BWDSP 是一款针对高性能计算领域设计的处理器,采用多簇超长指令字(VLIW)体系结构和 SIMD 架构,同时也提供了很多向量化指令.然而现有的编译框架无法对这些向量化指令提供支持,因此本文提出了一种向量化优化算法,可以显著提高一些在 DSP 领域有着广泛应用的计算密集型程序的性能.最终实验结果表明,该优化算法能够平均取得6.60倍的加速比.%This paper presents a vectorization algorithm for multi-cluster and VLIW(very long instruction word) DSP and this algorithm can significantly improve the performance of some compute-intensive programs which are widely used in DSP field. Via pre-handling the chain of the instruction and synthesizing the special instruction if needed, at last the algorithm synthesizing the vectorized instruction which the DSP provided. The experimental result shows this vectorization algorithm achieves 6.60 times performance improvement on average.
展开▼