diff --git a/keccak/CHANGELOG.md b/keccak/CHANGELOG.md
index b0e1721..c5d7645 100644
--- a/keccak/CHANGELOG.md
+++ b/keccak/CHANGELOG.md
@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+### 0.2.2 (UNRELEASED)
+### Changed
+- Improved performance of software backend on some targets ([#129])
+
+[#129]: https://github.com/RustCrypto/sponges/pull/129
+
 ## 0.2.1 (2026-08-05)
 ### Fixed
 - Warning on softfloat AArch64 targets by not enabling `aarch64_sha3` backend on them ([#126])
diff --git a/keccak/src/backends/simd.rs b/keccak/src/backends/simd.rs
index 23ec35d..c9ee7da 100644
--- a/keccak/src/backends/simd.rs
+++ b/keccak/src/backends/simd.rs
@@ -12,11 +12,18 @@ use core::simd::u64x4 as u64xN;
 use core::simd::u64x8 as u64xN;
 
 impl LaneSize for u64xN {
-    const KECCAK_F_ROUND_COUNT: usize = crate::consts::F1600_ROUNDS;
+    const RC: &[Self] = &{
+        use crate::consts::{F1600_ROUNDS, RC};
 
-    fn truncate_rc(rc: u64) -> Self {
-        Self::splat(rc)
-    }
+        let mut res = [Self::splat(0); F1600_ROUNDS];
+        let mut i = 0;
+        #[allow(clippy::cast_possible_truncation, trivial_numeric_casts)]
+        while i < res.len() {
+            res[i] = Self::splat(RC[i]);
+            i += 1;
+        }
+        res
+    };
 
     fn rotate_left(self, n: u32) -> Self {
         self << Self::splat(n.into()) | self >> Self::splat((64 - n).into())
diff --git a/keccak/src/backends/soft.rs b/keccak/src/backends/soft.rs
index f3ad673..7c0c906 100644
--- a/keccak/src/backends/soft.rs
+++ b/keccak/src/backends/soft.rs
@@ -15,12 +15,10 @@ pub trait LaneSize:
     + BitXorAssign
     + BitXor<Output = Self>
     + Not<Output = Self>
+    + 'static
 {
-    /// Number of rounds of the Keccak-f permutation.
-    const KECCAK_F_ROUND_COUNT: usize;
-
-    /// Truncate function.
-    fn truncate_rc(rc: u64) -> Self;
+    /// Round constants
+    const RC: &[Self];
 
     /// Rotate left function.
     #[must_use]
@@ -30,12 +28,16 @@ pub trait LaneSize:
 macro_rules! impl_lanesize {
     ($type:ty, $round:expr) => {
         impl LaneSize for $type {
-            const KECCAK_F_ROUND_COUNT: usize = $round;
-
-            #[allow(clippy::cast_possible_truncation, trivial_numeric_casts)]
-            fn truncate_rc(rc: u64) -> Self {
-                rc as Self
-            }
+            const RC: &[Self] = &{
+                let mut res = [0; $round];
+                let mut i = 0;
+                #[allow(clippy::cast_possible_truncation, trivial_numeric_casts)]
+                while i < res.len() {
+                    res[i] = RC[i] as Self;
+                    i += 1;
+                }
+                res
+            };
 
             fn rotate_left(self, n: u32) -> Self {
                 self.rotate_left(n)
@@ -53,6 +55,7 @@ impl_lanesize!(u64, F1600_ROUNDS);
 macro_rules! unroll5 {
     ($var: ident, $body: block) => {
         #[cfg(not(keccak_backend_soft = "compact"))]
+        #[allow(non_upper_case_globals)]
         {
             { const $var: usize = 0; $body; }
             { const $var: usize = 1; $body; }
@@ -71,6 +74,7 @@ macro_rules! unroll5 {
 macro_rules! unroll24 {
     ($var: ident, $body: block) => {
         #[cfg(not(keccak_backend_soft = "compact"))]
+        #[allow(non_upper_case_globals)]
         {
             { const $var: usize = 0; $body; }
             { const $var: usize = 1; $body; }
@@ -108,14 +112,14 @@ macro_rules! unroll24 {
 ///
 /// # Panics
 /// If the `ROUNDS` is greater than `L::KECCAK_F_ROUND_COUNT`.
-#[allow(non_upper_case_globals, unused_assignments)]
 pub(crate) fn keccak_p<L: LaneSize, const ROUNDS: usize>(state: &mut [L; PLEN]) {
+    const { assert!(ROUNDS <= L::RC.len()) };
+
     // https://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf#page=25
     // "the rounds of KECCAK-p[b, nr] match the last rounds of KECCAK-f[b]"
-    let round_consts = RC[..L::KECCAK_F_ROUND_COUNT]
+    let round_consts = L::RC
         .last_chunk::<ROUNDS>()
-        .expect("Number of rounds greater than `KECCAK_F_ROUND_COUNT` is not supported!")
-        .map(L::truncate_rc);
+        .expect("Number of rounds is checked above");
 
     // Not unrolling this loop results in a much smaller function, plus
     // it positively influences performance due to the smaller load on I-cache
@@ -142,7 +146,10 @@ pub(crate) fn keccak_p<L: LaneSize, const ROUNDS: usize>(state: &mut [L; PLEN])
         unroll24!(x, {
             array[0] = state[PI[x]];
             state[PI[x]] = last.rotate_left(RHO[x]);
-            last = array[0];
+            #[allow(unused_assignments)]
+            {
+                last = array[0];
+            }
         });
 
         // Chi
@@ -159,7 +166,7 @@ pub(crate) fn keccak_p<L: LaneSize, const ROUNDS: usize>(state: &mut [L; PLEN])
         });
 
         // Iota
-        state[0] ^= rc;
+        state[0] ^= *rc;
     }
 }