ICall-Bench：评测大模型在二进制程序中的跨过程推理
原标题：ICALL-BENCH: Benchmarking LLM Interprocedural Reasoning in Binary Programs
AI 摘要
研究团队提出 ICall-Bench 基准，用于直接评估大语言模型及智能体在二进制程序中进行跨过程语义推理的能力，以间接调用目标恢复作为可验证的探测任务。该基准覆盖 50 个真实 C/C++ 程序，平均每个程序约 7.5 万行代码、941 个函数，并设置纯模型推理、静态工具增强智能体、联网开放世界智能体三种分析模式。对五个近期模型的评测显示，程序级可验证成功率在 23% 到 81% 之间，信息获取更多并不必然带来更好结果，最困难案例的瓶颈仍是深层跨过程推理。
正文节选
ICall-Bench: Benchmarking LLM Interprocedural Reasoning in Binary Programs Abstract Recent benchmarks show that large language models (LLMs) achieve strong performance in both function-level binary understanding and end-to-end binary reverse engineering tasks, but whether LLMs can reason over interprocedural semantics in compiled binaries remains unclear. Function-level benchmarks offer limited visibility into cross-function relationships, while whole-program benchmarks typically evaluate final