Transcript PPT

snick

snack
CPSC 121: Models of Computation
2013W2
Number Representation
Steve Wolfman, based on notes by
Patrice Belleville and others
1
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
Outline
• Prereqs, Learning Goals, and Quiz Notes
• Prelude: “Additive Inverse”
• Problems and Discussion
– Clock Arithmetic and Two’s Complement
– 1/3 Scottish and Fractions in Binary
– Programs and Numbers
– Programs as Numbers
• Next Lecture Notes
2
Learning Goals: Pre-Class
By the start of class, you should be able to:
– Convert positive numbers from decimal to binary
and back.
– Convert positive numbers from hexadecimal to
binary and back.
– Take the two’s complement of a binary number.
– Convert signed (either positive or negative)
numbers to binary and back.
– Add binary numbers.
3
Learning Goals: In-Class
By the end of this unit, you should be able to:
– Critique the choice of a digital representation
scheme—including describing its strengths,
weaknesses, and flaws (such as imprecise
representation or overflow) —for a given type of
data and purpose, such as (1) fixed-width binary
numbers using a two’s complement scheme for
signed integer arithmetic in computers or (2)
hexadecimal for human inspection of raw binary
data.
4
Where We Are in
The Big Stories
Theory
How do we model
computational systems?
Now: showing that our
logical models can
connect smoothly to
models of number
systems.
Hardware
How do we build devices to
compute?
Now: enabling our
hardware to work with
data that’s more
meaningful to
humans. (And once
we have numbers,
we can represent
pictures, words, sounds,
and everything else!)
5
Motivating Problem
Understand and avoid cases like those at:
http://www.ima.umn.edu/~arnold/455.f96/disasters.html
Photo by Philippe Semanaz
(CC by/sa)
Death of 28 people caused by failure of an anti-missile system, caused
in turn by the misuse of one representation for fractions.
Explosion of a $7 billion space vehicle caused by failure of the
guidance system, caused in turn by misuse of a 16-bit signed binary
value.
(Both representations are discussed in these slides.)
6
Outline
• Prereqs, Learning Goals, and Quiz Notes
• Prelude: “Additive Inverse”
• Problems and Discussion
– Clock Arithmetic and Two’s Complement
– 1/3 Scottish and Fractions in Binary
– Programs and Numbers
– Programs as Numbers
• Next Lecture Notes
7
Prelude: Unsigned Integers
We can choose any
arrangement of Ts
and Fs to represent
numbers...
But, we might as well
choose something
convenient.
If we let F correspond to
0 and T to 1, then our
representation is...
#
p
q
r
0
F
F
F
1
F
F
T
2
F
T
F
3
F
T
T
4
T
F
F
5
T
F
T
6
T
T
F
7
T
T
T
8
Prelude: Unsigned Integers
...base 2 numbers.
When we represent
negative numbers, the
choice is also arbitrary,
but may as well be
convenient:
• Just one representation
for zero
• Easy to tell negative
from positive (or nonnegative?)
• Basic operations easy.
#
p
q
r
0
0
0
0
1
0
0
1
2
0
1
0
3
0
1
1
4
1
0
0
5
1
0
1
6
1
1
0
7
1
1
1
But... What does it mean for basic operations to be “easy”?
9
Prelude: Additive Inverse
The “additive inverse” of a number x is
another number y such that x + y = 0.
What is the additive inverse of 3?
What is the additive inverse of -7?
We want to be able to add signed binary numbers.
We need x + -x to be 0.
10
And, we want subtraction to be just addition “plus” negation.
Outline
• Prereqs, Learning Goals, and Quiz Notes
• Prelude: “Additive Inverse”
• Problems and Discussion
– Clock Arithmetic and Two’s Complement
– 1/3 Scottish and Fractions in Binary
– Programs and Numbers
– Programs as Numbers
• Next Lecture Notes
15
Problem: Clock Arithmetic
Problem: It’s 0500h.
How many hours until
midnight? Give an
algorithm that
requires a 24-hour
clock, a level, and no
arithmetic.
A level is a carpentry tool, essentially a straightedge
that indicates when it is either horizontal or vertical.
16
Clock Arithmetic
0500 is five hours from midnight.
1900 is five hours to midnight.
5 and 19 are “additive inverses”
in clock arithmetic: 5 + 19 = 0.
So are any other numbers that
are “across the clock” from
each other.
That’s even true for 12. Its additive inverse is itself!
17
Clock Arithmetic Problem
It’s 18 hundred. Without using numbers
larger than 24 in your calculations, what
time will it be 22*7 hours from now?
(Don’t multiply 22 by 7!)
a. 0 hundred (midnight)
b. 4 hundred
c. 8 hundred
d. 14 hundred
(Clock arithmetic is also known as 18
e. None of these
modular arithmetic in mathematics.)
