benchmark: avoid TurboFan deopt in arrays bench

[targos]

Something unidentified at the moment is causing the arrays benchmarks to
deopt when run with the TurboFan compiler. Refactor the test to use an
inner function that can be correctly optimized by TurboFan and
Crankshaft.
I think this change also makes the benchmarks generally better. I will optimize the run function instead of the whole main that is run only once anyway.

Refs: #11851 (comment)

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• commit message follows commit guidelines

Affected core subsystem(s)

benchmark

[targos]

There are other benchmarks with the same issue. For example https://github.com/nodejs/node/blob/master/benchmark/assert/deepequal-prims-and-objs-big-array.js#L36. This one is optimized I think during the loop to initialize the arrays. Then it deopts when it reaches conf.method (or bench.start() if I move the lookup to the top).

[mscdex]

If we backport these changes, we'd want to make sure that run() is inlined since older branches probably still might use Crankshaft for those functions?

[jasnell]

Ping @mscdex ... is that an objection to landing this or does it LGTY?

[mscdex]

I'm +0 really, the functions don't get compiled by TurboFan currently (tested with node v6.x, master with V8 5.6 and 5.7) unless forced via V8 command line options. I would guess that this is because the V8 team knows about such deopts and has not enabled TurboFan for those situations.

[targos]

I'm not sure we necessarily want the run() function to be inlined. The goal of this change is to make the benchmark fair between TurboFan and Crankshaft. I tried to run it with both compilers and I think none inlined it (used --trace-inlining, is that correct?)

[mcollina]

awesome work.
LGTM

[targos]

A linter run is enough, right?
CI: https://ci.nodejs.org/job/node-test-linter/8005/

[jasnell]

Landed in eefdf45

[MylesBorins]

assuming we don't need to backport this to v6. thoughts?