1482212 - Understand Shield study COMPOSITE_TIME results

Status: RESOLVED WONTFIX

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

Description

[Jeff Muizelaar [:jrmuizel]]

The current results are somewhat surprising. We should figure out the cause of the difference.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

It looks like WebRender only contributes to this metric when for nop frames as DecPendingFrameCount(aWindowId) is only called from wr_notifier_nop_frame_done. Thoughts Sotaro?

Comment 2

[Jeff Muizelaar [:jrmuizel]]

As Matt points out accumulation also happens in FrameRenderingComplete() so my previous idea was garbage.

Comment 3

[Sotaro Ikeda [:sotaro]]

wr_notifier_nop_frame_done() is rarely used. It is used in the following situation
- RenderBackend handles frame generation and calls notifier.new_frame_ready but it does not have root_pipeline_id.
- RenderBackend handles frame generation and calls notifier.new_frame_ready but document is not ready for render yet.
     this happens if we are building the first scene asynchronously and scroll at the same time.

Then wr_notifier_new_frame_ready() is normally used. When I tested locally on linux and win10, I did not saw wr_notifier_nop_frame_done() call.

Comment 4

[Glenn Watson [:gw]]

Probably no longer relevant.