Optimization Of Daala Encoder

Daala is the code-name for a new video compression technology. The effort is a collaboration between Mozilla Foundation, Xiph.Org Foundation and other contributors. The goal of the project is to provide a free to implement, use and distribute digital media format and reference implementation with technical performance superior to h.265.This thesis introduces the principle of Daala codec and new technology used in it. Based on this, we analyze the performance and computational complexity of the encoding algorithm, find the major parts that lead to the highest computation to optimize. The key parts are Lapped Transformed, Sub-pixel interpolation in Motion Compensation,Quantization and the compute of Rate Distortion. We analyzed the modules carefully and proposed data level parallel algorithm to optimize the modules by using Intel SIMD technology. Lapped Transform is composed of pre-filter and DCT, so it is key to optimize DCT. Pixels in Sub-pixel interpolation is independent and Interpolation in it is parallel in data. Multi-group pixels can be interpolated by SSE instruction. There is no data correlation in the process of Quantization and the compute of Rate Distortion. Finally, we test the optimized encoder by video of various resolution.The results show that the proposed SIMD algorithms for those modules could improve the performance of Daala encoding about 50%. The effect of Optimization for this modules is very obvious.