Partial Image Decoding On The GPU For Mobile Web Browsers

Abstract

This thesis will investigate how the GPU can be used to facilitate image decoding for mobile web browsers. In the mobile web browsers of today, the majority of computations are performed on the CPU, with only some parts offloaded to the GPU. This thesis will move computations from the CPU, by partially decoding JPEG images on the CPU, and offloading the rest of the process to the GPU. Both the decoding- and encoding process of the JPEG format will be investigated to determine which parts of the decoding process would be suited to be performed on the GPU. The study will show that upsampling and color space transformation, which are the two last steps in the JPEG decoding process, are particularly suited for the GPU. The implementation will be done in the Opera web browser for Android. Opera for Android uses the Chromium back-end for handling decoding and rendering of images. In Chromium the Blink component handles image decoding and the Skia component handles rendering of images. The results of the implementation will, depending on how the encoding of the image was done, show memory savings of up to 62.5.