Low GPU occupancy approach to fast arithmetic coding in JPEG2000

Arithmetic coding, and especially adaptive MQ-Coding of JPEG2000, is a serial process, which does not match specifics of GPUs as massively parallel processors well. In this paper we study and evaluate several approaches to acceleration of the MQ-Coding using commodity GPU hardware, including our proposal of a new enhanced renormalization procedure. We conclude with a "low occupancy approach" and 5.6---16× average speedup when compared to the state of the art multi-threaded CPU implementations.