Support arbitrary rasterization coordinate roots. #3030

gw3583 · 2018-09-06T08:37:34Z

This allows us to choose to rasterize any picture in an
arbitrary coordinate space.

This fixes a large number of real world pages that use 3d
transforms. It also simplifies fixing any remaining 3d
transform related issues.

The initial patch uses this to promote any picture that has
perspective into a rasterization root.

There are a number of benefits to this:

It simplifies a lot of the math involving plane splitting,
since we don't need to deal with splits drawn in screen-space.
It makes caching perspective rendered pictures much easier,
since we can cache them between frames even when the
root coordinate system is animating the picture (this isn't
implemented yet, but is a simple-ish follow up).

We can make use of this to render text shadows in local space,
as a performance optimization (which can then be cached between
frames and display lists).

Part of this change involves simplifying the plane splitting
code to work in a similar way to traditional brush images. It's
likely that in the future we can completely remove ps_split_composite
and draw plane splits with brush_image for better batching.

Introducing a new coordinate space allows for some more optimal
calculations of required raster rectangles, which reduces the
number of pixels that we draw for clip masks and transformed
pictures, in some cases.

Follow ups:

Support arbitrary rasterization scales for raster roots, for
better quality / performance.
Support caching of pictures as discussed above.

This change is

gw3583 · 2018-09-06T08:39:07Z

r? @kvark

This needs some more code documentation - but I thought I'd open the PR now anyway, and work on the remaining bits tomorrow. This fixes a lot of real world pages that have 3d transforms on them.

Try run:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=5274c24e237d09c87f21693974d11b7aa1332250

There are 2 FAILs in R6 that I need to look into, and 1 new PASS in Wr1.

(I'll do a full try run on all platforms once I resolve those 2 FAILs tomorrow).

🎉

gw3583 · 2018-09-06T08:51:37Z

(includes the fix in #3029)

kvark

My overall impression is a bit of worry about complexity. We are getting another space (yes, I asked for this back in the day, nobody else to blame here :) ) and an extra indirection to think about (primitive -> surface/picture -> raster -> world) instead of (primitive -> picture -> world). The mutability of transform palette passed around in a lot of places is not making the code simpler either.

This fixes a large number of real world pages that use 3d transforms.

I'm trying to understand why exactly this change fixes a large number of real world pages. I thought that the previous approach, conceptually, would only have troubles on supporting 3D -> 2D -> 3D nested chains. Are those used often in practice? If you are thinking of a wider (and simpler) set of cases, do you know why the PR makes them better, i.e. what was wrong with the previous approach?

Reviewed 31 of 31 files at r1.
Reviewable status: all files reviewed, 23 unresolved discussions (waiting on @gw3583)

webrender/res/brush_blend.glsl, line 37 at r1 (raw file):

    vec2 texture_size = vec2(textureSize(sColor0, 0).xy);
    vec2 f = (vi.local_pos - local_rect.p0) / local_rect.size;
    ImageResourceExtra extra_data = fetch_image_resource_extra(user_data.x);

where did the snapping go?

webrender/res/brush_blend.glsl, line 42 at r1 (raw file):

    f = mix(x, y, f.y);
    vec2 uv = mix(uv0, uv1, f);
    vUv = vec3(uv / texture_size, res.layer);

we are missing the gl_Position.w component here. Is this intentional?

webrender/res/brush_image.glsl, line 156 at r1 (raw file):

    vUv.xy /= texture_size;
    vUv.xy *= repeat.xy;
    if ((brush_flags & BRUSH_FLAG_PERSPECTIVE_INTERPOLATION) == 0) {

why is non-perspective interpolation removed for brushes?

webrender/res/clip_shared.glsl, line 82 at r1 (raw file):

    vec2 world_pos = device_pos / uDevicePixelRatio;

    vec4 pos = prim_transform.m * vec4(world_pos, 0.0, 1.0);

could you clarify this part? We are snapping based on clip transform but then applying the primitive transform. Will it end up badly snapped in some cases then?

webrender/res/clip_shared.glsl, line 83 at r1 (raw file):

    vec4 pos = prim_transform.m * vec4(world_pos, 0.0, 1.0);
    pos.xyz /= pos.w;

we can't do this, unless you know that W is always positive

webrender/res/cs_clip_box_shadow.glsl, line 63 at r1 (raw file):

    vec2 texture_size = vec2(textureSize(sColor0, 0));
    vec2 local_pos = vLocalPos.xy / vLocalPos.z;

