20-02-2024

Arithmetic for Computers

Binary Representation
Unsigned :
• The binary number

01011000 00010101 00101110 11100111

MSB LSB

represents the quantity 0 × 2 +1×2 + ⋯+ 1 × 2

• A 32-bit word can represent 232 numbers between 0 and 2 − 1 Numbers

are always positive

1
 20-02-2024

Binary Representation
2’s Complement:
32 bits can only represent 232 numbers – if we wish to also Each number represents the quantity
represent negative numbers, we can represent 231 positive -231 x31 + x30 230 + x29 229 + … + x1 21 + x0 20
numbers (incl zero) and 231 negative numbers
Why is this representation favorable?
0000 0000 0000 0000 0000 0000 0000 0000two = 0ten Consider the sum of 1 and -2 …. we get -1
0000 0000 0000 0000 0000 0000 0000 0001two = 1ten Consider the sum of 2 and -1 …. we get +1
… This format can directly undergo addition
0111 1111 1111 1111 1111 1111 1111 1111two = 231-1 without any conversions!

1000 0000 0000 0000 0000 0000 0000 0000two = -231

1000 0000 0000 0000 0000 0000 0000 0001two = -(231 – 1)
1000 0000 0000 0000 0000 0000 0000 0010two = -(231 – 2)
…
1111 1111 1111 1111 1111 1111 1111 1110two = -2
1111 1111 1111 1111 1111 1111 1111 1111two = -1

Binary Representation
2’s Complement: 0000 0000 0000 0000 0000 0000 0000 0000two = 0ten
0000 0000 0000 0000 0000 0000 0000 0001two = 1ten
…
0111 1111 1111 1111 1111 1111 1111 1111two = 231-1

1000 0000 0000 0000 0000 0000 0000 0000two = -231

1000 0000 0000 0000 0000 0000 0000 0001two = -(231 – 1)
1000 0000 0000 0000 0000 0000 0000 0010two = -(231 – 2)
…
1111 1111 1111 1111 1111 1111 1111 1110two = -2
1111 1111 1111 1111 1111 1111 1111 1111two = -1

Note that the sum of a number x and its inverted representation x’ always equals a string of 1s (-1)
x + x’ = -1
x’ + 1 = -x … hence, can compute the negative of a number by
-x = x’ + 1 inverting all bits and adding 1

Similarly, the sum of x and -x gives us all zeroes, with a carry of 1

In reality, x + (-x) = 2n … hence the name 2’s complement

2
 20-02-2024

ALU Design
1bit ALU: Q. How to do subtraction?

Q. How to design 32-bit ALU

from 1-bit ALU?

Tailoring the 32-bit ALU to MIPS

slt rd, rs, rt -- if (rs < rt) rd = 1; else rd = 0 Idea: (a-b) < 0 → a < b

ALU 31
ALU 0 – ALU 30

Problem: Overflow e.g., rs = (1001)2 = -7 and rt = (0110)2= 6

How to modify the 1-bit ALU to handle slt correctly?

3
 20-02-2024

Tailoring the 32-bit ALU to MIPS

Supporting conditional branch: beq rs, rt, L1 -- if (rs == rt) branch to instruction labeled L1
Idea: (a-b) = 0 → a = b

ALU 0 – ALU 30

ALU control lines Function

000 and
001 or
010 add
110 subtract
111 Set on less than
ALU 31

Unsigned Multiplication (Combinational)

1 1 0 1 Multiplicand (13)
1 0 1 1 Multiplier (11)
--------- Ripple-carry array
1101
1101
0000
1101
-----------------
1 0 0 0 1 1 1 1 Product (143)

Carry-save array

4
 20-02-2024

Unsigned Multiplication (Sequential)

1 1 0 1 Multiplicand (13)
1 0 1 1 Multiplier (11)
---------
1101
1101
0000
1101
-----------------
1 0 0 0 1 1 1 1 Product (143)

Unsigned Multiplication
Unsigned multiplication:
M
1101
C Q
0 0000 1011 Initialization
A

