ECSE 343 Numerical Methods in

Engineering
Roni Khazaka
Dept. of Electrical and Computer Engineering
McGill University
 Number
Representation

2 Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering

Number Representation
Abstract Quantity Stored Quantity
𝑥 𝑥"

0.6
𝑥 = 4/7 0.57
7𝑥 = 4 𝑥" 0.57143
0.571429
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering
0.57142857 3
Unsigned Integers
103 102 101 100

8 3 0 2 à 8×10! + 3×10" + 0×10# + 2×10$ = 8,302

27 26 25 24 23 22 21 20
1 0 1 0 0 1 0 1 à 1×2% + 0×2& + 1×2' + 0×2( + ⋯
⋯ + 0×2! + 1×2" + 0×2# + 1×2$ = 165#$

8 bit binary
%
d7 d6 d5 d4 d3 d2 d1 d0 à - 𝑑) 2 ) Range: 0 – 255
)*$

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 4

 Unsigned Integers
uint8() has a range
uint8(5) à 0 0 0 0 0 1 0 1
from 0 to 255
8 bits

uint16(5) à 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1

16 bits
uint16() has a range
from 0 to 65,535

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 5

Signed Integers: Sign Bit
± 26 25 24 23 22 21 20 &

s d6 d5 d4 d3 d2 d1 d0 à −1 +
- 𝑑) 2 )
)*$
Range: −127 → +127

8 bit binary

± 26 25 24 23 22 21 20
1 0 1 0 0 1 0 1 à − 0×2& + 1×2' + 0×2( + 0×2! + 1×2" + 0×2# + 1×2$ =
= −37

0 0 0 0 0 0 0 0 à +0

1 0 0 0 0 0 0 0 à −0

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 6

Signed Integers: Two’s Complement
26 25 24 23 22 21 20
0 1 0 1 1 0 1 1 à 1×2& + 0×2' + 1×2( + 1×2! + 0×2" + 1×2# + 1×2$ = 91

Flip all bits and add 1

à

1 0 1 0 0 1 0 1 à 1×2% + 0×2& + 1×2' + 0×2( + 0×2! + 1×2" + 0×2# + 1×2$

= 16565+)7548
à 165 − 2, = −91-./ + 0/1234145-
!

à −2, + 1×2% + 0×2& + 1×2' + 0×2( + 0×2! + 1×2" + 0×2# + 1×2$ = −91
à −1×2% + 0×2& + 1×2' + 0×2( + 0×2! + 1×2" + 0×2# + 1×2$ = −91
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 7
Signed Integers: Two’s Complement
-27 26 25 24 23 22 21 20 &
Range: −128 → +127
s d6 d5 d4 d3 d2 d1 d0 à −𝑠×2% + - 𝑑) 2)
)*$

8 bit binary

± 26 25 24 23 22 21 20
1 0 1 0 0 1 0 1 à −1×2% + 0×2& + 1×2' + 0×2( + 0×2! + 1×2" + 0×2# + 1×2$ =
= −91

0 0 0 0 0 0 0 0 à 0

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 8

Integers
Type Number of Bits Range
uint8() 8 0 à 255
uint16() 16 0 à 65,535
uint32() 32 0 à 4,294,967,295
uint64() 64 0 à 18,446,744,073,709,551,615
int8() 8 –128 à 127
int16() 16 –32768 à 32767
int32() 32 –2,147,483,648 à 2,147,483,647
int64() 64 –9,223,372,036,854,775,808 à
9,223,372,036,854,775,807

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 9

Set of Signed/Unsigned Integers

0 1 2 3 4 251 252 253 254 255

uint8()
1

−128 −127 −126 −1 0 1 125 126 127

int8()
1

Distance between two adjacent stored values: 1

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 10

 Number Representation
Abstract Quantity Stored Quantity
𝑥 𝑥"

𝑥∈ℝ 𝑥3 = 𝑢𝑖𝑛𝑡32(𝑥)

0 ≤ 𝑥 ≤ 4,294,967,295 𝑥" ∈ 𝑢𝑖𝑛𝑡32

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 11

Error
Absolute Error: 𝜖 = 𝑥 − 𝑥"
𝑥 − 𝑥"
Relative Error: 𝜂=
𝑥
Rounding scheme à Round to nearest integer.
𝜖 = 𝑥 − 𝑥" ≤ 0.5
What about relative error?