I know this is copied from the old code, but do we have a guarantee that Z is positive here?

webrender/res/ps_split_composite.glsl, line 78 at r1 (raw file):

    vec4 final_pos = vec4(
        dest_origin + world_pos.xy * uDevicePixelRatio,

we need to multiply the offset by world_pos.w here

webrender/res/ps_split_composite.glsl, line 82 at r1 (raw file):

#ifdef WR_FRAGMENT_SHADER
void main(void) {
    bvec4 inside = lessThanEqual(vec4(vUvTaskBounds.xy, vUv.xy),

what's the guarantee that we aren't going to be rendering outside of the task bounds?

webrender/res/ps_split_composite.glsl, line 99 at r1 (raw file):

    vec2 f = (local_pos - geometry.local_rect.p0) / geometry.local_rect.size;

    vec2 x = mix(extra_data.st_tl, extra_data.st_tr, f.x);

can we use bilerp here?

webrender/src/batch.rs, line 9 at r1 (raw file):

use api::{YuvColorSpace, YuvFormat, WorldPixel, WorldRect};
use clip::{ClipNodeFlags, ClipNodeRange, ClipItem, ClipStore};
use clip_scroll_tree::{ClipScrollTree, SpatialNodeIndex};

I find it a bit worrying that batching now depends on CST

webrender/src/batch.rs, line 1723 at r1 (raw file):

        clip_store: &ClipStore,
        clip_scroll_tree: &ClipScrollTree,
        transforms: &mut TransformPalette,

why do we need to mutate it here?

webrender/src/gpu_types.rs, line 432 at r1 (raw file):

    }

    pub fn set_world_transform(

reads a bit weird that it's called set_world_transform but actually gets LayoutToPictureTransform

webrender/src/gpu_types.rs, line 465 at r1 (raw file):

            let transforms = &mut self.transforms;

            *self.map

now I see why the mutable state is needed

webrender/src/gpu_types.rs, line 599 at r1 (raw file):

    //           some investigation on if this ever happens?
    let inv_transform = transform.inverse()
                                 .unwrap_or(PictureToLayoutTransform::identity());

we can error! at least in this case

webrender/src/picture.rs, line 42 at r1 (raw file):

    pub surface: Option<PictureSurface>,

    pub raster_spatial_node_index: SpatialNodeIndex,

nit: not clear if raster_ is needed here given that we are in RasterConfig struct

webrender/src/picture.rs, line 240 at r1 (raw file):

        let xf = frame_context.clip_scroll_tree.get_relative_transform(
            raster_spatial_node_index,

uh, I'm getting a little lost here. If this primitive spawns a surface, wouldn't the rasterized space match the surface space?
I reads like there can be more surfaces down the road based on the same raster space if they aren't perspectively transformed. Is there a benefit to this versus just treating Surface=Raster at all times?

webrender/src/picture.rs, line 268 at r1 (raw file):

        let map_raster_to_world = SpaceMapper::new_with_target(
            SpatialNodeIndex(0),

of all the spatial node indices in scope, why do we choose 0 here? :)

webrender/src/picture.rs, line 419 at r1 (raw file):

                                               .unwrap_or(RasterRect::max_rect());

                let map_to_raster = SpaceMapper::new_with_target(

there is quite a bit of common code between this and take_context

webrender/src/prim_store.rs, line 2909 at r1 (raw file):

    let unclipped_raster_rect = match map_to_raster.map(&pic_rect) {
        Some(rect) => rect,
        None => return None,

this function gives us a great opportunity to abuse ?, e.g.

let unclipped_raster_rect = map_to_raster(&pic_rect)?;

webrender/src/prim_store.rs, line 2942 at r1 (raw file):

    );

    let transform = map_to_raster.get_transform();

In at least 2 cases the caller only cares about the first result. Can/should we avoid the extra computation in this case?

webrender/src/renderer.rs, line 446 at r1 (raw file):

                name: "aClipTransformId",
                count: 1,
                kind: VertexAttributeKind::I32,

it is rather tempting to reduce the number of attributes by having ivec4

webrender/src/util.rs, line 68 at r1 (raw file):

         self.m24.abs() > NEARLY_ZERO ||
         self.m34.abs() > NEARLY_ZERO ||
         (self.m44.abs() - 1.0) > NEARLY_ZERO

what if m44 = 0.5?

