milksad
/
rust-bitcoin-unsafe-fast
14
0
Fork 0
Code Issues Pull Requests Activity

chacha20_poly1305: remove mod operator 

* Swaps out the mod operator for a switch statement for a 5% performance
boost.
This commit is contained in:
Nick Johnson 2025-02-19 10:45:42 -08:00
parent 6c286e32d4
commit 36d45bf360
No known key found for this signature in database
GPG Key ID: 97B34267D0DBC8BF
1 changed files with 28 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
chacha20_poly1305/src/chacha20.rs
View File

@ -31,6 +31,14 @@ impl Nonce {
pub const fn new(nonce: [u8; 12]) -> Self { Nonce(nonce) } pub const fn new(nonce: [u8; 12]) -> Self { Nonce(nonce) }
} }
// Const validation trait for compile time check with max of 3.
trait UpTo3<const N: u32> {}
impl UpTo3<0> for () {}
impl UpTo3<1> for () {}
impl UpTo3<2> for () {}
impl UpTo3<3> for () {}
/// A SIMD-friendly structure which holds 25% of the cipher state. /// A SIMD-friendly structure which holds 25% of the cipher state.
/// ///
/// The cipher's quarter round function is the bulk of its work /// The cipher's quarter round function is the bulk of its work
@ -81,21 +89,29 @@ impl U32x4 {
} }
#[inline(always)] #[inline(always)]
fn rotate_elements_left<const N: u32>(self) -> Self { fn rotate_elements_left<const N: u32>(self) -> Self
let mut result = [0u32; 4]; where
(0..4).for_each(|i| { (): UpTo3<N>,
result[i] = self.0[(i + N as usize) % 4]; {
}); match N {
U32x4(result) 1 => U32x4([self.0[1], self.0[2], self.0[3], self.0[0]]),
2 => U32x4([self.0[2], self.0[3], self.0[0], self.0[1]]),
3 => U32x4([self.0[3], self.0[0], self.0[1], self.0[2]]),
_ => self, // Rotate by 0 is a no-op.
}
} }
#[inline(always)] #[inline(always)]
fn rotate_elements_right<const N: u32>(self) -> Self { fn rotate_elements_right<const N: u32>(self) -> Self
let mut result = [0u32; 4]; where
(0..4).for_each(|i| { (): UpTo3<N>,
result[i] = self.0[(i + 4 - N as usize) % 4]; {
}); match N {
U32x4(result) 1 => U32x4([self.0[3], self.0[0], self.0[1], self.0[2]]),
2 => U32x4([self.0[2], self.0[3], self.0[0], self.0[1]]),
3 => U32x4([self.0[1], self.0[2], self.0[3], self.0[0]]),
_ => self, // Rotate by 0 is a no-op.
}
} }
#[inline(always)] #[inline(always)]