Add picture caching support.

[gw3583]

This change is 

[gw3583]

r? @kvark

This patch has 3 commits - the first two are simple bug fixes and functionality, and shouldn't be controversial.

The last commit has a couple of horrible hacks in it to enable the first version of picture caching (when the renderer option is enabled, which is off by default), where the code relies on knowledge of what the Gecko display list looks like.

I don't like this change at all, but I think that getting it merged while we work out a better solution (during the all-hands) has a number of benefits:

• It's off by default, so shouldn't affect anyone.
• It lets us dogfood the first iteration of picture caching internally by flipping a preference, which allows us to find bugs, compare performance etc.

[kvark]

hopefully, this is temporary :)

Reviewed 2 of 2 files at r1, 1 of 1 files at r2, 1 of 1 files at r3.
Reviewable status: all files reviewed, 3 unresolved discussions (waiting on @gw3583)

---

webrender/src/display_list_flattener.rs, line 252 at r3 (raw file):

        }

        // This method is basically a hack to set up picture caching in a minimal

I can see how all of this would be removed if Gecko could just tell us what to cache

---

webrender/src/display_list_flattener.rs, line 377 at r3 (raw file):

        while node_index != ROOT_SPATIAL_NODE_INDEX {
            let node = &self.clip_scroll_tree.spatial_nodes[node_index.0];

if we are to keep this hack, might want to move it in sode the clip_scroll_tree

---

webrender/src/display_list_flattener.rs, line 387 at r3 (raw file):

                    // and keep looking up the tree.
                    if let ScrollFrameKind::Explicit = info.frame_kind {
                        scroll_root = node_index;

don't we need to break here?

[gw3583]

Reviewable status: all files reviewed, 3 unresolved discussions (waiting on @kvark)

---

webrender/src/display_list_flattener.rs, line 252 at r3 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

I can see how all of this would be removed if Gecko could just tell us what to cache

Yup, it's fairly isolated at least :)

---

webrender/src/display_list_flattener.rs, line 377 at r3 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

if we are to keep this hack, might want to move it in sode the clip_scroll_tree

Yea, that'd be a good idea. If we keep it, we should also add caching since the vast majority of the time the spatial node doesn't change between primitives.

---

webrender/src/display_list_flattener.rs, line 387 at r3 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

don't we need to break here?

No, we want to keep walking up the tree to find the closest scroll root to the root node (since we're not creating caches for sub-scroll roots at this point). It won't pick the scroll root in the root pipeline, since that's not an Explicit scroll frame.

[gw3583]

@kvark Thank you for the quick review! 👍

[kvark]

@bors-servo r+

[bors-servo]

📌 Commit 2c89d39 has been approved by kvark

[bors-servo]

⌛ Testing commit 2c89d39 with merge 6ca6341...

[bors-servo]

☀️ Test successful - status-appveyor, status-taskcluster
Approved by: kvark
Pushing 6ca6341 to master...

[pcwalton]

@gw3583 Did you get a chance to profile this by any chance? I'm curious what kinds of perf improvements we saw, if any. (I wouldn't expect much on most sites, based on our experience, but it'd be good to confirm.)

[gw3583]

@pcwalton Nothing detailed yet - anecdotally it does seem like quite a win on some sites, but I need to do some proper profiling next week.