TY - GEN
T1 - Adaptive reduced bit-width Instruction Set Architecture (adapt-rISA)
AU - Neves Soares, Sandro
AU - Halambi, Ashok
AU - Shrivastava, Aviral
AU - Wagner, Flávio Rech
AU - Dutt, Nikil
PY - 2009
Y1 - 2009
N2 - RISA (reduced bit-width Instruction Set Architecture) is an important architectural feature to reduce code size, which continues to be an extremely important concern for low-end embedded systems. rISA reduces code size by expressing parts of the application in terms of low bit-width instructions. ARM-Thumb, ARCcompact and MIPS16/32 are popular examples. With the intent to exploit the dynamically changing working instruction set of today's complex software, ARM 11 now comes with two rISAs, which can be interleaved in the application binary. However, it was demonstrated that the code compression achieved by rISA is extremely sensitive on the selected rISA design. Therefore, it is important to design the optimal rISA for a given embedded application. The one optimal rISA per application approach has already been explored by previous works. In this paper, we present a scheme to design a multiple rISA architecture for embedded systems. Our experiments on MiBench report an average of 19% code compression and up to 7% power reduction of instruction memory when compared to previous approaches using only one optimal rISA.
AB - RISA (reduced bit-width Instruction Set Architecture) is an important architectural feature to reduce code size, which continues to be an extremely important concern for low-end embedded systems. rISA reduces code size by expressing parts of the application in terms of low bit-width instructions. ARM-Thumb, ARCcompact and MIPS16/32 are popular examples. With the intent to exploit the dynamically changing working instruction set of today's complex software, ARM 11 now comes with two rISAs, which can be interleaved in the application binary. However, it was demonstrated that the code compression achieved by rISA is extremely sensitive on the selected rISA design. Therefore, it is important to design the optimal rISA for a given embedded application. The one optimal rISA per application approach has already been explored by previous works. In this paper, we present a scheme to design a multiple rISA architecture for embedded systems. Our experiments on MiBench report an average of 19% code compression and up to 7% power reduction of instruction memory when compared to previous approaches using only one optimal rISA.
KW - Embedded systems design
KW - code compression
KW - power reduction
UR - http://www.scopus.com/inward/record.url?scp=80555154157&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=80555154157&partnerID=8YFLogxK
U2 - 10.1109/VLSISOC.2009.6041329
DO - 10.1109/VLSISOC.2009.6041329
M3 - Conference contribution
AN - SCOPUS:80555154157
SN - 9781457702365
T3 - Proceedings - 17th IFIP International Conference on Very Large Scale Integration, VLSI-SoC 2009
SP - 47
EP - 52
BT - Proceedings - 17th IFIP International Conference on Very Large Scale Integration, VLSI-SoC 2009
PB - IEEE Computer Society
T2 - 17th IFIP International Conference on Very Large Scale Integration, VLSI-SoC 2009
Y2 - 12 October 2009 through 14 October 2009
ER -