From b89a7657a9e1820b0e95b322b00b60dece15cb04 Mon Sep 17 00:00:00 2001
From: DaniPopes <57450786+DaniPopes@users.noreply.github.com>
Date: Sat, 7 Feb 2026 05:47:32 +0100
Subject: [PATCH] perf: specialize Ord::cmp for small sizes

---
 CHANGELOG.md  |  6 ++++++
 src/bits.rs   |  4 ++--
 src/cmp.rs    | 12 ++++++++++++
 src/from.rs   |  2 +-
 src/macros.rs | 30 ++++++++++++++++++++----------
 5 files changed, 41 insertions(+), 13 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 4cd83e55..097070bc 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -9,6 +9,12 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Changed
+
+- Specialize `Ord::cmp`, `const_eq`, `const_is_zero` for small sizes ([#561])
+
+[#561]: https://github.com/recmo/uint/pull/561
+
 ## [1.17.2] - 2025-12-28
 
 ### Fixed
diff --git a/src/bits.rs b/src/bits.rs
index fef5179d..a2ea24fb 100644
--- a/src/bits.rs
+++ b/src/bits.rs
@@ -157,10 +157,10 @@ impl<const BITS: usize, const LIMBS: usize> Uint<BITS, LIMBS> {
     pub const fn leading_zeros(&self) -> usize {
         let fixed = Self::MASK.leading_zeros() as usize;
 
-        as_primitives!(self, {
+        as_primitives!(self; {
             u64(x) => return x.leading_zeros() as usize - fixed,
             u128(x) => return x.leading_zeros() as usize - fixed,
-            u256(lo, hi) => return (select_unpredictable_u32(hi != 0,
+            u256((lo, hi)) => return (select_unpredictable_u32(hi != 0,
                 hi.leading_zeros(),
                 lo.leading_zeros() + 128
             )) as usize - fixed,
diff --git a/src/cmp.rs b/src/cmp.rs
index 47ede059..a263e99a 100644
--- a/src/cmp.rs
+++ b/src/cmp.rs
@@ -11,6 +11,10 @@ impl<const BITS: usize, const LIMBS: usize> PartialOrd for Uint<BITS, LIMBS> {
 impl<const BITS: usize, const LIMBS: usize> Ord for Uint<BITS, LIMBS> {
     #[inline]
     fn cmp(&self, rhs: &Self) -> Ordering {
+        as_primitives!(self, rhs; {
+            u64(x, y) => return x.cmp(&y),
+            u128(x, y) => return x.cmp(&y),
+        });
         crate::algorithms::cmp(self.as_limbs(), rhs.as_limbs())
     }
 }
@@ -82,6 +86,10 @@ impl<const BITS: usize, const LIMBS: usize> Uint<BITS, LIMBS> {
     #[inline]
     #[must_use]
     pub const fn const_is_zero(&self) -> bool {
+        as_primitives!(self; {
+            u64(x) => return x == 0,
+            u128(x) => return x == 0,
+        });
         self.const_eq(&Self::ZERO)
     }
 
@@ -92,6 +100,10 @@ impl<const BITS: usize, const LIMBS: usize> Uint<BITS, LIMBS> {
     #[inline]
     #[must_use]
     pub const fn const_eq(&self, other: &Self) -> bool {
+        as_primitives!(self, other; {
+            u64(x, y) => return x == y,
+            u128(x, y) => return x == y,
+        });
         // TODO: Replace with `self == other` and deprecate once `PartialEq` is const.
         let a = self.as_limbs();
         let b = other.as_limbs();
diff --git a/src/from.rs b/src/from.rs
index 2ab2a17e..e92e8eb9 100644
--- a/src/from.rs
+++ b/src/from.rs
@@ -807,7 +807,7 @@ impl<const BITS: usize, const LIMBS: usize> Uint<BITS, LIMBS> {
     /// Convert to IEEE 754 double precision float bit representation.
     #[inline]
     fn as_f64_bits(&self) -> u64 {
-        as_primitives!(self, {
+        as_primitives!(self; {
             u64(x) => return f64::to_bits(x as f64),
         });
 
diff --git a/src/macros.rs b/src/macros.rs
index a8ff8fb1..04ef869d 100644
--- a/src/macros.rs
+++ b/src/macros.rs
@@ -116,29 +116,39 @@ macro_rules! let_double_bits {
 
 /// Specialize an operation for u64, u128, u256 ([u128; 2])...
 macro_rules! as_primitives {
-    ($uint:expr, { $($arm:ident $t:tt => $e:expr),* $(,)? }) => {
+    ($($uints:expr),* $(,)?; $($rest:tt)*) => {
+        as_primitives!(@inner ($($uints),*); $($rest)*);
+    };
+
+    (@inner $uints:tt; { $($arm:ident $t:tt => $e:expr),* $(,)? }) => {
         $(
-            as_primitives!(@arm $uint; $arm $t => $e);
+            as_primitives!(@arm $uints; $arm $t => $e);
         )*
     };
 
-    (@arm $uint:expr; u64($n:ident) => $e:expr) => {
+    (@arm ($($uint:expr),*); u64($($n:pat),*) => $e:expr) => {
         if LIMBS == 1 {
-            let $n = $uint.limbs[0];
+            $(
+                let $n = $uint.limbs[0];
+            )*
             $e
         }
     };
-    (@arm $uint:expr; u128($n:ident) => $e:expr) => {
+    (@arm ($($uint:expr),*); u128($($n:pat),*) => $e:expr) => {
         if LIMBS == 2 {
-            let $n = $uint.as_double_words()[0].get();
+            $(
+                let $n = $uint.as_double_words()[0].get();
+            )*
             $e
         }
     };
-    (@arm $uint:expr; u256($lo:ident, $hi:ident) => $e:expr) => {
+    (@arm ($($uint:expr),*); u256($(($lo:pat, $hi:pat)),*) => $e:expr) => {
         if LIMBS == 4 {
-            let &[lo, hi] = $uint.as_double_words() else { unreachable!() };
-            let $lo = lo.get();
-            let $hi = hi.get();
+            $(
+                let &[lo, hi] = $uint.as_double_words() else { unreachable!() };
+                let $lo = lo.get();
+                let $hi = hi.get();
+            )*
             $e
         }
     };