Earticle

In this paper, we propose a task-parallel programming extension for the JPEG decoder of the libjpeg-turbo library. Efficient JPEG decoding is especially important for resource-constrained mobile devices such as smartphones, where decoding (e.g., browsing of web pages containing images, image search aso) is far more common than image encoding. The aim of our work is to utilize multiple CPU cores for JPEG decompression from a single client thread. Our method is orthogonal to libjpeg-turbo's support for data-parallelism (SIMD). Experimental evaluation of our approach on a 4-core Intel i7-2600K CPU shows speed-ups of up to 2.5x over the sequential, and up to 34% over the SIMD-version of the libjpeg-turbo JPEG decoder.