Assignment 15: Matrix Operations and OpenMP

Weeks 13–14 · Nov 30 – Dec 9

Note

Not published yet. This page fills in before week 13 begins. The outline below is what the assignment is being built to cover, a preview, not a specification. Nothing here is final.

This is the only assignment that spans two weeks: it starts in week 13 and concludes in week 14, so budget accordingly.

What it will cover

This assignment closes out the course by taking a data structure you already know and making it faster on hardware you already have, the last of the five course outcomes, use parallel computing techniques to improve algorithm performance and measure the improvement.

By the end of it, you will be able to:

Describe Moore’s law, and why manufacturers stopped raising clock speed
Describe the fork-join paradigm, and where it fits a loop over a matrix
Use #pragma omp parallel for to parallelise a loop
Reason about how memory access patterns affect parallel speed-up
Compare timings across scheduling strategies and justify the difference
Explain the speed-up you measured, including the cases where parallelism did not help

Shape of the work

Matrix operations are the vehicle. They are easy to state, expensive enough to time honestly, and they reward thinking about how memory is walked rather than just how many cores are thrown at the problem.

As with every assignment this term, the measuring and the write-up carry real weight. A correct parallel loop that you cannot explain is not the goal.

Before it opens

Nothing to do yet. When the page fills in, it will follow the same structure as the rest: objectives, getting the code, the parts in order, and what to submit.

In the meantime, the schedule lists the readings and slides for weeks 13 and 14, and Skills & Check-Ins shows which skills this assignment gives you evidence for.