@wds15 mentioned the idea of using the parallel versions of std algorithms (mainly for_each, reduce, transform, etc.) instead of OpenMP (actually the parallel versions of std algorithms use OpenMP under the hood) like I did on. [image] stan-dev/math The Stan Math Library is a C++, reverse-mode automatic differentiation library designed to be usable, extensive and extensible, efficient, scalable, stable, portable, and redistributable in order t... I thought this suggestion was impractical because neither g++ nor clang++ implements the parallel versions of std algorithms yet, despite its being part of the C++17 standard. However, it seems as if g++ is in the process of merging Intel’s implementation of the parallel versions of std algorithms, so that raises the possibility that we could just include [image] intel/parallelstl Contribute to intel/parallelstl development by creating an account on GitHub. in the Stan Math sources like Boost and Eigen. Intel’s implementation only requires C++11 and OpenMP 4.0, so we should be good there (except with Apple’s Xcode). I have run one example on Windows with g++-4.9.

Parallel STL

yizhang February 18, 2019, 5:33pm 21

TBB and OpenMP have overlaps but largely targeted toward different goals. It all depends on the application.

Topic		Replies	Views
Proposed parallelism RFC - Stan language bits Developers	14	1316	July 9, 2019
Benefits of parallelization with a threadpool of the Intel TBB Developers	39	6345	October 25, 2019
Historical record on including TBB Developers math	0	902	July 1, 2020
Threaded AD Developers math	39	2540	April 8, 2018
RStan (PyStan) & MPI / GPU Developers features	43	4067	September 24, 2017