high performance computing on graphics processing units: hgpu.org

In this thesis several steps towards the optimization and acceleration of a Monte Carlo code for the simulation of light propagation in particulate scattering media have been taken. This is performed by parallelizing a Monte Carlo code originally written by E. Berrocal [1] and running the simulation on a modern computer graphic card; a process known as general-purpose computing on graphics processing units (GP-GPU). It is demonstrated that the new Monte Carlo code, can speedup the simulation time by a factor ~40 times, when the calculations are offloaded on the GPU. This acceleration in computational speed is obtained for a scattering medium containing a collection of micrometric droplets at an optical depth of OD = 10. It is also deduced from this study that higher speeding-up of the code is reached at higher OD. This can be explained by the fact that more calculations are, in this case, run on the GPU. A second optimization concerns the improvement of the user interface. The code is now able to handle hundreds of detection setting schemes simultaneously. Input and output files are stored into separate folders. The output folder contains sub-folders where all the output files can be organized as a function of the detection conditions. The resulting matrices can now be directly viewed when the simulation is ending, without the need of plotting them in Matlab. Also, imaging at the Fourier plane has been implemented as a detection option. All these improvements allow more complex calculations and detections schemes to be run during a reasonable time frame. This open-up the possibility of simulating more realistic and challenging cases of study. Thanks to these new features, the improved Monte Carlo code has strong potential to further help the understanding of light scattering in turbid media and the development of modern optical diagnostics.