Program Autotune <2026>

This paper presents a statistical modeling-based approach for autotuning parallel programs. The authors demonstrate their approach on several benchmark applications.

Before 2000, a record label had to find a singer who was perfect. You needed the lung capacity of Whitney Houston or the pitch accuracy of a tuning fork. Autotune changed the economics of music. Suddenly, a producer could hire a vocalist with incredible tone and emotion —even if their pitch drifted slightly—and fix the technical errors in post-production. program autotune

This survey paper provides an overview of autotuning techniques, including machine learning-based approaches, and their applications in high-performance computing. You needed the lung capacity of Whitney Houston

Source: Ghosh, S., Sharma, A., & Banerjee, S. (2018). Autotuning of parallel programs using statistical modeling. Journal of Parallel and Distributed Computing, 114, 33-44. This survey paper provides an overview of autotuning

Program autotuning, also known as automatic performance tuning, is a technique to automatically optimize the performance of a program by adjusting its parameters, such as compiler flags, algorithm choices, or even rewriting parts of the code.