NISTfit: A Natively Multithreaded C++11 Framework for Model Development

The current trend in computer architecture is for increasingly parallel computation while the clock frequency stagnates. The increase in computing speed is achieved by dividing a process into several threads which are executed in parallel on multiple processors, processors with multiple cores, cores that are able to handle multiple threads (hyper-threading), graphical processing units (GPU), or co-processors. In order to take advantage of these new architectures, algorithms that have historically been implemented for serial evaluation need to be refactored for parallelization. In this work, a native multithreading framework in C++11 for scientifc and engineering model development is presented. The motivation for NISTfit is to develop a modern C++11-based library for this problem that is: • Cross-platform: NISTfit has only very minimal header-only dependencies (Eigen and ThreadPool), and builds reliably on all major architectures; a CMake build fle is provided. • Easy to use: There are a multitude of similar libraries for model ftting (e.g., MINPACK1, levmar2, Eigen's LevenbergMarquardt module3, to name but a few) that a) are based on archaic FORTRAN/C/C++ constructs, b) require signifcant boilerplate to solve simple problems, or c) have diffcult-to-build dependencies. It is the opinion of the authors of NISTfit that NISTfit strikes a good balance of power and ease-of-use for simple ftting problems. The code utilizes modern C++11 constructs and will build on any C++11 compliant compiler. • Parallelizable: The future is parallel, and NISTfit is able to achieve near-theoretical speedup as more cores are made available to the ftting for suffciently expensive models. © 2018 National Institute of Standards and Technology. All rights reserved.