0 1101 1011 q0 = 1 → Add

0 0110 1101 Shift right

1 0011 1101 q0 = 1 → Add

0 1001 1110 Shift right
1 1 0 1 Multiplicand (13)
1 0 1 1 Multiplier (11)
0 1001 1110 q0 = 0 → No Add ---------
0 0100 1111 Shift right 1101
1101
1 0001 1111 q0 = 1 → Add 0000
1101
0 1000 1111 Shift right -----------------
1 0 0 0 1 1 1 1 Product (143)

5
 20-02-2024

Signed Multiplication

• Recall grade school trick

Booth’s encoding:
• When multiplying by 9: Current Bit to Explanation Example Operation
• Multiply by 10 (shift digits left) bit right
• Subtract once 1 0 Begins run of ‘1’ 0000111100 Subtract
• E.g., 12345 x 9 = 12345 x (10 - 1) 1 1 Middle of run of ‘1’ 0000111100 Nothing
= 123450 - 12345 0 1 End of a run of ‘1’ 0000111100 Add
• Booth’s algorithm applies same principle 0 0 Middle of a run of ‘0’ 0000111100 Nothing
• Except no ‘9’ in binary, just ‘1’ and ‘0’
• Search for a run of ‘1’ bits in the multiplier
• E.g. ‘0110’ has a run of 2 ‘1’ bits in the middle
• Multiplying by ‘0110’ (6 in decimal) is equivalent to multiplying by 8 and subtracting twice,
since 6 x m = (8 – 2) x m = 8m – 2m

Booth’s Algorithm
Example: 2 x -3 = -6, or 0010 x 1101 = 1111 1010

M
0010 Q-1
Q
0000 1101 0 Initialization Current bit Bit to right Operation
A
1 0 Subtract

1110 1101 0 Q0Q-1 = 10 → Subtract 1 1 Nothing

0 1 Add
1111 0110 1 Arithmetic right shift
0 0 Nothing

0001 0110 1 Q0Q-1 = 01 → Add

0000 1011 0 Arithmetic right shift

1110 1011 0 Q0Q-1 = 10 → Subtract 1111 0101 1 Q0Q-1 = 11 → Nothing

1111 0101 1 Arithmetic right shift 1111 1010 1 Arithmetic right shift

6
 20-02-2024

Why Booth’s Algorithm Works?

𝒂𝒊 𝒂𝒊 𝟏 Operation
Assume 𝑎 be the multiplier and 𝑏 be the multiplicand
1 0 Subtract
1 1 Nothing
Booth’s algorithm can be written as
0 1 Add
𝑎 −𝑎 ×𝑏×2 0 0 Nothing
+ 𝑎 −𝑎 ×𝑏×2
+ 𝑎 −𝑎 ×𝑏×2 Note: −𝑎 × 2 + 𝑎 × 2 = 𝑎 × 2 and 𝑎 =0 𝑎 − 𝑎 = 0 Nothing
… … 𝑎 − 𝑎 = 1 Add
𝑎 − 𝑎 = −1 Subtract
+ 𝑎 −𝑎 ×𝑏×2
+ 𝑎 −𝑎 ×𝑏×2

Factoring out 𝑏 from each term: 𝑏 × 𝑎 × −2 + 𝑎 ×2 +⋯+ 𝑎 × 2 + 𝑎 ×2

2’s complement of 𝑎

Unsigned Division

1101 100010010 Dividend

Divisor 1101
10000
1101
1110
1101
1 Remainder

Restoring division:
Do the following 𝒏 times
1. Shift 𝐴 and 𝑄 left one bit
2. 𝐴 ← 𝐴 − 𝑀
3. If the sign of 𝐴 is 1 i.e., 𝑎 = 1
𝑞 ←0
𝐴 ← 𝐴 + 𝑀 //Restore After division 𝑛-bit quotient is in 𝑄 and the remainder is in 𝐴
else
𝑞 ←1

7
 20-02-2024

Unsigned Division (Restoring)

