Preserve Smi representation of non-escaping fields.

Preserve Smi representation of non-escaping fields.

R=mstarzinger@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=20971

[Michael Starzinger, 2014-04-25 08:35:09]

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-escape-analysis.cc
File src/hydrogen-escape-analysis.cc (right):

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-escape-analysis...
src/hydrogen-escape-analysis.cc:202: if (representation.IsSmi()) {
Can we factor this out into a NewLoadReplacement helper (similar to the
NewMapCheckAndInsert) to preserve readability of the giant AnalyzeDataFlow
method?

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-escape-analysis...
src/hydrogen-escape-analysis.cc:212: load_value->id(), load_value->Mnemonic());
nit: The logging should print the actual replacement IMHO.

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (left):

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-instructions.h#...
src/hydrogen-instructions.h:1731: private:
Let's keep this constructor private and use HForceRepresentation::New instead,
that gives us implicit constant folding.

[Jarin, 2014-04-25 11:05:07]

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-escape-analysis.cc
File src/hydrogen-escape-analysis.cc (right):

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-escape-analysis...
src/hydrogen-escape-analysis.cc:202: if (representation.IsSmi()) {
On 2014/04/25 08:35:10, Michael Starzinger wrote:
> Can we factor this out into a NewLoadReplacement helper (similar to the
> NewMapCheckAndInsert) to preserve readability of the giant AnalyzeDataFlow
> method?

Done.

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-escape-analysis...
src/hydrogen-escape-analysis.cc:212: load_value->id(), load_value->Mnemonic());
On 2014/04/25 08:35:10, Michael Starzinger wrote:
> nit: The logging should print the actual replacement IMHO.

Done.

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-instructions.h
File src/hydrogen-instructions.h (left):

https://codereview.chromium.org/251493004/diff/1/src/hydrogen-instructions.h#...
src/hydrogen-instructions.h:1731: private:
On 2014/04/25 08:35:10, Michael Starzinger wrote:
> Let's keep this constructor private and use HForceRepresentation::New instead,
> that gives us implicit constant folding.

Done.

[Michael Starzinger, 2014-04-25 11:08:09]

LGTM.

https://codereview.chromium.org/251493004/diff/20001/src/hydrogen-escape-anal...
File src/hydrogen-escape-analysis.cc (right):

https://codereview.chromium.org/251493004/diff/20001/src/hydrogen-escape-anal...
src/hydrogen-escape-analysis.cc:177: load->DeleteAndReplaceWith(replacement);
nit: I was actually thinking along the lines of having the DeleteAndReplaceWith
plus the printing outside. And the helper would really just be a
"NewLoadReplacement" that returns some other node as a replacement. Thereby
mutation of the existing graph is in the data-flow analysis code.

[Jarin, 2014-04-25 11:29:10]

Committed patchset #4 manually as r20971 (presubmit successful).