| CSC 161 | Grinnell College | Spring, 2012 |
| Imperative Problem Solving and Data Structures | ||
Summary: We consider the underlying binary representation of numbers.
Contents:
In your previous experiences with C or various other programming languages, you may have noticed that you got some fairly odd results when you add to the largest integer (or int) or subtract from the smallest integer (or int). Why? Because C, Java, and various other languages have chosen an interesting way to represent these values.
Behind the scenes on almost every computer, every value is represented
as a series of bits (0's and 1's). Computers use such a two-value system
because it is easy to represent physically or electronically. However,
the two-value system means that designers of languages and computers must
figure out how to convert from and to other representations (such as the
traditional
representation of integers). In the next
few sections, we will consider some basic issues of binary numbers and
the representation of integers in binary. We will then return to the
details of these representations in C, Java, and various other languages.
To understand binary numbers, begin by recalling elementary school math. When we first learned about numbers, we were taught that, in the decimal system, things are organized into columns:
H | T | O 1 | 9 | 3
such that "H" is the hundreds column, "T" is the tens column, and "O" is the ones column. So the number "193" is 1-hundreds plus 9-tens plus 3-ones. Years later, we learned that the ones column meant 100, the tens column meant 101, the hundreds column 102 and so on, such that
| 102 | | | 101 | | | 100 |
| 1 | | | 9 | | | 3 |
The number 193 is really [(1*102)+(9*101)+(3*100)].
As you know, the decimal system uses the digits 0-9 to represent numbers. If we wanted to put a larger number in column 10n (e.g., 10), we would have to multiply 10*10n, which would give 10(n+1), and be carried a column to the left. For example, putting ten in the 100 column is impossible, so we put a 1 in the 101 column, and a 0 in the 100 column, thus using two columns. Twelve would be 12*100, or 100(10+2), or 101+2*100, which also uses an additional column to the left (12).
The binary system works under the exact same principles as the decimal system, only it operates in base 2 rather than base 10. In other words, instead of columns being
102|101|100
they are
22|21|20
Instead of using the digits 0-9, we only use 0-1 (again, if we used anything larger it would be like multiplying 2*2n and getting 2(n+1), which would not fit in the 2n column. Therefore, it would shift you one column to the left. For example, "3" in binary cannot be put into one column. The first column we fill is the right-most column, which is 20, or 1. Since 3>1, we need to use an extra column to the left, and indicate it as "11" in binary (1*21) + (1*20).
What would the binary number 1011 be in decimal notation?
Try converting these numbers from binary to decimal
Remember:
| 24 | | | 23 | | | 22 | | | 21 | | | 20 |
| | | | | | | 1 | | | 0 | |||
| | | | | 1 | | | 1 | | | 1 | ||
| 1 | | | 0 | | | 1 | | | 0 | | | 1 |
| 1 | | | 1 | | | 1 | | | 1 | | | 0 |
Consider the addition of decimal numbers:
23
+48
___
We begin by adding 3+8=11. Since 11 is greater than 10, a one is put into the 10's column (carried), and a 1 is recorded in the one's column of the sum. Next, add [(2+4) +1] (the one is from the carry)=7, which is put in the 10's column of the sum. Thus, the answer is 71.
Binary addition works on the same principle, but the numerals are different. Begin with one-bit binary addition:
0 0 1 1
+0 +1 +0 1
___ ___ ___ ___
0 1 1 ??
1+1 carries us into the next column. In decimal form, 1+1=2. In binary, any digit higher than 1 puts us a column to the left (as would 10 in decimal notation). The decimal number "2" is written in binary notation as "10" (1*21)+(0*20). Record the 0 in the ones column, and carry the 1 to the twos column to get an answer of "10." In our vertical notation,
1 +1 ___ 10
The process is the same for multiple-bit binary numbers:
1010 +1111 ______
Alternately:
11 (carry)
1010
+1111
______
11001
Always remember
Try a few examples of binary addition:
111 101 111
+110 +111 +111
______ _____ _____
Multiplication in the binary system works the same way as in the decimal system.
For one-digit values:
101 * 11 ____ 101 1010 _____ 1111
Note that multiplying by two is extremely easy. To multiply by two, just add a 0 on the end.
Follow the same rules as in decimal division. Normally, for the sake of simplicity, the remainder is ignored when dividing integers.
