基于图形处理器的STM研究与实现  

Research and Implementation of Software Transactional Memory Based on Graphics Processor Unite

作　　者：梁飞[1] 胡大裟[1] 蒲亦非[1] 

机构地区：[1]四川大学计算机学院,成都610065

出　　处：《成都工业学院学报》2013年第2期13-17,共5页Journal of Chengdu Technological University

基　　金：国家自然科学基金"分数阶微积分应用于医学核磁共振图像处理的理论与技术"(60972131)

摘　　要：多核处理器和基于图形处理器通用计算(GPGPU)的发展,提出了简化并行编程的需求,而软件事务存储(STM)通过标记代码段并保证其执行的原子性为简化并行编程提供了很好的选择。为降低图形处理器(GPU)并行编程的复杂性,在分析GPU编程中存在的同步问题,结合统一计算设备架构(CUDA)的特点以及影响STM重要因素的基础上,提出在编程环境中引入STM模型的编程方法,测试结果表明相对基于CPU的计算依然具有良好的加速比。The development of multi-core processor and GPGPU （ general purpose computing on graphics processors） creates a demand for ease of parallelization. STM （Software transactional memory ） provides a good choice to simplify the development of concurrent code by allowing the programmer to mark sections of code to be executed atomically. To simplify the relatively complex of parallel programming on GPU （ Graphics Processing Unit）, synchronization problems of GPU programming are analyzed. Based on the comprehensive consideration of significant factors of STM and characteristics of CUDA （Compute Unified Device Architecture）, the introduction of STM in GPU programming environment is proposed and the test results show that speedup ratio sustains well by comparison with computing on CPU.

关 键 词：图形处理器 软件事务存储 通用计算 统一计算设备架构 

分 类 号：TP334.7[自动化与计算机技术—计算机系统结构]