Vectorized PoCA algorithm for faster image reconstruction for Muon Tomography

Abstract

Muon Tomography is a promising technique for detecting the presence of high atomic number (Z) material, using naturally available cosmic muons. The reconstruction algorithms used in Muon Tomography need to process a considerably high number of muon tracks to reconstruct the image of the object under test. This makes the image reconstruction process computation-intensive and time-consuming. The algorithms to process the muon tracks are event-based algorithms, where all the events are independent and can be processed in any order. The traditional form of parallelism tries to distribute a small segment of work to multiple computing cores, but does not utilize the data level parallelism. The event-based algorithm can be processed on vector architecture and can provide a path to better utilize the data level parallelism. This approach provides another level of parallelism on standard computers without the need for specialized hardware. In this paper, we have tried to exploit the vector architecture provided by the current generation of CPUs to implement the vectorized version of Point of Closest Approach (PoCA) reconstruction algorithm for Muon Tomography. Results are presented using different vector instruction sets, for single-precision and double-precision calculations.