Remove AlphaBatchTask, and introduce start of raster mode. #2053
Conversation
Some render tasks can have a source task of different kinds. The most common example is that the source task to a blur task can be any of (a) a picture (b) an alpha task (c) a downscaling task. To ensure this works correctly, we separate render task data into a common base, followed by a small amount of task-specific data. The blur tasks now only fetch the common task data, which is all that they require. This fixes a bug (that I don't think is actually occurring in any tests), and is also prep work for the next step. Now that everything is using Picture internally, we can remove the AlphaBatchTask completely. This will allow us to start converting more primitives over to use a smaller subset of brush shaders, which will alow us to start doing the primitive segmentation and also get better batching results.
Most of the difficult work related to this was done in the large picture PR that landed previously. This removes the remnants of AlphaBatch tasks, and makes them all run through Picture render tasks. This also introduces the first part of the picture and shader side support for selecting whether primitives on a given picture are to be rasterized in screen space or a local space.
|
r? @kvark This patch isn't as big as it seems - it just includes #2052 which is a prerequisite. If it's easier to review and land #2052 first, and then review this - that's cool, otherwise I can close #2052 and just get them both reviewed here at the same time. I'll kick off a gecko try run with this shortly. |
|
Pending try is here https://treeherder.mozilla.org/#/jobs?repo=try&revision=8b82697f5d521d9e07fc45dd874b36c88ccd57cb - all the orange results are expected due to previous PRs, from what I can tell (but the try is not quite finished yet). |
|
Where does this one come from? https://treeherder.mozilla.org/logviewer.html#?job_id=145504579&repo=try&lineNumber=4811 |
|
The first one is known, see https://bugzilla.mozilla.org/show_bug.cgi?id=1417062#c18 . |
|
And |
|
Reviewed 14 of 14 files at r2. webrender/res/prim_shared.glsl, line 290 at r1 (raw file):
hooray! webrender/src/picture.rs, line 44 at r2 (raw file):
shouldn't this be webrender/src/render_task.rs, line 286 at r2 (raw file):
any reason this is not just Comments from Reviewable |
|
Review status: 14 of 18 files reviewed at latest revision, 3 unresolved discussions. webrender/src/picture.rs, line 44 at r2 (raw file): Previously, kvark (Dzmitry Malyshau) wrote…
Done. webrender/src/render_task.rs, line 286 at r2 (raw file): Previously, kvark (Dzmitry Malyshau) wrote…
When rasterization space is Screen - it's a device space content origin, not a layer origin. As a follow up, I'd like to make this type safe - by moving content origin to be part of the rasterization space enum. Comments from Reviewable |
DocumentAPI support on the Renderer side This is a second batch of changes related to Document API, following #1509. The idea is to allow Gecko to have separate documents for the chrome UI, page content, and bottom status bar, as opposed to using different pipelines in the same document. This change would allow minimal scene rebuilds per frame when UI is affected. WIP TODO: - [x] strict ordering API: 532bff5 - [x] Servo patch and test runs: kvark/servo@0263fa8 - [x] Firefox patch and try push: https://treeherder.mozilla.org/#/jobs?repo=try&revision=f939fbdab0231e6328061c12903db0c45847dd18 - ~~`box-shadow/boxshadow-skiprect.html`~~ (from #1943) - ~~`css-filter-chains/moz-element.html`~~ (same as in #2053) - ~~`transform/animate-layer-scale-inherit-1.html`~~ (from #2043) - [x] example - [x] `FrameBuilder` and `RenderedDocument` refactor https://tools.taskcluster.net/groups/DEvThfCaQWmt5Hp806GDLg/tasks/VLCZI7hVTQCc1xrDR3PgOQ/details# - [x] review comments cc @glennw @mstange @jrmuizel <!-- Reviewable:start --> --- This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/webrender/2050) <!-- Reviewable:end -->
Most of the difficult work related to this was done in the
large picture PR that landed previously.
This removes the remnants of AlphaBatch tasks, and makes them
all run through Picture render tasks.
This also introduces the first part of the picture and shader
side support for selecting whether primitives on a given picture
are to be rasterized in screen space or a local space.
This change is