John’s Reviews > Parallel and High Performance Computing > Status Update

Skimmed vectorization, threading, and multiprocessing chapters. Multiprocessing and threading are tightly-bound to implementations (OpenMP and MPI) and not very general. Discussion of halo cells / ghost cells buried in OpenMP chapter was interesting.

Verdict on algorithms chapter: wrong level of abstraction. Very concrete: here's an incomplete description of a problem that we solved with spatial hashing in our research, here's the code for the spatial hashing technique, graph of speedup. Preconditions unstated. Poor job conveying general structure or intuition behind techniques. Critical parts of eg parallel prefix-sum buried in diagram and not described in text.

I did like the note on better cache behavior of quadratic probing over double hashing at least

This algorithm for finding neighbors of irregular polygons doesn't work correctly if the edge of a single cell continues without bending while in contact with edges of two more cells (like, say, the eastern edge of Wyoming abutting both Nebraska and South Dakota).

Forward reference to prefix sum, a concept introduced later in the chapter, makes this paragraph on performance characteristics of sorting using spatial hashing require rereading later

"The neighbor list of each cell on each side is just one of the [possibly multiple] neighbor cells and the choice is to be the lower cell or the cell to the left of each pair" - sounds like knowingly sacrificing correctness?

This explanation of hashing and hashtables is atrocious; provides a concrete example of inputs and outputs but doesn't clearly describe the hash function being used, and terminologically unclear around hash vs hashtable.