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 12

Fixed-Point Arithmetics
102 101 100 10-1 10-2 10-3

1 3 2 0 4 5 à 1×10" + 3×10# + 2×10$ + 0×109# + 4×109" + 5×109! =

= 132.045
Distance between two adjacent stored values : 10+,

23 22 21 20 2-1 2-2 2-3 2-4

0 1 0 1 1 1 0 1
à 0×2! + 1×2" + 0×2# + 1×2$ + ⋯
⋯ + 1×29# + 1×29" + 0×29! + 1×29( = 5.8125#$

Distance between two adjacent stored values : 2+-

è Difficulty with dynamic range

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 13

Fixed-Point Arithmetics
ü Can use similar (same) hardware as integer arithmetic.
ü Fast computation.
o Numbers are equally spaced.
Loss of precision.
Limited Dynamic Range.

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 14

 Floating Point
Arithmetics

15 Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering

Example: Planck’s Law
2ℎ𝜈 * 1
𝐵) 𝜈, 𝑇 = + ,)
𝑐
𝑒 -=. −1

Planck’s Constant: ℎ = 6.626070×10+,- 𝐽. 𝑠

Speed of light: 𝑐 = 299,792,458𝑚/𝑠
Boltzmann’s constant: 𝜅. = 1.380649×10+/, 𝐽/𝐾
For an IR wavelength of 1000𝑛𝑚: 𝜈 ≅ 300𝑇𝐻𝑧 = 3×100- 𝐻𝑧

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 16

Scientific Notation
Normalized value
1123 à 1.123×10!
11.23 à 1.123×10"
0.1123 à 1.123×10#"

1.123×10!$ 1.123×10#!$

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 17

Floating Point Example: Binary

Sign bit Always ‘1’ Exponent

±1.101001×2!

‘Significand’ or ‘Mantissa’

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 18

Floating Point: Binary

Sign bit Always ‘1’ Exponent

±1. 𝑀×2%

‘Significand’ or ‘Mantissa’

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 19

 Float: Single Precision 32-bit
Sign bit 23-bit mantissa

𝑏!" 𝑏!# 𝑏$% 𝑏$& 𝑏$' 𝑏$( 𝑏$) 𝑏$* 𝑏$! 𝑏$$ 𝑏$" 𝑏$# 𝑏"% 𝑏"& 𝑏"' 𝑏"( 𝑏") 𝑏"* 𝑏"! 𝑏"$ 𝑏"" 𝑏"# 𝑏% 𝑏& 𝑏' 𝑏( 𝑏) 𝑏* 𝑏! 𝑏$ 𝑏" 𝑏#

8-bit exponent 𝐸 Bias

𝑒 = 𝑏*F 𝑏+G 𝑏+H 𝑏+I 𝑏+J 𝑏+K 𝑏+L 𝑏+* − 127 = 𝐸 − 127
Except for special cases (00000000 and 11111111)
è −126 ≤ 𝑒 ≤ 127
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 20
 Float: Single Precision 32-bit
Sign bit 23-bit mantissa 𝑀

𝑏!" 𝑏!# 𝑏$% 𝑏$& 𝑏$' 𝑏$( 𝑏$) 𝑏$* 𝑏$! 𝑏$$ 𝑏$" 𝑏$# 𝑏"% 𝑏"& 𝑏"' 𝑏"( 𝑏") 𝑏"* 𝑏"! 𝑏"$ 𝑏"" 𝑏"# 𝑏% 𝑏& 𝑏' 𝑏( 𝑏) 𝑏* 𝑏! 𝑏$ 𝑏" 𝑏#

8-bit exponent 𝐸 Bias

&!"
−1 1. 𝑏'' 𝑏'" … 𝑏" 𝑏$ ×2(#"')
&!"
−1 1. 𝑀×2(#"')

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 21

 Special Cases, IEEE 754
𝑀
𝑏!" 𝑏!# 𝑏$% 𝑏$& 𝑏$' 𝑏$( 𝑏$) 𝑏$* 𝑏$! 𝑏$$ 𝑏$" 𝑏$# 𝑏"% 𝑏"& 𝑏"' 𝑏"( 𝑏") 𝑏"* 𝑏"! 𝑏"$ 𝑏"" 𝑏"# 𝑏% 𝑏& 𝑏' 𝑏( 𝑏) 𝑏* 𝑏! 𝑏$ 𝑏" 𝑏#

