GRS - GPU radix sort for multifield records

We extend the number sorting algorithms on the GPU to sort large multi-field records. We notice that traditional way of sorting the records by first sorting a (key, index) pair to obtain the sorted permutation of the records followed by actually rearranging the entire records to their final position might not actually be the most efficient way to sort them depending on the type of sorting algorithm used and the layouts of the records in the memory.