1 paper across 1 session
We combine LLM-synthesized performance-characterizing constraints with fuzzing to uncover difficult-to-find code inefficiencies and generate performance-stressing tests.
