We'll be running a BoF session on Performance tracking and benchmarking infrastructure at the LLVM developer's meeting next week. Similar BoFs in previous years have been very helpful in raising, discussing and refining the best ideas for improving our infrastructure, especially LNT and the test-suite. Good progress has been made in the last year, which will be summarized as part of my presentation on the topic. However, there are still some chunky pieces of work that need to come together before we'll have an infrastructure that consistently produces clear actionable information when changes in the performance of generated code is detected.
We'd like to discuss what improvements or features we still need before we can very efficiently make use of llvm.org/perf specifically and LNT-based performance tracking in general.
With this email we'd like to start collecting specific topics that you'd like to discuss at the BoF, or any insights or experiences you've got with using LNT, the test-suite or other infrastructure to track performance or related metrics.