Overflow Controlled SIMD Arithmetic

Abstract

Although the ”SIMD within a register” parallel architectures have existed for almost 10 years, the automatic optimizations for such architectures are not well developed yet. Since most optimizations for SIMD architectures are transplanted from traditional vectorization techniques, many special features of SIMD architectures, such as packed operations, have not been thoroughly considered. As operands are tightly packed within a register, there is no spare space to indicate overflow. To maintain the accuracy of automatic SIMDized programs, the operands should be unpacked to preserve enough space for interim overflow. By doing this, great overhead would be introduced. Furthermore, the instructions for handling interim overflows can sometimes prevent other optimizations. In this paper, a new technique, OCSA (overflow controlled SIMD arithmetic), is proposed to reduce the negative effects caused by interim overflow handling and eliminate the interference of interim overflows. We have applied our algorithm to the multimedia benchmarks of Berkeley. The experimental results show that the OCSA algorithm can significantly improve the performance of ADPCM-Decoder (110%), MESA-Reflect (113%) and DJVU-Encoder (106%).

Supported by the National Natural Science Foundation of China under Grant No. 60273046; Shanghai Science and Technology Committee of China Key Project Funding (02JC14013).