[turbofan] Initial version of allocation folding and write barrier elimination.

[turbofan] Initial version of allocation folding and write barrier elimination.

This adds a new pass MemoryOptimizer that walks over the effect chain
from Start and lowers all Allocate, LoadField, StoreField, LoadElement,
and StoreElement nodes, trying to fold allocations into allocation
groups and eliminate write barriers on StoreField and StoreElement if
possible (i.e. if the object belongs to the current allocation group and
that group allocates in new space).

R=hpayer@chromium.org, jarin@chromium.org
BUG=v8:4931, chromium:580959
LOG=n

Committed: https://crrev.com/b8229ec446b80fd3ff9a08362fe1cef36e9d8346
Cr-Commit-Position: refs/heads/master@{#36128}

[Benedikt Meurer, 2016-05-10 06:59:00]

Hey Hannes, Jaro,

The very first version.
Please take a look.

Thanks,
Benedikt

[Benedikt Meurer, 2016-05-10 07:26:40]

Description was changed from

==========
[turbofan] Initial version of allocation folding and write barrier elimination.

This adds a new pass MemoryOptimizer that walks over the effect chain
from Start and lowers all Allocate, LoadField, StoreField, LoadElement,
and StoreElement nodes, trying to fold allocations into allocation
groups and eliminate write barriers on StoreField and StoreElement if
possible (i.e. if the object belongs to the current allocation group and
that group allocates in new space).

R=hpayer@chromium.org, jarin@chromium.org
BUG=v8:4931
LOG=n
==========

to

==========
[turbofan] Initial version of allocation folding and write barrier elimination.

This adds a new pass MemoryOptimizer that walks over the effect chain
from Start and lowers all Allocate, LoadField, StoreField, LoadElement,
and StoreElement nodes, trying to fold allocations into allocation
groups and eliminate write barriers on StoreField and StoreElement if
possible (i.e. if the object belongs to the current allocation group and
that group allocates in new space).

R=hpayer@chromium.org, jarin@chromium.org
BUG=v8:4931, chromium:580959
LOG=n
==========

[Hannes Payer (out of office), 2016-05-10 07:49:33]

Awesome, LGTM! Just nits.

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
File src/compiler/memory-optimizer.cc (right):

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
src/compiler/memory-optimizer.cc:128: int32_t const state_size = state->size() +
object_size;
Hoist this computation and use state_size already in the if check above.

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
src/compiler/memory-optimizer.cc:229: // Load allocation top and limit.
This code is similar to the one in the else case above. I guess it would make
sense to refactor the common path of bump pointer allocation. WDYT?

https://codereview.chromium.org/1963583004/diff/1/src/heap/spaces.cc
File src/heap/spaces.cc (left):

https://codereview.chromium.org/1963583004/diff/1/src/heap/spaces.cc#oldcode2474
src/heap/spaces.cc:2474: DCHECK(owner_->top() == NULL && owner_->limit() ==
NULL);
These changes will be committed in the crankshaft change.

[Benedikt Meurer, 2016-05-10 08:01:45]

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
File src/compiler/memory-optimizer.cc (right):

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
src/compiler/memory-optimizer.cc:128: int32_t const state_size = state->size() +
object_size;
Cannot do because state->size() + object_size can overflow.

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
src/compiler/memory-optimizer.cc:229: // Load allocation top and limit.
It's just almost the same, but not 100%. For now I'd like to keep it this way
and refactor it once the top is tagged.

https://codereview.chromium.org/1963583004/diff/1/src/heap/spaces.cc
File src/heap/spaces.cc (left):

https://codereview.chromium.org/1963583004/diff/1/src/heap/spaces.cc#oldcode2474
src/heap/spaces.cc:2474: DCHECK(owner_->top() == NULL && owner_->limit() ==
NULL);
Yep, will rebase.

[Jarin, 2016-05-10 09:24:49]

Very nice! lgtm.

One thing I am not too wild about is the AllocationState constructors,
especially the one with just a group. Could not you give these guys some
descriptive name? (I do not have a good idea here, perhaps Empty/Open/Closed?)

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
File src/compiler/memory-optimizer.h (right):

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
src/compiler/memory-optimizer.h:36: void Optimize();
Don't we normally call these methods 'Run'?

