weft / warp

continuous energy monte carlo neutron transport in general geometries on GPUs

WARP

Ryan M. Bergmann, 2014.

WARP is a GPU-powered library for continuous energy Monte Carlo neutron transport in general 3D geometries. The host-side code has been written in C++, CUDA, and Python. The C++ API is exposed to Python via SWIG. The Python C API is used within WARP to load ACE-formatted cross section libraries via the PyNE package. Geometry routines are implemented using the NVIDIA OptiX ray tracing framework. Currently, WARP is only able to run on a single NVIDIA GPU.

If the name "WARP" were an acronym, it would stand for "weaving all the random particles," with the word "weaving" referring to the lock-step way in which "all the random particles," i.e. the neutrons, are sorted into coherent bundles and transported. Using the word "warp" is also a nod to NVIDIA's terminology for a group of 32 concurrent threads.

Please refer to the usage guide for examples of how to use WARP, and to the doxygen documentation for detail descriptions of the API.

Related Publications:

Performance and accuracy of criticality calculations performed using WARP – A framework for continuous energy Monte Carlo neutron transport in general 3D geometries on GPU

Algorithmic choices in WARP – A framework for continuous energy Monte Carlo neutron transport in general 3D geometries on GPUs

Miscellaneous

Logo provided by designer extrodinaire Daniel Castro, http://www.danieldanieldaniel.com/