ECE 353 ECE 353 Fall 2011 Lab C Pipeline Simulator October 20, 2011.

31
ECE 353 ECE 353 Fall 2011 Lab C Pipeline Simulator October 20, 2011
  • date post

    19-Dec-2015
  • Category

    Documents

  • view

    217
  • download

    0

Transcript of ECE 353 ECE 353 Fall 2011 Lab C Pipeline Simulator October 20, 2011.

ECE 353

ECE 353 Fall 2011Lab C

Pipeline Simulator

October 20, 2011

2ECE 353

Aims of Lab C

Reinforce your understanding of pipelining Provide additional experience in C programming

• Managing queues Introduce you to time-driven simulation

3ECE 353

Outline of Lab

Write a simulator for the MIPS five-stage pipeline (covered in ECE 232) that does the following:• Implements a subset of the instruction set (LW, SW,

BEQ, ADD, SUB, MUL, ADDI) as well as ``quasi-instructions”

• Reads from a file an assembly language program• Simulates, cycle by cycle, the activity in all registers

associated with that program• Displays the values of all the integer registers (except

for register 0) and the PC• Gives the user the option of stepping through the

simulation, cycle by cycle, or just checking the registers at the end of the program and the utilization of each stage

4ECE 353

Types of Simulator

Event-Driven Simulator: Identifies events of interest in the system being simulated, orders them in time, and moves down the ordered list of events, simulating the machine as it goes.• Example: Queuing network simulator for a computer

network. Time-Driven Simulator: Has a central clock and

moves from one clock cycle to the next, simulating all the activity of interest in that clock cycle.• Example: Architecture simulator

5ECE 353

Five-Stage Pipeline

IF Stage: Fetches instructions • If a conditional branch is encountered, the fetch unit stops

fetching until that branch is resolved.• The IF stage places fetched instructions in an instruction

queue, to be consumed by the ID stage.• The instruction queue can hold up to q instructions, where q is

an input to the simulator.• Assume the PC has its own dedicated adder to increment the

byte address by 4 (or word address by 1). ID Stage: Decodes the instructions in the instruction queue,

one by one. • Immediate operands are sign-extended to 32 bits • Makes operands available to the EX stage• Generates control signals

6ECE 353

Five-Stage Pipeline (contd.)

EX Stage:• Executes arithmetic/logical instructions• BEQ instructions are resolved in this stage

MEM Stage:• Carries out access to the data memory

WB Stage:• Writes back into the register file (if necessary)

7ECE 353

Simplifying Assumptions

Separate instruction and data memories; all memory accesses are hits. No need to model cache misses.

There is just one ALU and it is NOT pipelined. It takes • m cycles for MUL• n cycles for all other arithmetic and logic operations

• m and n are input parameters to the simulator• Integer operations only: FP is not implemented

No forwarding is available in this pipeline Ignore all interrupts Register writes are done in the first half of a clock cycle;

register reads in the second

8ECE 353

Interaction between IF and ID stages

IF must not overrun the ID unit: if the instruction queue fills up, IF must stop fetching until there is at least one empty slot in the queue

ID must not overrun the IF unit

9ECE 353

Parsing Instructions

The assembly language program must be read by the simulator• You can use a case statement associated with each

possible instruction, which tells the simulator what to do with that instruction.

When an instruction reaches the ID stage, controls governing the rest of its activity will be generated.

10ECE 353

A Straightforward Approach

Mimic within the simulator what happens in the MIPS pipeline• Generate the control signals associated with each

instruction in the ID stage• Pass these signals along, stage by stage, along with the

instruction.• Don’t forget to check for data hazards• Use a branch_pending signal to guide the IF unit.

