A new parallel algorithm for the prefix minima problem is presented for inputs drawn from the range of integers $[1..s]$. For an input of size n, it runs in $O(\log \log \log s)$ time and $O(n)$ work (which is optimal). A faster algorithm is presented for the special case $s = n$; it runs in $O(\log ^ * n)$ time with optimal work. Both algorithms are for the Priority concurrent-read concurrent-write parallel random access machine (CROW PRAM). A possibly surprising outcome of this work is that, whenever the range of the input is restricted, the prefix minima problem can be solved significantly faster than the $\Omega (\log \log n)$ time lower bound in a decision model of parallel computation, as described by Valiant [SIAM J. Comput., 4 (1975), pp. 348–355]. The top-bottom routing problem, which is an important subproblem of routing wires around a rectangle in two layers, is also considered. It is established that, for parallel (and hence for serial) computation, the problem of top-bottom routing is no harder than the prefix minima problem with $s = n$, thus giving an $O(\log ^ * n)$ time optimal parallel algorithm for the top-bottom routing problem. This is one of the first nontrivial problems to be given an optimal parallel algorithm that runs in sublogarithmic time.

It is established that several natural problems have triply logarithmic, or even faster, optimal parallel algorithms. These problems include: merging two sorted lists, where the values are drawn from a large, but restricted, domain on a CREW PRAM; finding all prefix minima, where the values are drawn from a restricted domain; and top-bottom global routing around a rectangle, a well-investigated problem in VLSI routing for which only highly involved serial algorithms were known.>

The channel-routing of a set of multiterminal nets in the standard two-layer model is considered. The sequential algorithms based on the greedy strategy do not seem to be easily parallelizable. Proposed is an efficient parallel algorithm for routing channels with cyclic constraints. The algorithm runs in time O(n/sup 2//p+log/sup 2/p), with p processors, on a shared memory parallel random access machine (PRAM) model where 1>