| AWB-GCN (Geng et al., 2020) |
Three load balancing techniques |
Large |
GCN |
PyG-CPU |
| Fine-grained pipelining of aggregation and combination. |
PyG-GPU |
|
HyGCN |
| LW-GCN (Tao et al., 2021) |
Apply data Quantization and workload tiling |
Small |
GCN |
PyG-CPU |
| Works effectively on resource limited edge devices. |
GraphSAGE |
PyG-GPU |
|
|
AWB-GCN |
| SPA-GCN (Sohrabizadeh, Chi & Cong, 2022) |
Four levels of parallelization |
Small |
GCN |
PyG-CPU |
| GCN-based graph matching. |
SimGNN |
PyG-GPU |
|
|
PyG-CPU |
| FP-GNN (Tian et al., 2022) |
Support flexible execution order |
Large |
GCN |
PyG-GPU |
| Adaptive graph partition strategy |
GraphSAGE |
HyGCN |
|
GAT |
BoostGCN |
| FPGAN (Yan, Tong & Zhi, 2020) |
Accelerate GAT inference |
Large |
|
|
| Shift addition unit |
GAT |
PyG-CPU |
| SoftMax approximation |
|
PyG-GPU |
| BoostGCN (Zhang, Kannan & Prasanna, 2021) |
|
Large |
GCN |
PyG-CPU |
| PCFA with 3-D partitioning |
PyG-GPU |
| Two types of feature update modules. |
DGL-CPU |
| Task scheduling optimization for aggregation and combination |
DGL-GPU |
|
HyGCN |
|
PyG-CPU |
| I-GCN (Geng et al., 2021b) |
Graph restructuring algorithm—islandization |
Large |
GCN |
PyG-GPU |
| Improve data locality |
GraphSAGE |
DGL-CPU |
| Avoiding redundant aggregation |
GIN |
DGL-GPU |
|
|
HyGCN |
|
|
AWB-GCN |
|
GCN |
|
| BlockGNN (Zhou et al., 2021) |
CirCore architecture for matrices computation |
Large |
GCN |
HyGCN |
| Performance and resource model |
GraphSAGE |
| Reduce the computational complexity of GNNs |
GAT |
|
G-GCN |
|
GCN |
|
GIN |
| FlowGNN (Sarkar et al., 2022) |
Generic GNN acceleration framework |
Large |
GAT |
PyG-CPU |
| Developed by using high-level synthesis (HLS) |
PNA |
PyG-GPU |
|
DGN |
I-GCN |
|
VN |
|
