diff --git a/src/k_smallest.rs b/src/k_smallest.rs
index f5817700b..e1bb1114b 100644
--- a/src/k_smallest.rs
+++ b/src/k_smallest.rs
@@ -14,6 +14,7 @@ where
     where
         F: FnMut(&T, &T) -> bool,
     {
+        #[inline]
         fn children_of(n: usize) -> (usize, usize) {
             (2 * n + 1, 2 * n + 2)
         }
@@ -47,7 +48,7 @@ where
 
     let mut is_less_than = move |a: &_, b: &_| comparator(a, b) == Ordering::Less;
 
-    // Rearrange the into a valid heap by reordering from the second-bottom-most layer up to the root
+    // Rearrange the storage into a valid heap by reordering from the second-bottom-most layer up to the root
     // Slightly faster than ordering on each insert, but only by a factor of lg(k)
     // The resulting heap has the **largest** item on top
     for i in (0..=(storage.len() / 2)).rev() {
@@ -59,8 +60,7 @@ where
         // So feed them into the heap
         // Also avoids unexpected behaviour with restartable iterators
         iter.for_each(|val| {
-            // `for_each` is potentially more performant for deeply nested iterators, see its docs.
-            if is_less_than(&val, &mut storage[0]) {
+            if is_less_than(&val, &storage[0]) {
                 // Treating this as an push-and-pop saves having to write a sift-up implementation
                 // https://en.wikipedia.org/wiki/Binary_heap#Insert_then_extract
                 storage[0] = val;
@@ -88,6 +88,7 @@ where
     storage
 }
 
+#[inline]
 pub(crate) fn key_to_cmp<T, K, F>(key: F) -> impl Fn(&T, &T) -> Ordering
 where
     F: Fn(&T) -> K,