一句话：提出 Panoptes 框架，通过 Profile 聚类识别相似瓶颈的二进制，以促进上下文感知的二进制优化复用。
激进的编译器优化策略（如自动调优和 Profile-Guided Optimization）虽能提升应用效率，却面临可扩展性问题。程序聚类可通过识别具有相似瓶颈的二进制来促进优化重用，从而缓解该问题。然而，现有程序聚类技术无法在能代表有意义优化目标的粒度上进行聚类，限制了其实际效用。为此，论文提出 Panoptes，一个面向上下文感知二进制优化的 Profile 聚类框架，旨在围绕优化目标对二进制进行 Profile 聚类，以发现可复用优化配置的相似程序。摘要未报告具体实现细节、实验评估与量化结果。
Optimization strategies are crucial for application efficiency, yet aggressive strategies such as compiler autotuning and profile-guided optimization suffer from scalability issues. Clustering techniques can mitigate this by identifying binaries with similar bottlenecks and promoting optimization reuse. However, existing program clustering techniques fail to cluster at granularities that represent meaningful optimization targets, thus hindering their utility.
@inproceedings{paperbot11912,
title = {Panoptes: A Profile Clustering Framework for Context-Aware Binary Optimization},
author = {Edwin P. Kayang and Eric Jahns and Mishel Jyothis Paul and Michel A. Kinsy},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3744916.3787767}
}