ELEC1601 Week 6

Introduction to Computer Systems

The University of Sydney

 Lecture 6

Transistors RAMs/
Gates ROMs Actuators
ALUs
Flip-Flops/ Computer Sensors
Compilers
Registers Architecture Bus

Real-world
Ones and Zeros Words Instructions Programs
interactions

FSMs Pipelining Software Calibration

Floating
Combinational Point Memory/
Digital Logic Fixed Point Stack Subroutines
Addressing High-level
Propositional Binary Modes Constructs
Logic ISA/
Assembly
Machine
Code Simulation
Code
The University of Sydney
Outline:
– Lab 3
– Complete Floating-point from Week 5
– Memory

The University of Sydney Page 3

Floating-point
– What do you need to know?
– How to read understand a floating-point description
• What is Sign, Mantissa, Exponent
• What is an implicit 1
• What is Overflow/Underflow
– What are the characteristics of floating-point vs fixed-point
• What is the relationship between bits, range, precision and rounding error for
floating-point numbers vs fixed-point numbers
• How to perform floating-point arithmetic
– When to use floating-pont.

The University of Sydney Page 4

Floating-point
– What do you need to know?
– How to read understand a floating-point description
• What is Sign, Mantissa, Exponent
• What is an implicit 1
• What is Overflow/Underflow
– What are the characteristics of floating-point vs fixed-point
• What is the relationship between bits, range, precision and rounding error for
floating-point numbers vs fixed-point numbers
• How to perform floating-point arithmetic
– When to use floating-pont.

The University of Sydney Page 5

What is Floating-point?
– A number system with three key components:
– Sign
– Mantissa
– Exponent

– You need to read the definition of the number system to

understand what it means.
– Best understood by example

The University of Sydney Page 6

Floating-point example 3
– An 8-bit number format is:
– 1 sign bit, 4 mantissa bits, 3 exponent bits
– There is is an exponent bias of -2 with the exponent in unsigned binary
– There is an implicit one for the mantissa in the encoding (1.mantissa)

– Convert 11010100 to decimal, use as many decimal places as

necessary
– What is its absolute maximum, minimum. What is it’s worst-case
round-off error?

The University of Sydney Page 7

Why the implicit one
– A 16-bit number format is:
– 1 sign bit, 10 mantissa bits, 5 exponent bits
– Two’s complement exponent, no bias
– There is no implicit one for the mantissa in the encoding (0.mantissa)

– How do you represent 7.125 in floating point?

The University of Sydney Page 8

Why the implicit one
– A 16-bit number format is:
– 1 sign bit, 10 mantissa bits, 5 exponent bits
– Two’s complement exponent, no bias
– There is no implicit one for the mantissa in the encoding (0.mantissa)

– How do you represent 16.625 in floating point?

The University of Sydney Page 9

Why the implicit one
– How many bits did you need to accurately represent 7.125?
– How many bits did you need to accurately represent 16.625?

– Is there any other way you could have represented either

number?

The University of Sydney Page 10

Floating-point design example
– You must represent numbers in the range [0 127].
– Design an 8-bit number format for this range:

– How do you represent the integer 117 in this number system?

– What is the maximum and minimum value you can represent?
– What is the worst-case round-off error?

The University of Sydney Page 11

Why should I learn custom floating-point?
– IEEE 754 Standard Floating Point Single precision:

– IEEE 754 Standard Floating Point Double precision

The University of Sydney Page 12

Why should I learn custom floating-point?
– Single precision:

– Half precision

The University of Sydney Page 13

Why should I learn custom floating-point?
– IEEE standard single precision
– Implicit leading 1
• Except for subnormals…
– Exponent is unsigned with a bias of -127
– Special codes:
• Zero: exponent = 0, mantissa=0
• NaN (0/0, sqrt(-10))
• Infinity (100/0, -100/0)

The University of Sydney Page 14

Why should I learn custom floating-point?

The University of Sydney Page 15

Overflow/Underflow
– Overflow: When a number is too large to be represented.
– What causes this?
– Underflow: When a number is too small to be represented.
– What causes this?

The University of Sydney Page 16

Floating-point
– What do you need to know?
– How to read understand a floating-point description
• What is Sign, Mantissa, Exponent
• What is an implicit 1
• What is Overflow/Underflow
– What are the characteristics of floating-point vs fixed-point
• What is the relationship between bits, range, precision and rounding error for
floating-point numbers vs fixed-point numbers
• How to perform floating-point arithmetic
– When to use floating-pont.

The University of Sydney Page 17

Floating-point vs Fixed-point
– Let’s start with an example.

– You have an 5 bits.

– You can design a fixed or floating-point number system.
– What is the range/precision of each number system?
– What is the round-off error associated with each number system?

– You want to represent numbers from 0 to 7.

