VectorICs: use a vector slot to aid in array literal processing.

VectorICs: use a vector slot to aid in array literal processing.

The lack of a vector slot for the keyed store operation in filling in
non-constant array literal properties led to undesirable contortions in
compilers downwind of full-codegen. The use of a single slot to initialize all
the array elements is sufficient.

BUG=

Committed: https://crrev.com/f2f46aff8b776365e0bafd7826f23c8b9c16d928
Cr-Commit-Position: refs/heads/master@{#31242}

[mvstanton, 2015-10-13 09:47:22]

mvstanton@chromium.org changed reviewers:
+ mstarzinger@chromium.org

[mvstanton, 2015-10-13 09:47:22]

Hi Michael,
Here is the CL that adds a vector slot for the store case in VisitArrayLiteral.
It simplies much code downwind, and helps out the interpreter effort too.

Looks like I might be exposing an existing bug in mips/mips64 __ RecordWrite()
implementation. I'm investigating that issue so take your time...
Thanks,
--Michael

[Michael Starzinger, 2015-10-13 11:59:18]

https://codereview.chromium.org/1405503002/diff/40001/src/compiler/ast-graph-...
File src/compiler/ast-graph-builder.cc (right):

https://codereview.chromium.org/1405503002/diff/40001/src/compiler/ast-graph-...
src/compiler/ast-graph-builder.cc:1931: VectorSlotPair pair =
CreateVectorSlotPair(expr->LiteralFeedbackSlot());
nit: Can we move this into the loop (ideally yo line 1940) so that
ArrayLiteral::LiteralFeedbackSlot is only queried when we actually have a slot?

https://codereview.chromium.org/1405503002/diff/40001/src/full-codegen/arm/fu...
File src/full-codegen/arm/full-codegen-arm.cc (right):

https://codereview.chromium.org/1405503002/diff/40001/src/full-codegen/arm/fu...
src/full-codegen/arm/full-codegen-arm.cc:1756:
expr->BuildConstantElements(isolate());
This shouldn't be needed anymore, since the numbering-phase takes care of it,
correct? If so, then this applies to all architecture ports.

[mvstanton, 2015-10-13 12:56:10]

Thanks Michael, here is response plus turning off vector stores. I'll keep
trying to reproduce those flaky runs, it's likely a side effect from some
sub-component of --vector-stores which hides a lot of code.
--Michael

https://codereview.chromium.org/1405503002/diff/40001/src/compiler/ast-graph-...
File src/compiler/ast-graph-builder.cc (right):

https://codereview.chromium.org/1405503002/diff/40001/src/compiler/ast-graph-...
src/compiler/ast-graph-builder.cc:1931: VectorSlotPair pair =
CreateVectorSlotPair(expr->LiteralFeedbackSlot());
On 2015/10/13 11:59:18, Michael Starzinger wrote:
> nit: Can we move this into the loop (ideally yo line 1940) so that
> ArrayLiteral::LiteralFeedbackSlot is only queried when we actually have a
slot?

Done.

https://codereview.chromium.org/1405503002/diff/40001/src/full-codegen/arm/fu...
File src/full-codegen/arm/full-codegen-arm.cc (right):

https://codereview.chromium.org/1405503002/diff/40001/src/full-codegen/arm/fu...
src/full-codegen/arm/full-codegen-arm.cc:1756:
expr->BuildConstantElements(isolate());
On 2015/10/13 11:59:18, Michael Starzinger wrote:
> This shouldn't be needed anymore, since the numbering-phase takes care of it,
> correct? If so, then this applies to all architecture ports.

Done.

[Michael Starzinger, 2015-10-13 14:44:35]

LGTM.

[mvstanton, 2015-10-13 14:46:49]

The CQ bit was checked by mvstanton@chromium.org

[commit-bot: I haz the power, 2015-10-13 14:47:04]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1405503002/60001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1405503002/60001

[commit-bot: I haz the power, 2015-10-13 14:48:39]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2015-10-13 14:48:54]

Patchset 4 (id:??) landed as
https://crrev.com/f2f46aff8b776365e0bafd7826f23c8b9c16d928
Cr-Commit-Position: refs/heads/master@{#31242}