Radix-2r Arithmetic for Multiplication by a Constant.

International audienceIn this paper, radix-2r arithmetic is explored to minimize the number of additions in the multiplication by a constant. We provide the formal proof that for an N-bit constant, the maximum number of additions using radix-2r is lower than Dimitrov's estimated upper-bound (2.N/log(N)) using double base number system (DBNS). In comparison to canonical signed digit (CSD) and DBNS, the new radix-2r recoding requires an average of 23.12% and 3.07% less additions for 64-bit constant, respectively