Resumen
GPGPU (General Purpose computing for Graphical Processing Units) technology allows one to harness the computational power of a GPU (Graphical Processing Unit) and apply it to practically any computationally-intensive task benefiting from parallelization.Software relying on GPGPU inevitably runs in performance problems as the complexity of the program grows and new functionality is introduced. This paper proposes a method to alleviate that particular issue, improving overall GPU program performance. Proposed method entails the creation of problem-oriented programs from the code of the original program.A concept of problem-oriented program is introduced, and the key parts differentiating them from original programs are discussed. The preferable degree of program?s specialization is covered.Various aspects of practical application of this approach are presented. Comparison with existing methods for enhancing the software performance is made, presenting the similarities and differences between proposed approach and said methods, as well their general applicability on GPU.