TY - GEN
T1 - An inter-frame macroblock schedule for memory access reduction in H.264/AVC bi-directional prediction
AU - Rhee, Chae Eun
AU - Kim, Hyun
AU - Lee, Hyuk Jae
PY - 2012
Y1 - 2012
N2 - For an H.264/AVC encoder, excessive external memory access operations often cause a performance bottleneck because excessive access increases the encoding time as well as the power consumption. In order to reduce the degree of external memory access, this paper proposes a new macroblock (MB) schedule which changes the processing order of MBs across frames. In the proposed algorithm, MBs accessing the same reference data are processed in sequence although they belong to different frames. As a result, the reference data can be used multiple times when they are loaded from external memory. Despite the change in the MB processing order, all prediction operations for the current MB are completed before the next MB is processed. Thus, no loss of compression efficiency is introduced. Simulation results show that the overall memory traffic is reduced by as much as 48.71% for full-HD (1920×1080) video encoded in the PBB Group of Picture (GOP) structure. The saving grows as the number of B frames increases such that traffic is reduced by 70.91% when using the PBBBBB GOP structure.
AB - For an H.264/AVC encoder, excessive external memory access operations often cause a performance bottleneck because excessive access increases the encoding time as well as the power consumption. In order to reduce the degree of external memory access, this paper proposes a new macroblock (MB) schedule which changes the processing order of MBs across frames. In the proposed algorithm, MBs accessing the same reference data are processed in sequence although they belong to different frames. As a result, the reference data can be used multiple times when they are loaded from external memory. Despite the change in the MB processing order, all prediction operations for the current MB are completed before the next MB is processed. Thus, no loss of compression efficiency is introduced. Simulation results show that the overall memory traffic is reduced by as much as 48.71% for full-HD (1920×1080) video encoded in the PBB Group of Picture (GOP) structure. The saving grows as the number of B frames increases such that traffic is reduced by 70.91% when using the PBBBBB GOP structure.
UR - http://www.scopus.com/inward/record.url?scp=84873947696&partnerID=8YFLogxK
U2 - 10.1109/ISOCC.2012.6406918
DO - 10.1109/ISOCC.2012.6406918
M3 - Conference contribution
AN - SCOPUS:84873947696
SN - 9781467329880
T3 - ISOCC 2012 - 2012 International SoC Design Conference
SP - 33
EP - 36
BT - ISOCC 2012 - 2012 International SoC Design Conference
T2 - 2012 International SoC Design Conference, ISOCC 2012
Y2 - 4 November 2012 through 7 November 2012
ER -