Semantic Merge comes to mind, although it only does an AST merge on the structural level while blocks are still merged traditionally (I guess it's a good tradeoff of complexity, though). It has however proven to be rather slow with large merges (or large files; the source file I tried to merge had ~4 kLOC), in my recent test (was merging a major version of one product into derived ones, with lots of refactoring, so I thought SM could help).
