论文部分内容阅读
这里所说的光计算是指与完成矩阵运算(首先是矩阵—矩阵乘法和矩阵—矢量乘法)有关的计算。众所周知,所有的数值分析首先必须通过采样使连续变量离散化,其结果常归结为线性方程组或矩阵运算。此外,许多工程技术问题的数学模型亦常归结为矩阵运算。鉴于矩阵运算在数值分析中的重要性,80年代初在电子计算机领域内发明了一种节律(Systolic)算法,首次跳出了传统的Von Neumann结构。按照Von Neumann的概念,运算器在进行每次运算前先访问存储器取得待运算数,经运算后再次存入存储器,如此频繁地访问存储器必然成为提高运算速度的瓶颈。随着大规模集成电
The calculation of light here refers to the calculation related to the completion of the matrix operations (above all matrix-matrix multiplication and matrix-vector multiplication). It is well-known that all numerical analysis must first discretize continuous variables by sampling, the result of which is often summarized as a system of linear equations or matrix operations. In addition, many mathematical models of engineering problems are often summarized as matrix operations. In view of the importance of matrix operations in numerical analysis, a Systolic algorithm was invented in the field of electronic computers in the early 1980s, jumping out of the traditional Von Neumann structure for the first time. According to the concept of Von Neumann, the operator accesses the memory to obtain the operand before performing each operation, and then stores it in the memory after being calculated. Therefore, frequently accessing the memory becomes a bottleneck to increase the speed of operation. With large scale integration