MARS使用手册

MIPS

指令、伪指令、

directives

和

syscalls

。

12

、

修改程序，使得

程序运行时提示用户输入菲波那契序列的长度。

11

、

Page 2

?

选择

Ed ition

标签页，返回程序编辑界面。

?

MIP S

注释符号是

#

。一行中

#

后面的所有符号都是注释内容。

?

去掉第

1 2

行到第

19

行代码的注释标志，这几 行用于提示用户输入

菲

波那契序列的 长度，允许输入的长度范围为

2-19

（主要是受

32

位数的

大小限制）。

?

查找帮助文件，找出执行读入整数 功能的

syscall

指令的参数值。

13

、

重新运行程序，观察结果值。

Page 3

Part 2 : MARS Tools

You may have noticed that MARS has a

Tools

menu. The capabilities provided

through this menu really catapult MARS into a different league of computer science

educational software.

We call each of the items in the Tools menu a MARS Tool. A MARS Tool is best

described as a pop-up application that observes MIPS memory and/or register activity

during MIPS program execution then communicates that activity to the tool user to serve

a particular purpose. This is best seen by example.

MARS Tools Activity 1 : Running the Data Cache Simulator tool

1.

2.

Open the program

from the

Examples

folder. This

program will traverse a 16 by 16 element integer matrix in row-major order, assigning

elements the values 0 through 255 in order. It performs the following algorithm:

for (row = 0; row < 16; row++)

for (col = 0; col < 16; col++)

data[row][col] = value++;

3.

Assemble the program.

4.

From the

Tools

menu, select

Data Cache Simulator

. A new frame will

appear in the middle of the screen.

Close any MIPS programs you are currently using.

Page 4

This is a MARS Tool that will simulate the use and performance of cache

memory when the underlying MIPS program executes. Notice its three major

sections:

?

Cache Organization:

You can use the combo boxes to specify

how the cache will be configured for this run. Feel free to explore the

different settings, but the default is fine for now.

?

Cache Performance:

With each memory access during program

execution, the simulator will determine whether or not that access can be

satisfied from cache and update the performance display accordingly.

?

Tool Control:

These buttons perform generic control functions as

described by their labels.

5.

Click the tool's

Connect to MIPS

button. This causes the tool to register

as an observer of MIPS memory and thus respond during program execution.

6.

Back in MARS, adjust the

Run Speed slider

to 30 instructions per second.

It is located at the right side of the toolbar. This slows execution so you can watch

the Cache Performance animation.

7.

In MARS, run the program using the

Run

toolbar button

, the menu

item or keyboard shortcut. Watch the Cache Performance animate as it is updated

with every access to MIPS memory.

8.

What was the final cache hit rate?

_____________. With each miss, a

block of 4 words are written into the cache. In a row-major traversal, matrix elements

are accessed in the same order they are stored in memory. Thus each cache miss is

followed by 3 hits as the next 3 elements are found in the same cache block. This is

followed by another miss when Direct Mapping maps to the next cache block, and the

patterns repeats itself. So 3 of every 4 memory accesses will be resolved in cache.

9.

Given that explanation,

what do you predict the hit rate will be if the block

size is increased from 4 words to 8 words?

______________.

Decreased from 4

words to 2 words?

___________.

10.

Verify your predictions by modifying the block size and re- running the

program from step 7.

NOTE:

when you modify the Cache Organization, the performance values are

automatically reset (you can always use the tool's

Reset

button).

NOTE:

You have to

reset

the MIPS program before you can re-run it.

NOTE:

Feel free to adjust the

Run Speed slider

to maximum speed anytime you

want.

Page 5

11.

Repeat steps 2 through 10 for program

from the

Examples

folder. This program will traverse a 16 by 16 element integer matrix in

column-major order, assigning elements the values 0 through 255 in order. It

performs the following algorithm:

for (col = 0; col < 16; col++)

for (row = 0; row < 16; row++)

data[row][col] = value++;

NOTE:

You can leave the Cache Simulator in place, move it out of the way, or close

it. It will not interfere with the actions needed to open, assemble, or run this new

program and will remain connected to MIPS memory. If you do not close the tool,

then skip steps 4 and 5.

12.

What was the cache performance for this program?

____________. The

problem is the memory locations are now accessed not sequentially as before, but

each access is 16 words beyond the previous one (circularly). With the settings we've

used, no two consecutive memory accesses occur in the same block so every access is

a miss.

13.

Change the block size to 16. Note this will reset the tool.

14.

Create a second instance of the Cache Simulator by once again selecting

Data Cache Simulator

from the

Tools

menu. Adjust the two frames so you can

view both at the same time. Connect the new tool instance to MIPS, change its block

size to 16 and change its number of blocks to 16.

15.

Re-run the program.

What is the cache performance of the original tool

instance?

____________. Block size 16 didn't help because there was still only one

access to each block, the initial miss, before that block was replaced with a new one.

What is the cache performance of the second tool instance?

____________. At this

point, the entire matrix will fit into cache and so once a block is read in it is never

replaced. Only the first access to a block results in a miss.

