Replace DeoptimizeOnUndefined with whitelisting AllowUndefinedAsNan

Replace DeoptimizeOnUndefined with whitelisting AllowUndefinedAsNan

R=danno@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=14894

[Toon Verwaest, 2013-05-29 13:58:25]

PTAL

[danno, 2013-05-29 15:33:16]

https://chromiumcodereview.appspot.com/15952007/diff/13/src/hydrogen.cc
File src/hydrogen.cc (right):

https://chromiumcodereview.appspot.com/15952007/diff/13/src/hydrogen.cc#newco...
src/hydrogen.cc:1589: holey_store->SetFlag(HValue::kAllowUndefinedAsNaN);
Maybe a comment, e.g. "Allow NaN hole values to be converted to their tagged
counterparts."  Also, maybe add an if? e.g.

if (IsFastHoleyDoubleElementsKind(to_element_kind)) {
  holey_store->SetFlag(HValue::kAllowUndefinedAsNaN);
}

https://chromiumcodereview.appspot.com/15952007/diff/13/src/hydrogen.cc#newco...
src/hydrogen.cc:10109: store->SetFlag(HValue::kAllowUndefinedAsNaN);
Are you sure you need this? Since there is no change involved, I don't see why
this flag needs to be cleared.

[Toon Verwaest, 2013-05-29 16:19:58]

Addressed comments.

https://chromiumcodereview.appspot.com/15952007/diff/13/src/hydrogen.cc
File src/hydrogen.cc (right):

https://chromiumcodereview.appspot.com/15952007/diff/13/src/hydrogen.cc#newco...
src/hydrogen.cc:1589: holey_store->SetFlag(HValue::kAllowUndefinedAsNaN);
On 2013/05/29 15:33:16, danno wrote:
> Maybe a comment, e.g. "Allow NaN hole values to be converted to their tagged
> counterparts."  Also, maybe add an if? e.g.
> 
> if (IsFastHoleyDoubleElementsKind(to_element_kind)) {
>   holey_store->SetFlag(HValue::kAllowUndefinedAsNaN);
> }

Done.

https://chromiumcodereview.appspot.com/15952007/diff/13/src/hydrogen.cc#newco...
src/hydrogen.cc:10109: store->SetFlag(HValue::kAllowUndefinedAsNaN);
On 2013/05/29 15:33:16, danno wrote:
> Are you sure you need this? Since there is no change involved, I don't see why
> this flag needs to be cleared.

Done.

[Toon Verwaest, 2013-05-29 16:38:38]

I undid my second change. The flag is required because otherwise the double-load
will deopt on the-hole-NaN, making copying holey double arrays fail.

[danno, 2013-05-30 08:43:42]

lgtm

[Toon Verwaest, 2013-05-30 09:11:16]

Committed patchset #4 manually as r14894 (presubmit successful).