In crochet, a stitch marker is a mnemonic device used to distinguish important locations on a work in progress. Crochet patterns have a mathematical basis, so stitch markers serve as a visual reference that takes the place of continuous stitch counting and reduces a crocheter's error rate.
Stitch Marker is a collection of threading benchmarks. It measures an abstract machine‘s low-level capabilities for atomic instructions, higher-level waiting, and general scalability under contention. It collects state-of-the-art approaches to locking and lock-free code as well as older approaches, and a few naive ones because in the real world everyone loses a stitch or two from time to time. The point isn’t to compare different lock implementations, or fancy lock-free algorithms, it's rather to compare how they lower onto different (virtual) ISAs.
These tests define the boundary of correctness for the JavaScript SharedArrayBuffer memory model, which WebAssembly mostly re-uses. The tests themselves are probabilistic in that the programs are run and their outcome is validated against a list of known-allowed outcomes. Running the tests many times will, over time, find any implementation flaws.
Running these tests in a loop has attractive properties typical microbenchmarks do not have: it validates that an implementation can perform as fast as possible at the boundary of correctness, but no faster. By definition this microbenchmark cannot be gamed because doing so out cause it to fail.
45245cb293e3c61deb8b952956507b62e88b0cf9Facebook's folly library has multiple classes related to concurrency and parallelism, and benchmark many of them.
8b2c4ea06952084913ae18d6bb66b0a957d96cd4Code and benchmark for a C++ standards committee proposal adding a high-performance semaphore class to C++.
0d1e86d18ee58534bea976489276d75053279100Concurrency primitives, safe memory reclamation mechanisms and non-blocking data structures for the research, design and implementation of high performance concurrent systems.
r220151