https://en.wikipedia.org/wiki/IEEE_754
cross-posted from: https://lemmy.ml/post/24332731
StolenCross-posted from here: https://fosstodon.org/@foo/113731569632505985
25 days
IEEE 754
English
You must log in or register to comment.
- 25 days
Son of a bitch. Actual fifteen seconds of real laughter. Maybe even twenty.
25 daysAnd other times 0.1 + 0.2 = 0.3000004
0.1 + 0.2 != 0.3000004
You’re off by 0.000000399999999955991114575226674787700176239013671875
(0.1 + 0.2 is actually 0.3000000000000000444089209850062616169452667236328125)
(And 0.1 is actually 0.1000000000000000055511151231257827021181583404541015625, and so on. People unfamiliar with IEEE 754 may be surprised by the output of
python -c 'print("%.60f" % 0.1)')
- 24 days
Funny meme, never got why they used a bias instead of two’s complement for the exponent though
- 24 days
they could have used two’s complement for the exponent, but the biased exponent is objectively a better encoding for what floating point hardware needs to do.
This is paraphrased like mad, but basically that means hardware can compare exponent fields as plain unsigned integers. With twos complement, unsigned comparison says 255 > 1, even though -1 < 1. So every comparison would need special signed comparison logic.
