Optimize Rng::fill

Benchmarks:

  test fill             ... bench:          29 ns/iter (+/- 5)
  test fill_naive       ... bench:         343 ns/iter (+/- 33)

Author: cloudhead

benches/bench.rs

@@ -73,3 +73,26 @@ fn u32_fastrand(b: &mut Bencher) {
         sum
     })
 }
+
+#[bench]
+fn fill(b: &mut Bencher) {
+    let rng = fastrand::Rng::new();
+    b.iter(|| {
+        // Pick a size that isn't divisble by 8.
+        let mut bytes = [0u8; 367];
+        rng.fill(&mut bytes);
+        bytes
+    })
+}
+
+#[bench]
+fn fill_naive(b: &mut Bencher) {
+    let rng = fastrand::Rng::new();
+    b.iter(|| {
+        let mut bytes = [0u8; 367];
+        for item in &mut bytes {
+            *item = rng.u8(..);
+        }
+        bytes
+    })
+}

src/lib.rs

@@ -444,7 +444,15 @@ impl Rng {
     /// Fill a byte slice with random data.
     #[inline]
     pub fn fill(&self, slice: &mut [u8]) {
-        for item in slice {
+        // Filling the buffer in chunks of 8 is much faster.
+        let mut chunks = slice.chunks_exact_mut(8);
+        for items in chunks.by_ref() {
+            let r = self.u64(..);
+            items.copy_from_slice(&r.to_le_bytes());
+        }
+
+        let remainder = chunks.into_remainder();
+        for item in remainder {
             *item = self.u8(..);
         }
     }