-
Notifications
You must be signed in to change notification settings - Fork 1.1k
/
Copy path002_grammar_numeric_limits.cpp
65 lines (59 loc) · 4.14 KB
/
002_grammar_numeric_limits.cpp
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
#include "numeric_limits.hpp"
#include <limits>
#include <iostream>
int main()
{
std::cout << std::boolalpha;
std::cout << "Minimum value for int: " << std::numeric_limits<int>::min() << std::endl;
std::cout << "Maximum value for int: " << std::numeric_limits<int>::max() << std::endl;
std::cout << "int is signed: " << std::numeric_limits<int>::is_signed << std::endl;
std::cout << "Non-sign bits in int: " << std::numeric_limits<int>::digits << std::endl;
std::cout << "int has infinity: " << std::numeric_limits<int>::has_infinity << std::endl;
std::cout << "Minimum value for float: " << std::numeric_limits<float>::min() << std::endl; // min returns the smallest positive value the type can encode, not the lowest
std::cout << "Lowest value for float: " << std::numeric_limits<float>::lowest() << std::endl; // the lowest value
std::cout << "Maximum value for float: " << std::numeric_limits<float>::max() << std::endl;
std::cout << "float is signed: " << std::numeric_limits<float>::is_signed << std::endl;
std::cout << "Non-sign bits in float: " << std::numeric_limits<float>::digits << std::endl;
std::cout << "float has infinity: " << std::numeric_limits<float>::has_infinity << std::endl;
std::cout << "Minimum value for unsigned short: " << std::numeric_limits<unsigned short>::min() << std::endl;
std::cout << "Maximum value for unsigned short: " << std::numeric_limits<unsigned short>::max() << std::endl;
std::cout << "is_specialized(float): " << std::numeric_limits<float>::is_specialized << std::endl;
std::cout << "is_integer(float): " << std::numeric_limits<float>::is_integer << std::endl;
std::cout << "is_exact(float): " << std::numeric_limits<float>::is_exact << std::endl;
std::cout << "is_bounded(float): " << std::numeric_limits<float>::is_bounded << std::endl;
std::cout << "is_modulo(float): " << std::numeric_limits<float>::is_modulo << std::endl;
std::cout << "is_iec559(float): " << std::numeric_limits<float>::is_iec559 << std::endl;
std::cout << "digits10(float): " << std::numeric_limits<float>::digits10 << std::endl;
std::cout << "radix(float): " << std::numeric_limits<float>::radix << std::endl;
std::cout << "min_exponent(float): " << std::numeric_limits<float>::min_exponent << std::endl;
std::cout << "max_exponent(float): " << std::numeric_limits<float>::max_exponent << std::endl;
std::cout << "min_exponent10(float): " << std::numeric_limits<float>::min_exponent10 << std::endl;
std::cout << "max_exponent10(float): " << std::numeric_limits<float>::max_exponent10 << std::endl;
std::cout << "epsilon(float): " << std::numeric_limits<float>::epsilon() << std::endl;
std::cout << "round_style(float): " << std::numeric_limits<float>::round_style << std::endl;
std::cout << "The smallest nonzero denormalized value for float: "
<< std::numeric_limits<float>::denorm_min()<< std::endl;
std::cout << "The difference between 1 and the smallest value greater than 1 for float: "
<< std::numeric_limits<float>::epsilon()<< std::endl;
std::cout << "Whether float objects allow denormalized values: "
<< std::numeric_limits<float>::has_denorm << std::endl;
std::cout << "Whether float objects can detect denormalized loss: "
<< std::numeric_limits<float>::has_denorm_loss << std::endl;
std::cout << "Whether float objects have quiet_NaN: "
<< std::numeric_limits<float>::has_quiet_NaN << std::endl;
std::cout << "Whether float objects have a signaling_NaN: "
<< std::numeric_limits<float>::has_signaling_NaN << std::endl;
std::cout << "The base for type float is: "
<< std::numeric_limits<float>::radix << std::endl;
std::cout << "The maximum rounding error for type float is: "
<< std::numeric_limits<float>::round_error() << std::endl;
std::cout << "The rounding style for a double type is: "
<< std::numeric_limits<double>::round_style << std::endl;
std::cout << "The signaling NaN for type float is: "
<< std::numeric_limits<float>::signaling_NaN() << std::endl;
std::cout << "Whether float types can detect tinyness before rounding: "
<< std::numeric_limits<float>::tinyness_before << std::endl;
std::cout << "Whether float types have implemented trapping: "
<< std::numeric_limits<float>::traps << std::endl;
return 0;
}