cc: More consistent reasoning about display list memory usage.

cc: More consistent reasoning about display list memory usage.

This refines "picture memory usage" methods to be more consistent.
In particular:

* DisplayItem::external_memory_usage_ replaces picture_memory_usage_, and only
  refers to memory owned by the display item but not allocated within it (since
  that memory can already be accounted for by measuring the ListContainer).

* DisplayItemList separately tracks the memory allocated within items_,
  external memory usage due to those items, and memory usage due to the cached
  SkPicture, when one is used. These are added to produce total memory usage.

BUG=481083
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Committed: https://crrev.com/b0caed2bedfc19825ba34024c4678adf351d561c
Cr-Commit-Position: refs/heads/master@{#337911}

[jbroman, 2015-07-07 20:15:36]

jbroman@chromium.org changed reviewers:
+ enne@chromium.org

[jbroman, 2015-07-07 20:15:37]

If this looks good, my intention is for Blink to assume a similar methodology,
plumb the total there via ContentLayerDelegate -> WebContentLayerImpl ->
DisplayListRecordingSource -> DisplayListRasterSource, and then add it to the
memory metric we're recording.

This will make SP's apparent memory usage increase, but I think that's only
fair.

[enne (OOO), 2015-07-07 21:12:37]

lgtm

https://codereview.chromium.org/1226503006/diff/20001/cc/debug/rasterize_and_...
File cc/debug/rasterize_and_record_benchmark.cc (right):

https://codereview.chromium.org/1226503006/diff/20001/cc/debug/rasterize_and_...
cc/debug/rasterize_and_record_benchmark.cc:226: DCHECK(memory_used ==
display_list->ApproximateMemoryUsage());
DCHECK_EQ?

[jbroman, 2015-07-07 21:17:09]

The CQ bit was checked by jbroman@chromium.org

[jbroman, 2015-07-07 21:17:09]

The patchset sent to the CQ was uploaded after l-g-t-m from enne@chromium.org
Link to the patchset: https://codereview.chromium.org/1226503006/#ps60001
(title: "DCHECK_EQ")

[commit-bot: I haz the power, 2015-07-07 21:19:38]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1226503006/60001

[jbroman, 2015-07-07 21:39:03]

https://codereview.chromium.org/1226503006/diff/20001/cc/debug/rasterize_and_...
File cc/debug/rasterize_and_record_benchmark.cc (right):

https://codereview.chromium.org/1226503006/diff/20001/cc/debug/rasterize_and_...
cc/debug/rasterize_and_record_benchmark.cc:226: DCHECK(memory_used ==
display_list->ApproximateMemoryUsage());
On 2015/07/07 21:12:37, enne wrote:
> DCHECK_EQ?

Haha, the hazards of modifying an existing line of code. Done.

[jbroman, 2015-07-07 23:08:40]

On some platforms, this broke the (admittedly brittle) way that the test was
avoiding double-counting. Will investigate further tomorrow.

[commit-bot: I haz the power, 2015-07-07 23:54:32]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2015-07-07 23:54:33]

Try jobs failed on following builders:
  win_chromium_rel_ng on tryserver.chromium.win (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_...)

[jbroman, 2015-07-08 16:02:59]

Patchset #5 (id:80001) has been deleted

[jbroman, 2015-07-08 18:04:31]

Patchset #5 (id:100001) has been deleted

[jbroman, 2015-07-08 19:12:52]

Patchset #5 (id:120001) has been deleted

[jbroman, 2015-07-08 19:26:54]

PTAL at DisplayItemListTest.ApproximateMemoryUsage.

The crux of the issue is that it depends on implementation details of how both
SkRecord and ListContainer preallocate memory.

I have enlarged the SkPicture, so that the external memory is "enough" larger
than the ListContainer data to clearly tell whether SkPicture was
double-counted. I've also removed the upper-bound sanity check on the size of
that SkPicture, because SkRecord was allocating more memory than that (and so
the assertion was failing).

I'm not super thrilled, but I don't see an awesome way to make this less
brittle. WDYT?

[enne (OOO), 2015-07-08 20:20:15]

Yeah, this test.  :C

This seems pretty reasonable, though! lgtm

[jbroman, 2015-07-08 20:21:25]

The CQ bit was checked by jbroman@chromium.org

[commit-bot: I haz the power, 2015-07-08 20:22:12]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1226503006/140001

[commit-bot: I haz the power, 2015-07-08 21:27:18]

Committed patchset #5 (id:140001)

[commit-bot: I haz the power, 2015-07-08 21:28:12]

Patchset 5 (id:??) landed as
https://crrev.com/b0caed2bedfc19825ba34024c4678adf351d561c
Cr-Commit-Position: refs/heads/master@{#337911}