Clock Arithmetic:
Food for Thought
If we wanted negative
numbers on the clock,
we’d probably put them
“across the clock” from
the positives.
0
-3
-6
After all, if 3 + 21 is
already 0, why not put -3
where 21 usually goes?
-9
-12
19
Unsigned Binary Clock
000
111
001
110
Here’s a 3-bit unsigned
binary clock,
numbered from 0
(000) to 7 (111).
010
101
011
100
20
Crossing the Clock
000
111
001
110
010
101
011
100
To “cross the clock”, go
as many ticks left
from the top as you
previously went right
from the top.
Here’s a clock labelled
with 0 (000) to 3 (011)
and -1 (111) to -4
(100).
21
Two’s Complement
Taking two’s complement of B = b1b2b3...bn:
1 1 1 ...1
- b1b2b3...bn
---------x1x2x3...xn
Add
1
one +
----------B
Flip
the
bits
22
A Different View of
Two’s Complement
Taking two’s complement of B = b1b2b3...bn:
1 1 1 ...1
- b1b2b3...bn
---------x1x2x3...xn
Add
1
one +
----------B
Flip
the
bits
1 1 1 ...1
+
1
---------1 0 0 0 ...0
- b1b2b3...bn
----------B
Or... Just subtract from 100...0
23
Two’s Complement vs.
Crossing the Clock
Two’s complement with k
bits:
1 1 1 ...1
+
1
---------1 0 0 0 ...0
- b1b2b3...bn
----------B
“Crossing the clock” with k
bits:
111
000
110
001
010
101
011
100
Equivalent to subtracting
Equivalent to subtracting
from 100...000 with k 0s.
from 100...000 with k 0s.
Two’s complement turns numbers into their
24
“normal”, “cross-the-clock” additive inverses.
Problem: Why Two’s
complement?
Why make the negation of 010 be 110?
(In other words: why use the two’s complement scheme?)
a. 010 + 110 already equals 0 (that is, 000).
000
b. 110 isn’t being used
111
for any other purpose.
c. 110 is the easiest
110
negation to calculate.
101
001
010
011
100
25
Summary Questions (1 of 2)
• With n bits, how many distinct values can we represent?
• What’s the smallest/largest n-bit unsigned binary integer?
• What’s the smallest/largest n-bit signed (using two’s complement)
binary integer?
• Why the “extra” negative number?
• How many representations for each number do we have with
unsigned/signed binary integers?
26
Summary Questions (2 of 2)
• How do we tell if an unsigned binary integer is: negative, positive,
zero?
• How do we tell if a signed binary integer is: negative, positive, zero?
• How do we negate a signed binary integer?
• On what value does this “negation” not behave like negation on
integers?
• How do we perform the subtraction x – y?
27
Outline
• Prereqs, Learning Goals, and Quiz Notes
• Prelude: “Additive Inverse”
• Problems and Discussion
– Clock Arithmetic and Two’s Complement
– 1/3 Scottish and Fractions in Binary
– Programs and Numbers
– Programs as Numbers
• Next Lecture Notes
28
Problem: 1/3 Scottish
Problem: Can you be 1/3 Scottish?
29
Problem: 1/3 Scottish
Problem: Can you be 1/3 Scottish?
To build a model, we must clearly specify
the problem. Many problems admit
multiple models that lead to fundamentally
different results.
We’re going to use the model of
parentage “endowing” 50% of each
parent’s “ish-ness”. That’s a coarse, even
silly model. (By that model, none of us
are Canadian, since humans did not
originate in Canada.)
Our model is handy for us, but it’s not
necessarily what people’s identity is
about!
30
Can you be one-third Scottish?
Focus on Mom (and Mom’s Mom and so on).
We’ll just make Dad “Scot” or “Not” as needed at each step.
Mom:
??
Dad:
??
Can you be one-third Scottish?
Mom:
??
Dad:
??
Mom:
2/3
Dad:
Not
Can you be one-third Scottish?
Mom:
??
Dad:
??
Mom:
1/3
Dad:
Scot
Mom:
2/3
Dad:
Not
And so on...
Mom:
1/3
Can you be one-third Scottish?
Dad:
Scot
Dad:
Not
Mom:
2/3
Mom:
1/3
Dad:
Scot
What’s
happening
here?
Mom:
2/3
Dad:
Not
Dad:
Scot
Mom:
2/3
Dad:
Not
Mom:
1/3
Dad:
Scot
Mom:
2/3
Dad:
Not
0 . 0 1 0 1..
Mom:
1/3
Now, focus on Dad...