– Your floating point is of the form: x  2e  0.b1b2 ...bm
– The exponent is unsigned with no bias

The University of Sydney Page 18

Fixed vs Float a visualisation
5 bit fixed point, 3 bits integer, 2 bits fractional
Roundoff Overflow

Number system must not overflow (error unbounded)

Accumulation of round-off errors must lie below tolerable region

x  2e  0.b1b2 ...bm , bi  {0,1}

5 bit floating point, 2 bits exponent, 3 bits mantissa

The University of Sydney Page 19

Floating-point
– What do you need to know?
– How to read understand a floating-point description
• What is Sign, Mantissa, Exponent
• What is an implicit 1
• What is Overflow/Underflow
– What are the characteristics of floating-point vs fixed-point
• What is the relationship between bits, range, precision and rounding error for
floating-point numbers vs fixed-point numbers
• How to perform floating-point arithmetic
– When to use floating-pont.

The University of Sydney Page 20

Multiplication in floating-point
– Consider some examples:
– 1 sign-bit, 3 exponent bits (unsigned), 4 mantissa bits.
– We assume there is an implicit 1

The University of Sydney Page 21

Addition in floating-point
– Consider some examples:
– 1 sign-bit, 3 exponent bits (unsigned), 4 mantissa bits.
– We assume no implicit 1

– How do you add 0 100 1101 + 0 001 1001?

The University of Sydney Page 22

How do you add numbers in FP
Evaluate 3+0.625+0.875, with 3 bit mantissa
– Compute 3+0.625 – Compute 3.5+0.875.
x  2e  0.b1b2 ...bm
– 3 is 22x0.110 – 3.5is 22x0.111
– 0.625 is 20x0.101 – 0.875 is 20x0.111

– Align exponents – Align exponents

– 3 is 22x0.110 – 3.5 is 22x0.111
– 0.625 is 22x0.00101 – 0.875 is 22x0.00111
– Perform addition – Perform addition
– 22x0.11101 – 22x1.00011
– Re-normalise to 5 bits – Re-normalise to 5 bits
– 22x0.111 = 3.5 – 23x0.100 = 4
– Roundoff error of 0.125 – Roundoff error of 0.375

The University of Sydney Page 23

How do you add numbers in FP
Evaluate 0.625+0.875+3, with 3 bit mantissa
– Compute 0.625+0.875 – Compute 1.5+3
x  2e  0.b1b2 ...bm
– 0.625 is 20x0.101 – Result is 21x0.110
– 0.875 is 20x0.111 – 3 is 22x0.110

– Align exponents – Align exponents

– 0.625 is 20x0.101 – 3 is 22x0.110
– 0.875 is 20x0.111 – 0.875 is 22x0.0110
– Perform addition – Perform addition
– 20x1.100 – 22x1.001
– Re-normalise to 5 bits – Re-normalise to 5 bits
– 21x0.110 = 1.5 – 23x0.101 = 5
– No roundoff error – Roundoff error of 0.5

The University of Sydney Page 24

Floating-point
– What do you need to know?
– How to read understand a floating-point description
• What is Sign, Mantissa, Exponent
• What is an implicit 1
• What is Overflow/Underflow
– What are the characteristics of floating-point vs fixed-point
• What is the relationship between bits, range, precision and rounding error for
floating-point numbers vs fixed-point numbers
• How to perform floating-point arithmetic
– When to use floating-pont.

The University of Sydney Page 25

Encoding Real Numbers
– Do we need floating-point to encode real numbers ?

The University of Sydney Page 26

Why Floating-point?
– You want to represent a wide range with a fixed number of
bits?

– Reason for floating-point: You want to represent a variable

range with a fixed number of bits

The University of Sydney Page 27

Example:
– Your memory can store 16 bits.
– Your algorithm involves some calculations with many
variables
– Variables include numbers as large as 30000.
– Your algorithm has an exit condition var1-var2>1e-6
– You need more fixed-point bits than you can handle

The University of Sydney Page 28

Bisection algorithm

The University of Sydney Page 29

Why not use floating-point?

The University of Sydney Page 30

A bonus – floating-point games

The University of Sydney Page 31

Floating point is not associative
1e 7

i
i 1
Which answer is correct?

The University of Sydney Page 32

Floating point is not associative

The University of Sydney Page 33

Floating point games

2 25

 1
2  25

i 1

The University of Sydney Page 34

 Memory

The University of Sydney Page 35

Memory
– What do you need to know?
– Why is memory necessary
– Main types of memory in a computer architecture
• Registers
• RAM (ROM)
– How to visualize a RAM/ROM
• What is the relationship between total data storage, number/size of
cells/number of address bits
– How to store data in memory
• Reading/writing
• Endianness
• Indirection
The University of Sydney Page 36
Memory
– What do you need to know?
– Why is memory necessary
– Main types of memory in a computer architecture
• Registers
• RAM (ROM)
– How to visualize a RAM/ROM
• What is the relationship between total data storage, number/size of
cells/number of address bits
– How to store data in memory
• Reading/writing
• Endianness
• Indirection
The University of Sydney Page 37
Why/What memory
– Every computer has memory

