7 总结与展望本文围绕 NPU 下的核内调度问题展开研究重点针对硬件平台设计 了高效、通用的核内调度算法。随着 AI 技术在边缘计算场景的广泛应用NPU 成为支撑 神经网络推理任务的关键硬件。然而其软件模型适配的复杂性尤其是计算图中异构算子的调度与缓存管理已成为制约其大规模商用的瓶颈。本文通过建模、算法设计与实验验证系统解决了核内调度中的三个核心问题最小缓存驻留调度、缓存分配与换入换出 机制、以及性能优化策略。本研究不仅为 NPU 软件栈开发提供了理论依据与算法支撑也 为未来智能计算系统的高效调度指明了方向。未来我们将在现有工作基础上进一步探索动态、自适应、多目标的调度机制推动 NPU 在更复杂场景下的高效应用为构建下一代智能计算平台贡献力量。参考文献[1] Voulodimos A, Doulamis N, Doulamis A, et al., Deep learning for computer vision: A brief review, Computational intelligence and neuroscience, 2018(1):7068349, 2018.[2] Zhao W X, Zhou K, Li J, et al., A survey of large language models, arXiv preprint arXiv:2303.18223, 1(2), 2023.[3] Mehrish A, Majumder N, Bharadwaj R, et al., A review of deep learning techniquesforspeech processing, Information Fusion, 99:101869, 2023.[4] Muyal T A, OpenNPU: an open source platform for automatic neural network synthesis for FPGAs., Universidade de São Paulo, 2023.[5] Park S S, Chung K S, Conna: Conffgurable matrix multiplication engine for neural network acceleration, Electronics, 11(15):2373, 2022.[6] Abbasi-Esfeden R, Nurkanovic A, Diehl M, et al., An Effffcient Mixed-Integer Formulation and an Iterative Method for Optimal Control of Switched Systems Under Dwell Time Constraints, arXiv preprint arXiv:2501.05158, 2025.[7] Kahn A B, Topological sorting of large networks, Communications of the ACM, 5(11): 558-562, 1962.[8] Belady L A, A study of replacement algorithms for a virtual-storage computer, IBM Systems journal, 5(2):78-101, 1966.[9] Van Laarhoven P J, Aarts E H, Lenstra J K, Job shop scheduling by simulated annealing, Operations research, 40(1):113-125, 1992.[10] Sampson J R, Adaptation in natural and artiffcial systems (John H. Holland), Society for Industrial and Applied Mathematics, 1976.[11] Wu Z, Pan S, Chen F, et al., A comprehensive survey on graph neural networks, IEEE transactions on neural networks and learning systems, 32(1):4-24, 2020.
核内调度问题的分层优化：缓存管理与性能均衡策略 总结与展望+参考文献
7 总结与展望本文围绕 NPU 下的核内调度问题展开研究重点针对硬件平台设计 了高效、通用的核内调度算法。随着 AI 技术在边缘计算场景的广泛应用NPU 成为支撑 神经网络推理任务的关键硬件。然而其软件模型适配的复杂性尤其是计算图中异构算子的调度与缓存管理已成为制约其大规模商用的瓶颈。本文通过建模、算法设计与实验验证系统解决了核内调度中的三个核心问题最小缓存驻留调度、缓存分配与换入换出 机制、以及性能优化策略。本研究不仅为 NPU 软件栈开发提供了理论依据与算法支撑也 为未来智能计算系统的高效调度指明了方向。未来我们将在现有工作基础上进一步探索动态、自适应、多目标的调度机制推动 NPU 在更复杂场景下的高效应用为构建下一代智能计算平台贡献力量。参考文献[1] Voulodimos A, Doulamis N, Doulamis A, et al., Deep learning for computer vision: A brief review, Computational intelligence and neuroscience, 2018(1):7068349, 2018.[2] Zhao W X, Zhou K, Li J, et al., A survey of large language models, arXiv preprint arXiv:2303.18223, 1(2), 2023.[3] Mehrish A, Majumder N, Bharadwaj R, et al., A review of deep learning techniquesforspeech processing, Information Fusion, 99:101869, 2023.[4] Muyal T A, OpenNPU: an open source platform for automatic neural network synthesis for FPGAs., Universidade de São Paulo, 2023.[5] Park S S, Chung K S, Conna: Conffgurable matrix multiplication engine for neural network acceleration, Electronics, 11(15):2373, 2022.[6] Abbasi-Esfeden R, Nurkanovic A, Diehl M, et al., An Effffcient Mixed-Integer Formulation and an Iterative Method for Optimal Control of Switched Systems Under Dwell Time Constraints, arXiv preprint arXiv:2501.05158, 2025.[7] Kahn A B, Topological sorting of large networks, Communications of the ACM, 5(11): 558-562, 1962.[8] Belady L A, A study of replacement algorithms for a virtual-storage computer, IBM Systems journal, 5(2):78-101, 1966.[9] Van Laarhoven P J, Aarts E H, Lenstra J K, Job shop scheduling by simulated annealing, Operations research, 40(1):113-125, 1992.[10] Sampson J R, Adaptation in natural and artiffcial systems (John H. Holland), Society for Industrial and Applied Mathematics, 1976.[11] Wu Z, Pan S, Chen F, et al., A comprehensive survey on graph neural networks, IEEE transactions on neural networks and learning systems, 32(1):4-24, 2020.
