Using ChatGPT on Improving Program Performance with pprof and Benchmark
Wei-Cheng Lei, Luo-You Jian, Yanwen Chen, Li‐Der Chou · 2023
In the context of limited computing resources, optimizing program architecture is crucial. Therefore, this paper proposes to apply the powerful analytical capabilities of large language models (LLM) to the field of systematic performance optimization. The output of pprof and benchmark is fed into ChatGPT, and the program's performance is improved based on feedback. In the case study, the number of memory allocations for the objective function was successfully reduced from 99 to 1, resulting in a reduction of the test execution time from 8.6 microseconds to 0.36 microseconds. At the same time, the memory allocation was also reduced from 53.5KB to approximately 1KB.