非线性双曲Schrodinger函数单浮点Cache加速器优化  

Single Floating-point Cache Accelerator Optimization Based on Nonlinear Hyperbolic Schrodinger Function

作　　者：张帆[1] 刘志红[1] 

机构地区：[1]郑州经贸职业学院计算机系,郑州450044

出　　处：《科技通报》2014年第10期1-3,共3页Bulletin of Science and Technology

摘　　要：非线性双曲Schr？dinger函数在计算机编码和通用处理器设计中具有广阔用途，双曲微分方程求解作为单浮点Cache加速器设计的核心算法。传统方法采用稀疏矩阵向量乘方法进行单浮点Cache加速器设计，当微分矩阵的阶数较大时，不能直接进行随机搜索，对具有单浮点数据格式的算法加速性能不好。提出一种基于非线性双曲Schr？dinger函数的单浮点Cache加速器优化设计方法，采用非线性双曲Schr？dinger函数进行矩阵重排序方法，并进行非线性编码，引入稀疏矩阵向量乘单浮点Cache数据结构。采用三级流水线结构设计单浮点Cache加速器，基于Xilinx Virtex-5平台进行数据并行处理性能测试，得出该算法单浮点Cache并行运算加速比比传统方法大1.37~2.60倍，且有优越的数据吞吐性能，稳定性好，在数据库执行算法的并行处理和外部存储器的带宽利用率提高等方面有很好的应用价值。Nonlinear hyperbolic Schr？dinger function has a wide use in the computer code and general processor design, hy-perbolic differential equations is the a core algorithm for single floating-point Cache accelerator design. Traditional method using sparse matrix vector multiplication method for single floating-point Cache accelerator design, when the differential matrix is large, it cannot directly carry out random search, acceleration performance is not good. A kind of single floating-point Cache accelerator optimization method sis designed is proposed base on nonlinear hyperbolic Schr？dinger function, the matrix reordering method is used, and nonlinear coding is taken, the sparse matrix vector multiplication single floating-point Cache data structure is introduced. Three stage pipeline structures are used to design a single floating-point Cache accelerator, data parallel processing performance test is taken based on Xilinx Virtex-5, and it shows that the parallel com-puting speed is more than 1.37-2.60 times comparing to the traditional method, the throughput performance is good with su-perior stability, and it has good application value in the database parallel processing and bandwidth utilization.

关 键 词：非线性双曲方程 Schrodinger函数 单浮点数据 

分 类 号：O246[理学—计算数学]