By pass object boxing for all js primitives.

By pass object boxing for all js primitives.
Committed: https://code.google.com/p/dart/source/detail?r=510

[John Lenz, 2011-10-17 18:44:20]

On 2011/10/17 18:34:03, John Lenz wrote:

This is going to wait until the performance tests are online again for DartC,
but I wanted to follow up on the last code review.

[floitsch, 2011-10-18 10:48:40]

LGTM.
But I just looked at the assembly and got the impression that typeof(x) ==
"number" is much more optimized than typeof(x) == typeof(y).
It could be that I'm wrong (maybe the type-feedback wasn't kicking in yet).
If typeof(x) == "number" really is more optimized I would also investigate for
typeof(x) == "string". If necessary write it then as:
if (typeof(x) == 'number') {
  return (typeof(y) == 'number') && x === y;
} else if (typeof(x) == 'string') {
  return ...
...
} else {
  return (typeof(y) != 'number') && (typeof(y) != 'string') ... &&
         x.eq$operator(y);
}

[John Lenz, 2011-10-18 16:18:02]

On 2011/10/18 10:48:40, floitsch wrote:
> LGTM.
> But I just looked at the assembly and got the impression that typeof(x) ==
> "number" is much more optimized than typeof(x) == typeof(y).
> It could be that I'm wrong (maybe the type-feedback wasn't kicking in yet).
> If typeof(x) == "number" really is more optimized I would also investigate for
> typeof(x) == "string". If necessary write it then as:
> if (typeof(x) == 'number') {
>   return (typeof(y) == 'number') && x === y;
> } else if (typeof(x) == 'string') {
>   return ...
> ...
> } else {
>   return (typeof(y) != 'number') && (typeof(y) != 'string') ... &&
>          x.eq$operator(y);
> }

Lets see what Golem makes of it.