Advanced Photon Source

An Office of Science National User Facility

Laue GPU Depth Reconstruction

Laue diffraction microscopy uses the polychromatic Laue micro-diffraction technique to examine the structure of materials with sub-micron spatial resolution in all three dimensions. This GPU-based implementation of a depth reconstruction algorithm for Laue diffraction patterns provides up to a 10 - 15 times speed-up as compared to a CPU-implementation.

Description

The Laue diffraction microscopy uses the polychromatic Laue microdiffraction technique to examine the structure of materials with sub-micron spatial resolution in all three dimensions. The materials that are investigated include inter- and intra-granular orientation distributions in polycrystals, elastic strain tensors in elastically deformed materials, and plastic deformation microstructures under microindents in Cu single crystals. This structural microscopy technique is very powerful for detailed investigation of the microstructure and evolution in materials, especially including local crystallographic orientations, orientation gradients and strains.

Data sets are first collected and stored in HDF5 files. The data is then read by the application, written in C using the CUDA GPU programming library. Individual images are assembled into a 3D cube, and then split for processing on the GPU, usually a few rows at a time, based on available GPU memory. After creating initial data structures in the GPU, the application processes each pixel independently. It first resolves the pixel depth for a given pixel location. It then measures depth along the incident beam from the origin. The intensities are then combined. The result is keep and added to the result of subsequent chucks of data until the entire data set is processed.

This GPU implementation of single-threaded CPU code drastically improves performance by approximately 10 – 15 times using one Tesla M2070 with 512 cores, enabling nearer real-time reconstructions.

Distribution

https://subversion.xray.aps.anl.gov/DepthReconstruction/

Accelerating Laue Depth Reconstruction Algorithm With CUDA, Ke Yue, Nicholas Schwarz, Jonathan Tischler, Proceedings of IEEE High Performance Extreme Computing Conference (HPEC ‘15), Waltham, MA, 9/15/15 - 9/17/15.

Accelerating Laue Depth Reconstruction Algorithm With CUDA, Ke Yue, Nicholas Schwarz, Jonathan Tischler, Proceedings of IEEE Cluster 2015, Chicago, IL, 9/8/15 – 9/11/15.

Acknowledgements

The CUDA Laue Depth Reconstruction software package was developed, and is supported and maintained by Ke Yue in the XSD Scientific Software Engineering & Data Management group (XSD-SDM) in collaboration with Jonathan Tischler, Ruqing Xu, and Wenjun Liu of the APS Surface Scattering and Microdiffraction group with funding from DE-AC02-06CH11357.

Argonne National Laboratory is supported by the Office of Science of the U.S. Department of Energy. The Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit science.energy.gov.