Port inlined version of swap primitive for sorting from ia32 to x64....

Port inlined version of swap primitive for sorting from ia32 to x64.

Original code review for ia32 version: http://codereview.chromium.org/1709008

Committed: http://code.google.com/p/v8/source/detail?r=4569

[William Hesse, 2010-05-03 14:59:52]

LGTM with comments.

http://codereview.chromium.org/1858002/diff/1/4
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/1858002/diff/1/4#newcode4544
src/x64/codegen-x64.cc:4544: deferred->Branch(less);
Should be below, not less, I think, since we do cmpb.

http://codereview.chromium.org/1858002/diff/1/4#newcode4556
src/x64/codegen-x64.cc:4556: Condition both_smi = __ CheckBothSmi(index1.reg(),
index2.reg());
Shouldn't you check that both are positive smis, and below the size of the
array?  Or do you know that, because of the caller of this function?  If so, it
should be commented, I think.  I checked and this is not commented in
codegen-x64.h.

http://codereview.chromium.org/1858002/diff/1/4#newcode4580
src/x64/codegen-x64.cc:4580: // (or them and test against Smi mask.)
You could test them separately, and omit one RecordWrite for each, but if both
smi or both non-smi is the common case, your idea is better.

[Mads Ager (chromium), 2010-05-03 17:43:36]

http://codereview.chromium.org/1858002/diff/1/4
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/1858002/diff/1/4#newcode4544
src/x64/codegen-x64.cc:4544: deferred->Branch(less);
On 2010/05/03 14:59:52, William Hesse wrote:
> Should be below, not less, I think, since we do cmpb.

Yes, good catch, thanks.

http://codereview.chromium.org/1858002/diff/1/4#newcode4556
src/x64/codegen-x64.cc:4556: Condition both_smi = __ CheckBothSmi(index1.reg(),
index2.reg());
On 2010/05/03 14:59:52, William Hesse wrote:
> Shouldn't you check that both are positive smis, and below the size of the
> array?  Or do you know that, because of the caller of this function?  If so,
it
> should be commented, I think.  I checked and this is not commented in
> codegen-x64.h.

I know because of the caller. This inlined code is only generated for the sort
routine in array.js.  I'll add a comment to both the ia32 and x64 versions.  The
indices are expected to be positive and within bounds - there are no bounds
checks in this code either.