Bonjour,

Je vous invite à la présentation du prédoc 3 d'Olivier Melançon:

Jury

Abstract

Dynamic languages like Python, JavaScript, Ruby, and R offer advantages in rapid development but often suffer from slower execution speeds compared to statically-typed languages such as C. Traditional Just-In-Time compiler techniques, such as Lazy Basic Block Versioning, have made strides in mitigating this issue by propagating forward context to optimize runtime type checks. However, these techniques are limited by their inability to consider the future use of computed results in the optimization process. This research aims to address this limitation by introducing Demand-Driven Basic Block Versioning, an extension of LBBV that utilizes backward propagation of context to enhance code generation. The work is expected to contribute an algorithm for DDBBV applicable to highly dynamic languages and an implementation within a JIT compiler for a representative subset of the Python language.