Next generation SIMD hardware and software. This research focuses on bringing the best features of GPGPUs to next generation SIMD engines; designing and supporting better programming interfaces for next generation SIMD.
- Generic SIMD intrinsic library (open-sourced on github)
- Fiorano JIT (Python) by repurposing a production Java JIT infrastructure to optimize Python. Main findings of the work is the “repurposed JIT phenomenon” (discussions on quora)
- R JIT: Optimizing R (a popular data analysis language) via Interpreter-level Specialization. Collaboration w/ Dr Padua and Ph.D candidate Haichuan Wang at UIUC.
Large-scale graph runtime. Many analytics software today are developed by mathematicians that focus on performance at the algorithm level. In this work, we bring decades of performance engineering experiences from the HPC community to the analytics domain. The work aims at providing a highly optimized distributed graph runtime (System G) with built-in support for parallelism, persistence, and fault tolerance for a broad class of graph-based workloads.