New add_crc pack function

[rurban]

Dependent on the msgpack/msgpack#114 pull request

[rurban]

Performance loss by adding a seperate add_crc call: 15%.
Which is still much faster than anything else.

I also tried a CRC32 lookup implementation, which was 20% slower than without crc, 5% slower than the current inlined version.

-- serialize
JSON::XS: 2.32
Data::MessagePack: 0.46_01
Storable: 2.30
Benchmark: running json, mp, mp_crc, storable for at least 1 CPU seconds...
json: 2 wallclock secs ( 1.12 usr + 0.00 sys = 1.12 CPU) @ 139635.71/s (n=156392)
mp: 2 wallclock secs ( 1.03 usr + 0.00 sys = 1.03 CPU) @ 196495.15/s (n=202390)
mp_crc: 3 wallclock secs ( 1.03 usr + 0.00 sys = 1.03 CPU) @ 167020.39/s (n=172031)
storable: 5 wallclock secs ( 1.07 usr + 0.00 sys = 1.07 CPU) @ 89320.56/s (n=95573)
Rate storable json mp_crc mp
storable 89321/s -- -36% -47% -55%
json 139636/s 56% -- -16% -29%
mp_crc 167020/s 87% 20% -- -15%
mp 196495/s 120% 41% 18% --

benchmark/serialize.pl:
cmpthese timethese(
-1 => {
json => sub { JSON::encode_json($a) },
storable => sub { Storable::freeze($a) },
mp => sub { Data::MessagePack->pack($a) },
mp_crc => sub { my $p = Data::MessagePack->pack($a); Data::MessagePack->add_crc($p); },
}
);

lookup table:
json: 5 wallclock secs ( 1.10 usr + 0.01 sys = 1.11 CPU) @ 129152.25/s (n=143359)
mp: 3 wallclock secs ( 1.03 usr + 0.00 sys = 1.03 CPU) @ 196495.15/s (n=202390)
mp_crc: 5 wallclock secs ( 0.99 usr + 0.01 sys = 1.00 CPU) @ 156392.00/s (n=156392)
storable: 4 wallclock secs ( 1.07 usr + 0.03 sys = 1.10 CPU) @ 86884.55/s (n=95573)
Rate storable json mp_crc mp
storable 86885/s -- -33% -44% -56%
json 129152/s 49% -- -17% -34%
mp_crc 156392/s 80% 21% -- -20%
mp 196495/s 126% 52% 26% --