Port YUV primitive to be a brush.

[glennw]

Also refactor and optimize the YUV shader implementation to do
less per-pixel work.

Support segment building for YUV primitive - this is a large
performance win for any YUV images with a complex clip on them.

This is part of the prep work to allow selection of whether
a stacking context is rasterized in local or screen-space.

---

This change is 

[glennw]

r? @nical and/or @kvark

If someone else on Gecko knows the YUV code better, please get them to take a look too as we don't have a huge amount of test coverage in-tree for YUV images.

Pending try:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=0c5b905907fb40875452aa702cd9ff0d8cc90a96

[glennw]

There are a couple of failures in some of the MP4 tests, which are probably related to this. I'll investigate those and see if I can reproduce locally.

[glennw]

Fixed a typo - was using the uv_bounds.z for the texture layer index instead of the uv.z. Another gecko try coming shortly to confirm it's all fixed.

[glennw]

Pending try with typo fix:

https://hg.mozilla.org/try/rev/e36a9175fff173fa054e24cdc972d0980c67ee77
https://treeherder.mozilla.org/#/jobs?repo=try&revision=e36a9175fff173fa054e24cdc972d0980c67ee77

[glennw]

Try looks good now, I think - just the expected failure in R8 from previous PR, and an unrelated intermittent in R4 (that test has no YUV images).

[kvark]

---

Reviewed 10 of 10 files at r1.
Review status: all files reviewed at latest revision, 2 unresolved discussions.

---

webrender/res/brush_yuv_image.glsl, line 146 at r1 (raw file):

    vec2 uv_u = clamp(vUv_U.xy, vUvBounds_U.xy, vUvBounds_U.zw);
    vec2 uv_v = clamp(vUv_V.xy, vUvBounds_V.xy, vUvBounds_V.zw);
    yuv_value.x = TEX_SAMPLE(sColor0, vec3(uv_y, vUv_Y.z)).r;

it almost feels like our TEX_SAMPLE macro should get an extra parameter of UV bounds and do the clamping

---

webrender/res/brush_yuv_image.glsl, line 159 at r1 (raw file):

    // https://www.khronos.org/registry/OpenGL/extensions/APPLE/APPLE_rgb_422.txt
    vec2 uv_y = clamp(vUv_YUV.xy, vUvBounds_YUV.xy, vUvBounds_YUV.zw);
    yuv_value = TEX_SAMPLE(sColor0, vec3(uv_y, vUv_YUV.z)).gbr;

uh, the gbr order here is confusing

---

Comments from Reviewable

[kvark]

@bors-servo r+

[bors-servo]

📌 Commit e6dc591 has been approved by kvark

[bors-servo]

⌛ Testing commit e6dc591 with merge 9291910...

[bors-servo]

☀️ Test successful - status-appveyor, status-taskcluster, status-travis
Approved by: kvark
Pushing 9291910 to master...

[staktrace]

This PR seems to have introduced shader compilation failures on Windows. See https://treeherder.mozilla.org/logviewer.html#?job_id=162670488&repo=try&lineNumber=1823

This also probably explains the orange windows test jobs in the try push noted earlier in this PR. Those jobs should be green.

[kvark]

@staktrace sorry about that! I'm fixing this.