Every clock tick, mimic what is supposed to happen in each of the five stages and collect statistics on which stages are doing anything useful

11ECE 353

Maintain Data Memory State

Keep track of the contents of the data memory Each data memory access takes c cycles Some data (as specified by the lab handout) will

already be in the data memory when the program starts• Make any reasonable assumption about the addresses

of your data No virtual memory: assume all addresses are

physical Data and instruction address spaces are separate

12ECE 353

Instruction Memory

The program starts at location 0 of the instruction memory

The PC will point to this location at the beginning of the simulation

All addresses are physical: no virtual addresses (no need for TLBs or page tables)

Each memory access takes c cycles

13ECE 353

Test Program 1

Multiplication of two 10X10 matrices: C=A X B. Starting address of matrices A, B, and C are already in

registers $1, $2, and $3 when execution begins (since you don’t have a virtual memory management system, don’t worry about how these addresses were loaded into these registers): pick any appropriate starting data addresses

The values of A and B for the test program are: A[i][j] = B[i][j] = i*2+j. Initialize your simulator by loading these values into the data memory: should be a separate function called d_initialize.

14ECE 353

Simplifications

To make it easier for you to read in the assembly language program:• The assembly language is redefined to do away with commas;

spaces will do instead.• No register names will be used: register numbers only.

• Example: add 3 2 1 means add register 1 contents to those of register 2 and put the result in register 3.

• The SW and LW instructions are redefined to remove their offset field• Example: lw 5 3 means to load into register 5 from the

memory at the location pointed to by register 3. For 10% extra credit, remove all these limitations and

enable the simulator to accept the standard MIPS assembly language format.

15ECE 353

Simplifications (contd.)

Don’t use labels in your program: instead the beq instruction will identify its target by an offset. • Remember from ECE 232 that the offset is with respect

to the instruction following beq and is in units of words, not bytes

• Example: beq 4 5 -2 means that if the contents of registers 4 and 5 are identical, we branch to the instruction immediately preceding beq.

For 5% extra credit, endow the simulator with the ability to handle instruction labels.

16ECE 353

Reading the Assembly Language Program

C provides a number of ways in which to read input. For example,

fscanf(fp, “%s %d %d %d”, opcode, &field2, &field3, &field4);

where fp is a file pointer, opcode is a character array, and field2, field3, and field4 are integers

17ECE 353

Parsing the Assembly Language Program

Some useful library functions (remember to #include <string.h>):• *strcpy: Copies one string into another• strcmp: Compares two strings. Note that this will give

you an output of 0 if the two strings match.

18ECE 353

Test Program 2

Use quasi-instructions: these don’t do anything other than take up time in the pipeline stages.

A quasi-instruction:• Consumes one cycle to decode and takes one cycle to

pass through each of the MEM and WB stages.• Takes a random amount of time in the EX stage: picks

any of the integers in {a, a+1, …, b} with equal probability and uses that as its execution time.

• Has no impact on the register contents.• Does not suffer from data hazards (since it neither uses

nor produces register contents)

19ECE 353

Generating Random Numbers

Computers are deterministic, so how can we use them to generate random numbers?• Approach 1: Provide random input.

• Example 1: Use time interval between keystrokes.• Example 2: Have the user move the mouse randomly

in a given area and extract numbers from that.• Example 3: Use a device (e.g., vacuum tube) whose

noise characteristic is known and measure the noise to generate the random number

• Approach 2: Generate pseudo-random numbers.• This is the technique we will use

20ECE 353

Pseudo-Random Numbers

These are generated iteratively, from a given seed.• Example: X_n+1 = (aX_n + b) mod M

• X_1, X_2,… are the outputs of the generator• X_0 is the seed, which is given to the generator• a, b and M are integer constants: these determine

how good the generator is.• The above stream of numbers is:

• Clearly NOT random: given X_i, you can predict the rest of the stream

• The stream repeats itself: the cycle cannot be more than M long

21ECE 353

Pseudo-Random Numbers (contd.)

Given that pseudo-random numbers are not random, how can we use them as if they were random?• We check the statistical properties of these numbers.

