Guy Blelloch's PH.D thesis is a bit naive. The young professor back then seems to think that a magic compiler-compiler will turn his parallel-lisp into high-speed algorithms of many different types.

Nonetheless, this PH.D thesis is exceptionally high quality, and serves as a great introduction to which algorithms can be parallelized (and parallelized on a SIMD architecture, like a GPU, to boot), in a simple and easy concept.

The idea of using not only "scan" or "prefix-sum" as a baseline... but also going forth and making more advanced algorithms on top as "segmented-scan", seems to have wide applicability in many fields. From Quicksort, Mergesort, RadixSort, Graph traversals, Graphics Programming, 2d image algorithms and more.

This one PH.d thesis covers so many algorithms using this one methodology that it is clear that the segmented-scan "mindset" is one for GPU-programmers to learn and apply.