Experimental: use register allocation in the code for keyed property

Experimental: use register allocation in the code for keyed property
loads.

Committed: http://code.google.com/p/v8/source/detail?r=1058

[Mads Ager (chromium), 2009-01-13 07:09:09]

LGTM

http://codereview.chromium.org/17615/diff/206/210
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/17615/diff/206/210#newcode4580
Line 4580: // reuse the index or elements register for the value.
It seems that we do use the index register for the value.  Update the comment?

[William Hesse, 2009-01-13 08:41:09]

LGTM

http://codereview.chromium.org/17615/diff/206/210
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/17615/diff/206/210#newcode4581
Line 4581: Result value = index;
I see that assignment of Results does not invalidate the RHS.  If we ever change
that, the following use of index will become invalid.  Does the assembler catch
that?

[Kevin Millikin (Chromium), 2009-01-13 08:44:20]

http://codereview.chromium.org/17615/diff/206/210
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/17615/diff/206/210#newcode4580
Line 4580: // reuse the index or elements register for the value.
On 2009/01/13 07:09:09, Mads Ager wrote:
> It seems that we do use the index register for the value.  Update the comment?

I want to reflect that there are two possibilities.  We might revisit this code
when we have some heuristic that we want to use here---like if one is eax reuse
that one, because the value coming from the deferred code will be in eax.

I'll make the comment more explicit.

[Kevin Millikin (Chromium), 2009-01-13 08:56:35]

http://codereview.chromium.org/17615/diff/206/210
File src/codegen-ia32.cc (right):

http://codereview.chromium.org/17615/diff/206/210#newcode4581
Line 4581: Result value = index;
On 2009/01/13 08:41:10, William Hesse wrote:
> I see that assignment of Results does not invalidate the RHS.  If we ever
change
> that, the following use of index will become invalid.  Does the assembler
catch
> that?

I'm not sure we will want to change it---copy constructing a  result creates an
alias and that's probably how we want it.

There is an assertion in Result::reg() that will catch it in debug builds.