user0/rust
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Shift then mask, rather than mask then shift

Browse source 
This may allow for small optimizations with larger float types since
`u32` math can be used after shifting. LLVM may be already getting this
anyway.
This commit is contained in:
Trevor Gross 2025-01-13 11:58:17 +00:00
parent 51496423b5
commit 5ac2f99954
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
library/compiler-builtins/libm/src/math/support/float_traits.rs
View file

@ -1,6 +1,6 @@
use core::{fmt, mem, ops};
use super::int_traits::{CastFrom, CastInto, Int, MinInt};
use super::int_traits::{CastFrom, Int, MinInt};
/// Trait for some basic operations on floats
#[allow(dead_code)]
@ -108,7 +108,7 @@ pub trait Float:
/// Returns the exponent, not adjusting for bias, not accounting for subnormals or zero.
fn exp(self) -> i32 {
((self.to_bits() & Self::EXP_MASK) >> Self::SIG_BITS).cast()
(u32::cast_from(self.to_bits() >> Self::SIG_BITS) & Self::EXP_MAX).signed()
}
/// Extract the exponent and adjust it for bias, not accounting for subnormals or zero.