optimize varint encoded len calculation

Based on the cpp Protobuf implementation. This is a big win: encoded_len_varint
is now constant time WRT the input integer size.

 name                       control ns/iter  variable ns/iter  diff ns/iter   diff %  speedup
 encoded_len_varint_small   111 (7207 MB/s)  110 (7272 MB/s)             -1   -0.90%   x 1.01
 encoded_len_varint_medium  361 (2216 MB/s)  110 (7272 MB/s)           -251  -69.53%   x 3.28
 encoded_len_varint_mixed   420 (1904 MB/s)  111 (7207 MB/s)           -309  -73.57%   x 3.78
 encoded_len_varint_large   464 (1724 MB/s)  110 (7272 MB/s)           -354  -76.29%   x 4.22

Author: danburkert

src/encoding.rs

@@ -140,16 +140,9 @@ fn decode_varint_slow<B>(buf: &mut B) -> Result<u64, DecodeError> where B: Buf {
 /// The returned value will be between 1 and 10, inclusive.
 #[inline]
 pub fn encoded_len_varint(value: u64) -> usize {
-         if value < 1 <<  7 { 1 }
-    else if value < 1 << 14 { 2 }
-    else if value < 1 << 21 { 3 }
-    else if value < 1 << 28 { 4 }
-    else if value < 1 << 35 { 5 }
-    else if value < 1 << 42 { 6 }
-    else if value < 1 << 49 { 7 }
-    else if value < 1 << 56 { 8 }
-    else if value < 1 << 63 { 9 }
-    else { 10 }
+    // Based on [VarintSize64][1].
+    // [1]: https://github.com/google/protobuf/blob/3.3.x/src/google/protobuf/io/coded_stream.h#L1301-L1309
+    ((((value | 1).leading_zeros() ^ 63) * 9 + 73) / 64) as usize
 }
 
 #[derive(Clone, Copy, Debug, PartialEq)]