The Research on SLP Optimization Technique towards DSP

According to the application of Super word Level Paralleism (SLP) auto-vectorization compiling system in Digital Signal Processing (DSP), due to the specialized functions of DSP frame, such as the specific addressing model, a wide variety of registers, irregular data branch, the obstacle of dependence relation to vectorization non-aligned data or other reasons, which resulted in the compiler can not support SIMD auto-vectorization with the feature of DSP. In order to solve this problem, the SLP auto-vectorization method was studied towards DSP frame. Based on the specialized functions of DSP frame, the instruction analysis and optimization algorithm in SLP auto-vectorization were improved, so as to transform more high-efficiency source program. The experimental results show that the optimization method of this paper has a tremendous role on improving DSP performances and reducing power consumption.