Memory access profiling for HEVC encoders

Abstract

In this work we present a framework that profiles HEVC (High Efficiency Video Coding) encoders modules focusing on cache memory performance and energy. This framework considers the HEVC reference software (HM) and analyzes the impact of some coding parameters on the cache hierarchy. HEVC was proposed in 2013, presenting new video coding techniques to deal with the demand for higher resolutions. The tools included in the HEVC increase significantly the computational effort and energy consumption required to encode videos when compared to its predecessor, H.264/AVC. For this analysis we used the proposed framework MAP-HEVC (Memory Access Profiling for HEVC) considering seven different video coding configurations and four video resolutions. In the prediction module, the results showed that using Full Search (FS) results in 75% more accesses than using Test Zonal (TZ) Search. The results also suggest that using 16×16 search ranges is a very viable option to reduce the memory accesses achieving a better compression. For the residual coding, the Rate-Distortion Optimized Quantization (RDOQ) was evaluated and our analysis showed that this tool adds 10% more accesses to the memory.