Page 19 - 《软件学报》2025年第12期

5400 软件学报 2025 年第 36 卷第 12 期

5 总 结

激进编译优化会改变程序中的浮点计算行为, 导致计算结果不一致性问题. 本文提出了一种浮点指令序列差
异性引导的 Delta-Debugging 定位方法 FI3D, 能够高效定位程序中导致编译优化结果不一致性的代码. 一方面,
FI3D 基于 Delta-Debugging 中的回溯搜索机制有效支持了多源问题代码的定位. 另一方面, FI3D 利用函数编译优
化前后浮点指令序列的差异性过滤掉冗余代码并优先搜索浮点计算行为改变程度高的代码, 提升了问题代码搜索
定位效率. 本文基于 LLVM 编译器、Python difflib 模块和编译优化结果不一致性代码定位工具 PLiner 实现了
FI3D, 并基于 NPB 测试集、GSL 数学库和 floatsmith 测试集中的典型用例进行了评测. 相比当前最先进的工具
PLiner, FI3D 能够有效定位 PLiner 失效的 4 个用例, 且对 PLiner 成功定位的 14 个用例 FI3D 获得了平均 26.8%
的性能提升. 实验结果显示了 FI3D 的有效性和高效性. 未来主要工作是将 FI3D 的分析定位思想移植支持其他开
源编译器, 例如 GNU 编译器 GCC 等.

References:
[1] Corden MJ, Kreitzer D. Consistency of floating-point results using the intel compiler or why doesn’t my application always give the same
answer. 2009. https://software.intel.com/sites/default/files/article/164389/fp-consistency-102511.pdf
[2] Sawaya G, Bentley M, Briggs I, Gopalakrishnan G, Ahn DH. FLiT: Cross-platform floating-point result-consistency tester and workload.
In: Proc. of the 2017 IEEE Int’l Symp. on Workload Characterization. Seattle: IEEE, 2017. 229–238. [doi: 10.1109/IISWC.2017.
8167780]
[3] Bentley M, Briggs I, Gopalakrishnan G, Ahn DH, Laguna I, Lee GL, Jones HE. Multi-level analysis of compiler-induced variability and
performance tradeoffs. In: Proc. of the 28th Int’l Symp. on High-performance Parallel and Distributed Computing. Phoenix: ACM, 2019.
61–72. [doi: 10.1145/3307681.3325960]
[4] Guo H, Laguna I, C. Rubio-González. PLiner: Isolating lines of floating-point code for compiler-induced variability. In: Proc. of the 2020
Int’l Conf. for High Performance Computing, Networking, Storage and Analysis. Atlanta: IEEE, 2020. 1–14. [doi: 10.1109/SC41405.
2020.00053]
[5] Laguna I. Varity: Quantifying floating-point variations in HPC systems through randomized testing. In: Proc. of the 2020 IEEE Int’l
Parallel and Distributed Processing Symp. NewOrleans: IEEE, 2020. 622–633. [doi: 10.1109/IPDPS47924.2020.00070]
[6] Yu HB, Yi X, Yin BH, Li F, Chen ZB, Huang C. Efficient generation of floating-point inputs for compiler-induced variability. In: Proc.
of the 2023 IEEE Int’l Conf. on Software Analysis, Evolution and Reengineering. Macao: IEEE, 2023. 224–235. [doi: 10.1109/
SANER56733.2023.00030]
[7] Miao D, Laguna I, Rubio-González C. Expression isolation of compiler-induced numerical inconsistencies in heterogeneous code. In:
Proc. of the 38th Int’l Conf. on High Performance Computing. Hamburg: Springer, 2023. 381–401. [doi: 10.1007/978-3-031-32041-5_20]
[8] Panchekha P, Sanchez-Stern A, Wilcox JR, Tatlock Z. Automatically improving accuracy for floating point expressions. In: Proc. of the
36th ACM SIGPLAN Conf. on Programming Language Design and Implementation. Portland: ACM, 2015. 1–11. [doi: 10.1145/2737924.
2737959]
[9] Yi X, Chen LQ, Mao XG, Ji T. Efficient automated repair of high floating-point errors in numerical libraries. Proc. of the ACM on
Programming Languages, 2019, 3(POPL): 56. [doi: 10.1145/3290369]
[10] Zou DM, Gu YC, Shi YF, Wang MZ, Xiong YF, Su ZD. Oracle-free repair synthesis for floating-point programs. Proc. of the ACM on
Programming Languages, 2022, 6(OOPSLA2): 159. [doi: 10.1145/3563322]
[11] Zeller A. Yesterday, my program worked. Today, it does not. Why? In: Proc. of the 7th European Software Engineering Conf. Held
Jointly with the 7th ACM SIGSOFT Symp. on the Foundations of Software Engineering. Toulouse: Springer, 1999. 253–267. [doi: 10.
1007/3-540-48166-4_16]
[12] Galassi M, Davies J, Theiler J, Gough B, Jungman G, Alken P, Booth M, Rossi F. GNU Scientific Library Reference Manual. 3rd ed.,
Godalming: Network Theory Ltd., 2009. 1–572.
[13] Lam MO, Vanderbruggen T, Menon H, Schordan M. Tool integration for source-level mixed precision. In: Proc. of the 3rd Int’l
Workshop on Software Correctness for HPC Applications. Denver: IEEE, 2019. 27–35. [doi: 10.1109/CORRECTNESS49594.2019.
00009]
[14] Benz F, Hildebrandt A, Hack S. A dynamic program analysis to find floating-point accuracy problems. In: Proc. of the 33rd ACM
SIGPLAN Conf. on Programming Language Design and Implementation. Beijing: ACM, 2012. 453–462. [doi: 10.1145/2254064.
2254118]
[15] Sanchez-Stern A, Panchekha P, Lerner S, Tatlock Z. Finding root causes of floating point error. In: Proc. of the 39th ACM SIGPLAN
Conf. on Programming Language Design and Implementation. Philadelphia: ACM, 2018. 256–269. [doi: 10.1145/3192366.3192411]