cc: Construct the RSLL without using layer hierarchy

cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-level list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799, 603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Committed: https://crrev.com/5c8f5db211972b428577a3e268d25dccef61c366
Cr-Commit-Position: refs/heads/master@{#388233}

[ajuma, 2016-04-14 23:24:51]

Description was changed from

==========
WIP

BUG=
==========

to

==========
WIP

BUG=
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

[ajuma, 2016-04-15 13:15:55]

Patchset #1 (id:1) has been deleted

[ajuma, 2016-04-15 13:26:31]

Description was changed from

==========
WIP

BUG=
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

to

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the list of surfaces in order to compute
their content rects, and then iterate over the list of surfaces to
produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

[ajuma, 2016-04-15 13:34:19]

Description was changed from

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the list of surfaces in order to compute
their content rects, and then iterate over the list of surfaces to
produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

to

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-leve list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

[ajuma, 2016-04-15 13:34:30]

Description was changed from

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-leve list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

to

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-level list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

[ajuma, 2016-04-15 13:57:01]

ajuma@chromium.org changed reviewers:
+ weiliangc@chromium.org

[ajuma, 2016-04-15 14:17:12]

Patchset #1 (id:20001) has been deleted

[ajuma, 2016-04-15 17:32:38]

https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
File cc/trees/layer_tree_host_common.cc (right):

https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
cc/trees/layer_tree_host_common.cc:260: int
current_render_surface_layer_list_id) {
Using a sequence number to track which layers are drawn RSLL members doesn't
seem to get us much, now that every layer gets visited anyway. I'm planning on
replacing this with a bool in a follow-up, but I can do that here too if you'd
prefer.

https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
cc/trees/layer_tree_host_common.cc:681: target_list.erase(it);
One alternative to erasing this way would be to have preliminary and final layer
lists at each surface. But I'd expect surfaces that initially get put into the
RSLL but later turn out to have an empty content rect to be rare, so avoiding
the complexity of two lists at each surface seemed like a better approach.

[weiliangc, 2016-04-19 15:05:59]

Overall looks reasonable. Thanks.

https://codereview.chromium.org/1882413006/diff/40001/cc/layers/render_surfac...
File cc/layers/render_surface_impl.h (right):

https://codereview.chromium.org/1882413006/diff/40001/cc/layers/render_surfac...
cc/layers/render_surface_impl.h:64: RenderSurfaceImpl*
nearest_occlusion_immune_ancestor() {
Is this non-const version really needed?

https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
File cc/trees/layer_tree_host_common.cc (right):

https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
cc/trees/layer_tree_host_common.cc:681: target_list.erase(it);
On 2016/04/15 at 17:32:37, ajuma wrote:
> One alternative to erasing this way would be to have preliminary and final
layer lists at each surface. But I'd expect surfaces that initially get put into
the RSLL but later turn out to have an empty content rect to be rare, so
avoiding the complexity of two lists at each surface seemed like a better
approach.

At this point, target_list might be empty, and we would need to remove target as
well. Maybe we should walk the initial_surface_list backwards? Though that would
mean we have to erase instead having a copy of final_surface_list.

[ajuma, 2016-04-19 16:35:13]

https://codereview.chromium.org/1882413006/diff/40001/cc/layers/render_surfac...
File cc/layers/render_surface_impl.h (right):

https://codereview.chromium.org/1882413006/diff/40001/cc/layers/render_surfac...
cc/layers/render_surface_impl.h:64: RenderSurfaceImpl*
nearest_occlusion_immune_ancestor() {
On 2016/04/19 15:05:59, weiliangc wrote:
> Is this non-const version really needed?

Removed. Turns out I just needed to make SetNearestOcclusionImmuneAncestor take
a const pointer.

https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
File cc/trees/layer_tree_host_common.cc (right):

https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
cc/trees/layer_tree_host_common.cc:681: target_list.erase(it);
On 2016/04/19 15:05:59, weiliangc wrote:
> On 2016/04/15 at 17:32:37, ajuma wrote:
> > One alternative to erasing this way would be to have preliminary and final
> layer lists at each surface. But I'd expect surfaces that initially get put
into
> the RSLL but later turn out to have an empty content rect to be rare, so
> avoiding the complexity of two lists at each surface seemed like a better
> approach.
> 
> At this point, target_list might be empty, and we would need to remove target
as
> well. Maybe we should walk the initial_surface_list backwards? Though that
would
> mean we have to erase instead having a copy of final_surface_list.

