This research was not just about creating another tool—it was about empowering users in the optimization process, ensuring that automated parallelization aligns with real-world needs rather than rigid compiler assumptions.
The journey began with a frustrating reality: automatic parallelization tools often fail to deliver optimal results because they lack contextual knowledge of the application domain. Manual parallelization, on the other hand, is time-consuming and error-prone, making it an impractical solution at scale. Researchers and developers were left navigating a gray area—needing automation but also requiring control over the process.
This dilemma fueled the development of iCetus, an interactive parallelization environment that combines the strengths of both worlds—leveraging compiler analysis while keeping the user in the loop. Unlike traditional tools that passively apply optimizations, iCetus actively involves developers, enabling them to fine-tune, validate, and enhance code performance for greater efficiency.