enable SIMD when comparing to simd-json benchmarks

[Licenser]

Hi,

first of all, this is a really impressive creation I love the trick of delaying number parsing darn smart :D.

That said I would suggest enabling SIMD when comparing against simd-json, otherwise, it's a bit of a pointless comparison. The best way to do that in a non too CPU-dependent way is to use RUSTFLAGS="-C target-feature=+avx,+avx2,+sse4.2" those features are present on all modern CPUs.

[jorgecarleitao]

Thanks!

There is a bench using it on the README:

string json_deserializer 2^18 time:   [10.106 ms 10.138 ms 10.173 ms]
string serde_json 2^18        time:   [23.177 ms 23.209 ms 23.243 ms]
string simd_json 2^18         time:   [10.924 ms 10.941 ms 10.959 ms]

# with `RUSTFLAGS='-C target-cpu=native'` (skilake in this case)
string simd_json 2^18         time:   [8.0735 ms 8.0887 ms 8.1046 ms]

basically, simd_json is faster for strings (best case for this crate) by ~20%.

The background of this crate is jorgecarleitao/arrow2#1024 - the delay of number parsing came as a requirement from arrow2 since it allows the user to decide to which integer/float to parse them into.

I like that this crate is forbid unsafe, but despite using it bringing a 50% improvement over serde_json, 20% is still 20%, so it is a tradeoff that I am thinking about.

[Licenser]

Absolutely, and it works in non-simd-enabled environments, which is a big plus. Over in simd-lite/simd-json#218 we've been talking about having a fallback for cases where no SIMD acceleration is available perhaps json-deserializer would be a better target then serde for that :)

I'm curious how you feel about implementing the value trait for the json-deserializer Value? It has been pretty nice for swapping between borrowed / owned value types.

[jorgecarleitao]

wow, I am humbled by this - that would be awesome!

Note that this still does not support surrogates encoded strings. Need to spend some time to do that (and find a json with them).

Sorry, not very familiar with the ecosystem - could you clarify which "value trait" you mean?

[Licenser]

Oh sorry,

the value-trait is basically this repo. The idea is to allow users to have a trait that can represent different implementations for JSONesque values so libraries can be implemented w/o targeting a specific implementation.

It basically offers all the functions needed to access, manipulate and traverse the datastructures.

[jorgecarleitao]

Got it. Thanks! How do you recommend? feature-flag value-trait on this crate and expose the implementation?

[Licenser]

That's the way I'd go ja 👍