Design of Tile-based ARGB Image Lossless Compressor and Decompressor

The utilization of lossless compression is an essential technique, and the compression and decompression of images without any loss have consistently remained a prominent subject of research. In the field of chip design, such as GPU and AI, memory access often becomes a bottleneck for system performance. The frequent read and write operations on color buffer data have a significant impact on performance. This paper aims to improve memory access efficiency by reducing the amount of data accessed in the color buffer from the perspective of data compression. However, with the development of the gaming and special effects industries, GPUs need to process images rendered with special effects. Existing algorithms are good at compressing natural images but have poor compatibility and suboptimal compression ratio for this type of image. In this paper, we first design the QOI+ algorithm based on the Quite Ok Image (QOI) algorithm, which is good at dealing with the compression of non-natural images. Then, we propose a fusion optimization algorithm called QOI++, which covers both natural images and synthetic images. We provide a dedicated accelerator for QOI++ compression and decompression, which conforms to the AXI4-Stream interface. It can be selected for compression of tile sizes of 4x4, 8x8, and 16x16. The compression ratio achieved are 48.72% for 4x4 tiles, 41.22% for 8x8 tiles, and 38.60% for 16x16 tiles, with a clock frequency of 400MHz.