No abstract

OpenMM is a software toolkit for performing molecular simulations on a range of high performance computing architectures. It is based on a layered architecture: the lower layers function as a reusable library that can be invoked by any application, while the upper layers form a complete environment for running molecular simulations. The library API hides all hardware-specific dependencies and optimizations from the users and developers of simulation programs: they can be run without modification on any hardware on which the API has been implemented. The current implementations of OpenMM include support for graphics processing units using the OpenCL and CUDA frameworks. In addition, OpenMM was designed to be extensible, so new hardware architectures can be accommodated and new functionality (e.g., energy terms and integrators) can be easily added.

We describe a complete implementation of all-atom protein molecular dynamics running entirely on a graphics processing unit (GPU), including all standard force field terms, integration, constraints, and implicit solvent. We discuss the design of our algorithms and important optimizations needed to fully take advantage of a GPU. We evaluate its performance, and show that it can be more than 700 times faster than a conventional implementation running on a single CPU core.

No abstract

The rigorous application of static timing analysis\ud requires a large and costly amount of detail knowledge on the\ud hardware and software components of the system. Probabilistic\ud Timing Analysis has potential for reducing the weight of that\ud demand. In this paper, we present a sound measurement-based\ud probabilistic timing analysis technique based on Extreme Value\ud Theory. In all the experiments made as part of this work, the\ud timing bounds determined by our technique were less than\ud 15% pessimistic in comparison with the tightest possible bounds\ud obtainable with any probabilistic timing analysis technique.\ud As a point of interest to industrial users, our technique also\ud requires a comparatively low number of measurement runs of\ud the program under analysis; less than 650 runs were needed for\ud the benchmarks presented in this paper