Perf comparison

[jasongin]

I ported the perf measurement script from edge-js, and took some measurements using release builds. Numbers are microseconds per function call, averaged over 100,000 iterations.

Run #	js-baseline	edge-js	napi-clr	napi-aot
1	16.78	92.67	52.9	51.93
2	16.37	95.84	53.34	51.77
3	15.78	87.93	49.67	48.65
4	16.24	95.13	51.97	48.70
5	15.31	95.09	49.51	48.53
average	16.10	93.33	51.48	49.92

	Relative performance
edge-js/baseline	579.8%
napi-aot/baseline	310.1%
napi-aot/edge-js	53.5%
napi-clr/napi-aot	97.2%

Interpretation

• The edge-js/baseline ratio is consistent with the "6x" difference reported in the original Edge project.
• napi-dotnet is about twice as fast as edge-js, or 3x slower than plain in-engine JS function calls, in this scenario.
• The napi-dotnet AOT compiled build is consistently about 3% faster than the CLR build. I'd expect a greater difference for a smaller number of iterations, since AOT has the biggest impact on startup time.
• I'd expect more advanced scenarios to show a larger advantage for napi-dotnet because it supports shared-memory typed-arrays and marshalling classes by reference, whereas edge-js serializes everything.

[vmoroz]

LGTM

[vmoroz]

Wow! These are awesome and promising results!

[jasongin]

Update: #28 cuts this call time by ~15% to around 42 microseconds.