For his M.S. thesis research, Sam investigated store-load forwarding in conventional superscalar processors. Leveraging the insight that store multi-versioning in the store queue incurs significant complexity but rarely increases performance, Sam proposed the forwarding cache, a low-energy, complexity-effective alternative to the store queue. The forwarding cache achieves the same performance as the store queue while dissipating 10X less energy. Sam's M.S. thesis research was advised by Dr. Matthew Frank.

As a member of the IMPACT research group, Sam plans to develop compiler analyses and tools for feedback-driven parallelization of data-parallel applications. Given source code that contains syntactic errors, modern compilers identify those errors with great specificity, typically displaying the file name and line number in which each error is found, along with a description of the error. This information allows the programmer to quickly understand and correct syntax errors. By contrast, given source code that exhibits little parallelism, the compiler behaves as a black box, providing little or no useful feedback to the programmer. Programmers are left to blindly restructure and re-execute their algorithms until the desired level of performance is observed. Sam proposes to develop compiler tools that will assist the programmer by providing useful information about the compiler's view of the parallelism in the program. He is currently studying the parallelism in medical imaging applications in an effort to understand what types of compiler-provided feedback would be useful to programmers of data-parallel applications.

• Medical imaging applications
• Feedback-driven parallelization of implicitly parallel programs
• Energy-efficient store-load forwarding