SELF-PROFILING MECHANISMS FOR REAL-TIME CODE COMPILERS

Abstract

This paper explores the concept of self-profiling compilers as a means of adaptive real-time code optimization. The approach relies on collecting dynamic performance metrics during program execution and analyzing the collected data to choose the most effective compilation strategies. We propose a compiler architecture capable of automatically detecting performance-critical code regions (hotspots), adjusting the configuration of optimization passes, and recompiling code with updated metrics. A prototype was implemented based on LLVM with an embedded runtime agent responsible for code instrumentation, metric collection, and interaction with a dynamic Pass Manager. A series of experiments were conducted across various hardware platforms, including desktop CPUs and ARM-based architectures. The results demonstrated significant performance gains without noticeable increases in compilation time or resource usage. These findings confirm the feasibility of integrating self-profiling into next-generation compilers targeting high-performance computing, edge systems, and mobile devices. The paper presents the concept of self-profiling as a tool for real-time code optimization. A prototype based on LLVM with embedded runtime analysis has been implemented. The results demonstrate the advantages of the proposed approach.