Instructor

Textbook:

• Optimizing Compilers for Modern Architectures by Randy Allen and Ken Kennedy

Course prerequisites: CSC 451 or CSC 501 (Operating Systems). Helpful: CSC 253 (C and C++ for Java Programmers), CSC 224 (Applied Discrete Mathematics), CSC 234 (Computer Organization & Assembly Language), and MA 121 (Calculus), CSC 512 (Compiler Construction), CSC 548 (Parallel Systems).

Course purpose: This class prepares you to understand advanced research issues in code optimization for scalar and parallel programs. You will be introduced to program analysis, scalar and parallel compilation as well as various advanced topics. The material will cover numerous research papers besides the textbook. Current research will be presented by students and discussed to give a better understanding of open issues in parallel systems.

Course objectives: By the end of the course, you should be able to do the following things:

• Program Analysis and Code Generation. To design and implement program analysis algorithms, such as scanners, parsers, semantic analyzers and code generators; to analyze control flow and data flow of programs; to critique different implementation choices and analysis approaches; to comprehend the limitations of static program analysis.
• Scalar Optimizations. To understand the fundamental principles in code optimization for uniprocessors; to design, implement and evaluate selected optimizations, such as peephole optimization, register allocation, instruction scheduling, code motion etc.; to comprehend the challenges in code optimizations; to incorporate extensions to optimizations into existing compiler back-ends.
• Parallel Optimizations. To understand the fundamental principles in optimizations for parallel architectures; to comprehend data dependence problems, to design, implement and evaluate selected high-level transformations, such as loop transformation, data transformations or a combination of both, such as in tiling; to comprehend and assess the benefits of optimizations on performance with respect to architectural components, such as caches; to incorporate extensions to optimizations into existing compiler back-ends.
• Research Topics. To reiterate hot topics in research on parallel computing; to judge the relevance and impact of novel approaches on the field; to present recent results in a comprehensive manner; to identify open problems; to prototype solutions to selected hot topics and open problems; to demonstrate the benefits of such implementations by methodological evaluations.