Software pipelining of nested loops for real-time DSP applications

Modern DSP processors have been integrated with instruction-level parallelism (lLP), which presents a challenge to exploit ILP within DSP applications. Software pipelining is an efficient technique used to expose ILP for loop programs and has been widely used for current microprocessors. It has been also used in DSP compilers, but only for the innermost loops. This paper proposes a new approach which extends software pipelining from innermost loops to whole nested loops in DSP applications. Given a perfect loop, we apply an existing software pipelining approach for the innermost loops, then use the so-called pipelining-dovetailing transformation to extend software pipelining to the outer loops. We also present a transformation to convert a non-perfect nested loop into a perfect one. We have verified the above transformations with some nested loops selected from DSP compiler-challenge C code. The preliminary results are further presented in this paper