Using Heterogeneous Computing for Astronomical Image Subtraction : Improving on X-BACH

This thesis investigates how heterogeneous computing techniques can be used to accelerate astronomical image subtraction. Earlier work such as BACH or it's follow-up project X-BACH only utilize the GPU which leads to performance limitations in some parts of the algorithm. A new version, EMX-BACH, combines OpenMP and OpenCL to utilize both the CPU and the GPU in parallel, addressing previous performance limitations in X-BACH. Benchmarks show that EMX-BACH achieves a relative speedup of 2 over X-BACH and 2 to 6 over the original HOTPANTS implementation, without compromising output accuracy. Some part of the algorithm show speedups of over 25 compared to X-BACH.