FuzzyFlow: Leveraging Dataflow for Program Optimization Testing and Debugging

Watch a conference talk from Supercomputing '23 that introduces FuzzyFlow, a fault localization and test case extraction framework for program optimization testing. Explore how dataflow program representations can capture reproducible system states and optimization effects, enabling faster semantic equivalence checking. Discover techniques for minimizing test inputs and achieving up to 528 times faster optimization testing compared to traditional approaches. Learn about key concepts including differential fuzzing, program cutouts, parametric dataflow analysis, input configuration reduction, and coverage-guided fuzzing. Follow along as speaker Philipp Schaad from ETH Zurich's Scalable Parallel Computing Lab demonstrates real-world applications and explains how FuzzyFlow helps performance engineers verify and debug complex program optimizations.