Parallel implementations of the Cooley-Turkey Fast Fourier Transform (FFT) algorithm in C for my University of Bristol masters unit in Advanced Computational Physics, and as a way to teach myself C.
There is a serial approach along with three parallel approaches. A shared memory approach which uses OpenMP is included, along with a distributed memory approach using MPI. A combination of these approaches is also included, although this is limited to 1D.
The effectiveness of these approaches are investigated using the Blue Crystal 4 Supercomputer.
Data on the speed of approach as a function of dimension, problem size and number of threads/proceses is included, alongside a notebook for guided analysis.
My full report can be found under reports/
- The GNU Scientific Library (GSL) is required for the handling of 2D matrices.
- FFTW is required to test the accuracy in the files under
accuracy/ - OpenMP and OpenMPI are also required for the parallel implementations.
Compilation instructions are included in compilation_instructions.txt.
Instructions are provided for gcc and icc, although any compiler supporting the c99 standard will suffice.