Express an array index reordering as mapping

[bernhardmgruber]

For the LHCB analysis example, presorting the dataset gives a speedup. As an alternative, we could also define this permutation inside a dedicated mapping using extra memory to store an index table. LLAMA should be powerful enough to express this. It would be interesting though to see the effect on the runtime.

My guess: it will probably be worse than the unsorted dataset. The indexing introduces an extra indirection, so more memory needs to be loaded and data access now requires an indirection in-between inhibiting prefetching. Furthermore, since the actually needed data is not grouped together into hot and cold parts, cacheline utilization will stay low.