𝐸
𝑬=𝟎 𝟎 < 𝑬 < 𝟐𝟓𝟓 𝑬 = 𝟐𝟓𝟓
1!" 1.0×22+0/3
𝑴=𝟎 ±0 −1 1.0×2 ±∞
1!" 1. 𝑀×22+0/3
𝑴≠𝟎 Denormalized −1 NaN

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 22

Distance between two adjacent numbers
23-bit mantissa

+1.100101011100101010010×2$ Next higher number

+1.100101011100101010011×2$
Difference is 𝜖* = 2#'! ×2$ ≅ 1.1921×10#)
Valid for all 𝑥 where 1 < 𝑥 < 2

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 23

Distance between two adjacent numbers
23-bit mantissa

+1.100101011100101010010×2" Next higher number

+1.100101011100101010011×2"
Difference is 𝜖 = 2#'! ×2" ≅ 2.3842×10#)
Valid for all 𝑥 where 2 < 𝑥 < 4

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 24

Distance between two adjacent numbers
23-bit mantissa

+1.100101011100101010010×2' Next higher number

+1.100101011100101010011×2'
Difference is 𝜖 = 2#'! ×2' ≅ 4.7684×10#)
Valid for all 𝑥 where 4 < 𝑥 < 8

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 25

Distance between two adjacent numbers
23-bit mantissa

+1.100101011100101010010×2"') Next higher number

+1.100101011100101010011×2"')
Difference is Δ = 2#'! ×2"') ≅ 2.0282×10!"
Valid for all 𝑥 where 2"') < 𝑥 < 2"'+

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 26

Smallest Normalized Value
23-bit mantissa

+1.000000000000000000000×2#"', Next higher number

+1.000000000000000000001×2#"',
Difference is 𝜖 = 2#'! ×2#"', ≅ 1.1013×10#-.
What about next smaller number?
Next number is zero if we do not de-normalize.
è Δ = 2#"', ≅ 1.1755×10#!+
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 27
 Smallest Normalized Value
0
2+0/4 ≅ 1.1755×10+,6

+1.000000000000000000000×2#"',
2+/, ×2+0/4 ≅ 1.4013×10+-5

+1.000000000000000000001×2#"',

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 28

 Denormalize 𝐸 = 0

+0.111111111111111111111×2#"',
2+/, ×2+0/4 ≅ 1.4013×10+-5

+1.000000000000000000000×2#"',
2+/, ×2+0/4 ≅ 1.4013×10+-5

+1.000000000000000000001×2#"',

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 29

 Denormalized values 𝐸 = 0

+0.111111111111111111111×2#"',

+0.000000000000000000001×2#"',

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 30

 Half-Precision 16-bit
5-bit exponent 𝐸 = 𝑏0- 𝑏0, 𝑏0/ 𝑏00 𝑏07
Sign bit 10-bit mantissa 𝑀 = 𝑏G 𝑏H … 𝑏P 𝑏F

𝑏") 𝑏"* 𝑏"! 𝑏"$ 𝑏"" 𝑏"# 𝑏% 𝑏& 𝑏' 𝑏( 𝑏) 𝑏* 𝑏! 𝑏$ 𝑏" 𝑏#

Bias
𝑒 = 𝑏>? 𝑏>@ 𝑏>A 𝑏>> 𝑏>B − 15 = 𝐸 − 15
Except for special cases (00000 and 11111) è −14 ≤ 𝑒 ≤ 15

&"#
−1 1. 𝑀×2(#".

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 31

 Quarter Precision 8-bit (non-standard)
3-bit exponent 𝐸 = 𝑏4 𝑏5 𝑏-
Sign bit 4-bit mantissa 𝑀 = 𝑏* 𝑏+ 𝑏P 𝑏F
𝑏' 𝑏( 𝑏) 𝑏* 𝑏! 𝑏$ 𝑏" 𝑏#
Bias
𝑒 = 𝑏C 𝑏D 𝑏? − 3 = 𝐸 − 3
Except for special cases (000 and 111) è −2 ≤ 𝑒 ≤ 3

&$
−1 1. 𝑀×2(#!
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 32
 Spacing (Powers of 2)
𝑀 = 0000 −1 $ 1.0000×2% −2 ≤ 𝑒 ≤ 3

0 1 2 4 8 16

𝐸 = 110 à 𝑒 =6−3=3 à 2@ =8
𝐸 = 101 à 𝑒 =5−3=2 à 2A =4

𝐸 = 010 à 𝑒 = 2 − 3 = −1 à 2E> =0.5

𝐸 = 001 à 𝑒 = 1 − 3 = −2 à 2EA =0.25
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 33
 Spacing (𝐸 = 110)
−1 $ 1. 𝑀×2! = 1𝑏! 𝑏' 𝑏" . 𝑏$ 0000 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

