Foundations of Empirical Memory Consistency Testing
summary
Claims a reproducible, efficient GPU litmus-testing workflow that finds weak-memory behaviors by combining a new OpenCL tool, randomized 737000 parameter sets search space, and an exploratory dataset of 18 million iterations (154 hours). It shows tunable stress is essential, uses data-peeking to cut runtimes (about 130 hours speedup vs full counts), defines scoring metrics to pick robust parameter sets, and proves the method by discovering an actual Intel OpenCL compiler bug that received an official patch.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:26Empirical Testing
- 5:37Exploration
- 9:13Process Tuning
- 11:51Conformance Testing
- 13:56Conclusion
related talks
20mScalable and Accurate Application-Level Crash-Consistency Testing(…)
15mProgramming and Execution Models for Parallel Bounded Exhaustive Testing
55mUsing the Valgrind Framework to Build a Persistent Memory Error Detector
42mSurprise yourself with Property Based Testing
29mHow Hard Is It to Test the GPU Driver?
19mVinter: Automatic Non-Volatile Memory Crash Consistency Testing for Full Systems
15mMimesys: Generating Realistic Executable Testing Environments from Resource Usage Traces
35mBenchmarking: You're Doing It Wrong