives optimized for AMD GPUs and can be easily integrated into applications. 
RCCL (pronounced “Rickel”) implements routines such as all-reduce, all-gather, reduce, broadcast, reduce-scatter, gather, scatter, all-to-allv, and all-to-all as well as direct point-to-point (GPU-to-GPU) send and receive operations. The provided collective communication routines are implemented using Ring and Tree algorithms and are optimized to achieve high bandwidth and low latency over PCIe and xGMI high-speed interconnects for intra-node communication and over InfiniBand, RoCE, and TCP/IP for inter-node communication. 
RCCL supports an arbitrary number of GPUs installed in a single-node or multi-node platform and can be used in either single- or multi-process (e.g., MPI) applications.