We can represent fractions in
binary by making “Scottish family trees”:
Mom:
0.0
Dad:
Scot
Mom:
0.5
Dad:
Scot
Mom:
0.75
Dad:
Not
0.0 1 1
Here’s 0.375 in binary...
“Fixed-Point” Binary Values
0.101101…
Many languages use “floating point” numbers (scientific notation in base 2, essentially)
to represent “real numbers”: Java float and double, Racket inexact.
“Fixed-Point” Binary Values
0.101101
Problem:
Base 10 vs. Base 2
Can we represent 1/9 with a finite number of
digits in base 10 or base 2?
a. Yes (in both).
b. Yes in base 10, but not in base 2.
c. Yes in base 2, but not in base 10.
d. No (in both).
e. I don’t know.
39
Problem:
Base 10 vs. Base 2
Can we represent 1/8 with a finite number of
digits in base 10 or base 2?
a. Yes (in both).
b. Yes in base 10, but not in base 2.
c. Yes in base 2, but not in base 10.
d. No (in both).
e. I don’t know.
40
Problem:
Base 10 vs. Base 2
Can we represent 1/5 with a finite number of
digits in base 10 or base 2?
a. Yes (in both).
b. Yes in base 10, but not in base 2.
c. Yes in base 2, but not in base 10.
d. No (in both).
e. I don’t know.
41
What’s Representable?
In base 10: Anything that, multiplied by some power of 10,
becomes an integer can be represented with a finite
number of digits with our fixed-point scheme.
In base 2: The same except multiplying by a power of 2.
So why does it seem that base 10 can represent with a
finite number of digits anything that base 2 can? 42
So... Computers Can’t
Represent 1/3?
No! Using a different representation scheme (e.g.,
a rational number with a separate integer
numerator and denominator), computers can
perfectly represent 1/3!
(You know that from Racket!)
The point is: Representations that use a finite
number of bits (all of them) have weaknesses.
Know those weaknesses and their impact on
your computations!
43
Outline
• Prereqs, Learning Goals, and Quiz Notes
• Prelude: “Additive Inverse”
• Problems and Discussion
– Clock Arithmetic and Two’s Complement
– 1/3 Scottish and Fractions in Binary
– Programs and Numbers
– Programs as Numbers
• Next Lecture Notes
44
What Doesn’t Work is
Not Always Obvious (1 of 2)
Class Main {
public static void main(String[] args) {
// Let's add up 4 quarters.
System.out.println("4 quarters gives us:");
System.out.println(0.25 + 0.25 + 0.25 + 0.25);
// Let's do something a hundred times.
int i = 100;
do {
// Make i one smaller.
i--;
} while (i > 0);
System.out.println("Done!");
System.out.println("i ended up with the value: " + i);
System.out.println("It went down by: " + (100 - i));
}
}
45
What Doesn’t Work is
Not Always Obvious (2 of 2)
Class Main {
public static void main(String[] args) {
// Let's add up 10 dimes.
System.out.println("10 dimes gives us:");
System.out.println(0.1 + 0.1 + 0.1 + 0.1 + 0.1 +
0.1 + 0.1 + 0.1 + 0.1 + 0.1);
// Let's try do something a hundred times..
// but accidentally go forever
int i = 100;
do {
// Make i one LARGER. Oops!
i++;
} while (i > 0);
System.out.println("Done!");
System.out.println("i ended up with the value: " + i);
System.out.println("It went down by: " + (100 - i));
}
}
46
Number Representation Prediction
// Let's add up 10 dimes.
System.out.println("10 dimes gives us:");
System.out.println(0.1 + 0.1 + 0.1 + 0.1 + 0.1 +
0.1 + 0.1 + 0.1 + 0.1 + 0.1);
// Let's try do something a hundred times..
// but accidentally go forever
int i = 100;
do {
// Make i one LARGER. Oops!
i++;
} while (i > 0);
What will this print?
a. First 1.0 and then nothing because it runs forever.
b. First 1.0 and then some other value, because it won’t run forever.
c. First something OTHER than 1.0 and then nothing because it runs forever.
d. First something OTHER than 1.0 and then some other value,
because it won’t forever.
47
e. None of these
But Racket solves this…
right?
Racket’s number representation scheme is flexible and
powerful (but mostly not as “efficient” or “compact” as
Java’s, which usually doesn’t matter!).
But no representation is perfect, and every representation
could be more flexible.
;; What do these evaluate to?
(+ 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1)
(* (sqrt 2) (sqrt 2))
48
Outline
• Prereqs, Learning Goals, and Quiz Notes
• Prelude: “Additive Inverse”
• Problems and Discussion
– Clock Arithmetic and Two’s Complement
– 1/3 Scottish and Fractions in Binary
– Programs and Numbers
– Programs as Numbers
• Next Lecture Notes
49
Preface: Java Byte Code
Programs in the Java programming language are compiled to a
language called “byte code” that is then interpreted on a particular
computer.
