Memory prefetching for tomography acceleration on FPGAs through HLS tools
Résumé
Backward projection is one of the most timeconsuming steps in method-based iterative reconstruction computed tomography. The 3D back-projection memory access pattern is potentially enough to efficiently exploit the computation power of acceleration boards based on GPU or FPGA. However, exploiting the full potential of these architectures has always been a major concern. Therefore, an algorithm architecture co-design approach is necessary to harness these parallel architectures sufficiently. This paper proposes an OpenCL acceleration of the voxel-driven 3D backprojection algorithm on an Arria 10 FPGA. We perform an offline study of the algorithm memory access pattern to prefetch sinogram data to the on-chip BRAM before performing reconstruction.
Origine : Fichiers produits par l'(les) auteur(s)