Add a faster API for creating v8::Integer objects

Add a faster API for creating v8::Integer objects

In WebKit, we have a small integer cache because calling v8::Integer::New is
slow. This patch adds a faster API for creating integers that requires the
caller to supply the v8::Isolate, saving us the work of looking up the isolate
in thread-local storage.

BUG=

Committed: https://code.google.com/p/v8/source/detail?r=12773

[abarth-chromium, 2012-10-17 21:32:11]

I wasn't sure whether we want to supply the isolate as the first parameter or
the last parameter. I didn't find any other examples in the API of functions
that took an isolate and another parameter.

[abarth-chromium, 2012-10-17 21:33:04]

Also, I haven't benchmarked the performance difference between the two APIs. 
I'm secretly hoping haraken will run the same benchmark he used when he
introduced WebKit's integer cache.

[abarth-chromium, 2012-10-18 01:12:41]

Benchmarks have verified that this API is faster than the old API, but perhaps
not quite as fast as the integer cache we're using currently.

I need to do some more careful measurements to see whether we can make this as
fast as the integer cache.

[Jakob Kummerow, 2012-10-18 07:35:07]

[-dano +danno]

[abarth-chromium, 2012-10-18 23:55:19]

This API isn't fast enough to replace the integer cache we have on the WebKit
side, but it is faster than the existing API.  IMHO, we should add the API and
use it in cases where we miss the integer cache.

[danno, 2012-10-19 07:31:18]

If you are going to go through all this trouble to optimize for the isolate, you
probably should make these inlined for the fast case of looking up a 31-bit
signed integer. If you inline that case and then fall back to api.cc only when
you need to allocate a heap number. If you do that, I suspect that it will as
fast as the integer cache, unless webkit often caches integers where the than
don't fit in the 31-bit signed range.

[abarth-chromium, 2012-10-19 08:32:27]

I wanted to do that, but I wasn't sure how.  I couldn't figure out how to call
i::Smi::IsValid and i::Smi::FromInt from the header.  Should I just do the bit
mashing manually or is there a more structured approach I should use in the
header?

Thanks for the review!

[danno, 2012-10-19 11:00:34]

Actually, I don't know if you will ever be able to replace the int cache with
this code, since it allocates handles. Even if it's inlined, the handle
allocation will probably make it more than trivially expensive.

I'll land this in its current form which LGTM. At some point, mstarzinger or
svenpanne might want to make it inlinable, but that won't change the API.