Design And Implementation Of SIMD Unaligned Memory Access Structure

Multimedia and high-performance applications have been important workloads in general-purpose processors and embedded processors,how quickly and effectively to complete these workloads currently become the main issue.SIMD is an effective approach to exploit data level parallelism,but accessing unaligned data without an efficient unaligned access structure support seriously affect vectorization of the program,and cause a lot of processor performance loss.However,there are certain restrictions on the structure of memory access in most of SIMD architecture processors.These architectures either does not provide hardware support for unaligned memory accesses,either support unaligned memory accesses with great performance penalty.In order to reduce the latency of unaligned memory access,and improve performance of accessing memory in processor,we design and implement the "split buffer line unaligned memory access structure" and "dual Cache memory unaligned memory access structure".The main work and innovations are as follows:1,SIMD unaligned memory access case a lot of performance loss.The existing solutions to SIMD unaligned memory access fall into two categories: software solutions and hardware solutions.In allusion to the long latency of unaligned memory access in existing solutions,we design and implement the "split buffer line unaligned memory access structure" and "dual-bank cache memory unaligned memory access structure".2,This paper models on the features of SPLASH-2 test suite,and test on the "split buffer line unaligned memory access structure" and "dual-bank cache memory unaligned memory access structure" with SPLASH-2.SPLASH-2 experimental results show that the "split buffer line unaligned memory access structure" and "dual-bank cache memory unaligned memory access structure" both can improve memory access performance.Dual-bank cache unaligned memory access structure with respect to the split buffer line unaligned memory access structure has some performance advantages,but the complexity of hardware is also higher.The latency of SIMD unaligned memory has a great impact on program performance,when unaligned memory access latency is more than twice the latency of aligned access,some applications do have a performance degradation.