| commit | 0e78a596101c41956e2129031cb40232600ff74b | [log] [tgz] |
|---|---|---|
| author | Andrew Dye <andrewdye@fb.com> | Mon Feb 27 14:00:54 2017 -0800 |
| committer | Pieter Noordhuis <pietern@devgpu244.prn2.facebook.com> | Thu Mar 02 10:16:38 2017 -0800 |
| tree | 48b399442fa5952582e1f019ec9a142542d8daf5 | |
| parent | 2d4d3b18ddcb6fb70fc4216f50e86eb3d33339e2 [diff] |
add mutex getter/setter to synchronize CUDA and NCCL ops Summary: Allow gloo consumers to assign a mutex to synchronize CUDA malloc/free and NCCL operations. Reviewed By: pietern Differential Revision: D4622135 fbshipit-source-id: 60acd7c01a677a0df5415fe38e6ef5a2e7c8606a
Gloo is a collective communications library. It comes with a number of collective algorithms useful for machine learning applications. These include a barrier, broadcast, and allreduce.
Transport of data between participating machines is abstracted so that IP can be used at all times, or InifiniBand (or RoCE) when available.
Where applicable, algorithms have an implementation that works with system memory buffers, and one that works with NVIDIA GPU memory buffers. In the latter case, if the InfiniBand transport is used, GPUDirect can be used to accelerate cross machine GPU-to-GPU memory transfers.
Gloo is built to run on Linux and has no hard dependencies other than libc.
Optional dependencies are:
You can build Gloo using CMake.
Since it is a library, it is most convenient to vendor it in your own project and include the project root in your own CMake configuration.
For standalone builds (e.g. to run tests or benchmarks), first populate the third-party directory with a few dependencies to compile both the tests and the benchmark tool:
cd third-party ./fetch.sh
Then, to build:
mkdir build cd build cmake ../ -DBUILD_TEST=1 -DBUILD_BENCHMARK=1 ls -l gloo/gloo_{test,benchmark}
Please refer to docs/ for detailed documentation.
Gloo is BSD-licensed. We also provide an additional patent grant.