webrender_api/src/units.rs, line 55 at r1 (raw file):

#[derive(Hash, Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd)]
pub struct RasterPixel;

need to document what this space is about

pcwalton · 2018-09-06T22:10:37Z

Note that this will possibly be needed to optimize blur as well, because right now there's no way to render contents at a lower resolution without introducing more transforms, which necessitate stacking contexts and reference frames.

gw3583 · 2018-09-07T02:26:11Z

Current try:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=57331aff6e49b2ec1003a85abff4b4130dd0e3ff

Down to one failure - working through that now, then will follow up on and address review comments :)

kvark · 2018-09-07T02:49:28Z

I confirm (by downloading the last try artifact) that the PR fixes https://bugzilla.mozilla.org/show_bug.cgi?id=1487570
🎉 🎉 🎉

gw3583

Reviewable status: 28 of 31 files reviewed, 23 unresolved discussions (waiting on @kvark and @gw3583)

webrender/res/brush_blend.glsl, line 37 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

where did the snapping go?

This was needed when we were deriving the screen UV from the render task device rect. However, we now use the UVs provided by calculate_uv_rect_kind(), which handles the snapping when calculating the UVs.

webrender/res/brush_blend.glsl, line 42 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

we are missing the gl_Position.w component here. Is this intentional?

Yes - any pictures that have perspective now form a rasterization root in their local space, so we don't need to apply perspective correction when compositing them.

webrender/res/brush_image.glsl, line 156 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

why is non-perspective interpolation removed for brushes?

Same as above - we only ever composite with source images that were not drawn with perspective now.

webrender/res/clip_shared.glsl, line 82 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

could you clarify this part? We are snapping based on clip transform but then applying the primitive transform. Will it end up badly snapped in some cases then?

I think this is wrong, yes - I have left it as is, since it's the same behavior as current (previous code only looked at the clip transform here). But it does seem potentially incorrect to me too, and should be considered as a follow up.

webrender/res/clip_shared.glsl, line 83 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

we can't do this, unless you know that W is always positive

I think this is (somewhat) related to the last failure I have in try run - working on a solution to that now.

webrender/res/cs_clip_box_shadow.glsl, line 63 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

I know this is copied from the old code, but do we have a guarantee that Z is positive here?

I'm not totally sure - this might be related to the current reftest failure on try.

webrender/res/ps_split_composite.glsl, line 78 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

we need to multiply the offset by world_pos.w here

dest_origin is already in device pixels. I ran into this because I found the existing code was failing on my hi-dpi screen - this is the fix for that.

webrender/res/ps_split_composite.glsl, line 82 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

what's the guarantee that we aren't going to be rendering outside of the task bounds?

In theory the clip mask generated on the surface handles this, but I want to do some more testing - this might actually be the issue with the failing reftest too.

webrender/res/ps_split_composite.glsl, line 99 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

can we use bilerp here?

Yes, and your comment helped me notice I had the parameter order here wrong (compared to the bilerp above) - fixing this actually fixes most of the last failing reftest I have!

webrender/src/batch.rs, line 9 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

I find it a bit worrying that batching now depends on CST

Yea - it's because of the way we generate relative transforms in the transform palette on demand. A tidier solution, as a follow up, is that the culling and clipping generation code should pre-register all the needed relative transforms. Then they will exist in the transform palette.

webrender/src/batch.rs, line 1723 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

why do we need to mutate it here?

See above comment about CST - we can fix this as a follow up.

webrender/src/gpu_types.rs, line 432 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

reads a bit weird that it's called set_world_transform but actually gets LayoutToPictureTransform

True, fixed.

webrender/src/gpu_types.rs, line 465 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

now I see why the mutable state is needed

Yea - it's a bit of a hack to get this landed. We can pre-register them in the future as described above.

webrender/src/gpu_types.rs, line 599 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

we can error! at least in this case

Fixed

webrender/src/picture.rs, line 42 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

nit: not clear if raster_ is needed here given that we are in RasterConfig struct

Added a comment to clarify.

webrender/src/picture.rs, line 240 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

uh, I'm getting a little lost here. If this primitive spawns a surface, wouldn't the rasterized space match the surface space?
I reads like there can be more surfaces down the road based on the same raster space if they aren't perspectively transformed. Is there a benefit to this versus just treating Surface=Raster at all times?