Example: Divisor: 11 Dividend: 1000
A Q M
00000 1000 00011 Initialization
00001 000? Shift left 𝐴𝑄
11101 000? Subtract 𝐴 ← 𝐴 − 𝑀
00001 0000 𝑞 ← 0, 𝐴 ← 𝐴 + 𝑀

00010 000? Shift left 𝐴𝑄

11111 000? Subtract 𝐴 ← 𝐴 − 𝑀
00010 0000 𝑞 ← 0, 𝐴 ← 𝐴 + 𝑀
00100 000? Shift left 𝐴𝑄
Do the following 𝒏 times
00001 000? Subtract 𝐴 ← 𝐴 − 𝑀 1. Shift 𝐴 and 𝑄 left one bit
00001 0001 𝑞 ←1 2. 𝐴 ← 𝐴 − 𝑀
3. If the sign of 𝐴 is 1 i.e., 𝑎 = 1
00010 001? Shift left 𝐴𝑄 𝑞 ←0
𝐴 ← 𝐴 + 𝑀 //Restore
11111 001? Subtract 𝐴 ← 𝐴 − 𝑀
else
00010 0010 𝑞 ← 0, 𝐴 ← 𝐴 + 𝑀 𝑞 ←1
Remainder Quotient

Unsigned Division (Non-restoring)

In restoring division after subtraction what we do?

If A is positive – Shift left AQ and subtract M

 2A - M
If A is negative – First we restore A by adding M,
then we shift left AQ and subtract M
 2(A + M) – M = 2A + M

Summary: Restore operation is eliminated

In each step, we need to shift and then Add/Subtract
Restoring Division
Non-restoring division Do the following 𝒏 times
Step 1: Do the following 𝑛 times 1. Shift 𝐴 and 𝑄 left one bit
1. If the sign of A is 0, Shift left AQ by one bit, and subtract M from A 2. 𝐴 ← 𝐴 − 𝑀
otherwise, Shift left AQ by one bit and add M from A 3. If the sign of 𝐴 is 1 i.e., 𝑎 = 1
2. Now, if the sign of A is 0, set 𝑞 to 1; otherwise set 𝑞 to 0 𝑞 ←0
𝐴 ← 𝐴 + 𝑀 //Restore
Step 2 [Sign correction of reminder]: If the sign of A is 1, add M to A else
𝑞 ←1

8
 20-02-2024

Unsigned Division (Non-restoring)

Example: Divisor: 11 Dividend: 1000
A Q M
00000 1000 00011 Initialization
00001 000? Shift left 𝐴𝑄
11110 000? Subtract 𝐴 ← 𝐴 − 𝑀
11110 0000 𝑞 ← 0 (A is –ve)

11100 000? Shift left 𝐴𝑄

11111 000? Add 𝐴 ← 𝐴 + 𝑀
11111 0000 𝑞 ← 0 (A is –ve) Non-restoring division
Step 1: Do the following 𝑛 times
11110 000? Shift left 𝐴𝑄 1. If the sign of A is 0, Shift left AQ by one bit, and subtract
00001 000? Add 𝐴 ← 𝐴 + 𝑀 M from A
otherwise, Shift left AQ by one bit and add M from A
00001 0001 𝑞 ← 1 (A is +ve) 2. Now, if the sign of A is 0, set 𝑞 to 1; otherwise set 𝑞 to 0
00010 001? Shift left 𝐴𝑄 Step 2 [Sign correction of reminder]:
If the sign of A is 1, add M to A
11111 001? Subtract 𝐴 ← 𝐴 − 𝑀
11111 0010 𝑞 ← 0 (A is –ve)
Remainder(?) Quotient After sign correction, reminder is 00010

Signed Division
Simplest solution: Use unsigned division and negate the quotient if signs of divisor and dividend disagree.

