Introduce faster swapping primitives.

Introduce faster swapping primitives.

Keyed store stub sits high in sorting profiles.

Swapping allows to save us additional type checks as we could both read and
write elmenets (on fast path) without them.

Committed: http://code.google.com/p/v8/source/detail?r=4551

[antonm, 2010-04-26 14:55:43]

Lasse,

may you have a look?

[antonm, 2010-04-27 11:47:54]

Kevin,

may you have a look as well?  I've addressed the first batch of issues you
spotted

[Lasse Reichstein, 2010-04-29 12:56:33]

LGTM

http://codereview.chromium.org/1709008/diff/3001/4003
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/1709008/diff/3001/4003#newcode6680
src/ia32/codegen-ia32.cc:6680: if (FLAG_debug_code) __
AbortIfNotSmi(index2.reg());
You should still test that they are valid. I.e., that they are smis and below
elements length.
(You are relying on the calling code to be safe, which is not stable under
rewriting).

[antonm, 2010-04-29 14:22:39]

Lasse,

I've add as well a warning about this intrinsic assumptions (indices within
bounds, no hole values)---I emphasis it here as you usually don't like those
assumptions.

Starting on either full ports or stubs for x64 and ARM.

Thanks a lot for comments.

http://codereview.chromium.org/1709008/diff/3001/4003
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/1709008/diff/3001/4003#newcode6680
src/ia32/codegen-ia32.cc:6680: if (FLAG_debug_code) __
AbortIfNotSmi(index2.reg());
On 2010/04/29 12:56:34, Lasse Reichstein wrote:
> You should still test that they are valid. I.e., that they are smis and below
> elements length.
> (You are relying on the calling code to be safe, which is not stable under
> rewriting).

Ok, done.

[antonm, 2010-04-29 14:57:13]

Trivial ports to x64 and ARM (just going to runtime immediately) added.  Will
file a bug for proper ports.

I'm going to submit it as is right---let me know if you'd like to have more
checks in _SwapElements to make it a generic primitive.