National Natural Science Foundation of China Key Program (No.61033008);National Natural Science Foundation of China (No.60903041, No.61103080);Ph.D. Programs Foundation of Ministry of Education of China (No.20104307110002);National University of Defense Technology Outstanding Youth Fund (No.CJ11-06-01)
The huge VLIW code size has brought serious problems to the capacity and energy consumption of instruction memory in stream architecture.Through analyzing the characteristics of instruction
this paper proposes a novel domain-divided VLIW compression scheme
designs a distributed on-chip instruction memory for stream architecture and proposes a new SIMD pipeline execution model.The experiment results show that about 38% of the off-chip instruction accessing and 65% of the on-chip instruction memory space demand can be reduced by the compression;the distributed instruction memory depresses about 37% of the area of on-chip instruction memory
and about 8.92% of the area of MASA stream processor.At the same time
the energy consumption of the instruction memory is reduced by 61%.