Closed Bug 1416645 Opened 7 years ago Closed 2 years ago

Bad performance on running glterrain talos suite

Categories

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

Product:
Core
Component:
Graphics: WebRender
Platform:
x86_64
All
Type:
defect

Tracking

()

Status:
RESOLVED INCOMPLETE
Tracking Flags:
Tracking Status
firefox57 --- unaffected
firefox58 --- unaffected

People

(Reporter: vliu, Unassigned)

References

(Depends on 1 open bug, Blocks 1 open bug)

Details

(Whiteboard: [wr-reserve] [gfx-noted][needs-investigation])

See [1] for the Comparison link in detail. It showed that glterrain opt e10s: 11.95% glterrain opt e10s stylo_disabled: 12.99% [1]: Comparison link on Linux x64 opt https://treeherder.mozilla.org/perf.html#/compare?originalProject=try&originalRevision=226893bbeb3e&newProject=try&newRevision=624d8ef848c2d76eed6a2854e03e3c54950ffd12&framework=1
Blocks: 1416635
Whiteboard: [wr-mvp] [triage]
Whiteboard: [wr-mvp] [triage] → [wr-mvp]
Whiteboard: [wr-mvp] → [wr-mvp] [gfx-noted]
Priority: P2 → P3
Whiteboard: [wr-mvp] [gfx-noted] → [wr-reserve] [gfx-noted]
Actually this is bad on Windows too: https://treeherder.mozilla.org/perf.html#/graphs?series=autoland,1653796,1,1&series=autoland,1684187,1,1 It used to be good until bug 1452845 landed, so marking this as dependent on bug 1465784 which is tracking that regression.
Blocks: stage-wr-trains
No longer blocks: stage-wr-nightly
Depends on: 1465784
OS: Linux → All
Priority: P3 → P1
Summary: Bad performance on running glterrain talos suite in Linux x64 opt → Bad performance on running glterrain talos suite
The WR update in bug 1466549 seems to help this test on windows.
Depends on: 1466549
The Windows numbers are now comparable, it's just linux that's worse with WR enabled. Downgrading this to P3 and making it linux specific.
OS: All → Linux
Priority: P1 → P3
Summary: Bad performance on running glterrain talos suite → Bad performance on running glterrain talos suite on Linux64-qr
This is bad again on Windows, from servo/webrender#2814, but we needed that for correctness.
OS: Linux → All
Priority: P3 → P1
Summary: Bad performance on running glterrain talos suite on Linux64-qr → Bad performance on running glterrain talos suite
Bug 1475010 improved the non-WR perf on this test, and also degraded the WR perf on this test. So it increased the size of the regression on Windows.
Blocks: 1475010
Depends on: 1477916
We can't release this to the field, but we can let this ride to beta. However, we want to investigate and understand the bad perf numbers asap. Added whiteboard tag of [needs-investigation].
Priority: P1 → P2
Whiteboard: [wr-reserve] [gfx-noted] → [wr-reserve] [gfx-noted][needs-investigation]
Priority: P2 → P3
Priority: P3 → P2
glterrain talos seems to be improved with wip patch in Bug 1461239 on Windows.
Depends on: 1461239
(In reply to Sotaro Ikeda [:sotaro] from comment #11) > glterrain talos seems to be improved with wip patch in Bug 1461239 on > Windows. With Bug 1461239, glterrain talos became closer to non-WR.
Looking at a recent revision in m-c: https://treeherder.mozilla.org/#/jobs?repo=mozilla-central&selectedJob=208889873&revision=9e08e4f36d5e016e2b5c1959984fbe1af789552f There is a 5.1% difference between QR linux opt and Linux64 opt on glterrain. Much better than the initial ~12% but still. In the short term a good way to improve this test is to optimize the CPU time spent in WebRender during frame building.
(In reply to Nicolas Silva [:nical] from comment #13) > Looking at a recent revision in m-c: > https://treeherder.mozilla.org/#/jobs?repo=mozilla- > central&selectedJob=208889873&revision=9e08e4f36d5e016e2b5c1959984fbe1af78955 > 2f > > There is a 5.1% difference between QR linux opt and Linux64 opt on > glterrain. FWIW, we should focus on windows here, and mostly ignore the linux results. Anyway, bug 1500017 made a big dent here, taking us from ~1.5 to ~1.1, with a non-WR baseline of 0.9.
Depends on: 1500017
Priority: P2 → P4
Depends on: 1505259
bug 1505259 made from 1.08 -> 0.97.

It looks like this is a change of 0.1-0.2ms/frame (which on 1.0ms sounds scary as "~10%"!), which is tolerable, though requires eventual follow-up.

Priority: P4 → P3

Dropping back to P4 because of a misunderstanding about what the priorities mean.

Priority: P3 → P4
Depends on: 1526527

Bug 1526213 is going to improve glterrain on windows10-64-qr about 26%. See Bug 1526213 comment 8.

Depends on: 1526213
Summary: Bad performance on running glterrain talos suite → Bad performance on running glterrain talos suite on Linux64-qr

:jrmuizel, can we remove this bug from stage-wr-trains?

Flags: needinfo?(jmuizelaar)

Yep. Done.

Blocks: stage-wr-next
No longer blocks: stage-wr-trains
Flags: needinfo?(jmuizelaar)

(In reply to Sotaro Ikeda [:sotaro] from comment #19)

Bug 1526213 is going to improve glterrain on windows10-64-qr about 26%. See Bug 1526213 comment 8.

It is sad that the glterrain improvement was a fake. Bug 1531294 addressed it and reverted the glterrain improvement :(

Summary: Bad performance on running glterrain talos suite on Linux64-qr → Bad performance on running glterrain talos suite

:jrmuizel, should this bug block Bug 1386669 again, because of comment 24?

Flags: needinfo?(jmuizelaar)

I made it block wr-67. It would be nice to fix for 67 but I don't think we need to block shipping on it.

Blocks: wr-67
Flags: needinfo?(jmuizelaar)
Priority: P4 → P3

Sotaro, do you think it's likely that we'd do anything about this for 67?

Flags: needinfo?(sotaro.ikeda.g)

I do not think there is something to do for 67.

Flags: needinfo?(sotaro.ikeda.g)
Blocks: wr-68
No longer blocks: wr-67
Blocks: wr-70
No longer blocks: wr-68
No longer depends on: 1526213
No longer blocks: wr-70
Depends on: 1546823

There is only WR.

Status: NEW → RESOLVED
Closed: 2 years ago
Resolution: --- → INCOMPLETE
You need to log in before you can comment on or make changes to this bug.