Skip to content

Latest commit

 

History

History
95 lines (77 loc) · 5.38 KB

README.md

File metadata and controls

95 lines (77 loc) · 5.38 KB

About

FlagCX is a scalable and adaptive cross-chip communication library developed with the backing of the Beijing Academy of Artificial Intelligence (BAAI).

FlagCX is also a part of FlagAI-Open, an open-source initiative by BAAI that aims to foster an open-source ecosystem for AI technologies. It serves as a platform where developers, researchers, and AI enthusiasts can collaborate on various AI projects, contribute to the development of cutting-edge AI solutions, and share their work with the global community.

FlagCX leverages native collective communications libraries to provide the full support of single-chip communications on different platforms. In addition to its native x-CCL support, FlagCX provides an original device-buffer RDMA design to offer advanced support for cross-chip high-performance sendrecev operations (CORE module), which can also be integrated with native x-CCL backends to enable optimized cross-chip collective communications. A comprehensive list of currently supported communication backends and their different capabilities are listed as follows:

Backend NCCL IXCCL CNCL GLOO CORE+x-CCL
Mode Homo Homo Homo Hetero Hetero
send ✓ ✓ ✓ ✓ ✓
recv ✓ ✓ ✓ ✓ ✓
broadcast ✓ ✓ ✓ ✘ ✓
gather ✓ ✓ ✓ ✘ ✓
scatter ✓ ✓ ✓ ✘ ✓
reduce ✓ ✓ ✓ ✘ ✓
allreduce ✓ ✓ ✓ ✓ ✓
allgather ✓ ✓ ✓ ✓ ✓
reducescatter ✓ ✓ ✓ ✘ ✓
alltoall ✓ ✓ ✓ ✓ ✓
group ops ✓ ✓ ✓ ? ?

Note that Homo and Hetero modes refer to communications among homogeneous and heterogeneous clusters. All supported native collective communication libraries can be referenced through the links below:

  • NCCL, NVIDIA Collective Communications Library.
  • IXCCL, Iluvatar Corex Collective Communications Library.
  • CNCL, Cambricon Communications Library.
  • GLOO, Gloo Collective Communications Library.

FlagCX also develops plugins to integrate with upper-layer applications such as PyTorch based on its unified APIs. The table below presents the communication operations currently supported by the plugins of their corresponding frameworks, where the batch_XXX and XXX_coalesced ops refer to the usage of group primitives.

Plugin PyTorch
send ✓
recv ✓
batch_isend_irecv ✓
broadcast ✓
all_reduce ✓
all_reduce_coalesced ✘
reduce ✓
all_gather ✓
all_gather_into_tensor_coalesced ✘
gather ✓
scatter ✓
reduce_scatter ✓
reduce_scatter_tensor_coalesced ✘
all_to_all ✓

Quick Start

Build

  1. Clone the repository:

    git clone https://github.com/FlagOpen/FlagCX.git

  2. Build the library with different flags targeting to different platforms:

    cd FlagCX
make [USE_NVIDIA/USE_ILUVATAR_COREX/USE_CAMBRICON/USE_GLOO]=1

    The default install path is set to build/, you can manually set BUILDDIR to specify the build path. You may also define DEVICE_HOME and CCL_HOME to indicate the install paths of device runtime and communication libraries.

Tests

Tests for FlagCX are maintained in test/perf.

cd test/perf
make [USE_NVIDIA/USE_ILUVATAR_COREX/USE_CAMBRICON]=1
./test_allreduce -b 128M -e 8G -f 2

Note that the default MPI install path is set to /usr/local/mpi, you may specify the MPI path with:

make MPI_HOME=<path to mpi install>

All tests support the same set of arguments:

  • Sizes to scan
    • -b <min size in bytes> minimum size to start with. Default: 1M.
    • -e <max size in bytes> maximum size to end at. Default: 1G.
    • -f <increment factor> multiplication factor between sizes. Default: 2.
  • Performance
    • -w, <warmup iteration count> number of warmup iterations (not timed). Default: 5.
    • -n, <iteration count> number of iterations. Default: 20.
  • Utils
    • -p, <0/1> print buffer info. Default: 0.
    • -h print help message. Default: disabled.

License

This project is licensed under the Apache License (Version 2.0).