The process of performance tuning is time consuming and costly even if it is carried out automatically. It is crucial to learn from the experience of experts. Our long-term goal is to construct a database of facts extracted from specific performance tuning histories of computation-intensive applications such that we can search the database for promising optimization patterns that fit a given kernel. In this study, as a significant step toward our goal, we explored a thousand computation-intensive applications in terms of the distribution of kernel classes, each of which is related to expected efficiency and specific tuning patterns. To statistically estimate the distribution of the kernel classes, 100 loops were randomly sampled and then manually classified by experienced performance engineers. The result indicates that 50-70% of the kernels are memory-bound and hence difficult to run efficiently on modern scalar processors. In addition, based on the classification results, we constructed experimental classifiers for identifying loop kernels and for predicting kernel classes, which achieved cross-validated classification accuracy of 81% and 65%, respectively.
Reference
Masatomo Hashimoto, Masaaki Terai, Toshiyuki Maeda and Kazuo Minami. An Empirical Study of Computation-Intensive Loops for Identifying and Classifying Loop Kernels. In Proceedings of the 8th ACM/SPEC International Conference on Performance Engineering (ICPE’17), 2017.