v.2.21.0 (18 October 2019)
We are now using TBB for threading in map_rect. With performance tests on a non-trivial map_rect model we have observed speedups of up to 20% on Windows, 70% on Linux and 30% on MacOS. Speedups were observed for both Intel and AMD CPUs. This has a downstream license effect too - we can no longer be bundled in a single distributed binary with GPLv2 code.
Here are some speedups on a non-trivial model with 12 cores:
New Features
- Add Intel TBB (#2769)
- Add automated maintanence and cleanup via clang-tools to makefile (#2805)
- Use Welford's algorithm for Variance Estimates in src/stan/analyze (#2819)
- Adds additional no-u-turn checks across subtrees to avoid missing u-turns for approximately iid normal models (#2800)
- Add gp kernels to language (#2739)
- Update rhat and ess (#2794)
Bug Fixes
- Fix const logic and size check (#2822)
- Add ifdef for R new_model (#2798)
- Add g_ to name of g vector name in ps_point (#2832)
Documentation
HTML