Remove (H|L)JSArrayLength instructions

Remove (H|L)JSArrayLength instructions


BUG=

Committed: https://code.google.com/p/v8/source/detail?r=14127

[Dmitry Lomov (no reviews), 2013-03-20 14:38:08]

Replace *JSArrayLength instructions with HLoadNamedField. Perf looks good.

[danno, 2013-03-22 15:13:41]

https://codereview.chromium.org/12491023/diff/1/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/12491023/diff/1/src/hydrogen-instructions.h#n...
src/hydrogen-instructions.h:5149: HLoadNamedField(HValue* object, bool
is_in_object, int offset)
I don't this is totally ok. I think you need to add a HValue* typecheck
parameter that prevents the length load from getting hoisted above other
instructions. You can leave it unused for other load variants.

[Dmitry Lomov (no reviews), 2013-03-28 14:07:15]

PTAL

[danno, 2013-04-02 11:10:43]

https://codereview.chromium.org/12491023/diff/5001/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (right):

https://codereview.chromium.org/12491023/diff/5001/src/hydrogen-instructions....
src/hydrogen-instructions.h:5171: result->SetGVNFlag(kDependsOnArrayLengths);
Clear the flag for kDependsOnInobjectFields, might require a slight bit of
restructuring.

https://codereview.chromium.org/12491023/diff/5001/src/hydrogen.cc
File src/hydrogen.cc (right):

https://codereview.chromium.org/12491023/diff/5001/src/hydrogen.cc#newcode1169
src/hydrogen.cc:1169: HLoadNamedField::NewArrayLength(zone, object, mapcheck));
I think you are losing the fact that the type of this instruction is a Smi. It
is used for optimization purposes later in the compilation pipeline. Some array
lengths are known to be Smis (if their map is guaranteed to fast elements), but
otherwise, they may be a tagged heap number, which will require more heavy
checks in the generated code. You will need to preserve the ability to specify
the return type of the load.

[Dmitry Lomov (no reviews), 2013-04-02 14:19:35]

Propagated the type, in course of that found a bug and fixed it, test included

[danno, 2013-04-03 15:49:44]

lgtm

[Dmitry Lomov (no reviews), 2013-04-03 16:25:45]

Committed patchset #4 manually as r14127 (presubmit successful).