In what courses might an exercise like this one be useful for your students? I have

used a variation on this exercise as a student exercise in Operating Systems, and for a

lecture illustration of the cache concept in Otterbein's CS 0 course,

Computer Science

Page 6

MARS Tools Activity 2 : Running the Cache Simulator as a stand-

alone

1.

In command mode, traverse to the directory containing and enter

the command:

java -classpath imulator

2.

The cache simulator tool is launched. Its Tool Control section is replaced

by Application Control, which contains additional controls for loading, assembling

and running MIPS programs. It uses MARS' MIPS assembler and runtime simulator

in the background to control MIPS execution.

3.

Click the

Open MIPS program

button and a File Open dialog will pop up.

Browse to and select a MIPS program to run. Select

row_

again if you

wish.

4.

The

Assemble and Run

button is now enabled. Click it to assemble and

run the program. The animation will be very rapid.

5.

Use the

Run Speed slider

to adjust the running speed, click the

Reset

button then click

Assemble and Run

again. While the program is running, the

Stop

button is enabled. Program status is updated in the single line text field.

We plan to implement a small MARS Tool Suite application to simplify the

selection and launching of tools such as the Cache Simulator that are capable of running

Page 7

outside the MARS integrated development environment.

Page 8

MARS Tools Activity 3 : The Memory Reference Visualization tool

1.

Open the program

from the

Examples

folder if it is

not already open.

2.

Assemble the program.

3.

From the

Tools

menu, select

Memory Reference Visualization

. A new

frame will appear in the middle of the screen.

This tool will paint a grid unit each time the corresponding MIPS memory word

is referenced. The base address, the first static data segment (

.data

directive) word,

corresponds to the upper-left grid unit. Address correspondence continues in row-

major order (left to right, then next row down).

The color depends on the number of times the word has been referenced. Black

is 0, blue is 1, green is 2, yellow is 3 and 4, orange is 5 through 9, red is 10 or higher.

View the scale using the tool’s slider control. You can change the color (but not the

reference count) by clicking on the color patch.

4.

Click the tool's

Connect to MIPS

button. This causes the tool to register

as an observer of MIPS memory and thus respond during program execution.

5.

Back in MARS, adjust the

Run Speed slider

to 30 instructions per second.

Page 9

6.

Run the program. Watch the tool animate as it is updated with every

access to MIPS memory.

Feel free to stop the program at any time.

7.

Hopefully you observed that the animation sequence corresponded to the

expected memory access sequence of the program.

If you have

trouble seeing the blue,

reset the tool, move the slider to position 1, change the color

to something brighter, and re-run.

8.

Repeat steps 2 through 7, for

. You should

observe that the animation sequence corresponded to the expected memory access

sequence of this program.

9.

Repeat again for

to observe the animated pattern of

memory references. Adjust the run speed and re-run if necessary.

10.

(Optional)

Create a new instance of the Data Cache Simulator. Move the

two frames around so you can see both. Connect the cache simulator to MIPS and

reset the Memory Reference Visualization. Re-run the program. This exercise

illustrates that two different tools can be used simultaneously.

The Memory Reference Visualization tool could be useful in an operating systems

course to illustrate spatial and temporal locality and memory reference patterns in general.

Page 10

Part 3 : Extending MARS Capabilities

Our session today is not long enough for interactive activities in this area, but we’ll

provide you with enough detail that you can pursue them on your own if you desire.

Abstract

MARS can be customized and extended in four different ways: (1) writing new

MARS Tools, (2) writing new MIPS system calls, (3) writing new MIPS pseudo-

instructions, and (4) writing new MIPS basic instructions. Techniques for all four are

described here.

You have the ability to extend and customize certain MARS capabilities to make it

more useful in your courses. This document describes four different techniques for

extending MARS capabilities:

1.

Ability to write MARS Tools for inclusion in the Tools menu and stand-

alone use.

2.

Ability to define and add new system calls for subsequent use by MIPS

programs.

3.

Ability to customize the instruction set by adding, removing or modifying

pseudo (macro) instruction specifications.

4.

Ability to customize the basic instruction set by adding, removing or

modifying basic instruction specifications.

These procedures apply to MARS 3.2.1, released January 2007. Some may be

streamlined in future releases.

The ability to define and plug in new MARS Tools will be used to develop new

learning aids for students in a variety of computer science courses and provide micro-

worlds for assembly language students to target in their projects. Through those tools it

is possible to deeply engage students in both settings.

The ability to modify the set of system calls, basic instructions, and pseudo-

instructions can be used to define a complete instruction set for an alternative RISC

architecture. MARS was not designed with this in mind however, so support is uneven.

For example, instruction syntax analysis and code generation is driven by the example

and template provided with each instruction, whereas lexical analysis such as the

recognition of register names is embedded in program logic and cannot easily be

Page 11

modified. A customized instruction set has to use MIPS conventions for labeling,

register naming, and so forth. Hopefully MARS can be refactored in future releases to

facilitate its use for alternative instruction sets.

Page 12