Use a #noinline wrapper from the benchmarks to get consistent results

benches/bench.rs

@@ -8,10 +8,15 @@ use self::test::Bencher;
 
 #[bench]
 fn bench_push(b: &mut Bencher) {
+    #[inline(never)]
+    fn push_noinline(vec: &mut SmallVec<[u64; 16]>, x: u64) {
+        vec.push(x)
+    }
+
     b.iter(|| {
         let mut vec: SmallVec<[u64; 16]> = SmallVec::new();
         for x in 0..100 {
-            vec.push(x);
+            push_noinline(&mut vec, x);
         }
         vec
     });
@@ -19,10 +24,15 @@ fn bench_push(b: &mut Bencher) {
 
 #[bench]
 fn bench_insert(b: &mut Bencher) {
+    #[inline(never)]
+    fn insert_noinline(vec: &mut SmallVec<[u64; 16]>, x: u64) {
+        vec.insert(0, x)
+    }
+
     b.iter(|| {
         let mut vec: SmallVec<[u64; 16]> = SmallVec::new();
         for x in 0..100 {
-            vec.insert(0, x);
+            insert_noinline(&mut vec, x);
         }
         vec
     });
@@ -39,11 +49,16 @@ fn bench_extend(b: &mut Bencher) {
 
 #[bench]
 fn bench_pushpop(b: &mut Bencher) {
+    #[inline(never)]
+    fn pushpop_noinline(vec: &mut SmallVec<[u64; 16]>, x: u64) {
+        vec.push(x);
+        vec.pop();
+    }
+
     b.iter(|| {
         let mut vec: SmallVec<[u64; 16]> = SmallVec::new();
         for x in 0..100 {
-            vec.push(x);
-            vec.pop();
+            pushpop_noinline(&mut vec, x);
         }
         vec
     });