Solving some telemetry tests' slowness

Solving some telemetry tests' slowness

After investigation, it turns out that most telemetry test perfomance
regressions were linked to an observer indefinitely waiting for gpu
info. This comes from SystemInfoHandler::GetInfo(), where, when
SwiftShader was being used,
GpuDataManager::IsEssentialGpuInfoAvailable() would always return false,
so the gpu info collection would think it needs to set a GpuObserver in
order to wait for the info to be loaded. By having
GpuDataManager::IsEssentialGpuInfoAvailable() return true when
SwiftShader is used, the gpu info is returned immediately and this
solves the hang.

BUG=702417
Review-Url: https://codereview.chromium.org/2770933008
Cr-Commit-Position: refs/heads/master@{#459651}
Committed: https://chromium.googlesource.com/chromium/src/+/7ae2ba1892a8b4dee8d7c231ac0e0fbe66588912

[sugoi1, 2017-03-24 19:03:20]

Description was changed from

==========
Solving some telemetry tests' slowness

After investigation, it turns out that most telemetry test perfomance
regressions were linked to an observer indefinitely waiting for gpu
info. This comes from SystemInfoHandler::GetInfo(), where, when
SwiftShader was being used,
GpuDataManager::IsEssentialGpuInfoAvailable() would always return false,
so the gpu info collection would think it needs to set a GpuObserver in
order to wait for the info to be loaded. By having
GpuDataManager::IsEssentialGpuInfoAvailable() return true when
SwiftShader is used, the gpu info is returned immediately and this
solves the hang.

BUG=702417
==========

to

==========
Solving some telemetry tests' slowness

After investigation, it turns out that most telemetry test perfomance
regressions were linked to an observer indefinitely waiting for gpu
info. This comes from SystemInfoHandler::GetInfo(), where, when
SwiftShader was being used,
GpuDataManager::IsEssentialGpuInfoAvailable() would always return false,
so the gpu info collection would think it needs to set a GpuObserver in
order to wait for the info to be loaded. By having
GpuDataManager::IsEssentialGpuInfoAvailable() return true when
SwiftShader is used, the gpu info is returned immediately and this
solves the hang.

BUG=702417
==========

[sugoi1, 2017-03-24 19:03:20]

sugoi@chromium.org changed reviewers:
+ chcunningham@chromium.org, crouleau@chromium.org, jbauman@chromium.org,
kbr@chromium.org, zmo@chromium.org

[sugoi1, 2017-03-24 19:03:49]

The CQ bit was checked by sugoi@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-03-24 19:04:28]

Dry run: CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-03-24 20:50:03]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-03-24 20:50:04]

Dry run: This issue passed the CQ dry run.

[Ken Russell (switch to Gerrit), 2017-03-24 21:57:52]

I'm going to lgtm this because it sounds like there's a significant performance
regression on these tests and I think this workaround is non-fatal to any of the
existing GPU tests, but please think about the comment below and perhaps talk
with me and zmo@ about it next week.

https://codereview.chromium.org/2770933008/diff/1/content/browser/gpu/gpu_dat...
File content/browser/gpu/gpu_data_manager_impl_private.cc (right):

https://codereview.chromium.org/2770933008/diff/1/content/browser/gpu/gpu_dat...
content/browser/gpu/gpu_data_manager_impl_private.cc:433: use_swiftshader_;
What does GpuDataManager::GetGPUInfo() return now in this case?

If it returns a GPUInfo with no information at all in it -- I think that may
violate the contract of the API.

We should consider populating the GPUInfo, if empty, with information indicating
SwiftShader's being used. (The context_info in particular.)

[sugoi1, 2017-03-25 13:56:20]

The CQ bit was checked by sugoi@chromium.org

[commit-bot: I haz the power, 2017-03-25 13:56:25]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-03-25 13:59:50]

CQ is committing da patch.

Bot data: {"patchset_id": 1, "attempt_start_ts": 1490450180329500, "parent_rev":
"a9461a0586471e6fa16c02201f69eb366acc8350", "commit_rev":
"7ae2ba1892a8b4dee8d7c231ac0e0fbe66588912"}

[commit-bot: I haz the power, 2017-03-25 14:00:40]

Description was changed from

==========
Solving some telemetry tests' slowness

After investigation, it turns out that most telemetry test perfomance
regressions were linked to an observer indefinitely waiting for gpu
info. This comes from SystemInfoHandler::GetInfo(), where, when
SwiftShader was being used,
GpuDataManager::IsEssentialGpuInfoAvailable() would always return false,
so the gpu info collection would think it needs to set a GpuObserver in
order to wait for the info to be loaded. By having
GpuDataManager::IsEssentialGpuInfoAvailable() return true when
SwiftShader is used, the gpu info is returned immediately and this
solves the hang.

BUG=702417
==========

to

==========
Solving some telemetry tests' slowness

After investigation, it turns out that most telemetry test perfomance
regressions were linked to an observer indefinitely waiting for gpu
info. This comes from SystemInfoHandler::GetInfo(), where, when
SwiftShader was being used,
GpuDataManager::IsEssentialGpuInfoAvailable() would always return false,
so the gpu info collection would think it needs to set a GpuObserver in
order to wait for the info to be loaded. By having
GpuDataManager::IsEssentialGpuInfoAvailable() return true when
SwiftShader is used, the gpu info is returned immediately and this
solves the hang.

BUG=702417

Review-Url: https://codereview.chromium.org/2770933008
Cr-Commit-Position: refs/heads/master@{#459651}
Committed:
https://chromium.googlesource.com/chromium/src/+/7ae2ba1892a8b4dee8d7c231ac0e...
==========

[commit-bot: I haz the power, 2017-03-25 14:00:41]

Committed patchset #1 (id:1) as
https://chromium.googlesource.com/chromium/src/+/7ae2ba1892a8b4dee8d7c231ac0e...