Optimizing compilers often sacrifice binary size in pursuit of higher run-time performance. In the absence of method execution profiles, they uniformly apply performance-oriented optimizations, typically various forms of code duplication. Duplications in methods that are rarely or never executed only increase binary size without improving performance. Modern static profiler use ML to predict branch profiles, yet they do not identify which methods will be frequently executed at run time. Doing so would enable more selective optimizations, reducing binary size while preserving or only minimally affecting run-time performance.

Accurate static profile prediction is crucial for achieving optimal program performance in the absence of dynamic profiles. However, existing static profiling methods struggle to fully exploit the complex structure of the compiler’s intermediate representation and fail to effectively utilize the calling context information needed for accurate profile inference. To address these limitations, we introduce GraalNN, a Graph Neural Network-based static profiling framework that directly learns structural information from control-flow graphs. This reduces the reliance on handcrafted features and minimizes the effort required for feature engineering while improving the model’s ability to predict profiles. GraalNN adopts a two-stage approach: it predicts context-insensitive profiles during parsing and uses contextual information from the call graph to refine profiles during inlining. This methodology achieves a 10.13% runtime speedup across a diverse set of industry-standard benchmarks, surpassing state-of-the-art static profiling techniques by more than 2.5%. Furthermore, GraalNN improves throughput by 3.7% over other static profiling methods on real-world microservices.