Why? Byte code is hard for humans to write, read, and understand...
but it’s easy to write a program that reads and executes it
(compared to writing a program to directly read and execute Java
source code).
So, if you create a new type of computer tomorrow, and I want to run
Java code on it, I don’t have to write a program that works on your
computer and knows how to execute Java; I just need to write a
program that knows how to execute byte code.
Java byte code is also designed to be compact so it’s cheap to transmit
across the internet.
Problem: Java Byte Code
Problem: When compiled to bytecode, i = 100 might be
“push 100; store in variable 1”. The “opcode” for
bipush (push a byte) is 1610. The opcode for
istore_1 is 6010. Here’s a typical “hex” view of ~1/5th
of the previous program’s byte code. Where is i =
100?
a
b
b
d
c
e: None of these.
51
Problem:
Binary Byte Code
Why would the same task (finding a particular snippet of
code in a bytecode file) be much more difficult if the
file were represented in binary?
a. Because we would have to translate all the opcodes
and values to binary.
b. Because many bytecode files would have no binary
representation.
c. Because the binary representation of the file would
be much longer.
d. Because data like 1100100 (100 in base 2) might not
show up as the sequence of numbers 1 1 0 0 1 0 0.
e. It wouldn’t be much more difficult.
52
Problem:
Decimal Byte Code
Why would the same task (finding a particular
snippet of code in a bytecode file) be much more
difficult if the file were represented in decimal?
a. Because we would have to translate all the
opcodes and values to decimal.
b. Because many bytecode files would have no
decimal representation.
c. Because the decimal representation of the file
would be much longer.
d. Because data like 100 might not show up as the
sequence of numbers 1 0 0.
e. It wouldn’t be much more difficult.
53
Outline
• Prereqs, Learning Goals, and Quiz Notes
• Prelude: “Additive Inverse”
• Problems and Discussion
– Clock Arithmetic and Two’s Complement
– 1/3 Scottish and Fractions in Binary
– Programs and Numbers
– Programs as Numbers
• Next Lecture Notes
54
Learning Goals: In-Class
By the end of this unit, you should be able to:
– Critique the choice of a digital representation
scheme—including describing its strengths,
weaknesses, and flaws (such as imprecise
representation or overflow) —for a given type of
data and purpose, such as (1) fixed-width binary
numbers using a two’s complement scheme for
signed integer arithmetic in computers or (2)
hexadecimal for human inspection of raw binary
data.
55
Next Lecture Learning Goals:
Pre-Class
By the start of class, you should be able to:
– Use truth tables to establish or refute the
validity of a rule of inference.
– Given a rule of inference and propositional
logic statements that correspond to the rule’s
premises, apply the rule to infer a new
statement implied by the original statements.
56
Next Lecture Prerequisites
Read Section 1.3 (Epp 3rd ed) or 2.3 (Epp 4th
ed).
Complete the open-book, untimed quiz on
Connect that is due before next class.
57
snick

snack
Some Things to Try...
(on your own if you have time,
not required)
58
Representing Characters
59
Problem: Weighty Numbers
Problem: You have a balance scale and
four weights. You may choose the mass
of the weights, as long as they’re in whole
units of grams. What’s the largest number
n such that you can exactly measure
every weight 0…n?
?g
?g
?g
?g
60
Problem: Representing Data
Problem: Devise two different ways to
represent each of the following with bits:
• black-and-white images
• text
• the shape of your face
61
Problem:
256-hour Clock Arithmetic
Problem: Imagine you’ve built a computer
that uses 256-hour clock faces, each with
a single dial, as storage units. How would
you store, add, subtract, and negate
integers?
62
Concept Q: 256-hour Clock
Arithmetic
Java has a type called “byte” that is an 8-bit signed
integer. What will the following code print?
byte b = 70;
b = b + 64;
b = b + 64;
b = b + 64;
a. A positive number, greater than 70.
b. 70.
c. A positive number, less than 70.
d. 0.
e. A negative number.
63
Problem: Number Rep
Breakdown
Problem: Explain what’s happening in each
of these…
64
Program for Introductions:
Testing
Java version with 100: 9900
Java version with 8675309: 265642364
Java version with 1526097757: -645820308
Reminder: this is to calculate n*(n-1), the number of
introductions for a group of size65 n.
Program for Introductions:
Testing
Racket version with 100: 9900
Racket version with 8675309:
75260977570172
Racket version with 1526097757:
2328974362394333292
Reminder: this is to calculate (* n (- n 1)), the
number of introductions for a group of size66 n.