– Memory operations:
– Read
– Write

The University of Sydney Page 38

Memory
– What do you need to know?
– Why is memory necessary
– Main types of memory in a computer architecture
• Registers
• RAM (ROM)
– How to visualize a RAM/ROM
• What is the relationship between total data storage, number/size of
cells/number of address bits
– How to store data in memory
• Reading/writing
• Endianness
• Indirection
The University of Sydney Page 39
What is a register?
– Remember a D-flip flop?
D Q

– A register is just a collection of D-Flip-flops

– It allows you to represent “words” (your encodings e.g. 4-bit
unsigned binary)

The University of Sydney Page 40

What is a RAM (ROM)?
– A denser (but slower) way of storing data

N
Address Array

Data

The University of Sydney Page 41

Memory
– What do you need to know?
– Why is memory necessary
– Main types of memory in a computer architecture
• Registers
• RAM (ROM)
– How to visualize a RAM/ROM
• What is the relationship between total data storage, number/size of
cells/number of address bits
– How to store data in memory
• Reading/writing
• Endianness
• Indirection
The University of Sydney Page 42
What does a memory conceptually look like?
2:4
Decoder bitline2 bitline1 bitline0
wordline3
N 11
Address Array 2 stored stored stored
Address bit = 0 bit = 1 bit = 0
wordline2
10
stored stored stored
wordline1 bit = 1 bit = 0 bit = 0
01
M stored stored stored
bit = 1 bit = 1 bit = 0
wordline0
Data 00
stored stored stored
bit = 0 bit = 1 bit = 1

Data2 Data1 Data0

The University of Sydney Page 43

What does a memory conceptually look like?

DRAM bit cell:

bitline SRAM bit cell:

wordline
bitline bitline
stored wordline
bit

The University of Sydney Page 44

How does a memory (ROM) work
– (Flash is a variant of ROM) word lines (only one
is active – decoder is
just right for this)

1 1 1 1

n
2 -1

i word[i] = 0011
decoder
j word[j] = 1010

0 n-1
Address
bit lines (normally pulled to 1 through
resistor – selectively connected to 0
by word line controlled switches)

The University of Sydney Page 45

Memory
– What do you need to know?
– Why is memory necessary
– Main types of memory in a computer architecture
• Registers
• RAM (ROM)
– How to visualize a RAM/ROM
• What is the relationship between total data storage, number/size of
cells/number of address bits
– How to store data in memory
• Reading/writing
• Endianness
• Indirection
The University of Sydney Page 47
First – an aside. Some notation
– How many grams in a kilogram?

– How many metres in a kilometre?

– How many bytes in a kilobyte?

– KB: 2^10 Bytes,
– Also
– MB: 2^20 Bytes,
– GB: 2^30 Bytes,
The University of Sydney Page 48
Memory visualisation
– Output can be M bits (often described as A cells of B bytes
such that A*B/8=M)
N
Address Array

Data

The University of Sydney Page 49

Memory visualisation

N
Address Array

Data
– How many cells in a 2GB memory, where each cell is 1 Byte.
– 2^30 *2 = 2^31

The University of Sydney Page 50

Memory visualisation

N
Address Array

Data
– How many cells in a 2GB memory, where each cell is 2 Bytes.
– 2^30 *2 = 2^30

The University of Sydney Page 51

Memory visualisation

N
Address Array

Data
– How much data can a memory with 10 address bits, 1 byte
cells store?
– 2^10
The University of Sydney Page 52
Memory visualisation

N
Address Array

Data
– 10 bit address, 2-byte cells. How many cells are there in the
memory?

The University of Sydney Page 53

Memory Read/Write
– Write 4 bits into 8-bit cell. What happens?

The University of Sydney Page 54

How do we store things in memory

The University of Sydney Page 55

How do we store things in memory?
– What if I want to store a big thing in memory

The University of Sydney Page 56

How do we store things in memory?
– Can a big endian architecture share memory with a little
endian memory?

The University of Sydney Page 57

How to store arrays in memory
– Suppose
Imagine you just get
we have an given
array
the memory on the right.
of integers
– What
A[] = is stored
{10, 267,in39,
it?40}
– How many bits do we need
to store the array?
– What does it look like in
memory?

The University of Sydney Page 58

Design decisions with arrays
– How do you know size of array if stored in memory?
– Memory has no notion of size
– It does not know how many bits are the integers
– It does not know they are storing integers

– Options
– Create a symbol to say end of array
• Typically used for strings (null character)
• Not great for integers (already have a 0)
– Store the size
• First put a number
– Do nothing…

The University of Sydney Page 61

How to find elements in memory
– How do I look up A[5]?
– Address base
+ bytes to encoding array size
+ index * size of element

– E.g. What if look up A[7] in a

array declared of size 6.
– Error message
arrayindexoutofbounds

The University of Sydney Page 62

Working with memory addresses (Indirection)
– Common when you use pointers
(common for structures/arrays):

The University of Sydney Page 63

Working with memory addresses (Indirection)
– Common when you use pointers
(common for structures/arrays):

The University of Sydney Page 64