Not always - for example, if you have a rotated picture on the main frame, the surface space comes from the picture, but the raster space could be the surface space (if drawing in local space for caching / performance) or the root world space (if drawing in screen space for quality).

webrender/src/picture.rs, line 268 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

of all the spatial node indices in scope, why do we choose 0 here? :)

We know that node index 0 is world space - I changed these references to use ROOT_SPATIAL_NODE_INDEX for clarity.

webrender/src/picture.rs, line 419 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

there is quite a bit of common code between this and take_context

Agreed, I'll tidy this up a bit.

webrender/src/prim_store.rs, line 2909 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

this function gives us a great opportunity to abuse ?, e.g.
let unclipped_raster_rect = map_to_raster(&pic_rect)?;

Agreed, I'll tidy this up a bit.

webrender/src/prim_store.rs, line 2942 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

In at least 2 cases the caller only cares about the first result. Can/should we avoid the extra computation in this case?

It's not called that often, but I will tidy it up to avoid the extra work in those cases since it's the more common case.

webrender/src/renderer.rs, line 446 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

it is rather tempting to reduce the number of attributes by having ivec4

Yep, this could definitely be tidied up in a follow up.

webrender/src/util.rs, line 68 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

what if m44 = 0.5?

That still works? All we're trying to do here is see if it's 1 or very close to 1?

webrender_api/src/units.rs, line 55 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

need to document what this space is about

Done.

gw3583 · 2018-09-07T06:01:11Z

Addressed most of those review comments. Still working on one reftest failure, and then a couple of tidy up comments from above still need to be addressed.

kvark

Reviewed 1 of 3 files at r2, 9 of 9 files at r3.
Reviewable status: all files reviewed, 12 unresolved discussions (waiting on @kvark)

webrender/res/brush_image.glsl, line 156 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Same as above - we only ever composite with source images that were not drawn with perspective now.

Hold on. Brush image is our general hammer of drawing image-ish nails, not just compositing. If there is a transformed stacking context with an image in it, we'll still need to draw the brush_image primitive (generated by this image) begin transformed. What am I missing?

webrender/res/clip_shared.glsl, line 82 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

I think this is wrong, yes - I have left it as is, since it's the same behavior as current (previous code only looked at the clip transform here). But it does seem potentially incorrect to me too, and should be considered as a follow up.

please add a TODO comment since we are at it

webrender/res/ps_split_composite.glsl, line 78 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

dest_origin is already in device pixels. I ran into this because I found the existing code was failing on my hi-dpi screen - this is the fix for that.

It's in device pixels, but it's not in the homogeneous space, so it will end up divided by world_pos.w by the hardware unless you premultiply it here

webrender/res/ps_split_composite.glsl, line 99 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Yes, and your comment helped me notice I had the parameter order here wrong (compared to the bilerp above) - fixing this actually fixes most of the last failing reftest I have!

yay! glad to help :D

webrender/src/picture.rs, line 240 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Not always - for example, if you have a rotated picture on the main frame, the surface space comes from the picture, but the raster space could be the surface space (if drawing in local space for caching / performance) or the root world space (if drawing in screen space for quality).

Thank you for clarification! So a picture may need to spawn a surface. And then a surface may need to spawn a raster space (if it has perspective component). Correct?

webrender/src/picture.rs, line 419 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Agreed, I'll tidy this up a bit.

this is still TODO, right?

webrender/src/util.rs, line 68 at r1 (raw file):

All we're trying to do here is see if it's 1 or very close to 1?

That's what we should be doing, but that's not what the code does :)

webrender_api/src/units.rs, line 55 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Done.

I think "screen space" should be replaced by "world space" here to match our strongly typed spaces.
In general, we should be careful with interchanging "screen" and "device", preferring the latter for there being a type for it.

gw3583 · 2018-09-08T07:23:02Z

OK, I have a fix for the remaining reftest failure - the fix also resolves the concerns above about clipping outside the boundaries of a plane splitting polygon.

I'll tidy up the patch on Monday (the implementation is a bit of a hack, but the result is correct) and post it here.

