Accelerated ray-tracing simulations using McXtrace

TL;DR

This paper introduces GPU-accelerated ray-tracing simulations in McXtrace, significantly boosting performance for simulating X-ray instruments and beamlines, with speed-up factors up to 600 times.

Contribution

It presents the implementation of GPU acceleration in McXtrace using openACC and demonstrates substantial speed improvements across various instruments.

Findings

Speed-up factors of 250 to 600 times achieved

GPU acceleration benefits depend on instrument complexity

Memory access optimization is crucial for performance

Abstract

McXtrace is an established Monte Carlo based ray-tracing tool to simulate synchrotron beamlines and X-ray laboratory instruments. This work explains and demonstrates the new capability of GPU-accelerated McXtrace ray-tracing simulations. The openACC implementation is presented, followed by a demonstration of the achieved speed-up factor for several types of instruments across different types of hardware. The instruments achieve speed-up factors around \SIrange{250}{600}{} dependent on the instrument complexity. Instruments requiring repeated memory access might require optimised memory access procedures to avoid severe penalties in the simulation time when using GPUs. The importance of reducing the simulations was demonstrated for an aviation security application by comparing the simulation time of a projection of an energy-dispersive X-ray computed tomography instrument.