Clustering method for the identification of convex disconnected Multiple Input Multiple Output instructions

The instruction-set extensions problem has been one of the topics which has became popular over the last years and it entails the addition of a set of new complex instructions to a given instruction-set. In this paper, we present an efficient algorithm for the automatic generation of convex multiple input multiple output instructions (MIMOs). The proposed algorithm is based on a two-step method which generates a coverage of the application with single output clusters of instructions and subsequently generates an optimal set of complex MIMO instructions to implement in hardware through an integer linear programming (ILP) formulation of the instruction selection problem. The proposed approach can be applied directly to large kernels and does not impose limitations neither on the number of inputs and/or outputs, nor on the number of new instructions generated. Our results on four well known kernels show that the extended instruction-set allows to execute applications more efficiently and needing fewer cycles. On average between 50% and 70% of the initial execution cycles can be saved depending on the kernel and the reconfigurable platform addressed.