Compare Unit64 and BigInt64

[tshemsedinov]

Compare performance of two implementation #164 and #181
We are interested in binary operations, not math operations.

[lundibundi]

Here comparing old = bigint and new = uint64, therefore improvement relates the
difference between old(bigint) and new(uint64).

Comparison was done with uint64-vs-bigint-mod.csv (it has 'kind' value removed
from options and put as 'binary') and node/benchmarks/compare.R with the patch available in the link below. Compared using https://github.com/nodejs/node/blob/master/benchmark/compare.R.

Unit64 at a point of 3rd commit in branch.

Results (you can see tests and data in the link at the bottom)

                                                                                              confidence improvement accuracy (*)   (**)  (***)
tempBench/uint64-vs-bigint.js val='0: 0x123ff, 0xab' op='and' n=3000000                              ***    -89.27 %       ±2.60% ±3.49% ±4.60%
tempBench/uint64-vs-bigint.js val='0: 0x123ff, 0xab' op='dec' n=3000000                              ***    -89.27 %       ±2.78% ±3.73% ±4.91%
tempBench/uint64-vs-bigint.js val='0: 0x123ff, 0xab' op='div' n=3000000                              ***    -97.29 %       ±2.47% ±3.32% ±4.37%
tempBench/uint64-vs-bigint.js val='0: 0x123ff, 0xab' op='inc' n=3000000                              ***    -88.55 %       ±2.76% ±3.70% ±4.88%
tempBench/uint64-vs-bigint.js val='0: 0x123ff, 0xab' op='mul' n=3000000                              ***    -97.47 %       ±2.35% ±3.16% ±4.16%
tempBench/uint64-vs-bigint.js val='0: 0x123ff, 0xab' op='xor' n=3000000                              ***    -88.24 %       ±3.38% ±4.54% ±5.99%
tempBench/uint64-vs-bigint.js val='1: 0xffffffffff, 0xffff' op='and' n=3000000                       ***    -89.06 %       ±2.69% ±3.61% ±4.76%
tempBench/uint64-vs-bigint.js val='1: 0xffffffffff, 0xffff' op='dec' n=3000000                       ***    -89.58 %       ±2.65% ±3.56% ±4.69%
tempBench/uint64-vs-bigint.js val='1: 0xffffffffff, 0xffff' op='div' n=3000000                       ***    -97.20 %       ±2.53% ±3.39% ±4.47%
tempBench/uint64-vs-bigint.js val='1: 0xffffffffff, 0xffff' op='inc' n=3000000                       ***    -88.61 %       ±3.54% ±4.75% ±6.26%
tempBench/uint64-vs-bigint.js val='1: 0xffffffffff, 0xffff' op='mul' n=3000000                       ***    -99.10 %       ±2.81% ±3.76% ±4.96%
tempBench/uint64-vs-bigint.js val='1: 0xffffffffff, 0xffff' op='xor' n=3000000                       ***    -88.49 %       ±4.08% ±5.47% ±7.22%
tempBench/uint64-vs-bigint.js val='2: 0xffffeeee00ff11, 0x3ffffeeee00ff11' op='and' n=3000000        ***    -88.70 %       ±3.48% ±4.67% ±6.15%
tempBench/uint64-vs-bigint.js val='2: 0xffffeeee00ff11, 0x3ffffeeee00ff11' op='dec' n=3000000        ***    -89.41 %       ±2.08% ±2.80% ±3.68%
tempBench/uint64-vs-bigint.js val='2: 0xffffeeee00ff11, 0x3ffffeeee00ff11' op='div' n=3000000        ***    -94.38 %       ±2.92% ±3.92% ±5.16%
tempBench/uint64-vs-bigint.js val='2: 0xffffeeee00ff11, 0x3ffffeeee00ff11' op='inc' n=3000000        ***    -88.32 %       ±3.33% ±4.47% ±5.89%
tempBench/uint64-vs-bigint.js val='2: 0xffffeeee00ff11, 0x3ffffeeee00ff11' op='mul' n=3000000        ***    -99.65 %       ±3.90% ±5.24% ±6.91%
tempBench/uint64-vs-bigint.js val='2: 0xffffeeee00ff11, 0x3ffffeeee00ff11' op='xor' n=3000000        ***    -87.72 %       ±4.85% ±6.51% ±8.58%

Related files: https://github.com/metarhia/common/tree/uint64-vs-bigint/benchmarks

/cc @tshemsedinov @nechaido @belochub

[lundibundi]

Well, this test shows that using our own uint64 is probably useless. If you have any other tests to add I'll gladly try them.
Also, each configuration was run for 3e6 times 36 times. (results are pretty stable as you see) Though it took smth like 100 min to complete so if you want to add more suites don't expect to get the results vary fast =).

[lundibundi]

This has the results and benchmarking code if needed. Closing.