For Example: 111011/11
10011 r 10
_______
11)111011
-11
______
101
-11
______
101
11
______
10
Converting from decimal to binary notation is slightly more difficult conceptually, but can easily be done once you know an algorithm. Begin by thinking of a few examples. We can easily see that the number 3= 2+1, and that this is equivalent to (1*21)+(1*20). This translates into putting a "1" in the 21 column and a "1" in the 20 column, to get "11". Almost as intuitive is the number 5: it is obviously 4+1, which is the same as saying [(2*2) +1], or 22+1. This can also be written as [(1*22)+(1*20)]. Looking at this in columns,
| 22 | | | 21 | | | 20 |
| 1 | | | 0 | | | 1 |
or 101.
What we're doing here is finding the largest power of two within the number (22=4 is the largest power of 2 in 5), subtracting that from the number (5-4=1), and finding the largest power of 2 in the remainder (20=1 is the largest power of 2 in 1). Then we just put this into columns. This process continues until we have a remainder of 0. Let's take a look at how it works. We know that:
| 20=1 |
| 21=2 |
| 22=4 |
| 23=8 |
| 24=16 |
| 25=32 |
| 26=64 |
| 27=128 |
and so on. To convert the decimal number 75 to binary, we would find the largest power of 2 less than 75, which is 64. Thus, we would put a 1 in the 26 column, and subtract 64 from 75, giving us 11. The largest power of 2 in 11 is 8, or 23. Put 1 in the 23 column, and 0 in 24 and 25. Subtract 8 from 11 to get 3. Put 1 in the 21 column, 0 in 22, and subtract 2 from 3. We're left with 1, which goes in 20, and we subtract one to get zero. Thus, our number is 1001011.
Making this algorithm a bit more formal gives us:
This algorithm is a bit awkward. Particularly step 3, "filling in the zeros." Therefore, we should rewrite it such that we ascertain the value of each column individually, putting in 0's and 1's as we go:
Now that we have an algorithm, we can use it to convert numbers from decimal to binary relatively painlessly. Let's try the number D=55.
1-----.
11----.
110---
1101--
11011-
110111
However, this is not the only approach possible. We can start at the right, rather than the left.
All binary numbers are in the form
a[n]*2n + a[n-1]*2(n-1)+...+a[1]*21 + a[0]*20
where each a[i] is either a 1 or a 0 (the only possible digits for the binary system). The only way a number can be odd is if it has a 1 in the 20 column, because all powers of two greater than 0 are even numbers (2, 4, 8, 16...). This gives us the rightmost digit as a starting point.
Now we need to do the remaining digits. One idea is to "shift" them. It is also easy to see that multiplying and dividing by 2 shifts everything by one column: two in binary is 10, or (1*21). Dividing (1*21) by 2 gives us (1*20), or just a 1 in binary. Similarly, multiplying by 2 shifts in the other direction: (1*21)*2=(1*22) or 10 in binary. Therefore
{a[n]*2n + a[n-1]*2(n-1) + ... + a[1]*21 + a[0]*20}/2
is equal to
a[n]*2(n-1) + a[n-1]*2(n-2) + ... + a[1]20
Let's look at how this can help us convert from decimal to binary. Take the number 163. We know that since it is odd, there must be a 1 in the 20 column (a[0]=1). We also know that it equals 162+1. If we put the 1 in the 20 column, we have 162 left, and have to decide how to translate the remaining digits.
Two's column: Dividing 162 by 2 gives 81. The number 81 in binary would also have a 1 in the 20 column. Since we divided the number by two, we "took out" one power of two. Similarly, the statement a[n-1]*2(n-1) + a[n-2]*2(n-2) + ... + a[1]*20 has a power of two removed. Our "new" 20 column now contains a1. We learned earlier that there is a 1 in the 20 column if the number is odd. Since 81 is odd, a[1]=1. Practically, we can simply keep a "running total", which now stands at 11 (a[1]=1 and a[0]=1). Also note that a 1 is essentially "remultiplied" by two just by putting it in front of a[0], so it is automatically fit into the correct column.
Four's column: Now we can subtract 1 from 81 to see what remainder we still must place (80). Dividing 80 by 2 gives 40. Therefore, there must be a 0 in the 4's column, (because what we are actually placing is a 20 column, and the number is not odd).