Some of these are:• Probability distribution function.• Measures of correlation between random numbers

• We can use pseudo-random numbers in simulations if their statistical properties are sufficiently close to a truly random stream

22ECE 353

Pseudo-Random Numbers (contd.)

Random Number Generators (RNGs) are an active area of research: there are many available.

Using a good RNG is often vital to the correctness of the simulation

The C standard library (accessed by including stdlib.h in the header) has a function, rand()• rand() is an integer function, generating values from 0

to RAND_MAX, where RAND_MAX is a constant defined within the library

• It is not the best RNG around, but it is available and easy to use and good enough for our purposes

23ECE 353

Using rand() to pick exec time at random

Quasi-instructions take a random amount of time in the set {a, a+1, …, b} in the EX stage.

rand() generates random integers uniformly distributed over the set {0, 1, …., RAND_MAX}.

Use rand() to generate the execution time of each of the quasi-instructions

24ECE 353

Intermediate Checkpoints

Show a TA the following items: Code for reading the assembly program (by

November 2 – carries deadline-sensitive 5% credit):• Write assembly code to multiply two 10X10 matrices• Demonstrate that your code for reading the assembly

code from a file works correctly Code for Instruction Queue Management: (by

November 8 – carries deadline-sensitive 5% credit):• Interaction between IF and ID units• Handling unresolved branches• Demonstrate that your code works correctly

25ECE 353

Simulator Output

The simulator can be set to operate in either of two modes: single-step or batch

Output consists of • Contents of each of the 31 registers $1 to $31 and that

of the PC (in decimal)• The number of clock cycles that have elapsed• Utilization of each stage IF, ID, EX, MEM, WB.

• In single-step mode, this will be the utilization so far; in batch mode, this will be the utilization for the entire program

• The utilization of a stage is the number of cycles during which that stage did anything useful divided by the total number of cycles

26ECE 353

Lab Report

Description of the structure of your simulator: Include• How the IF stage manages branches and how it knows that

branches have been resolved• How hazards are handled• How instructions are prevented from progressing if there’s a

data or structural hazard• How you implement single-step mode

Source code for your simulator• Fully document your code• Keep the code as well-structured as possible

Matrix multiplication program: For the parameter values specified in the lab document, plot the following:• Assembly language code• Total execution time• Fraction of fetch buffer utilized• Utilization of each pipeline stage

Your strategy for testing the simulator for correctness.

ECE 353

Doug Frazer

Tips from TA

28ECE 353

Tips from TA (Doug Frazer)

Start with the queue:• Write enqueue and dequeue functions• Test them thoroughly against a controlled set of

instructions• Be sure to test things such as:

• Calling dequeue on an empty queue• Handling NULL next and prev pointers on dequeue

and enqueue Handle dependencies on queue Handle different types of MIPS commands Test with MIPS

29ECE 353

Tips from TA

The assembly and C are relatively independent• One person can work on each

If you need to test the C and do not have working assembly, try running it against MIPS from your hardware book.

30ECE 353

Intro to Structures and enums

Generic data structure: struct• Define necessary information to be grouped together

Precompiler ‘labels’: enum• Makes the code easier to read and debug by using BEQ

instead of an integer to represent BEQ• Gets compiled to an integer, first element is 0 and it

increments by 1 unless you specify a different value Useful for this lab:

• Define a struct for each instruction• Define an enum for registers• Define an enum for instruction type

31ECE 353

Enumerated Data Types

enum OPCODE {add, addi, sub, mul, lw, sw, beq}• Defines OPCODE as a data type, taking values just from the

above set• With this definition, you can declare variables to be of type

OPCODE, e.g., enum OPCODE op1, op2, op3;• Internally represents them as integers, starting from 0• Can use equality checks or case statements

• if (op1 == add)• case op2:

You can define enum types for the opcode and the registers You’ll still need to read the instruction opcode from the file as a

character string