Skip to content
Light-weight GPU kernel interface for graph operations
CMake Cuda C++ Python Shell
Branch: master
Clone or download
Latest commit 9f5e381 Jun 13, 2019

MiniGun: Light-weight GPU kernel interface for graph operations

The project is heavily influenced by the Gunrock project. The goal is to define a general kernel library that can cover a wide range of graph operations used in DGL.

Current roadmap:

  • Port basic advance operator (all edge policy).
  • Port advance operator with dynamic frontiers.
  • Port more advance kernel policies
    • TWC
    • Inbound edge partition
    • outbound edge partition
  • CPU implementation for advance
  • Integration with DGL
  • Filter


Require CUDA.

mkdir build
cmake ..
make -j4


See codes in samples. Currently, there are four:

  • spmv
  • spmm
  • masked_mm
  • edge_softmax

You could run sample by (in the build folder):


Disclaimer: Minigun project is not related to the "mini-gunrock" project.

You can’t perform that action at this time.