[Benedikt Meurer, 2016-05-10 09:33:06]

On 2016/05/10 09:24:49, Jarin wrote:
> Very nice! lgtm.
> 
> One thing I am not too wild about is the AllocationState constructors,
> especially the one with just a group. Could not you give these guys some
> descriptive name? (I do not have a good idea here, perhaps Empty/Open/Closed?)

Did that. Thanks for the suggestion.

[Benedikt Meurer, 2016-05-10 09:34:04]

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
File src/compiler/memory-optimizer.h (right):

https://codereview.chromium.org/1963583004/diff/1/src/compiler/memory-optimiz...
src/compiler/memory-optimizer.h:36: void Optimize();
Followed the pattern of ControlFlowOptimizer here. Let's rename both if it's
important.

[Benedikt Meurer, 2016-05-10 09:34:10]

The CQ bit was checked by bmeurer@chromium.org

[Benedikt Meurer, 2016-05-10 09:34:11]

The patchset sent to the CQ was uploaded after l-g-t-m from hpayer@chromium.org,
jarin@chromium.org
Link to the patchset: https://codereview.chromium.org/1963583004/#ps20001
(title: "Jaros comments;")

[commit-bot: I haz the power, 2016-05-10 09:34:22]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1963583004/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1963583004/20001

[commit-bot: I haz the power, 2016-05-10 10:11:15]

Description was changed from

==========
[turbofan] Initial version of allocation folding and write barrier elimination.

This adds a new pass MemoryOptimizer that walks over the effect chain
from Start and lowers all Allocate, LoadField, StoreField, LoadElement,
and StoreElement nodes, trying to fold allocations into allocation
groups and eliminate write barriers on StoreField and StoreElement if
possible (i.e. if the object belongs to the current allocation group and
that group allocates in new space).

R=hpayer@chromium.org, jarin@chromium.org
BUG=v8:4931, chromium:580959
LOG=n
==========

to

==========
[turbofan] Initial version of allocation folding and write barrier elimination.

This adds a new pass MemoryOptimizer that walks over the effect chain
from Start and lowers all Allocate, LoadField, StoreField, LoadElement,
and StoreElement nodes, trying to fold allocations into allocation
groups and eliminate write barriers on StoreField and StoreElement if
possible (i.e. if the object belongs to the current allocation group and
that group allocates in new space).

R=hpayer@chromium.org, jarin@chromium.org
BUG=v8:4931, chromium:580959
LOG=n
==========

[commit-bot: I haz the power, 2016-05-10 10:11:17]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2016-05-10 10:12:33]

Description was changed from

==========
[turbofan] Initial version of allocation folding and write barrier elimination.

This adds a new pass MemoryOptimizer that walks over the effect chain
from Start and lowers all Allocate, LoadField, StoreField, LoadElement,
and StoreElement nodes, trying to fold allocations into allocation
groups and eliminate write barriers on StoreField and StoreElement if
possible (i.e. if the object belongs to the current allocation group and
that group allocates in new space).

R=hpayer@chromium.org, jarin@chromium.org
BUG=v8:4931, chromium:580959
LOG=n
==========

to

==========
[turbofan] Initial version of allocation folding and write barrier elimination.

This adds a new pass MemoryOptimizer that walks over the effect chain
from Start and lowers all Allocate, LoadField, StoreField, LoadElement,
and StoreElement nodes, trying to fold allocations into allocation
groups and eliminate write barriers on StoreField and StoreElement if
possible (i.e. if the object belongs to the current allocation group and
that group allocates in new space).

R=hpayer@chromium.org, jarin@chromium.org
BUG=v8:4931, chromium:580959
LOG=n

Committed: https://crrev.com/b8229ec446b80fd3ff9a08362fe1cef36e9d8346
Cr-Commit-Position: refs/heads/master@{#36128}
==========

[commit-bot: I haz the power, 2016-05-10 10:12:34]

Patchset 2 (id:??) landed as
https://crrev.com/b8229ec446b80fd3ff9a08362fe1cef36e9d8346
Cr-Commit-Position: refs/heads/master@{#36128}