From c6307a2fa55c3d62c06b85b349257a8194093442 Mon Sep 17 00:00:00 2001
From: Djzin <djzin@users.noreply.github.com>
Date: Sat, 27 May 2017 14:29:41 +0100
Subject: [PATCH] copy tail bytes better for aligned types

---
 src/libcore/mem.rs | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

diff --git a/src/libcore/mem.rs b/src/libcore/mem.rs
index 87206152a0d2..2975d2e28dd3 100644
--- a/src/libcore/mem.rs
+++ b/src/libcore/mem.rs
@@ -479,6 +479,27 @@ pub fn swap<T>(x: &mut T, y: &mut T) {
             ptr::copy_nonoverlapping(t, y.offset(i), block_size);
             i += block_size as isize;
         }
+
+        // Swap remaining bytes 8 at a time if x & y are properly aligned
+        if align_of::<T>() % 8 == 0 {
+            while i + 8 <= len as isize {
+                let t = *(x.offset(i) as *mut u64);
+                *(x.offset(i) as *mut u64) = *(y.offset(i) as *mut u64);
+                *(y.offset(i) as *mut u64) = t;
+                i += 8;
+            }
+        }
+
+        // Swap remaining bytes 4 at a time if x & y are properly aligned
+        if align_of::<T>() % 4 == 0 {
+            while i + 4 <= len as isize {
+                let t = *(x.offset(i) as *mut u32);
+                *(x.offset(i) as *mut u32) = *(y.offset(i) as *mut u32);
+                *(y.offset(i) as *mut u32) = t;
+                i += 4;
+            }
+        }
+
         if i < len {
             // Swap any remaining bytes
             let mut t: UnalignedBlock = uninitialized();