1000.0 ≤ 𝑥< ≤ 1111.1 Δ = 0.1A = 0.5>B

+1.0000×2@ = 1000.0A = 8>B +1.0100×2@ = 1010.0A = 10>B

+1.0001×2@ = 1000.1A = 8.5>B
+1.0010×2@ = 1001.0A = 9>B
+1.0011×2@ = 1001.1A = 9.5>B +1.1111×2@ = 1111.1A = 15.5>B
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 34
 Spacing (𝐸 = 101)
−1 $ 1. 𝑀×2' = 1𝑏! 𝑏' . 𝑏" 𝑏$ 0000 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

100.00 ≤ 𝑥< ≤ 111.11 Δ = 0.01A = 0.25>B

+1.0000×2A = 100.00A = 4>B +1.0100×2A = 101.00A = 5>B

+1.0001×2A = 100.01A = 4.25>B
+1.0010×2A = 100.10A = 4.5>B
+1.0011×2A = 100.11A = 4.75>B +1.1111×2A = 111.11A = 7.75>B
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 35
 Spacing (𝐸 = 100)
−1 $ 1. 𝑀×2" = 1𝑏! . 𝑏' 𝑏" 𝑏$ 0000 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

10.000 ≤ 𝑥< ≤ 11.111 Δ = 0.001A = 0.125>B

+1.0000×2> = 10.000A = 2>B +1.0100×2> = 10.100A = 2.5>B

+1.0001×2> = 10.001A = 2.125>B
+1.0010×2> = 10.010A = 2.25>B
+1.0011×2> = 10.011A = 2.375>B +1.1111×2A = 11.111A = 3.875>B
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 36
 Spacing (𝐸 = 011)
−1 $ 1. 𝑀×2" = 1. 𝑏! 𝑏' 𝑏" 𝑏$ 0000 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

1.0000 ≤ 𝑥< ≤ 1.1111 Δ = 0.0001A = 2E? = 0.0625>B

+1.0000×2B = 1.0000A = 1>B +1.0100×2B = 1.0100A = 1.25>B

+1.0001×2B = 1.0001A = 1.0625>B
+1.0010×2B = 1.0010A = 1.125>B
+1.0011×2B = 1.0011A = 1.1875>B +1.1111×2B = 1.1111A = 1.9375>B
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 37
 Spacing (𝐸 = 010)
−1 $ 1. 𝑀×2#" = 0.1𝑏! 𝑏' 𝑏" 𝑏$ 0000 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

0.10000 ≤ 𝑥< ≤ 0.11111 Δ = 0.00001A = 2ED = 0.03125>B

1.0000×2E> = 0.10000A = 0.5>B 1.0100×2E> = 0.10100A = 0.625>B
1.0001×2E> = 0.10001A = 0.53125>B
1.0010×2E> = 0.10010A = 0.5625>B
1.0011×2E> = 0.10011A = 0.59375>B 1.1111×2E> = 0.11111A = 0.96875>B
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 38
 Spacing (𝐸 = 001)
−1 $ 1. 𝑀×2#' = 0.01𝑏! 𝑏' 𝑏" 𝑏$ 0000 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

0.010000 ≤ 𝑥< ≤ 0.011111 Δ = 2E? 2EA = 2EC = 0.015625>B

1.0000×2+/ = 0.010000/ = 0.2507 0.10100×2+/ = 0.25 + 4×2+4
1.0001×2+/ = 0.010001/ = 0.25 + 2+4
1.0010×2+/ = 0.010010/ = 0.25 + 2×2+4
1.0011×2+/ = 0.010011/ = 0.25 + 3×2+4 0.11111×2+/ = 0.25 + 15×2+4

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 39

 Spacing (𝐸 = 001)
−1 $ 1. 𝑀×2#' 0000 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

0.010000 ≤ 𝑥< ≤ 0.011111 Δ = 2E? 2EA = 2EC = 0.015625>B

1.0000×2+/ = 0.010000/ = 0.2507 0.10100×2+/ = 0.25 + 4×2+4
1.0001×2+/ = 0.010001/ = 0.25 + 2+4
1.0010×2+/ = 0.010010/ = 0.25 + 2×2+4
1.0011×2+/ = 0.010011/ = 0.25 + 3×2+4 0.11111×2+/ = 0.25 + 15×2+4

