A versatile tomographic forward- and back-projection approach on multi-GPUs

Iterative tomographic reconstruction gets more and more into the focus of interest for x-ray computed tomography as parallel high-performance computing finds its way into compact and affordable computing systems in form of GPU devices. However, when it comes to the point of high-resolution x-ray computed tomography, e. g. measured at synchrotron facilities, the limited memory and bandwidth of such devices are soon stretched to their limits. Especially keeping the core part of tomographic reconstruction, the projectors, both versatile and fast for large datasets is challenging. Therefore, we demonstrate a multi-GPU accelerated forward- and backprojector based on projection matrices and taking advantage of two concepts to distribute large datasets into smaller units. The first concept involves splitting up the volume into chunks of slices perpendicular to the axis of rotation. The result is many perfectly independent tasks which then can be solved by distinct GPU devices. A novel ultrafast precalculation kernel prevents unnecessary data transfers for cone-beam geometries. Datasets with a great number of projections can additionally take advantage of the second concept, a split-up into angular wedges. We demonstrate the portability of our projectors to multiple devices and the associated speedup on a high-resolution liver sample measured at the synchrotron. With our splitting approaches, we gained factors of 3.5 - 3.9 on a system with four and 7.5 - 8.0 with eight GPUs. The computing time for our test example decreased from 23:5 s to 2:94 s in the latter case.