https://github.com/dmitryyurov/bitonic-cuda
An implementation of bitonic search on CUDA
https://github.com/dmitryyurov/bitonic-cuda
cuda gpu-programming sorting-algorithms
Last synced: 9 months ago
JSON representation
An implementation of bitonic search on CUDA
- Host: GitHub
- URL: https://github.com/dmitryyurov/bitonic-cuda
- Owner: DmitryYurov
- License: unlicense
- Created: 2025-08-08T15:32:02.000Z (11 months ago)
- Default Branch: main
- Last Pushed: 2025-08-31T20:23:33.000Z (10 months ago)
- Last Synced: 2025-08-31T21:25:57.769Z (10 months ago)
- Topics: cuda, gpu-programming, sorting-algorithms
- Language: Cuda
- Homepage:
- Size: 39.1 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
[](https://github.com/DmitryYurov/bitonic-cuda/actions/workflows/cmake-single-platform.yml)
# Bitonic Sort
This project implements the [bitonic sorting algorithm](https://en.wikipedia.org/wiki/Bitonic_sorter) using CUDA.
Bitonic sort is a sorting algorithm particularly well-suited for parallel implementation.
## What does this project do?
The `main` function initializes a CUDA device and executes the bitonic sort algorithm on it.
Additionally, it verifies the correctness of the results by comparing them with `std::sort`,
and prints the execution times for both bitonic sort and `std::sort`.
The algorithm sorts a given array of integers in ascending order in two main stages:
1. Initial sorting using shared memory within each thread block
2. Merging sorted sequences using global memory operations (for the arrays which do not fit in shared memory)
Input array size must always be a power of 2 (a limitation of the algorithm).
## Project structure
1. `include` - header files. Currently, only `bitonic.cuh` is provided, which contains the declarations of the bitonic sort
wrapper as well as some helper functions.
2. `src` - source files. `bitonic.cu` contains the implementation of the bitonic sort, while `main.cu` contains
all the utilities necessary for running the sample program.
All the source files and headers are thoroughly documented. Please refer to the built-in documentation in case you want
to investigate the implementation details.
## Requirements
- CMake 3.26 or higher
- CUDA-capable GPU
- CUDA Toolkit
## Building
```bash
mkdir build
cd ./build
cmake -S .. -B .
cmake --build .
```
## Running
If run without arguments, the program will compare the results of bitonic sort with `std::sort` for a number of
array sizes (from `256` to `256 * 2^18`):
```bash
cd ./build
./bitonic_sort
```
An example output for NVIDIA GeForce GTX 1660 Ti and Intel Core i7-9750H CPU is shown below:
```
NVIDIA GeForce GTX 1660 Ti
Device capability: 7.5
Data size | GPU time | CPU time
1 kB 0.25 ms 0.01 ms
2 kB 0.17 ms 0.02 ms
4 kB 0.16 ms 0.04 ms
8 kB 0.20 ms 0.08 ms
16 kB 0.22 ms 0.17 ms
32 kB 0.26 ms 0.36 ms
64 kB 0.29 ms 0.78 ms
128 kB 0.43 ms 1.69 ms
256 kB 0.87 ms 3.64 ms
512 kB 0.82 ms 7.81 ms
1024 kB 1.43 ms 16.56 ms
2048 kB 3.50 ms 35.39 ms
4096 kB 7.00 ms 74.67 ms
8192 kB 14.81 ms 157.83 ms
16384 kB 32.00 ms 326.89 ms
32768 kB 69.85 ms 693.16 ms
65536 kB 152.75 ms 1422.30 ms
131072 kB 282.02 ms 3052.01 ms
262144 kB 748.82 ms 6356.76 ms
```
A user can also specify the size of the array to sort:
```bash
cd ./build
./bitonic_sort 1024
```
In this case, the program will execute the sorting and comparison for a single array with the specified number of
elements:
```
NVIDIA GeForce GTX 1660 Ti
Device capability: 7.5
Data size | GPU time | CPU time
4 kB 0.23 ms 0.04 ms
```