DisplayList tracks maximum render op depths

[flar]

The DisplayList will now track the "render depth" of all rendering ops and provide the total content depth for DisplayList and save[Layer]/restore pairs to enable depth clipping in the back ends.

The depth accounting process is described in the class comment for DlOpReceiver

[flar]

I've updated the unit tests to double-check the max_depth values on the DisplayList in the existing unit tests. While I was in there I did some cleanup. There used to be 4 values on each "test snippet", as follows:

• display list op count
• display list byte count
• dl->skia picture->dl recapture op count
• dl->skia picture->dl recapture byte count

We haven't done "skia picture recapture" testing for a long time since we eliminated the need to convert Skia calls into DisplayList calls so the last 2 values have been obsolete for a while. I needed to add "depth" counts to all of the snippets, though, so while I was making that sweeping change I deleted the "sk recapture" numbers. The 3 numbers at the start of each snippet are now:

• display list op count
• display list byte count
• display list depth

While doing that, I also deleted all comments related to fudging the numbers to satisfy the Skia recapture logic in the tests.

[bdero]

As an aside: I'm not sure there would be any utility in hooking this feature up to the current version of the DL dispatcher. We should hook this up to the new DL dispatcher as it's being formed, because

1. Aiks/EntityPass will need to be almost entirely rewritten anyhow,
2. we need to keep Aiks working without the DL until we're able to migrate them to use the DL.

[bdero]

Ah, this is something I should have specified. SaveLayers have a draw call that also needs a unique depth. But they are drawn after all of their children have been rendered, so the depth we use for them must be greater than all contained draws.

I recommend replicating this test against the DL depth values to ensure that:

1. SaveLayers are reserved a depth value, and
2. clips interact correctly with SaveLayers.

[flar]

I noticed that while I was updating dl_dispatcher.cc...

The odd thing is that they only allocate the depth on the restore after recording the max_depth, right? The clips inside them will be rendered to the "restore copy - 1" max depth and then the restore copy happens above that depth so it is only clipped by the enclosing clips.

[flar]

I think I've accounted for this in the latest commit.

[flar]

As an aside: I'm not sure there would be any utility in hooking this feature up to the current version of the DL dispatcher. We should hook this up to the new DL dispatcher as it's being formed, because

1. Aiks/EntityPass will need to be almost entirely rewritten anyhow,
2. we need to keep Aiks working without the DL until we're able to migrate them to use the DL.

I prototyped it, but now I'm working through adjusting the new max_depth testing in the DL unit tests to account for saveLayer depths.

And, in the end, the prototype was only useful for checking to see how it could be used. Since the impeller::Canvas class is used internally for a lot of unit tests which don't supply the new depth information, the new code isn't directly useful without a lot of modifications to the unit tests. It is possible we could do something in the future such that if the depth information was provided we keep it, but if not then we compute it, but it isn't important for this first step.

[jonahwilliams]

LGTM

[bdero]

Could you add a test that checks the depth of clip ops added to both the root and to the root and within a SaveLayer? That way we know they close with the correct depth (max depth of everything they affect).

[flar]

What is the "depth of clip ops"? They aren't accounted in DL. That's an Impeller thing...?

All DL does is count how many render calls are within a save/restore or an entire DL. Clips are invisible to that.

[flar]

As in the class comment in dl_op_receiver, the only places depth information is reported is the DL total_depth and the argument passed to save/saveLayer.