[Stepanov] Already well along at SGI. Matt Austern's Dagstuhl lecture reported considerable progress. See also NUMA iterators.
[Stepanov]
[Stepanov]
[Stepanov]
[Stepanov]
[Stepanov] Dave Musser and Gor Nishanov have essentially solved this problem, with a fast generic sequence searching algorithm obtained by combining the Knuth-Morris-Pratt algorithm with a hashed version of Boyer and Moore's skip loop.
[Musser and Stepanov] This project is almost complete, with Musser's introsort algorithm replacing the original quicksort implementation of sort, and the Musser-Nishanov algorithm replacing the original straightforward sequence search algorithm. The only remaining algorithm with an O(N2) worst-case time bound is the Hoare find algorithm used to implement nth_element. This can be replaced by some variant of Musser's introselect algorithm, but some experimentation remains to be done.
[Stepanov] For example, a new implementation of stable_sort should be developed that works on forward iterators and improves its cache behavior by using binary-counter instead of parallel-reduction merging.
[Stepanov]
[Stepanov] Examples:
[Stepanov]
[Stepanov] This assumes a solution for the NUMA-iterator problem.