Dr. Junqiao Qiu will present a Computer Science department lecture on Friday, November 19, 2021, at 3:00 p.m. in Rekhi 214. Dr. Qiu is an assistant professor in the Computer Science department. His research focuses on parallel computing, programming systems, and compiler optimization.
Unleashing the Power of Parallelism: Extracting Fine-grained SIMD Parallelism from Data-intensive Computations
Exposing parallelism is key to computing efficiency and scalability of software application. Modern microprocessors feature a variety of hardware parallelism from on-chip multiprocessors to instruction levels. Effectively leveraging such rich hardware parallelism critically affects the performance. As increasingly wide single instruction multiple data (SIMD) units are ubiquitous in modern commodity processors, SIMD vectorization of programs becomes crucial for achieving high performance as well as energy efficient.
This talk focuses on exploiting fine-grained SIMD parallelism to accelerating the data-intensive computations. We will discuss three mainstream techniques for extracting SIMD parallelism, including loop vectorization, basic block or superword level parallelism (SLP) based vectorization, and single program multiple data (SPMD) execution on the SIMD extensions. We will also discuss existing technical challenges and future research opportunities, especially the ones about SIMD vectorization of those inherently sequential computations.