speed up pure-Go inner loop by allowing improved bounds check hints

name          old time/op    new time/op    delta
Metro/100-8     28.8ns ± 6%    23.6ns ± 4%  -18.30%  (p=0.000 n=27+28)
Metro/1024-8     158ns ± 6%     118ns ± 5%  -25.58%  (p=0.000 n=30+30)
Metro/8192-8    1.17µs ± 5%    0.88µs ± 7%  -24.73%  (p=0.000 n=30+30)

name          old speed      new speed      delta
Metro/100-8   3.47GB/s ± 6%  4.25GB/s ± 3%  +22.48%  (p=0.000 n=27+27)
Metro/1024-8  6.47GB/s ± 5%  8.68GB/s ± 4%  +34.23%  (p=0.000 n=30+30)
Metro/8192-8  7.03GB/s ± 5%  9.34GB/s ± 6%  +32.86%  (p=0.000 n=30+30)

Author: dgryski

metro64.go

@@ -21,18 +21,15 @@ func Hash64(buffer []byte, seed uint64) uint64 {
 		v := [4]uint64{hash, hash, hash, hash}
 
 		for len(ptr) >= 32 {
-			v[0] += binary.LittleEndian.Uint64(ptr) * k0
-			ptr = ptr[8:]
+			v[0] += binary.LittleEndian.Uint64(ptr[:8]) * k0
 			v[0] = rotate_right(v[0], 29) + v[2]
-			v[1] += binary.LittleEndian.Uint64(ptr) * k1
-			ptr = ptr[8:]
+			v[1] += binary.LittleEndian.Uint64(ptr[8:16]) * k1
 			v[1] = rotate_right(v[1], 29) + v[3]
-			v[2] += binary.LittleEndian.Uint64(ptr) * k2
-			ptr = ptr[8:]
+			v[2] += binary.LittleEndian.Uint64(ptr[16:24]) * k2
 			v[2] = rotate_right(v[2], 29) + v[0]
-			v[3] += binary.LittleEndian.Uint64(ptr) * k3
-			ptr = ptr[8:]
+			v[3] += binary.LittleEndian.Uint64(ptr[24:32]) * k3
 			v[3] = rotate_right(v[3], 29) + v[1]
+			ptr = ptr[32:]
 		}
 
 		v[2] ^= rotate_right(((v[0]+v[3])*k0)+v[1], 37) * k1