Reduce the amount of full code generated to fill out array literals.

Reduce the amount of full code generated to fill out array literals.

This is achieved by tuning the calling convention of StoreArrayLiteralStub:

 * The map of the array literal can be loaded in the stub from the
   array literal itself, there is no need to pass it at all.

 * The array literal is already on the stack, so there is no need to
   pass it again via a register.

 * The literal index is unchanged while filling the literal, so we can
   push it on the stack once and avoid passing it every time. Note that
   we need to mirror this change in the stack layout in crankshaft, too.

R=mstarzinger@chromium.org

Committed: http://code.google.com/p/v8/source/detail?r=15138

[Sven Panne, 2013-06-13 12:46:15]

The auto-increment optimization we talked about offline is still missing, I will
add this later in this CL or a separate one.

[Michael Starzinger, 2013-06-13 13:57:27]

LGTM with comments.

https://codereview.chromium.org/16950004/diff/1/src/hydrogen.cc
File src/hydrogen.cc (right):

https://codereview.chromium.org/16950004/diff/1/src/hydrogen.cc#newcode6105
src/hydrogen.cc:6105: new(zone()) HConstant(expr->literal_index(),
Representation::Smi())));
I would just not specify a particular representation in HConstant constructor
and let the inferencer figure it out.

https://codereview.chromium.org/16950004/diff/1/src/ia32/full-codegen-ia32.cc
File src/ia32/full-codegen-ia32.cc (right):

https://codereview.chromium.org/16950004/diff/1/src/ia32/full-codegen-ia32.cc...
src/ia32/full-codegen-ia32.cc:1797: if (false &&
IsFastObjectElementsKind(constant_elements_kind)) {
This "false" looks like a debugging left-over.

[Sven Panne, 2013-06-14 06:04:32]

Comments addressed, landing...

https://codereview.chromium.org/16950004/diff/1/src/hydrogen.cc
File src/hydrogen.cc (right):

https://codereview.chromium.org/16950004/diff/1/src/hydrogen.cc#newcode6105
src/hydrogen.cc:6105: new(zone()) HConstant(expr->literal_index(),
Representation::Smi())));
On 2013/06/13 13:57:27, Michael Starzinger wrote:
> I would just not specify a particular representation in HConstant constructor
> and let the inferencer figure it out.

Done.

https://codereview.chromium.org/16950004/diff/1/src/ia32/full-codegen-ia32.cc
File src/ia32/full-codegen-ia32.cc (right):

https://codereview.chromium.org/16950004/diff/1/src/ia32/full-codegen-ia32.cc...
src/ia32/full-codegen-ia32.cc:1797: if (false &&
IsFastObjectElementsKind(constant_elements_kind)) {
On 2013/06/13 13:57:27, Michael Starzinger wrote:
> This "false" looks like a debugging left-over.

Done.

[Sven Panne, 2013-06-14 06:06:09]

Committed patchset #2 manually as r15138 (presubmit successful).