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 40

Normalized Positive Values

0 1 2 4 8 15.5 16

• Smallest normalized value: 0.25

• Largest normalized value: 15.5
• Spacing is the same between powers of 2
• There are 16 equally spaced values at each value of 𝐸
• There are 6×16 = 96 normalized positive floating-point values
• There is a ‘large’ gap between zero and the smallest normalized value

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 41

 Subnormal Positive Values (𝐸 = 000)
−1 $ 0. 𝑀×2#' = 0.00𝑏! 𝑏' 𝑏" 𝑏$ 0001 ≤ 𝑀 ≤ 1111

0 1 2 4 8 15.5 16

0.000001 ≤ 𝑥< ≤ 0.001111 Δ = 2E? 2EA = 2EC = 0.015625>B

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 42

Special Cases, “8-bit Floating Point”

𝑬=𝟎 𝟎<𝑬<𝟕
𝟐𝟓𝟓 𝑬𝑬 == 𝟐𝟓𝟓
𝟕
11!" 2+0/3
# 1.0×22+,
𝑴=𝟎 ±0 −1
−1 1.0×2 ±∞
11!" 2+0/3
𝑀×22+,
𝑴≠𝟎 Denormalized
Subnormal −1
−1 1.𝑀×2
# 1.
NaN

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 43

 Set of Floating-Point Values 𝔽
NaN

−∞ −0 +0 +∞

−16 −8 −4 −2 −1 0 1 2 4 8 16
−15.5 15.5

Abstract Quantity Stored Quantity

𝑥 𝑥∈ℝ 𝑥" 𝑥" ∈ 𝔽

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 44

Floating Point
Type Bits Exponent Mantissa Exponent Bias 𝒆𝒎𝒊𝒏 𝒆𝒎𝒂𝒙
half() 16 5 bits 10 bits 15 -14 15
single() 32 8 bits 23 bits 127 -126 127
double() 64 11 bits 52 bits 1,023 -1,022 1,023

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 45

Double() Largest Value
11 bits
M = 111 ⋯ 11 𝐸 = 111 ⋯ 10 𝑒 = 2046 − 1023 = 1023
52 bits

+1.111 ⋯ 11×2PF+* = +1111 ⋯ 11×2GIP =

53 bits
= 1.797693134862316×10*FH
Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 46
Double() Smallest Normal Value
11 bits
M = 000 ⋯ 00 𝐸 = 000 ⋯ 01 𝑒 = 1 − 1023 = −1022
52 bits

+1.000 ⋯ 00×2QPF++ = 2.225073858507201×10Q*FH

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 47

Double() Smallest Subnormal Value
11 bits
M = 000 ⋯ 01 𝐸 = 000 ⋯ 00 𝑒 = −1022
52 bits

+0.000 ⋯ 01×2QPF++ = 2QPFIL

= 4.940656458412465×10Q*+L

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 48

 Example 32-bit Float

Sign bit

𝑏!" 𝑏!# 𝑏$% 𝑏$& 𝑏$' 𝑏$( 𝑏$) 𝑏$* 𝑏$! 𝑏$$ 𝑏$" 𝑏$# 𝑏"% 𝑏"& 𝑏"' 𝑏"( 𝑏") 𝑏"* 𝑏"! 𝑏"$ 𝑏"" 𝑏"# 𝑏% 𝑏& 𝑏' 𝑏( 𝑏) 𝑏* 𝑏! 𝑏$ 𝑏" 𝑏#

8-bit exponent 𝐸 Bias

&!"
−1 1. 𝑏'' 𝑏'" … 𝑏" 𝑏$ ×2(#"')

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 49

Exercise 1
Fill the values of the exponent bias, minimum exponent 𝑒FGH and
maximum exponent 𝑒FIJ in the table

Type Bits Exponent Mantissa Exponent Bias 𝒆𝒎𝒊𝒏 𝒆𝒎𝒂𝒙

half() 16 5 bits 10 bits
single() 32 8 bits 23 bits
double() 64 11 bits 52 bits

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 50

Floating Point
Type Bits Exponent Mantissa Exponent Bias 𝒆𝒎𝒊𝒏 𝒆𝒎𝒂𝒙
half() 16 5 bits 10 bits 15 -14 15
single() 32 8 bits 23 bits 127 -126 127
double() 64 11 bits 52 bits 1,023 -1,022 1,023

Copyright ©2020 R. Khazaka ECSE 343 Numerical Methods in Engineering 51