Eight's column: We can divide by two again to get 20. This is even, so we put a 0 in the 8's column. Our running total now stands at a[3]=0, a[2]=0, a[1]=1, and a[0]=1.
We can continue in this manner until there is no remainder to place.
Let's formalize this algorithm:
1. Let D= the number we wish to convert from decimal to binary. 2. Repeat until D=0 a. If D is even, put "0" in the leftmost open column. a. Else if D is odd, put "1" in the leftmost open column, and subtract 1 from D. c. Divide D by 2.
For the number 163, this works as follows:
| Algorithm Step | Processing Description | Result so far | New D Value | Final Comments | |
|---|---|---|---|---|---|
| Start | 163 | ||||
| 2. | b. | D is odd Put a 1 in the 20 column. Subtract 1 from D to get 162. | 1 | 162 | On to step 2c within this loop. |
| c. | Divide D=162 by 2. | 81 | D does not equal 0, so we repeat step 2. | ||
| 2. | b. | D is odd Put a 1 in the 21 column. Subtract 1 from D to get 80. | 11 | 80 | On to step 2c within this loop. |
| c. | Divide D=80 by 2. | 40 | D does not equal 0, so we repeat step 2. | ||
| 2. | a. | D is even Put a 0 in the 22 column. | 011 | 40 | On to step 2c within this loop. |
| c. | Divide D by 2. | 20 | D does not equal 0, so we repeat step 2. | ||
| 2. | a. | D is even Put a 0 in the 23 column. | 0011 | 20 | On to step 2c within this loop. |
| c. | Divide D by 2. | 10 | D does not equal 0, so we repeat step 2. | ||
| 2. | b. | D is even Put a 0 in the 24 column. | 00011 | 10 | On to step 2c within this loop. |
| c. | Divide D by 2. | 5 | D does not equal 0, so we repeat step 2. | ||
| 2. | b. | D is odd Put a 1 in the 25 column. Subtract 1 from D to get 4. | 100011 | 4 | On to step 2c within this loop. |
| c. | Divide D by 2. | 2 | D does not equal 0, so we repeat step 2. | ||
| 2. | a. | D is even Put a 0 in the 26 column. | 0100011 | 2 | On to step 2c within this loop. |
| c. | Divide D by 2. | 1 | D does not equal 0, so we repeat step 2. | ||
| 2. | b. | D is odd Put a 1 in the 27 column. Subtract 1 from D to get D=0. | 10100011 | 0 | On to step 2c within this loop. |
| c. | Divide D by 2. | 0 | D does not equal 0, so we repeat step 2. | ||
| D=0 | Loop terminates | 10100011 | |||
Conclusion: the decimal number 163 is equivalent to the binary number 10100011.
Since we already knew how to convert from binary to decimal, we can easily verify our result.
The techniques discussed above work well for non-negative integers, but how do we indicate negative numbers in the binary system? Before we investigate negative numbers, we note that the computer uses a fixed number of "bits" or binary digits. An 8-bit number is 8 digits long. For this section, we will work with 8 bits.
The simplest way to indicate negation is sign-magnitude notation. In sign-magnitude notation, the left-most bit is not actually part of the number, but is just the equivalent of a +/- sign. "0" indicates that the number is positive, "1" indicates negative. In 8 bits, 00001100 would be 12 (break this down into (1*23) + (1*22) ). To indicate -12, we would simply put a "1" rather than a "0" as the first bit: 10001100.
In one's complement, positive numbers are represented as usual in regular binary. However, negative numbers are represented differently. To negate a number, replace all zeros with ones, and ones with zeros - flip the bits. Thus, 12 would be 00001100, and -12 would be 11110011. As in signed magnitude, the leftmost bit indicates the sign (1 is negative, 0 is positive). To compute the value of a negative number, flip the bits and translate as before.
Two's complement is an interesting variant of one's complement that is more procedural than anything. You can negate a number by flipping all the bits and then adding 1 with the techniques of binary addition. (A little math will tell you that this technique works correctly when you doubly negate a number.)
In this notation, twelve would be represented as 00001100, and -12 as 11110100. To verify this, let's subtract 1 from 11110100, to get 11110011. If we flip the bits, we get 00001100, or 12 in decimal.
