←back to thread

C++ proposal: There are exactly 8 bits in a byte

(www.open-std.org)
288 points Twirrim | 1 comments | 17 Oct 24 22:21 UTC | HN request time: 0.274s | source
Show context
favorited ◴[17 Oct 24 23:54 UTC] No.41875023[source]▶
Previously, in JF's "Can we acknowledge that every real computer works this way?" series: "Signed Integers are Two’s Complement" <https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p09...>
replies(1): >>41875200 #
jsheard ◴[18 Oct 24 00:25 UTC] No.41875200[source]▶
Maybe specifying that floats are always IEEE floats should be next? Though that would obsolete this Linux kernel classic so maybe not.

https://github.com/torvalds/linux/blob/master/include/math-e...

replies(9): >>41875213 #>>41875351 #>>41875749 #>>41875859 #>>41876173 #>>41876461 #>>41876831 #>>41877394 #>>41877730 #
conradev ◴[18 Oct 24 03:46 UTC] No.41876173[source]▶
I was curious about float16, and TIL that the 2008 revision of the standard includes it as an interchange format:

https://en.wikipedia.org/wiki/IEEE_754-2008_revision

replies(1): >>41877684 #
tialaramex ◴[18 Oct 24 09:27 UTC] No.41877684[source]▶
Note that this type (which Rust will/ does in nightly call "f16" and a C-like language would probably name "half") is not the only popular 16-bit floating point type, as some people want to have https://en.wikipedia.org/wiki/Bfloat16_floating-point_format
replies(1): >>41886241 #
1. adrian_b ◴[19 Oct 24 07:23 UTC] No.41886241[source]▶
The IEEE FP16 format is what is useful in graphics applications, e.g. for storing color values.

The Google BF16 format is useful strictly only for machine learning/AI applications, because its low precision is insufficient for anything else. BF16 has very low precision, but an exponent range equal to FP32, which makes overflows and underflows less likely.