DETERMINING SIMILARITIES IN COMPUTER SOFTWARE CODES FOR PERFORMANCE ANALYSIS

摘要

Execution performance of software code is analyzed by converting values of profiling metrics, obtained by at least one software profiling tool in respect of at least one computer software code, into images using a profiling metric conversion process. The conversion process may comprise identifying a set of profiling metrics with synonymous labels within a plurality of labeled profiling metrics whose values are to be converted, using at least one natural language processing scheme, matching all profiling metrics in the set to the same label, and, for each set of profiling metrics, converting values corresponding to the profiling metrics in the set into respective images such that all the converted images have the same image format. The comparison method further comprises computing at least one of similarities and differences between at least two of the images, and comparing execution performance of the software code(s) corresponding to the two images based on the results of the computation.