GNN-Acceleration-Language

Directory structure

project-root/
├── codegen/ # output path for GALA code generation
├── Data/ # path where datasets are stores
├── docs/ # Has the artifact documentation
├── scripts/ # various scripts for setup and reproduce results
│   └── Data # download datasets
│       └── get_all_datasets.py # Main script that downloads datasets
│       └── other auxiliary scripts
│   └── e2e # end-2-end bash scripts to generate results and setup
│       └── setup # setup scripts
│       └── scripts for figures and tables (f<number>.sh or t<number>.sh)
│       └── simple-test.sh # script for a simple test of GALA
│   └── Environments # scripts/code necessary for the setup of baselines
│   └── Evaluations # scripts to reproduce Figures and Tables 
│   # NOTE: These require multiple runs (to get results, then create figure) with different input parameters
│       └── Figure-<number>.py # script to generate results and figure
│       └── Table-<number>.py # script to generate results and table
│       └── WiseGraph.py # script to specifically generaet WiseGraph results
├── src/ # Source code for GALA
├── tests/ # Baseline test code and GALA DSL and Compiler for testing
│   └── Baselines # Code for baselines
│   └── GALA-DSL # GALA front-end code for testing
│       └── ablations
│           └── input-optimize # DSL for input optimizer tests
│           └── memory-consumption  # DSL for memory consumption tests
│           └── sampling # DSL for sampling tests
│           └── scalability # DSL for scalability tests
│           └── speedups # DSL for other speedup breakdown tests
│       └── gat # DSL for the GAT model
│       └── gcn # DSL for the GCN model
│       └── gin # DSL for the GIN model
│       └── sage # DSL for the SAGE model
│   └── *.cpp # GALA Compiler executables
├── utils/ # Utils used in GALA
└── README.md # readme

Prerequisites

• Cuda Toolkit
• Conda
• g++ >= 11.4
• Cmake >= 3.25

Installation (Linux/Ubuntu)

cd scripts/e2e/setup/
./gala.sh
./data.sh # the minimal necessary to run GALA (there are other scripts for baselines in the same directory)

Quick test for GALA

cd scripts/e2e/
./simple-test.sh
# OR
cd build
tests/gala_inference <path to GALA DSL code> <output path> # run the codegen 
cd <output path>
mkdir build
cd build
cmake -DCMAKE_PREFIX_PATH=<path to libtorch> ..
make -j6
./gala_model

Data

Scripts necessary for downloading data can be found in scripts/Data.

Documents

An artifact documentation is available in docs.