This allows us to choose to rasterize any picture in an arbitrary coordinate space. This fixes a large number of real world pages that use 3d transforms. It also simplifies fixing any remaining 3d transform related issues. The initial patch uses this to promote any picture that has perspective into a rasterization root. There are a number of benefits to this: * It simplifies a lot of the math involving plane splitting, since we don't need to deal with splits drawn in screen-space. * It makes caching perspective rendered pictures much easier, since we can cache them between frames even when the root coordinate system is animating the picture (this isn't implemented yet, but is a simple-ish follow up). We can make use of this to render text shadows in local space, as a performance optimization (which can then be cached between frames and display lists). Part of this change involves simplifying the plane splitting code to work in a similar way to traditional brush images. It's likely that in the future we can completely remove ps_split_composite and draw plane splits with brush_image for better batching. Introducing a new coordinate space allows for some more optimal calculations of required raster rectangles, which reduces the number of pixels that we draw for clip masks and transformed pictures, in some cases. Follow ups: * Support arbitrary rasterization scales for raster roots, for better quality / performance. * Support caching of pictures as discussed above.

gw3583

Reviewable status: 24 of 32 files reviewed, 12 unresolved discussions (waiting on @kvark and @gw3583)

webrender/res/brush_image.glsl, line 156 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

Hold on. Brush image is our general hammer of drawing image-ish nails, not just compositing. If there is a transformed stacking context with an image in it, we'll still need to draw the brush_image primitive (generated by this image) begin transformed. What am I missing?

The transform is still applied, but we don't need to specifically correct for perspective interpolation, since the source images from render tasks are no longer drawn with a perspective component. So it's just like drawing a normal image source in perspective.

webrender/res/clip_shared.glsl, line 82 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

please add a TODO comment since we are at it

Added a comment.

webrender/res/ps_split_composite.glsl, line 78 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

It's in device pixels, but it's not in the homogeneous space, so it will end up divided by world_pos.w by the hardware unless you premultiply it here

Ah, indeed. Fixed!

webrender/res/ps_split_composite.glsl, line 99 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

yay! glad to help :D

Done.

webrender/src/picture.rs, line 240 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

Thank you for clarification! So a picture may need to spawn a surface. And then a surface may need to spawn a raster space (if it has perspective component). Correct?

Yep, exactly :)

webrender/src/picture.rs, line 419 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

this is still TODO, right?

Yep, fixed in the most recent commit.

webrender/src/util.rs, line 68 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

All we're trying to do here is see if it's 1 or very close to 1?

That's what we should be doing, but that's not what the code does :)

Ooops! Fixed.

webrender_api/src/units.rs, line 55 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

I think "screen space" should be replaced by "world space" here to match our strongly typed spaces.
In general, we should be careful with interchanging "screen" and "device", preferring the latter for there being a type for it.

Agreed, fixed.

gw3583 · 2018-09-09T23:25:04Z

I addressed the remaining review comments, and rebased. I also added a (hopefully) final commit that resolves the remaining reftest failure and comments about clipping on split composite primitives. It's basically a hack / workaround for legacy API issues with the clipping APIs. However, I think it's a reasonable implementation detail for now, that we can tidy up at a later point. Commit note here for details:

The split composite shader is now quite similar to a normal
brush shader (uses prim header, supports normal clip masks
etc). The only real difference is that it draws a quad polygon
from local points instead of a rectangle. It should be easy
to unify this into the brush_image shader in the future.

Add a temporary "ClipNodeCollector" interface to correctly
handle clips that are ancestors of the current rasterization
root. Although it's not enforced by the API, I have verified
that any time a clip from a spatial node that is an ancestor
of the current rasterization root is applied on all primitives
that are part of that picture. This means we can safely apply
those clips during compositing of the rasterization root rather
than on each individual primitive. This fixes correctness
issues with clips across rasterization roots, and is also a
quite significant performance win in some cases, since the
clip mask is now applied once when the rasterization root
is drawn, rather than on each indidvidual primitive.

The end result of this is actually quite good - it will allow
us to simplify some of the clipping code in the future, and is
a performance win. The implementation leaves a lot to be
desired, but gets the required result without having to fix
up both gecko and wr clipping API mismatches. The long term
solution for this is to update gecko and WR to make better use
of clip chains on stacking contexts, rather than putting
clips from the stacking context on each individual primitive (we
have discussed doing this previously, but it's low-ish priority
for now given other bugs).

I'll kick off an updated try run shortly. Hopefully after that and review of the last two commits, this should be ready to go. Although not ideal, it fixes a heap of existing 3d transform bugs we have.

gw3583 · 2018-09-09T23:43:38Z

Try run looks good so far (still in progress):

https://treeherder.mozilla.org/#/jobs?repo=try&revision=3e29fc7c6924f333afd5418aed4854ac32255937

1 new PASS in Wr1.
2x new PASS in R1 (OSX).
Unrelated intermittent in R8 and 15.

