Move primitive origin from template data to the instance.

[gw3583]

This allows sharing of primitive templates where the only difference
between two primitives is the local space position.

This also means that interning works as expected for primitives
sent by Gecko that contain baked in scroll offsets.

This patch changes the primitive rect only - the next step is a
follow up patch that makes the clip rect, and a few other params
in PrimitiveTemplateKind to also be in a true local space.

---

This change is 

[gw3583]

r? @kvark

Pending try:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=f1c82c5322f7cacd2be464b83295a9caeda02a36

[gw3583]

Pushed a follow up commit that normalizes the local clip rect. For some reason I can't push a try run at the moment, but I'll paste one here once I can.

[gw3583]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=39398d6009368bf910da37d5bdd1d1e9463fca16

[gw3583]

Hmmm, that second try run is very broken - investigating.

[gw3583]

Fixed up the last commit, and kicked off a new try:

https://treeherder.mozilla.org/#/jobs?repo=try&revision=08c72f439b92112bec5d68ce06598d06869836a3

[gw3583]

That try run looks much better 👍

[gw3583]

Re-implemented on top of the separate interning PR, since the rebase had too many conflicts.

Kicked off a new pending try:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=63766afea3cd0d5f1a8f929bd9528d4c52e817fb

[kvark]

One concern

Reviewed 6 of 6 files at r1.
Reviewable status: all files reviewed, 4 unresolved discussions (waiting on @gw3583)

---

webrender/src/display_list_flattener.rs, line 1048 at r1 (raw file):

        DocumentResources: InternerMut<P>,
    {
        let normalized_clip_rect = info.clip_rect

the "normalized" part is a bit confusing to me for having a different meaning in vector math
can we call it relative_clip_rect?
We might need to go as far as creating a separate space for that

---

webrender/src/display_list_flattener.rs, line 1049 at r1 (raw file):

    {
        let normalized_clip_rect = info.clip_rect
            .translate(&LayoutVector2D::new(-info.rect.origin.x, -info.rect.origin.y))

nit: can we use to_vector()?

---

webrender/src/picture.rs, line 626 at r1 (raw file):

        let clip_rect = prim_data
            .normalized_clip_rect
            .translate(&LayoutVector2D::new(prim_instance.prim_origin.x, prim_instance.prim_origin.y));

nit: can we use prim_instance.prim_origin.to_vector()?

---

webrender/src/picture.rs, line 1345 at r1 (raw file):

                let clip_rect = prim_data
                    .normalized_clip_rect
                    .translate(&LayoutVector2D::new(prim_instance.prim_origin.x, prim_instance.prim_origin.y));

same here

[gw3583]

@kvark Thanks, addressed those comments. And I agree, I think introducing a primitive local space in the future probably makes sense. 👍

[gw3583]

Try looks good.

@bors-servo r=kvark

[gw3583]

@bors-servo r=kvark

[jdm]

@bors-servo ping

[bors-servo]

😪 I'm awake I'm awake

[jdm]

@bors-servo r=kvark

[bors-servo]

📌 Commit aaeb3d8 has been approved by kvark

[bors-servo]

⌛ Testing commit aaeb3d8 with merge 694726e...

[gw3583]

This looks like another 2-3% win on dl_mutate from the try run.

[bors-servo]

☀️ Test successful - status-appveyor, status-taskcluster
Approved by: kvark
Pushing 694726e to master...