School of Engineering Department of Computer Science and Engineering 121 Compilation, Optimization, and Metaprogramming Supervisor: PARREAUX Lionel / CSE Student: QIU Jiangkun / ISDN Course: UROP1100, Spring As developers, we seek the triangular balance of writing fast, reading fast, and running fast. Abstraction is an excellent way to achieve this, as we abstract and encapsulate well-written algorithms and optimizations into functions, classes, and libraries. However, many abstractions come with a penalty. Many languages state that they have zero-cost abstraction, but are they really zero-cost? Compilers fail to perform some optimizations behind abstractions and pay a runtime performance cost. Even as simple as invoking a function without inlining has some runtime cost and may influence CPU cache and branch predictions; generics in GoLang have runtime performance penalty; Java relies on virtual tables to invoke methods once failed to infer the specific method. This leaves space for our Programmable Zero-Cost Abstractions, which heavily utilized macro, comp-time interpretation, and specialization. We can avoid many abstraction costs at runtime by leaving more work to comp-time. Compilation, Optimization, and Metaprogramming Supervisor: PARREAUX Lionel / CSE Student: YUAN Keyu / COMP Course: UROP1000, Summer Specialization is a useful technique for program optimization, and on top of that, in PZCA(Programmable Zero-cost Abstractions), program monomorphization is implemented to produce better optimization results. In this UROP we implemented a regular expression matcher as the front end of PZCA, which demonstrated the power of this program optimizer. The matcher takes in a static regex as a class hierarchy and a dynamic string, and applies a naive matching function to determine whether the string matches the regex. The original matching function was slow, but after adequately specialized, the algorithm become as fast as compiling regex.
RkJQdWJsaXNoZXIy NDk5Njg=