Indexed Color History Many-Core Engines for Display Stream Compression Decoders

This paper describes and compares 9 many-core designs and software implementations of the Indexed Color History (ICH) module, which is part of VESA Display Stream Compression (DSC) decoders. The smallest design is mapped to only 8 small processors. Other designs use a new algorithm to split the ICH table update process into index update and entry update tasks. This algorithm is implemented with a variety of parallel and optimized architectures to provide a range of throughputs and energy efficiencies utilizing from 9 to 53 processors. The proposed ICH designs deliver frame rates in 1080p (1920×1080) up to 75, 74, and 38 frames per second (fps) in 4:2:0, 4:2:2, and 4:4:4 modes, while dissipating 15 mJ, 16 mJ, and 30 mJ per frame respectively at 1.75 GHz at 1.1 V. Compared to reference designs implemented on an Intel i7-7700HQ, the proposed designs achieve up to 3.4×, 3.9×, and 5.3× higher frame rates, and up to 177×, 193×, and 261× lower energy per frame in 4:2:0, 4:2:2, and 4:4:4 modes respectively.