Bugzilla bugs that are affected by this change:

FIXED:
https://bugzilla.mozilla.org/show_bug.cgi?id=1487570
https://bugzilla.mozilla.org/show_bug.cgi?id=1470905
https://bugzilla.mozilla.org/show_bug.cgi?id=1421526
https://bugzilla.mozilla.org/show_bug.cgi?id=1451087 (#2587)
https://bugzilla.mozilla.org/show_bug.cgi?id=1454706 (still some flickering in places)
https://bugzilla.mozilla.org/show_bug.cgi?id=1483647

IMPROVED:
https://bugzilla.mozilla.org/show_bug.cgi?id=1415272
https://bugzilla.mozilla.org/show_bug.cgi?id=1422051

gw3583 · 2018-09-10T01:00:46Z

Try run looks good. The same new PASS results on Windows, plus a couple unrelated intermittents. re-r? @kvark

kvark

I got just a few concerns left. Code is looking great for the most part 👍

Reviewed 14 of 14 files at r4.
Reviewable status: all files reviewed, 5 unresolved discussions (waiting on @kvark and @gw3583)

webrender/res/clip_shared.glsl, line 83 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

I think this is (somewhat) related to the last failure I have in try run - working on a solution to that now.

This concern doesn't appear addressed?

webrender/src/clip.rs, line 472 at r4 (raw file):

                let clip_node = &self.clip_nodes[clip_node_index.0 as usize];

                // Check if any this clip node index should actually be

"Check if any this clip node" - something is missing here :)

webrender/src/clip.rs, line 1120 at r4 (raw file):

    clip_node_index: ClipNodeIndex,
    spatial_node_index: SpatialNodeIndex,
    local_clip_rect: &mut LayoutRect,

It seems like this would be cleaner if instead of &mut clip rectangle we returned Option<LayoutRect>. This is a nit, not necessary for immediate addressing.

webrender/src/clip.rs, line 1140 at r4 (raw file):

        let xf = clip_scroll_tree.get_relative_transform(
            clip_node.spatial_node_index,
            ROOT_SPATIAL_NODE_INDEX,

shouldn't this be spatial_node_index, given that you are trying to find the relation between ref and clip?

gw3583

Reviewable status: all files reviewed, 5 unresolved discussions (waiting on @gw3583 and @kvark)

webrender/res/clip_shared.glsl, line 83 at r1 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

This concern doesn't appear addressed?

It's addressed by the clip node collector change - we don't rasterize clips across perspective roots now. In fact this means we should be able to simplify this code significantly, but I left this as a follow up, since I want to get this stuff into gecko asap.

webrender/src/clip.rs, line 472 at r4 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

"Check if any this clip node" - something is missing here :)

Done.

webrender/src/clip.rs, line 1120 at r4 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

It seems like this would be cleaner if instead of &mut clip rectangle we returned Option<LayoutRect>. This is a nit, not necessary for immediate addressing.

The problem with that is that we also need to signal if the clip completely culled the primitive - I guess we could have Result<Option, ()> or something like that as a follow up?

webrender/src/clip.rs, line 1140 at r4 (raw file):

Previously, kvark (Dzmitry Malyshau) wrote…

shouldn't this be spatial_node_index, given that you are trying to find the relation between ref and clip?

In this case (the complex clip spanning a complex transform), we map both the prim and clip to world space in order to do our trivial clip accept / reject / partial tests. So this transform is used to get the clip inner / outer rects in world space.

gw3583 · 2018-09-10T19:33:27Z

@kvark Thanks! Responded to those comments / quesions and pushed an updated commit.

kvark

Reviewed 1 of 1 files at r5.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @kvark)

webrender/res/cs_clip_box_shadow.glsl, line 63 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

I'm not totally sure - this might be related to the current reftest failure on try.