Q. What should be the sign of remainder?

Note: The equation 𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑 = 𝑄𝑢𝑜𝑡𝑖𝑒𝑛𝑡 × 𝐷𝑖𝑣𝑖𝑠𝑜𝑟 + 𝑅𝑒𝑚𝑎𝑖𝑛𝑑𝑒𝑟 must always hold

Example: −7 ÷ 2: 𝑄𝑢𝑜𝑡𝑖𝑒𝑛𝑡 = −3,

𝑅𝑒𝑚𝑎𝑖𝑛𝑑𝑒𝑟 = 𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑 − 𝑄𝑢𝑜𝑡𝑖𝑒𝑛𝑡 × 𝐷𝑖𝑣𝑖𝑠𝑜𝑟 = −7 − −3 × 2 = −1

Note: −4 as quotient also satisfies the formula i.e., 𝑅𝑒𝑚𝑎𝑖𝑛𝑑𝑒𝑟 = −7 − −4 × 2 = 1.

However, the absolute value of the quotient would change depending on the sign of the
divisor and the dividend. Here, − 𝑥 ÷ 𝑦 ≠ −𝑥 ÷ 𝑦, which is problematic!
Example: 7 ÷ −2: 𝑄𝑢𝑜𝑡𝑖𝑒𝑛𝑡 = −3,
𝑅𝑒𝑚𝑎𝑖𝑛𝑑𝑒𝑟 = 𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑 − 𝑄𝑢𝑜𝑡𝑖𝑒𝑛𝑡 × 𝐷𝑖𝑣𝑖𝑠𝑜𝑟 = 7 − −3 × −2 = 1

Example: −7 ÷ −2: 𝑄𝑢𝑜𝑡𝑖𝑒𝑛𝑡 = 3,

𝑅𝑒𝑚𝑎𝑖𝑛𝑑𝑒𝑟 = 𝐷𝑖𝑣𝑖𝑑𝑒𝑛𝑑 − 𝑄𝑢𝑜𝑡𝑖𝑒𝑛𝑡 × 𝐷𝑖𝑣𝑖𝑠𝑜𝑟 = −7 − 3 × −2 = −1
Rule: Negate the quotient if the sign of divisor and dividend are opposite.
Sign of non-zero remainder matches the dividend.

9
 20-02-2024

MIPS Multiplication and Division

Multiplication: Division:
• Two 32-bit registers for product • Use HI/LO registers for result
• HI: most-significant 32 bits • HI: 32-bit remainder
• LO: least-significant 32-bits • LO: 32-bit quotient
• Instructions • Instructions
• mult rs, rt / multu rs, rt • div rs, rt / divu rs, rt
• 64-bit product in HI/LO • No overflow or divide-by-0 checking
• mfhi rd / mflo rd • If divisor is 0, result is unpredictable
• Move from HI/LO to rd • Software must perform checks if required
• Can test HI value to see if product • Use mfhi, mflo to access result
overflows 32 bits

Floating Point Numbers

Normalized scientific notation: single non-zero digit to the left of the decimal (binary) point
– example: 3.5 x 109

Floating-point standard:
• Defined by IEEE Std 754-1985
• Developed in response to divergence of representations
• Portability issues for scientific code
• Now almost universally adopted
• Two representations
• Single precision (32-bit)
• Double precision (64-bit)

10
 20-02-2024

IEEE Floating Point Format

single: 8 bits single: 23 bits
double: 11 bits double: 52 bits
S Exponent Fraction

x  ( 1)S  (1 Fraction)  2(Exponent Bias)

• S: sign bit (0  non-negative, 1  negative)

• Normalize significand: 1.0 ≤ |significand| < 2.0
• Always has a leading pre-binary-point 1 bit, so no need to represent it explicitly (hidden bit)
• Significand is Fraction with the “1.” restored
• Exponent: excess representation: actual exponent + Bias
• Ensures exponent is unsigned
• Single: Bias = 127; Double: Bias = 1203 Q. Why excess-127 representation
for exponent is used?

