Vector ICs: Make the Oracle gather feedback for vector stores.

Vector ICs: Make the Oracle gather feedback for vector stores.

Also, polymorphic element stores have a slightly different shape for the array
attached to a vector slot. It's of the form [map, map, handler], where the 2nd
map is either a transition map or undefined (the maps are actually in
WeakCells).

Committed: https://crrev.com/0302e63e7f467a677466653378449d00ea096c56
Cr-Commit-Position: refs/heads/master@{#30432}

[mvstanton, 2015-08-27 12:20:45]

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

[mvstanton, 2015-08-27 12:20:45]

Hi Benedikt, here is the change we discussed. Thanks for the suggestion on
regularity of the array, I've addressed that.
--Michael

[Benedikt Meurer, 2015-08-28 05:48:05]

LGTM, however there's one major question that I'm not sure about:

Both the receiver map and the transition map is in a weak cell. Now
what happens if the receiver map is still alive, but the transition map
dies? I guess we currently handle this with special code in the IC
handler. But ideally the transition maps would be strong referenced
as long as the receiver map is alive, right? So can we use some kind
of Ephemeron for this, i.e. a WeakMap with the receiver map as key
and (transition map, handler) as value?

[mvstanton, 2015-08-28 08:25:28]

Hi Benedikt, thanks!

Okay, I just checked the ic code, and they will MISS if the transitioned map has
disappeared from the weak cell. I'll do the same in the vector dispatcher.
--Michael

https://codereview.chromium.org/1316953003/diff/1/src/type-info.cc
File src/type-info.cc (right):

https://codereview.chromium.org/1316953003/diff/1/src/type-info.cc#newcode86
src/type-info.cc:86: obj->IsSimd128Value() || (FLAG_vector_stores &&
obj->IsMap())) {
I refactored StoreIsUninitialized(FeedbackVectorICSlot) to be more
"vector-style" code. This eliminated the need for this strange flag-based
addition to the conditional above. The only purpose for that was to send a
signal that there is "something interesting" in the vector slot. Good vector
code should better use a Nexus for those decisions.

[Benedikt Meurer, 2015-08-28 08:57:39]

Still LGTM!

[mvstanton, 2015-08-28 08:59:55]

The CQ bit was checked by mvstanton@chromium.org

[commit-bot: I haz the power, 2015-08-28 09:00:04]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1316953003/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1316953003/20001

[commit-bot: I haz the power, 2015-08-28 09:01:27]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2015-08-28 09:01:43]

Patchset 2 (id:??) landed as
https://crrev.com/0302e63e7f467a677466653378449d00ea096c56
Cr-Commit-Position: refs/heads/master@{#30432}