Reuse the mem::swap optimizations to speed up slice::rotate

Exposes the swapping logic from PR 40454 as `pub unsafe fn ptr::swap_nonoverlapping` under feature swap_nonoverlapping

This is most helpful for compound types where LLVM didn't vectorize the loop.  Highlight: bench slice::rotate_medium_by727_strings gets 38% faster.

1 files changed, 1 insertions, 8 deletions

diff --git a/src/libcore/slice/rotate.rs b/src/libcore/slice/rotate.rs
index 3b9ae5652c5..e4a4e33c172 100644
--- a/src/libcore/slice/rotate.rs
+++ b/src/libcore/slice/rotate.rs
@@ -76,7 +76,7 @@ pub unsafe fn ptr_rotate<T>(mut left: usize, mid: *mut T, mut right: usize) {
             break;
         }
 
-        ptr_swap_n(
+        ptr::swap_nonoverlapping(
             mid.offset(-(left as isize)),
             mid.offset((right-delta) as isize),
             delta);
@@ -103,10 +103,3 @@ pub unsafe fn ptr_rotate<T>(mut left: usize, mid: *mut T, mut right: usize) {
         ptr::copy_nonoverlapping(buf, mid.offset(-(left as isize)), right);
     }
 }
-
-unsafe fn ptr_swap_n<T>(a: *mut T, b: *mut T, n: usize) {
-    for i in 0..n {
-        // These are nonoverlapping, so use mem::swap instead of ptr::swap
-        mem::swap(&mut *a.offset(i as isize), &mut *b.offset(i as isize));
-    }
-}