Note: More exponent bits → wider range of numbers

More fraction bits → higher precision

Example
• Represent 0.75 • What number is represented by the single-
• 0.75 = 1.12 × 2–1 precision float
• S=0 11000000101000…00
• Fraction = 1000…002 • S=1
• Exponent = –1 + Bias • Fraction = 01000…002
• Single: –1 + 127 = 126 = 011111102 • Exponent = 100000012 = 129
• Single: 0011111101000…00
• x = (–1)1 × (1 + 012) × 2(129 – 127)
Q. Represent 0.5 = (–1) × 1.25 × 22
= –5.0
Q. How to represent 0 ?
A. Exponent 00…0, Fraction: 00…0

±Infinity: Exponent = 111...1, Fraction = 000...0

NAN: Exponent = 111...1, Fraction ≠ 000...0
Indicates illegal or undefined result e.g., 0.0/0.0

11
 20-02-2024

Floating Point Numbers

Single Precision Range: Double Precision Range:
• Exponents 00000000 and 11111111 reserved • Exponents 0000…00 and 1111…11 reserved
• Smallest value
Smallest value
• Exponent: 00000000001
Exponent: 00000001
 actual exponent = 1 – 1023 = –1022
 actual exponent = 1 – 127 = –126
• Fraction: 000…00  significand = 1.0
Fraction: 000…00  significand = 1.0
±1.0 × 2–126 ≈ ±1.2 × 10–38 • ±1.0 × 2–1022 ≈ ±2.2 × 10–308
• Largest value
Largest value • Exponent: 11111111110
exponent: 11111110  actual exponent = 2046 – 1023 = +1023
 actual exponent = 254 – 127 = +127 • Fraction: 111…11  significand ≈ 2.0
Fraction: 111…11  significand ≈ 2.0 • ±2.0 × 2+1023 ≈ ±1.8 × 10+308
±2.0 × 2+127 ≈ ±3.4 × 10+38

Summary
Value inf 2 special cases up top that use the 0 255 00…0
Value NAN reserved exponent field of 255 0 255 xx….x
Highest Norm ~2 x 2127 0 254 11….1

Value 1 0 127 00…0

Exponent field < 127, i.e., after
subtracting bias, they are negative
exponents, representing numbers < 1
Smallest Norm ~1x 2-126 0 0..01 00…0
Largest Denorm ~1 x 2-126 Special case with exponent field 0, used to 0 0..00 11…1
Smallest Denorm ~2-149 represent denorms, that help us gradually approach 0 0 0..00 00…1
Value 0 0 0..00 00…0
Same rules as above, but the sign bit is 1
Same magnitudes as above, but negative numbers

12
 20-02-2024

Floating Point Addition

• Consider a 4-digit binary example (0.5 + –0.4375)
• 1.0002 × 2–1 + –1.1102 × 2–2

1. Align binary points

Shift number with smaller exponent
1.0002 × 2–1 + –0.1112 × 2–1
2. Add significands
1.0002 × 2–1 + –0.1112 × 2–1 = 0.0012 × 2–1

3. Normalize result & check for over/underflow

1.0002 × 2–4, with no over/underflow
4. Round and renormalize if necessary
1.0002 × 2–4 (no change) = 0.0625 Assumptions of associativity may fail
(x+y)+z x+(y+z)
x -1.50E+38 -1.50E+38
y 1.50E+38 0.00E+00
z 1.0 1.0 1.50E+38
1.00E+00 0.00E+00

Floating Point Multiplication

• Consider a 4-digit binary example
• 1.0002 × 2–1 × –1.1102 × 2–2 (0.5 × –0.4375)
• 1. Add exponents
• Unbiased: –1 + –2 = –3
• Biased: (–1 + 127) + (–2 + 127) = –3 + 254 – 127 = –3 + 127
• 2. Multiply significands
• 1.0002 × 1.1102 = 1.1102  1.1102 × 2–3
• 3. Normalize result & check for over/underflow
• 1.1102 × 2–3 (no change) with no over/underflow
• 4. Round and renormalize if necessary
• 1.1102 × 2–3 (no change)
• 5. Determine sign: +ve × –ve  –ve
• –1.1102 × 2–3 = –0.21875

13