1372083 - Some advanced layers are not rendered on windows

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect, P3)

Description

[Sotaro Ikeda [:sotaro]]

Bug 1371557 addressed some rendering problems of webrender on Windows. But even with Bug 1371557, I still saw rendering problems. Some advanced layers seemed not rendered.

Comment 1

[Sotaro Ikeda [:sotaro]]

Some ui was still not correctly with a patch of Bug 1371557.
  https://www.youtube.com/watch?v=jhOVibLEDhA

And I often saw a flickering with layout/reftests/border-radius/color-layer-1a.html

Comment 2

[Sotaro Ikeda [:sotaro]]

(In reply to Sotaro Ikeda [:sotaro] from comment #1)
> 
> And I often saw a flickering with
> layout/reftests/border-radius/color-layer-1a.html

I did not saw the problem only one Tab of the "color-layer-1a.html" was opened. When I opened another tab, I saw the flicker.

Comment 3

[Darkspirit]

See bug 1357734 comment 37 (video in bug 1357734 comment 27). The shadows on the right side of bugzilla are not shown or disappear with scrolling down.

Comment 4

[Sotaro Ikeda [:sotaro]]

When the problem happen, FramebufferState in ANGLE check failed.

The callstack was the following.

 libGLESv2.dll!gl::FramebufferState::attachmentsHaveSameDimensions() Line 196
 libGLESv2.dll!rx::FramebufferD3D::checkStatus() Line 317
 libGLESv2.dll!gl::Framebuffer::checkStatusImpl(const gl::ContextState & state) Line 710
 libGLESv2.dll!gl::Framebuffer::checkStatus(const gl::ContextState & state) Line 482
 libGLESv2.dll!gl::ValidateClear() Line 1920
 libGLESv2.dll!gl::Clear(unsigned int mask) Line 295
 libGLESv2.dll!glClear(unsigned int mask) Line 97
 xul.dll!gleam::ffi_gles::Gles2::Clear() Line 1525
 xul.dll!gleam::gl::{{impl}}::clear() Line 1213
 xul.dll!webrender::device::Device::clear_target_rect() Line 1943
 xul.dll!webrender::renderer::Renderer::draw_alpha_target() Line 1879
 xul.dll!webrender::renderer::Renderer::draw_tile_frame() Line 2127
 xul.dll!webrender::renderer::{{impl}}::render::{{closure}}(closure) Line 1332
 xul.dll!webrender::profiler::TimeProfileCounter::profile<webrender::device::FrameId,closure>(closure self) Line 154
 xul.dll!webrender::renderer::Renderer::render() Line 1309
 xul.dll!webrender_bindings::bindings::wr_renderer_render() Line 804
 xul.dll!mozilla::wr::RendererOGL::Render() Line 132
 xul.dll!mozilla::wr::RenderThread::UpdateAndRender(WrWindowId aWindowId) Line 242
 xul.dll!mozilla::wr::RenderThread::NewFrameReady(WrWindowId aWindowId) Line 170



"if (hasMismatchedSize(mDepthAttachment))" was failed in FramebufferState::attachmentsHaveSameDimensions().
  https://dxr.mozilla.org/mozilla-central/source/gfx/angle/src/libANGLE/Framebuffer.cpp#194

Comment 5

[Sotaro Ikeda [:sotaro]]

The attachmentsHaveSameDimensions() failed because dimensions of ColorAttachments and mDepthAttachment were different.

> sizes of mColorAttachments = {width=0x000004f4 height=0x00000400 depth=0x00000002 }
> size of mDepthAttachment = {width=0x000004f4 height=0x00000400 depth=0x00000001 }


They were created in create_fbo_for_texture_if_necessary().
  https://dxr.mozilla.org/mozilla-central/source/gfx/webrender/src/device.rs#1225

When the problem happened, layer_count was 2 in create_fbo_for_texture_if_necessary() and ColorAttachment was created by gl.tex_image_3d() and framebuffer_texture_layer(). Then the depth of ColorAttachment became 2. But depth of mDepthAttachment was i since it was created with gl.renderbuffer_storage().

From it, it seems necessary to create DepthAttachment with gl.tex_image_3d() and framebuffer_texture_layer() like ColorAttachment.

Comment 6

[Sotaro Ikeda [:sotaro]]

From the following comment, attachmentsHaveSameDimensions() seems to have a bug.
  https://github.com/servo/webrender/issues/1470#issuecomment-314828465

 > All in all, I'd bet this to be a bug in Angle, which takes the depth of the color texture
 > as opposed to the depth of the color attachment.

Comment 7

[Sotaro Ikeda [:sotaro]]

Attachment: patch - Do not compare depth in attachmentsHaveSameDimensions()

I confirmed that the patch addressed the problem locally.

Comment 8

[Sotaro Ikeda [:sotaro]]

Comment on attachment 8885985 [details] [diff] [review]
patch - Do not compare depth in attachmentsHaveSameDimensions()

:jgilbert, can you feedback to the patch?

Comment 9

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/81f667d3a504
Do not compare depth in attachmentsHaveSameDimensions() r=jgilbert

Comment 10

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/81f667d3a504

Comment 11

[Michael Leu[:Lenzak](UTC+8)]

This bug has been upstreamed to google.
https://github.com/google/angle/commit/8f8edd6eac1af12b7cdb2f69f2f00055bc4b5b31