On-the-fly bookkeeping of PagedSpace memory kept in free-lists.

On-the-fly bookkeeping of PagedSpace memory kept in free-lists.


BUG=

Committed: https://code.google.com/p/v8/source/detail?r=14197

[Hannes Payer (out of office), 2013-04-08 16:30:38]

Golem does not report any performance regressions.
http://108.170.219.0:8080/Comparison#targetA%3Dv8%3BmachineTypeA%3Dlinux-ia32...

[Michael Starzinger, 2013-04-09 16:56:04]

Looking good already ...

https://codereview.chromium.org/13798002/diff/2002/src/spaces.cc
File src/spaces.cc (right):

https://codereview.chromium.org/13798002/diff/2002/src/spaces.cc#newcode2225
src/spaces.cc:2225:
Page::FromAddress(node->address())->AddAvailableInSmallFreeList(
nit: Can we pull out the page into a local "page" variable here. I think that
increases readability. Same goes for the four occurrences below.

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h
File src/spaces.h (right):

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h#newcode551
src/spaces.h:551: 5 * kPointerSize;
As discussed offline: Can we check whether that still fits into the current page
header alignment, or if it bumps the size of the page header? If it does bump
the size we have wiggle-room with the unused "non_available_small_blocks" field
and there should be some room to squeeze the other fields together.

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h#newcode728
src/spaces.h:728: Page();
There should be no constructor for Pages, they should only be constructed
through the FromAddress() accessors.

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h#newcode808
src/spaces.h:808: void set_available_in_small_free_list(intptr_t
available_in_small_free_list) {
Just an idea. We could use macros to generate this long tail of accessors. I
would prefer the macro approach here, but I'll leave the final decision up to
you.

[Hannes Payer (out of office), 2013-04-10 07:53:49]

https://codereview.chromium.org/13798002/diff/2002/src/spaces.cc
File src/spaces.cc (right):

https://codereview.chromium.org/13798002/diff/2002/src/spaces.cc#newcode2225
src/spaces.cc:2225:
Page::FromAddress(node->address())->AddAvailableInSmallFreeList(
On 2013/04/09 16:56:04, Michael Starzinger wrote:
> nit: Can we pull out the page into a local "page" variable here. I think that
> increases readability. Same goes for the four occurrences below.

Done.

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h
File src/spaces.h (right):

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h#newcode551
src/spaces.h:551: 5 * kPointerSize;
On 2013/04/09 16:56:04, Michael Starzinger wrote:
> As discussed offline: Can we check whether that still fits into the current
page
> header alignment, or if it bumps the size of the page header? If it does bump
> the size we have wiggle-room with the unused "non_available_small_blocks"
field
> and there should be some room to squeeze the other fields together.

It still fits.

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h#newcode728
src/spaces.h:728: Page();
On 2013/04/09 16:56:04, Michael Starzinger wrote:
> There should be no constructor for Pages, they should only be constructed
> through the FromAddress() accessors.

Done.

https://codereview.chromium.org/13798002/diff/2002/src/spaces.h#newcode808
src/spaces.h:808: void set_available_in_small_free_list(intptr_t
available_in_small_free_list) {
On 2013/04/09 16:56:04, Michael Starzinger wrote:
> Just an idea. We could use macros to generate this long tail of accessors. I
> would prefer the macro approach here, but I'll leave the final decision up to
> you.

Done.

[Michael Starzinger, 2013-04-10 07:59:31]

LGTM with one final nit.

https://codereview.chromium.org/13798002/diff/10001/src/spaces.h
File src/spaces.h (right):

https://codereview.chromium.org/13798002/diff/10001/src/spaces.h#newcode825
src/spaces.h:825: intptr_t non_available_small_blocks_;
Let's move the fields to MemoryChunk, because that's where the actual header
size is also computed. The accessors can stay here.

[Hannes Payer (out of office), 2013-04-10 08:05:57]

https://codereview.chromium.org/13798002/diff/10001/src/spaces.h
File src/spaces.h (right):

https://codereview.chromium.org/13798002/diff/10001/src/spaces.h#newcode825
src/spaces.h:825: intptr_t non_available_small_blocks_;
On 2013/04/10 07:59:31, Michael Starzinger wrote:
> Let's move the fields to MemoryChunk, because that's where the actual header
> size is also computed. The accessors can stay here.

Done.

[Hannes Payer (out of office), 2013-04-10 08:08:08]

Committed patchset #6 manually as r14197 (presubmit successful).