[CS] Byron Zhong MS Presentation/Dec.10

[via cs]

This is an announcement of Byron Zhong's MS Presentation
===============================================
Candidate: Byron Zhong

Date: Tuesday, December 10th

Time: 2PM-3PM 

Location: JCL 298

Remote Location: https://uchicago.zoom.us/j/97852770072?pwd=2AGHPFaOw70GYXUo0S9qYgmnXSF3Jh.1
Meeting ID: 978 5277 0072
Passcode: 246107

Title: Flow-directed Space-efficient Closure Conversion

Abstract: In higher-order languages, a function captures its free variables in a closure, often implemented as a flat record. While this simple structure is effective, more sophisticated representations can significantly reduce overhead. To manage the multiplicity of possible closure representation decisions, this thesis introduces a model that represents closure decisions and a transformation that applies these decisions to a program. This paper presents several optimization passes that rewrites the closure representations, each of which greedily minimizes the dynamic allocation required for closures. Higher-order control-flow analysis provides more information about the flows of function values. This information is used to justify soundness for some of the proposed rewriting passes as well as to provide heuristics for others. This paper presents an efficient implementation of 0-CFA with supports for ref-cells and separate compilation. This novel approach is implemented in the Standard ML of New Jersey compiler and evaluated against the existing implementation. The results show that this approach is scalable to large programs and effective in reducing dynamic allocation and overall performance.

Advisors: John Reppy

Committee Members: John Reppy, Ravi Chugh, Olin Shivers