Use an allocation site scratchpad to speed up allocaton site processing during gc.

Use an allocation site scratchpad to speed up allocaton site processing during gc.

BUG=
R=mvstanton@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=18367

[Hannes Payer (out of office), 2013-12-12 15:26:05]

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc
File src/heap.cc (right):

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc#newcode518
src/heap.cc:518: if (use_scratchpad) {
Should I make an Iterator for that duplicated piece of code?

[mvstanton, 2013-12-13 10:30:17]

Looks good. Lemme see the iterator.

https://codereview.chromium.org/99133017/diff/60001/src/heap-inl.h
File src/heap-inl.h (right):

https://codereview.chromium.org/99133017/diff/60001/src/heap-inl.h#newcode497
src/heap-inl.h:497: add_to_scratchpad) {
put add_to_scratchpad first in the if expression.

https://codereview.chromium.org/99133017/diff/60001/src/heap-inl.h#newcode499
src/heap-inl.h:499: heap->allocation_sites_scratchpad_pointer] =
nit: I'd be into the more compact
heap->allocation_sites_scratchpad[heap->allocation_sites_scratchpad_pointer++] =
memento->GetAllocationSite(). But it's up to you!

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc
File src/heap.cc (right):

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc#newcode151
src/heap.cc:151: allocation_sites_scratchpad_pointer(0),
How about using the word count or length instead of pointer for the name.

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc#newcode518
src/heap.cc:518: if (use_scratchpad) {
On 2013/12/12 15:26:05, Hannes Payer wrote:
> Should I make an Iterator for that duplicated piece of code?

Yep that is a good idea. Alternatively, a function that accepts an
AllocationSite, but I see it would have to accept all these statistics variables
which wouldn't be fun.

https://codereview.chromium.org/99133017/diff/60001/src/objects.h
File src/objects.h (right):

https://codereview.chromium.org/99133017/diff/60001/src/objects.h#newcode8138
src/objects.h:8138: inline bool IncrementMementoFoundCount();
I guess this needs a comment to explain the motivation and meaning of the bool
return.

[Hannes Payer (out of office), 2013-12-18 14:29:18]

https://codereview.chromium.org/99133017/diff/60001/src/heap-inl.h
File src/heap-inl.h (right):

https://codereview.chromium.org/99133017/diff/60001/src/heap-inl.h#newcode497
src/heap-inl.h:497: add_to_scratchpad) {
On 2013/12/13 10:30:17, mvstanton wrote:
> put add_to_scratchpad first in the if expression.

Done.

https://codereview.chromium.org/99133017/diff/60001/src/heap-inl.h#newcode499
src/heap-inl.h:499: heap->allocation_sites_scratchpad_pointer] =
On 2013/12/13 10:30:17, mvstanton wrote:
> nit: I'd be into the more compact
> heap->allocation_sites_scratchpad[heap->allocation_sites_scratchpad_pointer++]
=
> memento->GetAllocationSite(). But it's up to you!

Done.

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc
File src/heap.cc (right):

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc#newcode151
src/heap.cc:151: allocation_sites_scratchpad_pointer(0),
On 2013/12/13 10:30:17, mvstanton wrote:
> How about using the word count or length instead of pointer for the name.

Done.

https://codereview.chromium.org/99133017/diff/60001/src/heap.cc#newcode518
src/heap.cc:518: if (use_scratchpad) {
On 2013/12/13 10:30:17, mvstanton wrote:
> On 2013/12/12 15:26:05, Hannes Payer wrote:
> > Should I make an Iterator for that duplicated piece of code?
> 
> Yep that is a good idea. Alternatively, a function that accepts an
> AllocationSite, but I see it would have to accept all these statistics
variables
> which wouldn't be fun.

I decided to do the simplified version, not beautiful but an extra iterator is
overkill. WDYT?

https://codereview.chromium.org/99133017/diff/60001/src/objects.h
File src/objects.h (right):

https://codereview.chromium.org/99133017/diff/60001/src/objects.h#newcode8138
src/objects.h:8138: inline bool IncrementMementoFoundCount();
On 2013/12/13 10:30:17, mvstanton wrote:
> I guess this needs a comment to explain the motivation and meaning of the bool
> return. 

Done.

[mvstanton, 2013-12-18 14:34:59]

lgtm

[Hannes Payer (out of office), 2013-12-18 21:24:04]

Committed patchset #6 manually as r18367 (presubmit successful).