Linear benchmarks
Steve Vestal · ACM SIGAda Ada Letters · 1990
This paper presents a language feature benchmarking technique that is based on the use of multiple sampling loops and linear regression. Multiple performance estimates for multiple parameters can be obtained simultaneously. A heuristic is presented to automatically adjust the number of iterations for the sampling loops. It is also possible to compute values that give some insight into the accuracy of the estimates. This paper gives an overview of how such benchmarks may be coded and discusses some preliminary experiments with this approach.