OpenCL Image Processing Toolkit (2012)
Linux, C, OpenCL, GTK 3
GPU accelerated image processing application using OpenCL.
This project was started by me and Miłosz Kosobucki as a programming assignment for Image Processing course at Faculty of Mathematics and Computer Science, Adam Mickiewicz University in Poznań, Poland, under the supervision of Mr Wojciech Kowalewski PhD.
Image processing operations run in OpenCL kernels and results are displayed via OpenGL. The UI is made in GTK3.
I was responsible for the main program structure (subsystem initialization, configuration handling, main loop), OpenCL initialization code, a robust OpenGL/OpenCL buffer sharing system and a plugin API. I also wrote a universal GPU-accelerated convolution filter.
To avoid sending big images back and forth through PCI Express bus CLIPT uses buffer sharing between OpenCL and OpenGL introduced in OpenCL 1.1. Internal image format is 32 bits per channel RGB stored as floats. This fact establishes 1GB of VRAM as a reasonable minimum when working with large (4k x 4k) images, since two copies of the image must be stored in GPU memory. When user previews some filter/operation and adjusts parameters in real time, the data is not leaving the GPU. This makes interaction with the program very smooth. Only when the user hits Apply on an effect window, the result of the GPU computation is sent back to system RAM.