I/O系统主要有三种方式来与主机交换数据，其中主要用软件方法来实现，CPU的效率低的是(46)；要有硬件和软件两部分来实现，它利用专门的电路向CPU中的控制器发出 I/O服务请求，控制器则(47)转入执行相应的服务程序的是(48)；主要由硬件来实现，此时高速外设和内存之间进行数据交换(49)的是(50)。供选择的答案

(87)

(85)

(86)

In the following essay, each blank has four choices. Choose the best answer and write down on the answer sheet.

The purpose of the testing phase of software development is not to(71)the absence of errors in the programs being tested, but rather to find as many errors as possible. This statement may seem to contradict one's intuitive notion of a successful test of a piece of(72); however, a successful test is one in which errors have been found, net one in which no errors have been detected.

 There are generally two kinds of testing method: human testing and machine testing. In most companies where systems are developed, the first testing that occurs is not machine testing, but rather the detection of errors through(73), team efforts, referred to as walkthroughs and inspection. With these methods, the author of the program forms a group with others who represent programming expertise or who may be part of the project team.

(84)

(81)

(80)

(82)

(79)

For each blank, choose the best answer from the four choices and write down on the answer sheet.

It is claimed that Al applications have moved from laboratories to the real world. However, conventional von Neumann computers are(66)for Al applications, because they are de- signed mainly for numerical processing. In a larger yon Neumann computer, there is a larger ratio of memory to processing power and consequently it is even less efficient. This(67)remains no matter how fast we make the processor because the length of the computation becomes dominated by the time required to move data between(68)and memory. This is called the yon Neumann(69). The bigger we build the machines, the worse it gets. The way to solve the problem is to diverse from the traditional architectures and to design special ones for Al applications. In the research of future Al architectures, we can take advantages of many(70)or currently emerging concepts. in computer architecture, such as dataflow computation, stack machines, tagging, pipelining, systolic array, multiprocessing, distributed processing, database machines, and inference machines.