It's fairly obvious: the rewrite prevents parallelization because floating-point isn't associative. You'd need to parallelize it explicitly (which can be done

It's fairly obvious: the rewrite prevents parallelization because floating-point isn't associative. You'd need to parallelize it explicitly (which can be done by just unrolling the loop).