60.0.3095.6 is a Windows Clang build and I was unable to reproduce the issue on the reported version: 60.0.3095.6 on Windows-10 and Windows-7. No Logs as such were recorded under chrome://gpu. Attached is the chrome://gpu of Windows-7 on which this was tested.

Adding respective component for someone from the respective team to have a look at this.

Deleted: Gpu.txt 9.4 KB

Gpu.txt 9.4 KB View Download

jmadill: can you triage this?

Thanks for the report. This is almost certainly due to 

https://chromium-review.googlesource.com/c/453886/3/src/libANGLE/renderer/d3d/d3d11/RenderStateCache.cpp

Which cuts the cache sizes in half. I don't think this is a serious issue, since the cache is doing what we want and discarding the older entries, but there are concerns about performance and certainly we shouldn't be spamming about:gpu.

Options are to:

* revert the change
* pick somewhat larger sizes (this would ensure we don't overflow the cache)
* instead of dropping a single item, drop a larger number of items
* remove the warning entirely.

We can also consider moving to MRUCache in Chromium's base library. Having more info would be useful.

mikhail.v.gavrilov@gmail.com: what were you doing when you noticed the warnings? Was this just from ordinary browsing over a period of time, or were you using WebGL in someway?

Demostration issue: https://youtu.be/hNvxBYNgEfk

Thanks Mikhail, that's very helpful. I'll take a look.

I was having trouble reproing this. Just saw this was a Clang build. Possibly related somehow to initialization of variables or a copy constructor. Probably not a critical issue, but will keep looking.

I think I have a speculative fix for this - I noticed some undefined behaviours in RenderStateCache on D3D11.

The following revision refers to this bug:
  https://chromium.googlesource.com/angle/angle/+/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1

commit aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1
Author: Jamie Madill <jmadill@chromium.org>
Date: Mon May 29 20:51:09 2017

Ensure gl State structs are zero filled.

In some cases we would hash or memcmp against structs with bools or
other non-filled data. This could have implementation differences,
and may have been causing cache errors on Clang.

BUG= chromium:721648 
BUG= angleproject:2044 

Change-Id: I981a1e6e8d50a33f7fade568497b72b919accfce
Reviewed-on: https://chromium-review.googlesource.com/516383
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Commit-Queue: Jamie Madill <jmadill@chromium.org>

[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/renderer/d3d/d3d11/renderer11_utils.cpp
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/State.cpp
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/renderer/d3d/d3d11/RenderStateCache.cpp
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/Sampler.cpp
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/angletypes.inl
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/renderer/d3d/d3d11/RenderStateCache.h
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/angletypes.cpp
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/angletypes.h
[modify] https://crrev.com/aa0a5446e1a7ba2ae2cd08ae52493778c86d33c1/src/libANGLE/renderer/d3d/d3d11/renderer11_utils.h

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/07f6758930d23cf2790a7ee97bda1d97889638c9

commit 07f6758930d23cf2790a7ee97bda1d97889638c9
Author: geofflang <geofflang@chromium.org>
Date: Tue May 30 15:47:35 2017

Roll ANGLE 7d79fe9..63d8c26

https://chromium.googlesource.com/angle/angle.git/+log/7d79fe9..63d8c26

BUG=722684, chromium:721648 

TBR=jmadill@chromium.org

TEST=bots

CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.win:win_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.android:android_optional_gpu_tests_rel

Review-Url: https://codereview.chromium.org/2912963002
Cr-Commit-Position: refs/heads/master@{#475539}

[modify] https://crrev.com/07f6758930d23cf2790a7ee97bda1d97889638c9/DEPS

Should be fixed now. Not verified.

Still seen again message "[13652:17036:0602/090654.658:WARNING:angle_platform_impl.cc(41)] : getBlendState(157): Overflowed the limit of 2048 blend states, removing the least recently used to make room." on chrome://gpu tab.

Deleted: gpu4.htm 2.5 MB

gpu4.htm 2.5 MB View Download

Version 60.0.3112.11 (Official Build) dev (64-bit)

The fix will be visible in Chrome Canary, versions after 61.0.3116.0

Chrome/61.0.3124.11

Still see message:
[14932:12628:0613/091454.093:WARNING:angle_platform_impl.cc(41)] : TrimCache(35): Overflowed the limit of 2048 blend states, removing the least recently used to make room.

Deleted: gpu6.htm 170 KB

gpu6.htm 170 KB View Download

The following revision refers to this bug:
  https://chromium.googlesource.com/angle/angle/+/f1b47e89631848b074ae0892126e0d675aaa26e4

commit f1b47e89631848b074ae0892126e0d675aaa26e4
Author: Jamie Madill <jmadill@chromium.org>
Date: Wed Nov 22 05:43:51 2017

Fix ComputeGenericHash.

ANGLE's internal hash maps would run MurmurHash on c++ structs to
come up with hash values. Since the hash ran on 4 byte words only,
it would have no understanding that sometimes our structs would only
have meaninful data in the first N bytes, and would include the
garbage at the end in the hash calculation. This fixes the problem by
forcing our structs to be aligned at compile-time.

It also adds custom copy operators for a few classes to ensure that
all bits are copied when the struct is initialized, including the
padding.

Bug:  angleproject:1569 
Bug:  chromium:721648 
Change-Id: I4d11f6e12d9a067b36e1416c7ed15586eff99aab
Reviewed-on: https://chromium-review.googlesource.com/783990
Commit-Queue: Jamie Madill <jmadill@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>

[modify] https://crrev.com/f1b47e89631848b074ae0892126e0d675aaa26e4/src/libANGLE/renderer/d3d/d3d11/RenderStateCache.cpp
[modify] https://crrev.com/f1b47e89631848b074ae0892126e0d675aaa26e4/src/libANGLE/renderer/d3d/d3d11/InputLayoutCache.h
[modify] https://crrev.com/f1b47e89631848b074ae0892126e0d675aaa26e4/src/libANGLE/angletypes.h
[modify] https://crrev.com/f1b47e89631848b074ae0892126e0d675aaa26e4/src/libANGLE/angletypes.cpp
[modify] https://crrev.com/f1b47e89631848b074ae0892126e0d675aaa26e4/src/libANGLE/SizedMRUCache.h
[modify] https://crrev.com/f1b47e89631848b074ae0892126e0d675aaa26e4/src/libANGLE/renderer/d3d/d3d11/renderer11_utils.h
[modify] https://crrev.com/f1b47e89631848b074ae0892126e0d675aaa26e4/src/libANGLE/renderer/d3d/d3d11/Clear11.cpp

It could fix https://bugs.chromium.org/p/chromium/issues/detail?id=765576 ?

No, issue 765576 is some minor D3D compiler warning, not related.

It's possible this could fix the cache overflow warnings, but it's speculative. It should roll out to Canary in a few days, then you could test it.

mikhail.v.gavrilov@gmail.com, can you verify if this is fixed for you now?