Cache-aware GPU memory scheduling scheme for CT back-projection

Graphic process units (GPUs) are well suited to computing-intensive tasks and are among the fastest solutions to perform Computed Tomography (CT) reconstruction. As previous research shows, the bottleneck of GPU-implementation is not the computational power, but the memory bandwidth. We propose a cache-aware memory-scheduling scheme for the back-projection, which can ensure a better load-balancing between GPU processors and the GPU memory. The proposed reshuffling method can be directly applied on existing GPU-accelerated CT reconstruction pipelines. The experimental results show that our optimization can achieve speedup ranging from 1.18-1.48. Our cache-optimization method is particular effective for low-resolution volumes with high resolution projections.