I don't think this call can make the target_list be empty (unless the target
list is the root, in which case we don't want to remove the target anyway). If
we reach here, the surface must have an empty content rect, so if this was the
only thing in the target list, the target would have an empty content rect too,
and would already have been removed when we visited it.

I've added a DCHECK for this.

[weiliangc, 2016-04-19 17:03:42]

On 2016/04/19 at 16:35:13, ajuma wrote:
>
https://codereview.chromium.org/1882413006/diff/40001/cc/layers/render_surfac...
> File cc/layers/render_surface_impl.h (right):
> 
>
https://codereview.chromium.org/1882413006/diff/40001/cc/layers/render_surfac...
> cc/layers/render_surface_impl.h:64: RenderSurfaceImpl*
nearest_occlusion_immune_ancestor() {
> On 2016/04/19 15:05:59, weiliangc wrote:
> > Is this non-const version really needed?
> 
> Removed. Turns out I just needed to make SetNearestOcclusionImmuneAncestor
take a const pointer.
> 
>
https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
> File cc/trees/layer_tree_host_common.cc (right):
> 
>
https://codereview.chromium.org/1882413006/diff/40001/cc/trees/layer_tree_hos...
> cc/trees/layer_tree_host_common.cc:681: target_list.erase(it);
> On 2016/04/19 15:05:59, weiliangc wrote:
> > On 2016/04/15 at 17:32:37, ajuma wrote:
> > > One alternative to erasing this way would be to have preliminary and final
> > layer lists at each surface. But I'd expect surfaces that initially get put
into
> > the RSLL but later turn out to have an empty content rect to be rare, so
> > avoiding the complexity of two lists at each surface seemed like a better
> > approach.
> > 
> > At this point, target_list might be empty, and we would need to remove
target as
> > well. Maybe we should walk the initial_surface_list backwards? Though that
would
> > mean we have to erase instead having a copy of final_surface_list.
> 
> I don't think this call can make the target_list be empty (unless the target
list is the root, in which case we don't want to remove the target anyway). If
we reach here, the surface must have an empty content rect, so if this was the
only thing in the target list, the target would have an empty content rect too,
and would already have been removed when we visited it.
> 
> I've added a DCHECK for this.

Ah right, makes sense. LGTM

[ajuma, 2016-04-19 17:05:37]

The CQ bit was checked by ajuma@chromium.org

[commit-bot: I haz the power, 2016-04-19 17:05:56]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1882413006/80001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1882413006/80001

[commit-bot: I haz the power, 2016-04-19 17:29:20]

Description was changed from

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-level list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

to

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-level list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

[commit-bot: I haz the power, 2016-04-19 17:29:21]

Committed patchset #3 (id:80001)

[commit-bot: I haz the power, 2016-04-22 19:13:20]

Description was changed from

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-level list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel
==========

to

==========
cc: Construct the RSLL without using layer hierarchy

This removes recursion over the layer tree from the construction of the
render surface layer list. Instead, we iterate over the layers in
stacking order to produce an initial render surface layer list (that
might include surfaces that later turn out to have empty content rects),
then iterate backwards over the top-level list of surfaces in order to
compute their content rects, and then iterate over the top-level list of
surfaces to produce the final RSLL that includes only non-empty surfaces.

BUG=568799,603147
CQ_INCLUDE_TRYBOTS=tryserver.blink:linux_blink_rel

Committed: https://crrev.com/5c8f5db211972b428577a3e268d25dccef61c366
Cr-Commit-Position: refs/heads/master@{#388233}
==========

[commit-bot: I haz the power, 2016-04-22 19:13:21]

Patchset 3 (id:??) landed as
https://crrev.com/5c8f5db211972b428577a3e268d25dccef61c366
Cr-Commit-Position: refs/heads/master@{#388233}