Artifact for GraphChallenge'2024 champion🏆

MERCURY: Efficient Subgraph Matching on GPUs with Hybrid Scheduling, HPEC'2024

1. Dataset preparation

1.1. Download datasets.(amazon0302)

wget https://graphchallenge.s3.amazonaws.com/snap/amazon0302/amazon0302_adj.tsv

1.2. Convert tsv to mtx.

amazon0302_adj.tsv -> amazon0302_adj.mtx

python3 ./datasets/tsv2mtx.py amazon0302_adj

1.3. Convert mtx to binary file.

 amazon0302_adj.mtx -> amazon0302_adj/graph.meta.txt  
                    -> amazon0302_adj/graph.edge.bin 
                    -> amazon0302_adj/graph.vertex.bin

sh ./datasets/converter.sh amazon0302_adj.mtx amazon0302_adj

2. Compile implementation.

2.1 For single gpu triangle counting

make tc_challenge

2.2 For multi-gpu triangle counting

cd tc_multigpu && make tc_challenge_multigpu

2.3 For multi-gpu subgraph matching

cd subgraph_matching && make sm_multigpu

3. Running code.

2.1 For single gpu triangle counting

Usage: ./tc_challenge <graph_path>

./tc_challenge  ~/data/cit-Patents/graph

2.2 For multi-gpu triangle counting

Usage: mpirun -n <process_num> --bind-to numa ./tc_challenge_multigpu <graph_path>

mpirun -n 4 --bind-to numa ./tc_challenge_multigpu ~/data/cit-Patents/graph

2.3 For multi-gpu subgraph matching

Usage: mpirun -n <process_num> --bind-to numa ./sm_multigpu <graph_path> <pattern_name>
Support Patterns(pattern_name): Pattern-Enumeration(P1,P2,P3,P4,P5,P6,P7,P8)

mpirun -n 4 --bind-to numa ./tc_challenge_multigpu ~/data/cit-Patents/graph P1