Does it mean we are still blocked?

kvark · 2018-09-10T22:10:39Z

@bors-servo r+

gw3583 · 2018-09-10T22:11:51Z

🚀

gw3583 · 2018-09-10T23:55:41Z

@bors-servo r=kvark

The split composite shader is now quite similar to a normal brush shader (uses prim header, supports normal clip masks etc). The only real difference is that it draws a quad polygon from local points instead of a rectangle. It should be easy to unify this into the brush_image shader in the future. Add a temporary "ClipNodeCollector" interface to correctly handle clips that are ancestors of the current rasterization root. Although it's not enforced by the API, I have verified that any time a clip from a spatial node that is an ancestor of the current rasterization root is applied on all primitives that are part of that picture. This means we can safely apply those clips during compositing of the rasterization root rather than on each individual primitive. This fixes correctness issues with clips across rasterization roots, and is also a quite significant performance win in some cases, since the clip mask is now applied once when the rasterization root is drawn, rather than on each indidvidual primitive. The end result of this is actually quite good - it will allow us to simplify some of the clipping code in the future, and is a performance win. The implementation leaves a lot to be desired, but gets the required result without having to fix up both gecko and wr clipping API mismatches. The long term solution for this is to update gecko and WR to make better use of clip chains on stacking contexts, rather than putting clips from the stacking context on each individual primitive (we have discussed doing this previously, but it's low-ish priority for now given other bugs).

gw3583 · 2018-09-11T00:01:01Z

@bors-servo r=kvark

gw3583 · 2018-09-11T00:03:36Z

Just pushed an amended commit to try and get bors to respond, but no luck so far 😢

jdm · 2018-09-11T01:39:08Z

@bors-servo r=kvark

bors-servo · 2018-09-11T01:39:08Z

📌 Commit a99f733 has been approved by kvark

bors-servo · 2018-09-11T01:41:18Z

⌛ Testing commit a99f733 with merge dede39f...

Support arbitrary rasterization coordinate roots. This allows us to choose to rasterize any picture in an arbitrary coordinate space. This fixes a large number of real world pages that use 3d transforms. It also simplifies fixing any remaining 3d transform related issues. The initial patch uses this to promote any picture that has perspective into a rasterization root. There are a number of benefits to this: * It simplifies a lot of the math involving plane splitting, since we don't need to deal with splits drawn in screen-space. * It makes caching perspective rendered pictures much easier, since we can cache them between frames even when the root coordinate system is animating the picture (this isn't implemented yet, but is a simple-ish follow up). We can make use of this to render text shadows in local space, as a performance optimization (which can then be cached between frames and display lists). Part of this change involves simplifying the plane splitting code to work in a similar way to traditional brush images. It's likely that in the future we can completely remove ps_split_composite and draw plane splits with brush_image for better batching. Introducing a new coordinate space allows for some more optimal calculations of required raster rectangles, which reduces the number of pixels that we draw for clip masks and transformed pictures, in some cases. Follow ups: * Support arbitrary rasterization scales for raster roots, for better quality / performance. * Support caching of pictures as discussed above.  --- This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/webrender/3030)

gw3583 · 2018-09-11T04:01:24Z

@bors-servo retry

bors-servo · 2018-09-11T04:01:27Z

⌛ Testing commit a99f733 with merge 04d63e7...

Support arbitrary rasterization coordinate roots. This allows us to choose to rasterize any picture in an arbitrary coordinate space. This fixes a large number of real world pages that use 3d transforms. It also simplifies fixing any remaining 3d transform related issues. The initial patch uses this to promote any picture that has perspective into a rasterization root. There are a number of benefits to this: * It simplifies a lot of the math involving plane splitting, since we don't need to deal with splits drawn in screen-space. * It makes caching perspective rendered pictures much easier, since we can cache them between frames even when the root coordinate system is animating the picture (this isn't implemented yet, but is a simple-ish follow up). We can make use of this to render text shadows in local space, as a performance optimization (which can then be cached between frames and display lists). Part of this change involves simplifying the plane splitting code to work in a similar way to traditional brush images. It's likely that in the future we can completely remove ps_split_composite and draw plane splits with brush_image for better batching. Introducing a new coordinate space allows for some more optimal calculations of required raster rectangles, which reduces the number of pixels that we draw for clip masks and transformed pictures, in some cases. Follow ups: * Support arbitrary rasterization scales for raster roots, for better quality / performance. * Support caching of pictures as discussed above.  --- This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/webrender/3030)

bors-servo · 2018-09-11T08:02:30Z

☀️ Test successful - status-appveyor, status-taskcluster
Approved by: kvark
Pushing 04d63e7 to master...