In excess 2(m-1), "m" indicates the total number of bits. For us (working with 8 bits), it would be excess 27. To represent a number (positive or negative) in excess 27, begin by taking the number in unsigned binary representation. Then add 27 (=128) to that number. For example, 7 would be 128 + 7=135, or 27+22+21+20, and, in binary,10000111. We would represent -7 as 128-7=121, and, in binary, 01111001.
To see the advantages and disadvantages of each method, let's try working with them.
Click here to see the answers.
C and Java use two's complement to represent the various forms of integers,
with different numbers of bits for the different forms.
A byte has eight bits, a short sixteen,
an int thirty-two, and a long sixty-four. Why do
many languages use two's complement? Because it provides the advantage
that addition is simple (you can use the standard algorithm for positive
and negative numbers), subtraction is simple (subtraction can be
implemented as negate and add
), and because you can easily tell the
sign of a number (if the leftmost bit is 0, the number is non-negative; if
the leftmost bit is 1, the number is negative).
Now, why do we get a negative number when we add to the largest value
in each? Let's consider the largest 32-bit integer,
01111111111111111111111111111111. All addition is done
using the simple additional algorithm. Let's consider what happens
when we add 2.
111111111111111111111111111111 (carry) 01111111111111111111111111111111 + 10 --------------------------------- 10000000000000000000000000000001
What number is that? Well, we know it's negative because it starts with a 1. Hence, we flip all the bits and then add 1 to find the corresponding positive number.
01111111111111111111111111111110 + 1 --------------------------------- 01111111111111111111111111111111
So, in this notation, 231 + 2 = -231
Since some programmers use the underlying bits in different ways, C provides a wide variety of infix binary operations that manipulate those bits, including
i << n
i >> n
x & y. The "and" of
1 and 1 is 1, and the "and" of any other combination is 0.
x | y. The "inclusive
or" of 0 and 0 is 0, and the "inclusive or" of any other
combination is 1.
x | y. The "exclusive
or" of two different bits is 1 and the "exclusive or" of
two identical bits is 0, and the inclusive or of any other combination is 1.
What would the binary number 1011 be in decimal notation?
| 1011= | (1*23)+(0*22)+(1*21)+(1*20) |
| = | (1*8) + (0*4) + (1*2) + (1*1) |
| = | 11 (in decimal notation) |
Try converting these numbers from binary to decimal.
| 10 = | (1*21) + (0*20) = 2+0 = 2 |
| 111 = | (1*22) + (1*21) + (1*20) = 4+2+1=7 |
| 10101= | (1*24) + (0*23) + (1*22) + (0*21) + (1*20)=16+0+4+0+1=21 |
| 11110= | (1*24) + (1*23) + (1*22) + (1*21) + (0*20)=16+8+4+2+0=30 |
Try a few examples of binary addition:
step 1 step 2 step 3
1 1
111 111 111
+110 +110 +110
______ ______ _____
1 01 1101
1 1 1
101 101 101
+111 +111 +111
_____ _____ ______
0 00 1100
1 1 1
111 111 111
+111 +111 +111
_____ _____ _____
0 10 1110
Click here to return to the question
Sign-Magnitude:
5+12 -5+12 -12+-5 12+-12
00000101 10000101 10001100 00001100
00001100 00001100 10000101 10001100
__________ ________ ________ _________
00010001 10010001 00010000 10011000
17 -17 16 -24
One' Complement:
00000101 11111010 11110011 00001100
00001100 00001100 11111010 11110011
_________ ________ ________ ________
00010001 00000110 11101101 11111111
17 6 -18 0
Two's Complement:
00000101 11111011 11110100 00001100
00001100 00001100 11111011 11110100
________ ________ ________ ________
00010001 00000111 11101111 00000000
17 7 -17 0
Click here to return to the question
Circa 1995 [Christine R. Wright]
The Binary System: A pretty damn clear guide to a quite confusing concept by Christine R. Wright with some help from Samuel A. Rebelsky.
Monday, 6 February 2006 [Samuel A. Rebelsky]
Thursday, 31 August 2006 [Marge M. Coahran]
Thursday, 15 February 2010 [Henry M. Walker]
| For more information, please contact Henry M. Walker at walker@cs.grinnell.edu. |
|