A Hardware-Friendly Compression Algorithm for Profiling DDR4 Memory Accesses
Xuan Truong Nguyen, Ji-Woong Choi, Hyuk‐Jae Lee, Hyun Kim · 2019 International Conference on Electronics, Information, and Communication (ICEIC) · 2019
Profiling or tracking memory accesses is a critical task in analyzing memory behavior and thereby improving memory technology. However, profiling memory accesses is a challenging task, especially from memory side, due to a large bandwidth and a huge memory storage requirement. To address this problem, this paper proposes an efficient compression algorithm for memory trace profile. A single command is firstly represented under a format of variable run length coding. Based on this, simple and efficient scheduling is presented to compress memory trace. Experimental results show that the proposed compression algorithm achieves up to 50.61% memory saving on average with a negligible overhead.