gw3583 force-pushed the raster-17 branch from 6cd0575 to de73670 Compare September 6, 2018 08:51

kvark requested changes Sep 6, 2018

View reviewed changes

gw3583 force-pushed the raster-17 branch from de73670 to 16b00dc Compare September 7, 2018 05:18

gw3583 commented Sep 7, 2018

View reviewed changes

gw3583 force-pushed the raster-17 branch from 586a750 to 83d6d9d Compare September 7, 2018 06:27

kvark requested changes Sep 7, 2018

View reviewed changes

gw3583 added 4 commits September 10, 2018 06:26

Fix a bug

e68e263

Address some review comments

d900696

Address more review comments

37d2b19

gw3583 force-pushed the raster-17 branch from 83d6d9d to 37d2b19 Compare September 9, 2018 21:19

gw3583 commented Sep 9, 2018

View reviewed changes

kvark requested changes Sep 10, 2018

View reviewed changes

gw3583 commented Sep 10, 2018

View reviewed changes

gw3583 force-pushed the raster-17 branch from 3418958 to f740b57 Compare September 10, 2018 19:32

kvark approved these changes Sep 10, 2018

View reviewed changes

gw3583 force-pushed the raster-17 branch from f740b57 to a99f733 Compare September 11, 2018 00:00

bors-servo merged commit a99f733 into servo:master Sep 11, 2018

bors-servo mentioned this pull request Sep 11, 2018

Remove flat varying #3012

Closed

Darkspirit mentioned this pull request Sep 19, 2018

Deformed and glichy terminal on stripe.com (at 280+% zoom) #3087

Closed

kvark mentioned this pull request Sep 20, 2018

Preserve3D baking space #1447

Closed

Darkspirit mentioned this pull request Dec 12, 2018

Crash when scrolling down on https://www.magicleap.com/magic-leap-one #3406

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support arbitrary rasterization coordinate roots. #3030

Support arbitrary rasterization coordinate roots. #3030

gw3583 commented Sep 6, 2018 •

edited by larsbergstrom

Loading

gw3583 commented Sep 6, 2018

gw3583 commented Sep 6, 2018

kvark left a comment

pcwalton commented Sep 6, 2018

gw3583 commented Sep 7, 2018

kvark commented Sep 7, 2018

gw3583 left a comment

gw3583 commented Sep 7, 2018

kvark left a comment

gw3583 commented Sep 8, 2018

gw3583 left a comment

gw3583 commented Sep 9, 2018

gw3583 commented Sep 9, 2018

gw3583 commented Sep 10, 2018

kvark left a comment

gw3583 left a comment

gw3583 commented Sep 10, 2018

kvark left a comment

kvark commented Sep 10, 2018

gw3583 commented Sep 10, 2018

gw3583 commented Sep 10, 2018

gw3583 commented Sep 11, 2018

gw3583 commented Sep 11, 2018

jdm commented Sep 11, 2018

bors-servo commented Sep 11, 2018

bors-servo commented Sep 11, 2018

gw3583 commented Sep 11, 2018

bors-servo commented Sep 11, 2018

bors-servo commented Sep 11, 2018

Support arbitrary rasterization coordinate roots. #3030

Support arbitrary rasterization coordinate roots. #3030

Conversation

gw3583 commented Sep 6, 2018 • edited by larsbergstrom Loading

gw3583 commented Sep 6, 2018

gw3583 commented Sep 6, 2018

kvark left a comment

Choose a reason for hiding this comment

pcwalton commented Sep 6, 2018

gw3583 commented Sep 7, 2018

kvark commented Sep 7, 2018

gw3583 left a comment

Choose a reason for hiding this comment

gw3583 commented Sep 7, 2018

kvark left a comment

Choose a reason for hiding this comment

gw3583 commented Sep 8, 2018

gw3583 left a comment

Choose a reason for hiding this comment

gw3583 commented Sep 9, 2018

gw3583 commented Sep 9, 2018

gw3583 commented Sep 10, 2018

kvark left a comment

Choose a reason for hiding this comment

gw3583 left a comment

Choose a reason for hiding this comment

gw3583 commented Sep 10, 2018

kvark left a comment

Choose a reason for hiding this comment

kvark commented Sep 10, 2018

gw3583 commented Sep 10, 2018

gw3583 commented Sep 10, 2018

gw3583 commented Sep 11, 2018

gw3583 commented Sep 11, 2018

jdm commented Sep 11, 2018

bors-servo commented Sep 11, 2018

bors-servo commented Sep 11, 2018

gw3583 commented Sep 11, 2018

bors-servo commented Sep 11, 2018

bors-servo commented Sep 11, 2018

gw3583 commented Sep 6, 2018 •

edited by larsbergstrom

Loading