CODO: An Automated Compiler for Comprehensive Dataflow Optimization
Weichuang Zhang, Yiquan Wang, Xinzhou Zhang, Chi Zhang, Yu Feng, Xiaofeng Hou, Chao Li, Jieru Zhao, Minyi Guo
Abstract
FPGAs are well-suited for dataflow architectures that process data in a streaming or pipelined manner, thus satisfying the high computational and communication demands of emerging applications. However, manually implementing an efficient dataflow architecture for large-scale applications is still challenging, even for specialists who use high-level synthesis (HLS) to simplify FPGA programming. To address this, we introduce CODO, an automated compiler that generates feasible and efficient dataflow accelerators on FPGAs. CODO features a systematic method for detecting and eliminating both coarse-grained and fine-grained dataflow violations. Building on this, CODO performs both on- and off-chip data movement optimizations to maximize transfer efficiency. To guarantee a higher design quality, CODO performs automatic scheduling to generate high-performance dataflow accelerators, ensuring a balanced performance-resource trade-off. Synthesis results show that CODO delivers to latency speedups on typical computation kernels and to speedups on DNN models compared to SOTA frameworks. In on-board evaluations, CODO achieves average speedup on CNN models and average speedup on the GPT-2 model over SOTA frameworks. The compiler is open-sourced at https://github.com/sjtu-zhao-lab/codo-artifact.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
Builds on11
- Heterogeneous Dataflow Accelerators for Multi-DNN WorkloadsHyoukjun Kwon, Liangzhen Lai, Michael Pellauer, Tushar Krishna et al.HPCA 2021 · 143 citations
- DFX: A Low-latency Multi-FPGA Appliance for Accelerating Transformer-based Text GenerationSeongmin Hong, Seungjae Moon, Junsoo Kim, Sungjae Lee et al.MICRO 2022 · 107 citations
- Think Fast: A Tensor Streaming Processor (TSP) for Accelerating Deep Learning WorkloadsDennis Abts, Jonathan Ross, Jonathan Sparling, Mark Wong-VanHaren et al.ISCA 2020 · 91 citations
- ScaleHLS: A New Scalable High-Level Synthesis Framework on Multi-Level Intermediate RepresentationHanchen Ye, Cong Hao, Jianyi Cheng, Hyunmin Jeong et al.HPCA 2022 · 77 citations
- Predictable accelerator design with time-sensitive affine typesRachit Nigam, Sachille Atapattu, Samuel Thomas, Zhijing Li et al.PLDI 2020 · 58 citations
Related papers
- Memory and Computation Coordinated Mapping of DNNs onto Complex Heterogeneous SoCSize Zheng, Siyuan Chen, Yun LiangDAC 2023 · 10 citations
- A Streaming Collectives Interface Targeting Dataflow Acceleration and HPC WorkloadsNicholas Contini, Jake Queiser, Bharath Ramesh, Hari Subramoni et al.SC 2025 · 1 citation
- HIDA: A Hierarchical Dataflow Compiler for High-Level SynthesisHanchen Ye, Hyegang Jun, Deming ChenASPLOS 2024 · 21 citations
- StreamTensor: Make Tensors Stream in Dataflow Accelerators for LLMsHanchen Ye, Deming ChenMICRO 2025 · 5 citations
- Cocktailer: Analyzing and Optimizing Dynamic Control Flow in Deep LearningChen Zhang, Lingxiao Ma, Jilong Xue, Yining Shi et al.OSDI 2023 · 28 citations
