A tool for autotuning CUDA and OpenCL kernels called the Kernel Tuning Toolkit (KTT) is developed in collaboration between Faculty of Informatics, and Institute of Computer Science, MU. The framework includes a library of examples that demonstrate its use, but also serve as benchmarks used for further research. These examples were created gradually, are not programmed in a completely uniform style and have a lot of redundant code.

The task of the bachelor's thesis is to unify the library of examples so that individual examples are programmed in a uniform programming style and do not contain unnecessary redundancies. The student first analyzes the code of the examples, then summarizes the functionality that they should contain and separate it into a general class from which individual examples would inherit and specialize only the necessary parts of the code (e.g., creating a tuning space). Next, the student implements appropriate functionality that is currently missing in the examples (e.g., configuring tuning space searcher, or dynamic autotuning).