Finding Optimal Hardware Configurations For C Code

To look at a piece of C code especially an uncommented one is often not very enlightening. Even small pieces of code with one or two functions will often appear to perform the same array initializations, loops over the data, and basic arithmetic functions, without clearly indicating what the overall purpose of the routine is. This can be frustrating to those using the code as part of a package who are forced to understand it as a “black box,” without completely understanding its properties. More specifically, attempting to find the most optimal way to run the code becomes a very difficult problem if there are, for example, several different choices of harware to choose from, and the actions of the program are a mystery. In this problem, we choose to attack a tractable subproblem that we hope will be helpful for deciding how to run this software for a user with several different hardware options. Specifically, we have observed that most single code files perform one of two functions: data manipulation and computation. We hope to be able to successfully classify C code files into these two main categories to allow a user who has not been informed of its functionality to quickly determine its core purpose, and use this information to enhance performance of the overall project. We choose to look at the assembly representation of these files, whose simple structure and more limited vocabulary we believe contain the key clues to functionality in order to perform this classification. Generally, the conversion to assembly is done “behind the scenes,” and only a small percentage of programmers actually feel comfortable reading any particular type of assembly and understanding what it does, so we believe that assembly is underutilized in understanding code’s functionality. Thus, we also hope that classifying code in this manner will provide a strong argument for using assembly to extract features of code for use in future machine learning applications pertaining to C or other programming languages, and eliminate the need to look at assembly directly.