Coursework repository for the High Performance Computing course. This course contains a number of C++ based assignments, all oriented towards speeding up the execution of computational code using CPU and GPU (OpenCL) based multi-threading.

• Introduction to tbb::parallel_for

• Using tbb::parallel_for to parallelise a DFT
• Using tbb::task_group to parallelise an FFT

• Introduction to OpenCL programming
• Introduction to GPU specific performance optimisations (reducing memory accesses, read/writes, etc)

• Demonstration of non-trivial parallelisation
• Problem reformulation in order to expose parallelism

• Project consisted of 6 small algorithms

  1. Gaussian Blur: Compute a gaussian blur over an arbitrarily large input image.
  2. Mining (e.g. hashing): Search for m such that hash(m) satisfies a target condition.
  3. Edit Distance: Compute the edit distance betwen two string of bits.
  4. Vector projection: Compute the projection of a vector onto a randomly generated space.
  5. Finite Difference algorithm seen in Coursework 3.
  6. Hold time: Given a set of gates and flipflops, compute the minimum hold time.

• Objective was to get them to run as fast as possible using software or hardware acceleration

• Very large speedups over most of the puzzles, combination of:

  • Algorithmic Improvements (Reduced complexity)
  • Porting the algorithm to OpenCL

• Real-world application of the knowledge gained from the course
• Unknown and complicated codebase (C with a C++ wrapper)
• Objective was to optimise three benchmarks