Zhanyuan Di
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Zhanyuan Di has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- En Shao
State Key Lab of Processors, Institute of Computing Technology, CAS
5 shared publications - Guangming Tan
Chinese Academy of Sciences
4 shared publications - Leping Wang
State Key Lab of Processors, Institute of Computing Technology, CAS
4 shared publications - Ninghui Sun
State Key Lab of Processors, Institute of Computing Technology, CAS
4 shared publications - Dingwen Tao
Chinese Academy of Sciences
3 shared publications - Jie Zhao
Hunan University
3 shared publications
Latest Publications
- Open Access
- 26 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 23 (4) , 1-24
EcoMLC: An Energy-Efficient Machine Learning Compilation Framework via Cache Reuse and Frequency Scaling
While modern machine learning compilers (MLCs) are pushing deep neural network (DNN) inference latency to its theoretical limits, the accompanying energy overhead has become a critical bottleneck for large-scale deployment. Excessive energy consumption not only increases electricity costs, but also affects the sustainability of edge platforms. Traditional performance-first optimizations often lead to significant energy waste, as they overlook the complex interplay between kernel configurations, cache behaviors and frequency scaling. This article presents EcoMLC, an energy-efficient machine learning compilation framework. EcoMLC introduces a reverse scheduling policy combined with an explicit cache priority mechanism, mitigating cache thrashing inherent in standard least recently used (LRU) policies. EcoMLC employs an end-to-end frequency-sweeping analysis to identify optimal energy operating points and enables fine-grained tradeoffs between performance and energy under diverse service level objective (SLO) constraints. Experimental results demonstrate that EcoMLC effectively optimizes various workloads, achieving an average of 43% and 26% energy savings (in J/request and J/token) over NVIDIA’s TensorRT and vLLM, and 22% energy savings on an AMD GPU compared with vLLM. For certain large language model (LLM) configurations, EcoMLC achieves a 33.9% reduction in energy consumption at the cost of a marginal 7% decline in performance.
- 26 October 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
- Vol. 23 (4) , 1-24
Open AccessEcoMLC: An Energy-Efficient Machine Learning Compilation Framework via Cache Reuse and Frequency Scaling
While modern machine learning compilers (MLCs) are pushing deep neural network (DNN) inference latency to its theoretical limits, the accompanying energy overhead has become a critical bottleneck for large-scale deployment. Excessive energy consumption not only increases electricity costs, but also affects the sustainability of edge platforms. Traditional performance-first optimizations often lead to significant energy waste, as they overlook the complex interplay between kernel configurations, cache behaviors and frequency scaling. This article presents EcoMLC, an energy-efficient machine learning compilation framework. EcoMLC introduces a reverse scheduling policy combined with an explicit cache priority mechanism, mitigating cache thrashing inherent in standard least recently used (LRU) policies. EcoMLC employs an end-to-end frequency-sweeping analysis to identify optimal energy operating points and enables fine-grained tradeoffs between performance and energy under diverse service level objective (SLO) constraints. Experimental results demonstrate that EcoMLC effectively optimizes various workloads, achieving an average of 43% and 26% energy savings (in J/request and J/token) over NVIDIA’s TensorRT and vLLM, and 22% energy savings on an AMD GPU compared with vLLM. For certain large language model (LLM) configurations, EcoMLC achieves a 33.9% reduction in energy consumption at the cost of a marginal 7% decline in performance.
- Open Access
- 16 June 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
Accelerating Parallel Structures in DNNs via Parallel Fusion and Operator Co-Optimization
Parallel structures have become a key pattern in deep neural networks (DNNs), offering improved efficiency and scalability. However, existing machine learning compilers (MLCs) face challenges in optimizing these structures due to limited parallel fusion scope and insufficient analysis of intra-operator characteristics. This paper introduces Magneto, a framework designed to accelerate DNN inference by co-optimizing parallel operators. Magneto broadens the fusion scope and incorporates a specialized co-tuning algorithm to optimize operators jointly. Our approach addresses the unique challenges inherent in optimizing parallel structures, enabling significant performance improvements across various hardware platforms. Experimental results show that Magneto outperforms state-of-the-art NVIDIA TensorRT and AMD MIGraphX, achieving geometric mean speedups of 2.27 × and 2.88 ×, respectively.
- 16 June 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Architecture and Code Optimization
Open AccessAccelerating Parallel Structures in DNNs via Parallel Fusion and Operator Co-Optimization
Parallel structures have become a key pattern in deep neural networks (DNNs), offering improved efficiency and scalability. However, existing machine learning compilers (MLCs) face challenges in optimizing these structures due to limited parallel fusion scope and insufficient analysis of intra-operator characteristics. This paper introduces Magneto, a framework designed to accelerate DNN inference by co-optimizing parallel operators. Magneto broadens the fusion scope and incorporates a specialized co-tuning algorithm to optimize operators jointly. Our approach addresses the unique challenges inherent in optimizing parallel structures, enabling significant performance improvements across various hardware platforms. Experimental results show that Magneto outperforms state-of-the-art NVIDIA TensorRT and AMD MIGraphX, achieving geometric mean speedups of 2.27 × and 2.88 ×, respectively.
- Open Access
- 30 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 719-733
Optimizing Deep Learning Inference Efficiency through Block Dependency Analysis
- 30 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 719-733
Open AccessOptimizing Deep Learning Inference Efficiency through Block Dependency Analysis
- Open Access
- 28 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 563-565
Magneto: Accelerating Parallel Structures in DNNs via Co-Optimization of Operators
Deep neural networks (DNNs) increasingly rely on parallel structures to enhance performance and efficiency. However, existing machine learning compilers (MLCs) face challenges in optimizing these structures due to limited parallel fusion scopes and insufficient consideration of intra-operator information. This paper introduces Magneto, a novel framework designed to accelerate parallel structures in DNNs through the co-optimization of parallel operators. By expanding the scope of parallel operator fusion and introducing a dedicated co-tuning algorithm, Magneto unlocks new opportunities for co-optimization. Experimental results demonstrate that Magneto outperforms NVIDIA TensorRT and AMD MIGraphX, achieving speedups of 3.02× and 4.19×, respectively.
- 28 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 563-565
Open AccessMagneto: Accelerating Parallel Structures in DNNs via Co-Optimization of Operators
Deep neural networks (DNNs) increasingly rely on parallel structures to enhance performance and efficiency. However, existing machine learning compilers (MLCs) face challenges in optimizing these structures due to limited parallel fusion scopes and insufficient consideration of intra-operator information. This paper introduces Magneto, a novel framework designed to accelerate parallel structures in DNNs through the co-optimization of parallel operators. By expanding the scope of parallel operator fusion and introducing a dedicated co-tuning algorithm, Magneto unlocks new opportunities for co-optimization. Experimental results demonstrate that Magneto outperforms NVIDIA TensorRT and AMD MIGraphX, achieving speedups of 3.02× and 4.19×, respectively.
- 23 June 2021
- Published by Springer Nature
- p. 263-275
Reducing the Time of Live Container Migration in a Workflow
- 23 June 2021
- Published by Springer Nature
- p. 263-275
Reducing the Time of Live Container Migration in a Workflow