Assignments for the Parallel Algorithms course, taken in Fall 2019. This repository consists of two parts,

1. An initial assignment primes where we develop a parallel prime number sieve.
2. A final assignment ccl, where we develop a parallel algorithm for connected components labelling (CCL) in a binary, sparse 3D image.

In /primes, we provide a sequential and parallel implementation of the sieve of Eratosthenes. Several branches relate to this, as follows,

• baseline contains the initial implementation of our sequential and parallel sieves, without any further enhancements.

• odd-k2 contains an initial search-space reduction, by considering only odd primes (and two, the only even prime).

• six-k contains an optimisation where we limit the search space even further.

• twin-primes contains code for generating only twin primes, that is, primes that surround an even number k as (k - 1, k + 1).

• goldbach contains code for verifying the Goldbach conjecture in parallel.

Finally, we refer the reader to the paper in primes/report.pdf, which explains the various algorithms in considerably more detail.

In /ccl, we provide a sequential and parallel implementation of a connected component labelling (CCL) algorithm. All relevant code is in the master branch. We refer the reader to the paper in ccl/report.pdf, which explains the various algorithms in considerably more detail.

Regarding development,

• For primes, there was not really any one code style, though an attempt was certainly made to be reasonable. For ccl, the style is enforced by clang-format - and a Travis check.

• CMake is used as a build tool.

• Unity is our testing framework.

Finally, you will also need on your machine,

• MulticoreBSP-for-C as it handles the nitty-gritty details of parallel communication.

First run e.g. cmake primes and make. This produces an executable, which must be called with positive integer argument: the upper bound for the sieve (exclusive). Additionally, several optional arguments are available:

• l, a value for the lower bound. When specified, the sieve operates over the interval [lower bound, upper bound), rather than [0, upper bound).
• p, a value for the number of processors to use. When left unspecified, this defaults to the maximal number available.

Thus, bin/primes 100 -l 20 -p 3 would run the parallel sieve over the interval [20, 100), dividing the work between three processors.

First run e.g. cmake ccl and make. This produces an executable, which must be called with two file arguments: the input and output files. Some example input files are given in /ccl/examples, with the .mat extension. The first line indicates the number of non-zeroes, each subsequent line a tuple of x y z indices. Several optional arguments are available:

• p, a value for the number of processors to use. When left unspecified, this defaults to the maximal number available.

• s, run the sequential algorithm, rather than the parallel one. If both p and s are passed, s takes precedence.

Thus, bin/ccl ccl/examples/hilbert2.mat hilbert2.ccl -p 4 would run the parallel algorithm with four processes on ccl/examples/hilbert2.mat, and output the result to hilbert2.ccl.