Development of a personal accelerator using reconfigurable LSIs.

在这项研究中，我们利用本研究预算购买的台式计算机和CAD工具，研究了一个将C程序翻译成HDL语言程序的编译器，一个用于为FPGA综合HDL语言程序的工具，以及用于图像处理和元胞自动机的硬件库。在硬件编译器方面，我们研究了处理结构和指针的算法，利用更多的程序对编译器进行评估，并通过避免内存块上的访问冲突来实现高效的流水线处理。我们还开发了一种算法，将C程序中的循环自动转换为硬件上的流水线处理。对于综合工具，我们已经开发了一个原型，并使用简单的HDL语言程序对其性能进行了评估。在硬件库方面，我们开发了以下内容。在图像处理方面，我们开发了一种可用于实时立体视觉的一维互相关函数电路。我们还研制了两个二维互相关函数的电路。这两个电路可用于PIV(粒子图像测速)和光流的实时处理。为了检测图像中的目标，我们开发了基于Line Hough变换的直线检测电路和基于广义Hough变换的任意形状检测电路。我们还开发了二维和三维元胞自动机的库。

In this research, we have researched on a compiler which translates C programs to HDL programs, a tool to synthesize the HDL program for FPGAs, and hardware libraries for image processing and cellular automaton using desk top computers and CAD tools which we purchased by the budget of this research. As for the hardware compiler, we have researched on algorithms to handle structures and pointers to evaluate the compiler using more programs, arid to achieve efficient pipeline processing by avoiding access conflicts on memory banks. We have also developed an algorithm to automatically translate loops in C programs to pipeline processing on the hardware. As for the synthesis tool, we have developed a prototype, and evaluated its performance using simple HDL programs. As for the hardware libraries, we have developed the followings. In the image processing, we have developed a circuit for one dimensional cross correlation function which can be used for real-time stereo vision. We have also developed two circuits for two dimensional cross correlation function. These two circuits can be used for real-time processing of PIV (Particle Image Velocimetry) and optical flow. In order to detect objects in images, we have developed circuits to find lines based on Line Hough Transform, and arbitrary shapes based on Generalized Hough Transform. We have also developed libraries for two and three dimensional cellular automata.