| 赵士彭* ** ***,张立志* ** ***,章隆兵* ** ***.一种基于图形处理器压缩结构的预取结构设计[J].高技术通讯(中文),2022,32(4):351~357 |
| 一种基于图形处理器压缩结构的预取结构设计 |
| A prefetch architecture design based on graphics processor compression architecture |
| |
| DOI： |
| 中文关键词: 图形处理器(GPU)； 访存子系统； 预取结构； 压缩结构 |
| 英文关键词: graphic processing unit (GPU), memory access subsystem, prefetch architecture, compressed architecture |
| 基金项目: |
| 作者 | 单位 | | 赵士彭* ** *** | | | 张立志* ** *** | | | 章隆兵* ** *** | |
|
| 摘要点击次数: 1090 |
| 全文下载次数: 732 |
| 中文摘要: |
| 图形处理器(GPU)访存利用率已经成为影响其性能的关键瓶颈之一。在处理器设计中，访存的预取结构设计成为了提高访存利用率的主要方法之一。结合图形处理器的访存密集的特点，在提高预取性能的前提下，减小影响图形流水线正常效率成为热门的研究方向。本文基于一种图形处理器无损压缩的结构，提出了一套图形处理器的预取结构设计。本预取结构设计可在访存密集型的图形流水线中有效提高访存利用率，并不影响当前图形流水线的效率。实验结果表明，在Godson GPU图形处理器平台上，与传统预取结构相比，针对访存密集型测试程序，cache命中率可以提高15%以上。针对访存空闲的测试程序，该设计不会对流水线产生负面影响。 |
| 英文摘要: |
| Graphics processing unit (GPU) memory access utilization has become one of the key bottlenecks affecting performance. In processor design, memory access prefetch architecture design has become one of the main methods to improve memory access utilization. Combined with graphics processor memory access, due to the dense features, under the premise of improving the prefetch performance, reducing the influence on the normal efficiency of the graphics pipeline has become a popular research direction. Based on a graphics processor lossless compression architecture, this paper proposes a set of graphics processor prefetch architecture design. The design of the prefetch architecture can effectively improve the memory access utilization in the memory-intensive graphics pipeline, and does not affect the efficiency of the current graphics pipeline. The experimental results show that on the Godson graphic processing unit (GSGPU) graphics processor platform, compared with the traditional prefetch architecture, the cache hit rate can be increased by more than 15% for the memory-intensive test program. For the test program with idle memory, it will not have a negative impact on the pipeline. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
