4 releases
0.2.1 | Nov 15, 2022 |
---|---|
0.2.0 | Oct 18, 2022 |
0.1.1 | Dec 12, 2021 |
0.1.0 | Dec 12, 2021 |
#180 in No standard library
27KB
430 lines
nanval
A no_std
, zero-dependency crate for the creation and handling of NaN-tagged 64-bit floating-point values.
Inspired by this article and this crate.
How does this work?
When a 64-bit floating-point number is set to NaN
/0x7FF8000000000000
, its bits are as follows:
s111 1111 1111 1qxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx
^ ^\____________________________________________________________/
| | ^
| Sign Bit | Quiet Bit | Data Bits
As long as the data bits aren't all set to 0
, indicating the original/sentinel NaN
value, they can be literally anything else! This gives us 50 bits to mess with/use as we please...
UInts / Unsigned Integers
Look at the module
crate::uint
for this.
TODO: Add explanation.
Cells / Pointers
Look at the module
crate::cell
for this.
Since it doesn't matter what the sign-bit s
is set to, we can use it as a flag/marker that indicates that the value is some kind of cell
or ptr
.
Combine this with the fact that basically all x64-platforms only use the lower 48 or 50 bits for addressing (ignoring CHERI shenanigans), we are left with 3 bits (that includes the 'quiet' bit) to store some kind of type-tag for the cell; look at the crate::cell::CellTag
.