In recent years, academic research increasingly requires high performance computing for large amounts of data. In computer centers, accelerators such as GPUs and vector processors are being introduced in addition to general-purpose processors as high performance computing systems to answer such computational needs.
On the other hand, in many cases, the code written by users of such high performance computing systems does not fully exploit the performance of the processors and accelerators. Many users are experts in their respective fields of academic research and are not experts in computer systems. Therefore, it is difficult for users to extract performance from increasingly complex high performance computing systems such as accelerators, and they either rely on automatic optimization techniques using tools such as compilers to increase speed, or they ask computer system experts to perform code optimization. In the former case, the performance of the computer system, especially the performance of the accelerator, is not fully exploited in many cases. Although the latter approach can bring out the performance, it has not been widely implemented due to the lack of sufficient personnel to perform the optimization and the economic cost of using it. Therefore, bridging the gap between these two approaches is becoming increasingly important.
This research group aims to construct code optimization support technology that enables users themselves to perform performance optimization without specialized knowledge of high performance computing systems, targeting user code for which automatic optimization by compilers has not sufficiently brought out the performance of high performance computing systems.