CUDA-bitonic-sort

Bitonic Sort implementation using CUDA

There are 3 different versions, v0, v1 and v2. The differences between them are explain in docs/report.pdf.

Compile and run

1. First, clone the repository

git clone [email protected]:NontasBak/CUDA-bitonic-sort.git

2. Make sure you have the CUDA drivers installed and run

make

3. Run any of the 3 versions while also specifying the number of elements in the array as a power of 2 ($2^{power}$)

./bitonic_sort_v0 <power>
./bitonic_sort_v1 <power>
./bitonic_sort_v2 <power>

For example running ./bitonic_sort_v0 23 will run the algorithm with $2^{23}$ elements in total

4. Once you finish, remove the binaries

make clean

Benchmarks

The complete benchmark results can be found in docs/report.pdf.

TL;DR: Compared to a CPU sorting algorithm running in multiple threads, there's a considerable speedup using CUDA, reaching up to 12 times faster.

Author

Bakoulas Epameinondas

Class: Parallel and Distributed Systems with professor Nikolaos Pitsianis

February 2025