1677211 - Bad performance on Google Cal transitions

Status: RESOLVED FIXED

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

Description

[Jim Mathies [:jimm]]

STR: opening then closing an existing google cal event. The event fades in and then fades out on top of the cal view. The performance here was better a week or so ago.

Profile:
https://share.firefox.dev/3lxbGsB

Comment 1

[Lee Salzman [:lsalzman]]

Looks like most of the time is spent in brush_opacity.

Comment 2

[Matt Woodrow (:mattwoodrow)]

I can reproduce this regression, and it bisects to bug 1676332.

Comment 3

[Jeff Muizelaar [:jrmuizel]]

Is this just because we use D3D11_CPU_ACCESS_WRITE when we should be using WRITE | READ so that we don't end up getting uncached memory?

Comment 4

[Matt Woodrow (:mattwoodrow)]

Just setting up an opt build locally to test that!

Comment 5

[Matt Woodrow (:mattwoodrow)]

Attachment: Bug 1677211 - Add Read access flags to D3D11 staging textures to avoid slow accesses when reading our intermediate results. r?jrmuizel

Comment 6

[Lee Salzman [:lsalzman]]

Does D3D11_MAP_WRITE_DISCARD have any positive effect here? In my PBO implementation for OpenGL, I generally like to utilize orphaning and/or GL_MAP_INVALIDATE_BUFFER_BIT to ensure that we don't stall on the previous contents of the buffer.

Comment 7

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/2298e1a23c6e
Add Read access flags to D3D11 staging textures to avoid slow accesses when reading our intermediate results. r=jrmuizel

Comment 8

[Cristina Coroiu [:ccoroiu]]

https://hg.mozilla.org/mozilla-central/rev/2298e1a23c6e

Comment 9

[BugBot [:suhaib / :marco/ :calixte]]

The patch landed in nightly and beta is affected.
:mattwoodrow, is this bug important enough to require an uplift?
If not please set status_beta to wontfix.

For more information, please visit auto_nag documentation.

Comment 10

[Lee Salzman [:lsalzman]]

SW-WR is not released yet beyond nightly, so doesn't require uplift.