Faster filling newly allocated arrays with the holes from the Array construction stub.

Faster filling newly allocated arrays with the holes from the Array construction stub.

Committed: http://code.google.com/p/v8/source/detail?r=3995

[antonm, 2010-02-27 12:41:36]

Mads, Søren,

may you have a look?

[Søren Thygesen Gjesse, 2010-03-01 12:50:50]

LGTM

I actually think we are missing a cld instruction before rep_movs in
StringStubBase::GenerateCopyCharactersREP.

http://codereview.chromium.org/661245/diff/1/4
File src/ia32/builtins-ia32.cc (right):

http://codereview.chromium.org/661245/diff/1/4#newcode806
src/ia32/builtins-ia32.cc:806: Label* gc_required) {
Please ASSERT array_size.is(ecx) at the begining of the function. Also make a
comment about it.

The same regarding the assumptions about array_function and scratch, see
StringStubBase::GenerateCopyCharactersREP()

[antonm, 2010-03-01 16:01:49]

http://codereview.chromium.org/661245/diff/1/4
File src/ia32/builtins-ia32.cc (right):

http://codereview.chromium.org/661245/diff/1/4#newcode806
src/ia32/builtins-ia32.cc:806: Label* gc_required) {
On 2010/03/01 12:50:50, Søren Gjesse wrote:
> Please ASSERT array_size.is(ecx) at the begining of the function. Also make a
> comment about it.
> 
> The same regarding the assumptions about array_function and scratch, see
> StringStubBase::GenerateCopyCharactersREP()

Søren,

I tried to do that, but there is a catch: for the case when > 1 arguments are
passed into Array's function, size of array is in eax (number of arguments) vs.
ecx (case of a single argument).  So to assert that for all the cases, I need to
move eax to ecx.  So for now I just added conditional assert, but it looks
slightly ugly to me, so I'd overall prefer to keep those asserts under if
(fill_with_hole).

Which solution do you prefer?

And I'll take care of CLD in another CL.