Resumen
Problem-oriented program generation is an approach to improving performance for certain class of GPGPU (General Purpose computing for Graphical Processing Units) programs. Creating problem-oriented program generators manually is a time-consuming and difficult task, one that requires the programmer to possess a certain degree of insight in the problem being solved by said program. This paper covers an approach to automating the creation of problem-oriented program generators. Meta-generator is introduced as a program creating said generators from GPU program?s source code. The key steps that meta-generator performs in the course of program analysis are discussed along with technologies most suitable for implementing on each step. Lastly, a comparison between proposed approach and similar existing means to enhancing software performance was made, outlining their key differences.