GPUDWT: Discrete Wavelet Transform accelerated on GPU

Discrete Wavelet Transform (DWT) is a broadly used digital signal processing technique with application in diverse areas. GPUDWT library implements forward and revers DWT 9/7 and 5/3 transforms and supports CUDA platform (version 3.2 and newer) and NVIDIA GPUs.

Performance results

On average a HD-sized (1920x1080) image can be transformed within 0.199ms using NVIDIA GeForce GTX 580 GPU.

Results for single color HD (1920x1080) 8bit image:

The times listed are in microseconds (μs) for one-level and three-levels transformations.

GPU	Transformation type	Avg. time (µs) (1 level / 3 levels)
GeForce GTX 580	Forward DWT 9/7	199/351
GeForce GTX 580	Reverse DWT 9/7	278/453
GeForce GTX 580	Forward DWT 5/3	173/277
GeForce GTX 580	Reverse DWT 5/3	190/315
GeForce 8600 GT	Forward DWT 9/7	5113/7790
GeForce 8600 GT	Reverse DWT 9/7	6439/9310
GeForce 8600 GT	Forward DWT 5/3	3877/5773
GeForce 8600 GT	Reverse DWT 5/3	4281/6301

Installation / Compilation

Download source from http://code.google.com/p/gpudwt/downloads/list
Unpack

$ tar xfvz gpudwt-XXX.tar.gz

Change current directory

$ cd gpudwt-XXX

Edit 'Makefile' so that 'CUDA_INSTALL_PATH' points to CUDA toolkit directory
Run 'make'

$ make

To get help run

$ ./gpudwt --help

References

Early gpudwt design was described in paper GPU-Based DWT Acceleration for JPEG2000 (http://www.sitola.cz/papers/858064.pdf).

Bibtex entry:

@inproceedings{858064,
   author = {Matela, Jiří},
   title = {{GPU-Based DWT Acceleration for JPEG2000}},
   booktitle = {MEMICS 2009 Proceedings},
   year = {2009},
   pages = {136-143},
   address = {Brno},
   isbn = {978-80-87342-04-6}
}

Note, we have not described the current design in any paper yet.

Acknowledgments

This project has been supported by a research intent Optical Network of National Research and Its New Applications (MSM 6383917201) (http://www.cesnet.cz) and Parallel and Distributed Systems (MSM 0021622419).