基于GPU的Figaro：两张表格
高能物理 - 实验
2025-08-06 v2
摘要
本文介绍了Figaro-GPU算法的实现，用于计算由两个表通过自然连接定义的联合矩阵上的QR和SVD分解。Figaro-GPU的主要新颖之处是对Figaro算法的GPU实现：符号变换结合GPU并行计算。这导致与联合大小与输入大小之比比例的理论性能提升。在合成表格的实验中，对于计算上三角矩阵和右奇异向量矩阵，Figaro-GPU在运行时间上相较于NVIDIA cuSolver库表现更好，提升幅度随着联合大小与输入大小之差的比例变化，分别为NVIDIA 2070为5倍-150倍，NVIDIA 4080可达最高160倍，而内存使用最多可比GPU cuSolver低出1000倍。对于计算奇异值，Figaro-GPU在运行时间上相较于NVIDIA cuSolver库提升范围为2.8倍-31倍（NVIDIA 4080）。
引用
@article{arxiv.2503.23384,
title = {First measurement of $\nu_e$ + $\bar{\nu}_e$ charged current single charged pion production differential cross sections on argon using the MicroBooNE detector},
author = {MicroBooNE collaboration and P. Abratenko and D. Andrade Aldana and L. Arellano and J. Asaadi and A. Ashkenazi and S. Balasubramanian and B. Baller and A. Barnard and G. Barr and D. Barrow and J. Barrow and V. Basque and J. Bateman and O. Benevides Rodrigues and S. Berkman and A. Bhat and M. Bhattacharya and M. Bishai and A. Blake and B. Bogart and T. Bolton and M. B. Brunetti and L. Camilleri and D. Caratelli and F. Cavanna and G. Cerati and A. Chappell and Y. Chen and J. M. Conrad and M. Convery and L. Cooper-Troendle and J. I. Crespo-Anadon and R. Cross and M. Del Tutto and S. R. Dennis and P. Detje and R. Diurba and Z. Djurcic and K. Duffy and S. Dytman and B. Eberly and P. Englezos and A. Ereditato and J. J. Evans and C. Fang and B. T. Fleming and W. Foreman and D. Franco and A. P. Furmanski and F. Gao and D. Garcia-Gamez and S. Gardiner and G. Ge and S. Gollapinni and E. Gramellini and P. Green and H. Greenlee and L. Gu and W. Gu and R. Guenette and P. Guzowski and L. Hagaman and M. D. Handley and O. Hen and C. Hilgenberg and G. A. Horton-Smith and A. Hussain and B. Irwin and M. S. Ismail and C. James and X. Ji and J. H. Jo and R. A. Johnson and D. Kalra and G. Karagiorgi and W. Ketchum and M. Kirby and T. Kobilarcik and N. Lane and J. -Y. Li and Y. Li and K. Lin and B. R. Littlejohn and L. Liu and W. C. Louis and X. Luo and T. Mahmud and C. Mariani and J. Marshall and N. Martinez and D. A. Martinez Caicedo and S. Martynenko and A. Mastbaum and I. Mawby and N. McConkey and L. Mellet and J. Mendez and J. Micallef and T. Mohayai and A. Mogan and M. Mooney and A. F. Moor and C. D. Moore and L. Mora Lepin and M. M. Moudgalya and S. Mulleria Babu and D. Naples and A. Navrer-Agasson and N. Nayak and M. Nebot-Guinot and C. Nguyen and J. Nowak and N. Oza and O. Palamara and N. Pallat and V. Paolone and A. Papadopoulou and V. Papavassiliou and H. Parkinson and S. F. Pate and N. Patel and Z. Pavlovic and E. Piasetzky and K. Pletcher and I. Pophale and X. Qian and J. L. Raaf and V. Radeka and A. Rafique and M. Reggiani-Guzzo and J. Rodriguez Rondon and M. Rosenberg and M. Ross-Lonergan and I. Safa and D. W. Schmitz and A. Schukraft and W. Seligman and M. H. Shaevitz and R. Sharankova and J. Shi and E. L. Snider and S. Soldner-Rembold and J. Spitz and M. Stancari and J. St. John and T. Strauss and A. M. Szelc and N. Taniuchi and K. Terao and C. Thorpe and D. Torbunov and D. Totani and M. Toups and A. Trettin and Y. -T. Tsai and J. Tyler and M. A. Uchida and T. Usher and B. Viren and J. Wang and M. Weber and H. Wei and A. J. White and S. Wolbers and T. Wongjirad and K. Wresilo and W. Wu and E. Yandel and T. Yang and L. E. Yates and H. W. Yu and G. P. Zeller and J. Zennamo and C. Zhang},
journal= {arXiv preprint arXiv:2503.23384},
year = {2025}
}
备注
Version accepted by PRL