- Use the unused top of old-space pages when allocating.

- Use the free top of old-space pages when allocating.
Committed: https://code.google.com/p/dart/source/detail?r=6405

[ricow1, 2012-04-11 10:23:30]

LGTM, but doesn't this mean that we will iterate all heap pages continuously
when for every allocation in some cases. Example, we are allocation a ton of
objects of size between  kAllocatablePageSize / 2 + 1 and kAllocatablePageSize.
Assuming that we don't have room anywhere we will first traverse all pages, then
allocate a new page resetting bump_page to NULL, and on the next allocation do
it again..

https://chromiumcodereview.appspot.com/10057001/diff/1/vm/pages.cc
File vm/pages.cc (right):

https://chromiumcodereview.appspot.com/10057001/diff/1/vm/pages.cc#newcode290
vm/pages.cc:290: in_use += page_in_use;
Why this change? (easier debugging?)

[Ivan Posva, 2012-04-11 11:53:39]

On 2012/04/11 10:23:30, ricow1 wrote:
> LGTM, but doesn't this mean that we will iterate all heap pages continuously
> when for every allocation in some cases. Example, we are allocation a ton of
> objects of size between  kAllocatablePageSize / 2 + 1 and
kAllocatablePageSize.
> Assuming that we don't have room anywhere we will first traverse all pages,
then
> allocate a new page resetting bump_page to NULL, and on the next allocation do
> it again..

Remember that we only allocate in old-gen when promoting objects from new to
old. Direct allocation into old is very infrequent. If the scenario you outline
really becomes a problem, then we could add largest available bump space
tracking. This is a complication that I do not want to introduce before
addressing other bigger issues with the growth policy.

-Ivan

> 
> https://chromiumcodereview.appspot.com/10057001/diff/1/vm/pages.cc
> File vm/pages.cc (right):
> 
> https://chromiumcodereview.appspot.com/10057001/diff/1/vm/pages.cc#newcode290
> vm/pages.cc:290: in_use += page_in_use;
> Why this change? (easier debugging?)

[Ivan Posva, 2012-04-11 11:56:52]

https://chromiumcodereview.appspot.com/10057001/diff/1/vm/pages.cc
File vm/pages.cc (right):

https://chromiumcodereview.appspot.com/10057001/diff/1/vm/pages.cc#newcode290
vm/pages.cc:290: in_use += page_in_use;
On 2012/04/11 10:23:30, ricow1 wrote:
> Why this change? (easier debugging?)

Yes, and preparation for the next step. Freeing of normal pages if they are
completely empty.

[cshapiro, 2012-04-11 19:55:21]

If we are planning on doing compaction this is probably an okay policy.  If not,
it is an opportunity for increased fragmentation.