Why choose STORMM?

A Workbench for Implementing Molecular Models

STORMM provides a rich C++ framework for reading molecular coordinates and model parameters (topologies), manipulating their class objects with built-in routines, and parsing them into work units that distribute the work over a wide GPU. A versatile API offers simple commands for exchanging information between systems, on the CPU host or GPU device.

A New Molecular Dynamics Package

There are many codes for molecular dynamics with deep histories, all of which have their notable accomplishments. However, in most cases the histories of these codes predate the Graphics Processing Unit (GPU) technology that has proven so amenable to classical particle simulations, and as a consequence these codes build on designs which may be as much as forty years old. Even in the past ten years, GPU architectures have grown so rapidly that what used to be considered medium to large problems are not big enough to utilize the entirety of the most powerful cards.

Unlike other codes, STORMM stages multiple calculations in the same program instance, scaling to any size of card, present and future, by batching the problems. The required commonalities among the calculations are minimal, such as that they all take similar boundary conditions or the same thermostating method and time step.

Built-in Facilities for Taking Your Algorithms to the GPU

Without calling an external library like Thrust, STORMM provides a single Hybrid class for managing memory on the CPU host and the GPU device. Each Hybrid object can be allocated and resized much like a C++ Standard Template Library vector and creates space for data on one or both resources in a format which can be specified by the developer, including options for the best data transfer rates, maximum CPU memory space, and automatic synchronization. A Hybrid class object can also operate as the holder of its own data or as a pointer to existing data in another Hybrid object, formalizing C programming's pointer/array duality with an enumerator setting and allowing more agile manipulation of GPU data in the face of high-latency memory allocation for the underlying device architecture.

All implementations require testing to validate, and to build the upon existing features without breaking established methods. STORMM offers a built-in unit testing feature reminiscent of catch2. This unit testing library offers a means to compare single values or arrays of values and analyze failures, offering clues such as a miss scaled array indicating a misstated constant. The checks include methods for confirming that certain code statements or inputs raise exceptions.

Why Open-Source?

Psivant Therapeutics releases STORMM as free and open-source software to bring together a community and advance molecular science. Marketing the softare does not hold as much value as running an active development project that can invite collaborations with academics or other experts based on mutual goals.

Python Integration On the Horizon

In a future release, all coordinate and topology class objects will have the option to export or import their information to and from Python objects, connecting STORMM to a world of computer-aided chemistry.