Interactive computer-aided code vectorization

Abstract

Traditionally, programmer learns to write code using arrays and loops to perform complex computation. Languages designed specifically for scientific computation such as Matlab/Octave support vector based operations which are much faster than arrays and loops. However, converting the loop oriented code to a vector based operation requires skilled programmer, takes time, and introduces many errors. This paper presents an easy to use, interactive web application based system that automatically detect for loops that can be automatically vectorized, allowing the programmer to simply selects which for loop to convert into a vector equivalent operation, debug the output and compare the performance gained. The goal is to ease the task of learning and applying code vectorization techniques. We extended Octave with additional operations to automatically detect possible loop vectorization, and implemented a web application that hides the Octave command line complexity by automatically generating necessary Octave command. Therefore, the user is able to use the tool to learn and use code vectorization tool with ease. The system is built with two goals in mind, ease the manual work of vectorizing loop operations, and enable interactive learning of code vectorization concept.