Reputation: 4227
I'm trying to extract the byte value of a double from the computer's memory. The reason for this is accuracy, as the final value that is represented to the user has some rounding performed.
Ideally, I want to be able to extract the sign, exponent and mantissa from the number (IEEE 754 standard)
From what I understand, casting to an unsigned char is the way to go. I have the following code snippet (stolen elsewhere), but I don't trust the results - there's no byte output for any integer values:
double d = 2;
unsigned char *p = (unsigned char*)&d;
Can anyone guide me in the right direction in order to extract byte representation of double numbers accurately, or give any advice/comments on how to proceed?
Upvotes: 4
Views: 2422
Reputation: 80265
If your motivation is only accuracy, use printf("%a\n", d);
It displays the same information as going to the binary representation would, but it displays it in a format that's almost human-readable. Here (that is, for d=2;
), it displays 0x1p+1
. The number after 0x
is the mantissa, normalized between 1 and 2, and in hexadecimal. The number after p
is the exponent. It is displayed in decimal but it represents a power of two.
Upvotes: 3
Reputation: 476950
You're doing it right. You can treat p
as an array unsigned char[sizeof(double)]
now.
For example:
for (int i = 0; i != sizeof(double); ++i) printf("%02X ", p[i]);
For d = 0.125
this prints 00 00 00 00 00 00 C0 3F
. Reversing endianness and decomposing into parts, this is:
00111111 11000000 0 0 0 0 0 0
3F C0 0 0 0 0 0 0
0 01111111100 0...<52 times>...0
S Exponent Mantissa
+ 1023 - 3 1.0
The exponent field's value is 1020, which gives an exponent of −3 after correcting by the bias of 1023, and the mantissa is 1.0, after including the implicit leading 1. So the value is 2−3 = 1/8.
Upvotes: 6