FPBench is a community project to facilitate comparisons and cooperation between tools that help programmers write accurate floating point code.
FPBench currently consists of three parts: a language for expressing benchmarks, a benchmark suite written in this language, and a language for describing ways to measure floating point error. You can find the current draft of the FPBench standards, including FPCore and the measures, here.
If you're interested in using or helping develop FPBench, please consider joining the mailing list (archives). Currently, the best way to access the existing benchmark suite and tools is from the FPBench GitHub repo.
The first FPBench paper was published at NSV’16.